Search is not available for this dataset
pmcid
large_stringlengths 8
11
| pmid
large_stringlengths 0
9
| authors
large_stringlengths 0
33.4k
| title
large_stringlengths 0
2.49k
| publication_date
large_stringlengths 10
10
| keywords
large_stringlengths 0
17k
| abstract
large_stringlengths 0
122k
| text
large_stringlengths 9
8.39M
|
|---|---|---|---|---|---|---|---|
PMC9647705 | Lignesh Durai,Sushmee Badhulika | Current Challenges and Developments in Perovskite-Based Electrochemical Biosensors for Effective Theragnostics of Neurological Disorders | 27-10-2022 | Early-stage diagnosis of neurological disease and effective therapeutics play a significant role in improving the chances of saving lives through suitable and personalized courses of treatment. Biomolecules are potential indicators of any kind of disorder in a biological system, and they are recognized as a critical quantitative parameter in disease diagnosis and therapeutics, collectively known as theragnostics. The effective diagnosis of neurological disorders solely depends on the detection of the imbalance in the concentration of neurological biomarkers such as nucleic acids, proteins, and small metabolites in bodily fluids such as blood serum, plasma, urine, etc. This process of neurological biomarker detection can lead to an effective prognosis with a prediction of the treatment efficiency and recurrence. While review papers on electrochemical, spectral, and electronic biosensors for the detection of a wide variety of biomarkers related to neurological disorders are available in the literature, the prevailing challenges and developments in perovskite-based biosensors for effective theragnostics of neurological disorders have received scant attention. In this Mini-Review, we discuss the topical advancements in design strategies of perovskite-based electrochemical biosensors with detailed insight into the detection of neurological disease or disorder-specific biomarkers and their trace-level detection in biological fluids with high specificity and sensitivity. The tables in this Review give the performance analysis of recently developed perovskite-based electrochemical biosensors for effective theragnostics of neurological disorders. To conclude, the current challenges in biosensing technology for early diagnosis and therapeutics of neurological disorders are discussed along with a forecast of their anticipated developments. | Current Challenges and Developments in Perovskite-Based Electrochemical Biosensors for Effective Theragnostics of Neurological Disorders
Early-stage diagnosis of neurological disease and effective therapeutics play a significant role in improving the chances of saving lives through suitable and personalized courses of treatment. Biomolecules are potential indicators of any kind of disorder in a biological system, and they are recognized as a critical quantitative parameter in disease diagnosis and therapeutics, collectively known as theragnostics. The effective diagnosis of neurological disorders solely depends on the detection of the imbalance in the concentration of neurological biomarkers such as nucleic acids, proteins, and small metabolites in bodily fluids such as blood serum, plasma, urine, etc. This process of neurological biomarker detection can lead to an effective prognosis with a prediction of the treatment efficiency and recurrence. While review papers on electrochemical, spectral, and electronic biosensors for the detection of a wide variety of biomarkers related to neurological disorders are available in the literature, the prevailing challenges and developments in perovskite-based biosensors for effective theragnostics of neurological disorders have received scant attention. In this Mini-Review, we discuss the topical advancements in design strategies of perovskite-based electrochemical biosensors with detailed insight into the detection of neurological disease or disorder-specific biomarkers and their trace-level detection in biological fluids with high specificity and sensitivity. The tables in this Review give the performance analysis of recently developed perovskite-based electrochemical biosensors for effective theragnostics of neurological disorders. To conclude, the current challenges in biosensing technology for early diagnosis and therapeutics of neurological disorders are discussed along with a forecast of their anticipated developments.
Biomarkers or biological markers are the measurable alterations of the biochemical or biomolecule concentrations in any biological medium such as bodily fluids, tissues, or cells. In recent decades, biomarkers have been recognized as potential indicators for various diseases. The biomarkers can be classified into two types, known as biomarkers of exposure and biomarkers of disease. This classification of biomarkers includes the subclasses named susceptibility, diagnostic, prognostic, and predictive biomarkers. The susceptibility biomarkers are the indicators of objectively measured environmental agents in the biological system. The diagnostic biomarkers are known as indicators for diseases related to disorders in a biological system. The prognostic biomarkers are the indicators of disease recurrence in a biological system irrespective of the treatment provided. Finally, the predictive biomarkers provide information related to the response of a biological system to targeted therapy. This above-mentioned classification of biomarkers was recognized as an order of events by Perera et al. A superlative biomarker should be noninvasive, simple, accurate to measure reproducibly, highly related to a biological process of interest, and predictive in the progression of the same. Generally, the anomalous values of the biomarkers in the biological fluids provide basic information on the development of irreversible injury in the biological system and aid the consideration of the pathophysiology of the respective progression. Major examples of the biomarkers range from the pulse and blood pressure of a biological system over elementary chemistries to the highly multifaceted laboratory examination of biological fluids and tissues. The recent development in the field of biomarkers has become more disease-specific, which can be used for the detection of deadly diseases such as cancers, cardiovascular diseases, neurologic disorders, oxidative stress, and metabolic instabilities such as metabolic syndrome, diabetes, chronic gout, cystic fibrosis, etc. The early detection of biomarkers in biological fluids such as blood, urine, and plasma would be highly valued for the early diagnosis and treatment of any specific diseases. Here, the major limitation observed is the selectivity of the biosensor toward the trace-level concentration of biomarkers in the presence of various other biological compounds in the body fluids. The wide variety of biomarkers includes antigens, enzymes, deoxyribonucleic acid (DNA), messenger ribonucleic acid (mRNA), amino acids, proteins, etc. The clinical significance of developing a reliable, rapid, cost-effective, and powerful biomarker detection technique has garnered the substantial attention of research toward biomarker detection which can effectively assist the process of prognosis, diagnosis, and monitoring of the recurrence of diseases. Various analytical techniques developed over a decade for the detection of biomarkers include enzyme-linked immunosorbent assay (ELISA), mass-sensing biodegradable citrus degreaser (Bio-CD), optical detection techniques (such as fluorescence, chromatography analysis, etc.), electrochemical detection, and gel-electrophoresis analysis. These conventional techniques with immunoassays facilitate the detection of targeted biomarkers via capture-antibody functionalization on a solid substrate for target capturing and assay readouts. Although some of these techniques provide a stable quantification of targeted biomarkers in biological fluids, they unanimously possess some major drawbacks owing to the nonspecific adsorption of nontargeted proteins onto the surface of a biosensor, which impacts the selectivity, sensitivity, and accuracy of the technique. On the other hand, these techniques necessitate trained manpower, expensive apparatus, and a time-consuming pretreatment process of the sample. To overcome the major drawbacks encountered by the conventional biomarker detection techniques, it is indispensable to achieve a low-cost, simple, rapid, highly sensitive, selective, accurate, and portable sensing technique for point-of-care and clinical diagnostics. These predominant requirements of biomarker detection for clinical and point-of-care diagnostic applications were achieved via wide research and development in the field of biosensors. Generally, an excellent biosensor should possess high specificity, sensitivity, and reusability. The biosensors can be classified into two types depending on the immobilization process, namely, enzymatic and nonenzymatic biosensors. In an enzymatic biosensor, the transducer has been immobilized with an enzyme to produce a signal proportional to target analyte concentrations. Nonenzymatic electrochemical biosensors depend on the redox reaction of the targeted analyte on the surface of the transducer. Generally, enzymatic biosensors have been reported with some significant limitations such as instability and low readout signal strength. Depending on the detection techniques employed for the sensing of biomarkers, the biosensors can be broadly classified into electrochemical, spectral, electronic, magnetoresistive, and piezoelectric biosensors, etc. This article aims to deliver a Review of the current challenges and recent developments in the design strategies using perovskite materials for different types of biosensors and their application in the theragnostics of neurological disorders. It is anticipated that this article will stimulate a broader research interest and insight into perovskite-based biosensors and their efficacy in the successful detection of complex biomarkers in biological fluids. This Review primarily consists of a brief overview of recently reported perovskite-based electrochemical biosensors for the early diagnosis and effective therapeutics of neurological disorders.
Perovskite materials are also known as ceramic oxide materials with the general formula of ABX3, where A and B are the cations and X is an anion. The X anion can be oxygen or any other element with larger ionic radii such as halides, nitrides, and sulfides. Generally, there are three types of perovskite materials reported depending on the structure of the material: (1) containing localized electrons; (2) containing delocalized energy bands, and (3) the transition state. These ABO3 perovskite materials form a cubic or nearly cubic crystal structure like any other transition metal oxide materials facilitated with a low-temperature phase transition process. Additionally, these materials are applied for a wide range of applications with respect to their simple crystal structures and extraordinary ferroelectric and dielectric properties. Apart from the ABO3 perovskite structure, there are several other perovskite structures such as layered (A2BO4), double-layered (A2BB′O6), triple-layered (A2A′B2B′O9), etc. Alhough the oxide perovskite materials possess the basic transition metal oxide structures, their inherent properties highly differ from the basic metal oxides, such as ionic conduction characteristics, insulator–metal transition, dielectric, metallic, and superconducting. Apart from the special structures of perovskites and their tunable electromechanical properties, these materials have gathered a wide spectrum of research interest owing to the large availability of cations and anions across the Periodic Table. Owing to these merits of perovskite materials, they have been widely used in a variety of applications such as random-access memories, actuators, tunable microwave devices display, piezoelectric devices, transducers, wireless communications, sensors, and capacitors. An interesting property of perovskite materials is the phase transition of the material with respect to temperature. Fu and Itoh have reported the phase transition of BaTiO3 (an ABO3-type perovskite material) between cubic and rhombohedral with an impact on the dielectric constant of the material. This proves that the phase transition mechanism depending on temperature and displacement of the B cation impacts the dielectric property of perovskite materials. Owing to the tunable crystal structure with a wide range of combinational availability and distinguished phase transition mechanisms, perovskite materials have been reported with various physicochemical properties such as electrical conductivity (ReO3, SrFeO3, LaCrO3, LaCoO3, and LaNiO3), piezoelectricity property (Pb(Zr, Ti)O3 and (Bi, Na)TiO3), ion conductivity (La(Ca)AlO3, BaZrO3, CaTiO3, and SrZrO3), superconductivity (La0.9Sr0.1CuO3 and YBa2Cu3O7), magnetic property (LaMnO3 and La2NiMnO6), ferromagnetic property (BaTiO3 and PdTiO3), and catalytic property (LaCoO3 and LaMnO3), etc. A variety of synthesis techniques such as solid-state reactions, the Pechini method, the coprecipitation method, hydrothermal synthesis, gas-phase preparations, and wet chemical and sol–gel methods have been evolved in the past few decades to overcome drawbacks such as inhomogeneity, defects in the lattice space, incorporation of chemical impurities, and coarseness of synthesized particles. Perovskite materials are widely used in different kinds of applications as they possess a variety of physiochemical properties via tunable crystal structures depending on the synthesis techniques and the elemental combination. The major applications developed and reported depending on the stability and unique physiochemical properties of the perovskite materials are as follows;: they are used in capacitors, photoelectrochemical cells, drug delivery, catalysts, recording applications, high-temperature heating applications, spintronics devices, thermal barrier coatings, laser applications, frequency filters for wireless communications, sensors, nonvolatile memories, ultrasonic imaging biosensors, actuators, and transducers, etc. Among these mentioned applications of perovskite materials, the electrochemical biosensor has gathered major attention owing to its sensitivity, selectivity, excellent reproducibility, and long-term stability, etc.
In the past few decades, a significant number of electrochemical biosensors were reported for various biomarker/biomolecule sensing uses for clinical diagnosis and other applications. The vital component in any electrochemical sensor is the working electrode; this determines the efficacy of the sensor. To improve the performance of these sensors toward the targeted analytes, conventional electrodes such as glassy carbon electrode (GCE), screen-printed carbon electrode (SPE), etc. were reported with different surface chemical modification processes and materials. Apart from these electrode modification processes via different functional materials, the surface of the electrodes was additionally immobilized with different chemical groups such as amine and carboxyl (1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide: EDC), aldehyde (hydrazide), thiol (maleimide), etc. to ensure the solid support for the labeling biomarkers such as enzyme, antibody, and nucleic acid. Although these kinds of immobilization processes improve the selectivity, sensitivity, and chemical and electrochemical stability in larger usable potential windows with resistance to fouling, the unsuitable immobilization may cause loss of activity, less specificity, and low biocompatibility of the sensor. Recently, various electrochemical biosensors were developed and reported with high sensitivity and selectivity using carbonaceous functional materials for electrode modifications. Similarly, a variety of 2D materials with different morphologies were also studied as potential functional materials for chemically modified electrochemical biosensors. The major drawbacks encountered by both the carbonaceous and 2D materials were the surface fouling effect known as adlayer formation along with the chemical and electrochemical instability of the materials after the process. Along with various developments in electrochemical biosensors, the induction of label-free biosensors garnered significant research interest because of its simple analysis without involving any complex labeling procedures. The label-free electrochemical biosensors have been widely reported for biomarker detection, DNA and protein sensing, etc. Here in this Review, we focus on the development of a label-free electrochemical biosensor using a perovskite material which was employed in the detection of a range of biomolecules in biological fluids amid various clinical diagnoses. Neurotransmitters are well-known potential biomarkers for various neurological diseases like Parkinson’s disease, Alzheimer’s disease, and depression. The low concentration of dopamine in biological fluids (10–8–10–6 M) demands an ultrasensitive and highly selective electrochemical biosensor for real-time clinical diagnosis. In recent years, perovskite materials such as NaNbO3, SrPdO3, LaFeO3, LaCoO3, LaFeO3, and β-NaFeO2 have been reposted as the functional material for the chemically modified based electrochemical biosensors targeting dopamine in biological fluids. Unlike 2D materials and other carbonaceous materials, perovskite materials are not known for their electrocatalytic activity, but they too have intense electrocatalytic active sites which favor the electrooxidation of dopamine into dopamine-o-quinone.
The electrochemical sensing mechanism using perovskite-based chemically modified electrodes reported depends on the electrocatalytic active sites. These are the octahedral sites of the perovskite structure with Nb5+ cations which favors the delocalization of charge carriers during the redox reaction at the electrode–electrolyte interface. Similarly, Atta et al. reported SrPdO3-modified CPE as an electrochemical biosensor for dopamine sensing in which the homogeneously distributed Pd4+ cations at the octahedral sites of the perovskite enhanced the rate of electron transfer during the redox reaction of the dopamine molecule in the medium. The β-NaFeO2 material was examined for dopamine sensing, and it was observed that the Fe3+/Fe4+ oxidation states of Fe in the respective octahedral and tetrahedral sites of the perovskite structure favored the delocalization of the charge in the electrocatalytic reaction. These concluding statements prove the structure-oriented electrocatalytic activity of the perovskite materials and their high chemical stability. On the other hand, the major imbalance between the free radicals and the antioxidants in a biological system indicates neurological disorders (such as Alzheimer’s, Parkinson’s, etc.). l-Tryptophan (Trp) is the major therapeutic biomarker in the treatment of various inflammatory diseases including schizophrenia, hallucinations, delusions, nausea, headache, Alzheimer’s, hepatic diseases, and Parkinson’s. Owing to improper metabolization, the high concentration of Trp can be very harmful to the brain. Govindasamy et al. reported a label-free electrochemical biosensor for the detection of Trp in biological fluids such as urine and blood serum samples using SrTiO3@RGO/GCE via amperometry analysis. The enhanced sensitivity and selectivity of the sensor were attributed to the large active surface area of RGO and π-to-π interaction between RGO and SrTiO3. The basic sensing mechanism via electrochemical oxidation of Trp is shown in Figure 1a. Wang et al. reported a LaNi0.5Ti0.5O3 (LNT) nanoparticle-modified carbon paste electrode for the nanomolar detection of l-cystine. The enhanced sensitivity and selectivity of the LNT/CPE toward l-cystine were attributed to the Ni centers in the perovskite structure facilitated by a Ni3+/Ni2+ electro-oxidation mechanism. The current–time responses of CPE and LNT/CPE upon the successive addition of different concentrations of l-cystine were given. Grace et al. reported the pristine graphene-decorated GdTiO3 perovskite, a binary composite-modified GCE for simultaneous trace-level detection of dopamine and acetaminophen in blood serum, tablets, and urine samples as shown in Figure 1b. The outstanding sensitivity, selectivity, and stability of the electrode have been ascribed to the π interactions between the graphene and the perovskite. The sensing performance of reported perovskite-based electrochemical biosensors for the theragnostics of neurological disorders is given in Table 1.
The major challenge of the electrochemical biosensor for the theragnostics of neurological disorders is the trace-level concentrations of the targeted biomarker in the presence of other interfering analytes. The secondary challenge that has attracted research interest toward the perovskite-based electrochemical sensor is the selection and synthesis of a suitable perovskite material as the sensing platform for different neurological biomarkers. Generally, perovskite materials are reported with a large number of electrocatalytic active sites which provides high sensitivity and stability. However, the selectivity of the sensor toward the targeted biomolecules would be compromised because of interfering biomolecules present in the biofluids. Generally, this drawback is addressed via the immobilization of antibodies through the enzymatic process. The enzymatic sensing of the biomarkers was reported with its drawbacks such as instability and operational constraints like pH, temperature, and humidity. On the other hand, the nonenzymatic detection of biomarkers based on nanomaterials was reported with high stability owing to the unique morphology and high surface area. Also, the sensors exhibited high selectivity and sensitivity owing to the presence of electrocatalytic active sites and variable oxidation states. However, the pH and operational temperatures are the two major factors that impact the efficacy of any perovskite-based chemically modified biosensors.
Despite the availability of a wide spectrum of biomarkers, the identification of suitable biomarkers for the relevant application of diagnosis is of the highest priority. First, the low physiological concentration of various neurotransmitters must be addressed, as the biomarkers demand an ultrasensitive detection technique. Second, the presence of these biomarkers in the biological fluids discloses the presence of other biomolecules and cell tissues in the complex. This demands a selective biosensing platform that is active toward the targeted analyte and inactive to other biomolecules. Thereby, a highly selective biosensor is required for clinical sample analysis. Third, stability and reusability are two vital parameters for any biosensor defining the efficacy of the sensor. These parameters should be improved when compared to the current progress status of the reported biosensor. This can be achieved through the application of various highly stable perovskite-based nanomaterials with unique structures and properties. The cost-effective development of biosensing technology is highly crucial for the commercialization of biosensors and related early point-of-care diagnosis applications. Finally, the miniaturization of the biosensors for the evolution of handy and wearable point-of-care diagnostic devices remains a vital challenge for early diagnosis and therapeutics of neurological disorders. The future scope in the field of biosensors is anticipated to rely on the development of highly stable, sensitive, and selective detection of complex biomarkers for early diagnosis of rare and deadly diseases. As mentioned above, the evolution of a low-cost, facile, highly selective, and sensitive biosensing platform has gathered a wide scope of research interest toward the point-of-care and wearable diagnostic applications. In this aspect, the perovskite-based nanostructured material is a promising special structured material with high stability, selectivity, and unique electronic and piezoelectric properties. Despite the advantages of the perovskite materials, only a few reports are presented for biosensor applications (discussed in this Review). The overall advancement expected in the field of biomarkers in the near future would be directed toward the following: (1) A user-friendly biosensing platform integrated into smartphone applications for the detection of complex biomarkers in various biological fluids. (2) Development in various active functional materials for microfluidic devices for in vivo and in vitro diagnosis and therapeutics. (3) Depending on the versatile properties of specially structured nanomaterials, an exponential development in the efficiency of biomarker detection is expected using new electrochemical sensing techniques such as photoelectrochemical analysis, sonoelectrochemical analysis microwave-activated electrochemical analysis, etc. 4. The development of wearable biosensing devices for early disease diagnosis and therapeutics of neurological disorders. (5) Provided with the miniaturization of these biosensing platforms, they could be integrated with a microelectromechanical system (MEMS) for a novel drug delivery application.
In summary, this Review provides insights into the current challenges and development in perovskite-based biosensors for effective theragnostics of neurological disorders in human beings. Biomarkers are the biological constituents of the bodily fluids which can indicate the normal and abnormal processes of a biological system. The detection of an abnormal concentration of targeted biomolecules in bodily fluids is associated with various diseases and disorders. In this regard, the most recent pertinent articles on different electrochemical biosensors using different perovskite-based chemically modified electrode-based biosensors and their analytical performance toward an early clinical diagnosis of neurological disorders were assessed. In short, every biosensing technique has its advantages and disadvantages in the detection of targeted biomolecules. However, recent research has progressed through simultaneous biomarker detection techniques for effective and low-cost clinical diagnosis of different diseases. On the other hand, the miniaturization process of the biosensors provided with the application of novel perovskite-based nanostructured materials has paved a new avenue of research for wearable and user-friendly point-of-care diagnosis. Also, the authors believe that this broad insight into different perovskite-based electrochemical biosensors and their applications would encourage more research in the field of biosensing based on stable, low-cost perovskite materials. |
||
PMC9647709 | Michael Gibbons,Jessica M. Hong,Mikelle Foster,Mariya Chavarha,Shirley Shao,Llyke Ching,Victoria A. Church,Lauren Schiff,Sara Ahadi,Marc Berndl,Phillip Jess,Annalisa Pawlosky | Million spot binding array platform for exploring and optimizing multiple simultaneous detection events | 08-11-2022 | Genomics,Sequencing,High throughput screening,Molecular biology | Summary Large-scale, high-throughput specificity assays to characterize binding properties within a competitive and complex environment of potential binder-target pairs remain challenging and cost prohibitive. Barcode cycle sequencing (BCS) is a molecular binding assay for proteins, peptides, and other small molecules that is built on a next-generation sequencing (NGS) chip. BCS uses a binder library and targets labeled with unique DNA barcodes. Upon binding, binder barcodes are ligated to target barcodes and sequenced to identify encoded binding events. For complete details on the use and execution of this protocol, please refer to Hong et al. (2022). | Million spot binding array platform for exploring and optimizing multiple simultaneous detection events
Large-scale, high-throughput specificity assays to characterize binding properties within a competitive and complex environment of potential binder-target pairs remain challenging and cost prohibitive. Barcode cycle sequencing (BCS) is a molecular binding assay for proteins, peptides, and other small molecules that is built on a next-generation sequencing (NGS) chip. BCS uses a binder library and targets labeled with unique DNA barcodes. Upon binding, binder barcodes are ligated to target barcodes and sequenced to identify encoded binding events. For complete details on the use and execution of this protocol, please refer to Hong et al. (2022).
The concept of the BCS workflow is outlined in the Graphical Abstract. The key components required for BCS include: 1) preparation of binders and targets for use on a NGS chip, 2) attachment of targets to the NGS chip, 3) a mechanism for binders to deposit their unique DNA barcodes onto the chip at the binding site, and 4) DNA sequencing of the binder barcodes may be used to identify the binders. Multiple rounds of binding and ligation (step 3) may be performed before DNA sequencing. The mechanism for barcode deposition requires the presence of a short oligonucleotide sequence called a “foundation”, which is positioned beside each peptide target to act as a receiver for binder barcodes. Each foundation contains a DNA barcode unique to its corresponding target, which is used to identify the target. This assay may be adapted for use with any molecular binder-target pair (ex. small molecule, quantum dot, protein and RNA, etc.) where both the binder and the target are amenable to conjugation with oligonucleotides, and where the presence of oligonucleotide tails does not significantly perturb binding affinity and specificity. The buffers described here were designed specifically to optimize binding dynamics between and structural stability of our binders and targets of interest, aptamers, and peptides. Because of the flexibility of the method to assay the binding properties of various types of molecules, it is important for users to adjust binding buffers and temperatures to facilitate binding and structural stability for their system. Users may choose to replace the recommended buffers and incubation parameters suitable to the user’s molecules of interest and the environment of their end application(s). Here we demonstrate a single cycle of BCS using the Spot Tag system, a 12-amino acid fSpot-Tag and anti-Spot-Tag nanobody engineered to recognize the Spot-Tag specifically and with high affinity (dissociation constant Kd of approximately 6 nM) (Virant et al., 2018; Braun et al., 2016). The nanobody is approximately 10-fold smaller than an antibody (approximately 15 kDa vs 150 kDa), is a stable and robust reagent, is commercially available from Chromotek as a bivalent construct (two anti-Spot-Tag nanobodies genetically fused via a linker), and includes a C-terminal recognition motif LPETG for site-specific sortase-mediated conjugation of the protein to small molecules (Antos et al., 2017). Furthermore, conjugation of nanobodies to oligonucleotide tails has been reported and does not appear to affect nanobody functionality (Fabricius et al., 2018). Peptide targets are displayed on the NGS chip through ligation to pre-existing P7 oligonucleotide adapters (Figure 1). Each peptide target exists as a peptide-oligonucleotide conjugate (POC), where a short, barcoded oligonucleotide is covalently attached to the C-terminus. Colocalization of the target and foundation on the chip is achieved by first forming a target-foundation complex in solution. Two short oligonucleotides called the “forward” and “reverse cololinkers” have regions of complementarity to the binder and target, as well as to pre-existing P7 oligonucleotide adapters on the Illumina DNA sequencing chip. The term cololinker is short for “colocalization linker,” since the cololinkers link together the POC and foundation (Figure 1, step 1a) so that they may colocalize on the chip. Experimental design can include negative control targets such as empty foundations (DNA foundation pieces without any targets colocalized) to measure signal-to-noise ratio and off-target binder-DNA binding, or scramble peptides to control for nonspecific, off-target binding. The chip is rinsed with a buffer solution (Figure 1, step 1b) and the target-foundation complex is flowed onto the chip (Figure 1, step 2). The cololinkers facilitate binding of the target-foundation complex to P7 adapters on the chip, and unbound elements are washed away (Figure 1, step 3). The target and foundation are ligated to the 3′ ends of two adjacent P7 adapters (Figure 1, step 4). Unligated elements are washed away (Figure 1, step 5), and the single-stranded regions of P5 and P7 are blocked using complementary DNA (Figure 1, step 6). Each nanobody binder contains a DNA barcode whose purpose is to identify the binder. Binding events are recorded on the chip by transferring the binder’s DNA barcode onto the foundation (Figure 2). Before introducing binders to the NGS chip, a nanobody complex (Figure 2, step 1) is created by annealing the nanobody 5′ end to a short “universal bridge” oligonucleotide to form a ligation site. Upon binding the target (Figure 2, step 2), the 5′ binder barcode is ligated to the 3′ end of the foundation, and the universal bridge is washed away (Figure 2, step 3). The binder tail is cleaved downstream of the binder barcode and washed off the chip (Figure 2, step 4), leaving the binder barcode ligated to the foundation. Optionally, multiple cycles of binding, ligation, and cleavage may be performed, where each new binder barcode is ligated to the 3′ end of the previous binder barcode (not depicted). Assisted by a short universal NGS adapter bridge oligonucleotide, the universal NGS adapter for DNA sequencing is then ligated to the binder barcode (Figure 2, steps 5 and 6), and the DNA construct is sequenced (Figure 2, step 7).
The following sections describe general principles which apply to designing DNA components of the BCS platform, including the foundation, POCs, cololinkers, binder barcodes, universal bridge, and blocking sequences. All oligonucleotide components may be ordered at 100 nM scale with high-pressure liquid chromatography purification and resuspended at 100 μM concentration in nanopure, nuclease-free (NF) H2O. All DNA components may be stored at −20°C.CRITICAL: Screen every designed sequence against other BCS DNA components to avoid undesired binding complexes. An online tool such as IDT Oligo Analyzer may be used for screening. We have observed as few as four consecutive complementary nucleotides between BCS components to induce off-target binding.
The purpose of this section is to design the foundation sequence. The foundation provides a site for a binder to deposit its barcode beside the peptide target. The 5′-phosphorylated end of the foundation is designed to be ligated to the existing P7 sequence on the chip. The foundation base is complementary to the 5′ end of the forward cololinker. The foundation barcode is a short oligonucleotide sequence unique to each peptide target. The bridge-binding sequence is complementary to the 3′ end of the universal bridge. The 3′ end of the foundation serves as the ligation site for a binder barcode (Figure 4A). 1. Design foundations. a. Generate a foundation base sequence according to general principles of primer design (https://eu.idtdna.com/pages/education/decoded/article/designing-pcr-primers-and-probes). The 5′ end must be phosphorylated for successful ligation. b. Generate a unique foundation barcode for each peptide target. i. Barcodes should be selected such that they minimize the number of homopolymers, constrain the GC content to a desired range, and minimize similarity between sequences. In practice we used a max homopolymer length of 2, GC range of 0.4–0.6, and a minimum distance of 2 between all barcode pairs. Examples of validated barcodes are provided in Table 1. Table 1 Examples of validated foundations Foundation name Sequence Fd7 /5Phos/CGACTGCGAGCTGATGGCCTTGATGATAACG Fd8 /5Phos/CGACTGCGAGCTGATGCGTACTAGGATAACG Fd11 /5Phos/CGACTGCGAGCTGATGTGTACGCAGATAACG Fd12 /5Phos/CGACTGCGAGCTGATGCGTTTGCAGATAACG Fd13 /5Phos/CGACTGCGAGCTGATGTCTTTCCGGATAACG Fd14 /5Phos/CGACTGCGAGCTGATGTTGCTCACGATAACG Fd15 /5Phos/CGACTGCGAGCTGATGGAGTTACGGATAACG Fd16 /5Phos/CGACTGCGAGCTGATGTGATATAGGATAACG Fd17 /5Phos/CGACTGCGAGCTGATGACCTTAGAGATAACG Fd18 /5Phos/CGACTGCGAGCTGATGAGTTGCTTGATAACG Fd19 /5Phos/CGACTGCGAGCTGATGAGGTACCAGATAACG Fd20 /5Phos/CGACTGCGAGCTGATGCACTTACGGATAACG Fd21 /5Phos/CGACTGCGAGCTGATGTTGGGCAAGATAACG Fd22 /5Phos/CGACTGCGAGCTGATGTTGGGCAAGATAACG Fd23 /5Phos/CGACTGCGAGCTGATGTTCCACGTGATAACG Fd24 /5Phos/CGACTGCGAGCTGATGAGGAGCAAGATAACG Fd25 /5Phos/CGACTGCGAGCTGATGTTCCCTTCGATAACG Fd26 /5Phos/CGACTGCGAGCTGATGTCTGAGGTGATAACG Fd27 /5Phos/CGACTGCGAGCTGATGTCATGTGGGATAACG Fd28 /5Phos/CGACTGCGAGCTGATGCACCAAACGATAACG Fd29 /5Phos/CGACTGCGAGCTGATGATTGTCCCGATAACG Fd31 /5Phos/CGACTGCGAGCTGATGTGGCATCTGATAACG Fd32 /5Phos/CGACTGCGAGCTGATGCTTCTAGCGATAACG Fd43 /5Phos/CGACTGCGAGCTGATGCAGCACATGATAACG c. Generate a bridge-binding sequence according to general principles of primer design. Note: We observed that certain foundation barcodes affected the efficiency of peptide deposition onto the chip. To quantify these differences, we performed a single cycle of BCS on one target using every foundation barcode we designed. For each target-foundation pair, we measured (1) the seeding efficiency of the target-foundation pair onto the P7 sequences on the flow cell, and (2) the efficiency of binder barcode capture. We ultimately selected foundations that produced reliable seeding density and binder barcode capture, as well as a low incidence of nonspecific binding. 2. Order the foundations and resuspend them at 100 μM in NF H2O. Pause point: Store stably at −20°C.
The purpose of this section is to design target POCs. Each target is displayed on the chip via covalent attachment to a short single-stranded oligonucleotide tail which is eventually ligated at the 5′-phosphorylated end to an existing P7 adapter on the chip. 3. Design the oligonucleotide tail (Figure 4A). a. Generate the target base sequence according to general principles of primer design. The 5′ end must be phosphorylated for successful ligation. b. Design a spacer region of approximately 21 nucleotides. 4. Purchase or synthesize target-oligonucleotide conjugates. a. Synthesis may be achieved through commercially available conjugation kits or chemical techniques such as click chemistry (Fantoni et al., 2021). Note: The Spot-tag peptide-oligonucleotide conjugates (POCs) used in this protocol were manufactured by GenScript. Users interested in creating their own POCs may contact lead contact to request further details about our in-house method developed for POC synthesis. 5. Resuspend POCs in 100 μL and separate into single-use 10 μM aliquots to avoid freeze-thaw cycles. Pause point: Store stably at −20°C.
The forward and reverse cololinkers anneal to one another at their 3′ ends to form a U-shaped complex between the foundation, POC, and P7 adapters on the NGS chip. Both the forward and reverse cololinkers include a region of complementarity to P7 on the chip, allowing the forward cololinker in the assembled complex to bring the 3′ end of the foundation into proximity with a P7 adapter on the chip and the reverse cololinker to bring the 3′ end of the POC into proximity with another nearby P7 adapter (Figure 4A). 6. Design the forward and reverse cololinkers with complementarity to the appropriate regions as depicted in Figure 4A. Note: These oligonucleotides are not modified with 5′ phosphorylation (to avoid non-specific ligation to the cololinkers). 7. Order the cololinkers and resuspend them at 100 μM in NF H2O. 8. Create a cololinker stock solution containing a 3:1 ratio of forward to reverse cololinker in the Hybridization Buffer. Pause point: Store stably at −20°C.
The purpose of this section is to design the binder tail that contains the DNA barcode used to identify the binder. On the 5′ phosphorylated end of the binder tail is a short “ligation spacer” complementary to the universal bridge. When the universal bridge anneals to the ligation spacer, a short double-stranded region is created to facilitate ligation between the binder tail and foundation. Following the ligation spacer is a 6- or 8-nt binder barcode unique to each binder and another short region of complementarity to the universal bridge. A downstream restriction enzyme site allows the binder to be cleaved and washed away, leaving the binder barcode attached to the foundation (Figure 4B). The 3′ end of the binder tail is conjugated to the C-terminus of the binder (anti-Spot-Tag nanobody). 9. Design the binder barcodes (see designing the foundation section for barcode design guidelines). Note: We noted similarities in performance between 6- and 8-nt barcodes. 8-nt barcodes may be useful in specific cases (ex. experiments with multiple binding cycles, NGS of the reverse read).Note: We tested binder tail sequences with and without the ligation and restriction site spacers. The addition of spacers improved enzyme activity, presumably by extending the double-stranded region surrounding the single-stranded break (Doherty and Suh, 2000). A shorter T-spacer may be used. 10. Conjugate the anti-Spot-Tag nanobody to the binder tail. a. Various conjugation methods may be used, including the SoluLINK Protein-Oligonucleotide Conjugation Kit (Vector Laboratories) and sortase-mediated conjugation (Antos et al., 2017). In Hong et al. (2022), we describe a conjugation method using the SoluLINK kit according to manufacturer instructions. Pause point: Store stably at −20°C.
Various DNA components of the BCS platform may unintentionally affect binder-target interactions. We found that blocking several ssDNA regions using complementary DNA reduced these effects. The blocked ssDDNA components included the POC oligonucleotide tail and the Illumina P7 and P5 adapters. These blocking oligonucleotides are used in the Chip Blocking Solution. 11. Design and order blocking sequences complementary to the POC tail, Illumina P5 sequence, and Illumina P7 sequence. 12. Resuspend each complementary sequence at 100 μM in NF H2O. Pause point: Store stably at −20°C.
The universal bridge anneals to the 3′ end of the foundation and the 5′ end of the binder to create a double-stranded region that facilitates ligation between the two. The universal bridge contains an 8- or 12-nt universal base region intended to anneal to the variable binder barcode, as well as the EcoRI restriction site to allow the binding region of the binder to be cleaved and eventually washed away for subsequent rounds of binding or NGS adaptor ligation for sequencing (Figure 4B). 13. Design the universal bridge. a. Design complementarity to the foundation and binder tail as shown in Figure 3B. Note: The signal to noise ratio observed after DNA sequencing was comparable between runs using the universal 5-nitroindole bridge and runs using bridge sequences with complementarity to specific binder barcodes. The universal sequence was chosen for ease of use.Note: We observed that ligation efficiency improved with universal bridges of longer length. Our bridge length was optimized for ligation efficiency balanced against purity and reagent cost. 14. Order the universal bridge and resuspend at 100 μM in NF H2O. Pause point: Store stably at −20°C.
Timing: varies The purpose of this section is to describe considerations for NGS chip selection and sequencer reprogramming. When choosing an NGS chip, the user should consider the number of desired sites for target-foundation pairs and the number of successful reads balanced against the cost of each sequencing run. For experiments with read counts under one million, we recommend a V2 nano MiSeq chip. For experiments requiring greater read counts, we recommend V3 chips. 15. Remove the initial washing and library loading steps from the cartridge to the sequencing chip on the MiSeq instrument. CRITICAL: Modified run instructions on a commercial sequencer may no longer be appropriate for standard DNA sequencing experiments. Users should be familiar with storing a copy of the original code file and reverting the sequencer to run its original code if they wish to continue utilizing the sequencer for standard procedures.
Timing: 90–120 min In this section, the cololinkers, foundation, and target are hybridized in solution to form a U-shaped complex (Figure 3A). The creation of this complex is required to ensure that targets and their respective foundations colocalize on the sequencing chip. In this protocol, the total concentration of foundations is 120 pM. Concentrations should be optimized according to the number of targets included (see troubleshooting section problem 1 for more details). 1. Thaw cololinker stock solution, foundation, and Spot-tag peptide oligonucleotide conjugate on ice. 2. Prepare buffers detailed in the materials and equipment section and the PhiX solution according to manufacturer instructions. Note: Buffers may be prepared in bulk for subsequent runs, except for the chip blocking solution. 3. In a 96-well plate, combine the Hybridization Buffer, cololinker stock solution, and foundations in the order listed below. Add the target last, immediately prior to hybridization. a. For positive controls, we used DNA targets (DNA Target 4.O1 and 6.O1), where binding occurs through complementary regions on the target and binder. For null control targets, we used a version of the peptide oligonucleotide tail without 5′ phosphorylation (CLR.Null.Block) and another without the peptide attached (5′Phos.01). 4. Thermocycle under the following conditions:
Timing: 45–60 min In this section, the target-foundation complex is ligated onto the chip, and the chip is blocked to decrease non-specific binding to the chip surface and ligated sequences (Figure 2). 5. Create the Foundation Mix. a. Mix gently by pipetting up and down. 6. Dilute each target mixture to 0.5 nM. 7. Carefully remove the chip from its original cylindrical container. Blot dry to remove salts left from the chip storage buffer. Note: Retain the original storage buffer in the cylindrical container and store at 4°C until sequencing has been performed. This buffer can be used to temporarily store the chip if the user needs to pause the experiment. See Pause Point after step 43. 8. Wash the sequencing chip twice. a. To wash, inject 100 μL of the Hybridization Buffer into either port on the sequencing chip (Figure 5). Figure 5 NGS chip manual fluid manipulation Fluid is pipetted into the chip through either of the ports, and exits through the other port. The last 5 μL of fluid remains inside the channel. 9. Wash the sequencing chip with 30 μL of Foundation Mix twice. Note: Blocking components should be added in greater than or equal to 2 times in excess of the component the user intends to block.Note: The BSA in blocking solution is intended to block off-target binding to the glass surface of the sequencing chip. 10. Incubate the sequencing chip at 28°C for 15 min on a hotplate. 11. Wash the sequencing chip once with 100 μL of 100% formamide to remove unligated elements. CRITICAL: All steps involving formamide should be performed in an appropriate certified chemical fume hood. 12. Heat the sequencing chip at 40°C for 90 s on a hotplate. 13. Wash the sequencing chip with 500 μL of Blocking Buffer. 14. Wash the sequencing chip with 30 μL of Chip Blocking solution twice. 15. Incubate the sequencing chip at 37°C for 15 min on a hot plate.
Timing: 75–90 min In this section, binder tails are hybridized to the universal bridge. The universal bridge is required to form a double-stranded nick between the binder tail and the foundation so the binder barcode can be ligated to the foundation. Positive control binders DNA Binder 4.2 and DNA Binder 6 bind to DNA Target DNA Target 4.O1 and 6.O1, respectively. The negative control DNA Binder 9 contains a binding region consisting of a scrambled DNA sequence that should bind to none of the targets present, serving as a control for noise. 16. Hybridize the positive and negative control binders to the universal bridge. This may be performed in a single reaction. a. In one Eppendorf tube, combine: 17. Heat the mixture of DNA binders to 95°C for 5 min. 18. Cool to 25°C for 1 h. 19. While waiting for the controls to hybridize to the universal bridge, hybridize the anti-Spot-Tag nanobody to the universal bridge. a. Combine the anti-Spot-Tag nanobody with 5 times excess universal bridge. Note: A 1:1 ratio of binder to universal bridge may also be used here. We used a 1:5 ratio of nanobody to universal bridge because the method used to conjugate the nanobody to the oligonucleotide tail made it possible for multiple oligonucleotide tails to attach to each POC and because excess oligonucleotide tail used during conjugation was not purified away. 20. Heat the anti-Spot-Tag nanobody solution to 37°C for 30 min. 21. Cool the anti-Spot-Tag nanobody solution to 25°C for 30 min. 22. After cooling, combine the solutions to generate the Binder Incubation Solution as below.
Timing: 90 min In this section, binders are incubated with targets on the NGS chip, binder barcodes are ligated to foundations to “record” the binding event, and restriction enzyme digestion is performed to remove binders from the platform. 23. Wash the sequencing chip with 100 μL of Hybridization Buffer for 60 s twice. 24. Wash the sequencing chip with 100 μL of Incubation Buffer for 60 s. 25. Gently mix the prepared binder library by pipetting it up and down. 26. Slowly load 30 μL of the Binder Incubation Solution onto the sequencing chip twice. 27. Incubate the sequencing chip on a hotplate at 25°C for 30 min. 28. Wash the sequencing chip with 100 μL of Incubation Buffer for 90 s three times. 29. Dilute 7 μL of 2× Blunt/TA MM Ligase solution in 63 μL of Hybridization buffer and mix gently. 30. Load 30 μL of the diluted ligase solution onto the sequencing chip twice. 31. Incubate the chip for 5 min in a hotplate at 28°C. 32. Terminate the ligation reaction by washing the sequencing chip with 100 μL of 1× CutSmart solution for 60 s three times. 33. Prepare the restriction enzyme mix by combining reagents in the order below. Reagent Stock concentration Amount 77 μL of NF H2O N/A 77 μL 10× CutSmart solution N/A 10 μL Restriction bridge (5′-CTGCGCCTATACGAATTCGTTATC-3′) 10 μM 3 μL EcoRI 20 units/μL 10 μL Total N/A 100 μL a. Gently mix after combining. 34. Load 30 μL of the restriction enzyme mix onto the sequencing chip twice. 35. Incubate the sequencing chip at 40°C on a hotplate for 30 min. Note: Begin preparing reagents required for DNA sequencing of binder barcodes (see step 39). 36. Terminate the restriction digestion reaction by loading the sequencing chip with 100 μL of 100% formamide. CRITICAL: All steps involving formamide should be performed in a certified chemical fume hood. 37. Incubate the sequencing chip at 40°C on a hotplate for 90 s. 38. Wash the sequencing chip with 500 μL of Hybridization Buffer. Optional: Additional cycles of binding and barcode capture (steps 16–38) may be repeated for a multi-cycle experiment.
Timing: 4–5 h In this section, a universal NGS adapter is ligated to the binder barcode with the assistance of a universal NGS adapter bridge (Figure 2, steps 5–7). The NGS adapter facilitates direct amplification and sequencing of the foundation with binder barcode. 39. Thaw the sequencing cartridge at 25°C. 40. Prepare the NGS ligation mix by combining the reagents below. 41. Load 30 μL of the NGS ligation mix onto the sequencing chip twice. 42. Incubate the sequencing chip on a hotplate at 40°C for 2 min and 24 s. 43. Wash the sequencing chip twice with 500 μL of NF H2O. a. Wait 90 s in between washes. Pause point: If users are unable to start the sequencing run after finishing NGS ligation, wash the chip with the buffer that the sequencing chip was originally stored in, return the chip to the cylindrical container, and keep at 4°C for no more than 24 h. When ready to sequence, wash the chip twice with 500 μL of NF H2O prior to continuing on to step 44. 44. Dilute 20 μL of 20 pM PhiX in 580 μL of HT1 Buffer (supplied with MiSeq cartridges). 45. Load the PhiX solution into the sample well of the sequencing cartridge. Note: This protocol conducts a 45–600 cycle read using a V2 nano chip. If the pre-run check produces a flow error, exchange the plastic hinged piece that contains the gasket on the flow cell with the same piece from an old flow cell after thorough rinsing with 70% Ethanol and NF H2O. 46. Load the sequencing chip, the cartridge, and the running buffer into the MiSeq according to the onscreen MiSeq sequencing run instructions. 47. Start the sequencing run.
While we have seen consistent ranges of sequencing counts in repeated experiments, sequencing counts may vary depending on a multitude of factors, including the binding properties of binder-target pairs, sequencer type, chip capacity, and similarities between binders or targets in the same run. The expected counts of a positive binder-target pair may range from the low thousands to high millions depending on the sequencer used and total chip spot capacity, as well as the sequencing data analysis parameters. In Table 2, DNA Binder 6 and DNA Target 6 yielded reads in the 28-thousands, while the Spot-tag binding pairs yielded sequencing counts in the lower range (thousands). The ideal heat map in Figure 6 illustrates the expected distribution of counts for target-binder pairs of varying affinities. For multi-cycle experiments, the match rate is expected to undergo exponential decay with an increasing number of cycles.
The code used to analyze the sequencing data is on GitHub. The dataset is on Mendeley (link).
Utility of this method toward certain applications may be limited by the binding kinetics, the physical constraints of the NGS chip, and interference of BCS DNA components with binding and DNA folding. The presence of multiple DNA components and targets in close proximity necessitates use of binders with specificity high enough to overcome potential off-target effects and affinity high enough to overcome signal from noise. In practice, this means that certain binders (ex. low Kd aptamers) are challenging to use in this assay. The Kd of the Spot-tag-nanobody binding pair demonstrated here is approximately 6 nM (Virant et al., 2018). When considering potential peptide binders to proteins, antibodies and antibody-based binding domains such as scFvs and nanobodies (Götzke et al., 2019) have been developed to bind peptides and small molecules with high affinity (Tabares-da Rosa et al., 2011; Finlay and Almagro, 2012; Cobaugh et al., 2008). ClpS, which preferentially recognizes the N-terminal over internal residues within a peptide, represents another promising scaffold of a potential N-terminal amino acid binding reagent (Tullman et al., 2019). While the chip-based platform allows for excellent spatial control of target-binder interaction, the number of target-binder pairs that can be screened at once is constrained by the need to maintain a certain distance between molecules on the chip. To maintain optimal clustering during sequencing, the concentration in solution cannot exceed 120 pM. For applications where multiple cycles of BCS must be performed, one should consider that the binder barcode must be unique for each binder for each cycle. In other words, one binder may require multiple distinctive barcodes to differentiate between binding in different cycles. Thus, an application of BCS requiring a large binder library may be limited by the need to experimentally validate a large number of unique binder barcodes for interference with BCS components and between binders. A final limitation is that targets used in BCS may not be recovered or removed from the chip after the assay is run.
Sequencing run failure. There are multiple reasons for why a sequencing run may fail. We found the leading cause of failure to be a high loading concentration of targets (step 3 of ‘before you begin’) leading to overclustering during amplification on the sequencer. Other things to consider include the presence of large air bubbles during BCS incubation steps (steps 10, 15, 27, 31, 35, 37, and 42) or use of a degraded PhiX solution (step 2 of ‘before you begin’).
Both overclustering and under clustering can compromise data quality and output. Ideal cluster densities will vary by machine and should be confirmed in the Illumina user manual (Optimizing cluster density on Illumina sequencing systems). Our preferred method for assessing under- or overclustering is the Sequencing Analysis Viewer, which includes multiple other modalities for assessing cluster density. Our constructs possess many shared sequences which leads to low nucleotide diversity that predispose the sequencing run to failure. Therefore, inclusion of a high-diversity library such as PhiX is especially important for sequencing run success. Regular preparations of fresh PhiX solution are necessary to avoid degradation from repeated freeze-thaw cycles. The chip should be visualized before each incubation to ensure that no air bubbles are visible, as these will interfere with proper functioning of BCS components. Bubbles may be flushed out with an additional injection of a compatible buffer solution. Automation using a liquid handler may be used to help reduce air bubbles. If successful colocalization of the foundation and target is questioned, one troubleshooting strategy is to couple BCS components with fluorescent probes and visualize their activity on the chip. We use single-molecule imaging to validate the physical colocalization of targets and their respective foundations in our previous paper (Hong et al., 2022). When interrogating more complex sources of failure or optimizing components at specific steps, couple the components with fluorescent probes and use an appropriate imaging modality to visualize each component.
Off-target binding. Sequencing data can reveal unexpected binding interactions such as a single binder candidate binding to a significant percentage of targets or to negative controls (relevant to step 27).
Components of the platform may have non-specific or off-target binding properties. Exposed adaptors or empty foundation bases are examples of off-target binding sites. The barcode designs of the targets and binders should be checked for unintended complementarity to self or other. As a first-line solution, sequences may be redesigned accordingly. Other areas to optimize through unit testing include incubation times and enzyme concentrations, especially for steps involving Blunt/TA MM Ligase. The Chip Blocking Solution contains sequences that block the single-stranded P5 and P7 adapters on the Illumina chip (see the section "designing chip blocking components"). The concentration of blocking components may be optimized through unit testing, and further blocking components may be added (such as blocking the POC ssDNA region or blocking part of the foundation). In cases involving DNA components, hybridization between individual components can be assessed using a Bioanalyzer RNA assay, which can differentiate between double and single stranded DNA (Unpublished data). In other cases, a Kd measurement assay between binder and target (Plach and Schubert, 2019) may be considered to quantify true off-target binding.
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Annalisa Pawlosky [email protected].
This study did not generate new unique reagents. Example buffers and sequences of oligos, peptides, and oligo-peptides used in this protocol are listed in the key resources table, though these should be interchanged with materials compatible with the user’s subject(s) of study. |
|
PMC9647731 | Jeane do Nascimento Moraes,Aleff Ferreira Francisco,Leandro Moreira Dill,Rafaela Souza Diniz,Claudia Siqueira de Oliveira,Tainara Maiane Rodrigues da Silva,Cleópatra Alves da Silva Caldeira,Edailson de Alcântara Corrêa,Antônio Coutinho-Neto,Fernando Berton Zanchi,Marcos Roberto de Mattos Fontes,Andreimar Martins Soares,Leonardo de Azevedo Calderon | New multienzymatic complex formed between human cathepsin D and snake venom phospholipase A2 | 04-11-2022 | Cathepsin D,Phospholipases A2,Snake venom,Enzyme complex | Abstract Background Cathepsin D (CatD) is a lysosomal proteolytic enzyme expressed in almost all tissues and organs. This protease is a multifunctional enzyme responsible for essential biological processes such as cell cycle regulation, differentiation, migration, tissue remodeling, neuronal growth, ovulation, and apoptosis. The overexpression and hypersecretion of CatD have been correlated with cancer aggressiveness and tumor progression, stimulating cancer cell proliferation, fibroblast growth, and angiogenesis. In addition, some studies report its participation in neurodegenerative diseases and inflammatory processes. In this regard, the search for new inhibitors from natural products could be an alternative against the harmful effects of this enzyme. Methods An investigation was carried out to analyze CatD interaction with snake venom toxins in an attempt to find inhibitory molecules. Interestingly, human CatD shows the ability to bind strongly to snake venom phospholipases A2 (svPLA2), forming a stable muti-enzymatic complex that maintains the catalytic activity of both CatD and PLA2. In addition, this complex remains active even under exposure to the specific inhibitor pepstatin A. Furthermore, the complex formation between CatD and svPLA2 was evidenced by surface plasmon resonance (SPR), two-dimensional electrophoresis, enzymatic assays, and extensive molecular docking and dynamics techniques. Conclusion The present study suggests the versatility of human CatD and svPLA2, showing that these enzymes can form a fully functional new enzymatic complex. | New multienzymatic complex formed between human cathepsin D and snake venom phospholipase A2
Cathepsin D (CatD) is a lysosomal proteolytic enzyme expressed in almost all tissues and organs. This protease is a multifunctional enzyme responsible for essential biological processes such as cell cycle regulation, differentiation, migration, tissue remodeling, neuronal growth, ovulation, and apoptosis. The overexpression and hypersecretion of CatD have been correlated with cancer aggressiveness and tumor progression, stimulating cancer cell proliferation, fibroblast growth, and angiogenesis. In addition, some studies report its participation in neurodegenerative diseases and inflammatory processes. In this regard, the search for new inhibitors from natural products could be an alternative against the harmful effects of this enzyme.
An investigation was carried out to analyze CatD interaction with snake venom toxins in an attempt to find inhibitory molecules. Interestingly, human CatD shows the ability to bind strongly to snake venom phospholipases A2 (svPLA2), forming a stable muti-enzymatic complex that maintains the catalytic activity of both CatD and PLA2. In addition, this complex remains active even under exposure to the specific inhibitor pepstatin A. Furthermore, the complex formation between CatD and svPLA2 was evidenced by surface plasmon resonance (SPR), two-dimensional electrophoresis, enzymatic assays, and extensive molecular docking and dynamics techniques.
The present study suggests the versatility of human CatD and svPLA2, showing that these enzymes can form a fully functional new enzymatic complex.
Cathepsins compose a family of lysosomal proteases mainly found in acidic endo/lysosomal compartments and are implicated in a broad spectrum of physiologic processes, such as intracellular protein degradation, energy metabolism, hormonal regulation, bone resorption, and immune responses [1]. These proteins belong to three protease families, characterized based on differences in the following amino acids at their active site: aspartic proteases (D and E), serine proteases (A and G), or cysteine proteases (B, C, H, F, K, L, O, S, V, X, and W) [1-4]. Furthermore, cathepsins are essential to maintaining cell homeostasis [5]. The inactivation, loss of function, and overexpression of these proteases can result in inappropriate degradation and abnormal accumulation of lysosomal waste [1, 6]. In addition, extracellular oversecretion of cathepsins is associated with uncontrolled cell proliferation, invasion, and differentiation, which in turn may bring about the onset of fatal pathologies, including atherosclerosis, cancer, and tissue fibrosis [6-12]. Due to its physio-pathological functions, cathepsin D (CatD) is one of the most studied lysosomal proteases [13-15]. CatD is an aspartic endopeptidase with two conserved Asp residues in its active site; these residues tend to deprotonate, indicating that the pH-optimum of activity resides at pH values below 5 [16]. In addition, CatD has three distinct regions that are characteristic of aspartic proteases, an N-terminal domain (residues 1-188), a C-terminal domain (residues 189-346), and an interdomain, antiparallel/3-sheet formed by the N-terminus (residues 1-7), the C-terminus (residues 330-346), as well as the linker residues between domains (160-200) [17]. Considered a multifunctional enzyme due to its involvement in various biological processes, CatD operates in both cytosolic and extracellular environments [13, 18-22]. Studies have shown that CatD is involved in the activation of precursors of biologically active proteins in pre-lysosomal compartments of specialized cells [11, 20, 23]. This enzyme is indispensable for cellular functions such as cell migration, differentiation, growth, cycle progression, tissue remodeling, and neovascularization activation [6, 11, 12, 19, 22, 24-27]. Additionally, CatD is involved in initiating the apoptotic cascade [28, 29] in lysosomal cell death pathways [22, 25]. CatD is directly related to the pathogenesis and progression of degenerative diseases [6, 30], such as lymphoid cell degeneration [31], Parkinson’s [32] and Alzheimer’s disease [33], atherosclerosis [34], and different types of cancer [35, 36]. For instance, some cell types under pathological conditions overexpress and secrete CatD to the extracellular environment via lysosomal release [20]; this makes CatD an important tumor marker in breast, bladder, and mouth cancers, among others [35, 36]. Furthermore, due to the participation of cathepsins in a broad spectrum of diseases, these proteases are promising therapeutic targets for small molecules and peptide drugs [33, 36]. In order to investigate human CatD inhibitors for the design and development of tools and agents of scientific and therapeutic interest, snake venoms belonging to the genera Bothrops, Crotalus, and Lachesis have been used as natural sources of biologically active molecules able to act selectively and specifically on different cellular targets [37, 38]. Of all the bioactive molecules present in snake venoms, phospholipases A2 (svPLA2) are among the most frequently encountered and studied [39, 40]; these proteins have established physical-chemical properties and a variety of pharmacologic and toxic effects in snakebite envenomation, such as myonecrosis, anticoagulation, platelet aggregation inhibition, neurotoxicity, cardiotoxicity, hypotension and edema formation [41-45]. Interestingly, human CatD shows the ability to bind strongly to svPLA2s, forming a stable and functional complex that is able to remain active even at pH values higher than 5 and is also unaffected by the inhibitor pepstatin A. These results, presented and discussed below, demonstrate the multifunctionality and versatility of CatD, warranting many new possibilities for the understanding of cathepsin functions in cytosolic and extracellular environments during physiologic and pathologic processes. Therefore, the present study aims to demonstrate and characterize an enzymatic complex formed by human CatD and a snake venom phospholipase A2.
Cathepsin D (cod. C8696) was obtained from Sigma-Aldrich Ltda and prepared according to the manufacturer’s recommendations.
All snake venoms used in this study were acquired from the Venom Bank at CEBio/Fiocruz Rondônia/UNIR (Centro de Estudos de Biomoléculas Aplicadas a Saúde), Porto Velho, RO, under local government authorization license number: IBAMA nº 27131-3 and CGEN/CNPq 010627/2011-1.
The Bothropstoxin-I (BthTX-I) and Bothropstoxin-II (BthTX-II) from Bothrops jararacussu were obtained from the Venom Bank at CEBio (Centro de Estudos de Biomoléculas Aplicadas a Saúde/Fiocruz Rondônia/UNIR), located in Porto Velho, RO. PLA2 LmtTX from Lachesis muta provided by Diniz-Sousa et al. [46], PLA2 BnuTX-I from Bothrops urutu provided by Corrêa et al. [47], PLA2 Braziliase-I and Braziliase-II from Bothrops brazili provided by Kayano et al. [48] and Sobrinho et al. [49].
B. jararaca venom was solubilized in 50mM ammonium bicarbonate buffer (AMBIC), pH 8.0 and applied to an anion exchange column (CM-Sepharose 10 x 30 cm). The fractions were eluted in a linear gradient of 500 mM AMBIC, pH 8.0 under a flow of 1 mL/min. Absorbances were measured at 215 and 280 nm. The fractions were subjected to salt removal in a 15mL filter (AMICON ULTRA-15) with a 50 kDa cutoff.
Surface plasmon resonance (SPR) molecular interaction assays were performed in a Biacore T200 system (GE Healthcare). Cathepsin D immobilization was done using a CM5 S-type sensor chip via amine coupling. The contact time of each cycle was set at 60 seconds, with a flow rate of 30 µL/min, followed by 60 seconds of dissociation time. For the regeneration stage at the end of each cycle, a 0.5% TFA solution was used with 30 seconds of contact time at a flow of 30 µL/min. All experiments were performed at 25 ºC, and binding assays were conducted in phosphate-saline buffer (PBS), pH 7.4 and analytes at a concentration of 100 µg/mL.
The protein concentrations present in venom samples were determined using Bradford’s method [50]. For spectrophotometric measurements, the sample was aliquoted in a 1 mL disposable plastic cuvette along with 1:10 (v/v) Bradford reagent, which was incubated for 15 minutes. Absorbance was monitored at 595 nm using a Biomate 3 spectrophotometer. The calibration curve was performed using bovine albumin (Sigma).
The relative mass of proteins was determined by SDS-PAGE using discontinuous gels, with a stacking gel (4% acrylamide in 0.5 M Tris-HCl buffer, pH 6.8) (Sigma Aldrich, USA) and a resolving gel (12.5% acrylamide in 1.5 M Tris-HCl buffer, pH 8.8). The experimental buffer solution used to fill the wells was 0.06 M Tris-Base, 0.5 M Glycine, and 10% SDS (Sigma Aldrich, USA). The samples with 1M DTT were preheated to 95 °C for 5 min and applied to the stacking gel wells along with the Molecular Weight standard (7 to 175 kDa - BioLabs P7709S, USA). In the electrophoretic run, a current of 15 mA per gel and free voltage was fixed for 1 hour and 40 minutes. After this, the gel was washed for 15 minutes with a fixing solution (ethyl alcohol 50% and acetic acid 12%) and then stained with Coomassie G-250 blue solution (Sigma Aldrich, USA) for 10-30 minutes. After this period, the gel was bleached in a bleaching solution (20% ethyl alcohol and 3% acetic acid). The gels’ images were scanned using Image Scanner III (GE Lifescience Health Care). The 2D electrophoresis consisted of two steps: isoelectric focusing and 1D SDS-PAGE. For the first dimension, the sample was prepared in a rehydration solution (8 M urea, 2% CHAPS, 0.5/2% IPG buffer, 0.002% bromophenol blue, and 1 M DTT); this same solution was then incubated with a 7-cm strip (pH 3-10, linear) for 12-20 h. After rehydration, the strip was applied to an Ettan IPGphor 3 (GE Healthcare) isoelectric focusing system and later stored at − 80 °C. For the second dimension, the strip was washed with DTT and iodoacetamide diluted in 5 mL of equilibration buffer solution (6 M urea, 2% SDS, 30% glycerol, 50 mM Tris- HCl, pH 7.4, 0.002% bromophenol blue). Then, the strip was applied to a 15% polyacrylamide gel. The gel was stained with Coomassie Blue G-250 and scanned in a GE Image Scanner III apparatus.
Proteolytic activity was evaluated according to the method described by Rodrigues and coworkers [51], with adaptations, using casein as a substrate. Samples (12 µg/mL) were incubated with 250 µL of 2% casein in 0.1 M sodium citrate (pH 3, 4, 5, 6, 7) for 30 minutes at 37 °C, interrupted by the addition of 250 μL of 20% trichloroacetic acid (TCA). Similarly, sample contamination by metalloprotease at different pHs was analyzed by adding 10uL of ethylenediaminetetraacetic acid (EDTA). The solution was left to rest for 30 minutes at room temperature and then centrifuged at 10,000 x g for 15 minutes at 25 °C. The proteolytic activity was estimated based on the absorbance of the supernatant at 280 nm, with trypsin as a positive control.
The proteolytic activity was evaluated according to the method described by Rodrigues et al. [51], with adaptations, using casein as a substrate. Samples (6.3 µg/mL) were incubated with 250 µL of 2% casein in 0.1 M sodium citrate (pH 3, 4, 5, 6, 7) for 30 minutes at 37 °C and then interrupted by the addition of 250 µL of 20% trichloroacetic acid (TCA). The solution was left to stand for 30 minutes at room temperature and then centrifuged at 10,000 x g for 15 minutes at 25 °C. Proteolytic activity was estimated based on the absorbance of the supernatant at 280 nm. The proteolytic activity monitored in SDS-PAGE electrophoresis was performed according to the protocol described above. Inhibition was carried out by means of exposure to high temperatures (90 ºC).
This procedure was carried out as described by Petrovic and coworkers [52]. 5 mg of the substrate 4-nitro-3-octanoyloxy-benzoic acid (4N3OBA) (Enzo Lifescience, USA) was diluted in 5.4 mL of acetonitrile. 0.2 mL aliquots were dried and stored at -20 °C. Each tube containing 4N3OBA was diluted in 2 mL of sample buffer (0.01 M Tris-HCl at pH 8.0, 0.01 M CaCl2, and 0.1 M NaCl) (Sigma Aldrich, USA) and maintained on ice. In order to determine the phospholipasic activity, a total of 190 μL of 4N3OBA reagent combined with 10 μL of sample (cathepsin + BthTX-II, and inhibitor) was applied in a 1:1 ratio, pre-diluted in water and incubated at 37 °C; subsequently, the substrate was added to the samples and immediately incubated at 37 ºC. The absorbance was measured at 425 nm for 30 minutes (interval of 1 min). Phospholipase activity was considered directly proportional to the increase in absorbance values and expressed as the mean ± standard deviation; the results were submitted to analysis of variance (ANOVA) followed by Tukey’s post-test for p < 0.05.
All PLA2s used in the in vitro assays were assessed through molecular docking against cathepsin D (CatD). The available structures of CatD (4OD9), BthTX-I (3CXI), BthTX-II (2OQD), and Crotoxin B (3R0l) were extracted from the RCSB Protein Data Bank. The structures of Braziliase II (UniProtKB: P0DUN4) and LmutTX (UniProtKB: P0DUN7) were generated by means of comparative modeling using the Rosetta web server [53]. The structural conformation guiding the interaction and complexation of CatD and PLA2 were predicted through a consensus of 5 protein/protein docking tools (pyDock, ZDOCK, HDOCK, ClusPro, and GRAMX). The CatD + BthTX-II complex was subjected to molecular dynamics, with five replicas of 100 ns using GROMACS 2020.2 employing the CHARMM36-mar2019 force field [54]. All simulations were carried out with a neutral net charge box of 4 Å radius from the farthest atom, solvated with TIP3P water, and equilibrated with 100 mM NaCl. The system was minimized with the steeper descent minimization until it reaches the power levels below 100 kJ/mol/nm. Then, the box was equilibrated under an isochoric-isothermal (NVT) ensemble for 1 ns, generating speeds according to the distribution of Maxwell-Boltzmann at 310.15 K using the V-Rescale thermostat [55] followed by an isothermal-isobaric (NPT) ensemble using the Berendsen barostat at 1 bar [56]. Subsequently, five replicas of unrestrained 100 ns simulations were executed using the Nose-Hoover Thermostat [57] and Parrinello-Rahman barostat [58]. Nonbonded interactions were calculated within a radius of 12 Å using a switching function between 10 and 12 Å. Afterwards, the trajectories were analyzed, and radius of gyration and backbone RMSD measurements were extracted from the main interacting parties for stability assessment. Further, the trajectories were subjected to clusterization using the gromos method [59] with an RMSD distribution of 2 Å. All images and interaction maps were created using UCSF Chimera 1.13.1 [60].
Snake venoms were screened as to their potential interactions with human CatD, aiming to generate an extensive analysis of binding responses featuring the unique molecular content found in each venom. In this fashion, the bioactive compounds with the most affinity towards CatD could be inferred based on venom composition. For this purpose, thirteen venoms from different species were used (Fig. 1 and Table 1). Among these species, Bothrops brazili, B. jararaca, B. jararacussu, and B. leucurus stood out as promising due to their association and dissociation profiles and the maximum number of responses reached. Bothrops jararaca venom is one of the most well-characterized and studied venoms and showed a significant binding response (1,625 RU mg/mL) with CatD; for those reasons, it was selected for further analysis. In order to identify the venom components responsible for the majority of interaction signals, B. jararaca crude venom was fractionated through cation exchange chromatography (Fig. 2A). The chromatography resulted in 12 fractions that were later submitted to SPR assays against CatD. The subsequent assays revealed that only fractions 10, 11, and 12 presented significant interactions with CatD, showing responses from 25, 12, and 10 RUs at a concentration of 50 mM (Fig. 2D). Next, the protein profile of each fraction was determined by SDS-PAGE, resulting in clear monophoretic bands around 13 kDa for all three fractions (Fig. 2C), compatible with svPLA2 mass and bands between 30 to 40 kDa, suggesting snake venom metalloproteases (SVMPs) in the fractions 10 and 11 (Fig. 2C). When these fractions were tested for their phospholipase activity, fractions 3, 8, 10, 11, and 12 showed relevant activity against the substrate 4N3OBA (results not shown), confirming the presence of phospholipases in the fractions of interest. These data strongly suggested that human CatD has the ability to interact with svPLA2s. In order to investigate this tendency and evaluate the specific affinity between both proteins, six svPLA2s from the genera Bothrops and Lachesis were submitted to SPR assays at concentrations of 15 and 50 mM (Table 2). The binding analysis via SPR spectroscopy revealed that the toxins tested (except BthTX-I and Braziliase I) displayed tight binding to immobilized CatD (Fig. 3). For instance, BthTX-II (an enzymatic Asp-49-PLA2) [61] presented interaction showing dose-dependent SPR responses ranging from 420 to 1,420 at concentrations of 15 and 50 mM, respectively (Fig. 3C). Different from Braziliase-I, Braziliase-II showed a dose-dependent sensorgram of 245 RUs (15 mM) and 837 RUs (50 mM) with a prolonged dissociation phase suggesting a possible low dissociation rate constant (Kd) (Fig. 3A), which could be investigated through further analysis. Both BnuTX-I from B. urutu and LmutTX from L. muta also interacted with immobilized CatD (Fig. 3B), showing sensorgrams with different intensities of 552 and 2,180 RU at 50 mM [46, 47]. In any case, both showed a similar shape in their association and dissociation curves. Despite the high level of homology among svPLA2s, the binding analysis between CatD and these toxins exhibited interactions with different intensity profiles. Nevertheless, the binding profile of CatD towards svPLA2 displayed high similarity, suggesting a common recognition site. It is worth pointing out that overall, svPLA2s present a characteristic and consistent tridimensional structure, which could be the driving factor behind the ability of CatD to interact with the svPLA2s tested in this study [61, 62].
Initially, the apparent molecular mass and isoelectric point (pI) of the CatD + BthTX-II complex, as well as that of both enzymes separately, BthTX-II and CatD, were verified through two-dimensional electrophoresis (Fig. 4), determining a molecular mass of approximately 60 kDa and pI of 5.79 for the CatD + BthTX-II complex. Next, the proteolytic activity of CatD and of its complex with svPLA2 (BthTX-II) were evaluated using casein as a substrate at pH values of 3, 4, 5, 6, and 7, and Pepstatin A as a specific inhibitor. The optimal enzymatic activity of CatD was observed at pH 5, which is in agreement with previous studies [63]. On the other hand, the CatD + BthTX-II complex proved to be functional at different pH values reaching maximum activity at pH 6 (Fig. 5), revealing that the binding between these two proteins changes CatD’s functionalities, increasing its pH-dependent activity to higher values. Additionally, the CatD + BthTX-II complex is resistant to the inhibitor Pepstatin A at pH 6, suggesting the possibility of changes in enzyme specificity (Fig. 6A). Similar outcomes were observed in the SDS-PAGE assay, revealing that the bands formed after casein hydrolysis by CatD and CatD + BthTX-II are slightly different (Fig. 6B), suggesting potential differences in cleavage sites and further confirming the in vitro enzymatic activity. Furthermore, to rule out any residual contamination from the BthTX-II sample due to venom proteases, this sample was also submitted to the same conditions, and showed no proteolytic activity (results not shown). Regarding the effects of the interaction of the CatD + BthTX-II complex on BthTX-II’s catalytic function, the phospholipase activity assay (Fig. 7) shows that the complex’s formation does not interfere with nor hinder BthTX-II’s capability to cleave the artificial substrate 4N30BA. Interestingly, the presence of Pepstatin A slightly diminishes the catalytic output of the CatD + BthTX-II complex.
All svPLA2s showing interaction with CatD in the SPR assay and enzymatic assays were selected for further in silico investigation, seeking details about the mechanism coordinating these interactions at the atomic level and the existence of common recognition sites for svPLA2s on CatD’s surface. Thus, five molecular docking methodologies were applied, effectively employing a consensus approach, which generated sets of docking conformations (Fig. 8) for each of the svPLAs2 (BthTX-II, Braziliase-II and LmutTX). Additionally, the CatD + BthTX-II complex (Fig. 9) was subjected to a more intensive inspection due to its enzymatic activity. Molecular dynamics (MD) was used to evaluate the structural stability of this macromolecular assembly. Five independent replicas were simulated for 100 ns each. The processing and analysis of the generated trajectories included an assessment of the CatD + BthTX-II complex’s behavior in solution considering the radius of gyration (Fig. 10A) and RMSD (Fig. 10B) variations during the simulations. There were few noticeable fluctuations in the complex’s backbone and its compactness. Nevertheless, the assembly formed between these two proteins remained stable through all five replicas. The interaction between CatD and BthTX-II was evaluated, using as reference the central structures from the three most populated clusters generated in the clusterization performed with the sum of all five trajectories, exhibiting in that way an approximation of the most predominant conformation assumed by the CatD + BthTX-II complex during 500 ns of simulation (Fig. 10C). The absence of any remarkable shift in the complex’s shape suggests an overall stable and cohesive interaction.
In order to proceed with the characterization of the CatD + BthTX-II complex, different methodologies were used, such as Surface Plasmon Resonance (SPR), a detection method capable of performing real-time, label-free, and high-sensitivity monitoring of molecular interactions [64], and molecular docking, a key tool in structural molecular biology and computer-aided drug design, useful to predict structural data about a potential protein-protein interaction using known three-dimensional structures [65]. SPR assays carried out with immobilized human CatD showed different levels of interaction with components of all snake venoms tested, ranging from 34.1 RU mg/mL for C. d. cascavella to 2,258.3 RU mg/mL for B. jararacussu (Table 1). The interaction of venom components with human cathepsin D, especially those from bothropic venoms, strongly suggests that this could be an important and relevant new biological mechanism involving the participation of CatD and svPLA2 in snake envenomation and other physiopathological processes with the participation of homologous proteins. The use of B. jararaca venom cation exchange chromatographic fractions for further SPR assays (Fig. 2) showed that immobilized CatD interacted only with the last fractions (10, 11, and 12), which corresponds to well-known svPLA2s, according to the monophoretic bands observed in the electrophoresis profile. This data indicated that the svPLA2s presented in the samples tested in SPR binding assays with CatD could be the respective ligands. The SPR analyses carried out with the isolated svPLA2s BthTX-II, Braziliase-II, BnuTX-I, and LmutTX revealed their ability to bind with immobilized human CatD (Fig. 3). Two-dimensional electrophoresis showed that human CatD and BthTX-II form a stable complex of approximately 60 kDa and pI of 5.79. Initially, the apparent molecular mass and isoelectric point (pI) of the CatD + BthTX-II complex, as well as that of both enzymes separately, BthTX-II and CatD, were verified through two-dimensional electrophoresis (Fig. 4), determining a molecular mass for the CatD + BthTX-II complex. Next, the proteolytic activity of CatD and its complex with svPLA2 (BthTX-II) was evaluated using casein as a substrate at pH values from 3 to 7, and Pepstatin A as a specific inhibitor. The pH optimum of the CatD + BthTX-II complex was found to be 6, while isolated CatD shows optimal activity at pH 4 [66]. Furthermore, Pepstatin A doesn’t affect the CatD + BthTX-II complex activity with the substrate (Casein) at different pH values. Interestingly, the change in CatD pH-dependent activity, when compared to that of the CatD + BthTX-II complex, is consistent with previous CatD studies in tumoral cell lines [67], suggesting that in the physiologic scenario, CatD’s interaction with proteins such as svPLA2 might be the factor allowing it to function in different pH ranges. Additionally, the CatD + BthTX-II complex was not inhibited by Pepstatin A, with CatD’s catalytic activity remaining steady, further corroborating the CatD + BthTX-II complex’s increased activity capacity. Moreover, the investigation of the CatD + BthTX-II complex’s impact on BthTX-II’s phospholipase activity suggests that the orientation of BthTX-II when coupled with CatD is ideal and allows BthTX-II to remain fully functional. Computational simulations revealed a clear pattern of interaction between CatD and svPLA2s, in such a way that all svPLA2s tested in this study exhibited affinity by the concave surface formed between the heavy and light chain of CatD. This interaction profile was observed in every docking performed in this study. Furthermore, MD simulations done with the CatD + BthTX-II complex demonstrated that this may be the stable conformation assumed by CatD interacting with svPLA2s in solution. Alone, the CatD + svPLA2 complex’s interface of interaction observed in the simulations performed herein is not able to enlighten the molecular mechanisms behind the boost in CatD’s catalytic activity observed in the enzymatic assays. However, the conformation of the CatD + BthTX-II complex generated in the docking predictions and later validated in the 500 ns of simulations agrees with the phospholipase activity assays. The capability of the CatD + BthTX-II complex to retain svPLA2 makes perfect sense given BthTX-II’s orientation upon attachment to CatD (Fig. 9 and 10C), in such a way that BthTX-II’s hydrophobic channel and active site remain fully exposed to solvent. Taking into account all these data, the in silico exploration of CatD’s complex with svPLA2 provides a clear basis for these two enzymes’ interaction in the physiologic scenario. Nevertheless, it is necessary to carry out more experimental structural studies in order to confirm the modes of interaction between these enzymes. These results also raise new questions in the investigation of pathological and inflammatory symptoms of snake envenomation, in which CatD’s interaction with svPLA2 and the complexes formed could play an important role in the cascade of systemic and local effects present in snakebite accidents. The interaction between CatD and svPLA2 demonstrated herein will possibly have future implications for snakebite therapeutics. However, the most significant results extracted from this study may foreshadow more fundamental physiological issues involving the role of CatD in inflammatory processes, apoptosis and tumor progression. In this regard, the proteolytic process in neurons, in which CatD actively participates, is an essential maintenance step for the clearance of protein aggregates that reach the lysosomes through endocytosis and autophagy [24]. Di Domenico and coworkers proposed that the lack of control in protein repair (proteasome and lysosomal system) is a characteristic of degenerating neurons in Alzheimer’s disease (AD), which highlights CatD’s involvement in these conditions due to its essential role in the management of lysosomal integrity [33]. Thus, the rise in PLA2 (IIA) in the cerebrospinal fluid of patients with AD indicates these enzymes as potential biomarkers in neuroinflammation [68, 69]. Furthermore, human brains affected by AD present a significant increase in PLA2 mRNA in the hippocampus [70]. Interestingly, reports of PLA2s’ involvement in the destabilization of lysosomal membranes have been made in different experimental systems [29, 71, 72]. Overall, many approaches have discussed the involvement of PLA2s in inflammatory processes [73-75]. In addition, PLA2s also act on cell membrane metabolism and the production of arachidonic acid, a known precursor of prostaglandins, leukotrienes, and thromboxanes [76-78]. Johansson and coworkers demonstrated that incubation of PLA2s with rat liver lysosomes resulted in the extravasation of its lysosomal constituents [29]. Additionally, Beaujouin and coworkers demonstrated CatD’s involvement in apoptosis and showed that cancer cells that were pretreated with Pepstatin A, could not halt CatD nor hinder apoptosis, supporting the results described herein in the proteolytic activity assays. Moreover, CatD’s capability to induce cancer cell growth, even when mutated, suggests an alternative mechanism for this enzyme [79].
For the first time, this study describes the formation of a functional muti-enzymatic complex between the human protease cathepsin D and snake venom phospholipases A2. Collectively, the in vitro assays and in silico predictions carried out in this study demonstrated interaction and the formation of a new muti-enzymatic and catalytically active complex between CatD and svPLA2. Additionally, the agreement between the data from previous studies regarding the pathways in which these enzymes are involved and the new data presented herein indicates the possibility of PLA2 and CatD acting in conjunction in the extracellular environment [41]. Nevertheless, in the face of the many possible outcomes of this new enzymatic complex, the conclusions drawn must be taken with caution and, most importantly, warrant more extensive investigation on the subject. |
|
PMC9647767 | Mohammad Al-Tamimi,Shahed Altarawneh,Mariam Alsallaq,Mai Ayoub | Efficient and Simple Paper-Based Assay for Plasma Separation Using Universal Anti-H Agglutinating Antibody | 28-10-2022 | Background: Conventional laboratory tests require plasma separation using centrifugation by skilled personnel in well-equipped lab. Development of a simple, reliable, and cheap point-of-care (POC) test for plasma separation will overcome these limitations. Methods: Plasma separation was achieved in filter paper using the anti-H agglutinating antibody. Hydrophobic channels were created using a solid ink printer. The reproducibility, efficiency, recovery, and applicability of the assay were validated on a large number of blood samples. Results: A simple, fast, cheap, and direct paper-based assay for plasma separation from whole blood using universal anti-H agglutinating antibody was developed without equipment or pretreatment requirements. The purity of plasma separation using anti-H treated paper was confirmed by microscopy and biuret test for plasma albumin detection. Plasma separation was affected by paper structure, antibody concentration, donor gender, and hematocrit. The efficiency of the assay was 72% and the reproducibility was about 90% with minimal interassay and intra-assay variabilities. The assay successfully separated plasma from 116/119 samples, indicating high sensitivity (97.5%). Furthermore, the assay accurately recovers thyroid stimulating hormone from samples compared to standard methods with 107% recovery rate. Conclusions: Paper-based plasma separation using anti-H agglutinating antibodies would have numerous applications in paper-based POC tests and in resource limited areas. | Efficient and Simple Paper-Based Assay for Plasma Separation Using Universal Anti-H Agglutinating Antibody
Background: Conventional laboratory tests require plasma separation using centrifugation by skilled personnel in well-equipped lab. Development of a simple, reliable, and cheap point-of-care (POC) test for plasma separation will overcome these limitations. Methods: Plasma separation was achieved in filter paper using the anti-H agglutinating antibody. Hydrophobic channels were created using a solid ink printer. The reproducibility, efficiency, recovery, and applicability of the assay were validated on a large number of blood samples. Results: A simple, fast, cheap, and direct paper-based assay for plasma separation from whole blood using universal anti-H agglutinating antibody was developed without equipment or pretreatment requirements. The purity of plasma separation using anti-H treated paper was confirmed by microscopy and biuret test for plasma albumin detection. Plasma separation was affected by paper structure, antibody concentration, donor gender, and hematocrit. The efficiency of the assay was 72% and the reproducibility was about 90% with minimal interassay and intra-assay variabilities. The assay successfully separated plasma from 116/119 samples, indicating high sensitivity (97.5%). Furthermore, the assay accurately recovers thyroid stimulating hormone from samples compared to standard methods with 107% recovery rate. Conclusions: Paper-based plasma separation using anti-H agglutinating antibodies would have numerous applications in paper-based POC tests and in resource limited areas.
Point-of-care (POC) testing is the emerging diagnostic procedure performed in clinical diagnostic labs, and by patient’s bedsides. It is also called rapid testing or near-patient testing to describe its fast test results obtained and interpreted by medical and nonmedical professionals. Depending on the test target, it offers the diagnosis, screening, or monitoring of patient’s diseases status. For example, many POC devices are approved and marketed for monitoring diseases such as diabetes mellitus and hypertension. Using this testing approach is cost-effective for both its cheap cost compared to clinical lab testing and saving millions of dollars spent on disease monitoring and treatment leading to reduced morbidity and mortality rates. Furthermore, POC tests’ relatively easy procedure and result reading offer the ability to use it in hospitals, ambulances, specialized private clinics, public health-related campaigns, military centers, and at home by patients in rural areas with limited medical services. Although POC tests ease the detection of many pathological agents and the monitoring of many diseases, they still face many challenges that should be overcome to consider it a reliable and sensitive testing method. Achieving high sensitivity and precision of diagnosis depends on various factors related to presample processing and sample processing to ensure low to nontesting errors. Samples handled in POC testing range from blood, urine, serum, stool, or saliva. Hence, sample chemical composition variability requires proper treatment and separation approaches to target analytes in complex biofluids for qualitative and quantitative purposes. Recently, multiple studies have highlighted the promising use of bioactive paper for disease detection, diagnosis, monitoring food quality, detection of pathogens, and drug testing, especially in developing countries and for POC applications. Paper is widely available, flexible, disposable, and very cheap; it wicks fluid through capillary absorption flow, is biologically compatible and recyclable, and is suitable for colorimetric assays. As a result, over the past years, there has been an increased interest in bioactive paper-based low-cost sensor development and fabrication. Paper-based sensors provide affordable platforms for the simple, accurate, and rapid detection of biomarkers, cells, DNA, microorganisms, chemicals, and drugs. Blood plasma separation is one of the sample treatment steps that must be performed before target detection. Many separation techniques emerged using many approaches and resulted in varying accuracy levels. Complex approaches utilize separation devices such as microscale separation devices. These devices rely on mechanical separation methods (passive separation) such as sedimentation, cross-flow filtration, and cell deviation obstacles. However, some separation devices that use these techniques suffer from a long separation time, resulting in higher coagulation and filter clogging risks. Dynamic force-based devices (active separation) use more complex systems like magnetic or electric forces for separation achievement which results in a complex separation system, low input flow rate within short times, and could risk blood cell integrity. Membrane-based plasma separation is a technique used in both passive and active methods. Asymmetric pore-sized membranes facilitate the separation of large cellular-sized components without negatively affecting their integrity. Combining it with microfluidic channels fasten the separation process and limits any slow separation complications. Nevertheless, the separated plasma purity and volume remain the two challenging issues faced when using this approach. A plasma separation method should have a biomarker high extraction yield to ensure the accurate detection of low concentration targets. Moreover, the method must not change the target analyte concentration or cause blood cell hemolysis. Therefore, finding a simple yet fast and high yield separation method is a must for the sake of POC testing reliability. The H antigen is the precursor of ABO blood group antigens and present in people of all common blood types. The extremely rare “Bombay phenotype” does not express antigen H on red blood cells (RBCs) and can have circulating anti-H antibodies that could mediate the hemolytic transfusion reaction if they received H antigen positive blood. The medical uses of anti-H monoclonal antibodies are usually limited to forward blood grouping of the suspected Bombay group. In this paper, plasma blood separation using simple treated microfluidic filter paper with anti-H agglutinating antibody was studied as a preanalyte targeting requirement for POC testing. The efficiency, purity, reproducibility, and applicability of this approach were investigated.
Hydrophobic wax barriers were made using two methods. The first one was based on manually waxing channels on filter papers using a wood burner. The filter paper that has channels drawn on it was put on top of a wax-soaked paper. Following that, a wood burner device was used to melt the wax onto the filter paper to draw the required channels. The second method was waxing channels automatically using a Xerox 8580 solid ink printer, then melting the wax from the surface of the filter paper into its layers using a laminator device. Different filter papers were chosen for the membrane filtration approach (Whatman Qualitative filter paper grades 5 (2.5 μm) and Whatman Quantitative ashless filter papers grade 42 and 44 (2.5 and 3 μm, respectively)).
Anti-A, anti-B, and anti-H antibodies (Lorne laboratories, UK, Abcam, UK, respectively) were used for agglutinating RBCs using traditional slide and/or tube agglutination tests or on the surface of filter papers. Anti-H antibody was diluted into 20 μg/mL using phosphate-buffered saline (PBS) with sodium azide added to it (1% w/v). About 7 μL of the antibody (anti-A or anti-H antibodies) was added to the channel’s reaction zone followed by the addition of an equal volume of the blood sample. Plasma purity after separation using anti-H agglutinating antibody was analyzed using a 10× light microscope. The functionality of plasma separation was further tested using biuret reagent (Carolina Biological Supply Company, USA) for serum albumin detection with blue color for positive reaction. Assay optimization under different conditions (filter paper type and grade, anti-H concentration, antibody sample volumes, wax channel design and size, blood collection method, etc.) was investigated. Paper-based plasma separation using anti-H under optimized standard conditions were tested in duplicate (intra-assay variability) and in two different days (interassay variability). The coefficient of variability (CV%) was calculated by dividing the standard deviation on mean to determine reproducibility.
Blood samples were collected using standard laboratory procedures using ethylenediaminetetraacetic acid (EDTA) tubes from random donors or directly using finger prick. Age and gender were recorded, and collective blood count (CBC) analysis was performed including hematocrit (HCT) %, hemoglobin (Hb) levels, white blood cell (WBC) count, RBC count, and platelet count. The efficiency of plasma separation was calculated by dividing the plasma separation band length mean of 45 samples over the total blood band length and compared to mean HCT levels of the same samples. To determine the sensitivity of the new separation method, 119 random EDTA blood samples were collected and separated. The sensitivity rate was calculated as the percentage of number of samples separated successfully divided by the total number.
A thyroid stimulating hormone (TSH) enzyme-linked immunosorbent assay (ELISA) kit (DiaMetra, Italy) was incorporated in the study for TSH detection and comparing the result between the centrifuged plasma sample and filter paper-separated plasma sample (recovery samples) for the same patient. About 20 μL plasma was obtained using paper treated with anti-H antibody on wax designed channels followed by cutting plasma bands and adding them to 980 μL buffer (1:50 dilution ratio), then the plasma bands and buffer were vortexed for a minute. Following that, an ELISA assay was performed using original plasma and recovery samples in the same dilution ratio. The obtained sample concentrations were finally multiplied by the dilution factor. The TSH recovery rate was calculated as the percentage of TSH concentration measured by ELISA from recovered plasma separated by anti-H agglutination antibody on paper divided by TSH concentration measured by ELISA from plasma separated by standard centrifugation.
All the separated bands (in cm) were made in duplicate; then the mean of the results was taken and incorporated in the analysis. The used software for the statistical analysis was SPSS (version 21) using one sample T-test and one-way analysis of variance test.
Whole blood added directly to filter paper with no treatment (Figure 1A) or treated with buffer (Figure 1B) had no plasma separation, while whole blood (group A) added to filter paper treated with anti-A antibody (Figure 1C) or anti-H agglutinating antibody (Figure 1D) had a clear plasma separation zone (yellow color). Wax channels were printed on filter paper grade 5 and then used for plasma separation from EDTA samples. No plasma separation was seen with whole blood alone or treated with buffer only (Figure 2A,B), while plasma separation from whole blood was successful upon using anti-A antibody to blood sample group A, and most importantly, using anti-H antibody to all blood types (Figure 2C,D).
To confirm successful separation of plasma, biuret reagent for serum albumin detection was added to plasma zone. As shown in Figure 3, the blue circles pointed out by the arrows display a positive biuret test for albumin detection proving the separation of albumin containing serum through the channel. The purity of plasma separated using anti-H antibody was further investigated through light microscopy (10×) on different locations on the tested channel. RBCs were seen in the agglutination zone (Figure 4A,B, section 4). Less RBCs were present in the area between the red and yellow zones (Figure 4A,B, section 3). Most importantly, no RBCs were present along the plasma zone (yellow zone) (Figure 4A,B, section 2).
Plasma separation is dependent on two approaches. The membrane filtration approach uses filter papers (grades 44, 42, and 5) displaying pore sizes smaller than the RBC pore size (about 7 μm), and the cell agglutination approach uses anti-H antibody. No separation was seen using filter papers only but occurred on all filter papers using both antibody concentrations (1:25 and 1:50); thus, a lower antibody dilution was used in the rest of the reactions (1:50) (Figure 5A). Equal volumes of antibody and sample (7 μL antibody + 7 μL sample) resulted in the best separation results compared to the other ratios. Moreover, grades 42 and 5 had higher separation purity than grade 44 (Figure 5B). However, grade 5 had better membrane filtration ability within one paper layer fibers and with no sample loss from beneath the paper followed by grade 42, then grade 44 (Figure 5C). Therefore, filter paper grade 5 was used for the separation reactions using equal volumes of the blood sample and anti-H antibody in concentration 1:50.
This plasma separation approach under optimized conditions was tested on 119 random EDTA blood samples using the same volumes and antibody concentration on all samples. The separation was successful on 116 samples (sensitivity rate 116/119 × 100 = 97.5%) but failed on three hemolyzed samples (Figure 6).
Direct fresh blood from finger prick was tested for separation using anti-H antibody treated filter paper after its success on EDTA blood samples. The five different tested samples showed successful separation (Figure 7).
In total, 45 different patient’s samples were collected along with their CBC test results. They were used to correlate their separated plasma bands and RBC bands (in cm) across the wax channels with their age, gender, HCT, Hb levels, WBC count, RBC counts, and platelet count (Table 1). A positive association was found between both band’s levels with gender and HCT% (P value < 0.05). This indicates that both separated bands’ spread varies among samples and provide a relatively true reflection of their HCT% level on the channels. Also, a positive correlation was found between the Hb level and the RBC band (P value < 0.05). However, no correlation was seen between the separated bands and patient’s age, WBC count, RBC count, and platelet count (P value > 0.05). The efficiency of plasma separation was 72% (mean of plasma bands 0.61 cm /mean of total bands length 1.41 cm, compared to expected plasma percentage of 60.1% according to HCT of 39.1%, n = 45 samples).
This plasma separation approach’s ability to be incorporated in future diagnostic tests was investigated to evaluate its capacity to preserve biomarkers quantitatively in the plasma band using the TSH ELISA assay. TSH results obtained from centrifugated samples using standard laboratory procedure were compared with results obtained from filter paper-separated samples from the same patient. The assay was performed three times with samples tested in duplicate or triplicate in each run. Consistent results were obtained, as the average of TSH results from filter paper-separated samples (recovery samples) was 11.8 ± 0.3 mlU/L (n = 3), while the average of TSH results from centrifuged samples was around 11.03 ± 0.05 mlU/L (n = 3) indicating 107% recovery rate (11.8/11.03 × 100).
Sample variability across different days and attempts was investigated using this separation approach. The obtained 45 samples were separated twice for two days. Results showed very close separation levels for all 45 samples during day 1 and day 2 as well as very close results for repeated attempts within the same day for both the RBC and plasma bands (Table 2).
Standard laboratory diagnostic assays require blood collection by venipuncture and separation of plasma or serum by centrifugation as blood cells can complicate analyses. This process requires skilled technicians, well-equipped laboratories, electricity source, and large blood volume and is prone to artifacts. To overcome these limitations, multiple POC assays were developed to separate plasma including for example a simple centrifugation utilizing hand-powered fidget-spinner. Similarly, paper-based assays require plasma separation for similar reasons and because RBC color interferes with colorimetric assays. Paper-based plasma separation can be achieved using filtration, capillarity-driven force, microfluidic channels, membrane separation, and RBC aggregation, coagulation, and agglutination. Different limitations were reported for different methods including complexity, cost, poor yield, large blood volumes, requirement of devices or diluents, long time, protein loss, and others. In this study we report paper-based plasma separation assay using anti-H agglutinating antibody that would overcome most of these limitations. The mechanism of separation in this assay depends mainly on mechanical passive filtration as large, agglutinated RBCs will be fixed/trapped in paper interfibers allowing the fluid phase (plasma) to separate and wick easily. Furthermore, using filter or chromatography paper with pore size less than 3 μm will also trap nonagglutinated RBCs of 6–8 μm size enhancing the efficiency of separation. This mechanism was validated and characterized by many studies and were utilized in the development of many diagnostic applications. Paper-based RBC agglutination/aggregation was achieved using chitosan, blood grouping antibodies, or synthetic paper substrate. Using anti-A, anti-B, and anti-AB blood grouping antibodies is clearly not applicable for the separation of blood group O samples accounting for over 50% of samples. In this study, we use anti-H blood grouping antibody to induce universal RBC agglutination in all blood groups. The ability of anti-H monoclonal antibodies to induce the agglutination of RBCs using different methods was characterized previously. H antigen is present in virtually all RBCs including blood group A, B, AB, and O. Using chitosan as the inducer of blood aggregation necessitates the addition of a diluent and EDTA, requires about 4 min, or requires a specific pattern which slightly complicates the final assay. Anti-A, anti-B, and anti-AB blood grouping antibodies coupled with a synthetic paper substrate mediate efficient separation with minimal protein loss but require a larger blood volume and time for separation, increasing the required cost. Other studies used the principle of RBC agglutination for blood group detection rather than plasma separation. The paper-based separation with anti-H antibody assay reported in this study is simple, instant, and requires small volume. The assay requires the direct addition of one drop of blood about 7 μL obtained by finger prick to filter paper soaked with anti-H antibody on patterned or unpatterned paper. Blood sample, anti-H antibody, and paper did not require any further treatment. Results are obtained immediately within seconds with high purity as indicated by the absence of RBCs in plasma bands under a microscope similar to other studies. While other assays have successfully separated plasma from RBCs on paper, however, most reports are proof-of-principle with minimal validation regarding reproducibility, functionality, efficiency, and applicability. In this study, the reported reproducibility tested on 45 samples was high with minimum variabilities, the assay successfully separates 116/119 (97.5%) samples, the recovery rate of plasma separated on paper was consistent with standard assays and similar or higher than other studies, and the efficiency of plasma separation was higher (72%) compared to other studies which reported an efficiency of 30–60%. For the three samples that did not exhibit separation, RBC hemolysis which limits agglutination and mechanical filtration was found to be the reason. Similarly, RBC hemolysis will affect other conventional or POC separation assays. The assay was affected by many variables related to paper type and structure, antibody, microfluidic channels, and blood sample as reported by other studies. The assay was optimized under standard conditions to induce the best plasma separation. Factors related to blood sample including gender and HCT have a significant effect on plasma separation. This is expected and known to occur with conventional centrifugation or with other POC assay for plasma separation. Importantly, almost all samples were separated successfully using our assay regardless of blood related variabilities. The functionality and applicability of the anti-H treated paper separation assay was confirmed by the detection of plasma albumin by color change and by measuring TSH where levels obtained were comparable to levels obtained by standard centrifugation. Other studies using different methods of plasma separation on paper showed the functionality and applicability of these methods on different analytes like glucose, proteins, vitamin A and iron, and human immunoglobulin G, interferon gamma, and HIV-1 RNA. The limitation of the anti-H paper-based assay is the requirement for antibody addition to the paper which could increase the cost and limit field applications. The assay requires a small amount and low concentration of anti-H antibodies to limit cost. Other studies have shown the interesting ability of filter paper to preserve antibodies at ambient temperature which can be enhanced and prolonged by the addition of simple materials like glycerol, tween, and others. Furthermore, a mild dilution factor might occur due to anti-H antibody solution depending on application time and temperature. This can be easily overcome by drying the antibody solution before the application of the blood droplet. While H antigen is almost universal, individuals lacking H antigen reported as “Bombay” blood group are a rare occurrence at 1 of 10,000 individuals in India and 1 per million in Europe.
A simple, instant, and direct paper-based assay for plasma separation from whole blood using the universal anti-H agglutinating antibody was reported and validated in this study. The assay was significantly affected by paper structure, antibody concentration, and donor gender and HCT. The efficiency of the assay was 72%, the recovery rate was in the range of 90–110%, the sensitivity was 97.5%, and the reproducibility was about 90%. The assay would have numerous applications in paper-based POC tests and in resource limited areas. |
||
PMC9647768 | Makoto Kurano,Daisuke Jubishi,Koh Okamoto,Hideki Hashimoto,Eri Sakai,Yoshifumi Morita,Daisuke Saigusa,Kuniyuki Kano,Junken Aoki,Sohei Harada,Shu Okugawa,Kent Doi,Kyoji Moriya,Yutaka Yatomi | Dynamic modulations of urinary sphingolipid and glycerophospholipid levels in COVID-19 and correlations with COVID-19-associated kidney injuries | 10-11-2022 | COVID-19-associated kidney injuries,Urine,Sphingolipids,Glycerophospholipids,Lipidomics | Background Among various complications of coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), renal complications, namely COVID-19-associated kidney injuries, are related to the mortality of COVID-19. Methods In this retrospective cross-sectional study, we measured the sphingolipids and glycerophospholipids, which have been shown to possess potent biological properties, using liquid chromatography-mass spectrometry in 272 urine samples collected longitudinally from 91 COVID-19 subjects and 95 control subjects without infectious diseases, to elucidate the pathogenesis of COVID-19-associated kidney injuries. Results The urinary levels of C18:0, C18:1, C22:0, and C24:0 ceramides, sphingosine, dihydrosphingosine, phosphatidylcholine, lysophosphatidylcholine, lysophosphatidic acid, and phosphatidylglycerol decreased, while those of phosphatidylserine, lysophosphatidylserine, phosphatidylethanolamine, and lysophosphatidylethanolamine increased in patients with mild COVID-19, especially during the early phase (day 1–3), suggesting that these modulations might reflect the direct effects of infection with SARS-CoV-2. Generally, the urinary levels of sphingomyelin, ceramides, sphingosine, dihydrosphingosine, dihydrosphingosine l -phosphate, phosphatidylcholine, lysophosphatidic acid, phosphatidylserine, lysophosphatidylserine, phosphatidylethanolamine, lysophosphatidylethanolamine, phosphatidylglycerol, lysophosphatidylglycerol, phosphatidylinositol, and lysophosphatidylinositol increased, especially in patients with severe COVID-19 during the later phase, suggesting that their modulations might result from kidney injuries accompanying severe COVID-19. Conclusions Considering the biological properties of sphingolipids and glycerophospholipids, an understanding of their urinary modulations in COVID-19 will help us to understand the mechanisms causing COVID-19-associated kidney injuries as well as general acute kidney injuries and may prompt researchers to develop laboratory tests for predicting maximum severity and/or novel reagents to suppress the renal complications of COVID-19. Supplementary Information The online version contains supplementary material available at 10.1186/s12929-022-00880-5. | Dynamic modulations of urinary sphingolipid and glycerophospholipid levels in COVID-19 and correlations with COVID-19-associated kidney injuries
Among various complications of coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), renal complications, namely COVID-19-associated kidney injuries, are related to the mortality of COVID-19.
In this retrospective cross-sectional study, we measured the sphingolipids and glycerophospholipids, which have been shown to possess potent biological properties, using liquid chromatography-mass spectrometry in 272 urine samples collected longitudinally from 91 COVID-19 subjects and 95 control subjects without infectious diseases, to elucidate the pathogenesis of COVID-19-associated kidney injuries.
The urinary levels of C18:0, C18:1, C22:0, and C24:0 ceramides, sphingosine, dihydrosphingosine, phosphatidylcholine, lysophosphatidylcholine, lysophosphatidic acid, and phosphatidylglycerol decreased, while those of phosphatidylserine, lysophosphatidylserine, phosphatidylethanolamine, and lysophosphatidylethanolamine increased in patients with mild COVID-19, especially during the early phase (day 1–3), suggesting that these modulations might reflect the direct effects of infection with SARS-CoV-2. Generally, the urinary levels of sphingomyelin, ceramides, sphingosine, dihydrosphingosine, dihydrosphingosine l-phosphate, phosphatidylcholine, lysophosphatidic acid, phosphatidylserine, lysophosphatidylserine, phosphatidylethanolamine, lysophosphatidylethanolamine, phosphatidylglycerol, lysophosphatidylglycerol, phosphatidylinositol, and lysophosphatidylinositol increased, especially in patients with severe COVID-19 during the later phase, suggesting that their modulations might result from kidney injuries accompanying severe COVID-19.
Considering the biological properties of sphingolipids and glycerophospholipids, an understanding of their urinary modulations in COVID-19 will help us to understand the mechanisms causing COVID-19-associated kidney injuries as well as general acute kidney injuries and may prompt researchers to develop laboratory tests for predicting maximum severity and/or novel reagents to suppress the renal complications of COVID-19.
The online version contains supplementary material available at 10.1186/s12929-022-00880-5.
Coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is associated with various complications. Among them, renal complications, especially acute kidney injury (AKI), are associated with critical conditions. These renal complications are known as COVID-19-associated kidney injuries. A high incidence of AKI has been reported, especially among critically ill patients, and patients with COVID-19-associated kidney injuries reportedly have a higher risk of in-hospital death [7, 13, 14, 40]. Moreover, recent studies have suggested prolonged kidney dysfunction in some patients with COVID-19 [43]. Regarding the etiology of COVID-19-associated kidney injuries, the mechanisms have not been fully elucidated at present, and both direct and indirect mechanisms have been proposed [30]. ACE2 and TMPRSS2, which are key proteins for the entry of SARS-CoV-2 into human cells [15], are highly expressed in podocytes and proximal tubules in the kidney, and SARS-CoV-2 can directly infect and damage the kidney [6, 42, 47]. As indirect mechanisms, complications such as thrombosis and endotheliitis, which are frequently observed in COVID-19 patients [57, 62, 64], can cause renal impairment. In addition, other mechanisms not specific to COVID-19, such as right heart failure [5], cytokines, and nephrotoxins [30], have been proposed to be involved in COVID-19-associated kidney injuries. Since COVID-19-associated kidney injuries are important in terms of clinical outcomes and the underlying mechanisms have yet to be fully elucidated, as described above, investigating biomarkers of COVID-19-associated kidney injuries will be important to understand the pathogenesis of such complications. Urine samples rapidly and accurately reflect renal conditions. Actually, recent studies have revealed that urinary chemical biomarkers (urinary total protein [TP], N-acetyl-β-d-glucosaminidase [NAG], α1-microglobulin [α1-MG], neutrophil gelatinase-associated lipocalin [NGAL], and liver type fatty acid-binding protein [L-FABP]) and urine sediment findings are correlated with both the severity of COVID-19 and COVID-19-associated kidney injuries [16, 19, 35]. Therefore, in the present study, we investigated urinary biomarkers for COVID-19-associated kidney injuries. In this study, we focused on sphingolipids and glycerophospholipids. A series of basic and clinical studies have revealed the importance of these lipids in the pathogenesis of various human diseases including kidney diseases. Among the sphingolipids, the bioactivities of sphingosine 1-phosphate (S1P) and ceramides have been well studied. S1P possesses potent anti-apoptotic and pro-survival properties [27, 33]. A series of basic studies showed that the S1P1 signal might attenuate both acute and chronic kidney diseases through its pro-survival, anti-inflammation, anti-fibrosis, and vasoprotective properties [1, 12, 24, 25, 50], whereas the S1P2 signal might aggravate kidney disease by accelerating fibrosis and inflammation [24, 49]. Ceramides have both pro-apoptosis and pro-inflammation properties [44, 61] and have been shown to accelerate the pathological condition of both chronic and acute kidney diseases in basic studies [41, 56]. Regarding the metabolism of sphingolipids, ceramides are derived from sphingomyelin (SM) and can be converted into sphingosine (Sph). S1P is produced from Sph by S1P kinases [33]. Dihydrosphingosine 1-phosphate (dhS1P), another analog for S1P receptors, is produced from dihydrosphingosine (dhSph) by S1P kinases, and dhSph is processed into ceramides via dihydroceramides [2]. In clinical studies, urinary ceramide levels were reportedly associated with diabetic nephropathy [38, 51], and urinary SM levels are altered in chronic kidney diseases [67]. However, the modulation of urinary sphingolipid levels remains unknown, especially in terms of the pathogenesis of human AKI. Among glycerophospholipids, lysophosphatidic acids (LPA) and lysophosphatidylcholine (LPC) have been well studied in the fields of nephrology. LPA is produced from LPC by autotaxin, and six kinds of LPA receptors have been identified [68]. The roles of LPA in inflammation depend on its receptors. LPA can exacerbate the pathogenesis of chronic kidney diseases, resulting in inflammation and fibrosis [29, 53], while it can also reportedly protect against acute kidney diseases [9, 34]. The urinary LPA levels are elevated in diabetic nephropathy [55], while the urinary autotaxin levels increase in membranous nephropathy [36]. The urinary LPC levels are also positively correlated with kidney dysfunction, and LPC itself might exert lipotoxicity [69]. Lysophosphatidylinositol (LPI) reportedly exacerbates the pathogenesis of sepsis-associated AKI [22]. Regarding other glycerolysophospholipids, such as lysophosphatidylethanolamine (LPE), lysophosphatidylglycerol (LPG), and lysophosphatidylserine (LPS), their roles in kidney injuries remain unknown, and the modulation of urinary lysophospholipids in AKI also remains to be elucidated. Along with LPA, the GPR34, P2Y10, and GPR174 receptors have been shown to be specific for LPS [17], while the GPR55 receptor is specific for LPI and LPG [45]. LPC, LPS, LPE, LPI, and LPG are produced from phosphatidylcholine (PC), phosphatidylserine (PS), phosphatidylethanolamine (PE), phosphatidylinositol (PI), and phosphatidylglycerol (PG), respectively. Although only a limited number of studies are available, PC has been shown to possess protective effects against acute kidney injuries through its antioxidant properties [10, 28], and PS might reduce nephrotoxicity by suppressing inflammation [21]. The roles of other diacylphospholipids in the pathogenesis of AKI have not been reported. Regarding the modulations of urinary diacylphospholipids in humans, a recent study showed that urinary PC levels might be associated with adverse outcomes and mortality in patients with chronic kidney diseases [60]; however, their associations with human AKI are not well known at present. Although several lipidomics analyses using serum or plasma samples from patients with COVID-19 have been performed [23], only one lipidomics study has been conducted using urine samples from a very small number of COVID-19 patients [31]. With this background in mind, we performed lipidomic analyses using urine samples to investigate the mechanisms responsible for COVID-19-associated kidney injuries to understand the pathogenesis of COVID-19 and COVID-19-associated kidney injuries, as well as AKI, better and to help researchers develop reagents capable of preventing severe kidney injuries in the future. In this study, we measured the longitudinal urinary levels of sphingolipids and glycerophospholipids in 272 samples from 91 COVID-19 subjects and 95 samples from 95 control subjects without infectious disease.
We collected the residual urinary samples after routine clinical testing from 91 subjects who had been diagnosed as having COVID-19 using an RT-PCR assay between September 2020 and April 2021. The sampling times were classified into the following eight periods: day 1–3, day 4–6, day 7–9, day 10–12, day 13–15, day 16–18, day 19–24, and day 25–40 after symptom onset. Since the timing of RT-PCR testing varied largely among the patients, we used the day after symptom onset as the initial measurement in our investigation of lipid modulation. None of the subjects enrolled in the present study had been vaccinated at the time of sampling. The subjects were classified into three groups according to the maximum severity of COVID-19: maximum severity group 1 (did not require oxygen supplementation), maximum severity group 2 (required oxygen supplementation, but did not require mechanical ventilatory support), and maximum severity group 3 (required mechanical ventilatory support). As a control, we collected 95 urine samples from volunteers without infectious diseases. The current study was performed in accordance with the ethical guidelines established by the Declaration of Helsinki. Written informed consent for sample analysis was obtained from some of the patients. For the remaining participants from whom written informed consent could not be obtained (because they had been discharged or transferred out of the hospital), informed consent was obtained in the form of an opt-out on our institution’s website, as follows. Patients were informed of the study through the website, and those who were unwilling to be enrolled were excluded. The study design was approved by The University of Tokyo Medical Research Center Ethics Committee (2602 and 2020206NI).
We measured the levels of the lipid mediators listed below using four independent LC–MS/MS methods and the LC8060 system, consisting of a quantum ultra-triple quadrupole mass spectrometer (Shimadzu, Japan). We simultaneously measured six ceramide species (Cer d18:1/16:0 [C16:0], Cer d18:1/18:0 [C18:0], Cer d18:1/18:1 [C18:1], Cer d18:1/20:0 [C20:0], Cer d18:1/22:0 [C22:0], and Cer d18:1/24:0 [C24:0]), Sph, and dhSph, as previously described [38]. We also measured S1P and dhS1P, as described previously [52]. Furthermore, LPA, LPC, LPS, LPI, LPG, and LPE were also measured, as described previously [37]. We monitored 11 acyl chains (14:0, 16:0, 16:1, 18:0, 18:1, 18:2, 18:3, 20:3, 20:4, 20:5, and 22:6) for these lysophospholipids as well as 22:5 LPI. We also measured SM and diacylphospholipids, including PC, PE, PI, PG, and PS [26]. We monitored 17 diacyl chains (32:1, 32:2, 34:1, 34:2, 36:1, 36:2, 36:3, 36:4, 38:1, 38:2, 38:3, 38:4, 38:5, 38:6, 40:1, 40:2, and 40:7) for SM and 64 diacyl chains (28:0, 28:1, 28:2, 30:0, 30:1, 30:2, 32:0, 32:1, 32:2, 32:3, 32:4, 34:0, 34:1, 34:2, 34:3, 34:4, 34:5, 34:6, 36:0, 36:1, 36:2, 36:3, 36:4, 36:5, 36:6, 36:7, 38:0, 38:1, 38:2, 38:3, 38:4, 38:5, 38:6, 38:7, 38:8, 40:0, 40:1, 40:2, 40:3, 40:4, 40:5, 40:6, 40:7, 40:8, 40:9, 40:10, 42:0, 42:1, 42:2,42:3, 42:4, 42:5, 42:5, 42:6, 42:7, 42:8, 42:9, 42:10, 42:11, 44:0, 44:1, 44:2, 44:6, 44:7, and 44:12) for PC, PE, PI, PG, and PS. With the exceptions of SM and the diacylphospholipids, both the intra-day and inter-day coefficients of variation for the metabolites were below 20%, as validated in our previous papers [37, 38, 52]. The urinary levels of the measured lipids were adjusted to the urinary creatinine levels.
To measure the urinary clinical markers, we used the reagents as described previously [35]. Renal tubular epithelial cells (RTE) were counted per high-power field of view (/HPF); urinary casts were classified into hyaline casts (HyaC), granular casts (GraC), epithelial casts (RTEC), and waxy casts (WaxC), and their numbers were counted per whole field (/WF). The RTE findings were classified as rank 0 (absent), rank 1 (< 1/HPF), rank 2 (1–4/HPF), rank 3 (5–9/HPF), or rank 4 (> 10/HPF), those of HyaC were classified as rank 0 (absent), rank 1 (< 4/WF), rank 2 (5–19/WF), rank 3 (20–49/WF), or rank 4 (> 50/WF), those of GraC and RTEC were classified as rank 0 (absent), rank 1 (< 4/WF), rank 2 (5–49/WF), or rank 3 (> 50/WF), and those of WaxC were classified as rank 0 (absent) or rank 1 (present).
The data were analyzed using SPSS (Chicago, IL) or MetaboAnalyst (https://www.metaboanalyst.ca/). To examine differences in the time courses of urinary lipids among the control subjects and maximum severity groups 1, 2, and 3, we evaluated significant differences using the Kruskal–Wallis test, followed by the Steel–Dwass test as a post-hoc test. To examine differences in the urinary lipid levels longitudinally between specific time points in a specific maximum severity group, we used the paired Wilcoxon signed-rank test. To examine differences between the control subjects and the COVID-19 subjects, we performed non-parametric Volcano plot analyses. For the correlation studies, a Kendall rank correlation was performed to examine the correlations of lipids and clinical data with the maximum severity of COVID-19, using age, sex, and the presence of diabetes, hypertension, and current smoking as covariates of interest. To construct machine learning models, we used SPSS modeler ver. 18:2 (Chicago, IL) and performed CHAID analyses, SVM analyses, and neural network analyses. The Spearman rank correlation was performed to examine correlations between lipids and clinical data. The independent effects of the clinical properties and the results of urinary laboratory tests on urinary lipid levels were investigated with a stepwise multiple regression analysis, using urinary lipid levels as objective variables and clinical information, maximum severity, eGFR, CRP, d-Dimer, urinary chemical markers, urinary sediment findings, SG, pH, and urinary sodium levels as possible explanatory factors. To examine differences between the subjects treated with antiviral reagents and those without, we used the Mann–Whitney U test. The graphs shown in the figures were prepared using Graphpad Prism 9 (GraphPad Software, San Diego, CA) or MetaboAnalyst. P values of less than 0.05 were deemed as denoting statistical significance in all the analyses.
The characteristics of all the subjects and the numbers of samples analyzed at each specific time point are described in Additional file 1: Tables S1 and S2, respectively. As shown in the tables, differences in patient age were seen among the maximum severity groups, while differences in the percentage of patients with hypertension were seen between the control subjects and the maximum severity groups. We also observed differences in sex between the control subjects and the maximum severity groups on day 19–24 and day 25–40. In the control group, no differences in the sphingolipid and total glycerophospholipid levels were seen between subjects with hypertension and those without hypertension. Therefore, the presence of hypertension might not have had a large effect on the results in the present analyses. Regarding sex, the urinary levels of several monitored lipids were higher in female subjects. The ratios of the lipid levels in female subjects relative to those in male subjects were 140.3% for the total SM levels, 191.2% for the S1P levels, 154.0% for the dhS1P levels, 186.6% for the Sph levels, 152.1% for the dhSph levels, 150.9% for the C18:1 Cer levels, 141.1% for the total LPG levels, 145.5% for the total PC levels, 144.2% for the total PE levels, 165.6% for the total PG levels, 151.5% for the total PI levels, and 125.1% for the total PS levels. Age was positively correlated with the total LPC levels (r = 0.214, p = 0.037) and the total PS levels (r = 0.218, p = 0.034). Therefore, we think that these correlations were thought to have had a minimal impact on the interpretation of the dynamic modulations of the monitored lipid levels, as described below. The time courses for the urinalysis results and other clinical parameters are shown in Additional file 1: Fig. S1. Overall, the modulations of these parameters seemed reasonable, while a remarkable decline in eGFR was not observed in the COVID-19 subjects.
Figure 1 shows the time courses of the urinary sphingolipid levels in the COVID-19 subjects. C16:0 Cer and SM increased most rapidly from day 4–6. C18:1 Cer, C20:0 Cer, C22:0 Cer, C24:0 Cer, Sph, dhSph, and dhS1P significantly decreased or tended to decrease during the early phase (day 1–3) in maximum severity group 1 and then increased, especially in maximum severity group 3. Among the monitored sphingolipids, the C18:0 Cer levels rapidly decreased from day 1–3. The urinary levels of several sphingolipids seemed remarkably higher on day 25–40; however, several biases might be present, since samples were collected only from patients with severe COVID-19 who were still hospitalized on day 25–40. Regarding longitudinal comparisons, although we could compare the urinary lipid levels between only limited time points, the results showed the elevation of urinary sphingolipids during the time course of COVID-19, especially in day 19–40 in maximum severity group 3 (Additional file 1: Figs. S2A-D and S3).
Figure 2 shows an overview of the total glycerophospholipid modulations. The total levels of all the monitored glycerophospholipids increased, especially in severe COVID-19. Regarding the PC-LPC-LPA axis, the urinary PC levels rapidly increased in maximum severity group 3; they tended to decrease in maximum severity groups 1 and 2. The urinary LPA and LPC levels increased, especially during the later phase. Regarding the PS-LPS axis, PS increased rapidly from the early phase (day 1–3) and LPS increased from day 7–9, especially in patients with severe COVID-19. Regarding the PE-LPE axis, LPE increased from the early phase (day 4–6) and reached a peak on day 10–12 in maximum severity group 2. In maximum severity group 3, LPE increased, especially during the later phase (day 19–24). The urinary total PE levels were modulated in an almost similar manner to those of LPE. Regarding the PG-LPG axis and the PI-LPI axis, LPG and LPI increased in maximum severity groups 2 and 3 from around the middle phase (day 7–15), while PG and PI increased only in maximum severity group 3. Regarding longitudinal comparisons, the paired statistical analyses showed the elevation of urinary glycerophospholipids during the time course of COVID-19, especially in day 19–40 in maximum severity group 3 (Additional file 1: Fig. S2E-H, S4).
To investigate time-course-dependent lipid modulations in greater detail, we next created separate volcano plots for each sampling point, as shown in Additional file 1: Figs. S5–S8. To understand the lipid modulations that occur in patients with COVID-19 better, the lipids with the 20 lowest p values at each time period were selected; their log2(FC) and p values are shown in Fig. 3A. Among the sphingolipids, the C18:0 Cer level decreased markedly, especially during the early phase. Decreased levels of PC, LPA, and LPC species were clearly observed until day 16–18, while the levels of several species including 38:2 PC, 42:10 PC, and 44:2 PC increased. Regarding the PE-LPE axis, increases in PE and LPE species were clearly observed after day 7–9, especially on day 19–24 and day 25–40. The levels of several species including 28:0 PE, 30:0 PE, 34:3 PE, 38:8 PE, and 18:3 LPE decreased. Regarding the PS-LPS axis, specific PS species, such as 36:2 PS and 36:3 PS, increased, especially during the middle phase (day 16–18 and day 19–24), while 38:5 PS consistently increased almost throughout the time course. Decreases in 38:0 PS and 34:3 PS were observed on day 7–12 and day 4–6, respectively. 18:0 LPS increased on day 7–9. Regarding the PI-LPI axis, 14:0 LPI decreased on day 4–15 and 32:0 PI decreased on day 4–6, while 16:1 LPI, 18:2 LPI, and 18:3 LPI and several PI species containing 16:1, 18:1, and 18:3 acyl chains increased after day 10–12. Regarding the PG-LPG axis, the increase in 14:0 LPG after day 10–12 and the decrease in 34:5 PG throughout the time course seemed characteristic. The time courses of the representative lipids are shown in Fig. 3B–I and Additional file 1: Figs. S9, S10.
Next, we performed correlation analyses with the maximum severity of COVID-19, using age, sex, the presence of diabetes and hypertension, and current smoking as covariates of interest. Figure 4A shows the correlation coefficients and the p values of the lipids and clinical parameters with the 20 lowest p values at any specific time points. Among sphingolipids, SM (except for 36:3 SM) and Sph were positively correlated with maximum severity on day 10–12. On day 19–40, C18:0 Cer, C22:0 Cer, and C24:0 Cer were positively correlated with maximum severity. Regarding LPA, the 18:1 LPA, 20:3 LPA, and 16:1 LPA levels on day 10–12 and day 13–15 were negatively correlated with maximum severity. The 14:0 LPC and 16:1 LPC levels on day 7–9 and the 20:4 LPC levels on day 19–40 were positively correlated with maximum severity. Many PC species had strong positive correlations with maximum severity, while the 32:3 PC, 40:2 PC, 42:0 PC, 44:0 PC, 44:1 PC, and 44:2 PC levels during the middle phase (day 10–18) had rather strong negative correlations with maximum severity. Many PE species were negatively correlated with maximum severity, while 34:4 PE, 34:5 PE, and 38:6 PE at some sampling points were positively correlated with maximum severity. Several PG and PI species, especially on day 10–15, were negatively correlated with maximum severity. Among LPG species, 14:0 LPG was positively correlated with maximum severity. Meanwhile, LPS and PS species were, in general, negatively correlated with maximum severity. Additional file 1: Fig. S11 shows the time courses of characteristic lipids.
In addition to performing simple correlation studies, we investigated which lipids and clinical parameters were strongly correlated with the maximum severity of COVID-19 using machine learning techniques. Figure 4B and Additional file 1: Fig. S12A, B show the lipids or clinical parameters selected with high importance by CHAID analyses, SVM analyses, and neural network analyses, respectively. In the CHAID analyses, the PC-LPA-LPC axis throughout the time course, except day 16–18, had a high importance for determining maximum severity in the constructed models. The PE-LPE axis on day 7–15, PS on day 10–12, C16:0 Cer on day 1–6, SM on day 13–15, and 14:0 LPG on day 7–9 had some importance. In the SVM and neural network analyses, although the importance of each parameter was relatively low, the PC-LPA-LPC axis throughout the time course, sphingolipids during the early phase, and PS and the PE-LPE axis during the middle phase had importance for determining maximum severity in the constructed models.
Next, we investigated the correlations of the monitored lipids with clinical parameters. Figure 5 and Additional file 1: Figs. S13, S14 shows the time courses of the correlations. As shown in Fig. 5A, the urinary levels of several lipids had positive correlations with serum CRP and d-Dimer levels. The lipids which had a negative correlation with the urinary SG levels and a positive correlation with sodium levels are deemed to increase in the pathogenesis the renal factors or decrease in the pathogenesis of prerenal kidney injuries. As shown in Fig. 5B, the urinary SM, C18:1 Cer, PC, and PG levels had negative correlations with the urinary SG levels and positive ones with the urinary sodium levels in the middle phase (day 13–18). The urinary TP levels were positively correlated with ceramides, Sph, LPC, PS, LPS, PE, LPE, and LPG in the early phase (day 1–6 and/or day 7–9) and in the late phase (day 16–18 and/or day 19–40). They consistently had positive correlations with the urinary C16:0 Cer and LPC levels. Regarding the urinary chemical biomarkers, generally, urinary sphingolipids except SM had positive correlations with urinary chemical biomarkers. Especially, the C16:0 and C18:1 ceramides had positive correlations almost throughout the monitored periods. The urinary glycerophospholipids, except PC, LPA, and LPI, generally had positive correlations with the urinary chemical markers (Fig. 5C). Interestingly, the eGFR levels were positively correlated with the urinary levels of several sphingolipids, while they were negatively correlated with the urinary LPC, LPE, and PE levels in the early to the middle phase. In the late phase (day 19–40), the eGFR levels were negatively correlated with the urinary sphingolipids and glycerophospholipids (Fig. 5D). Regarding the urinary sediment findings, the urinary SM and dhS1P levels were negatively correlated with RTE, while the urinary ceramides levels were positively correlated with RTE and GraC, except the negative correlations observed in day 16–18. Among glycerophospholipids, the urinary PC, LPA, PG, LPG, PI levels had negative correlations with RTE in day 7–15. The urinary LPC levels had positive correlations with the urinary sediment findings in many time points. The urinary PS, LPS, PE, and LPE also had positive correlations with the urinary sediment findings in the early phase (day 1–9) (Fig. 5E and Additional file 1: Fig. S14). We further investigated the independent effects of the systematic severity of COVID-19, represented by d-Dimer and CRP, and renal injuries, represented by the results of urinary laboratory tests on urinary lipid levels, were evaluated with a multiple regression analysis, using urinary lipid levels as subjective variables. As shown in Additional file 1: Figs. S15–S18, urinary chemical makers such as NGAL and L-FABP were selected as positive explanatory factors with high β values for ceramides, S1P, dhS1P, dhSph, LPC, LPS, LPE, LPG, LPA, PC, PE, PG, PI, and PS, whereas d-Dimer or maximum severity were selected as positive explanatory variables for SM, dhSph, LPI, when all the samples were analyzed. Although, when samples were analyzed separately according to the days after the onset of COVID-19, the results were not always consistent, these results suggested that both renal injuries and systematic severity would affect the dynamic modulations of urinary sphingolipids and glycerophospholipids in COVID-19. In addition, although, since this is a cross-sectional study, we could not conclude the possible influences of antiviral therapy on urinary lipid levels, we also observed some differences in urinary lipid levels between subjects treated with antiviral reagents such as remdesivir [63] and favipiravir [11] and those without (Additional file 1: Figs. S19, S20).
Lastly, after we finished all the analyses, to validate the main results, we measured 46 additional urine samples collected from 31 independent subjects who had been diagnosed as having COVID-19 using an RT-PCR assay between April 2021 and August 2021. Additional file 1: Figs. S21 and S22 show the concentrations of lipids, overlayed on Figs. 1 or 2. As shown in these figures, the modulations of sphingolipids and glycerophospholipids were generally replicated. Moreover, when we investigate the accuracy of the predicting models for maximum severity described in Fig. 4B and Additional file 1: Fig. S12, using these independent samples, we obtained the accuracy of over 80% (Additional file 1: Table S3). Considering these results, we think that the modulations of these lipids could be replicated.
The present study examined the dynamic modulations of urinary sphingolipids and glycerophospholipids in COVID-19 subjects. The modulations of the monitored lipids are summarized in Fig. 6. The modulations of each representative specie of lipids which were not shown in the previous figures are described in Additional file 1: Figs. S23–S25. The urinary SM levels increased only in maximum severity group 3 and were positively correlated with the maximum severity of COVID-19, suggesting that SM modulations were not specific to COVID-19-specific factors but were instead related to kidney injuries accompanying severe infection. Contrary to a previous paper reporting that the urinary SM levels were positively correlated with the urinary TP levels [67], the urinary levels of the SM species were not consistently correlated with the urinary TP level. Of note, the urinary total SM level was rather strongly negatively correlated with the urinary SG and positively with sodium levels (Fig. 5). The negative correlations of the SM with the RTE suggested the possibility that the urinary SM levels decrease in response to the prerenal factors, although a previous study reported that sphingomyelinase activities declined in the model of ischemic renal injury [70]. The modulations of urinary ceramides largely depended on the species. Overall, the levels of the monitored ceramides increased, especially in severe COVID-19, with the exception of C18:0 Cer (Fig. 1). Many ceramide species were downregulated during the early phase (day 1–3) in maximum severity group 1, suggesting that the downregulation of ceramides might originate from the direct influence of infection with SARS-CoV-2. A recent report demonstrated that the envelope of SARS-CoV-2 is rich in cholesterol and phospholipids and poor in sphingolipids [54], suggesting that these modulations are unlikely to be explained by the expenditure of sphingolipids during virus replication. During the later phase, the elevation of ceramides in severe COVID-19 might reflect the progression of kidney injuries, since ceramide production was induced in kidney injury models and ceramides are known to induce the apoptosis of renal mesangial cells and renal tubular epithelial cells [3, 18, 32]. Actually, the urinary ceramide levels were positively correlated with urinary chemical biomarkers and the urinary sediment findings in the COVID-19 subjects in the present study (Fig. 5). The urinary Sph, dhSph, and dhS1P levels increased, especially during the later phase in maximum severity group 3, while they tended to decrease during the early phase (day 1–3) in maximum severity group 1 (Fig. 1). These results suggested that reductions in Sph and dhSph might be direct effects of infection with SARS-CoV-2, while increases in these sphingolipids may occur as responses to kidney injuries associated with COVID-19, as observed for ceramides. Actually, their urinary levels had positive correlations with the urinary chemical markers (Fig. 5). Considering the agonistic properties of dhS1P for S1P receptors and the potential protective properties of S1P receptors against kidney injuries [1, 12, 24, 25], the elevation in urinary dhS1P levels in severe COVID-19 might reflect a compensatory mechanism in response to COVID-19-associated kidney injuries. Regarding the PC-LPC-LPA axis, the urinary total PC levels increased from the early phase (day 4–6) only in maximum severity group 3, while the total LPA levels increased during the later phase (Fig. 2). However, when the lipid modulations were investigated in detail, many PC, LPC, and LPA species decreased in the COVID-19 subjects (Fig. 3). These results suggested that SARS-CoV-2 infection generally downregulated the PC-LPA-LPA axis in a direct manner, while kidney injuries caused by critical COVID-19 disease resulted in upregulation. Urinary SG levels had rather negative correlations with PC and urinary sodium levels had positive ones with PC, while the urinary PC levels were rather negatively correlated with the RTE, suggesting that the urinary PC levels decrease in response to the prerenal factors. Regarding the pathophysiological significance, since LPA is involved in renal fibrosis as well as inflammation [20, 29, 53] and LPC has been shown to have a strong lipotoxicity in the field of nephrology [69], increases in LPA and LPC, especially during the later phase, might result in the translation of acute kidney injuries into chronic kidney injuries, which has been observed as a sequelae of COVID-19 [43]. Regarding the PS-LPS axis, the urinary PS and LPS levels increased rapidly, especially in patients with severe COVID-19 (Fig. 2). Although modulations of the urinary PS levels in AKI have not been reported, considering that PS is involved in apoptosis [39] and exosome formation [59], the elevation of PS in COVID-19-associated kidney injuries seems reasonable. The urinary PS and LPS levels were positively correlated with urinary chemical biomarkers and urinary sediment findings, suggesting that these levels reflect kidney injuries that have been mainly caused by renal factors. The roles of LPS remain to be elucidated in the fields of nephrology, while we recently demonstrated the elevation of PS-PLA1, a producing enzyme for LPS, in the serum of COVID-19 patients [58]. Since LPS and LPS receptors are involved in the regulation of the immune system through three kinds of specific receptors [17, 46], LPS might possess important roles in the pathogenesis of COVID-19-associated kidney injuries, in which inflammation might at least partly be involved [30]. The urinary levels of PE and LPE increased in COVID-19 beginning at an early phase. The upregulation of the PE-LPE axis might be characteristic of COVID-19, as shown in the volcano plots (Fig. 3). PE is abundant in the envelope of SARS-CoV-2 [54] and is reportedly involved in the replication of RNA viruses [66]. Previous studies suggested that LPE might possess anti-inflammatory properties on macrophages [48], which might activate natural killer T cell-dependent protective immunity [71]. Considering these potential biological properties of LPE and the negative correlation between LPE levels during the early phase and maximum severity, LPE might have protective biological properties against the pathogenesis of COVID-19, and a failure to increase LPE levels might be one mechanism resulting in the aggravation of COVID-19. Regarding the PG-LPG axis and the PI-LPI axis, the roles of PG in AKI are unknown, while PG reportedly suppresses toll-like receptor-mediated inflammation [8], suggesting that a decrease in PG might promote kidney injury. In contrast, the upregulation of LPG in patients with severe COVID-19 might contribute to the acceleration of inflammation, since LPG exerts agonist activities with proinflammatory GPR55 [22, 45]. LPI also acts on GPR55 [45], suggesting that the elevation of LPI during the later phase of severe COVID-19 might accelerate the pathogenesis of COVID-19-associated kidney injuries. Regarding the correlation with clinical phenotypes, the PG-LPG axis and the PI-LPI axis show somehow strange correlations with the urinary chemical markers and the urinary sediment findings (Fig. 5). Some unknown mechanisms are involved for the opposite results in the associations with lipids between the urinary chemical markers and the urinary sediment findings. To the best of our knowledge, the modulations of sphingolipids and glycerophospholipids in the urine of AKI have not been well studied, while urinary levels of ceramides, SM, LPA, LPC, and PC have been demonstrated to be higher in chronic kidney diseases, especially diabetic nephropathy, as described in the Introduction [38, 51, 55, 60, 67, 69]. Although the number of the subjects was limited, when we investigated the association between diabetes and urinary lipid levels in the control subjects used in the present study, we observed that the urinary levels of total LPG and S1P were also higher in the subjects with diabetes as well as LPC and ceramides. These results together with the previous reports suggested that the mechanisms similar to diabetic nephropathy, such as inflammation, oxidative stress, and fibrosis, might be somehow involved in the modulations of sphingolipids and glycerophospholipids in the present study. Anyway, since significant elevation of urinary levels of LPS, PS, LPE, PE, PG, LPI, and PI have not been observed or reported in the urine of chronic kidney diseases, several mechanisms specific to COVID-19 or AKI may be involved in the dynamic modulations of these lipids. Since this was an observational study, the main limitation is that the possible involvement of these lipid modulations in the pathogenesis of COVID-19-associated kidney injuries remains unknown. However, the accuracy of the models predicting maximum severity that were constructed using machine learning methods was generally high, especially during the early phase of COVID-19 (Fig. 4B and Additional file 1: Fig. S12), and several lipids were selected as important factors, in addition to important clinical parameters that are typically used to predict severity. These results suggest the potential usefulness of these lipids as biomarkers for predicting the maximum severity of COVID-19. Moreover, since the PC-LPA-LPC axis was important throughout the time course, sphingolipids were important during the early phase, and PS and the PE-LPE axis were important during the middle phase in all three machine learning models for predicting maximum severity, these lipids might have some important physiological properties in the pathogenesis of COVID-19 or its associated kidney injuries. Another limitation is that, although we evaluated possible confounding factors affecting the modulations of urinary sphingolipids and glycerophospholipids as described in the first section of the Results, we could not completely match the backgrounds of both the COVID-19 and control groups, since some characteristics such as age differed largely among the maximum severity groups. In addition, since the eGFR levels were not obviously modulated in the present study (Additional file 1: Fig. S1H), we were unable to investigate cases with severe AKI. At present, we could not conclude whether the urinary modulations of the lipids would recover when the patients are cured of COVID-19. Although the data is preliminary since the number of samples were limited, the urinary lipid levels generally recovered to the range of control subjects and no obvious differences among the maximum severity were observed except that the total PE levels were still higher in the subjects who had recovered from severe COVID-19 (Additional file 1: Fig. S26). Considering that serum modulation of lipids maintained for a long time [4, 65], further studies with post-COVID-19 subjects are necessary in the future to elucidate the mechanisms for long COVID-19.
In summary, analyses of urine samples collected from COVID-19 subjects showed that decreases in the urinary levels of C18:0, C18:1, C22:0, and C24:0 ceramides, Sph, dhSph, PC, LPC, LPA, and PG and increases in those of PS, LPS, PE, and LPE, especially during the early phase, might be derived from the direct effects of SARS-CoV-2 infection, while increases in the urinary levels of SM, ceramides, Sph, dhSph, dhS1P, PC, LPA, PS, LPS, PE, LPE, PG, LPG, PI, and LPI, especially during the later phase, might result from kidney injuries accompanying severe COVID-19. We believe that these results may prompt researchers to perform further investigations to develop laboratory testing methods based on sphingolipid and glycerophospholipid modulations for predicting the maximum severity of COVID-19 and/or novel reagents to suppress the renal complications of COVID-19.
Additional file 1. Supplementary tables and figures. |
|
PMC9647780 | Rebecca A. Luchtel,Yongmei Zhao,Ritesh K. Aggarwal,Kith Pradhan,Shahina B. Maqbool | ETS1 is a novel transcriptional regulator of adult T-cell leukemia/lymphoma of North American descent | 09-06-2022 | Abstract Adult T-cell leukemia/lymphoma (ATLL) is an aggressive T-cell lymphoma associated with the human T-cell lymphotropic virus type 1 virus endemic in regions including Japan, the Caribbean islands, and Latin America. Although progress has been made to understand the disease, survival outcomes with current standard therapy remain extremely poor particularly in acute ATLL, underlying the need for better understanding of its biology and identification of novel therapeutic targets. Recently, it was demonstrated that ATLL of North American–descendent patients (NA-ATLL) is both clinically and molecularly distinct from Japanese-descendent (J-ATLL), with inferior prognosis and higher incidence of epigenetic-targeting mutations compared with J-ATLL. In this study, combined chromatin accessibility and transcriptomic profiling were used to further understand the key transcriptional regulators of NA-ATLL compared with J-ATLL. The ETS1 motif was found to be enriched in chromatin regions that were differentially open in NA-ATLL, whereas the AP1/IRF4 motifs were enriched in chromatin regions more open in J-ATLL. ETS1 expression was markedly elevated in NA-ATLL in both cell line and primary tumor samples, and knockdown of ETS1 in NA-ATLL cells resulted in inhibition of cell growth. CCR4, a previously identified oncogenic factor in ATLL, was found to be a direct ETS1 transcriptional target in NA-ATLL. As such, ETS1 provides an alternate mechanism to enhance CCR4 expression/activity in NA-ATLL, even in the absence of activating CCR4 mutations (CCR4 mutations were identified in 4 of 9 NA-ATLL cases). Taken together, this study identifies ETS1 as a novel dominant oncogenic transcriptional regulator in NA-ATLL. | ETS1 is a novel transcriptional regulator of adult T-cell leukemia/lymphoma of North American descent
Adult T-cell leukemia/lymphoma (ATLL) is an aggressive T-cell lymphoma associated with the human T-cell lymphotropic virus type 1 virus endemic in regions including Japan, the Caribbean islands, and Latin America. Although progress has been made to understand the disease, survival outcomes with current standard therapy remain extremely poor particularly in acute ATLL, underlying the need for better understanding of its biology and identification of novel therapeutic targets. Recently, it was demonstrated that ATLL of North American–descendent patients (NA-ATLL) is both clinically and molecularly distinct from Japanese-descendent (J-ATLL), with inferior prognosis and higher incidence of epigenetic-targeting mutations compared with J-ATLL. In this study, combined chromatin accessibility and transcriptomic profiling were used to further understand the key transcriptional regulators of NA-ATLL compared with J-ATLL. The ETS1 motif was found to be enriched in chromatin regions that were differentially open in NA-ATLL, whereas the AP1/IRF4 motifs were enriched in chromatin regions more open in J-ATLL. ETS1 expression was markedly elevated in NA-ATLL in both cell line and primary tumor samples, and knockdown of ETS1 in NA-ATLL cells resulted in inhibition of cell growth. CCR4, a previously identified oncogenic factor in ATLL, was found to be a direct ETS1 transcriptional target in NA-ATLL. As such, ETS1 provides an alternate mechanism to enhance CCR4 expression/activity in NA-ATLL, even in the absence of activating CCR4 mutations (CCR4 mutations were identified in 4 of 9 NA-ATLL cases). Taken together, this study identifies ETS1 as a novel dominant oncogenic transcriptional regulator in NA-ATLL.
T-cell lymphomas are a highly heterogeneous group of lymphomas with a generally poor response to standard therapy. Adult T-cell leukemia/lymphoma (ATLL) is a rare T-cell lymphoma that occurs in 2% to 5% of individuals infected with the human T-cell lymphotropic virus type 1 (HTLV-1) retrovirus., The acute and lymphomatous ATLL subtypes follow an aggressive clinical course with a dismal prognosis (overall survival < 1 year) despite intensive chemotherapy, whereas the less common chronic and smoldering subtypes are more indolent with favorable response to antiviral therapy. ATLL cases occur predominantly in HTLV-1 endemic regions, such as Japan, the Caribbean islands, and Latin America. Within the United States, New York City has disproportionately high ATLL incidence as it hosts a large population of Caribbean-born immigrants., Recent work has illuminated clinical and molecular differences between ATLL arising in patients of Japanese (J-ATLL) compared with North American (NA-ATLL) descent. Epidemiologic and genomic analysis of NA-ATLL cohorts reported a younger median age of diagnosis (40-50 years), larger proportion of aggressive subtypes (acute and lymphomatous; ∼90%), shorter overall survival (6 months), and greater frequency of epigenetic mutations (57%) compared with patients with ATLL born outside of North America.4, 5, 6, 7, 8, 9 However, despite disparate prognostic and mutation profiles, J-ATLL and NA-ATLL are currently treated similarly because of the limited understanding of NA-ATLL biology and pathogenesis. ATLL cells generally express Treg phenotypic markers,10, 11, 12 and tumorigenesis is thought to progress in HTLV-1–infected T cells through a series of epigenetic and genetic events. Interactions between HTLV-1 viral proteins and the epigenome further support a role for epigenetics in ATLL pathogenesis. HTLV-1 encodes several viral proteins, of which Tax and HBZ are thought to be critically involved in clonal expansion of infected cells and subsequent oncogenesis. Both Tax and HBZ interact with chromatin machinery to facilitate transcriptional and epigenetic alterations promoting tumorigenesis., This was illustrated by a recent study showing that HBZ interacts with an ATLL-specific BATF3 super enhancer to drive the BATF3/IRF4 transcriptional program that was critical to ATLL growth. However, epigenetic molecular profiling studies have been reported exclusively in J-ATLL models, whereas the epigenetic and transcriptional landscape of NA-ATLL remains unknown. Given the disproportionately high mutation rate of epigenetic regulators in NA-ATLL and the known role for chromatin regulation in ATLL pathogenesis, we sought to characterize chromatin accessibility and transcriptional regulation in NA-ATLL and compare it with J-ATLL. Combined assay for transposase-accessible chromatin (ATAC) and RNA sequencing on cell lines representing each subgroup demonstrated differential transcriptional regulation between J-ATLL and NA-ATLL. ETS1 was identified as a dominant and unique transcriptional regulator in NA-ATLL. We then confirmed expression and the oncogenic role of the transcription factor, ETS1, through analysis of primary tumor samples and in vitro assays.
NA-ATLL cell lines ATL13, ATL18, ATL21, and ATL29 were cultured in Iscove's Modified Dulbecco's Medium (IMDM) with 20% human serum (Valley Biomedical, Winchester, VA) and 100 U/mL interleukin 2 (IL2; BD Biosciences, San Jose, CA). Japanese ATLL cell lines were cultured in RPMI 1640 with 10% fetal bovine serum (Gemini Bio-Products, West Sacaramento, CA) in the absence (ATL34Tb- and ED40515-) or presence (ATL43T+ and ED41214+) of 100 U/mL IL2.
ATAC sample processing was performed on 25 000 cells per cell line by the Epigenomics Shared Facility at Albert Einstein College of Medicine using the Omni-ATAC protocol. Normal healthy T-cell ATAC-seq data for Teff (CD4+CD127+) and Treg (CD4+CD25+) cells with and without stimulation were obtained from GSE118189 (supplemental Table 1). Sequencing and analysis were performed by Genewiz (South Plainfield, NJ). All data were trimmed using Trimmomatic 0.38, and cleaned reads aligned to reference genome hg38 using bowtie2. Aligned reads were filtered using SAMtools 1.9, and polymerase chain reaction (PCR) or optical duplicates were marked using Picard 2.18.26 and removed. Before peak calling, reads mapping to mitochondria were called and filtered, and reads mapping to unplaced contigs were removed. MACS2 2.1.2 was used for peak calling to identify open chromatin regions. Valid peaks from each group were merged, and peaks called in at least 66% of samples are kept for downstream analyses. For each pairwise comparison, peaks from each condition were merged and peaks found in either condition were kept for downstream analyses. Reads falling beneath peaks were counted in all samples, and these counts were used for differential peak analyses using the R package Diffbind (Stark and Brown). DiffBind was used for differential region of interest detection using a false discovery rate (FDR) ≤ 0.05 as a cutoff.
Motif discovery was performed on cell line and normal T-cell ATAC-seq data. Top differentially accessible regions were filtered by (adjusted P < .05; fold change (fc) > 2) from each comparison using the MEME Suite tools: MEME-ChIP and DREME. Transcription factor binding prediction was performed using Tomtom against the JASPAR and ETS factors motif databases. Find Individual Motif Occurences (FIMO) was used to determine the percent of total sequences containing a given motif (P < .001), and Analysis of Motif Enrichment (AME) was used to determine the difference in motif enrichment between two groups (Fisher’s exact test).
Datasets and sample sources used in transcriptome analyses are listed in supplemental Table 1. Cell line RNA was isolated using Qiagen RNAeasy (Qiagen). RNA sequencing libraries were prepared using Stranded mRNA kit (Illumina). Sequencing (150 bp, paired end) and analysis was performed by Genewiz. Normal T-cell RNA sequencing data paired to ATAC-seq for normal T cells were obtained from GSE118165. For both cell line and normal T-cell samples, sequence reads were trimmed to remove possible adapter sequences and nucleotides with poor quality using Trimmomatic v.0.36. The trimmed reads were mapped to the Homo sapiens GRCh38 reference genome using the STAR aligner v.2.5.2b. Unique gene hit counts were calculated by using featureCounts from the Subread package v.1.5.2. Comparison of gene expression between sample groups was performed using DESeq2, and the Wald test was used to generate P values and log2 fold changes. RNA isoforms were visualized in IGV software (Broad Institute). NA-ATLL patient and normal donor peripheral blood mononuclear cell (PBMC) RNA-sequencing data were previously published, and differential analysis was performed using DESeq2. J-ATLL and normal CD4 T cell microarray data were accessed through GSE33615,, and data were analyzed using GEO2R.
Gene set enrichment analysis (GSEA) was performed as previously described. Briefly, genes were ranked for each comparison using the product of −log(P value) and the sign of fold change. The ranked dataset was then analyzed for enrichment of Hallmark and custom datasets using GSEA software (Broad Institute). ETS1 target genes were defined by the TRANSFAC curated transcription factor targets gene set for ETS1 (Harmonizome,), as well as a recently published ETS1 target dataset in T-ALL. Pathway analysis was performed on ETS1 target genes overexpressed in NA-ATLL compared with J-ATLL (fc > 0.5) and normal control cells (fc > 0.5; P < .1) using Ingenuity Pathway Analysis (IPA; Qiagen) software.
Mutations were detected in the previously published NA-ATLL RNA-seq dataset. SNP/INDEL analysis was performed using mpileup within the Samtools v.1.3.1 program followed by VarScan v.2.3.9. The parameters for variant calling were as follows: minimum frequency of 25%, P < .05, minimum coverage of 10, and minimum read count of 7. Manual review of mapped reads within the region of interest was also performed.
Chromatin immunoprecipitation sequencing (ChIP-seq) data for T-regulatory (Treg) cells were accessed through GSE43119 and the FANTOM5 project https://fantom.gsc.riken.jp/data/. H3K27ac, H3K4me1, and ETS1 ChIP-seq tracks were visualized using the UCSC Genome Browser. ETS1 ChIP-seq data for Jurkat cells were accessed through GSE17954.
Human ACtive Enhancer to interpret Regulatory variants (HACER) was used to annotate genomic coordinates with chromatin interactions and enhancer information. Identification of enhancers overlapping with specific genomic coordinates was assessed from Cap Analysis of Gene Expression data derived from a large range of cell types including T cells. Enhancer target genes were determined by FANTOM5 project.
Illustrator for Biological Sequences (GPS) was used to construct the ETS1 and CCR4 gene models.
Small interfering RNA (siRNA) targeting ETS1 or nontarget control (ON-TARGETplus SMARTpool, Dharmacon) was transfected by electroporation using Amaxa Nucleofector II (Lonza). Cells were harvested for functional assays, and knockdown efficiency was assessed by quantitative PCR (qPCR) 48 or 72 hours after transfection.
Equal number of viable cells were plated 24 hours after transfection. Cell viability was determined 48 to 72 hours after transfection using the CellTiter-Blue Cell Viability Assay (Promega).
Western blotting was performed on protein lysates as previously described. The following primary antibodies were used to detect proteins of interest: ETS1 (Proteintech, 66598-1-IG) and β-actin (Novus, NB600-501). Total protein was visualized using Revert Total Protein stain (Li-Cor). Proteins were visualized using Li-Cor IR-dye secondary antibodies on an Odyssey FC Imaging System (Li-Cor).
RNA was isolated using Qiagen RNAeasy (Qiagen) and reverse transcribed to cDNA using SuperScript III First-Strand Synthesis SuperMix (Invitrogen). qPCR was performed using EvaGreen qPCR Mastermix (Applied Biologic materials Inc) with the following primer sequences: ETS1 forward: GGCAGTTTCTTCTGGAATTA, ETS1 reverse: CACGGCTCAGTTTCTCATA; CCR4 forward: CTCTGGCTTTTGTTCACTGCTGC, CCR4 reverse: AGCCCACAGTATTGGCAGAGCA (Origene), ACTB forward: CATCCTGCGTCTGGACCT, ACTB reverse: TAATGTCACGCACGATTTCC. Results were normalized using ACTB and relative fold change was calculated using the ΔΔCt method.
Statistical analysis was performed using the Student t test (in vitro data) or univariate generalized linear model (GLM) models, as indicated, through SPSS (IBM, Armonk, NY) and JMP (SAS Institute Inc., Cary, NC) software. P values are adjusted for multiple testing where indicated in ATAC-sequencing, RNA-sequencing, and microarray comparisons.
To identify and compare chromatin accessibility profiles between NA-ATLL and J-ATLL, ATAC-sequencing was performed on 8 ATLL cell lines (4 NA-ATLL: ATL13, ATL18, ATL21, and ATL29; and 4 J-ATLL: ATL34Tb-, ED40515-, ATL43T+, and ED41214+) and compared with chromatin accessibility of regulatory CD4+ T cells (Treg; CD4+CD25+) and effector CD4+ T cells (CD4EF; CD4+CD127+) derived from healthy controls. The cell lines were chosen to represent the mutational spectrum of ATLL including mutations targeting JAK/STAT, TP53, EP300/epigenetic, and immune-related genes. Unsupervised clustering showed that ATLL cell lines clustered separately from normal cells (supplemental Figure 1A). In addition, NA-ATLL cell lines clustered separately from J-ATLL cell lines with the exception of the J-ATLL cell line, ATL-43T+, which clustered with NA-ATLLs. ATL-43T+ was retained in the J-ATLL group for all comparisons. Chromatin accessibility of stimulated Treg and CD4EFF cells were more closely correlated to ATLLs than unstimulated T cells (supplemental Figure 1A). Because ATLL is known to have a Treg-like immunophenotype, we chose stimulated Tregs as the normal control for subsequent comparisons.
Analysis of genes annotated to chromatin accessibility peaks, showed significant overlap between ATLL cell lines and stimulated Treg samples (72.8%). However, NA-ATLL and J-ATLL chromatin accessibility also mapped to 1,348 and 2,051 unique genes, respectively. Given these differences in global chromatin accessibility between NA-ATLL and J-ATLL, we investigated transcription factor binding motif enrichment in differentially accessibility regions. Differentially accessible regions between NA-ATLL and J-ATLL (n = 3876 regions) were mapped predominantly to gene promoters (29%), intronic (37%), and distal intergenic regions (28%; supplemental Figure 1B). Of the regions that were more accessible in NA-ATLL cell lines, the most significantly enriched motif, AGGAAGW, corresponded to the ETS family of transcription factors (Figure 1A). Based on prediction score and expression of ETS transcription factors predicted to bind to this sequence, ETS proto-oncogene 1 (ETS1) was presumed to be the dominant ETS transcription factor recognizing the enriched ETS motif in NA-ATLL (supplemental Table 2). This motif was also significantly more enriched in total NA-ATLL chromatin accessibility peaks compared with Treg and J-ATLL (Figure 1B). When restricted to promoter regions (2 kb ± TSS), enrichment of the RUNX1 motif was also identified in sites with increased accessibility in NA-ATLL (E-value = 0.029); however, the RUNX1 motif was not enriched in total NA-ATLL peaks compared with J-ATLL (supplemental Figure 2A). Of note, RUNX1 has been shown to interact with ETS1. In contrast, the chromatin regions with greater accessibility in J-ATLL cell lines were enriched in 4 motifs, of which AP1 and IRF4 motifs were the most significantly enriched (Figure 1A; suppl Figure 2C). These motifs were also enriched in total J-ATLL accessible regions compared with NA-ATLL and Treg cells (Figure 1B; supplemental Figures 2B and 3). This is consistent with a recent report in J-ATLL cell lines in which HBZ was shown to interact with the AP1 member, BATF3, and its heterodimer partner, IRF4. In this study, HBZ was shown to interact at the BATF3 locus superenhancer, driving expression of BATF3 and the downstream BATF3-IRF4 transcriptional program. Indeed, we observed chromatin accessibility of the same BATF3 intronic site in 3 of the 4 J-ATLL cell lines and significantly greater accessibility of this site in J-ATLL cell lines compared with NA-ATLL (FDR = 0.0013; Figure 1C). We then determined whether the ETS1 locus also had differential chromatin accessibility. Like BATF3, ETS1 contained differentially accessible intronic regions (Figure 1D). Peak 1 corresponds to a previously identified intronic enhancer that interacts with ETS1 (Figure 1E). Motif analysis of these regions identified enrichment of the ETS1 motif. Consistent with this, these regions are characterized by H3K3me1 and ETS1 binding in Treg ChIP seq (Figure 1F) and ETS1 binding in Jurkat ChIP-seq (supplemental Figure 4).
We then assessed RNA-sequencing data to determine the expression of these transcription factors in cell lines. ETS1 expression was significantly elevated in NA-ATLL cell lines compared with J-ATLL cell lines (2.8-fold; P = .02; Figure 2A) and was not increased by stimulation status in normal T cells (supplemental Figure 5). Analysis of transcriptome data showed the dominant isoform of ETS1 to be p51 (NM_005238; supplemental Figure 6) across all cell lines, and no nonsynonymous mutations or intron retention events were identified in ETS1. Cell line expression of ETS1 was confirmed by Western blot, with similar degree of upregulation in NA-ATLL (P = .03; Figure 2B; supplemental Figure 7). To confirm the differential regulation of ETS1 expression between NA-ATLL and J-ATLL, we analyzed primary tumor gene expression data. In PBMCs isolated from 9 patients with NA-ATLL and 4 healthy controls, ETS1 expression was significantly elevated in NA-ATLL (Figure 2C). Similar findings were observed in the comparison between NA-ATLL and healthy donor CD4 T cells (Figure 2C; logfc = 10.6; FDR < 0.001). In contrast, analysis of gene expression dataset of 52 J-ATLL tumors compared with healthy donor CD4 T cells showed downregulation of ETS1, (Figure 2D). Subtype distribution is known to differ between NA-ATLL and J-ATLL; therefore, we compared ETS1 across subtypes in the J-ATLL dataset. ETS1 expression was similarly downregulated in J-ATLL Acute subtype compared with CD4 T cells (supplemental Figure 8), and there were no subtype-specific differences in ETS1 expression between acute (N = 26), chronic (N = 20), and smoldering (N = 4) in a univariate analysis (P = .113; Figure 2E). There were insufficient numbers of each subtype for comparison within the NA-ATLL dataset (acute, N = 8; chronic/smoldering, N = 1). Given the previously shown association of ETS1 with the HTLV-1 viral protein, Tax, we segregated tumors into low and high proviral load (median cutoff) and compared expression of ETS1. Notably, ETS1 expression was higher (P = .0018) in the high proviral load subset (Figure 2F). Although the trend was maintained, significance was lost when comparing ETS1 between low and high proviral loads within the J-ATLL acute subtype (supplemental Figure 8B). Primary tumor expression also supported the upregulation of BATF3 in J-ATLL (supplemental Figure 9A). BATF3 was not different between NA-ATLL in PBMCs isolated from NA-ATLL patients and healthy controls but was significantly elevated in J-ATLL primary tumor cells compared with control CD4 cells (supplemental Figure 9B-C). Unlike ETS1, BATF3 was found to be significantly elevated in acute compared with smoldering ATLL subsets in a J-ATLL cohort (supplemental Figure 9D). In addition, BATF3 expression was slightly elevated in J-ATLLs with high vs low proviral load (P = .011; supplemental Figure 9E); however, this trend was lost when analyzed within the acute subtype only (data not shown). Together, these data support disparate transcriptional regulators between J-ATLL and NA-ATLL.
We next sought to characterize the function of ETS1 in NA-ATLL. ETS1 has been shown to act as an oncogene, promoting proliferation, invasiveness, migration, and chemoresistance in many tumor types, including lymphomas.45, 46, 47 GSEA confirmed upregulation of ETS1 target genes in NA-ATLL compared with J-ATLL (Figure 3A). Pathway analysis of ETS1 target genes upregulated in NA-ATLL identified lymphocyte proliferation as a top pathway (Figure 3B). To functionally assess the role of ETS1 in proliferation, siRNA-mediated knockdown of ETS1 was performed in NA-ATLL cell lines (supplemental Figure 10). Knockdown of ETS1 in ATL18, ATL21, and ATL29 cell lines resulted in 0.47-, 0.51-, and 0.57-fold cell growth relative to control siRNA cells, respectively (Figure 3C-E).
To identify potential target genes of ETS1 in NA-ATLL, we focused on genes experimentally demonstrated to be direct ETS1 target genes that were upregulated in NA-ATLL cell lines compared with J-ATLL cell lines (fc > 0.5) and in NA-ATLL primary tumor cells compared with normal PBMC/CD4 cells (log fc > 0.5; FDR < 0.1). For ETS1 target genes that were consistently upregulated in each comparison, we then assessed chromatin accessibility to identify genes with differential chromatin accessibility in NA-ATLL compared with J-ATLL cells. CCR4 was a primary candidate gene identified by this approach. A region of chromatin accessibility was identified exclusively in NA-ATLL overlapping the CCR4 TSS and containing multiple predicted ETS1 binding sites (Figure 4A-B). ETS1 has been shown to bind to the CCR4 TSS by ChIP-seq in Treg cells (Figure 4C) and Jurkat cells (supplemental Figure 11). CCR4 gene expression was found to be significantly elevated in NA-ATLL compared with J-ATLL cell lines and in NA-ATLL primary compared with normal CD4 T cells (Figure 4D). Moreover, a strong correlation was observed between CCR4 and ETS1 gene expression in both cell lines and primary tumor cells (Figure 4E). Together with chromatin accessibility encompassing ETS1 binding sites and demonstrated binding of ETS1 to the CCR4 TSS, these data strongly indicate CCR4 as a direct transcriptional target in NA-ATLL. To confirm this relationship in NA-ATLL, we assessed CCR4 expression after ETS1 knockdown and observed significant inhibition of CCR4 expression with loss of ETS1 in NA-ATLL cell lines (Figure 4F). GATA3 has been previously implicated as driver of CCR4 in J-ATLL and along with ETS1 has several DNA binding motifs in the CCR4 promoter region. Although GATA3 is elevated in NA-ATLL primary tumor compared with control cells, it is not correlated with CCR4 expression in NA-ATLL primary tumors (R2 = −0.14; P = .85; supplemental Figure 12). In contrast, and consistent with the prior reports in J-ATLL, GATA3 is strongly correlated with CCR4 expression in J-ATLL primary tumors (R2 = 0.54; P < .001; supplemental Figure 12). CCR4 is a well-characterized oncogene in J-ATLL that is recurrently mutated in ∼30% of J-ATLL. Expression of mutant CCR4 resulted in decreased receptor internalization in response to ligand and increased migration toward its ligands CCL17 and CCL22. However, mutational status of CCR4 in NA-ATLL has not been reported previously. We assessed CCR4 mutation status in NA-ATLL cell lines from RNA-sequencing data. All 4 NA-ATLL cell lines lacked CCR4 mutations (in contrast, exon 2 mutations have been reported in the 4 J-ATLL cell lines previously). We then used NA-ATLL patient RNA sequencing data (n = 9) to identify CCR4 exon 2 variants in this population. Similar to the frequency observed in J-ATLL, early stop/non-sense mutations in exon 2 were observed in 3 of the 9 cases (33%; supplemental Table 3). All 3 of these nonsense mutations were located in the C-terminal cytoplasmic region of the CCR4 protein. Two of these sites have been previously reported in ATLL (C329∗ and Y331∗). In addition to these early stop mutations, a missense mutation was also identified in exon 2, I226F (c.928A>T), in the transmembrane region. Functional implications of this mutation are unknown, and it has not been previously reported in the COSMIC or dbSNP databases. Germline and tumor DNA were not available to confirm variant calls in these cases. Expression of both ETS1 and CCR4 at the RNA level was higher in wild-type (n = 5) compared with mutated CCR4 cases (n = 4; Figure 4G-H, ETS1, P = .039; CCR4, P = .052). Put together, these data suggest that ETS1 transcriptional regulation of CCR4 may represent an alternate mechanism to enhance CCR4 expression/activity in NA-ATLL, in the absence of gain of function CCR4 mutations.
We report for the first time that chromatin accessibility of NA-ATLL is characterized by enrichment of ETS binding motifs and corresponding upregulation of ETS1 expression. Disparate chromatin accessibility was observed between J-ATLL and NA-ATLL cells with the BATF3/IRF4 pathway identified as a primary transcriptional regulator in J-ATLL. Analysis of primary tumor RNA sequencing data in both NA-ATLL and J-ATLL cohorts confirmed that ETS1 is elevated in NA-ATLL tumor cells compared with normal controls, whereas ETS1 expression is lower in J-ATLL compared with normal T cells. Functional assays revealed that ETS1 drives cell growth in NA-ATLL and is a direct transcriptional regulator for CCR4. Together, these data provide further support for NA-ATLL as a distinct entity with a unique molecular profile characterized by aberrant expression of ETS1 and identify ETS1 as a key transcriptional regulator of in NA-ATLL. ETS1 is an Ets family transcription factor expressed in multiple cell types, including T lymphocytes. Mice deficient for Ets1 have demonstrated a critical role for Ets1 in Treg development and function. Overexpression, association with poor prognosis, and oncogenic function of ETS1 has been shown in multiple malignancies. In PTCL, ETS1 expression has been reported to varying degrees across several subtypes, with the highest observed in extranodal NK/TCL and weaker positive staining in AITL and ALCL., In a Chinese population of PTCLs, ETS1 expression was not significantly associated with outcome, although ATLL cases were not included in this study. ETS1 has been shown to interact with the HTLV-1 viral protein, Tax,, and to cooperate with Tax to transactivate target genes. Moreover, ETS1 has been shown to bind and activate the HTLV-1 LTR, providing a possible mechanism for activation of latent HTLV-1., In addition to Tax, interactions have been reported between ETS1 and other transcription factors such as Runx1 and Notch to facilitate DNA binding., Of note, the RUNX1 motif was also identified in NA-ATLL cell lines suggesting potential for cooperation between ETS1 and RUNX1 in this context. Further work is required to investigate potential interactions between ETS1, HTLV-1 proteins, and other T-cell transcription factors, as well as role for other Ets family transcription factors. We identified CCR4 as a direct transcriptional target of ETS1 in NA-ATLL. This is consistent with a recent report of CCR4 as an ETS1 target gene in T-cell acute lymphoblastic leukemia. CCR4 is a chemokine receptor for ligands CCL17 and CCL22, expressed primarily on TH2 and Treg cells. CCR4 is frequently expressed in ATL and is associated with poor prognosis. HTLV-1 Tax has been shown to induce the CCR4 ligand, CCL22, in HTLV-1–infected cells, resulting in selective attraction of CCR4+ T cells. Both CCR4 and CCL22 are overexpressed in NA-ATLL compared with J-ATLL cell lines (data not shown). HBZ-induced GATA3 has also been shown to drive CCR4 expression in J-ATLL cells, and GATA3 and CCR4 are coexpressed in PTCL. Moreover, GATA3, Tax1, and ETS1/2 have been shown to interact to promote gene transcription in HTLV-1–transformed cells. Both Ets and GATA3 binding motifs are abundant within the CCR4 chromatin accessibility region. In J-ATLL primary samples and cell lines, both ETS1 and GATA3 are correlated with CCR4 expression. Whereas in NA-ATLL cell lines and primary samples, ETS1, but not GATA3 were correlated with CCR4, despite elevated GATA3 expression. CCR4 mutations have not been previously reported for NA-ATLL as CCR4 was not included in the original genotyping panel for this patient cohort. Recurrent gain of function mutations in CCR4 (29%) have been reported in J-ATLL and for all 4 J-ATLL cell lines used in this study., We found a similar rate of mutation in NA-ATLL with 3 of 9 cases harboring nonsense mutations located in the C-terminal cytoplasmic region and another case with a previously uncharacterized missense mutation in the transmembrane region. Expression of the truncated mutant CCR4 has been shown to result in decreased receptor internalization and increased migration in response to ligands., Further work is needed to comprehensively characterize CCR4 variant status in NA-ATLL. It is possible that ETS1-driven upregulation of CCR4 in NA-ATLL is a way to promote CCR4 function in the absence of a gain of function mutation. The prevalence of epigenetic mutations, in particular, EP300, are a differentiating feature of NA-ATLL. Robust chromatin accessibility differences were not observed on the basis of EP300 mutational status between EP300 mutant (n = 3) and wild-type (n = 5) cell lines. However, the study was not adequately powered for this comparison. Our finding of BATF3/IRF4 motif enrichment is consistent with a recent study showing that HBZ drove the BATF3/IRF4 transcriptional program through interactions with BATF3. This study by Nakagawa et al was performed using J-ATLL cell lines, and we show the intronic HBZ interacting site identified has increased chromatin accessibility in J-ATLL cell lines compared with NA-ATLL. These findings indicate differential expression and/or regulatory relationships between transcriptional regulators and HTLV-1 viral proteins in J-ATLL and NA-ATLL. The cause of the clinical and molecular differences between NA-ATLL and J-ATLL remains an important question. It is likely that these differences are being driven by the presence or interactions of various genetic and environmental factors. The geographic differences and specific roles of factors such as host genetic variants, HLA expression, route and timing of HTLV-1 infection, and presence of co-infections remain largely unknown and will be important to define. The role of HTLV-1 genotype is also an important consideration. Although the genetically stable Cosmopolitan subtype 1a is common in both Japan and North America, Cosmopolitan subtype 1a subgroup A (transcontinental) is dominant in North America, whereas subgroup B (Japanese) together with subgroup A are found in Japan. There is evidence that ATLL in South America, where Cosmopolitan subtype 1a subgroup A is dominant, has more aggressive clinical course and younger age of diagnosis compared with J-ATLL, although subtype distribution may favor lymphomatous rather than acute as observed in NA-ATLL. Of particular interest will be characterization of ATLL in patients of Melanesian descent with the more divergent HTLV-1c genotype and patients of African descent where HTLV-1b and a spectrum of HTLV-1a subgroups are endemic. It has been suggested that HTLV-1c may be less oncogenic than Cosmopolitan subtype 1a, but likely underreporting and differences in life expectancy and environmental factors, particularly in indigenous populations, have made this comparison difficult. Despite disparate prognostic and mutation profiles, J-ATLL and NA-ATLL are treated similarly because of the limited understanding of NA-ATLL biology. Our data demonstrate that NA-ATLL has a distinct molecular profile, characterized by a key role for the transcriptional regulator, ETS1. Further work is needed to validate this chromatin signature in primary tumors and to elucidate the clinical implications and association with response to treatment, including anti-CCR4 (mogamulizumab) to help guide therapeutic approaches for NA-ATLL.
Conflict-of-interest disclosure: The authors declare no competing financial interests. |
||
PMC9647781 | Till Jasper Meyer,Stephan Hackenberg,Marietta Herrmann,Thomas Gehrke,Magdalena Steber,Rudolf Hagen,Norbert Kleinsasser,Agmal Scherzad | Head and neck squamous cancer cells enhance the differentiation of human mesenchymal stem cells to adipogenic and osteogenic linages in vitro | 31-10-2022 | human mesenchymal stem cell,adipogenic differentiation,osteogenic differentiation,head and neck squamous cell carcinoma,tumor microenvironment | Human mesenchymal stem cells (hMSC) are multipotent cells with the ability to differentiate into a range of different cell types, including fat, bone, cartilage or muscle. A pro-tumorigenic effect of hMSC has been previously reported as part of the tumor stroma. In addition, studies have previously revealed the influence of hematopoietic and lymphoid tumors on hMSC differentiation to support their own growth. However, this possible phenomenon has not been explored in solid malignancies. Therefore, the aim of the present study was to investigate the effects of head and neck squamous cell carcinoma (HNSCC) lines Cal27 and HLaC78 on the induction of osteogenic and adipogenic differentiation in hMSCs. Native hMSCs were co-cultured with Cal27 and HLaC78 cells for 3 weeks. Subsequently, hMSC differentiation was assessed using reverse transcription-PCR and using Oil Red O and von Kossa staining. Furthermore, the effects of differentiated hMSCs on Cal27 and HLaC78 were examined. For this purpose, hMSCs differentiated into the adipogenic (adipo-hMSC) and osteogenic (osteo-hMSC) lineages were co-cultured with Cal27 and HLaC78. Cell viability, cytokine secretion and activation of STAT3 signaling were measured by cell counting, dot blot assay (42 cytokines with focus on IL-6) and western blotting (STAT3, phosphorylated STAT3, β-actin), respectively. Co-culturing hMSCs with Cal27 and HLaC78 cells resulted in both adipogenic and osteogenic differentiation. In addition, the viability of Cal27 and HLaC78 cells was found to be increased after co-cultivation with adipo-hMSCs, compared with that of cells co-cultured with osteo-hMSC. According to western blotting results, Cal27 cells incubated with adipo-hMSCs exhibited increased STAT3 activation, compared with that in cells co-cultured with native hMSCs and osteo-hMSCs. IL-6 concentration in the media of Cal27 and HLaC78 after co-cultivation with respectively incubation with conditioned media of hMSCs, adipo-hMSCs and osteo-hMSCs were also found to be increased compared with that in the media of Cal27 and HLaC78 cells incubated with DMEM. To conclude, HNSCC cell lines Cal27 and HLaC78 induced hMSC differentiation towards the adipogenic and osteogenic lineages in vitro. Furthermore, a proliferative effect of adipo-hMSCs on Cal27 and HLaC78 cells was revealed with STAT3 activation as a possible mechanism. These results warrant further investigation of the interaction between HNSCC cells and hMSCs, with focus on the mechanism underlying the differentiation of hMSCs. | Head and neck squamous cancer cells enhance the differentiation of human mesenchymal stem cells to adipogenic and osteogenic linages in vitro
Human mesenchymal stem cells (hMSC) are multipotent cells with the ability to differentiate into a range of different cell types, including fat, bone, cartilage or muscle. A pro-tumorigenic effect of hMSC has been previously reported as part of the tumor stroma. In addition, studies have previously revealed the influence of hematopoietic and lymphoid tumors on hMSC differentiation to support their own growth. However, this possible phenomenon has not been explored in solid malignancies. Therefore, the aim of the present study was to investigate the effects of head and neck squamous cell carcinoma (HNSCC) lines Cal27 and HLaC78 on the induction of osteogenic and adipogenic differentiation in hMSCs. Native hMSCs were co-cultured with Cal27 and HLaC78 cells for 3 weeks. Subsequently, hMSC differentiation was assessed using reverse transcription-PCR and using Oil Red O and von Kossa staining. Furthermore, the effects of differentiated hMSCs on Cal27 and HLaC78 were examined. For this purpose, hMSCs differentiated into the adipogenic (adipo-hMSC) and osteogenic (osteo-hMSC) lineages were co-cultured with Cal27 and HLaC78. Cell viability, cytokine secretion and activation of STAT3 signaling were measured by cell counting, dot blot assay (42 cytokines with focus on IL-6) and western blotting (STAT3, phosphorylated STAT3, β-actin), respectively. Co-culturing hMSCs with Cal27 and HLaC78 cells resulted in both adipogenic and osteogenic differentiation. In addition, the viability of Cal27 and HLaC78 cells was found to be increased after co-cultivation with adipo-hMSCs, compared with that of cells co-cultured with osteo-hMSC. According to western blotting results, Cal27 cells incubated with adipo-hMSCs exhibited increased STAT3 activation, compared with that in cells co-cultured with native hMSCs and osteo-hMSCs. IL-6 concentration in the media of Cal27 and HLaC78 after co-cultivation with respectively incubation with conditioned media of hMSCs, adipo-hMSCs and osteo-hMSCs were also found to be increased compared with that in the media of Cal27 and HLaC78 cells incubated with DMEM. To conclude, HNSCC cell lines Cal27 and HLaC78 induced hMSC differentiation towards the adipogenic and osteogenic lineages in vitro. Furthermore, a proliferative effect of adipo-hMSCs on Cal27 and HLaC78 cells was revealed with STAT3 activation as a possible mechanism. These results warrant further investigation of the interaction between HNSCC cells and hMSCs, with focus on the mechanism underlying the differentiation of hMSCs.
Head and neck squamous cell carcinoma (HNSCC) is the seventh most common cancer in the United States, with ~65,000 new cases in 2019 (1). Despite advancements in diagnostics and therapeutic strategies, the 5-year survival rate remains poor at ~50 % (2). All solid tumors typically consist of cancer cells and the surrounding, non-malignant tumor microenvironment (TME) (3). This TME consists of a mixture of the extracellular matrix, endothelial cells, fibroblasts, immune cells and human mesenchymal stem cells (hMSCs) (3). There is a complex interaction between tumor cells and the TME, which has the overall effect of regulating cancer cell proliferation and tumor growth (3). This interaction has been proposed to also regulate the potential for metastasis and resistance to cancer therapy (3). These hallmarks are predominantly mediated by cytokines, chemokines, growth factors and cell-cell contacts (4). Therefore, research focus is increasingly being placed on the TME to deepen the understanding in the complexity of interactions in addition to developing novel targeted therapies (5). One of the main components in this interaction between tumor cells and TME are hMSCs, which are pluripotent cells with broad self-renewal capacities (6) and are able to differentiate into osteogenic (osteo-hMSCs), chondrogenic or adipogenic (adipo-hMSCs) lineages (7,8). The possibility of in vitro expansion rendered this cell type to be major targets of scientific investigation, especially in the field of regenerative medicine and treatment of various diseases. For example autologous transplantation of adipose-derived mesenchymal stem cells were evaluated for the treatment of knee osteoarthritis (9). To date, >800 clinical trials exploring the therapeutic potential of hMSCs are already underway (https://clinicaltrials.gov/ct2/results?cond=&term=mesenchymal+stem+cell&cntry=&state=&city=&dist=). The differentiation capacity of hMSC can be modified by manipulating the profile of external factors, such as growth factors, activation and inhibition of signaling pathways or metabolic processes (10,11). Corre et al (12) previously characterized hMSCs isolated from healthy donors and from patients with multiple myeloma (MM). In total, 145 genes were found to be differentially expressed between MM and healthy hMSCs (12). Among these, 46% were involved in tumor-microenvironment cross-talk (12). Therefore, it was hypothesized that hMSCs can create a highly favorable niche for supporting the survival and proliferation of the MM cells (12). In another study, Fairfield et al (13) investigated the effects of MM cells on the differentiation capacity and gene expression profile of hMSCs. It was shown that MM cells altered the gene expression profiles of hMSCs (13). In addition, a marked increase in the expression of MM-supporting genes, including IL-6 and C-X-C motif chemokine ligand 12, was detected (13). This previous study also indicated that MM cells inhibited adipogenic differentiation whilst inducing the expression of senescence-associated secretory phenotype and pro-myeloma proteins including IL-6 and Cxcl12 (13). Battula et al (14) observed that acute myeloid leukemia (AML) can attract hMSCs through chemotaxis and subsequently induce osteogenic differentiation. Furthermore, it was shown that osteo-hMSCs could enhance cancer progression (14). Another previous study revealed that MSCs from the bone marrow of patients with primary myelofibrosis exhibited increased osteogenic potential ex vivo (15), which appeared to serve a particularly important role in the pathophysiology of this disease (15). However, to the best of our knowledge, the impact of osteogenic and adipogenic differentiation of hMSCs on solid tumors has not been previously investigated. Therefore, the aim of the present study was to evaluate the effects of the interaction between HNSCC and hMSCs in terms of the induction of differentiation and the effects of hMSC differentiation on cancer cell proliferation in vitro.
To ensure that the most important types of HNSCC tumors are adequately represented, the laryngeal tumor-based cell line HLaC78 and the tongue tumor-based cell line Cal27 were chosen. The HNSCC cell line HLaC78 was isolated from a larynx carcinoma of a male patient by Professor Hans-Peter Zenner in the Department of Oto-Rhino-Laryngology, Plastic, Aesthetic and Reconstructive Head and Neck Surgery of the University Hospital (Würzburg, Germany) (RRID: CVCL_6647) (16). The Cal27 cell line was first isolated from the tongue tumor of a 56-year-old male patient (17), which was purchased at American Type Culture Collection. The cells were cultured in RPMI-1640 (Biochrom, Ltd.) containing 10% fetal calf serum (FCS; Linaris Biologische Produkte GmbH), 1% penicillin and streptomycin (Sigma-Aldrich, Merck KGaA) at 37°C with 5% CO2. The medium was changed every 2 days. After reaching 70–80% confluence, cells were detached by trypsinization with 0.25% trypsin (Gibco; Thermo Fisher Scientific, Inc.), washed with PBS, counted before 1×106 cells were seeded into new 250-ml culture flasks. Cells in the exponential growth phase were used for subsequent experiments.
Bone marrow was donated by 10 voluntary patients (5 male and 5 female; mean age 63.2 years), who had undergone surgery in the Department of Orthopedics, Koenig-Ludwig-Haus (University Hospital Würzburg, Germany). All patients agreed by providing written informed consent. The present study was approved by the Ethics Committee of the Medical Faculty of the University of Würzburg (approval no. 91/19). Bone marrow was harvested from acetabular reaming material as waste material from patients undergoing total hip arthroplasty surgery at the Department of Orthopedic Surgery, under aseptic conditions, and patients with clinical signs of osteoporosis, cancer or infectious disease were excluded. hMSCs were isolated in accordance with the protocol of Lee et al (18), which was also described in detail previously (19). Briefly, hMSCs were isolated by Ficoll (density=1.077 g/ml; Biochrom, Ltd) density gradient centrifugation (30 min; at room temperature; 800 × g; brake and acceleration levels set to the lowest value). After centrifugation, a clear phase separation was observed with a clearly visible optical dense interphase containing mononuclear cells. Cells from this interphase were pipetted in a new 50-ml reaction tube and subsequently washed with PBS twice. Cell culture was performed in the expansion medium (DMEM-EM), which consisted of DMEM (Gibco; Thermo Fisher Scientific, Inc.) containing 4.5 g/l D-Glucose, 10% FCS (Linaris Biologische Produkte GmbH), 1% penicillin and streptomycin (Sigma-Aldrich; Merck KGaA), whereas incubation was at 37°C and 5% CO2. hMSC morphology was evaluated by capturing phase contrast images using an inverted light microscope at 100× magnification (DMI 4000b Inverted Microscope, Leica Microsystems GmbH).
According to the guidelines provided by the International Society of Cellular Therapy (ISCT), hMSC should be adherent to plastic surfaces and positive for the expression of surface markers CD105, CD90 and CD73 but negative for the expression of hematopoietic surface markers, including CD45 or CD34 (20,21). Furthermore, hMSC should demonstrate multipotency in vitro (20,21). Plastic adherence was assessed using inverted light microscopy at ×10-40 magnification (Leica DMI 4000b Inverted Microscope; Leica Microsystems GmbH). hMSC surface marker expression was evaluated by flow cytometry. After detachment, cells were washed with PBS and cultured with 5% FCS on ice for 1 h. Afterwards, hMSCs (1×106) were incubated with anti-CD90 (dilution 1:500; conjugate APC; cat. no. 559869; BD Biosciences), anti-CD73 (dilution 1:50; conjugate PE; cat. no. 550257; BD Biosciences), anti-CD45 (dilution 1:50; conjugate FITC; cat. no. 555482; BD Biosciences), anti-CD44 (dilution 1:50; conjugate FITC; cat. no. 555478; BD Biosciences) and anti-CD34 (dilution 1:50; conjugate PE; cat. no. 550761; BD Biosciences) antibodies for 1 h at 4°C and flow cytometric analysis was performed (BD FACSCanto™; BD Biosciences) and further analyzed by FACS Diva Software v5.0.3 (BD Biosciences).
The pluripotency of hMSCs was evaluated by staining. First, hMSC were cultured in osteogenic and adipogenic media. hMSC control was cultured in DMEM-EM medium. The osteogenic differentiation medium was comprised of DMEM-EM, supplemented with 10−7 M dexamethasone, 10−3 M β-glycerophosphate and 10−4 M ascorbate-2-phosphate (all Sigma-Aldrich; Merck KGaA). The adipogenic differentiation medium was comprised of DMEM-EM, combined with 10−7 M dexamethasone and 10−9 g/ml recombinant human insulin (Sigma-Aldrich; Merck KGaA). hMSCs incubated with the osteogenic medium were termed osteo-hMSCs whereas hMSCs incubated with the adipogenic medium were termed adipo-hMSC at 37°C with 5% CO2 for 3 weeks.
For the evaluation of the osteogenic differentiation, von Kossa and Alizarin-Red staining were performed to detect calcium mineral components. For von Kossa staining the cells were first washed with distilled water, incubated with 1% silver nitrate solution at room temperature (diluted in distilled water; cat. no. #7908.1; Carl Roth) under UV-light for 20 min, washed three times with distilled water, incubated with 5% sodium thiosulfate pentahydrate (diluted in distilled water; cat. no. #6516.0500; Merck KGaA), washed three times with distilled water, incubated with Nuclear Fast Red solution for 5 min at room temperature [5 g Aluminum sulfate hydrate (cat. no. #227617; Sigma-Aldrich; Merck KGaA) in 100 ml distilled water, 0.1 g Nuclear Fast Red (cat. no. #5188; Sigma-Aldrich; Merck KGaA)], washed three times with distilled water, incubated with ascending alcohol series and dried for microscopy. The Alizarin-Red stock solution was prepared by dissolving 2 g of Alizarin-Red S (cat. no. #K00332679; Merck KGaA) in 100 ml distilled water. The pH-value was adjusted at 4.1-4.3 by adding of glacial acetic acid (cat. no. #1000661000; Merck KGaA). For staining the cells were incubated with the stock solution for 5 min at room temperature. Before and after incubation the cells were washed with distilled water. Adipogenic differentiation was assessed using Oil Red O staining to reveal intracellular lipid droplets. For preparing of the Oil Red O staining stock solution 0.5 g Oil Red O (cat. no. #O0625; Sigma-Aldrich; Merck KGaA) was dissolved in 100 ml Propylenglycol (cat. no. #P4347; Sigma-Aldrich; Merck KGaA) at 60°C. For the staining procedure the cells was washed with distilled water, incubated with Propylenglycol for 5 min at room temperature, then with 60°C warm Oil Red O stock solution for 10 min, washed with Propylenglycol, washed three times with distilled water and stained with Mayers Hematoxylin-solution (cat. no. #1.09249; Merck KGaA) for 30 sec. Until microscopy the cells were covered with PBS. All images were acquired with a light microscope (DMI 4000b Inverted Microscope; Leica Microsystems GmbH).
RT-qPCR was used to verify hMSC differentiation. The following primers were chosen for osteogenic differentiation: i) Alkaline phosphatase (ALPL; cat. no. 4331182; assay ID, Hs01029144_m1); ii) osteocalcin (BGLP; cat. no. 4331182; assay ID, Hs01587814_g1); iii) collagen 1 (Col 1; cat. no. 4331182; assay ID, Hs0016004_m1); and iv) runt-related transcription factor 2 (RUNX-2; cat. no. 4331182; assay ID, Hs00231692_m1). The following primers were chosen for adipogenic differentiation: i) fatty acid binding protein 4 (FABP4; cat. no. 4331182; assay ID, Hs01086177_m1); ii) leptin (LEP; cat. no. 4331182; assay ID, Hs00174877_m1); and iii) lipoprotein lipase (LPL; cat. no. 4331182; assay ID, Hs00173425_m1). GAPDH (cat. no. 4331182; assay ID, Hs02758991_g1) was used as the housekeeping gene. All primers were purchased from Thermo Fisher Scientific, Inc., the primer sequences of which are not publicly available. RT-qPCR was performed as follows: For total RNA extraction from hMSCs an RNeasy Kit (Qiagen GmbH) was used according to the manufacturer's protocol. For reverse transcription, the isolated RNA was converted into cDNA using SuperScript™ VILO™ Master Mix (cat. no. #11755-500; Invitrogen; Thermo Fisher Scientific, Inc.). The following temperature protocol was used for reverse transcription: 25°C for 10 min; 42°C for 59 min; 85°C for 5 min; 4°C for 2 min. Subsequent qPCR was performed using SYBR Green Real-Time PCR Master Mix (Thermo Fisher Scientific, Inc.) in a StepOnePlus™ thermocycler system (Applied Biosystems; Thermo Fisher Scientific, Inc.). The first denaturation step was 10 min at 95°C. Afterwards the following thermocycling protocol was utilized for 40 cycles: 50°C for 2 min, 60°C for 1 min and 95°C for 15 sec. The 2−ΔΔCq method was applied to quantify the relative gene expression levels (22). The gene expression values are then normalized to those on of the hMSC control.
The co-culture experiments were performed in Transwell systems with a polyester membrane and pore size of 0.4 µm (Costar® Transwell®; Corning, Inc.). After seeding 60,000 hMSCs into the lower chambers of 12-well plates in DMEM-EM medium and microscopic confirmation of adherence, 60,000 HLaC78 and 60,000 Cal27 cells were seeded onto the Transwell inserts in DMEM-EM medium. hMSC without co-cultivation served as the control. Cells were kept in this co-culture system for 3 weeks at 37°C and 5% CO2. The DMEM-EM medium was changed every 2 days. After a period of 21 days, hMSC differentiation into osteogenic and adipogenic lineages were determined using staining and RT-qPCR. This experiment was repeated 10 times using hMSCs from all 10 different donors.
The viability of HLaC78 and Cal27 cells co-cultured with hMSCs, adipo-hMSCs or osteo-hMSCs was measured by counting the cells using an electronic cell counter (CASY Cell Counter; OMNI Life Science GmbH). Identical measurements were also performed after the cultivation of the two HNSCC tumor cell lines with the conditioned medium of hMSCs (hMSC-CM), adipo-hMSC (adipo-hMSC-CM) and osteo-hMSC (osteo-hMSC-CM) at 37°C for 3 days. Conditioned media were obtained after 3 days at 37°C of hMSC, adipo-hMSC and osteo-hMSC incubation with DMEM-EM. Before the conditioning process the differentiation media was removed.
The Human Cytokine Array C3 dot blot assay (cat. no. AAH-CYT-3-4; Raybiotech, Inc.) was used to measure hMSC, adipo-hMSC and osteo-hMSC cytokine secretion after incubation with their respective differentiation media for 3 weeks. After removing of the differentiation media hMSCs, adipo-hMSCs and osteo-hMSCs were first incubated with DMEM without any supplements. After a period of 48 h at 37°C, the supernatants of the hMSCs from the 10 patients were then collected and pooled. The cytokine profile was analyzed according to the manufacturer's protocol. The chemiluminescence was assessed using an X-ray film. Semi-quantitative detection of IL-6 concentration was performed by density measurements using the ImageJ software (version 1.52a; National Institutes of Health) in relation to the positive control dot density. According to the manufacturer's declarations, the signal of the positive control spots is associated with the amount of biotinylated antibody printed onto the array.
Supernatants of the Cal27 cell culture after co-culture with hMSCs or incubation with the hMSC-CM for 3 days at 37°C were collected and analyzed for IL-6 levels using the ELISA kit human IL-6 (cat. no. 950.030.096; Diaclone SAS). All experiments were repeated with hMSCs from seven donors. The plates were read out at 450 nm (Titertek Multiskan PLUS; Thermo Fisher Scientific, Inc.). The standard curve was created by recombinant IL-6.
Cal27 and HlaC78 cells were incubated with hMSC-CM with or without 5 µg/ml anti-IL6 (cat. No. MAB2061; R&D Systems, Inc.), adipo-hMSC-CM and osteo-hMSC-CM at 37°C for 2 days. After washing the Cal27 and HlaC78 cells with PBS, they were harvested using a cell scraper and dissolved in RIPA buffer (PBS containing 1% NP40, 0.5% sodium deoxycholate and 0.1% SDS) supplemented with 10 µg/ml phenylmethanesulfonyl fluoride. Protein determination was performed by bicinchoninic acid method (Pierce BCA Protein Assay Kit; cat. no. #23227; Thermo Fisher Scientific, Inc.) Equal amounts (20 µg) of the total protein lysates were separated in a 10% SDS-polyacrylamide gel, before they were transferred onto polyvinylidene difluoride membranes. The membranes were blocked for 1 h at room temperature with TBST (10 mM Tris, 150 mM NaCl and 0.05% Tween-20, pH 8.0) containing 5% non-fat dry milk. The membranes were next incubated with the primary antibodies against phosphorylated (p-) STAT3 (1:1,000; rabbit; cat. No. 9145; Cell Signaling Technology, Inc.), STAT3 (1:2,000; rabbit; cat. No. 12640; Cell Signaling Technology, Inc.) and β-actin (1:2,000; mouse; cat. No. MA5-15739; Thermo Fisher Scientific, Inc.) overnight at 4°C. The membranes were then washed with TBST (Tween 0.1%) and incubated with a species-specific HRP-conjugated IgG secondary antibody (1:10,000; cat. no. 7074; Cell Signaling Technology, Inc.) for 1 h at room temperature. The bands were visualized using a chemiluminescence system (iBright1500; Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol.
All data were transferred into standard spreadsheets. Differences between groups were examined for significance, with one-way ANOVA performed using GraphPad Prism 6.0 statistics software (GraphPad Software, Inc.). For post hoc testing Dunnett´s multiple comparisons test was used (Fig. 3), for multiple comparisons Tukey's test was used (Fig. 5, Fig. 6, Fig. 7). All results were presented as mean ± SD. P<0.05 was considered to indicate a statistically significant difference and marked with an asterisk.
The hMSCs exhibited a fibroblast-shaped morphology when observed using microscopy (Fig. 1A). The Oil Red O, von Kossa and Alizarin Red staining revealed characteristics of osteogenic and adipogenic differentiation in osteo-hMSCs and adipo-hMSCs, respectively (Fig. 1). The successful osteogenic and adipogenic differentiation of hMSC was verified by qPCR. To evaluate the extent of hMSC differentiation towards the osteogenic and adipogenic lineages, RT-qPCR was performed. hMSCs incubated in DMEM-EM without differentiation medium served as the control. After 1 week of incubation with either osteogenic or adipogenic media, the expression of adipogenic cell markers FABP4, LEP and LPL and osteogenic cell markers ALPL, BGLP, Col 1 and RUNX-2 were measured. Compared with that in the control group, adipogenic differentiation medium induced a 229.4-fold increase in FABP4 expression, a 275.4-fold increase of LPL expression and a 206.6-fold increase of LEP expression expression. In terms of osteogenic differentiation, there was a 4.2-fold increase in ALPL expression, a 3.7-fold increase in BGLP expression, a 1.5-fold increase in Col 1 expression and a 1.4-fold increase in RUNX2 expression (Table I). According to flow cytometry analysis, the hMSCs were found to express surface markers CD90, CD73 and CD44 (Fig. 2). By contrast, hematopoietic markers CD45 and CD34 could not be detected (Fig. 2).
Co-culturing hMSCs with Cal27 or HLaC78 increased the expression of osteogenic and adipogenic lineages markers measured by RT-qPCR, compared with that in control cells (Fig. 3). However there was only a slightly increase of adipogenic markers after co-culturing hMSCs with Cal27. Furthermore, compared with that in the control group, Oil Red O staining of hMSCs co-cultured with Cal27 and HLaC78 cells showed markedly higher lipid droplet production (Fig. 4A-C). According to von Kossa staining, the quantity of calcium deposits was only increased slightly after hMSC co-cultivation with Cal27, but was more notably increased after co-cultivation with HLaC78 cells (Fig. 4D-F).
After co-cultivation of hMSC, adipo-hMSC and osteo-hMSC with Cal27 and HLaC78 cells, the number of HNSCC cells were counted. The number of Cal27 cells was increased significantly after co-cultivation with adipo-hMSC and osteo-hMSC compared with that in the groups of Cal27 cells that were not co-cultured (Fig. 5). Furthermore, there was an increased Cal27 cell count after co-cultivation with adipo-hMSC compared with hMSC (Fig. 5). In addition, the count of viable HLaC78 cells was significantly higher after co-cultivation with adipo-hMSC (Fig. 5). However, co-culturing with undifferentiated hMSCs did not alter the number of Cal27 and HLaC78 cells compared with that in monoculture cells (Fig. 5). Furthermore, there was no significant difference in the viability of HLaC78 cells after co-cultivation with osteo-hMSCs (Fig. 5).
Following the treatment of Cal27 and HLaC cells with media conditioned by hMSCs, increased cell viability was observed. Adipo-hMSC-CM treatment significantly enhanced tumor cell viability compared with that in cells treated with osteo-hMSC-CM (Fig. 6). There was no statistically significant difference in the cell count compared to the incubation with the control groups DMEM or hMSC-CM (Fig. 6).
Dot blot assay was used to investigate the profile of cytokine secretion in hMSCs, adipo-hMSCs and osteo-hMSCs. Due to the high expression level and their central role in tumour growth stimulation and inflammation, IL-6 was chosen for further detailed analysis. The secretion of IL-6 by adipo-hMSCs was markedly higher compared with that by osteo-hMSCs, but lower compared with in hMSCs (Fig. 7).
According to ELISA, IL-6 levels in the Cal27 cell culture supernatants were markedly increased after incubation with hMSC-CM and co-cultivation of Cal27 with hMSCs, compared with those in the supernatant of control Cal27 cells incubated with DMEM (Fig. 7F and G). In addition, comparably high levels of IL-6 were detected after incubation of Cal27 cells with hMSC-CM and in those co-cultured with osteo-hMSCs compared to Cal27 cells incubated with DMEM (Fig. 7F and G).
STAT3 activation at protein level was next evaluated by western blotting. Cal27 and HLaC78 cells were cultured with hMSC-CM, adipo-hMSC-CM and osteo-hMSC-CM. Furthermore to evaluate the influence of IL-6 on STAT3-activation, anti-IL-6 was added to hMSC-CM, adipo-hMSC-CM and osteo-hMSC-CM. The cultivation of Cal27 and HLaC78 cells with DMEM-EM served as the control. Markedly enhanced activation of STAT3 was observed after the treatment of Cal27 cells with CM compared with that in the control group (Fig. 8). However this was not observable for HLaC78 cells. The adipogenic or osteogenic differentiation had no influence on the level of STAT3-activation. The addition of anti-IL-6 reduced the STAT3 phosphorylation.
Tumors are comprised of malignant cells surrounded by a complex TME that contains different cell types, including fibroblasts, endothelial cells, hMSCs, innate and adaptive immune cells (23). A number of studies previously reported an important role of hMSC in cancer pathology, such as head and neck cancer (3,24). hMSCs consist of a heterogenic cell population with a range of properties, including migration towards wounds, immune modulation and enhancement of wound repair (25–27). Furthermore, hMSC have the reported ability to differentiate into cancer-associated fibroblasts (28–30). Mishra et al (30) demonstrated the trans-differentiation of hMSCs after exposure to breast cancer cell-conditioned media. In addition, another previous study showed that AML cells can induce chemotactic effects on hMSCs and osteogenic differentiation of these migrated cells (14). Osteogenic differentiation mediated an important impact on AML cell proliferation by an enhanced leukemia engraftment in a transgenic mouse model (14). However, to the best of our knowledge, no evidence of HNSCC-induced hMSC differentiation exists to date. In the present study, differentiation of hMSC towards both adipogenic and osteogenic lineages was shown after co-culturing with Cal27 and HLaC78 cells. However, spontaneous differentiation was previously described as an effect of hMSC aging in long-term cultures (31). Compared with that in control cells, without co-culturing with tumor cells, a higher rate of differentiation of hMSC into adipogenic and osteogenic cells in terms of morphology in addition to the expression of their markers, which was revealed by RT-qPCR. The effects of these differentiated hMSCs on tumor biology remains poorly understood. Tu et al (32) found an inhibition of the cancer cell survival by hMSCs in an osteosarcoma model though TGF-β/Smad2/3 signaling. In this previous study, an increase of VEGF- and IL-6-expession in hMSCs was observed (32). Furthermore, Paino et al (33) previously investigated the potential effects of SAOS2 and MCF7 cancer cell lines on hMSC differentiation. Neither alterations in the expression hMSC surface markers, including CD90, CD29 and vimentin, nor variations in the expression of transcription factors Twist and Slug, could be observed (33). However, an upregulation in the expression of stemness genes, such as OCT3/4 and Nanog, was observed (33). During the pathogenesis of breast cancer, adipocytes serve an important role (34). They are one of the main components of the breast microenvironment, where they have the ability to provide pro-tumorigenic signals (34). In the present study, differentiation of hMSCs towards adipocytes led to an enhancement of HNSCC cell viability. Furthermore, an enhanced activation of STAT3 in Cal27 cells was found after cultivation with hMSC-CM, adipo-hMSC-CM or osteo-hMSC-CM. The STAT3-activation was reduced after adding anti-IL-6. A potential reason for this pro-tumorigenic effect of adipo-hMSC may be the paracrine secretion of IL-6. STAT3 is activated particularly by the IL-6 family of cytokines, which include IL-6, IL-8, IL-11 and Oncostatin (35). However, IL-6 is the most potent activator of STAT3 (36). Adipose tissue is a key source of IL-6, which produces 33% IL-6 found in the plasma (37). However, comparably low concentrations of IL-6 were found in the adipo-hMSC-CM and in the supernatant of Cal27 cells co-cultured with adipo-hMSCs. Therefore, differences in STAT3 activation and cell viability could not be explained solely by effects mediated by IL-6. The effects of osteo-hMSCs on HNSCC cells were found to be ambiguous. Cultivation of Cal27 cells with osteo-hMSCs resulted in a positive effect on cell viability, evaluating the bi-directional effects, based on the reciprocal influences of hMSCs and tumour cells. However, no such effects could be detected on HLaC78 cells. This raised the question of whether cell viability was influenced by the cultivation of Cal27 and HLaC78 cells with osteo-hMSC-CM. No statistically significant effects could be detected after the cultivation of Cal27 and HLaC78 cells with osteo-hMSC-CM. One possible explanation could be the low concentrations of IL-6 in the osteo-hMSC-CM, compared with those in hMSC-CM and adipo-hMSC-CM media as shown in the dot blot analysis. However, these results remain ambiogious according to the IL-6 ELISA-measurements. Therefore, IL-6 alone is not sufficient to explain the differences in STAT3 activation and cell viability, in addition to differences in the osteogenic differentiated lineages. A large degree of variability was found in the present study, with high standard deviations in almost all data. Furthermore, there was ambiguous observations in only a slightly increase of adipogenic markers in qPCR, but a clear uptake of lipid droplets in the Oil Red O staining after co-culturing hMSCs with Cal27, and in addition only a slightly increase of ossification in the von Kossa staining and at the same time an increase of osteogenic markers in qPCR for Cal27. One reason for the mismatch of pPCR and morphology results could be that qPCR results represent the mRNA level and the mRNA level does not in every case correlate with the protein level. Another explanation for this finding could be the biological behavior of hMSCs in vitro. Despite characterizing hMSCs by their ability to adhere to plastic, cellular morphology and expression of different cell surface markers, these hMSCs remain highly heterogenic. This heterogeneity can be influenced by age, sex or the immune status of donors in addition to the culture conditions (38,39). Since the donors of hMSCs exhibited high variability in age, sex and immune status, this heterogeneity may have led to these ambiguous results. For future investigations the impact of the chrondrogenic differentiation lineage of hMSCs should be focused upon. Furthermore, the use of ≥ two different cell lines from a specific cancer would be beneficial to focus any studies into the molecular mechanism. In conclusion, data from the present study suggest that co-cultivation of hMSCs with Cal27 and HLaC78 cells can promote hMSC differentiation into adipogenic and osteogenic lineages. Furthermore, pro-tumorigenic effect of hMSCs differentiated towards adipogenic lineage was observed. One possible mechanism was the increased STAT3 activation in Cal27 and HLaC78 cells incubated with adipo-hMSC-CM. Therefore, further investigations into the underlying mechanisms are highly warranted. |
|
PMC9647786 | Jin Lee,Eun Mi Hong,Jung Han Kim,Jung Hee Kim,Jang Han Jung,Se Woo Park,Dong Hee Koh | Ursodeoxycholic acid inhibits epithelial-mesenchymal transition, suppressing invasiveness of bile duct cancer cells: An in vitro study | 26-10-2022 | bile duct cancer,ursodeoxycholic acid,epithelial-mesenchymal transition,E-cadherin,N-cadherin,epidermal growth factor receptor | Epithelial-mesenchymal transition (EMT) features are associated with pathological severity in the progression and metastasis of various cancer types, including bile duct cancer (BDC). Our previous study demonstrated that ursodeoxycholic acid (UDCA) blocked the EGFR-MAPK signaling pathway and inhibited the invasion of BDC cells. The present study was performed to determine whether UDCA inhibits EMT and promotes the expression of E-cadherin to inhibit the invasion and aggressiveness of BDC. In addition, the present study aimed to confirm that the primary mechanism of inhibition of EMT by UDCA is related to the EGFR axis. Human extrahepatic BDC cells were cultured. The effect of UDCA on cell proliferation was evaluated using MTT assays. A cell death ELISA kit was used to measure apoptosis, and western blot assays or immunofluorescence staining assays measured the expression levels of various target proteins. The mRNA expression of Slug and ZEB1 was evaluated via reverse transcription-quantitative PCR. The invasiveness of BDC cells was estimated by invasion assays and western blot assays for focal adhesion kinase (FAK). UDCA inhibited the proliferation of BDC cells as effectively as gefitinib (an EGFR inhibitor), and the combination of UDCA and gefitinib revealed an additive effect on the proliferation of cells. UDCA and gefitinib induced apoptosis, and the combination of UDCA and gefitinib demonstrated an additive effect on apoptosis in BDC cells. UDCA restored the E-cadherin expression inhibited by EGF and suppressed N-cadherin expression increased by EGF as effectively as gefitinib. UDCA suppressed the Slug and ZEB1 mRNA expression induced by EGF in BDC cells. UDCA suppressed the invasiveness of BDC cells and FAK expression linked to the invasiveness of BDC. In conclusion, UDCA enhanced E-cadherin expression and suppressed N-cadherin expression through inhibition of the EGF-EGFR axis, contributing to the inhibition of EMT and invasiveness in BDC cells. Therefore, UDCA may be applied as an adjuvant or palliative antineoplastic agent and as a therapeutic option to enhance the effect of other chemotherapeutics. | Ursodeoxycholic acid inhibits epithelial-mesenchymal transition, suppressing invasiveness of bile duct cancer cells: An in vitro study
Epithelial-mesenchymal transition (EMT) features are associated with pathological severity in the progression and metastasis of various cancer types, including bile duct cancer (BDC). Our previous study demonstrated that ursodeoxycholic acid (UDCA) blocked the EGFR-MAPK signaling pathway and inhibited the invasion of BDC cells. The present study was performed to determine whether UDCA inhibits EMT and promotes the expression of E-cadherin to inhibit the invasion and aggressiveness of BDC. In addition, the present study aimed to confirm that the primary mechanism of inhibition of EMT by UDCA is related to the EGFR axis. Human extrahepatic BDC cells were cultured. The effect of UDCA on cell proliferation was evaluated using MTT assays. A cell death ELISA kit was used to measure apoptosis, and western blot assays or immunofluorescence staining assays measured the expression levels of various target proteins. The mRNA expression of Slug and ZEB1 was evaluated via reverse transcription-quantitative PCR. The invasiveness of BDC cells was estimated by invasion assays and western blot assays for focal adhesion kinase (FAK). UDCA inhibited the proliferation of BDC cells as effectively as gefitinib (an EGFR inhibitor), and the combination of UDCA and gefitinib revealed an additive effect on the proliferation of cells. UDCA and gefitinib induced apoptosis, and the combination of UDCA and gefitinib demonstrated an additive effect on apoptosis in BDC cells. UDCA restored the E-cadherin expression inhibited by EGF and suppressed N-cadherin expression increased by EGF as effectively as gefitinib. UDCA suppressed the Slug and ZEB1 mRNA expression induced by EGF in BDC cells. UDCA suppressed the invasiveness of BDC cells and FAK expression linked to the invasiveness of BDC. In conclusion, UDCA enhanced E-cadherin expression and suppressed N-cadherin expression through inhibition of the EGF-EGFR axis, contributing to the inhibition of EMT and invasiveness in BDC cells. Therefore, UDCA may be applied as an adjuvant or palliative antineoplastic agent and as a therapeutic option to enhance the effect of other chemotherapeutics.
Bile duct cancer (BDC) is a malignant tumor with a 20~30% 5-year survival rate, even after resection, where most patients who cannot receive resection die within 2 years (1,2). This is because symptoms occur during the late stages of disease progression and an early diagnosis prior to metastasis, particularly to the lymphatic system, is challenging. Non-surgical palliative chemotherapy and radiation therapy may be considered, but the results have not been satisfactory (1). Ursodeoxycholic acid (UDCA), an endogenous hydrophilic bile acid, protects cells by inhibiting apoptosis in various cell types, such as hepatocytes. Activation of the EGFR/MAPK survival pathway, prevention of mitochondrial dysfunction and apoptosis, and minimization of the pro-apoptotic cascade activation are all known biological mechanisms that utilize UDCA to protect cells (3–5). UDCA is known to induce, rather than inhibit, apoptosis in malignant cells (6). In particular, UDCA induces potent apoptosis through BAX gene activation and BCL2 inhibition in hepatoma cells (6). Furthermore, a mouse model study demonstrated that UDCA inhibited hepatocellular carcinoma cell growth (7). UDCA inhibits signaling of EGFR and COX-2, blocking the tumorigenic effect caused by deoxycholic acid (DCA), thereby inhibiting the progression of colon cancer cells (4,8,9). A study was conducted on whether the effects of UDCA on apoptosis and growth in malignant and normal cells. This study showed that normal oral epithelial cells were not affected by UDCA treatment up to a toxic concentration, whereas apoptosis was stimulated in oral cancer epithelial cells proportional to the treatment concentration (10). Studies on whether UDCA decreases the incidence of BDC in high-risk groups are controversial. However, several epidemiological studies agree that long-term UDCA treatment lowers the incidence of cancer (11,12). Epithelial-mesenchymal transition (EMT) is a complex reversible process wherein epithelial cells increasingly change to the functional and structural properties of mesenchymal cells (13–15). Although it is the basis of physiological biogenesis and wound healing, EMT is also an early mechanism of metastasis and invasion at the primary site of tumor cells. The primary EMT mechanism alters gene expression to suppress the epithelial phenotype, activating the mesenchymal phenotype (16,17). In other words, the first step of EMT is the internalization and inhibition of E-cadherin, which induces the rupture of adherens junctions. After acquiring mesenchymal traits, EMT-transcriptional factors (ZEB1/2, Slug, Twis and Snail, etc.) regulate the expression of E-cadherin (14,15,18,19). Several studies have shown that EMT features was highly associated with pathological severity in terms of the progression and metastasis of BDC (20–25). Disappearance of epithelial markers (such as E-cadherin) and acquisition of mesenchymal markers (such as N-cadherin, S100A4, and Slug) were associated with aggressive characteristics of BDC, including metastasis, vascular and neural invasion, advanced tumor stage, and poor differentiation (20–22,24). EGFR activation is known to destabilize the E-cadherin/β-catenin complex in several tumors, thereby interfering with cell-cell adhesion, promoting EMT, and helping acquire a motile phenotype (25–27). Additionally, over-expressed EGFR is correlated with the tumor progression in BDC as well (28–32), and the EGFR axis triggers EMT in BDC cells, the most crucial step in the progression of the cancer (33). Recently, our studies demonstrated that UDCA suppresses the proliferation of BDC cells through the induction of apoptosis and inhibition of the EGFR-PI3K-Akt signaling pathway. Moreover, we found that UDCA blocks the EGFR-MAPK p42/44 (ERK1/2) signaling pathway and inhibits the invasion of the cancer cells (34). Accordingly, this study was conducted to determine whether UDCA inhibits EMT and promotes the expression of E-cadherin to inhibit the invasion and aggression of BDC. In addition, the primary mechanism of inhibition of EMT by UDCA, believed to be related to the EGF/EGFR axis, was investigated.
Fetal bovine serum (FBS), Roswell Park Memorial Institute (RPMI) 1640 medium, penicillin-streptomycin, trypsin, and sodium bicarbonate were supplied by Gibco. Dimethyl sulfoxide (DMSO), 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT), and UDCA were procured from Sigma Chemicals. Goat anti-rabbit IgG-horseradish peroxidase (HRP, Cat# sc2004) and human EGF were supplied by Santa Cruz Biotechnology. The Western Blot Hyper HRP substrate (Cat# T7103A) was procured from Takara. Gefitinib was procured from Roche Diagnostics. E-cadherin (Cat# 3195), N-cadherin (Cat# 4160), FAK (Cat# 3285), phosphorylated FAK (pFAK, cat# 3283), and β-actin (Cat# 4967) antibodies were purchased from Cell Signaling Technology.
The Korean Cell Line Bank (KCLB) supplied the SNU-245 cells (cat. #00245) obtained from distal common BDC presenting well-differentiation. They reported that the cells did not have any mutations in the genes of p53, p15, p16, hMLH1, and K-Ras. Moreover, the gene and mRNA of E-cadherin without mutation were found. The KCLB authenticated the absence of bacterial or mycoplasma contamination and the short tandem repeat (35). We cultured the SNU-245 cells in RPMI 1640 medium supplemented with 2 mM glutamine, 10% FBS, 1.5 g/l sodium bicarbonate, 100 µg/ml streptomycin, and 100 IU/ml penicillin. The media was refreshed twice a week and the cells were incubated at 37°C in a humidified incubator with 5% CO2. The cells were dislodged from the vessel using EDTA (1 g/l) and trypsin (2.5 g/l) when the cells were confluent.
MTT assays were performed to estimate cell proliferation as previously described (36). Briefly, cells were plated at a density of 5×104 cells/ml in RPMI regular media in 96-wells and incubated for 24 h. The various concentrations of UDCA then treated the cells within the serum-free medium (SFM) for 24 or 48 h. MTT (0.5 mg/ml) was then loaded in each well, and the cells were incubated at 37°C for an extra 4 h. After removing of the culture media, 100 µl of DMSO was added to each well. The colorimetric response was estimated using an ELX800 (Biotek) at 570 nm.
Cell apoptosis was estimated using the Cell Death Detection ELISA Plus Kit (Roche Molecular Biochemicals) that detects histone-associated deoxyribonucleic acid (DNA) fragments as previously described (36). Cells were plated at a density of 2×104 cells/ml in 96-well plates and incubated for 24 h. Various concentrations of UDCA were applied to the cells at 37°C for 24 or 48 h. After removing the media, 100 µl lysis buffer were loaded onto the cells for 30 min and followed by centrifugation at 200 × g at 4°C for 10 min. The supernatant was placed in the wells of a streptavidin-coated plate. The cell lysate was treated with the antibodies for DNA-peroxidase and histone-biotin, and then incubated for 2 h. After washing, 2,2′-azinobis-3-ethyl-benzothiazoline-6-sulfonic acid (100 µl) was incubated with each well for 20 min. A BioTek ELX800 microplate reader (BioTek Instruments) measured the absorbance at 405 nm.
The Annexin V-FITC/PI (fluorescein isothiocyante/propidium iodide) Apoptosis Detection kit. (Cat. #ab14085; Abcam) was applied to the identification of apoptotic cells. Cells were treated with indicated concentrations of UDCA or gefitinib at 37°C for 48 h. The cells were collected (2×106 cells) and washed once with PBS, and then, resuspended in 500 µl binding buffer (1×). The harvested cells were stained with Annexin V-FITC and PI for 20 min at RT in the dark. The stained cells were measured by CytoFLEX Flow Cytometer (Beckman Coulter), and the cell apoptosis rate was analyzed using CytExert software version 2.4 (Beckman Coulter).
Western blot assays were conducted as previously described (36). Briefly, cells were treated with various concentrations of UDCA, gefitinib, or EGF when the confluence reached 90% for 24 or 48 h. Cells were collected and washed with cold PBS (Gibco). Protein samples were extracted with RIPA buffer (Cat# R0278, Sigma Chemicals) and centrifuged at 15,000 rpm for 20 min. Bradford assays (Sigma-Aldrich; Merck KGaA) were used to estimate the amount of protein content of the cell lysate. Blots were blocked using a blocking solution at room temperature and incubated overnight at 4°C in 5% bovine serum albumin (BSA) solution with rabbit polyclonal diluted antibodies for N-cadherin (1:1,000), E-cadherin (1:1,000), focal adhesion kinase (FAK, 1:1,000), phosphorylated FAK (pFAK, 1:1,000), and β-actin (1:1,000). The nitrocellulose membranes were incubated with goat anti-rabbit IgG-HRP (1:8,000 dilution) after rinsing with TBS-T at room temperature for 1 h. The Luminata Forte Western HRP Western Blotting Detection Kit (Millipore Sigma) was used to detect the specific bands on the blots. Amersham image 600 system detected the bands automatically (Amersham Biosciences-GE Healthcare). The signal intensities of bands were measured with the ImageJ (Version 1.43; National Institute of Health).
BDC cells were lysed by adding 1 ml of Tri-reagent (Sigma-Aldrich; Merck KGaA) over 1 min. The lysate was the treated with 200 µl chloroform (Sigma-Aldrich; Merck KGaA) and incubated for 10 min. It was then centrifuged at 12,000 × g at 4°C for 10 min. An equal volume of isopropanol (Sigma-Aldrich; Merck KGaA) was treated to the supernatant and incubated for 15 min. The final product was centrifuged again at 12,000 × g at 4°C for 15 min. The pelleted RNA was rinsed with 70% ethanol and 50 µl nuclease-free-water (Roche Diagnostics) was added to dissolve the RNA, which was quantified using ELX800 (Biotek) at 260 nm.
After mixing 1 µg of RNA [extracted using RNA to cDNA EcoDry Premix (Takara)] and 20 µl of RT Master Mix, cDNA was synthesized under reverse transcription conditions (one cycle at 42°C for 60 min, one cycle at 70°C for 10 min, and 4°C for 10 min). After completion cDNA synthesis, 20 µl of cDNA was saved at −70°C until use.
Slug and ZEB1 gene expressions in the cDNA of SNU-245 cells were measured using relative quantitative PCR (LightCycler 480 using SYBR®-Green I Master, Cat. #5081963001, Roche Diagnostics). Relative quantitation of expression was determined by comparative Ct (2−ΔΔCt) method (37). PCR was performed under the following conditions: 1 cycle at 50°C for 2 min, 1 cycle at 95°C for 10 min, and 40 cycles of 95°C for 15 sec and 60°C for 1 min. Primer sequences were as follows: Slug, forward, 5′-ACACATTAGAACTCACACGG-3′, reverse, 5′-GAGAGACATTCTGGAGAAGG-3′; ZEB1, forward, 5′-ACCTGCCAACAGACCAGACAGTGT-3′, reverse, 5′-GCCCTTCCTTTCTGTCATCCTCCCA-3′; GAPDH, forward, 5′-GAGTCAACGGATTTGGTCGT-3′, reverse, 5′-GACAAGCTTCCCGTTCTCAG-3′.
Cells were cultured on glass coverslips and treated with 250 µM UDCA, 10 nM gefitinib, or 50 nM EGF for 24 h. Cells were fixed with 4% paraformaldehyde (Sigma-Aldrich; Merck KGaA) for 10 min, and then, cells were permeabilized with 0.3% triton x-100 in PBS for 10 min. Cells were incubated further with 10% goat serum (Cat. #sc-2043) (Santa Cruz Biotechnology) solution at room temperature for 1 h. Cells were incubated with an E-cadherin antibody (1:200 dilution, Cat. #3195; Cell Signaling Technology) in PBS containing 1% BSA (PBS-A) at room temperature. After reaction, this was incubated further with 1% albumin for 1 h and then goat serum solution at room temperature for 1 h. The cells were treated with a FITC-conjugated secondary antibody (1:500 dilution, Cat. #sc-36869, Santa Cruz Biotechnology) for 1 h in PBS-A and rinsed with PBS three times. The cells on coverslips were then treated with DAPI (0.5 µg/ml) (Sigma-Aldrich; Merck KGaA) for 1 min, and images were captured (×400 optical and ×3 digital magnification) using a super-resolution confocal laser microscope (Carl Zeiss).
Invasion assays were performed as previously described to evaluate the invasiveness of cancer cells (34). Briefly, we coated the upper membranes of cell culture inserts (Cat# 3401, Corning Incorporated) with Matrigel (Cat# A14132-01, Gibco) for 1 h. Serum-free regular medium (described in cell culture) of 200 µl was plated to the upper compartment, and 500 µl regular medium containing 10% FBS was added to the lower compartment. The cells were plated at a density of 2×104 cells/ml in the upper inserts and incubated at 37°C for 24 or 48 h. The upper membranes containing invading cells were fixed using 100% methanol for 20 min and stained for 15 min with 0.1% crystal violet (Sigma Chemicals) at room temperature. The upper surface of the inserts was washed in PBS, and noninvasive cells were wiped with cotton swabs. The membranes containing invading cells were mounted on slides, and light microscopy (100×, magnification, Olympus BX51-p polarizing Microscope) was used to count the number of cells present.
All experiments in this study were performed at least in triplicate. All described results were representative data and expressed as the means ± SD of duplicate cultures. The data were considered to follow parametric distribution after performing normality test (Skewness and Kurtosis statistics). One-way ANOVA followed by Tukey post hoc test for multiple comparison was used to compare three or more unpaired groups, and Student's t-test was used to compare two unpaired groups. P-values of less than 0.05 were considered statistically significant. IBM-SPSS version 27 (Armonk) was used as a statistical software.
Suppression of BDC cell proliferation by UDCA and gefitinib, a known EGFR inhibitor, was evaluated by an MTT assay after incubation for 24 or 48 h. Both gefitinib and UDCA treatment inhibited the viability of BDC cells in a dose- and time-dependent manner (Fig. 1A and B). The combination of UDCA and gefitinib for 24 or 48 h treatment demonstrated an additive effect, although not synergistic, on the proliferation of SNU-245 cells (Fig. 1C and D). A Cell Death Detection ELISA assay measured the effect of UDCA and gefitinib on apoptosis. UDCA and gefitinib induced significant apoptosis of BDC cells after 24 or 48 h of incubation in a dose- and time-dependent manner as well (Fig. 2A and B). The combination of UDCA and gefitinib for 48 h treatment also demonstrated an additive, but not synergistic, effect on the apoptosis of SNU-245 cells (Fig. 2C). We confirmed that UDCA and gefitinib induce significant apoptosis in BDC cells using Flow cytometry assays (Fig. S1). These results revealed that UDCA induced apoptosis and inhibited cell proliferation as effectively as gefitinib, and the combination of UDCA and gefitinib had an additive effect on apoptosis.
Western blot assays were conducted to evaluate whether UDCA activates E-cadherin (primary epithelial marker) and suppresses N-cadherin (primary mesenchymal marker) in BDC cells. The BDC cells were loaded with the indicated concentrations of UDCA and/or gefitinib and co-treatment with EGF in regular media containing 1% FBS for 48 h. EGF-only treatment, as a control, inhibited E-cadherin expression and increased N-cadherin expression (Fig. S2) in a time and dose-dependent manner, as was expected. Gefitinib or UDCA treatment restored the E-cadherin expression inhibited by EGF (50 ng/ml) and suppressed the N-cadherin expression enhanced by EGF in a dose-dependent manner as well (Fig. S3). Even though co-treatment with UDCA (250 µM) and gefitinib (10 nM) did not show synergistic restoration of E-cadherin expression decreased by EGF (50 ng/ml), co-treatment with UDCA (250 µM) and gefitinib (10 nM) synergistically suppressed N-cadherin expression increased by EGF (50 ng/ml) (Fig. 3). An immunofluorescence staining study was performed to confirm that UDCA activates E-cadherin expression in BDC cells. We treated SNU-245 cells with the determined concentrations of UDCA and/or gefitinib with co-treatment of EGF in regular media containing 1% FBS for 48 h. UDCA (250 µM) treatment restored E-cadherin expression inhibited by EGF (50 ng/ml) (Fig. 4), which was similar to what was observed for the western blot assay.
Here, we evaluated whether UDCA inhibits the mRNA expression of Slug and ZEB1, main EMT-transcription factors, using qPCR. Cells were loaded with the determined concentrations of UDCA (250 µM) and/or gefitinib (10 nM) with or without co-treatment of EGF (50 ng/ml) in regular media containing 1% FBS for 24 h. UDCA treatment significantly inhibited Slug and ZEB1 mRNA expression slightly less effectively than gefitinib (Fig. 5). Although co-treatment with UDCA and gefitinib did not show synergistic suppression of Slug mRNA expression increased by EGF (50 ng/ml), co-treatment with UDCA and gefitinib synergistically decreased the ZEB1 mRNA expression enhanced by EGF (Fig. 5).
Invasion assays were conducted to estimate the effect of UDCA on the aggressiveness on invasion and migration of BDC cells. The cells were seeded on upper inserts of Transwell® (Corning Incorporated) and treated with the indicated concentration of gefitinib (10 nM) and/or UDCA (250 µM) in SFM for 24 (not shown data) or 48 h. This experiment revealed that the invasiveness of bile duct cancer cells was significantly decreased after treatment with UDCA and was just as effective as gefitinib. In addition, the combination of UDCA and gefitinib had an additive or synergistic effect on the suppression of invasiveness of BDC cells (Fig. 6). Another western blot assay was conducted to evaluate the expressional change of pFAK, known to be positively associated with cancer metastasis and invasion (38), following treatment with gefitinib and/or UDCA in SFM for 24 h with pre-treatment of IGF-1 (100 nM) for 15 min. Both UDCA and gefitinib treatment inhibited the expression of pFAK enhanced by IGF. In addition, the combination of UDCA and gefitinib had an additive or synergistic effect on the suppression of pFAK induced by IGF (Fig. S4).
UDCA shows antineoplastic effects as a result of the induction of apoptosis, which has been demonstrated in several studies using cells and xenograft models of malignances (6,7). Recently, we proved that UDCA suppresses the proliferation of BDC cells via the induction of apoptosis and inhibition of the pathways of the EGFR-ERK and the PI3K-AKT, while blocking the invasiveness (34). Epidermal growth factor receptors (EGFR, HER-1, ErbB-1) belong to the tyrosine kinase receptor family. These growth factors, such as the epidermal growth factor (EGFR), bind at their extracellular binding domain, initiating intracellular signaling involved in stimulating cell proliferation, differentiation, and survival (39). Increased signaling from EGFR linked to its overexpression and mutation is associated with various cancers, including breast, colorectal, lung, head, neck, pancreatic, and BDCs (31,40,41). Enhanced expression of EGFR is known to contribute to poor prognosis in these cancers (28,42–45). EGFR expression in total cholangiocarcinoma ranged from 10.7 to 86% (31,46–48). Among them, EGFR in intrahepatic cholangiocarcinoma is positive in 43.3±30.6% (mean ± SD) (46), and extrahepatic BDC in Korea, where the prevalence rate is high, showed 86% positivity for EGFR (48). The prognosis in gallbladder cancer is also influenced by enhanced EGFR expression (49,50). Therefore, EGFR can be a therapeutic target for human cancer. ATP-competitive tyrosine kinase inhibitors, such as erlotinib or gefitinib, have increased the therapeutic efficacy for colorectal non-small cell lung, and pancreatic cancer treatments (51–53). In addition, studies have demonstrated that the inhibition of EGFR signaling by gefitinib effectively suppressed the proliferation of cholangiocarcinoma cells (29). The SNU-245 cells used in this study are extrahepatic bile duct cancer cells that exhibit EGFR expression (34). The aim of this study was to evaluate how effectively UDCA inhibits tumor cell proliferation compared to gefitinib (an EGFR inhibitor) and determine whether UDCA works synergistically with gefitinib compared to the monotherapy groups. Our results revealed that UDCA-induced apoptosis and inhibited cell proliferation as effectively as gefitinib and the combination of UDCA and gefitinib had an additive effect on apoptosis. Therefore, UDCA can be suggested as an antineoplastic agent with or without combination with known chemotherapeutics in BDC. EGFR activation is known to destabilize the E-cadherin/β-catenin complex in several tumors, thereby interfering with cell-cell adhesion, promoting EMT, and acquiring a motile phenotype (25–27,54), through the induction of adherens junction rupture. Once mesenchymal traits are acquired, EMT-transcriptional factors, such as ZEB1/2, Slug, Twis and Snail, modulate the expression of E-cadherin (14). Weakening of epithelial markers (E-cadherin) and obtainment of mesenchymal markers (N-cadherin, S100A4, and Slug) were associated with aggressive characteristics of BDC including metastasis, vascular and neural invasion, advanced tumor stage, and poor differentiation (20–24). In addition, Clapéron et al (33) proved an association between EGFR and EMT in cholangiocarcinoma by demonstrating that EGFR is a major factor in cancer progression by triggering EMT. As UDCA effectively inhibits EGFR in bile duct cancer cells, it has the potential to also inhibit EMT (34). In addition, if UDCA can properly inhibit the EGFR axis and EMT, there is a possibility that it may contribute to the inhibition of BDC progression by suppressing aggressiveness. In this study, UDCA restored E-cadherin expression inhibited by EGF and suppressed N-cadherin expression increased by EGF as effectively as gefitinib. UDCA also suppressed the expression of Slug and ZEB1 mRNA induced by EGF in BDC cells. These data implicate that UDCA suppresses EMT as effectively as gefitinib, through EGF-EGFR axis inhibition. We demonstrated that UDCA inhibits EMT and EGFR, which are directly linked to invasiveness and metastasis in BDC cells. Additionally, we performed invasion assays and western blot assays to evaluate the expressional change of phosphorylated FAK for the purpose of verifying the suppression of BDC cell invasiveness by UDCA. The invasion assays showed that UDCA suppresses invasiveness, and the combination of UDCA and gefitinib has a synergistic or additive effect on the suppression of BDC cells invasiveness. In addition, FAK is a significant regulator of signals mediated by the growth factor receptor and integrin and modulates basic processes in cancers. Enhanced FAK expression has been noted in various metastatic cancers and is associated with a grave prognosis. Therefore, FAK is regarded as a potential determinant of aggressiveness and metastasis (38,55). In this study, both UDCA and gefitinib treatment inhibited expression of pFAK enhanced by IGF. In addition, the combination of UDCA and gefitinib had a synergistic or additive effect on the inhibition of FAK induced by IGF. Accordingly, we suggest that UDCA-induced EMT suppression can be a significant determinant in regulating the invasiveness of BDC cells. As this study was a cellular-level in vitro study, there is a limitation in proving the actual anticancer effect of UDCA in animal and human BDC. Accordingly, we intend to conduct a study to investigate the effect of UDCA, with or without combination with other existing chemotherapeutics on EGFR/EMT, and antineoplastic effects using a xenograft animal model for BDC. In addition, we hope that various future practical studies will reveal the synergistic or additive effect of UDCA with known chemotherapeutics for BDC. On the other hand, SNU-245 cells, a human common BDC cell line presenting well-differentiation, was chosen for testing in this study although there are more types of BDC cell lines. which can be another limitation of our study. We wanted to evaluate wild BDC cells that express E-cadherin and do not have mutations of p53, p15, p16, hMLH1, and K-Ras to avoid lots of elements originated from mutations. In the future study, we hope we examine other BDC cell lines. In addition, 250 µM UDCA treatment in the media corresponds to the dose of 98.14 mg/Kg of bodyweight. Usual dose of UDCA in the patient with primary biliary cirrhosis is up to 15~20 mg/Kg, which means that 250 µM UDCA dose in our experiments was approximately 4.9-6.5 times higher than general therapeutic dose. Considering that we had to demonstrate definite change in experiments for short-term period (24 or 48 h) and prove anti-neoplastic effects, and that 25 or 50 µM UDCA (0.49~0.98 times of usual dose) was also effective on suppression of BDC cell proliferation, the concentrations we loaded may be acceptable. In conclusion, this study demonstrated that UDCA enhanced E-cadherin expression and suppressed N-cadherin expression, contributing to the inhibition of EMT and invasiveness in BDC cells, through inhibition of EGF-EGFR axis. Accordingly, UDCA may be applied as an adjuvant or palliative chemotherapeutic agent and as a therapeutic combination option that enforces the effect of other antitumor agents in BDC. |
|
PMC9647793 | Qiu-Chen Cai,Da-Lun Li,Ying Zhang,Yun-Yi Liu,Pei Fang,Si-Qin Zheng,Yue-Yan Zhang,Ya-Kun Yang,Chun Hou,Cheng-Wei Gao,Qi-Shun Zhu,Chuan-Hai Cao | Expression level comparison of marker genes related to early embryonic development and tumor growth | 26-10-2022 | tumor,embryo,developmental biology,evolutionary biology | In tumor research, the occurrence and origin of tumors are the fundamental problems. In the 1970s, the basic discussion of the developmental biology problem of tumors was proposed, and it was believed that tumorigenesis is closely related to developmental biology. Tumors are abnormal biological structures in organisms, and their biological behavior is very similar to that of the early embryo. Many tumor-related genes also serve regulatory roles in the normal development and differentiation of embryos. However, it remains unclear whether gene expression in early embryos has any similarities with tumor cells. In this study, to compare the similarities and differences in gene expression between early embryos and tumor cells, reverse transcription-quantitative PCR was conducted to determine and compare the relative expression levels of nine tumor-related genes in the brain glioma cell line, T98G, and in the early embryo of Spodoptera litura, which is fast-growing, low-cost, easily accessible and easy to observe. The expression of tumor-related genes in early embryos and the similarity of regulatory mechanisms between early embryonic development and tumor growth were explored. In conclusion, tumor growth may be regarded as an abnormal embryogenic activation that happens in the organs of adult individuals. | Expression level comparison of marker genes related to early embryonic development and tumor growth
In tumor research, the occurrence and origin of tumors are the fundamental problems. In the 1970s, the basic discussion of the developmental biology problem of tumors was proposed, and it was believed that tumorigenesis is closely related to developmental biology. Tumors are abnormal biological structures in organisms, and their biological behavior is very similar to that of the early embryo. Many tumor-related genes also serve regulatory roles in the normal development and differentiation of embryos. However, it remains unclear whether gene expression in early embryos has any similarities with tumor cells. In this study, to compare the similarities and differences in gene expression between early embryos and tumor cells, reverse transcription-quantitative PCR was conducted to determine and compare the relative expression levels of nine tumor-related genes in the brain glioma cell line, T98G, and in the early embryo of Spodoptera litura, which is fast-growing, low-cost, easily accessible and easy to observe. The expression of tumor-related genes in early embryos and the similarity of regulatory mechanisms between early embryonic development and tumor growth were explored. In conclusion, tumor growth may be regarded as an abnormal embryogenic activation that happens in the organs of adult individuals.
The treatment and prognosis of tumors has been a major focus of research in tumor-related studies, and the study of tumor pathogenesis and mechanisms of tumorigenesis is even more difficult than the study of tumor treatment. In the 1970s, Pierce and colleagues proposed that tumors are a developmental biology problem, and he believed that tumorigenesis is closely related to developmental biology (1,2). In 1892, Lobstein and Recamier presented a fundamental discussion on whether tumors are embryonic physiological disorders and their origin (3–5). They argued that tumors are formed by the continuous proliferation of embryonic cells stored in the body for a long time and that there is a high degree of similarity between tumors and embryos (6). Moreover, a study has shown that genes related to tumors can affect the normal development and differentiation of cells (7). Tumors are the products of embryonic gene expression and the result of the activation and expression of numerous oncogenes in the body (7). Early studies have confirmed that interconversion between tumor cells and early embryonic cells is possible under specific conditions (8,9). In 2000, Hanahan and Weinberg (2) proposed six major characteristics of tumor cells, including unlimited replication, tissue invasion, insensitivity to growth resistance, self-sufficient growth, evasion of apoptosis and sustained angiogenesis. Subsequently, in 2011, they added four more features of tumor cells, including genomic instability, promotion of inflammation, avoidance of immune response and energy dysregulation (10). These characteristics are very similar to the biological features of early embryonic cells. For example, gene methylation and demethylation, cell implantation, functional gene expression, cellular immune evasion and other such aspects in the early embryonic growth and development are strikingly like the biological function and behavior of tumor cells (11–14). Both embryonic and tumor cells can be deprogrammed to achieve a proliferative stem cell state with potential for apoptosis and invasiveness. Therefore, it is hypothesized that the set of genes expressed in tumor cells may be the same as those expressed in embryonic cells, particularly those genes involved in deprogramming, proliferation and undifferentiation (4,15–21). The present study focused on the similarity of gene expression related to early embryonic development and tumor growth. Nine factors highly associated with tumor regulation, including MYC, MYB, BCL-2, BCL-2-interacting protein 3 (BNIP3), p53, PTEN, PI3K, AKT and mTOR were selected as experimental research subjects. These nine factors occupy important positions in the large regulatory network of the body (22–27). Changing of their expression may lead to changes or even loss of control of the regulatory network (22,26–37). In addition, these factors are highly related to the development and growth, regulatory mechanisms and microenvironment maintenance of early embryos (4,6,17,30,38–42). MYC (29,31,43), MYB (26), BCL-2 (44–46), BNIP3 (47), p53 (27,48), PI3K (22,28), AKT (23,49) and mTOR (32,50–53) are regarded as proto-oncogenes that serve important roles in cell proliferation and differentiation, apoptosis, cell cycle regulation and metabolic processes. In previous studies of mouse embryonic development, high mRNA and protein expression levels of these proto-oncogenes were also detected in mouse embryos. The expression of these proto-oncogenes was highly associated with the successful implantation of fertilized eggs into the uterus, which could be determined by observing whether the mice became pregnant (6,30,38–42,45,54–56). The main function of the anti-oncogenic biomarker, PTEN, is to promote apoptosis and hinder cell proliferation, migration and local adhesion (57). Downregulation or loss of PTEN expression was found in a variety of tumors, such as non-small cell lung cancer and glioma (37,57). A previous study showed that PTEN has low expression levels in the zygote and blastocyst stages in early embryonic development of mice (58). Since the early embryos used in the present study were those before gastrulation, it was impossible to obtain human or large primate early embryos due to ethical restrictions. Furthermore, retrieving the early embryos of mice could be painful for the mice (59). Therefore, the insect model was chosen in the present study. Drosophila could be a good model as much of the research on humans has been conducted with Drosophila (60). However, the eggs of Drosophila are too small to clearly distinguish the embryonic development stage within them (61). Spodoptera litura is an omnivorous and gluttonous lepidopteran pest and is closely related to Drosophila (62). Their eggs are ideal for early embryonic studies, as they are flat and hemispherical, with a diameter of about 0.4-0.5 mm; they are yellow and white when they are newly laid, turning black before hatching, and the eggs are neatly stacked together (63,64). S. litura early embryonic development occurs between 1 and 8 h after egg laying, with the earliest divisions generally occurring at 2 h after egg laying (63,64). In the present study, the newly laid eggs of S. litura were used to analyze the expression of genes related to tumor metabolism, to understand the similarities and differences between these early embryos and tumors.
Experiments involved two different cell lines, T98G human glioma (65) and human astrocyte (HA) (66), which were provided by the Kunming Institute of Zoology, Chinese Academy of Sciences (Kunming, China). The original T98G cells were purchased from American Type Culture Collection (CRL-1690) and the HA cells were purchased from ScienCell Research Laboratories, Inc. (#1800). S. litura was from the Key Laboratory of the University in Yunnan Province for International Cooperation in Intercellular Communications and Regulations, Yunnan University (Kunming, China). S. litura breeding. S. litura were maintained in an artificial climate incubator in the following conditions: Temperature, 27±1°C; humidity, 60–80%; and 12-h light/dark cycle (67). Larvae were feed with artificial synthetic diet and adult moths were feed with 10% honey solution. The formulation of the artificial diet was the same as that used by the Key Laboratory of the University in Yunnan Province for International Cooperation in Intercellular Communications and Regulations (67). Provision of food and water was ad libitum. Both the larvae and adult moths were fed with a diet containing a large amount of water, so additional drinking water was not provided. The feed and honey solution were refreshed every 2 days. Larvae between first and sixth instar were kept in breathable boxes until pupation began. Larvae were transferred to a fine sand box for pupation. The pupae were separated from the male and female according to the position of the cloaca on the pupae. After emergence, the adult S. litura were placed in glass boxes with a 1:1 sex ratio to mate and lay eggs; they were fed with honey solution at the bottom of the boxes. The mating and spawning of the eggs were recorded. Two hours after oviposition, the eggs for RNA extraction were harvested and put into liquid nitrogen for preservation. The hemocytes of 3rd instar larvae were extracted for use as a control. The hemolymph was allowed to escape by puncturing the hematopoietic cavity from the larval gastropods and collecting it in a 1.5-ml Eppendorf tube containing 5 µl 5% reduced glutathione. After obtaining 1 ml hemolymph, it was gently pipetted and centrifuged at 10,000 × g for 5 min at 4°C. The precipitated fraction was hemocytes.
T98G and HA cells were thawed at 37°C, centrifuged at 500 × g for 1 min at room temperature to remove DMSO, and cultured as follows: The T98G cells were cultured at 37°C in a 5% CO2 atmosphere in Eagle's Minimum Essential Medium containing L-glutamine (Gibco; Thermo Fisher Scientific, Inc.) with the addition of 10% (v/v) fetal bovine serum (FBS) (Lonza Group Ltd.) and 1% (v/v) solution of penicillin and streptomycin (Gibco; Thermo Fisher Scientific, Inc.) (65). HA cells were cultured in HA medium containing DMEM/F12 (Gibco; Thermo Fisher Scientific, Inc.), 10% (v/v) FBS (Gibco; Thermo Fisher Scientific, Inc.), 2% B27 supplements (Gibco; Thermo Fisher Scientific, Inc), 3.5 mM glucose (Sigma-Aldrich), 10 ng/ml fibroblast growth factor 2 (Alomone Labs), 10 ng/ml epidermal growth factor (Alomone Labs), and 1% penicillin/streptomycin (Gibco; Thermo Fisher Scientific, Inc.) (66).
Total RNA samples for qPCR were extracted with RNA extraction kit (R6934-01; Omega Bio-Tek, Inc.), reverse transcription was carried out using a PrimeScript™ RT reagent Kit (RR047Q; Takara Bio, Inc.) according to the manufacturer's protocol and the concentration was determined. The relative expression levels of nine tumor-related genes and β-actin in early embryos and hemocytes of S. litura, the T98G and HA cell lines were measured by qPCR using an Applied Biosystems 7500 Fast Real-Time PCR System (Thermo Fisher Scientific, Inc.). The qPCR reaction procedure was as follows: 2 min at 50°C and 10 min at 95°C to activate the enzyme; 5 sec at 95°C and 35 sec at 60°C for 40 PCR cycles; 15 sec at 95°C, 1 min at 60°C, 30 sec at 95°C and 15 sec at 60°C to determine the melt curve. mRNA levels were quantified using the 2−ΔΔCq method (68) and normalized to the internal quantitative reference gene β-actin. The primers used for T98G and HA cells are listed in Table SI, and those used for S. litura are listed in Table SII.
Hatching rate measurement, head width measurement and ART treatment were conducted under S. litura breeding conditions. Newly laid eggs of S. litura were soaked for 10 sec at 27°C in 1 ml ART solution [300 ng/µl dissolved in 0.1% (v/v)] three times each day until eggs started hatching. As a blank control, newly laid eggs were soaked for 10 sec at 27°C in 1 ml H2O, and as a negative control, newly laid eggs were soaked for 10 sec at 27°C in in 1 ml 0.1% DMSO solution. ART itself is slightly soluble in water and 0.1% DMSO was added to the solution to increase the solubility of ART, so a 0.1% DMSO negative control group was set up in this experiment. The hatching rate of eggs was counted after 2 days of treatment. For the experiment on larvae ART treatment, 270 healthy newly hatched larvae were reselected and were divided into three groups. The groups of larvae were fed with a normal diet, a diet containing 0.1% DMSO and a diet containing 300 ng/µl ART, respectively, for 14 days until they pupated. During the feeding period, the width of the larvae's head capsule was measured once a day. All ART treatment experiments included three independent replicates.
The gene sequences of MYB, MYC, BCL-2, BNIP3, p53, PI3K, AKT, mTOR and PTEN of S. litura, Homo sapiens and 18 other invertebrates and vertebrates were downloaded from the NCBI database (https://www.ncbi.nlm.nih.gov/gene/?term=). The accession numbers of all downloaded sequences are listed in Table SIII. The FASTA file was analyzed with MEGA7.0 (69), and the systematic cluster tree was constructed by NJ.
mRNA expression levels between T98G cells and HA cells, and between early embryos and larval hemocytes were compared by unpaired Student's t-test. mRNA expression of H2O-treated, DMSO-treated and artemether (ART)-treated eggs was compared using one-way ANOVA and Tukey. The hatch rate of H2O-treated, DMSO-treated and ART-treated eggs was compared using Fisher's test. The 14-day measurements of larvae head capsule width in the H2O-treated, DMSO-treated and ART-treated larvae were compared using two-way ANOVA and Tukey. The 14-day measurements of larvae head capsule width were also compared within all three groups using two-way ANOVA and Tukey to confirm normal larval growth within the group (70). GraphPad Prism 9.0 (GraphPad Software, Inc.) was used for data analysis and graph plotting. P<0.05 was considered to indicate a statistically significant difference.
As shown in Fig. 1, MYB, MYC, BCL-2, BNIP3, p53, PI3K, AKT, mTOR and PTEN genes are related between the 20 species, indicating the evolutionary developmental conservation of the nine tumor-related factors.
The mRNA expression levels of eight oncogenes including MYB, MYC, BCL-2, BNIP3, p53, PI3K, AKT and mTOR in T98G and HA cells were detected (Fig. 2A-H). The results demonstrated that the mRNA expression levels of the eight oncogenes in T98G cells were significantly higher compared with that in the control group. The mRNA expression level of the anti-oncogene, PTEN, in T98G cells was significantly lower compared with that in the HA cell control group (Fig. 2I). These results demonstrated that seven of these nine genes may be excellent indicators of tumor cells. As soon as the expression of these marker genes is detected, it is possible to distinguish cells which are more like tumor cells, and which are not.
The mRNA expression levels of the eight oncogenes (MYB, MYC, BCL-2, BNIP3, p53, PI3K, AKT and mTOR) and the anti-oncogene (PTEN) were determined in S. litura early embryos and larval hemocytes (Fig. 3). The results demonstrated that the mRNA expressions of the eight oncogenes in early embryos were significantly higher compared with that in the larval control group (Fig. 3A-H). mRNA expression of the anti-oncogene, PTEN, in early embryos was significantly lower compared with that in the larval hemocytes control group (Fig. 3I). The expression levels of these oncogenes in early embryos of S. litura showed the same trend in T98G cells, suggesting that the metabolisms of tumor cells are more like S. litura early embryos than differentiated cells such as hemocyte.
In our previous studies, ART was demonstrated to exhibit excellent antitumor effects in vitro and in vivo via targeting several oncogenes and anti-tumor genes (36,71–75). The results demonstrated that the hatching rate of eggs was significantly decreased after ART treatment compared with H2O treatment and DMSO treatment (Fig. 4A). Furthermore, following ART treatment, the mRNA expression levels of PI3K, AKT, mTOR and p53 of early embryos were also significantly downregulated (Fig. 4B). However, when S. litura 1st instar larvae, which had completed their embryonic development and emerged from eggs, were fed the same concentration of ART solution for 14 days, the growth and development of larvae were not significantly affected (Fig. 4C). These results suggested that embryonic development may share some similarities with tumor cells in terms of gene expression, which can be altered by ART.
Malignant glioma is one of the most serious tumors, yet little is known about the pathogenesis of malignant glioma and other tumors (36). The causes of tumor formation are a matter of developmental biology. In our perspective of view, oncogenic factors in the environment and oncogenes in cells may initiate the rapid cell division, an ability obtained by cells after they became embryonic stem cells to ensure survival (76). In the absence of oncogenic stimulation, the rate of cell division would be reduced. Therefore, simply looking for carcinogenic factors and cancer suppressing drugs is not a solution to the cancer treatment problems such as side effects and drug resistance. From this perspective, the present study focused mainly on the similarity of the regulatory mechanisms between early embryonic development and tumor growth. A number of in vitro systems have been established for the study of malignant gliomas, including the well-known U87, U251 and T98G cell lines. The glioma cell line used in this study was T98G because, morphologically, it is a fibroblast and is more likely to form cell clusters, which are more similar to the cellular division and proliferation of early embryos (77). In addition, this study detected expression levels of p53. Therefore, T98G cells, which could stably express p53, was an excellent candidate (78). Since p53 in S. litura has two opposite functions of both promoting and inhibiting apoptosis (79,80), we hypothesized that it could be determined whether similar mutations had occurred in p53 in early embryos and hemocytes of S. litura using the sequences of wild-type and mutant p53 in T98G cells as a reference. However, as the mutation sites could not be accurately identified by Sanger sequencing, transcriptome, proteome and SNP analyses will be performed in future studies. Since the early embryos used in this study were those before gastrula, it was not possible to obtain human early embryos, as it is contrary to ethics and the original intention of treating diseases. The most commonly used human embryonic cell lines are not at this stage. Therefore, human embryonic cell lines were not selected. Retrieving the early embryos of mice could also be extremely painful for mice, which should be avoided. Other vertebrates such as chicken were also considered; however, due to the lack of suitable husbandry facilities and larger breeding sites, and the difficulty and expense of obtaining other vertebrate materials, an insect model organism was finally chosen for the present research. Therefore, we chose the insect S. litura, which is closely related to Drosophila, as it grows fast, is inexpensive and is easy to obtain. Besides, S. litura lay larger eggs and it is easy to observe the embryonic stage in the eggs with a low magnification dissecting microscope (62,67). However, S. litura is mainly used for specific innate immunity studies (67,70,79,81–83), and, to the best of our knowledge, this is the first time that S. litura is being used in oncology research. Therefore, organisms, such as Drosophila, are also included in the NJ tree to show that S. litura is related to Drosophila and could potentially be used in oncology research. Nine factors highly associated with tumor regulation, including MYC, MYB, BCL-2, BNIP3, p53, PTEN, PI3K, AKT and mTOR, were selected for investigation. These nine factors occupy very important positions in the large regulatory network as changes in their expression and function may lead to changes or even loss of control of the regulatory network (22–27). There is an obvious difference between the present and previous studies-the use of invertebrates as experimental materials (23,25–27,29,37,45,84) According to the NJ phylogenetic trees of 20 species presented in this study, these nine tumor-related key regulators are evolutionarily conserved in these species, suggesting that their functions may also be conserved. The results of the present study revealed high expression levels of MYC, MYB, BCL-2 and BNIP3 mRNA in T98G cells, which was also observed in early embryos. The oncogenes MYC and MYB may serve similar roles in the growth-promoting regulation mechanisms of early embryogenesis and tumor growth (25,26,29), and the oncogenes BCL-2 and BNIP3 may serve anti-apoptotic and microenvironmental roles in both early embryonic stage and tumor growth (25,46,85,86). This conclusion supports the embryogenic concept of a tumor and indicated that functional genes that serve a dominant role in the tumor and early embryos may be identical. The high mRNA expression level of the BNIP3 in early embryos and tumor cells suggested that the microenvironment of the early embryo may be highly similar to that of a tumor. BNIP3 is a downstream target protein of hypoxia-inducible factor, HIF-1α (24). High expression of HIF-1α in hypoxic environments can directly promote the high expression of BNIP3. HIF-1α and BNIP3 serve important roles in the maintenance of the hypoxic microenvironment in early embryos and tumors (24,47,84). In a previous study, it was demonstrated that embryonic development and embryonic cell growth require a good healthy microenvironment rather than a hypoxic tumor-like microenvironment (87). Therefore, the high similarity of the microenvironment between early embryos and tumors suggests that there may be a high degree of similarity between the gene expression in early embryos and tumor cells, as the similar microenvironments are regulated by the similar expression of genes. p53 mRNA expression in the early embryos of S. litura was also examined. Previous studies have reported that bracovirus could upregulate p53 in S. litura larval hemocytes to induce apoptosis, suggesting that p53 in hemocytes might function similarly to human wild-type p53 (67,80). However, p53 mRNA expression in early embryos of S. litura and T98G cells were higher compared with larval hemocytes and HA cells, respectively, suggesting that p53 in early embryos, consistent with tumors, lost its apoptotic function and serves a role as growth promoter (48). The high expression of p53 in early embryos suggested similarities in the expression and function of p53 between early embryos and tumors. The expression of the anti-oncogene, PTEN mRNA in T98G cells and early embryos of S. litura exhibited the same trend of low expression, which suggested that PTEN is expressed at a very low level or not expressed at all in the early embryonic stage or in tumors, and the proapoptotic effect is inhibited. Previous studies demonstrated that PTEN expression was low in tumor cells compared with normal cells (37,57). The present results suggested that PTEN expression was also low in S. litura early embryos. The similarity of the early embryo and the tumor cell was further demonstrated by the low expression of PTEN in both compared with that in normal somatic cells. The high expression levels of the PI3K, AKT and mTOR in early embryos and T98G cells suggested that this signaling pathway may serve an important regulatory role in both, confirming the high similarity of signaling pathway regulation between early embryos and tumors. Previous studies have demonstrated that PI3K, AKT and mTOR expression was high in tumor cells compared with normal cells (23). The present results suggested that these genes were also highly expressed in S. litura early embryos. mTOR is an important node at which multiple signaling pathways intersect and is therefore an important link in the regulatory network (23,88,89). In the present study, the similarity in the expression trends of functionally important genes between the early embryos and tumor cells was discussed by comparing the mRNA expression levels of nine evolutionarily conserved tumor-associated regulatory factors in early embryos and T98G cell lines. In addition, when early embryos and developed larvae of S. litura were treated with artemether [It was considered a typical antitumor compound in our previous studies, which could cause apoptosis by inhibiting the expression of oncogenes such as mTOR and BCL-2 and increasing the expression of oncogenes such as PTEN in cancer cells; however, artemether has no effect on normal cells (36,72–75)], our results demonstrated that artemether killed the early embryos but not the larvae. In addition, the present study revealed that the expression of oncogenes was reduced, and the expression of the anti-tumor gene was increased in S. litura early embryos after treatment with artemether, and the hatching rate of eggs was reduced, and the mortality rate increased after treatment with artemether, which was similar to the increase in apoptosis of tumor cells after treatment with artemether (36). These data suggested that gene expression and metabolism of early embryos and glioma cells are extremely similar. The results of the present experiments preliminarily confirm the concept of the embryonic origin of tumors, and place tumors from a cancer-based perspective into the perspective of individual development and evolution, that is, tumor-related regulatory factors are used to protect and promote early embryonic development and ensure the normal growth of living individuals in the early stages of their development (4,5). However, towards the end of an individual's life, tumor-associated regulatory factors are reactivated to create an embryonic-like mechanism, which competes strongly with the host and eventually outcompetes the host (90,91). The tumor is a life-regulating mechanism that has evolved over a long period of time and has been selected by natural selection and is both the beginning and the end of life (91,92). The present study will help to reveal the gene expression regulating early embryonic development, expand the study of tumorigenesis, enrich the discourse that tumor-associated regulators are products of individual development and population evolution, and further contribute to the exploration of the nature of life and tumors and the complex relationship between them. |
|
PMC9647794 | Jinju Wang,Zhe Song,Li Ren,Bowei Zhang,Yun Zhang,Xianwei Yang,Tong Liu,Yi Gu,Chao Feng | Pan-cancer analysis supports MAPK12 as a potential prognostic and immunotherapeutic target in multiple tumor types, including in THCA | 26-10-2022 | MAPK12,pan-cancer,prognosis,immune,thyroid carcinoma | P38 mitogen-activated protein kinase (MAPK)12 (also known as P38 γ) is critical in the development and progression of various types of tumors. Despite the extensive literature on the subject, further studies are needed to elucidate its role in cancer progression. Here, a comprehensive bioinformatics analysis of a generalized cancer dataset was performed to explore the mechanism of MAPK12 regulation in tumorigenesis. Several tumor datasets and online analytical tools, including HPA, SangerBox, UALCAN, GEPIA2, STRING, ImmuCellAI, and MEXPRESS, were used to analyze the expression information on MAPK12 in several types of cancers. Western blotting and reverse transcription-quantitative PCR were used to verify the protein and mRNA expression levels of MAPK12, respectively, in human normal thyroid cells (HTORI-3) and thyroid carcinoma (THCA) cells. Cytotoxicity and EdU assays were used to verify the promoting effect of MAPK12 on cell proliferation in THCA cells. Analysis of several cancers found that MAPK12 was overexpressed in multiple cancer types. Upregulated MAPK12 mRNA expression levels were correlated with a worse prognosis in patients with several types of cancer. Cytotoxicity and EdU experiments showed that MAPK12 knockdown inhibited THCA cell proliferation. Gene Ontology-Biological Process and Kyoto Encyclopedia of Genes and Genomes analyses showed that the enrichment of MAPK12 genes was related to cell proliferation and the tumor immune microenvironment. These results showed that MAPK12 was closely related to the immune checkpoint, microsatellite instability, and tumor mutational burden and affected the sensitivity of the tumor to immunotherapy. This study showed that MAPK12 may be an immunotherapeutic and promising prognostic biomarker in certain types of tumors. | Pan-cancer analysis supports MAPK12 as a potential prognostic and immunotherapeutic target in multiple tumor types, including in THCA
P38 mitogen-activated protein kinase (MAPK)12 (also known as P38 γ) is critical in the development and progression of various types of tumors. Despite the extensive literature on the subject, further studies are needed to elucidate its role in cancer progression. Here, a comprehensive bioinformatics analysis of a generalized cancer dataset was performed to explore the mechanism of MAPK12 regulation in tumorigenesis. Several tumor datasets and online analytical tools, including HPA, SangerBox, UALCAN, GEPIA2, STRING, ImmuCellAI, and MEXPRESS, were used to analyze the expression information on MAPK12 in several types of cancers. Western blotting and reverse transcription-quantitative PCR were used to verify the protein and mRNA expression levels of MAPK12, respectively, in human normal thyroid cells (HTORI-3) and thyroid carcinoma (THCA) cells. Cytotoxicity and EdU assays were used to verify the promoting effect of MAPK12 on cell proliferation in THCA cells. Analysis of several cancers found that MAPK12 was overexpressed in multiple cancer types. Upregulated MAPK12 mRNA expression levels were correlated with a worse prognosis in patients with several types of cancer. Cytotoxicity and EdU experiments showed that MAPK12 knockdown inhibited THCA cell proliferation. Gene Ontology-Biological Process and Kyoto Encyclopedia of Genes and Genomes analyses showed that the enrichment of MAPK12 genes was related to cell proliferation and the tumor immune microenvironment. These results showed that MAPK12 was closely related to the immune checkpoint, microsatellite instability, and tumor mutational burden and affected the sensitivity of the tumor to immunotherapy. This study showed that MAPK12 may be an immunotherapeutic and promising prognostic biomarker in certain types of tumors.
The (MAPK) family is a highly conserved protein family, the members of which participate in several cytokine pathways (1–3). P38 has four isoforms encoded by different genes in mammalian cells: P38 α (MAPK14), P38 β (MAPK11), P38 γ (MAPK12), and P38 δ (MAPK13) (3). The MAPK signaling pathway partakes in numerous core biological functions such as in the regulation of cell proliferation, inflammation, survival, innate immunity, and other cellular processes related to cancer progression and development (4,5). Classical MAPKs primarily include P38, JNK1/2/3, and other subtypes, which have been studied in depth (6–8). Studies suggest that MAPK12 is expressed in multiple tissues and promotes tumorigenesis and tumor progression (9). For example, high MAPK12 expression promoted epithelial-mesenchymal transition (EMT) in breast cancer cells, and the downregulation of MAPK12 inhibited EMT (10,11). MAPK12 overexpression increased the number of cancer stem cells (CSCs), whereas MAPK12 knockdown decreased the proportion of CSCs in breast cancer cells (12). Chen et al (13) found that overexpression of MAPK12 enhanced the transformation to a malignant phenotype in renal cell carcinoma (RCC) cells, and MAPK12 may be a novel therapeutic target for the management of RCC. Here, systematic bioinformatics analysis was used to explore the functions and effects of MAPK12 in a variety of cancers. The significance of abnormal MAPK12 expression in multiple cancer types was comprehensively studied using mRNA expression analysis, patient prognostic indicators, functional analyses of MAPK12-related genes, tumor immunity, and methylation patterns. Additionally, the relationship between MAPK12 expression and thyroid carcinoma (THCA) proliferation was determined using cytotoxicity and EdU assays in vitro.
The mRNA expression profiles of MAPK12 in various normal tissues were obtained from the Human Protein Atlas (HPA) website (https://www.proteinatlas.org). Data sources are Tumor-Node-Metastasis standardized. The ‘Gene’ module of SangerBox (14), a web-based program (http://www.sanger box.com), was used to examine the mRNA expression levels of MAPK12 in normal and cancer tissues as reported by TCGA (https://cancergenome.nih.gov/abouttcga/overview). The GSE33630 (15), GSE27155 (16), and GSE65144 (17) datasets were downloaded to analyze the MAPK12 mRNA expression differences between normal thyroid cells and thyroid carcinoma cells.
The relationship between MAPK12 mRNA expression and overall survival (OS) in each tumor type in the TCGA database was analyzed using the GEPIA2 website (18). Patients were divided into MAPK12 low and MAPK12 high cohorts based on MAPK12 expression levels. Cox regression analysis was used in SangerBox to study the effects of MAPK12 expression on the OS and disease-free survival (DFS) of patients with different types of tumors.
First, the ‘similar Gene’ module in GEPIA was used to analyze the top 100 genes that were most commonly associated with MAPK12 pan-cancer, and we screened the top 50 genes. A network map of MAPK12 and the 50 genes was created using STRING (https://string-db.org/) (19). Gene Ontology (GO) (20,21) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (22) were used to perform enrichment analysis of these 100 MAPK12-related genes. The TCGA-THCA database was downloaded, and the genes that most significantly correlated with MAPK12 were screened based on thresholds of R>0.35 and P<0.05 using R-project (http://www.R-project.org/) and R studio (http://www.rstudio.com/). GO and KEGG enrichment analyses were performed for the screened differentially expressed genes using the GeneDenovo tool (https://www.omicshare.com/tools/).
The EPIC and QUANTISEQ (https://www.epicimmunea tlas.org) datasets from TCGA (https://icbi.i-med.ac.at/software/quantiseq/doc/index.html) were used for the immune analysis of all types of infiltrating immune cells. The relationship between the expression levels of MAPK12 mRNA and immune checkpoint (ICP), microsatellite instability (MSI), and tumor mutational burden (TMB) in different cancer types in TCGA were analyzed using the immuno-analysis module of the SangerBox website (http://vip.sangerbox.com/home.html). The ImmuCellAI (http://bioinfo.life.hust.edu.cn/ImmuCellAI#!/) portal was used to analyze the relationship between MAPK12 expression and immune-related cells in THCA using the dataset from TCGA. The P-values and partial correlation were obtained using the Spearman rank correlation test.
The MAPK12 promoter methylation level differences in tumor tissues and normal tissues were analyzed using UALCAN (http://ualcan.path.uab.edu). The transcripts per kilobase of exon model per million mapped reads (TPM) was used to normalize the methylation expression value of raw data from TCGA. The MEXPRESS website (https://mexpress.be/) (23) was used to obtain the DNA promoter methylation patterns of MAPK12 in THCA.
The human normal thyroid cell line HTORI-3 and human THCA cell lines (TPC-1, K-1, and HTH-83) were obtained from ATCC. All cell lines were tested for mycoplasma, and STR cell identification was performed. All cell lines used in this study were cultured in DMEM supplemented with 10% FBS (both from Thermo Fisher Scientific, Inc.) and maintained in an incubator at 37°C, supplied with 5% CO2, and 95% humidity.
The MAPK12 small interfering (si)RNAs were purchased from Shanghai GenePharma Co., Ltd. siRNAs were used to knock down the expression of MAPK12 in HTH-83 and K-1 cells. The pcDNA3.1-MAPK12 plasmid was purchased from Genewiz, Inc. and used to overexpress MAPK12. Transient transfection of 3 µl si-MAPK12 (20 µM) or 2 µl MAPK12 plasmid (1,000 ng/µl) was performed in 6-well plates with a cell density of 2×105 cells using Lipofectamine 3000 (Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. The empty vector (pcDNA3.1) and si-NC were also used as negative controls. Transfected cells were harvested 48 h after transfection for subsequent analysis and detection. The MAPK12 siRNA sequences were #1:'5′-AAGUAACACGCUUCCAUUCTT'3′ and #2:'5′-UACAAAAGGGUCUAUUUCCTT'3′; the si-NC sequence was sense: UUCUCCGAACGUGUCACGUTT and antisense: ACGUGACACGUUCGGAGAATT.
The total RNA was extracted using TRIzol® reagent (Thermo Fisher Scientific, Inc.). The GoScript RT system (Promega Corporation) was used to synthesize cDNA according to the manufacturer's protocol. The MAPK12 mRNA expression status was detected using GoTaq®qPCR Master Mix (Promega Corporation) on an ABI QuantStudio 3. The PCR system was 20 µl in total, including 2X SYBR Green qPCR Master Mix (10 µl; Bimake), cDNA (2 µl), primer mix (2 µl), DNase/RNase-free water (6 µl), and the following thermocycling conditions were used: 95°C for 3 min, 40 cycles of 95°C for 15 sec and 60°C for 1 min. The dissolution curve program was: 95°C for 15 sec, 60°C for 1 min, and 95°C for 1 sec. The relative expression levels of the target gene were calculated using the 2−∆∆Cq method (24). The oligonucleotide primers used for qPCR were: MAPK12 forward, 5′-CCCTGGATGACTTCACGGAC-3′ and reverse, 5′-GCTTCAGGTCCCTCAGCC-3′; GAPDH forward, 5′-GGTGGTCTCCTCTGACTTCAACA-3′ and reverse, 5′-GTTGCTGTAGCCAAATTCGTTGT-3′.
The plates were carefully washed twice with PBS. Total protein was extracted using RIPA buffer (Beijing Solarbio Science & Technology Co., Ltd.) containing protease inhibitors (Beijing Solarbio Science & Technology Co., Ltd.). The protein concentration was detected using a BCA kit (Beijing Solarbio Science & Technology Co., Ltd.) according to the manufacturer's instructions. The protein lysates (20 µg) were loaded on a 10% SDS gel, resolved using SDS-PAGE, and transferred to PVDF membranes (MilliporeSigma). Then membranes were blocked using TBS containing 5% BSA (cat. no. A7906, MilliporeSigma) at room temperature for 2 h followed by incubation with primary antibodies against MAPK12 (1:2,000; cat. no. 9212; Cell Signaling Technology, Inc.) or GAPDH (1:5,000; cat. no. 97166; Cell Signaling Technology, Inc.) overnight at 4°C, followed by incubation with the corresponding HRP-conjugated secondary antibody at 1 h (1:3,000; cat. no. 7074; Cell Signaling Technology, Inc.). The protein bands were visualized and detected using an enhanced chemiluminescence system (Bio-Rad Laboratories, Inc.). GAPDH was used as a loading control.
A cytotoxicity assay (Dojindo Molecular Technologies, Inc.) was used to detect cell proliferation. A total of 2×103 cells/well were plated in 96-well plates with 3 replicate wells/group. The treatment group was treated with siMAPK12 knockdown. After 0, 1, 2, 3, 4, and 5 days, 100 µl serum-free solution (ApexBio) containing 10% cytotoxicity reagent was added, and cells were further incubated at 37°C for 1 h. The optical density values were obtained at 450 nm using a microplate reader (Thermo Fisher Scientific, Inc.).
An EdU assay was used to detect cell proliferation. A total of 2×105 THCA cells/well (HTH-83, K-1, or TPC1) were plated in 96-well plates. After 24 h, the adherent cells were transfected. After 48 h, 100 µl EdU medium containing 10 µM (Guangzhou RiboBio Co., Ltd.) was added to each well for 2 h, and the culture medium was washed and fixed with 100 µl cell fixator (cat. no. P1110; Beijing Solarbio Science & Technology Co., Ltd.) at room temperature for 30 min. The fixed cells were permeabilized with 100 µl PBS containing 0.5% Triton X-100, followed by staining using an Apollo staining reaction solution (Guangzhou RiboBio Co., Ltd.) in the dark for 30 min at 37°C. A Hoechst 33342 reaction solution (100 µl 1×; Guangzhou RiboBio Co., Ltd.) was used for 10 min at 37°C. The dyed plate was placed under an inverted fluorescence microscope (×100 magnification) to obtain fluorescence images.
Statistical analyses were automatically calculated using the aforementioned online tools. Comparisons between two groups were made using an unpaired t-test. Comparisons between multiple groups were made by one-way ANOVA and comparisons between CCK8 groups were made using two-way ANOVA, followed by Bonferroni's post-hoc test. P<0.05 was considered to indicate a statistically significant difference.
The flowchart of this study is shown in Fig. 1. The HPA database revealed that MAPK12 mRNA expression was highest in skeletal muscle, followed by the tongue (nTPM >100; Fig. 2A). MAPK12 mRNA expression levels were detectable but low (nTPM <20) in most other normal human tissues (Fig. 2A). To understand and analyze the differences in mRNA expression levels of MAPK12 in normal tissues compared with the respective tumor tissues, the expression profiles of several cancers were obtained from TCGA and the differences in expression of MAPK12 mRNA in normal tissues and tumor tissues were determined. The mRNA expression levels of MAPK12 in all TCGA tumor datasets are shown in Fig. 2B. MAPK12 mRNA expression levels were higher in several TCGA tumor datasets compared with the corresponding normal tissues. The expression of MAPK12 mRNA was significantly higher in 12 cancers: Cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), head and neck cancer (HNSC), kidney renal clear cell carcinoma (KIRC), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), prostate adenocarcinoma (PRAD), rectum adenocarcinoma (READ), stomach adenocarcinoma (STAD), and THCA. However, MAPK12 mRNA expression was lower in kidney chromophobe (KICH), kidney renal papillary cell carcinoma (KIRP), and uterine corpus endometrial carcinoma (UCEC) (Fig. 2B). The expression levels of MAPK12 in THCA were further studied to add to the relevance of disciplinary research. First, three GEO datasets were obtained: GSE33630, GSE27155, and GSE65144. MAPK12 expression levels were higher in all three databases compared with the normal tissues (Fig. 2C-E). Three THCA cell lines and a normal thyroid follicular cell line were used to verify the expression of MAPK12 RNA and protein. The expression levels of MAPK12 protein and mRNA were higher in the THCA cell lines compared with the normal thyroid follicle cells (Fig. 2F-G). In conclusion, these results suggested that MAPK12 expression was upregulated in several tumors, including THCA.
First, the correlation between MAPK12 mRNA levels and prognosis using patient survival-related information obtained from TCGA was determined, and the OS curves were plotted. The results showed that higher levels of MAPK12 mRNA in multiple cancer types were associated with a poorer prognosis and shorter survival in patients with multiple tumors, including bladder urothelial carcinoma (BLCA), LIHC, mesothelioma (MESO), THCA, and uveal melanoma (UVM) (Fig. 3). Cox regression analysis was used to further analyze the relationship between MAPK12 mRNA levels with OS and DFS in tumor patients. The results showed that high mRNA levels of MAPK12 were associated with a shorter OS in the pan-kidney cohort (KIPAN), LAML, HNSC, LIHC, lung adenocarcinoma (LUAD), BLCA, LAML, COADREAD, COAD, ACC, MESO, THCA, and UVM, and a shorter DFS in KIPAN, STES, HNSC, BRCA, KIRP, BLCA, ACC, COAD, COADREAD, UVM, MESO, and THCA (Fig. S1A and B). Overall, the analyses suggested that high levels of MAPK12 mRNA were associated with a poorer prognosis pan-cancer, including in THCA.
To further examine the molecular biological mechanism of MAPK12 function in tumors, MAPK12 expression-related proteins were identified, a protein-protein interaction network was constructed, and functional enrichment analysis of the MAPK12 expression-related genes obtained above was performed. First, the interaction of the top 50 proteins associated with MAPK12 in the form of a network diagram was shown using STRING (Fig. 4A). Second, the top 100 genes with a significant correlation with the MAPK12 gene in the generalized carcinoma dataset from TCGA were determined using GEPIA2. Third, GO-biological process (BP) and KEGG functional enrichment analyses were performed using the top 100 positively related genes (Fig. 4B and C). The results showed that the top 100 genes were enriched in cell proliferation-related pathways and immune-related pathways, including ‘cell cycle process’ and ‘activation of immune response’. The THCA dataset from TCGA was downloaded and analyzed regarding MAPK12-related genes based on thresholds of R>0.35 and P<0.05. The related genes were analyzed using GO-BP and KEGG functional enrichment. The results were similar to that of the pan-cancer analysis. MAPK12 THCA was also enriched in cell proliferation, and in immune-related functions and pathways (Fig. 4D and E). Based on these results, it was speculated that MAPK12 promoted the development of tumors pan-cancer, particularly in THCA, by influencing cell proliferation and the tumor immune microenvironment (TIM). Therefore, the underlying mechanism was further examined.
Although the functional enrichment analysis results confirmed that MAPK12 promoted cancer progression in THCA, experimental verification was required to confirm the bioinformatics results. Three classical THCA cell lines, TPC-1, HTH-83, and K-1, were selected for the cell proliferation experiments. MAPK12 expression in HTH-83 and K-1 cell lines was knocked down using siRNA. The MAPK12 expression levels in the TPC-1 cell line were upregulated using a pcDNA.31-MAPK12 plasmid. The efficiencies of MAPK12 knockdown or overexpression were confirmed using RT-qPCR and western blot analyses (Figs. 5A, B, and S2A). Cytotoxicity and EdU assays were performed, and the results showed that knockdown of MAPK12 led to a significant decrease in cell proliferation, whereas overexpression of MAPK12 resulted in an increase in cell proliferation (Figs. 5C-F, S2B and C). Together, it was experimentally demonstrated that MAPK12 played a role in tumor occurrence and development by regulating cell proliferation.
The enrichment analysis of MAPK12-related genes also suggested that MAPK12 influenced the development of tumors by influencing the TIM. Therefore, the impact of MAPK12 on TIM pan-cancer and in THCA was assessed. First, the correlation between MAPK12 mRNA levels and the abundance of tumor immune cells that had invaded diffuse carcinoma tissues was determined. Tumor-infiltrating lymphocytes (TILs) are an important component of the TIM and are generally associated with the development of tumors. TILs are key predictors of metastatic lymph node status and prognosis in patients with cancer. First, the relationship between TIL abundance and MAPK12 mRNA levels was determined using the EPIC and QUANTISEQ datasets. MAPK12 mRNA levels showed significant correlations with multiple TIIs/TILs in THCA (Fig. 6A and B). MAPK12 expression levels also showed a positive correlation with various TILs in the TCGA-THCA dataset (Fig. S3). Macrophages are a significant constituent of the innate immune system and play an indispensable role in activating the body's first-line defense against infection and cancer (25). Therefore, the infiltration levels of macrophages were assessed. A positive correlation was obtained in the EPIC database between macrophages and MAPK12 mRNA expression in THCA. Macrophages polarize to antitumor M1 and protumor M2 macrophages. Therefore, the QUANTISEQ database was used to further analyze the levels of invading M1 and M2 macrophages (26) and found that the M2 invasion levels were positively correlated with the MAPK12 mRNA expression levels in THCA. Therefore, it was hypothesized that THCA tumor cells secreted chemokines to increase the infiltration of tumor-promoting M2 macrophages to increase the degree of tumor malignancy. We analyzed whether MAPK12 affected the sensitivity of cancer patients to immunotherapy. ICP proteins are often regarded as promising therapeutic targets in the field of cancer immunotherapy (27–29). Therefore, the relationship between ICP gene expression levels and MAPK12 in various cancer types were determined. MAPK12 mRNA levels were positively correlated with multiple ICP genes in THCA (Fig. 6C). Moreover, MAPK12 expression in THCA showed a positive association with several ICP receptors including HAVCR2, NCR3, CD27, CD40, CD47, CD48, PDCD1, TNFRSF18, TNFRSF4, CTLA4, and TNFRSF14, as well as ICP ligands such as CD274, CD86, ICOSLG, CD160, PVR, CD244, and LGALS9 (Fig. 7). TMB and MSI are also important indicators of whether patients with cancer will benefit from immunotherapy (30,31). The relationship between the expression levels of MAPK12 with TMB and MSI in the TCGA dataset was studied using SangerBox. The results suggested that MAPK12 affected the sensitivity of THCA cells to immune checkpoint inhibitor therapy (Fig. 6D and E). Therefore, it was hypothesized that MAPK12 altered the TIM in THCA tissues by regulating the expression levels of ICP receptors and ligands. These results suggest that MAPK12 mediates the activation of ICP genes and is thus an ideal target for immunotherapy in THCA patients. In conclusion, MAPK12 may affect the TIM by modulating the infiltration of immune cells within tumors and the sensitivity of multiple tumors to immunotherapy. Therefore, MAPK12 may serve as an immunotherapeutic target.
To further study the mechanism of abnormal MAPK12 expression, we also analyzed the DNA methylation patterns of the MAPK12 gene promoter. DNA methylation generally leads to increased expression levels, and upregulation of oncogenes promotes tumor development (32). The UALCAN online tool was used to explore methylation levels in the MAPK12 promoter region. MAPK12 promoter methylation levels were lower in BLCA, CESC, ESCA, READ, KIRC, KIRP, LUSC, TGCT, THCA, and UCEC (Fig. 8A). These results suggest that methylation of the MAPK12 promoter may lead to its upregulation in a variety of cancer tissues. Next, the MEXPRESS methylation analytical tool was used to analyze THCA promoter levels and the results showed that MAPK12 mRNA expression levels were negatively correlated with MAPK12 methylation levels in THCA. The mRNA levels of MAPK12 were negatively correlated with the MAPK12 methylation levels at probe ID: cg19816445 (r=−0.089, P<0.05), probe ID: cg21649580 (r=−0.109, P<0.001), probe ID: cg21028326 (r=−0.108, P<0.05), probe ID: cg02031597 (r=−0.090, P<0.05), and probe ID: cg17193921 (r=−0.175, P<0.0001) in THCA (Fig. 8B). Taken together, these results suggest that the carcinogenicity of high MAPK12 expression in multiple types of cancer was due to hypomethylation of its promoter, particularly in THCA.
There are four subtypes of p38 MAPK encoded by different genes in mammalian cells: P38α (MAPK14), P38β (MAPK11), P38γ (MAPK12), and P38δ (MAPK13) (33). P38 MAPKs exhibit different expression patterns in different tissues. P38α was detected in all cells and tissues, and P38β was specifically overexpressed in brain tissue, thymus tissue, and spleen tissues. P38β is expressed at low levels in several tissues, such as the adrenal gland, and it is not expressed in skeletal muscle. In contrast, P38γ is highly expressed in skeletal muscle, whilst being expressed at very low levels in other tissues (34–37). All P38 MAPKs are serine/threonine kinases that are activated by a variety of inflammatory factors in a variety of conditions. MAPK12, also known as P38γ, ERK6, and SAPK3, regulates some of the processes of malignant transformation in several human cancer cell lines, such as proliferation, cell cycle progression, and apoptosis (38,39). Several researchers found that MAPK12 promoted the development and progression of various types of cancer (40–42). However, the role of MAPK12 in THCA metastasis is not known. Therefore, data from TCGA was used to analyze the functional role of MAPK12 in various tumors, particularly THCA. The analysis performed in this study included the expression of MAPK12 at the RNA level and the effect of differential expression on prognosis, functional enrichment analysis of MAPK12-related genes, and further analysis of its effect on tumor cell growth and proliferation, and the TIM. The present study found that MAPK12 was overexpressed in several tumors. The mRNA and protein expression levels of MAPK12 were higher in THCA cell lines compared with normal thyroid follicles. Higher mRNA levels of MAPK12 were associated with a worse OS in KIPAN, LAML, HNSC, LIHC, LUAD, BLCA, LAML, COADREAD, COAD, ACC, MESO, THCA, and UVM, and a shorter DFS in KIPAN, STES, HNSC, BRCA, KIRP, BLCA, ACC, COAD, COADREAD, UVM, MESO, and THCA. GO-BP and KEGG enrichment analyses were performed using the MAPK12-related genes following analysis of the THCA data from TCGA, and the results showed that MAPK12-related genes were enriched in cell proliferation and tumor immune-related functions and pathways. MAPK12 is highly expressed in HNSC, and it promotes the proliferation of ESCC cells and prevents their apoptosis in vitro (24). Hou et al (25) found that MAPK12 expression was significantly elevated in human colorectal cancer tissues relative to the corresponding normal epithelial tissues, and it promoted the growth, proliferation, and migration of CRC cells whilst inhibiting cellular apoptosis via the direct phosphorylation of PTPH1. Xu et al (26) showed that MAPK12 was positively correlated with the grade of glioma and may be a tumorigenic factor that promotes the growth and progression of glioma. Based on these results, it was hypothesized that MAPK12 promoted the development of tumors pan-cancer, particularly in THCA, by affecting cell proliferation and antitumor immunity. The association between MAPK12 mRNA levels and immune cell infiltration based on the GO and KEGG results of MAPK12-related genes was assessed. As an important component of the TIM, tumor-infiltrating immune cells are generally associated with the occurrence, progression, treatment, and/or metastasis of tumors (43). The MAPK12 mRNA levels were significantly correlated with multiple TIIs/TILs in THCA. A positive correlation was observed between macrophage numbers and MAPK12 mRNA expression levels in THCA in the EPIC database. As macrophages can polarize to antitumor M1 and protumor M2 macrophages (27), the levels of invading M1 and M2 macrophages were analyzed and the results showed that the level of invading M2 macrophages was positively correlated with the expression levels of MAPK12 mRNA in THCA. Therefore, it was hypothesized that tumor cells secreted chemokines to increase the infiltration of tumor M2 macrophages in THCA, which increased the degree of tumor malignancy. Whether MAPK12 affected the sensitivity of cancer patients to immunotherapy was next assessed. ICP, MSI, and TMB analyses showed that MAPK12 may be an ideal target for the treatment of THCA patients, especially in immunotherapy. In conclusion, the results of the present study suggest that MAPK12 may be a promising prognostic marker and a potential factor for predicting sensitivity to immunotherapy in patients with malignant tumors, particularly THCA. |
|
PMC9647812 | 36386630 | Xiaoxia Li,Zhengyuan Zhai,Yanling Hao,Ming Zhang,Caiyun Hou,Jingjing He,Shaoqi Shi,Zhi Zhao,Yue Sang,Fazheng Ren,Ran Wang | The plasmid-encoded lactose operon plays a vital role in the acid production rate of Lacticaseibacillus casei during milk beverage fermentation | 06-10-2022 | Lacticaseibacillus casei,acid production rate,comparative genome,lac operon,fermented milk-beverage | Lacticaseibacillus casei is used extensively in the fermented milk-beverage industry as a starter culture. Acid production capacity during fermentation is the main criterion for evaluating starters although it is strain-dependent. In this study, the acid production rates of 114 L. casei strains were determined and then classified into high acid (HC), medium acid (MC), and low acid (LC) groups. Comparative genomics analysis found that the lac operon genes encoding the phosphoenolpyruvate-lactose phosphotransferase system (PTSLac) were located on plasmids in the HC strains; however, it is notable that the corresponding operons were located on the chromosome in LC strains. Real-time PCR analysis showed that the copy numbers of lac operon genes in HC strains were between 3.1 and 9.3. To investigate the relationship between copy number and acid production rate, the lac operon cluster of the HC group was constitutively expressed in LC strains. The resulting copy numbers of lac operon genes were between 15.8 and 18.1; phospho-β-galactosidase activity increased by 1.68–1.99-fold; and the acid production rates increased by 1.24–1.40-fold, which enhanced the utilization rate of lactose from 17.5 to 42.6% in the recombinant strains. The markedly increased expression of lac operon genes increased lactose catabolism and thereby increased the acid production rate of L. casei. | The plasmid-encoded lactose operon plays a vital role in the acid production rate of Lacticaseibacillus casei during milk beverage fermentation
Lacticaseibacillus casei is used extensively in the fermented milk-beverage industry as a starter culture. Acid production capacity during fermentation is the main criterion for evaluating starters although it is strain-dependent. In this study, the acid production rates of 114 L. casei strains were determined and then classified into high acid (HC), medium acid (MC), and low acid (LC) groups. Comparative genomics analysis found that the lac operon genes encoding the phosphoenolpyruvate-lactose phosphotransferase system (PTSLac) were located on plasmids in the HC strains; however, it is notable that the corresponding operons were located on the chromosome in LC strains. Real-time PCR analysis showed that the copy numbers of lac operon genes in HC strains were between 3.1 and 9.3. To investigate the relationship between copy number and acid production rate, the lac operon cluster of the HC group was constitutively expressed in LC strains. The resulting copy numbers of lac operon genes were between 15.8 and 18.1; phospho-β-galactosidase activity increased by 1.68–1.99-fold; and the acid production rates increased by 1.24–1.40-fold, which enhanced the utilization rate of lactose from 17.5 to 42.6% in the recombinant strains. The markedly increased expression of lac operon genes increased lactose catabolism and thereby increased the acid production rate of L. casei.
Dairy protein-based beverages have received considerable research interest, because of the increasing consumption of health-promoting foods containing protein. The dairy beverages market is a very dynamic segment of the dairy industry, and the global dairy-based beverages market reached US$13.9 billion in 2021 (Abbas et al., 2022). Fermented milk beverages supplemented or enriched with functional ingredients such as bio-active peptides or probiotics occupy the largest segment within the dairy-based beverage market (Champagne et al., 2018; Fazilah et al., 2018). In addition to basic nutrition, fermented milk beverages offer health benefits to the consumer, e.g., prevention of digestive diseases, enhancement of immunity, and reduction of infection risk (Matsuzaki et al., 2004; Sanggaard et al., 2004; Shiby and Mishra, 2013). The requirements of the Chinese industry standard (NY1799-2004) for fermented milk beverages are based on probiotic microbial counts (>106 CFU/ml) and acidity, which must be greater than 25°T. The main function of the starter in milk fermentations is to produce organic acids by fermenting sugars. As the major acid end product, lactic acid not only provides the required acidity for fermented milk drinks, but also provides their characteristic taste, through its content and proportion (Leongmorgenthaler et al., 1991). Thus, the acid production rate is an important characteristic of a starter, because slow acid production may have major effects on the quality of the final product and increase the costs of industrial lactic fermentation. Fermented milk beverages containing probiotic strains are now well-established products in the global market; the fermentation is usually promoted by starter lactic acid bacteria (LAB) species such as Lactobacillus, Streptococcus, and Bifidobacterium (Fooks et al., 1999; Berhe et al., 2018). Lacticaseibacillus casei is an LAB that can be isolated from a variety of diverse habitats and has remarkable ecological adaptability, which contributes to the stability and persistence of acid production by L. casei during fermentation (Samet-Bali et al., 2010). L. casei has been isolated from dairy products, plant materials, as well as the human oral cavity and gastrointestinal tract (Cai et al., 2009). L. casei is mainly used in fermented milk beverages as a separate-starter, because of its efficient acid production; some examples of commercial fermented milk beverages contain strains of L. casei Shirota, L. casei Danone, and L. casei 01™ (Duar et al., 2017; Turkmen et al., 2019). The acid production rate during L. casei fermentation is influenced by several factors:
Almost all L. casei strains are able to ferment common hexose sugars, including glucose, fructose, mannose, and N-acetylglucosamine, as well as the disaccharides lactose, maltose, and cellobiose. The ability of L. casei to ferment other sugars is strain-dependent, their carbohydrate utilization being influenced by the environment they are isolated from (Ganzle and Follador, 2012).
As a facultatively hetero-fermentative species, the energy required for L. casei growth depends on the hydrolysis and metabolism of carbohydrates. Lactic acid production depends on glycoside hydrolase activity, which is highly variable (Hidalgo-Morales et al., 2005; Felis and Dellaglio, 2007). For example, the rapid hydrolysis of lactose implies the presence of a highly active β-galactosidase (EC 3.2.1.23) that enables L. casei ATCC334/64H to rapidly hydrolyze lactose into glucose and galactose-6-phosphate (Witt et al., 1993). The galactose-6-phosphate is metabolized to organic acids, via the tagatose-6-phosphate pathway, which is coded by the galR-galKTEM gene cluster (Tsai and Lin, 2006; Bidart et al., 2018). In addition, a trehalose 6-phosphohydrase (EC 3.2.1.93) of the GH13 family catalyzes the hydrolysis of trehalose, a β-glucosidase (EC 3.2.1.21) is the rate-limiting biocatalyst of cellobiose hydrolysis, and β-D-fucosidase (EC 3.2.1.38) is the rate-limiting biocatalyst for fucose hydrolysis (Ganzle and Follador, 2012; Buron-Moles et al., 2019). These glycoside hydrolases increase the ability of L. casei to hydrolyze various different carbohydrates and produce organic acids, to increase the acid production rate. This study aimed to group L. casei isolates by performing phenotypic analysis of acid production rate and exploring the genetic basis of lactose metabolism by different strains. The association between acid production rate and the lac operon was established, which provides a theoretical basis for screening of starter strains for rapid acid production, for use in fermented milk beverages.
The products from which the isolates came originated from 29 different regions in 12 provinces (Yunnan, Tibet, Guangxi, Gansu, Anhui, Shanxi, Guizhou, Guidong, Beijing, Chongqing, Xinjiang, and Inner Mongolia) in China between 2006 and 2020. Samples were derived from dairy products (fermented milk, koumiss, and qula), plant materials (pickles, wine), cured meat, and human isolates (vaginal, feces). Detailed information on samples and isolates is listed in Supplementary Table S1. Suitable dilutions of sample isolates were inoculated in triplicate on De Man, Rogosa, and Sharpe (MRS) agar containing the pH indicator bromocresol purple (BCP; Sigma-Aldrich, St. Louis, MO) and cycloheximide (0.01% v/v) to prevent fungal growth. After incubation at 37°C for 48 h under anaerobic conditions, colonies with distinct morphological differences (color, shape, and size), which were producing acid, according to the BCP, were selected and purified by streaking on MRS agar (supplemented with 10 mg/l vancomycin). Representative colonies were selected (~10% of the observed count) and tested for positive Gram staining and the absence of catalase activity. The screened isolates were stored in cryoprotectant (12% w/v skim milk containing 10% glycerol) and cultured in MRS broth (Oxoid) at 37°C overnight before further experimentation. Microbiological analyses of the 114 isolates were conducted using methods including conventional phenotypic identification, 16S rRNA sequence, and species-specific PCR (Supplementary Table S2; Liu et al., 2012; Colombo et al., 2018, 2020).
The milk fermentation medium was reconstituted skim milk powder (13% w/v; Fonterra™, Auckland, New Zealand) with sucrose (5% w/v) and glucose (3% w/v). The medium was two-stage (5ΜPa, 15 MPa) pressure-homogenized (Sfakianakis et al., 2015), sterilized by heating for 60 min at 95°C, inoculated with 5 × 106 CFU g−1 of an L. casei isolate, and then incubated at 37°C for 96 h.
The titratable acidity of the fermented milk was determined according to the China National Standard GB 5009.239–2016 (Jia et al., 2016). Fermented milk (10 g) was added to phenolphthalein solution (2 ml, 1% w/v in ethanol) and the mixture was titrated with standardized 0.1 M NaOH. The titratable acidity was calculated and expressed as degrees Thorner (°T; Alferez et al., 2001). Viable microbial counts. The viable microbial count of L. casei isolates in milk fermentations was determined by counting colonies on MRS agar as described previously (Sah et al., 2014). Plates were incubated for 36–48 h at 37°C. The number of viable cells per gram was counted and expressed as logCFU g−1. Correlations were evaluated by the testing significance of Pearson’s correlation coefficient (titratable acidity, bioactive compounds, pH) and Spearman correlation coefficient (growth).
We proposed a new method for the determination of acid production rate based on the titratable acidity and viable microbial count. The acid production rate of 114 isolates was calculated by dividing the acidity by the viable microbial count. Cluster analysis of L. casei strains was performed using IBM SPSS statistics 26.
HC and LC strains were inoculated (5 × 106 CFU g−1) in MRS broth containing a carbohydrate carbon source (0.5% w/v) and bromocresol purple (0.1% w/v). All incubations were performed at 37°C and in quadruplicate. The strain was considered to grow successfully on the tested sugar when the turbidity at OD600 was >0.8. The blank was un-inoculated MRS broth.
Carrez clarification reagents 1 and 2 were prepared by separately dissolving potassium hexacyanoferrate (10.60 g, Sigma) and zinc acetate dihydrate (21.90 g, Sigma) in water (100 ml). Fermented milk (5 g) was weighed accurately into graduated tubes (100 ml) and dissolved in water (25 g). Carrez reagents 1 and 2 (2.5 ml each) were added sequentially with magnetic stirring for 30 min, then the mixture was centrifuged at 5000 × g for 15 min. Extracts were made to 100 ml with water, filtered (discarding initial filtrate) and an aliquot passed through a 0.22 μm membrane filter (Indyk et al., 1996). A control milk powder sample was included in each sample set to monitor method performance. Sugar content was measured by high-performance liquid chromatography (HPLC) on a Model 2,695 HPLC (Waters Corporation, Milford, MA), fitted with a Bio-Rad Aminex®HPX-87P column (300 mm × 7.8 mm × 9 μm; Bio-Rad, Hercules, CA) and a Model 2,414 differential refractive index detector (RID). The mobile phase was aqueous sulfuric acid (5 mM) at a flow rate of 550 μl/min, with the column temperature set at 60°C.
Bacterial DNA isolation and purification. Overnight cultures with an OD600 of 0.8–1.0 were collected by centrifugation and washed with TES buffer (50 mM Tris-Cl; 30 mM EDTA; 25% Sucrose, pH 8.0). The cells were lysed by the addition of 100 μl of 50 mg/ml lysozyme solution in TE, with subsequent incubation at 37°C for 30 min. A total L. casei DNA extract was prepared as described previously (Pushnova et al., 2000). Cell lysates (50 mM Tris-Cl, pH8.0; 10 mM EDTA, pH 8.0; 50 mM sodium chloride; 60 mM sodium acetate, pH 5.2; 1% w/v SDS) were heated at 65°C for 30 min and extracted with phenol/chloroform. DNA samples (1 μg) were cleaved into 400–500 bp fragments using a Covaris (Woburn, MA) M220 Focused Acoustic Shearer, following the manufacturer’s protocol. Illumina sequencing libraries were prepared from the sheared fragments using the NEXTflex™ Rapid DNA-Seq Kit (PerkinElmer Applied Genomics). Libraries then were used for paired-end Illumina sequencing (2 × 150 bp) on an Illumina NovaSeq 6,000 (San Diego, CA). Assembly of the clean reads was performed using SOAPdenovo2 (Koren et al., 2017). Glimmer (Delcher et al., 2007) was used for CDS prediction, tRNA-scan-SE (Borodovsky and Mcininch, 1993) was used for tRNA prediction, and Barrnap (Victorian Bioinformatics Consortium, Australia) was used for rRNA prediction. The predicted CDSs were annotated using the NR, Swiss-Prot, Pfam, GO, COG, and KEGG databases, using the sequence alignment tools BLASTP, DIAMOND, and HMMER. Each set of query proteins was aligned with the databases, and annotations of best-matched subjects (e-value <10−5) were obtained for gene annotation.
All protein sequences of the six exemplar L. casei genomes were subjected to an orthology prediction using OrthoMCL (Fischer et al., 2011), with thresholds of: E-Value: 1E-5, percent identity cutoff: 0, and Markov inflation index: 1.5. Comparative genomic analyses for COGs were performed using the similarity clustering program implemented in ERGO (Huerta-Cepas et al., 2016). The core genomes were mapped to the bacterial COGs and KEGG database to evaluate the main functional COG categories related to the genome (Huerta-Cepas et al., 2016). Screening of core genes (genes contained in all gene families) and unique genes (genes contained only in a specific gene family) was based on homologous gene analysis.
The relative copy number was assessed by the quantification method (Alvarez-Martin et al., 2007). The elongation factor Tu gene tuf (GenBank Accession No. AJ418937.2), identified as a chromosomally encoded single-copy gene, was used as the reference gene (Chavagnat et al., 2002). A 180 bp fragment of the tuf gene was amplified with primers tuf2-F and tuf2-R (Supplementary Table S2) and a 102 bp fragment of the lacG gene was amplified with the primers lacG-F and lacG-R (Supplementary Table S2). Real-time PCR was conducted with the following cycling conditions: 95°C for 5 min, followed by 40 cycles of 95°C for 20 s, 60°C for 30 s, and 68°C for 30 s each. The relative copy number was calculated using Nrelative = (1 + E)−△CT (Alvarez-Martin et al., 2007), where E and ΔCT represent the PCR amplification efficiency and the difference between the threshold cycle number (CT) of the tuf and lacG reactions, respectively.
General molecular techniques, including DNA electrophoretic analysis, recovery, and storage, were performed using standard protocols. pSIP600 (Supplementary Figure S2) was pSIP502 vector without nisRK (Sorvig et al., 2003). Plasmid isolation from both E. coli and L. casei was performed using a Plasmid Kit according to the manufacturer’s instructions (OMEGA Bio-tek Inc., Doraville, GA). PCR was performed using Q5 high-fidelity DNA polymerase (New England Biolabs, Ipswich, MA). Primers used in PCR reactions were synthesized by Sangon Biotech (Beijing, China) and are listed in Supplementary Table S2. The purified PCR products were cloned directly into BglII-HindIII-digested pSIP600, using a ClonExpress Ultra one-step cloning kit (Vazyme Biotech, Beijing, China). The recombinant vector pSIP601 (Supplementary Figure S3) was transformed into E. coli DH5α using standard heat shock transformation (Watanabe et al., 1994) and introduced into LC strains by electroporation (1.5 kV, 400 Ω, 25 mF; Welker et al., 2015). Erythromycin was added as follows: 500 μg/ml for E. coli and 10 μg/ml for L. casei strains. L. casei LC_N31/pSIP601, LC_N80/pSIP601, and LC_N88/pSIP601 recombinant strains were generated under erythromycin resistance selection. DNA sequencing was performed by Sangon Biotech and the results were analyzed with DNAMAN software.
The 6-phospho-β-galactosidase (LacG) activity of selected strains was assayed as described previously (Hengstenberg et al., 1969; Bidart et al., 2018). A measured amount of the microbial sample (104CFU) was dissolved in Tris-hydrochloride buffer (0.9 ml, 0.1 M, pH 7.6, containing 0.05 M NaCl and 0.05 M MgCl2). The reaction was started by adding o-nitrophenyl-β-D-galactopyranoside-6-phosphate (ONPG-6-P; 3 μmoles; Toronto Research Chemicals, Canada) in distilled water (0.1 ml) to the enzyme solution, then the mixture was incubated at 30°C. The assays were repeated twice and in triplicate and the increase in absorbance was monitored at 405 nm. One unit of LacG activity was defined as 1 nmol/h of ONP released at 30°C by 104CFU of L. casei.
Sequence similarity was detected with BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) and multiple sequence alignments were performed with Clustal W (Thompson et al., 1994). Putative sugar and amino acid metabolic pathways were predicted by KEGG (http://-www.genome.jp). Carbohydrate-active enzymes were identified using the CAZy database (http://www.cazy.org; Cantarel et al., 2009). ANIb calculation (Richter and Rossello-Mora, 2009) was performed using pyani (v0.2.7; https://cloud.majorbio.com/) to evaluate distances between all L. casei/paracasei genomes. The presence of an N-terminal signal peptide sequence was predicted using SignalP 6.0 (https://services.healthtech.dtu.dk/service.php?SignalP-6.0; Petersen et al., 2011). Nucleotide sequence accession numbers. The draft genome sequences of the six exemplar isolates were submitted to GenBank under the following accession numbers: LC_N16 (SAMN24532469), LC_N17 (AMN24532470), LC_N40 (SAMN24532471), LC_N31 (SAMN24532472), LC_N80 (SAMN24532473), and LC_N88 (SAMN24532474).
A total of 199 samples were collected from 29 geographical regions, in 12 Chinese provinces, for isolation and purification. Isolates were identified as LAB based on positive Gram staining assays, the absence of catalase and oxidase activity, and microscopic observation of cell morphology. A panel of 472 LAB isolates was obtained from the samples and LAB were chosen for taxonomical identification by sequencing of the PCR-amplified 16S rRNA; there were 171 strains of L. casei/parcasei/rhamnosus subspecies (Supplementary Table S1, set A; Liu et al., 2012). L. casei-related taxa strains were specifically identified with housekeeping loci (pheS, recA, rpoC, tuf, and uvrC; Supplementary Table S2; Cai et al., 2007; Colombo et al., 2018, 2020), yielding 114 identified L. casei strains from 99 samples that were isolated and freeze-dried in a cryoprotectant (Supplementary Table S1, set B).
All strains successfully induced coagulation of the milk after 114 l.casei strains were fermented in milk. The pH of all milk fermentations was reduced below 4.0, microbial counts were greater than 8.0 logCFU g−1, and acidity was greater than 160°T for all strains. Of these, 24 strains produced greater acidity than the commercial strain, LC_N00 (238.76 ± 0.46°T; Supplementary Table S3; Figure 1A). The acid production rate was calculated by dividing the acidity by the viable microbial count. Cluster analysis was used to classify the acid production rate of all L. casei strains into three groups, by comparison with the acid production rate of LC_N00. The high acid (HC) group (34 strains) had acid production rates higher than 26.03 ± 0.62, the medium acid (MC) group (38 strains) had acid production rates between 24.54 ± 0.15 and 25.88 ± 0.53 (i.e., similar to that of LC_N00), and the low acid (LC) group (42 strains) had acid production rates less than 24.50 ± 0.25 (Figure 1B). Three exemplar strains were selected from both the HC and LC groups, to compare their acid production rates during fermentation. The mean acid production rates of the HC and LC strains rapidly increased to 22.97 ± 0.97 and 15.38 ± 1.35, respectively, within 60 h, then slowly increased further to 28.68 ± 0.89 and 18.71 ± 0.90, respectively, between 60 h and 96 h (p < 0.05; Figure 1C). The mean acid production rates of the HC group were about nine times higher than those of the LC group at 96 h (Table 1).
Comparative genomic analysis of the six sequenced HC and LC strains revealed 2,187 core genes (Supplementary Table S4A). The core genes were mapped to the Clusters of Orthologous Groups (COG) database (Huerta-Cepas et al., 2016) to evaluate the main functional COG categories. A collection of abundant COG categories, which contain genes that typically have the same functional category, from the six strains is shown in Supplementary Tables S4B,C. The most common COGs were those associated with “Carbohydrate transport and metabolism” (G). 10.53% (HC) and 9.73% (LC) of the COGs were grouped into the G category (Supplementary Figure S5). Notably, LC_N16 (208), LC_N17 (207), and LC_N40 (207) had almost equal numbers of genes classified into the G category. That is, the homologous gene function annotations indicate that these three strains have similar carbohydrate metabolic pathways. A comparative analysis of metabolic pathways related to genes in the G category indicated that the genes of the HC and LC groups are involved in carbohydrate metabolism, including phosphoenolpyruvate-carbohydrate phosphotransferase (PTS) transporter systems, glycosyl hydrolases, ABC transporter permeases, and other carbohydrate-related proteins (Figure 2). PTS system genes (52 and 91 genes in HC and LC, respectively) were the most numerous. These genes are often arranged in clusters, each of which is involved in the uptake and subsequent hydrolysis of a particular carbohydrate; the ability of strains to metabolize carbohydrates is reflected in their rates of organic acid production and the final concentration achieved (Rasko et al., 2005; Buron-Moles et al., 2019). Further analysis was focused on lactose metabolism and revealed that there appear to be two main routes for lactose catabolism in HC and LC strains, namely, the phosphoenolpyruvate-lactose phosphotransferase (PTSLac) system and the lactose permease symport system (Supplementary Table S4). The genes involved in the hydrolysis of lactose are related to three lactose metabolic pathways: 00052MN-Galactose metabolism (Supplementary Figure S1A), 00010MN-Glycolysis Gluconeogenesis (Supplementary Figure S1B), and 00030 M-Pentose phosphate pathway (Supplementary Figure S1C). The main routes of lactose hydrolysis were the Embden–Meyerhof–Parnas (EMP) and pentose phosphate pathways (Figure 3). A comparison of the number, location, and amino acid sequence of the genes in these pathways revealed that the lac operon was present in all six strains of the HC and LC groups, but there were significant differences in the location of the lac operon encoding PTSLac. The lacT, E, G, and F genes, which constitute the lac operon, control lactose metabolism; lacT codes for a transcriptional anti-terminator, lacE and F for the PTSLac EIICBA domains, and lacG for the phospho-β-galactosidase (Alpert and Siebers, 1997; Zheng et al., 2015). Notably, the lac operon responsible for lactose transport and hydrolysis was located on plasmids in the HC strains, whereas that of the LC strains was located in the chromosome (Table 2).
The lactose catabolic capacity of the six HC and LC group exemplar strains was determined by measuring the lactose content of hydrolyzed milk by anion-exchange high-performance liquid chromatography (HPLC; Richmond et al., 1982; Zhang et al., 2010). The linearity of the chromatographic method was verified by the coefficient of determination (𝑅2), which was >0.99 and the retention time of lactose was 19.39 ± 0.12 min, which indicates good reproducibility (Supplementary Figure S6). In addition, all six strains reached a turbidity in MRS medium of OD600 > 0.8 and a pH < 5.0 within 24 h, indicating that lactose was catabolized to produce organic acids (results not shown). The lactose utilization rates of the HC strains (LC_N16, 17, 40) were 46.4, 47.5, and 49.4%, respectively, whereas those of the LC strains (LC_N31, 80, 88) were 11.4, 24.3, and 19.8%, respectively, so the lactose utilization rate of HC strains was markedly higher than that of LC strains (Figure 4; p < 0.05). The lac operons of HC strains were located on plasmids and it appears that this enabled higher expression of the genes responsible for lactose metabolism, enabling these strains to rapidly catabolize lactose into glucose and galactose-6 phosphate.
Comparison of copy numbers of lac operon in HC and LC strains. The lac operon copy number in the six exemplar strains was determined by real-time PCR. A standard curve was generated for the lacG and tuf genes by linear regression of a plot of serial five-fold dilutions of both plasmid and genomic DNA (Table 3). Theoretically, the slope of the standard curve should be computed as its absolute gradient (−1/log52 = −2.32; Lee et al., 2006); the standard curves obtained for the lacG and tuf genes were linear over the tested range (𝑅2 > 0.99) and the slopes were between −2.23 and −2.41, only slightly different (<2%) from the theoretical value (Table 3). The copy numbers of the HC strains (LC_N16, 17, 40) were 9.34 ± 2.09, 3.09 ± 1.02, and 6.39 ± 0.98, respectively, significantly higher than those of the LC strains (LC_N31, 80, 88), at 1.2 ± 0.16, 1.14 ± 0.12, and 1.16 ± 0.10, respectively (Table 3). Plasmids are genetic material independent of the chromosome, which can replicate autonomously and their genes can be expressed to much higher levels than chromosomal genes (Panya et al., 2012). High copy numbers of lac operon genes would facilitate HC strains to catabolize lactose more rapidly than LC strains, in agreement with the lactose utilization results (Figure 4). Copy numbers of lac operon in recombinant strains. To determine the relationship between copy number and acid production rate, the lacT, E, G, and F genes were cloned into the pSIP600 vector (Supplementary Figure S3), then transferred into the LC strains by electroporation (Welker et al., 2015). The recombinant vector was verified by PCR (lacT-lacE-F/R and lacG2-F/R) and digested with HindIII, to confirm successful construction of pSIP601 (Supplementary Figure S4). The copy numbers of lac operon genes in the recombinant strains LC_N31-601, LC_N80-601, and LC_N88-601 were 15.86 ± 1.34, 16.59 ± 0.38, and 18.12 ± 1.90, respectively, and were significantly higher than those of the control (LC_N31-600, LC_N80-600, LC_N88-600) and LC wild-type strains (LC_N31, LC_N80, LC_N88; Table 3; p < 0.05). The pSIP series vectors are one-plasmid systems with pUC(pGEM)ori-256rep replication determinants for Escherichia coli, Lactobacillus sakei, and Lactiplantibacillus plantarum; the replication determinants of pSIP600 vector can be changed easily, meaning that the system can be made to function well in Lactobacilli (Sorvig et al., 2003). Markedly increasing the copy numbers of lac operon genes in the recombinant strains should increase their expression.
Phospho-β-galactosidase activity of recombinant strains. Phospho-β-galactosidase (LacG) is involved in the catabolism of lactose and is required for lactose utilization by LAB. LacG activity was assayed with o-nitrophenyl-β-D-galactopyranoside-6-phosphate (ONPG-6-P). The LacG activities of the recombinant strains (LC_N31-601, LC_N80-601, LC_N88-601) were 3.31 ± 0.14, 3.31 ± 0.11, and 3.27 ± 0.12 U/104 CFU, respectively (Figure 5) and were not significantly different from the HC strains (p < 0.05), but were markedly higher than the LC-600 control and LC wild-type strains. These results suggested that the high copy numbers of the lacG gene in the recombinant LC strains increased both lacG expression and their LacG activity, which would increase their hydrolytic capacity for intracellular phosphorylated lactose. Performance of recombinant strains in milk fermentation. The acid production rates of the recombinant LC strains in milk fermentations, with and without erythromycin resistance, were compared with those of the corresponding wild-type strains. There was no significant difference between the recombinant LC strains with erythromycin resistance (LC_N31-601, LC_N80-601, LC_N88-601) and without (LC_N31-601-R−, LC_N80-601-R−, LC_N88-601-R−), indicating that the lac operon is stable in the recombinant strains (Figure 6A; p < 0.05). However, the acid production rates of the recombinant LC strains were significantly higher than the recombinant control and LC wild-type strains (Figure 6A; p < 0.05). The lactose utilization of the recombinant LC strains (35.0, 50.7, and 42.1%) was significantly higher than LC wild-type (13.1, 21.5, and 17.9%) and control strains (14.2, 22.9, and 21.4%; Figure 6B; p < 0.05). These results suggest that location of the lac operon on a plasmid in L. casei markedly accelerates lactose utilization and increases organic acid production during milk fermentation.
The acid production rate of L. casei is an important characteristic, which determines the quality of fermented milk beverages and influences the cost of industrial milk fermentation. Some measures of acid production rate have been reported, the pH, tVmax, tpH4.5, and the kinetic parameter Vm = dpH/dt were quantified to assess Streptococcus strains (Zanatta and Basso, 1992; Dashper et al., 2012) and Lactobacillus acidophilus (Almeida et al., 2008). Fugl et al. (2017) evaluated the fermentation performance of L. lactis, S. lutetiensis, S. infantarius, and P. acidilactici strains by acidity and pH. These parameters take into account one important indicator of the starter, but the final microbial count in the fermented probiotic product must also be considered. Therefore, this study used acid production rate to assess LAB strains, which can be calculated from the acidity and viable cell count. The resulting acid production rate data were used to classify the L. casei isolates into three groups: the “medium acid” (MC) strains had similar acid production capacity to the commercial starter strain LC_N00, which has industrially acceptable acid production and final microbial counts. The “high acid” (HC) and “low acid” (LC) strains had higher and lower acid production capacity, respectively, in milk fermentations, compared with LC_N00. The HC and MC strains not only met industrial acid production requirements, but also could reach a final viable count of >108 CFU/ml. Uptake of lactose into bacterial cells and initiation of its catabolism involves several metabolic pathways: ABC protein-dependent systems, lactose-galactose antiporters, lactose-H+ symport systems, and the PTSLac transporter system (Alpert and Siebers, 1997). The two main systems active in the exemplar L. casei strains studied here were the lactose permease symport and PTSLac systems. Although there are at least two sugar transport systems, most sugars are transported by PTSs, the primary sugar transport systems of Gram-positive bacteria (Ajdic et al., 2002). PTSs regulate overall carbohydrate metabolism through gene expression in L. casei, and are important under acidic conditions (Wu et al., 2012). Lactose fermentation is very widely employed and well-understood by the dairy industry (Cavanagh et al., 2015). LAB have evolved metabolic systems that ensure the preferential use of readily metabolizable carbon sources, such as carbon catabolite repression (CCR), which modulates gene expression in response to the availability of carbon compounds; CCR is well known in lactobacilli and can also interact with catabolite control protein A (CcpA), the master regulator of carbon metabolism (Monedero et al., 1997; Stefanovic et al., 2017). Inhibition of lac gene expression during growth on glucose is a consequence of PTSGlc-mediated inducer exclusion, a repressive effect of a functional glucose-phosphoenolpyruvate-dependent phosphotransferase system (Chassy and Thompson, 1983; Monedero et al., 1997). However, the L. casei strains studied here were able to utilize three sugars (lactose, sucrose, and glucose) simultaneously, but with lactose as the primary carbon source (HC and LC strains utilized mostly lactose; results not shown). During glucose fermentation by L. casei via the EMP, carbohydrates are preferentially transported by PTS systems, and metabolism of disaccharides is preferred over glucose fermentation via the pentose phosphate pathway (Ganzle et al., 2007; Ganzle and Follador, 2012). When combined, these two metabolic shifts increase the yield of ATP. Accordingly, metabolism of lactose and sucrose by disaccharide phosphorylases is not repressed by glucose. The lac operon genes lacT, E, G, and F, which constitute PTSLac, regulate lactose metabolism during milk fermentation (Bidart et al., 2018). The lac operon is induced by lactose through transcription antitermination, mediated by LacT (Gosalbes et al., 1999) and several L. casei strains (e.g., L. casei 334/64H) carry the lac operon on a plasmid (Gosalbes et al., 1997; Van Kranenburg et al., 2005). Unique functions of L. casei are encoded on mobile elements, such as plasmids or transposons, including exopolysaccharide biosynthesis and sugar metabolism; these functions contribute to the high viscosity, i.e., creamy texture of fermented milk beverages and to utilization of lactose in dairy fermentations (Davidson et al., 1996; Schroeter and Klaenhammer, 2009). Differences in L. casei lactose metabolism resulting from different locations of the lac operon (plasmid vs. chromosome) have not been previously reported. This is the first report that expression of the lacT, E, G, and F genes located on a plasmid is enhanced, compared with the same genes located on the chromosome, since the former can replicate autonomously. The high expression of the lacG gene increased LacG enzyme activity and consequently, lactose hydrolysis, which increased the acid production rate, an important attribute of a milk fermentation starter culture. In conclusion, the association between acid production rate and lactose metabolism enabled the identification of the genes involved in transport and catabolism of lactose. Further research is required to confirm these findings, as well as to elucidate the potential functions of these genes during milk fermentation, which is of great importance for industrial applications of L. casei.
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary material.
XL: conceptualization, data curation, methodology, software, and writing—original draft. ZZhai and YH: formal analysis, writing—review and editing, and resources. MZ and CH: software and supervision. JH, SS, ZZhao, and YS: validation and visualization. FR and RW: conceptualization, data curation, formal analysis, investigation, project administration, resources, and supervision. All authors contributed to the article and approved the submitted version.
This work was supported by the National Natural Science Foundation of China (contract 31972055) and the National Key Research and Development Program of China (2018YFC1604303).
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
PMC9647826 | Fatima Al Amrani,Reshma Amin,Jackie Chiang,Lena Xiao,Jennifer Boyd,Eugenia Law,Elisa Nigro,Lauren Weinstock,Ana Stosic,Hernan D. Gonorazky | Scoliosis in Spinal Muscular Atrophy Type 1 in the Nusinersen Era | 01-08-2022 | Background and Objectives The introduction of spinal muscular dystrophy (SMA)-modifying therapies, such as antisense oligonucleotide therapy, has changed the natural history of SMA. Most reports on treatment outcomes have focused on motor scores and respiratory function. The objective of this study is to document the development and progression of scoliosis in patients with SMA1 treated with nusinersen. Methods A descriptive single-center study was conducted in patients with SMA1 who were treated with nusinersen before 6 months of age. Data were collected on patients who met criteria, including age at the first nusinersen dose, number of nusinersen doses, degree of scoliosis, respiratory parameters, feeding route, and motor scores at baseline and follow-up. The Children's Hospital of Philadelphia Infant Test of Neuromuscular Disorders (CHOP-INTEND) was subanalyzed using axial (AxS) and appendicular motor (ApS) scores to evaluate a possible correlation between scoliosis and axial muscle strength. Results From our cohort, 31 percent (11/35) of patients had a diagnosis of SMA1. Sixty-three percent (7/11) met the inclusion criteria. All patients (7/7) showed initial improvement in their CHOP-INTEND scores in correlation with improvement on the ApS. Despite this, most patients did not show improvement in the AxS. Subsequently, all patients developed scoliosis in the first year of life with Cobb angles that ranged between 18° and 60°. Furthermore, total CHOP-INTEND scores had dropped in 2 patients alongside the development of a Cobb angle of >40°. Discussion Despite the significant improvement in functional motor assessment in patients with SMA1, there is a progression of significant scoliosis despite treatment. Subsequently, lack or minimal improvement on the axial CHOP-INTEND scores may predict worsening on the total motor scores. | Scoliosis in Spinal Muscular Atrophy Type 1 in the Nusinersen Era
The introduction of spinal muscular dystrophy (SMA)-modifying therapies, such as antisense oligonucleotide therapy, has changed the natural history of SMA. Most reports on treatment outcomes have focused on motor scores and respiratory function. The objective of this study is to document the development and progression of scoliosis in patients with SMA1 treated with nusinersen.
A descriptive single-center study was conducted in patients with SMA1 who were treated with nusinersen before 6 months of age. Data were collected on patients who met criteria, including age at the first nusinersen dose, number of nusinersen doses, degree of scoliosis, respiratory parameters, feeding route, and motor scores at baseline and follow-up. The Children's Hospital of Philadelphia Infant Test of Neuromuscular Disorders (CHOP-INTEND) was subanalyzed using axial (AxS) and appendicular motor (ApS) scores to evaluate a possible correlation between scoliosis and axial muscle strength.
From our cohort, 31 percent (11/35) of patients had a diagnosis of SMA1. Sixty-three percent (7/11) met the inclusion criteria. All patients (7/7) showed initial improvement in their CHOP-INTEND scores in correlation with improvement on the ApS. Despite this, most patients did not show improvement in the AxS. Subsequently, all patients developed scoliosis in the first year of life with Cobb angles that ranged between 18° and 60°. Furthermore, total CHOP-INTEND scores had dropped in 2 patients alongside the development of a Cobb angle of >40°.
Despite the significant improvement in functional motor assessment in patients with SMA1, there is a progression of significant scoliosis despite treatment. Subsequently, lack or minimal improvement on the axial CHOP-INTEND scores may predict worsening on the total motor scores.
Scoliosis is a challenging comorbidity associated with many neuromuscular disorders. It is defined as a 3-dimensional deformity of the spine. Scoliosis can be idiopathic or congenital or can develop secondary to vertebral deformity or tumor-related deformity or neuromuscular disease. Scoliosis prevalence among patients with neuromuscular disorders differs from one condition to the other. Neuromuscular scoliosis is more severe, with a rapid progression in most cases, and is usually associated with increased comorbidities. The combination of scoliosis and the limitation resulting from the underlying neuromuscular condition can lead to significant impairment in limb movements, cardiopulmonary function, gait, standing, sitting, balance, trunk stability, activities of daily living, pain, concerns about self-image, and social interactions. Furthermore, progressive scoliosis due to rapidly deteriorating axial muscle tone is a critical factor in all patients with spinal muscular atrophy (SMA) types 1 and 2, significantly affecting both respiratory and motor function. Historically, SMA had no curative or disease-modifying treatment. The mortality was almost 100% in SMA1 without respiratory support before 2 years old. Over the past decade, there have been significant advancements in understanding the genetics and pathogenesis of SMA, which have led to the development of novel therapies for this devastating disorder. In 2016, the first disease-modifying drug for SMA was approved by the Food and Drug Administration, and it was the first drug approved for patients with SMA in Canada. Nusinersen is an antisense oligonucleotide intrathecal therapy that enhances the inclusion of exon 7 into the mRNA transcript from the survival motor neuron 2 (SMN2) gene, thus resulting in full-length SMN protein production. The ENDEAR clinical trial of intrathecal nusinersen in SMA1 demonstrated improved motor function in treated patients and promoted prolonged survival of infants with SMA1. However, the assessment of treatment outcome focused on motor scores and respiratory function. Little is known about the impact of nusinersen on the progression of scoliosis. Recently, Young described a high correlation between the degree of scoliosis and the Revised Upper Limb Model and Hammersmith Functional Motor Scale Expanded for SMA (HFMSE) in patients who participated on the CHERISSH and SHINE studies. Exploring the effect of novel therapies on scoliosis progression is needed because findings will likely affect the current guidelines of care for this group of patients. We hypothesized that scoliosis progression would be highly correlated with motor outcomes in the SMA1 population treated with nusinersen. Our aim was to document the development and progression of scoliosis in patients with SMA1 who started treatment with nusinersen before 6 months of age.
We conducted a descriptive study of patients with a confirmed genetic diagnosis of SMA, who were treated with nusinersen in our institution. The inclusion criteria were (1) a confirmed genetic diagnosis of SMA with 2 homozygous deletions of exon 7 in the SMN1 gene and 2 copies of SMN2 and (2) patients who initiated treatment with nusinersen before 6 months of age. All patients received the standard of care according to the 2017 international SMA guidelines. SMA diagnosis was confirmed genetically by the detection of homozygous loss of function of the SMN1 gene and determined SMN2 gene copy numbers by multiplex ligation-dependent probe amplification analysis using the SALSA multiplex ligation-dependent probe amplification. Patients who met the inclusion criteria had received 4 loading doses (12 mg) of intrathecal nusinersen at day 1, 14, 30, and 60, followed by a maintenance dose of 12 mg every 4 months.
The institutional research ethics board approved this study at the Hospital for Sick Children (REB#1000069140).
Deidentified patient information was collected and managed using REDCap tools hosted at the Hospital for Sick Children. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies, providing (1) an intuitive interface for validated data capture, (2) audit trails for tracking data manipulation and export procedures, (3) automated export procedures for seamless data downloads to common statistical packages, and (4) procedures for data integration and interoperability with external sources. The clinical data gathered included SMA type, SMN1 and SMN2 copy number, age at symptom onset, age at the first dose, duration of the disease at the time of treatment initiation, treatment duration to date, the total number of nusinersen doses, and feeding route at the time of the first dose and the last dose.
The presence of scoliosis was evaluated clinically at baseline using spinal X-rays and followed serially every 6 to 9 months. The Cobb angle was used as the measure of scoliosis progression. Ideally, the radiographs were taken in the sitting position if possible or in the supine position if sitting position radiographs were challenging to obtain. Two views were obtained: anterior-posterior and lateral. In the case of S-shape scoliosis, the highest degree angle was the one used for assessment.
The patient's respiratory status at the first nusinersen dose and most recent nusinersen dose was obtained from the electronic health record. Respiratory parameters, including whether the patient required invasive or noninvasive ventilation, age when noninvasive or invasive ventilation was initiated, the indication for the initiation, and the number of hours per day, were all documented. We documented the use of cough assist therapy, the number of times used per day, and the timing of the first and most recent dose of nusinersen. The presence of pectus excavatum and polysomnography results was documented at baseline, around the starting nusinersen dose and during the most recent visit to the respiratory clinic.
We also documented the use of a gastrostomy tube (G-tube) and swallowing assessment results.
The Children's Hospital of Philadelphia Infant Test of Neuromuscular Disorders (CHOP-INTEND) is a validated measure to evaluate SMA patients' motor function. A trained physiotherapist used the CHOP-INTEND to evaluate each patient and determine a score at baseline and every 4 months. Because the development of scoliosis in neuromuscular disorders is related to the axial muscles, CHOP-INTEND scores were divided into (1) axial muscle scores (AxS) (items 4, 12, 14, 15, and 16) with a maximum score of 20 points and (2) appendicular muscle scores (ApS) (the rest of the items) with a maximum score of 44 points. Clinical status was compared with the most recent follow-up visit in the neuromuscular and respiratory clinics.
Clinical and demographic characteristics were used to summarize the study population. For the primary analysis data, we used median, range, and proportions using Excel.
Anonymized data not published within this article will be made available by request from any qualified investigator.
Among a total of 35 patients with SMA who were followed at the neuromuscular clinic in our institution, 11 (31%) had SMA type I. All 11 patients (100%) had 2 copies of SMN2. Seven patients met our inclusion criteria. Two patients were excluded because nusinersen was started after 6 months of age (1 at 63 months and 1 at 108 months). Two additional patients were excluded from analysis because care was withdrawn after a prolonged intensive care unit admission, and they died after elective extubation. Both patients were on the loading phase of nusinersen treatment. Seven of 11 (63.6%) were followed during the study period (Table 1). Four males and 3 females met the study criteria. The onset of symptoms ranged between 2 and 8 weeks (median 4 weeks). All patients presented with axial hypotonia and weakness in the upper and lower extremities. Four of 7 patients presented with bulbar symptoms, including choking with feeds, regurgitation, coughing, and swallowing difficulties. The age for receiving the first nusinersen dose ranged between 8 and 23 weeks (median of 12 weeks). The time from symptom onset to the first nusinersen dose ranged between 4 and 19 weeks, with a median of 8 weeks. Treatment was started for all of them except patient 7 before 12 weeks of disease progression. Each study participant received an average of 9 doses of nusinersen ranging between 5 and 15 doses. The follow-up period ranged between 10 and 57 months, with a median of 21 months. All patients were feeding orally at the first nusinersen dose (Table 1). Six of the 7 patients were breathing spontaneously when they received the first nusinersen dose, whereas patient 7 needed invasive ventilation at the first dose and remained on invasive ventilation at the most recent nusinersen dose (Table 1). Patients 1, 2, and 4 were initiated on nocturnal noninvasive ventilation at the last nusinersen dose, 2 for nocturnal hypoventilation (patients 1 and 2) and 1 for chest wall remodeling (patient 4). There were plans at the time of the most recent nusinersen dose for 3 of 7 patients to start noninvasive ventilation, 1 for nocturnal hypoventilation (patient 6) and 2 for chest wall remodeling (patients 3 and 5). Patients 3 and 7 were on cough assist at the first nusinersen dose, and all patients were on cough assist at the time of their most recent nusinersen dose. No patients had pectus excavatum at the first nusinersen dose, and patients 6 and 7 presented it at their most recent nusinersen dose (Table 2). Patient 7 underwent tracheostomy because of the ongoing need for invasive ventilation. Furthermore, this patient had received the first nusinersen dose at 23 weeks of age (19 weeks after symptom onset). The second patient with pectus excavatum (patient 6) showed evidence of obstructive sleep apnea on polysomnography at the age of 8 months, indicating the need for bilevel positive airway pressure initiation; however, compliance with therapy was not optimal until the age of 24 months. All 7 patients, except 1, were orally feeding at the time of the first nusinersen dose. The 1 patient (patient 7) who was not feeding orally had a nasogastric tube and subsequently acquired a gastrostomy tube. Furthermore, 4 of these patients had bulbar symptoms at the first nusinersen dose. All patients except one were provided enteral feeds at the most recent nusinersen dose. Five of these patients were exclusively on enteral feeds and were deemed unsafe to orally feed because of the risk of aspiration. One patient (patient 6) received both enteral and oral feeds. Only 1 patient (patient 5) remained on oral feeds and did not show any bulbar symptoms. Four of 7 participants had CHOP-INTEND testing completed at baseline and subsequently for follow-ups. Two patients (patients 6 and 7) had Hammersmith Infant Neurologic Examination (HINE) initially followed by CHOP-INTEND assessments. Patient 4 had HINE assessment initially followed by the HFMSE for subsequent follow-ups because independent sitting was achieved and was considered too advanced for the CHOP-INTEND assessment. This last patient was excluded from the CHOP-INTEND analysis (Table 2). All patients showed initial improvement in the motor scores between baseline and their assessment performed 5–10 months after the first loading dose (median age 8 months). The initial increment in CHOP-INTEND scores ranged between 1 and 16 points (average increase of 6.1 points). The 1 patient with HFMSE scores showed initial improvement by 12 points. Maximum increment on the total CHOP-INTEND scores varied between 4 to 33 points. On follow-ups, 2 of 6 patients (patients 6 and 7) showed a drop in their CHOP-INTEND scores by 4 points and 1 point at 24 and 42 months of age, respectively (16 months and 34 months after nusinersen initiation, respectively) (Figure 1A). Patient 5, who had HFMSE assessments, had dropped their score by 5 points at 45-month evaluation. The subanalysis of the AxS and ApS in the 6 patients who had completed CHOP-INTEND assessments showed that 3 of 6 presented improvement in the AxS (patients 2, 5, and 6) initially, 2 of 6 did not change (patients 1 and 7), and 1 of 6 (patient 3) showed a drop by 2 points in follow-up (Figure 1D). Only patient 2 had an increment of more than 5 points on the AxS. All patients had improvements in the CHOP-INTEND ApS initially with a median of 9.5 points (range 1–28). However, 2 patients (patients 6 and 7) presented a decline of 4 and 1 point at 24 and 42 months, respectively, in correlation with the total scores (Figure 1C). None of the patients had scoliosis at baseline when clinically evaluated in the neuromuscular clinic, but all patients developed scoliosis on subsequent follow-up (Figure 1B). Three of 7 (43%) had S-shape scoliosis, and 4 of 7 (57%) had C-shape scoliosis (3/4 to the left side and 1/4 to the right side) (Figure 2). All the patients (7/7) had a thoracolumbar scoliosis. We observed a rapid worsening of the spine curvature for most patients starting at 5 months of age with an average progression of 2.3° per month (range 1.1–3.8) (Figure 1B). All patients had scoliosis defined as Cobb angle >15° by 12 months of age. Patient 2 was the only one with an ApS of >15 points (16), with a total score of 57. At the same time, this child showed a reduction in the progression of scoliosis from down to 0.4° from 7 to 13 months of age. On the other hand, patients 6 and 7 who had >40° of scoliosis started to show a drop in their total CHOP-INTEND scores and their ApS. At the same time, patient 4 who had HFMSE showed a drop in the total score when a Cobb angle of >40° was reached.
Here, we are reporting on the progression of scoliosis in a cohort of children with SMA1 treated with nusinersen. All the children in our study developed scoliosis in the first year of life. We observed that the total CHOP score was highly influenced by the appendicular scores, and the lack or mild improvement of the axial scores correlated with the onset and/or progression of scoliosis. Furthermore, motor scores dropped in all patients who developed a Cobb angle of >40°, highlighting the clinical relevance of scoliosis even in children with SMA1 treated with nusinersen. Scoliosis emerges in nearly all nonambulatory neuromuscular patients, leading to chest wall deformities, severely reduced vital capacities, tilted hips, and loss of ambulation. One study showed that scoliosis progresses by 8° per year in patients with SMA2, 3° per year in nonambulatory SMA3 patients, and 0.6° per year in ambulatory type-III patients. On the other hand, another study evaluated the progression of scoliosis in ambulatory patients and showed that scoliosis progressed by 5°–15° per year in these patients. In addition, reference 20 reports on a cohort of patients with SMA2 who developed scoliosis with the main Cobb curvature for patients between 0 and 4 years was 26° with a median of 25°. It was recently shown that scoliosis progressed by 7.2° per year, and this progression markedly increased in the 18 months before scoliosis surgery by 10.1° per year. The average age at onset of scoliosis was 2 years or less in SMA1, between 1 and 7 years in type-II, and 4–14 years in type-III patients. All our patients developed scoliosis, which is similar to what has been reported in the medical literature. We observed a rapid progression in the first year of life, with a curvature increment of 2.3° per month starting around 5 months of age for most patients. It is unclear what the rate of scoliosis progression was in the natural history pretreatment era of patients with SMA1. However, comparing what has been reported in the literature for patients with SMA2, the progression of our cohort of patients with SMA1 appears significantly higher. We did not observe differences in severity in terms of the shape of the scoliosis between our patients. As anticipated, all our patients showed initial improvements in CHOP-INTEND total scores after receiving treatment with nusinersen. This parallels the effect of nusinersen reported in the ENDEAR study. We know from natural history studies that patients with SMA1 initially present with CHOP-INTEND scores of 20 and rarely achieve any score above 40. Within our cohort, 3 patients had a decline of the motor scores with subsequent doses despite a dramatic initial improvement. These 3 patients had Cobb angles of 40° or more when the decline in their motor scores was observed (Figure 1B). Patient 7 was already with invasive ventilation and with more than 12 weeks of progression at the moment of starting the intrathecal infusion. This last child presented the fastest worsening on the scoliosis. We know that patients with severe respiratory component and more than 12 weeks of progression will present poor outcome despite the initiation of treatment. As a secondary objective, we wanted to analyze the contribution of ApS vs AxS in the total CHOP-INTEND scores. Analysis of the AxS (items 4, 12, 14, 15, and 16) showed either stabilization or decline in the first 6 months after initiation of treatment for the 6 of 7 patients, although there was an overall improvement in the total CHOP-INTEND score. On the other hand, the CHOP-INTEND ApS initially showed dramatic improvement in parallel with the total CHOP-INTEND initial scores (Figure 1, A–C). Likewise, the 2 patients who showed decline in the total score (16 and 34 months after therapy initiation, respectively) presented parallel worsening in the ApS. Moreover, the only patient who presented an increment of more than 5 points on the AxS presented a reduction on the progression of the scoliosis. These observations have shed light on 4 main points. First, in proportion, the contribution of the CHOP-INTEND ApS outweighed the one originated from the AxS, creating an unintended bias. Second, scoliosis progression is likely a natural result of the deterioration of axial muscle scores that have not improved with nusinersen subsequent doses. Third, the lack of improvement of the AxS might be an important indicator for the onset of severe progression of scoliosis. Finally, a worsening of the Cobb angle (≥40°) precedes the worsening of the total scores and ApS subscore. Deterioration in respiratory function is a significant cause of morbidity and mortality in SMA, especially in children with SMA1. In addition to the respiratory muscle weakness resulting from motor neuron degeneration, scoliosis causes mechanical restriction of the chest wall and results in restrictive lung disease. There is a well–documented correlation between the decline of respiratory function and progression of scoliosis (Cobb angle); a decline in forced vital capacity (FVC) % predicted by 7.7% per year has been observed. Furthermore, another study documented a decline in FVC % predicted by 4.7% and similar 3.3% decrease in the peak flow for each 10° increase in the Cobb angle. Although our study is limited by the lack of a pulmonary function test because of the age of our participants, the development of scoliosis compounded with neuromuscular weakness may have contributed to the need for respiratory support in some of our patients, particularly those with more rapidly progressing scoliosis. Six of 7 of our patients did not improve their swallowing function independent of age at the initiation of treatment or improvement of the motor scores. This feature might be an independent factor of assessment that could be related to the lack of homogenous distribution of nusinersen through the spine. Despite the small sample of our cohort (n = 7), our population is a homogenous group and representative of the general SMA1 population receiving nusinersen. We have observed the rapid progression of scoliosis in patients with SMA1 who received nusinersen. All our patients with SMA1 who received nusinersen at <6 months of age developed scoliosis in the first year of life and had a Cobb angle of >15° by the end of the first year of life despite improvement in motor function. This may have a considerable impact on these patients' respiratory parameters and CHOP-INTEND scores. Moreover, the initial improvement in the CHOP-INTEND scores seems to be falsely reassuring and reflects mainly the ApS improvement. Scoliosis is a major factor that directly affects the prognosis of patients with SMA and warrants regular monitoring. Our study highlights the importance of understanding the subcomponents of the motor functional assessment. Based on the ENDEAR clinical trial, nusinersen improves the initial motor function scores; however, it does not seem to stop or slow down the progression of scoliosis in symptomatic individuals. In treated patients with SMA1, the improvement in the total CHOP-INTEND score seems to be primarily led by the rise in the ApS subscore, whereas AxS scores appear to be influenced by the presence of scoliosis. Although these results will need validation, it sheds light on the possible effect of this novel therapy on one of the most critical SMA comorbidities. |
||
PMC9647829 | Aimee C. Talleur,Amr Qudeimat,Jean-Yves Métais,Deanna Langfitt,Ewelina Mamcarz,Jeremy Chase Crawford,Sujuan Huang,Cheng Cheng,Caitlin Hurley,Renee Madden,Akshay Sharma,Ali Suliman,Ashok Srinivasan,M. Paulina Velasquez,Esther A. Obeng,Catherine Willis,Salem Akel,Seth E. Karol,Hiroto Inaba,Allison Bragg,Wenting Zheng,Sheng M. Zhou,Sarah Schell,MaCal Tuggle-Brown,David Cullins,Sagar L Patil,Ying Li,Paul G. Thomas,Caitlin Zebley,Benjamin Youngblood,Ching-Hon Pui,Timothy Lockey,Terrence L. Geiger,Michael M. Meagher,Brandon M. Triplett,Stephen Gottschalk | Preferential expansion of CD8+ CD19-CAR T cells postinfusion and the role of disease burden on outcome in pediatric B-ALL | 22-04-2022 | Visual Abstract | Preferential expansion of CD8+ CD19-CAR T cells postinfusion and the role of disease burden on outcome in pediatric B-ALL
Adoptive immunotherapy with T cells expressing CD19-specific chimeric antigen receptors (CD19-CARs) has resulted in impressive clinical responses for relapsed and/or refractory B-cell acute lymphoblastic leukemia (B-ALL) or lymphoma, resulting in US Food and Drug Administration (FDA) approval of 4 CD19-CAR T-cell products. For pediatric, adolescent, and young adult (AYA) patients, several CD19-CAR T-cell products have been evaluated in large clinical studies and have demonstrated robust antitumor activity with similar toxicity profiles.2, 3, 4, 5 Despite these successes, CD19-CAR T-cell therapy does not induce remission in all patients with B-ALL, and many who achieve an initial response eventually experience disease relapse.6, 7, 8 While the durability of responses has been correlated to the costimulatory signaling domain of the CAR, with CD19.41BBζ-CAR T cells better able to induce long-term remission as compared with CD19.CD28ζ-CAR T cells, CD19.41BBζ-CAR T cells do not induce durable remissions in all patients. Studies have also indicated that peak CAR T-cell expansion, T-cell subsets present in the CAR T-cell product, pretreatment leukemic burden, and/or absolute lymphocyte counts also influence outcome.,6, 7, 8, 9, 10, 11, 12, 13, 14 Furthermore, at present, the optimal management for patients who achieve a complete response after CD19-CAR T-cell therapy is unknown, including which patients should proceed to consolidative allogeneic hematopoietic cell transplantation (AlloHCT)., While a commercial CD19-CAR T-cell product is available for pediatric B-ALL, there is a continued need to study new CD19.41BBζ-CAR T-cell products, with the goal of identifying improvements in manufacturing techniques, modification of construct design, and optimization of administration conditions, which result in improvements in long-term response rates. We, therefore, developed an early phase clinical study (SJCAR19; NCT03573700) to evaluate the use of autologous T cells expressing CD19.41BBζ-CARs (CD19-CAR T cells) in patients ≤21 years of age with relapsed/refractory B-ALL. Here we report the safety and efficacy of this CAR T-cell product and a detailed immunophenotypic analysis of the first 12 consecutively enrolled patients in the phase 1 study.
Detailed information regarding study design, eligibility criteria, disease response, toxicity evaluation, and correlative studies can be found in the supplemental Methods section.
This is a single institution phase 1/2 clinical study evaluating the safety and efficacy of escalating doses of autologous CD19-CAR T cells in pediatric/AYA subjects ≤21 years old with relapsed/refractory CD19-positive B-ALL (SJCAR19; NCT03573700). The protocol was approved by the St. Jude Children’s Research Hospital institutional review board; written informed consent/assent was obtained from all participants/parents in accordance with institutional guidelines and the Declaration of Helsinki. Enrollment in the phase 1 study occurred between June 2018 and March 2020; enrollment in the phase 2 study is ongoing. CD19-CAR T-cell products were manufactured at the Children’s Good Manufacturing Practice (GMP) facility of St. Jude from CD4/CD8-selected autologous leukapheresis products. CD19-CAR T-cell production is described in detail in the supplemental Methods section. Briefly, activated T cells were transduced with a self-inactivating lentiviral vector that encoded a CAR consisting of the CD19-specific single-chain variable fragment FMC63, the hinge/transmembrane domain of CD8α, and 41BB and CD3ζ signaling domains. After transduction, CD19-CAR T cells were expanded with interleukin-7 (IL-7) and IL-15 for 7 to 8 days before cryopreservation. Protocol treatment included lymphodepletion (fludarabine [25 mg/m2, days −4 to −2] and cyclophosphamide [900 mg/m2, day −2]) followed by CAR T-cell infusion (day 0; infusion occurred after thawing of a cryopreserved product). The phase 1 study used a 3 + 3 design to evaluate 2 dose levels (DLs): 1 × 106 and 3 × 106 CAR-positive T cells/kg, with a maximum cell dose of 2.5 × 108 CAR-positive T cells. Pretherapy disease evaluation (bone marrow [BM] and cerebrospinal fluid [CSF]) occurred after completion of any bridging therapy and within 2 weeks before protocol treatment. Bridging chemotherapy was per the treating physician’s discretion. After infusion, routine supportive care included anticonvulsant, antiviral, and antifungal prophylaxis, and IV immunoglobulin supplementation, as per institutional guidelines. After infusion, leukemia-directed therapy was not protocol defined, including the use of consolidative AlloHCT. Three patients received additional CD19-CAR T-cell infusions for recurrent leukemia on individualized, FDA-reviewed single-patient investigational plans.
Adverse events were captured using NCI Common Terminology Criteria (Version 5.0), except for cytokine release syndrome (CRS) and neurotoxicity (NTX). CRS was graded according to the consensus grading of the American Society of Transplant and Cellular Therapy (ASTCT). Before the adoption of the ASTCT consensus grading, NTX was graded using the common terminology adverse event criteria, in conjunction with a protocol-defined grading scheme to determine an overall NTX grade. With institutional adoption of the ASTCT consensus grading, NTX was then graded using immune effector cell-associated syndrome (ICANs). Disease response was determined using pre and 4 weeks post infusion leukemic burden in the BM. The response was categorized as complete response (CR), either minimal residual disease (MRD)-negative, MRD-positive, or no response (NR). MRD testing included flow cytometry (cut off 0.01%) and, when available for a given patient, reverse transcription-polymerase chain reaction (RT-PCR) and/or next-generation sequencing (NGS; Adaptive Biotechnologies) techniques. Extramedullary (EM) disease response was determined independently of marrow response. Relapse was defined as the development of any recurrent detectable disease (including MRD) after initial CR.
Peripheral blood samples were collected before lymphodepleting chemotherapy on the day of CAR T-cell infusion and at weeks 1, 2, 3, and 4 after infusion; BM and CSF samples were obtained at 4 weeks after infusion. Quantitative polymerase chain reaction (qPCR) assays and flow cytometric and multiplex analyses were performed.
Overall survival (OS) and event-free survival (EFS) rates were estimated by the Kaplan-Meier method. Pairwise comparisons were made using t tests. Adjustment for multiple testing was not performed due to the small sample size and the exploratory nature of the analysis. OS was defined by time from CAR T-cell infusion to death, censoring at the time of the last follow-up. EFS was defined as the time from CAR T-cell infusion to NR at the week 4 evaluation, relapse, or death, with censoring at the time of the last follow-up. Statistical analyses were performed with SAS version 9.4 (SAS Institute Inc., Cary, NC).
Thirteen patients enrolled in the phase 1 study. Twelve patients received protocol therapy. The first enrolled patient did not proceed due to poor clinical status in the setting of rapidly progressive B-ALL and is not included in this analysis. The median age at the time of B-ALL diagnosis and CD19-CAR T-cell infusion was 8.2 (range: 0.01-19.3) and 13.9 (range: 1.8-21.7) years, respectively (Table 1). Most patients identified as White (n = 10), of whom 4 were non-Hispanic. Participants were heavily pretreated, including prior AlloHCT (n = 4; median days from HCT infusion: 196.5; range: 132-254) and/or treatment with CD19- and/or CD22-directed non-CAR therapies (n = 5). No patient had previously received a CAR T-cell product, though this was not an exclusion criterion for the clinical study. Most participants harbored high-risk leukemic genetic and/or cytogenetic abnormalities. Pretherapy leukemic burden in the BM included a median morphologic blast count and MRD of 11% (range: 0-98; n = 12) and 20.6% (flow cytometry; range: 0-82.9; n = 11), respectively, with a median CD19 positivity of 99.1% (range: 67.2-100; n = 11). Pretherapy EM disease included 2 patients with disease in the CSF (central nervous system [CNS]-2) and 5 patients with areas of increased metabolic uptake on positron emission tomography (PET) imaging concerning possible leukemic involvement (Table 1). The median time between apheresis and protocol treatment was 20.5 (range: 17-45) days; most patients received bridging therapies (n = 10) during this time. Bridging therapy included systemic chemotherapy (n = 8) or focal radiation therapy to non-CNS EM sites (n = 2; 24 Gy to multiple sites of disease).
CD19-CAR T cells were successfully generated from CD4/CD8-purified leukapheresis products for all patients. The median time to manufacture cell products was 7 (range: 7-8) days. All products had >90% viability, with >98% of cells being CD45+/CD3+ (supplemental Figure 1A,B). The median transduction efficiency was 37.1% (range: 29.5-67.7) as judged by flow cytometric analysis, and the median vector copy number was 1.17 (range: 0.85-3.39); all products passed functional release criteria with a CD19-specific cytolytic activity of >20% (supplemental Figure 1C-E). Within the CAR-positive T-cell population, there was a significantly higher percentage of CD4+ vs CD8+ T cells (supplemental Figure 2A). Phenotypically, CAR-positive T cells had a predominantly central memory (CD45RO+/CCR7−) or effector memory (TEM; CD45RO+/CCR7−) phenotype (supplemental Figure 2B). Within the TEM compartment, >95% of T cells were transitional memory (CD28+/CD95+), which include CD27+ (EM1) as well as CD27− (EM4) subsets (supplemental Figure 2C,D). Phenotypic analysis of CAR-negative T cells within the CAR T-cell product revealed a similar phenotype (supplemental Figure 2B-D). CAR+/CD4+ T cells had a significantly higher percentage of PD1+/TIM3+ cells than CAR−/CD4+ T cells (supplemental Figure 2E). Likewise, CAR-positive T cells had a significantly higher percentage of CD39+ T cells than their CAR-negative counterparts (supplemental Figure 2F).
Twelve patients received protocol-defined lymphodepletion followed by CAR T-cell infusion (DL1: 1 × 106 CAR-positive T cells per kg [n = 6] and DL2: 3 × 106 CAR-positive T cells per kg [n = 6]). Patients remained inpatient after CAR T-cell infusion for a median of 11.5 (range: 8-44) days (1 discharge was patient transfer back to the referring hospital). Two patients required escalation of care to the intensive care unit due to high-grade CRS requiring either high-flow nasal cannula oxygen or vasopressor support at 7 and 9 days after CAR T-cell infusion, respectively. No patients required readmission within 30 days of infusion. Toxicities were minimal (Table 2). Hematologic toxicities included 11 patients with grade 4 neutropenia (7 of which were preexisting) and 6 with concurrent grade 3 thrombocytopenia. Six patients developed CRS (grade 3 to 4; n = 2). Three patients developed NTX, including 1 patient at grade 3 (Table 2). The median time to onset of CRS and NTX was 5 (range: 0-11) days and 6 (range: 5-7) days, respectively. Two patients showed initial improvement of CRS (resolution of fever and down-trending CRP) and then developed a subsequent hyperinflammatory syndrome consistent with CAR-associated hemophagocytic lymphohistiocytosis (carHLH), diagnosed at 9 and 10 days after CAR T-cell infusion (Table 2). Treatment of immune-mediated side effects included tocilizumab (n = 5; grade ≥3 CRS [n = 2] and carHLH [n = 1]), corticosteroids (methylprednisolone and/or dexamethasone; n = 2; grade ≥3 CRS [n = 1] and carHLH [n = 1]), siltuximab (n = 1; grade ≥3 NTX), and/or anakinra (n = 2; carHLH). Three patients received tocilizumab for grade 1 CRS due to ongoing fevers and clinical concern for the risk of rapid CRS progression. One patient later developed carHLH, while the other 2 had resolution of CRS without symptom progression. Patients recovered from CRS and NTX without lasting complications. The patients with carHLH showed clinical stability in response to immunomodulatory agents. Two dose limiting toxicities (DLTs) were observed: 1 on DL1 (grade 3 acute kidney injury in the setting of carHLH and presumed renal leukemic infiltrates as judged by metabolic activity on pretherapy PET), and 1 on DL2 (grade 4 CRS that failed to resolve ≤72 hours of onset). Though patients treated on DL2 had higher rates of grade ≥3 CRS/NTX and carHLH (Table 2), only 1 patient on DL2 experienced a DLT. Therefore, DL2 was determined to be the maximum tolerated dose (MTD) as per the study design and was selected for the phase 2 portion of the study.
At 4 weeks after infusion, 9/12 patients (75%) achieved an MRD-negative CR (MRDneg CR) in the BM, of which 7 were confirmed negative by NGS (Table 1). No patients had progressive CNS disease. PET reimaging of 3 MRDneg CR patients with abnormal pretherapy findings showed mixed metabolic responses at 4 weeks after infusion (patients 4, 6, and 11). Two patients had subsequent resolution of metabolic uptake on serial imaging (patients 6 and 11). The third patient, patient 4, had progressive disease with a tibial chloroma, which was later biopsied, and immunohistochemistry demonstrated CD19-positivity of tumor cells and presence of infiltrating CD19-CAR T cells as judged by qPCR (data not shown). Thus, 8/12 patients (67%) achieved a CR based on the National Comprehensive Cancer Network (NCCN) pediatric B-ALL response criteria. The clinical course of all patients is summarized in Figure 1A. With a median follow-up of 15.2 months (range: 1.7-22.4), the median EFS and OS for the entire cohort were 7.2 and 15.2 months, respectively (Figure 1B). Among the 9 patients who achieved an MRDneg CR, 5 patients proceeded to a planned consolidative AlloHCT at a median of 2.7 months (range: 1.9-3) after CAR T-cell infusion. For these 5 patients, this was their first AlloHCT, and all remained in remission at the time of the last follow-up (median: 19.1 months after AlloHCT; range: 7.0-30.7), with 1 patient dying of transplant-related complications. The remaining 4 patients all experienced disease reoccurrence at a median of 4.4 months (range: 2.3-9.0) after CAR T-cell infusion. Three relapsed with CD19-positive B-ALL (sites: EM, BM, and BM/EM) and 1 with CD19-negative B-ALL (site: BM). The patients with CD19-positive relapse all had a history of prior AlloHCT, and 2 patients also had a history of extensive EM disease before CAR T-cell therapy, making them less desirable candidates for a planned consolidative AlloHCT after CAR. The patient with CD19-negative relapse experienced relapse before planned AlloHCT; this patient received further leukemia-directed therapy, attained remission, proceeded to AlloHCT, and remains alive and in remission. Additionally, loss of B-cell aplasia (BCA) was noted in 4 of the 9 CR patients at a median of 3.6 months after T-cell infusion (range: 2.4-5.8). Loss of BCA occurred just before planned AlloHCT (n = 2) or coincided with CD19-positive marrow relapse (n = 2). The remaining 5 patients had ongoing BCA at time of relapse (CD19-negative BM, n = 1; CD19-positive EM, n = 1) or at last evaluation before HCT (n = 3) (Figure 1C). These data suggest that mechanisms of CAR T-cell failure included limited persistence (loss of BCA) and leukemic evasion (antigen loss and EM disease). Evaluation of employed immune evasion mechanisms of leukemic blasts is ongoing. The 3 patients with CD19-positive relapse received treatment with repeat lymphodepletion and CD19-CAR T-cell infusions using the previously manufactured products at a median of 6.4 months (range: 5.8-10.4) from initial infusion (Table 3; supplemental Figure 3). While these patients initially were treated on DL1 with reinfusion, they were treated at DL2 (the determined MTD) and received intensified lymphodepletion (fludarabine [30 mg/m2, days −4 to −2] and cyclophosphamide [500 mg/m2, days −3 to −2]). The one exception to this was the second infusion for patient 6, who received protocol-defined lymphodepletion chemotherapy, CAR T-cell infusion per DL1, and concurrent pembrolizumab. With reinfusion, the 2 patients with loss of BCA/relapse achieved an MRDneg CR, and the third patient (ongoing BCA/EM relapse; patient 4) had NR. Of the 2 patients with MRDneg CR, 1 proceeded to a second AlloHCT and died due to transplant-related complications (patient 5). The other (patient 6) again developed recurrent CD19-positive disease (EM), received a third CD19-CAR T-cell infusion (intensified LD; DL2), and again achieved MRDneg CR. The patient subsequently had a loss of BCA, received a fourth infusion, and again achieved BCA. This patient remains alive and in remission at the time of data cutoff (Table 3; supplemental Figure 3).
Based on prior reports, we classified high disease burden as having ≥40% morphologic blasts on pre-CAR T-cell disease evaluation. Five out of 12 patients had a high disease burden before treatment, and 4 of these were treated at DL2 (Figure 1D). We then focused our analysis on evaluating the influence of disease burden and dose level on clinical response, toxicity, plasma cytokines/chemokines, and CD19-CAR T-cell expansion. All 7 patients with low disease burden had MRDneg CRs, in contrast to only 2 of the 5 patients with high disease burden (Figure 1D). Likewise, severe CRS/NTX (grade 3 to 4) or carHLH occurred only in patients with a high disease burden (Figure 1D). We determined the concentration of 38 plasma cytokines, chemokines, and growth factors by multiplex analysis before and after lymphodepletion and after CAR T-cell infusion (at 2 to 4 hours and 1, 2, 3, and 4 weeks after infusion). Before lymphodepletion, the concentration of fractalkine, MCP-1, and MIP-1β was significantly higher in high disease-burden patients (Figure 2A), with no significant differences observed after lymphodepletion (supplemental Figure 4). Lymphodepletion resulted in a significant increase in individual (fractalkine, IP-10, IL-7, IL-15, and Flt-3L), as well as the sum of all cytokines, chemokines, and growth factors (Figure 2B,C). Low and high disease-burden patients generally clustered into separate groups by principal component analysis both when scaling to unit variance or analyzing absolute concentrations, with segregation in the latter case largely driven by MCP-1, GRO, and IP-10 (Figure 2D). The 4 patients who developed severe CRS or carHLH after CAR T-cell infusion all had high disease burden before infusion and exhibited elevation of total cytokines, chemokines, and growth factors consistently throughout the study observation period (Figure 2E), including cytokines associated with CRS or carHLH (IL-6, IL-1RA, and IL-1) (supplemental Figure 5). CAR T cells expanded in all patients, except for 1 patient with NR (patient 7), as judged by qPCR with no difference between low and high disease burden or dose level (Figure 3A,B; supplemental Figure 6). Peak expansion occurred between 1 and 2 weeks after infusion with subsequent CAR T-cell contraction. We observed no significant difference in peak expansion or area under the curve analysis for the first 4 weeks after CAR T-cell infusion when stratifying by disease burden (Figure 3C,E) or dose level (Figure 3D,F). At 4 weeks after infusion, we detected CD19-CAR T cells in all evaluated BM samples (n = 11) and in 9/10 CSF samples. In the CSF, there was significant enrichment of CD19-CAR T cells as judged by a high CD19-CAR/TRAC locus qPCR ratio (Figure 3G,H). Data on long-term CAR T-cell persistence (qPCR) is limited since 5/9 patients who achieved an MRDneg CR proceeded to consolidative AlloHCT (supplemental Figure 6). At peak expansion, we performed the same phenotypic analysis that was performed on the CAR T-cell product. Compared with the infused product, there was a significant increase in the percentage of CD8+ CD19-CAR T cells, consistent with a preferential expansion of CD8+ vs CD4+ CD19-CAR T cells (Figure 4A). CD19-CAR T-cell subset analysis revealed predominantly effector memory T-cell subsets in both CD4+ and CD8+ CAR T-cell compartments (Figure 4B). Within the effector memory compartment, there was a significant decrease in transitional memory CD8+ CAR T cells (Figure 4C). In regard to exhaustion markers, the frequency of PD1+/TIM3+ cells was significantly increased in the CD8+ CAR T-cell compartment in comparison with the infused CD8+ CAR T cells, with no significant change in CD39 expression (Figure 4D, supplemental Figure 7A). Comparing phenotype and PD1, TIM3, and CD39 expression at peak expansion between CAR-positive and CAR-negative T-cells, we observed a significantly higher percentage of effector memory T cells within the CAR-positive T-cell compartment (Figure 4E,F; supplemental Figure 7B). In addition, the frequency of PD1+/TIM3+ cells within the CAR-positive T-cell compartment was increased (Figure 4G). Subgroup analyses based on low and high disease burden for patients who had achieved a CR did not reveal significant differences in the performed phenotypic analysis (data not shown).
In this phase 1 study of CD19-CAR T-cells with a 41BBζ signaling domain, we observed a high initial response rate, with 9/12 of patients (75%) achieving an MRDneg CR in the BM and 8/12 a CR based on the NCCN pediatric B-ALL response criteria. Notably, approximately half of this patient cohort would not have met inclusion criteria on the tisagenlecleucel registration trial, including (1) age ≤3 years old, (2) prior treatment with CD19-directed therapies, and/or (3) low preinfusion disease burden in the BM. High disease burden (≥40% morphologic blasts) before CAR T-cell infusion correlated with increased side effects and lower response rate, but not with CD19-CAR T-cell expansion. Additionally, cell dose did not impact CAR T-cell expansion. While patients treated at DL2 did experience more side effects, this dose level was also enriched for patients with a high disease burden. Furthermore, our phenotypic analysis revealed that CD8+ CAR T cells had a proliferative advantage over CD4+ CAR T cells, and that expanded CAR T cells had an effector memory phenotype with evidence of antigen-driven differentiation. The incidence of high grade (grade 3 to 4) CRS (16.6%) or NTX (8.3%) was lower than reported in the pivotal trial of tisagenlecleucel in a similar patient population, but is in line with recently reported data using the same product. As highlighted by other investigators, our use of the ASTCT consensus grading for CRS and NTX may partially account for such differences. Additionally, consistent with previous reports of early intervention for CRS,, 3 of our 4 patients with low-grade CRS received therapy with tocilizumab, and none had symptom progression. Furthermore, carHLH is increasingly described in the literature as a complication of CD19-CAR T-cell therapy, with varying diagnostic criteria employed.,25, 26, 27, 28, 29 In our cohort, 2 patients developed carHLH, and their detailed clinical course is described elsewhere. Further studies are needed to inform upon uniform diagnostic criteria and treatment strategies. Our experience highlights the critically important question regarding the role of consolidative HCT after CD19-CAR T-cell therapy. Similar to others, in our cohort, patients with no prior history of AlloHCT who underwent a planned consolidative AlloHCT had improved leukemia-free survival. Consolidative AlloHCT was well tolerated with no treatment-related mortality within the initial 100 days after HCT. Developing an algorithm to decide which patients would benefit from a consolidative AlloHCT after successful CD19-CAR T-cell therapy remains a major challenge. Pretreatment antigen burden is increasingly recognized as an important predictor of both short and long-term response after CD19-CAR T-cell therapy.,, We also found that a high disease burden correlated with a higher risk of early treatment failure. However, the definition of “high disease burden” is not consistent across studies,,,,30, 31, 32 and thus needs to be interpreted in the context of each study. NGS testing is a promising approach to assess the risk of relapse after CAR T-cell therapy, particularly when used in conjunction with close monitoring for loss of BCA.,, In our study, 7 patients with MRDneg CR had available NGS testing, and all were considered negative per study definitions (<10 clones per million). While 4 of these patients proceeded to a planned consolidative AlloHCT, recent data suggest that pediatric and AYA patients with B-ALL who achieve an NGS-negative MRDneg CR and have persistent BCA after CD19-CAR T-cell therapy have excellent outcomes and may require only close monitoring. Clearly, future prospective clinical studies are needed to further inform on clinically relevant definitions of MRD negativity and determine the best management of these patients. Before lymphodepleting chemotherapy, patients with a high disease burden had significantly higher levels of fractalkine, MCP-1, and MIP-1β. Elevated levels of MCP-1 have been reported in pediatric and adult patients with B-ALL,, whereas no data are currently available for fractalkine and MIP-1β. Given the role of these chemokines in inflammation and their impact on numerous immune cells, including NK cells, monocytes, and macrophages, their preinfusion elevations suggest a proinflammatory state. However, we observed no correlation with the development of CRS, ICANS, and/or carHLH. Clearly, larger studies are needed to explore if elevated levels of fractalkine, MCP-1, and/or MIP-1β before infusion correlate with side effects after CAR T-cell infusion. After lymphodepleting chemotherapy, we saw a significant increase in circulating chemokines and cytokines, including IL-15, which is consistent with reports in adult patients who received lymphodepleting chemotherapy before the adoptive transfer of CD19-CAR T cells. We used a standard qPCR assay to track CD19-CAR T cells, which we combined with qPCR analysis for the TRAC locus. Based on our analysis, we found a striking enrichment of CD19-CAR T cells within the CSF, highlighting their ability to traffic to the CNS even in the absence of active disease. The investigated CD19.41BBζ-CAR T-cell product had a high percentage of CD4+ CAR T cells with an effector and central memory phenotype. This is in contrast to 2 other studies, which reported that CAR T-cell products generated with IL-7 and IL-15, or no cytokines, resulted in T-cell products that contain a significant number of naïve-like T cells., Both of these studies used unselected peripheral blood mononuclear cells as starting material, whereas we used CD4/CD8-selected T cells from leukapheresis products. A recent integrative and single-cell RNA sequencing (scRNA-seq) analysis of pediatric autologous leukapheresis products used to generate CD19-CAR T-cell products highlighted the role of transcriptional networks driven by TCF7 in promoting CAR T-cell persistence and the negative impact of IRF7-regulated networks. Additional studies are needed, and we are conducting scRNA-seq and methylome analyses of CD19-CAR T-cell products and sorted CAR T cells after infusion to gain additional insights. Comparing CAR-negative and CAR-positive T-cell populations within the infused cell product revealed significant phenotypic differences, including a higher frequency of CD4+ T cells in the CAR-positive compartment and a higher frequency of PD1+/TIM3+ T cells. This finding is most likely explained by baseline signaling (aka tonic signaling) of the CD19.41BBζ-CAR within CD4+ T cells. Recent studies highlighted that CAR signaling can be blocked with the abl/src family tyrosine kinase inhibitor dasatinib during CAR T-cell production, improving their effector function.42, 43, 44 After infusion, CD8+ CAR T cells expanded to a greater extent than CD4+ CAR T cells. This finding is consistent with a previous study in which a CD19-CAR T-cell product with a defined CD4/CD8 ratio was infused into adults with ALL, highlighting that this observation does not depend on the separate manufacturing of CD4 and CD8 CD19-CAR T cells. Both CAR T-cell populations at peak expansion had a predominantly effector memory phenotype. Similar results have been reported for CD19.CD28ζ-CAR T cells in pediatric patients with B-ALL. Of note, without performing lymph node biopsies, it is impossible to assess if the decrease of central memory CAR T cells at the peak of expansion is due to differentiation or merely reflects the trafficking of central memory T cells to lymphoid tissues. At peak expansion, we observed evidence of antigen-driven differentiation consistent with a loss of developmental potential, with a significantly greater percentage of CAR-positive T cells expressing PD1 in comparison with the infused CAR T-cell product. In addition, at peak expansion, the frequency of PD1+/TIM3+ T cells was significantly higher in the CAR-positive vs CAR-negative compartment. A recent study demonstrated that murine and human effector memory T-cell populations are heterogenous, highlighting the need to perform additional studies to further resolve the functional state of infused CAR T cells. These necessary studies include scRNA-seq and whole-genome bisulfite sequencing,48, 49, 50 which are currently in progress for our patient cohort. In particular, scRNA-seq analysis of CD19.41BBζ-CAR T cells in adult patients suggests that at early time points after infusion, CD19-CAR T cells have a transcriptional profile that is consistent with an effector response and lack the coexpression of PD1 and TIM3. In summary, our phase 1 study shows a favorable toxicity and efficacy profile of our institutional autologous CD19.41BBζ-CAR T-cell product. Our correlative studies highlight the role of disease burden on the outcome and the preferential expansion of CD8+ CAR T cells after infusion. Likewise, we demonstrate antigen-driven CAR T-cell differentiation, providing impetus to evaluate approaches to modulate the effector function of CD19-CAR T cells before and/or after infusion.
Conflict-of-interest disclosure: A. Sharma consults/consulted for Spotlight Therapeutics and Medexus Inc. B.Y. consults/consulted for ElevateBio. S.G. consults/consulted for TESSA Therapeutics, TIDAL, Catamaran, and Novartis. S.G. is a Data Safety and Monitoring Board (DSMB) member of Immatics. M.P.V. is a member of the Medical Advisory Board for the Rally Foundation. M.P.V., B.Y., C.Z., J.C.C., P.G.T., and S.G. have patents/patent applications in he fields of T-cell and/or gene therapy for cancer. The remaining authors declare no competing financial interests. |
||
PMC9647836 | Pau Abrisqueta,Daniel Medina,Guillermo Villacampa,Junyan Lu,Miguel Alcoceba,Julia Carabia,Joan Boix,Barbara Tazón-Vega,Gloria Iacoboni,Sabela Bobillo,Ana Marín-Niebla,Marcos González,Thorsten Zenz,Marta Crespo,Francesc Bosch | A gene expression assay based on chronic lymphocytic leukemia activation in the microenvironment to predict progression | 19-08-2022 | Abstract Several gene expression profiles with a strong correlation with patient outcomes have been previously described in chronic lymphocytic leukemia (CLL), although their applicability as biomarkers in clinical practice has been particularly limited. Here we describe the training and validation of a gene expression signature for predicting early progression in patients with CLL based on the analysis of 200 genes related to microenvironment signaling on the NanoString platform. In the training cohort (n = 154), the CLL15 assay containing a 15-gene signature was associated with the time to first treatment (TtFT) (hazard ratio [HR], 2.83; 95% CI, 2.17-3.68; P < .001). The prognostic value of the CLL15 score (HR, 1.71; 95% CI, 1.15-2.52; P = .007) was further confirmed in an external independent validation cohort (n = 112). Notably, the CLL15 score improved the prognostic capacity over IGHV mutational status and the International Prognostic Score for asymptomatic early-stage (IPS-E) CLL. In multivariate analysis, the CLL15 score (HR, 1.83; 95% CI, 1.32-2.56; P < .001) and the IPS-E CLL (HR, 2.23; 95% CI, 1.59-3.12; P < .001) were independently associated with TtFT. The newly developed and validated CLL15 assay successfully translated previous gene signatures such as the microenvironment signaling into a new gene expression–based assay with prognostic implications in CLL. | A gene expression assay based on chronic lymphocytic leukemia activation in the microenvironment to predict progression
Several gene expression profiles with a strong correlation with patient outcomes have been previously described in chronic lymphocytic leukemia (CLL), although their applicability as biomarkers in clinical practice has been particularly limited. Here we describe the training and validation of a gene expression signature for predicting early progression in patients with CLL based on the analysis of 200 genes related to microenvironment signaling on the NanoString platform. In the training cohort (n = 154), the CLL15 assay containing a 15-gene signature was associated with the time to first treatment (TtFT) (hazard ratio [HR], 2.83; 95% CI, 2.17-3.68; P < .001). The prognostic value of the CLL15 score (HR, 1.71; 95% CI, 1.15-2.52; P = .007) was further confirmed in an external independent validation cohort (n = 112). Notably, the CLL15 score improved the prognostic capacity over IGHV mutational status and the International Prognostic Score for asymptomatic early-stage (IPS-E) CLL. In multivariate analysis, the CLL15 score (HR, 1.83; 95% CI, 1.32-2.56; P < .001) and the IPS-E CLL (HR, 2.23; 95% CI, 1.59-3.12; P < .001) were independently associated with TtFT. The newly developed and validated CLL15 assay successfully translated previous gene signatures such as the microenvironment signaling into a new gene expression–based assay with prognostic implications in CLL.
It is well accepted that patients with chronic lymphocytic leukemia (CLL) who are asymptomatic and in an early clinical phase do not require therapy. Nevertheless, cumulative data on the risk of clonal evolution2, 3, 4 renewed interest in early therapeutic intervention in patients at diagnosis who are likely to progress rapidly. Therefore, the identification of these patients at diagnosis has been an intense focus of clinical research in the field of CLL. Prognostication in this setting has classically relied on a myriad of laboratory values, cytogenetic abnormalities, gene mutations, or the mutational status of the IGHV genes.6, 7, 8, 9, 10 More recently, the International Prognostic Score for Early-stage CLL (IPS-E) has been developed employing 3 covariates: unmutated IGHV, absolute lymphocyte count > 15 × 109/L, and presence of palpable lymph nodes. Despite this extensive investigation, the accuracy of these models may be improved.11, 12, 13, 14, 15, 16 In addition, the emergence of novel targeted agents has attracted interest in the early treatment of patients at high risk of early progression. Gene expression profiles and the clinical course of patients with CLL have been correlated in various studies.,,17, 18, 19, 20, 21, 22, 23, 24, 25, 26 Unfortunately, biomarkers based on gene expression profiles exhibit several caveats that preclude them from being widely applied in the prognostication of patients with CLL. These include the lack of reproducibility and standardization and the complexity of bioinformatics analysis. Significantly, the prognostic value of clustering methods is limited by the fact that the assignment of an individual may vary when different patients are included in the clustering process, thus impeding the use of these methods in real time. In this regard, the development of new platforms that allow direct and reproducible quantification of gene expression, such as NanoString nCounter, should facilitate the attainment of gene expression biomarkers applicable in clinical settings., Among different gene signatures, and because CLL is a malignancy that is particularly dependent on interaction with the microenvironment for survival and proliferation, IGHV mutational status signature,,, and genes involved in the activation of malignant cells in the microenvironment, including stimulation of the B-cell receptor (BCR),24, 25, 26, are of particular interest. Indeed, this notion is reinforced by the standard use of different small molecules targeting CLL-microenvironment interactions, particularly Bruton’s tyrosine kinase inhibitors., Herein, we developed, evaluated, and validated a multigene expression signature using genes associated with the activation of CLL cells in the microenvironment and the IGHV mutational status. This assay, based on the NanoString platform, should facilitate its applicability in clinical settings.
The overall design of the process for developing and evaluating a new assay to assess the risk of progression in patients with CLL is shown in supplemental Figure 1. For the training cohort of the study, 156 untreated samples, 119 from the University Hospital Vall d’Hebron and 37 from the University of Salamanca, were used. The assay was validated using 112 samples from an independent cohort of patients from the German Cancer Research Center, Heidelberg, Germany. The details of the validation cohort have been reported elsewhere. Samples were obtained at diagnosis, whenever possible. For patients who did not have a sample at the time of diagnosis, samples were collected during follow-up but always before the patients received any treatment. Gene expression quantification was performed in blood samples from untreated patients diagnosed with CLL. Peripheral blood mononuclear cells were obtained using Ficoll-Paque Plus density gradient (GE Healthcare, Buckinghamshire, United Kingdom) and subsequently cryopreserved until analysis. Tumor cells were purified using immunomagnetic depletion by EasySep Human B Cell Enrichment Kit (StemCell Technologies), and the final tumor content was assessed by flow cytometry. The estimated median tumor content was 98.3% (range, 80-99.9) in the training cohort and 95.7% (range, 86.8%-99.4%) in the validation cohort. Written informed consent was obtained from all individuals in accordance with the Declaration of Helsinki. The study was approved by the clinical research ethics committee of the Vall d’Hebron Barcelona Hospital Campus.
Gene expression was quantified in 250 ng of RNA on the NanoString platform (NanoString Technologies, WA) using the “high sensitivity” setting on the nCounter PrepStation and 555 fields of view on the nCounter Digital Analyzer. A total of 178 genes were selected from the literature, including genes related to the activation of CLL cells in the microenvironment,23, 24, 25, 26 genes that were differentially expressed according to the mutational status of IGHV,,,, and other genes of prognostic interest in CLL (supplemental Methods, supplemental Table 1). Normalization for RNA loading was performed using the geometric mean of 22 housekeeping genes (supplemental Table 1). The normalized data were log10 transformed. The reference gene selection is further described in the “Data supplement.”
Detailed descriptions of model building and performance assessment are provided in the “Data supplement.” In brief, we used the gene expression data from the training cohort to produce a parsimonious predictive model for time to first treatment (TtFT) using a penalized Cox model. To evaluate the global performance of the multivariate Cox model obtained from the selected genes, different diagnostic parameters were calculated and are summarized in the “Data supplement” (supplemental Table 2), including R2, Brier score, iAUC (a summary measure of the area under the receiver operating characteristic curve calculated for the different times), and Harrell’s C-statistic, a generalization of the AUC., The graph obtained for the AUC values at the different time points is shown in supplemental Figure 2. For illustrative purposes, we dichotomized the predictive gene expression score in 3 risk groups using the R partykit package.
The statistical analysis plan was prespecified before the evaluation of the gene expression in the training and validation cohort. The primary end point of the study was TtFT, defined as the time from the date of obtaining the sample to the date of treatment onset. To study the predictive capacity of the gene expression score, we relaxed the linearity assumption using restricted cubic splines by means of rms R package (Harrell, F. E. Jr Package “rms” [The Comprehensive R Archive Network, 2016]). Harrell’s C-statistic was calculated to compare the discrimination capacities of different models. The analysis of deviance (analysis of variance R function) was used to study whether the inclusion of new factors had a significant improvement in the predictive capacity of the model. Survival curves were estimated using the Kaplan-Meier method to visualize gene expression risk groups and were compared by the log-rank test. Cox proportional-hazard models were used to obtain hazard ratios (HRs) with 95% CIs without dichotomizing continuous factors. To select prognostic variables with the highest impact in TtFT, we performed a least absolute shrinkage and selection operator regression using package glmnet in the R software to build the most parsimonious multivariate model. Imputation of random missing values was carried out via the mice R package (supplemental Table 3). The median follow-up was calculated using the reverse Kaplan-Meier method. All analyses were performed using the R statistical software version 3.6.2.
The training cohort was comprised of 156 patients with previously untreated CLL. The median age of the series was 66 years (range, 34-90 years), and 57% of the patients were men. In total, 37% of samples were obtained at the time of CLL diagnosis, whereas 63% were obtained during the follow-up of patients before any CLL treatment. The median time from CLL diagnosis to sample collection was 11.9 months (95% CI, 7.1-22.6). The analysis of TtFT was calculated from the date of collecting the sample to the date of treatment onset. The main clinical and biological characteristics of the series are shown in Table 1. Ninety-two cases (59%) were IGHV mutated, 54 cases (35%) were IGHV unmutated, and 9 cases (6%) were undetermined because of polyclonal, unproductive, or biclonal rearrangement. In 1 case, no IGHV mutational data were obtained. Digital gene expression for 178 genes of interest and 22 housekeeping genes (supplemental Table 1) was determined in 156 samples from the training cohort. Adequate gene expression was obtained in 154 (99%) samples. Two samples (1%) with not enough quality for expression testing were excluded from the analysis. The expression of 76 genes was significantly associated with TtFT in univariate Cox regression analysis (adjusted P value controlling for false discovery rate [FDR] < .05), and 88 with FDR < .1. A total of 46 genes (FDR < .1) met the prespecified inclusion criteria and were selected for further analysis (see “Methods”). Among them, a total of 15 genes (MYC, ITGA4, CERS6, ZNF471, ZNF667, SEPT10P1, ZAP70, LTK, CCL3, CNR1, EGR2, TNF, IL4R, FGL2, PPBP) were finally selected for a prognostic model of TtFT using a penalized Cox method. In addition, 15 housekeeping genes were selected based on their low variance across the samples. A final model, named CLL15, to predict TtFT in the training cohort was developed using the expression of the 15 predictive genes normalized with the 15 housekeeping genes (Figure 1). Subsequently, a linear equation comprising log-transformed, normalized gene expression levels of the 15 genes multiplied by their respective regression coefficients was established and calculated for each patient of the training cohort to obtain the CLL15 score. The C-statistic for the model was 0.77. Figure 2A shows the shape of the association between the CLL15 score and TtFT risk after relaxing the linearity assumption for continuous variables. As a continuous variable, the CLL15 assay score was associated with TtFT (HR, 2.83; 95% CI, 2.17-3.68; P < .001). To better stratify the risk of progression, the optimal thresholds for defining 3 groups with differentiated outcomes (TtFT) were determined using the R partykit package. The low-risk group (score ≤ 2.718, comprising 55% of the cohort) had a 5-year estimated risk of treatment initiation of 30.5%. In the intermediate-risk group (score ≤ 3.535 and >2.718, comprising 20% of the cohort), the 5-year estimated risk of treatment initiation was 57.8% (HR, 2.67; 95% CI, 1.39-5.10; P = .003). Finally, in the high-risk group (score > 3.535, comprising 25% of the cohort) the 5-year estimated risk of treatment initiation was 93.4% (HR, 10.9; 95% CI, 6.12-19.3; P < .001) (Figure 2B). Notably, the CLL15 score exhibited a similar prognostic capacity in the subgroup of patients with an early clinical stage (n = 116), with a 5-year estimated risk of treatment initiation of 18.2%, 44.8%, and 79.54% in the low-, intermediate-, and high-risk groups, respectively (Figure 2C).
We analyzed the association between the progression risk groups obtained by the CLL15 assay with known biological prognostic factors in CLL, including the most common chromosomal alterations determined by FISH (del17p, del11q, and trisomy 12), the level of protein expression of ZAP-70 and CD38 determined by flow cytometry, the mutations in TP53, NOTCH1, SF3B1, and MYD88 genes, the mutational status of IGHV, CLL-IPI, and the IPS-E CLL score. In the univariate analysis, several factors such as the SF3B1 mutations, IGHV status, the expression of ZAP-70 and CD38 by flow cytometry, clinical stage (RAI and Binet), the CLL-IPI, and the IPS-E score were associated with TtFT (Figure 3). In the final multivariate analysis, the CLL15 score, the IPS-E CLL, and the Binet stage were the only factors that maintained their independent statistical significance (Figure 3). We subsequently explored the introduction of the mutational status of the IGHV (mutated/unmutated) as a variable in an expression model and compared its performance with that of a previous model of only gene expression. The C-statistic for the combined model was 0.79, and the analysis of deviance showed that the addition of IGHV status to the gene expression score (and vice versa) provided significant predictive information (analysis of deviance P < .001). According to these results, the model combining gene expression with the IGHV variable performed better in predicting TtFT than the models of gene expression and IGHV by themselves. In the pairwise multivariate Cox models, both variables, IGHV mutational status, and the categorized groups of progression risk according to the gene expression model contributed prognostically (Figure 2D; supplemental Table 4). The inclusion of the CLL15 score also improved the capacity to predict TtFT of the IPS-E score. Figure 4A shows the increment in discrimination capacity in terms of C-statistic when the CLL15 score was included in the model concurrently with the IPS-E score or IGHV status. Moreover, in pairwise multivariate Cox models, the CLL-IPI and CLL15 also independently contributed to TtFT in the training cohort, with a C-statistic of 0.73 for the CLL-IPI alone and 0.81 for the combination. However, when the IPS-E score was included, the information on the CLL-IPI did not improve the model (supplemental Table 4). Finally, the (1) CLL15 score, (2) IGHV status, and (3) IPS-E score were all independent factors that improved the prediction of TtFT (all analyses of deviance pairwise comparison, P < .01) (Figure 4B).
The CLL15 assay was then validated in cryopreserved samples from 112 patients from an independent cohort from Heidelberg (supplemental Table 5). As a continuous variable, the CLL15 score was significantly associated with TtFT (HR, 1.71; 95% CI, 1.15-2.52; P = .007). Figure 5A shows the association between CLL15 score and TtFT risk after relaxing the linearity assumption in the validation cohort. Using the preestablished cut-off in the training cohort, the assay assigned 22 (19.6%) patients to the low-risk group, 42 (37.5%) to the intermediate-risk group, and 48 (42.9%) to the high-risk group. These 3 groups presented differentiated outcomes with a 60-month estimated risk of treatment initiation of 16.5%, 40%, and 58.1% in the low-, intermediate-, and high-risk groups, respectively (P = .03 overall log-rank test, Figure 5B). Moreover, as observed in the training cohort, the gene expression information, both as a continuous variable and as a risk group, was an independent prognostic factor in the presence of IGHV mutational status (supplemental Table 6). The C-statistic for the IGHV mutational status and the gene expression model was 0.6 and 0.63, respectively, whereas the C-statistic for the combined model was 0.67. As observed in the training cohort, 3 risk groups were identified by combining the CLL15 score and the IGHV mutational status information (supplemental Figure 3). To determine the reproducibility of the CLL15 assay, we selected 9 samples with scores distributed across the assay (low risk, intermediate risk, and high risk). The RNA from each of the samples was run on the CLL15 assay in triplicate, with each run performed on a different NanoString cartridge. The results showed 100% concordance of risk-group assignment across triplicates (supplemental Figure 4), with a standard deviation of 0.073 points.
In this study, we translated a gene expression prognostic signature comprising genes involved in the microenvironment activation and IGHV mutational status into a test applicable to categorize patients into the differentiated risk of progression and requiring treatment for their CLL. The assay demonstrated the ability to identify patients at a high risk of requiring treatment in a short time or with an extremely stable disease. Based on the enormous advances in the biology and treatment of CLL, classical staging systems have been complemented by a plethora of new prognostic parameters based on CLL genetics and biology, including gene expression profiles.,,, Despite the fact that gene expression profiles have been strongly correlated with the clinical course of the patients,,,17, 18, 19, 20, 21, 22, 23, 24, 25, 26 their translational value in clinical practice has been difficult to implement due to methodological reasons. The recent advent of new platforms such as NanoString nCounter, capable of digital, direct quantification on a real-time basis for individual patients, allows the attainment of gene expression analysis in a clinical setting., In this regard, we demonstrated the clinical strength and reproducibility of the CLL15 assay in an independent cohort of previously untreated patients with CLL and its analytical reproducibility by showing a very low variability across repeated measurements. Several in vitro and in vivo data indicate that CLL is a malignancy highly dependent on microenvironment signals for survival and proliferation, with BCR signaling being the most prominent pathway activated in CLL cells isolated from lymph nodes. The role of the microenvironment in CLL pathogenesis has been reinforced when molecules targeting CLL-microenvironment interactions have shown unprecedented therapeutic results., The CLL15 assay included genes coding for cytokines, chemokines, and cytokines receptors such as CCL3, TNF, PPBP, and IL4R; integrins such as ITGA4; and transcription regulatory factors such as MYC and EGR2, which are involved in microenvironment activation in different studies in CLL.23, 24, 25, 26,39, 40, 41 In addition, genes previously reported to be differentially expressed according to the IGHV mutational status, including CERS6, CNR1, FGL2, LTK, SEPT10P1, ZAP70, ZNF471, and ZNF667, were also selected in the CLL15 assay.,,,,41, 42, 43, 44 Notably, the levels of expression of the aforementioned genes could also be regulated in microenvironment activation processes.,, Thus, ZAP-70 expression has been associated with enhanced and prolonged BCR signaling,, higher responsiveness to chemokines [56-58], and enhanced migration of CLL cells,, reinforcing the notion that increased ZAP-70 expression is associated with a more aggressive clinical course of patients with CLL.,, It is worth mentioning that the CLL15 signature kept its predictive value independent of the IGHV mutational status, the CLL-IPI, and the IPS-E CLL score. More importantly, the inclusion of the CLL15 score improved the discrimination capacity to predict TtFT when IGHV or IPS-E was included in the model, suggesting that the CLL15 signature could complement the prognostic value of these other variables. In addition, the combination of the CLL15 and CLL-IPI provided independent predictive information; however, with the inclusion of the IPS-E score, the information of the CLL-IPI did not contribute prognostically to the model. In this sense, the combination of the IPS-E and the CLL15 assay was highly discriminative for the TtFT, with a C-statistic of 0.85. It appears that combining a more clinical–based score, such as the IPS-E, with a molecular score (CLL15) could increase the accuracy of both models. Unfortunately, the IPS-E score was not available for the validation cohort and this comparison could not be validated in this cohort. On the other hand, the combination of IGHV and CLL15 also improved the predictive capacity of the model. Three clearly different risk groups were identified after combining the CLL15 and IGHV status. However, a limited improvement of the C-statistic was observed, and the lower statistical power in the validation cohort did not allow for the validation of all findings. Currently, one of the moving fields is the possibility of early treatment of patients at early stages that are likely to progress within a short period. The selection of these patients is usually based on standard prognostic scores. The usage of more accurate methods for prognostication, such as the CLL15 score, should allow for better identification of patients with an increased risk of early progression and thus support future trials based on risk-adapted therapeutic intervention. In conclusion, the biological prognostication of CLL relies on the use of genetic aberrations together with the mutational status of IGHV. Unfortunately, the use of gene expression profiles has been difficult owing to its technical difficulties and reproducibility, precluding its use in clinical practice. The use of newer and more reproducible methods to assess gene expression could round off well-established prognostic parameters, appraising the entire biological profile for the prognostication of patients with CLL. The study presented herein successfully translates previously described gene expression signatures with strong prognostic value into a new gene expression–based assay, the CLL15, applicable in the routine diagnostic setting.
Conflict-of-interest disclosure: P.A. received honoraria from Janssen, Roche, Celgene, AbbVie, and AstraZeneca. G.V. received research honoraria for speaker activities from MSD and an advisory role from AstraZeneca. M.A. received honoraria for speaker activities from AstraZeneca, an advisory role from Janssen, and nonfinancial support from Janssen, and AbbVie. A.M.-N. received honoraria from Janssen, Roche, Takeda, Gilead, AbbVie, and Celgene for speaker activities and from Janssen, Takeda, Gilead, Kiowa Kirin, AstraZeneca, and Beigene for participating in advisory boards. M.C. received research funding from Janssen, Roche, and AstraZeneca. F.B. received honoraria and research grants from Roche, Celgene, Takeda, AstraZeneca, Novartis, AbbVie, Lilly, Beigene, and Janssen. The remaining authors declare no competing financial interests. |
||
PMC9647847 | Pradnya Kulkarni,Dhrubajyoti Datta,Krishna N. Ganesh | Gemdimethyl Peptide Nucleic Acids (α/β/γ-gdm-PNA): E/Z-Rotamers Influence the Selectivity in the Formation of Parallel/Antiparallel gdm-PNA:DNA/RNA Duplexes | 28-10-2022 | Peptide nucleic acids (PNAs) consist of an aminoethylglycine (aeg) backbone to which the nucleobases are linked through a tertiary amide group and bind to complementary DNA/RNA in a sequence-specific manner. The flexible aeg backbone has been the target for several chemical modifications of the PNA to improve its properties such as specificity, solubility, etc. PNA monomers exhibit a mixture of two rotamers (Z/E) arising from the restricted rotation around the tertiary amide N–CO bond. We have recently demonstrated that achiral gemdimethyl substitution at the α, β, and γ sites on the aeg backbone induces exclusive Z (α-gdm)- or E-rotamer (β-gdm) selectivity at the monomer level. It is now shown that γ/β-gdm-PNA:DNA parallel duplexes are more stable than the analogous antiparallel duplexes, while γ/β-gdm-PNA:RNA antiparallel duplexes are more stable than parallel duplexes. Furthermore, the γ/β-gdm-PNA:RNA duplexes are more stable than the γ/β-gdm-PNA:DNA duplexes. These results with γ/β-gdm-PNA are the reverse of those previously seen with α-gdm-PNA oligomers that stabilized antiparallel α-gdm-PNA:DNA duplexes compared to α-gdm-PNA:RNA duplexes. The stability of antiparallel/parallel PNA:DNA/RNA duplexes is correlated with the preference for Z/E-rotamer selectivity in α/β-gdm-PNA monomers, with Z-rotamers (α-gdm) leading to antiparallel duplexes and E-rotamers (β/γ-gdm) leading to parallel duplexes. The results highlight the role and importance of Z- and E-rotamers in controlling the structural preferences of PNA:DNA/RNA duplexes. | Gemdimethyl Peptide Nucleic Acids (α/β/γ-gdm-PNA): E/Z-Rotamers Influence the Selectivity in the Formation of Parallel/Antiparallel gdm-PNA:DNA/RNA Duplexes
Peptide nucleic acids (PNAs) consist of an aminoethylglycine (aeg) backbone to which the nucleobases are linked through a tertiary amide group and bind to complementary DNA/RNA in a sequence-specific manner. The flexible aeg backbone has been the target for several chemical modifications of the PNA to improve its properties such as specificity, solubility, etc. PNA monomers exhibit a mixture of two rotamers (Z/E) arising from the restricted rotation around the tertiary amide N–CO bond. We have recently demonstrated that achiral gemdimethyl substitution at the α, β, and γ sites on the aeg backbone induces exclusive Z (α-gdm)- or E-rotamer (β-gdm) selectivity at the monomer level. It is now shown that γ/β-gdm-PNA:DNA parallel duplexes are more stable than the analogous antiparallel duplexes, while γ/β-gdm-PNA:RNA antiparallel duplexes are more stable than parallel duplexes. Furthermore, the γ/β-gdm-PNA:RNA duplexes are more stable than the γ/β-gdm-PNA:DNA duplexes. These results with γ/β-gdm-PNA are the reverse of those previously seen with α-gdm-PNA oligomers that stabilized antiparallel α-gdm-PNA:DNA duplexes compared to α-gdm-PNA:RNA duplexes. The stability of antiparallel/parallel PNA:DNA/RNA duplexes is correlated with the preference for Z/E-rotamer selectivity in α/β-gdm-PNA monomers, with Z-rotamers (α-gdm) leading to antiparallel duplexes and E-rotamers (β/γ-gdm) leading to parallel duplexes. The results highlight the role and importance of Z- and E-rotamers in controlling the structural preferences of PNA:DNA/RNA duplexes.
Peptide nucleic acids (PNA) are achiral mimics of nucleic acids composed of a polyamide backbone with recurring units of 2-aminoethylglycine (aeg) to which nucleobases A, C, T, and G are linked via a tertiary amide linkage (Figure 1). Such a backbone with linear and flexible C–C and C–N bonds has the capacity to reorganize slowly into the preferred conformation for hybridization with both complementary DNA or RNA with similar avidity and a high sequence specificity. These characteristics of PNAs combined with their chemical stability to proteases and nucleases and low toxicity provide resourceful applications for in vitro diagnostics and antisense therapeutics. The poor cell penetration of PNAs has prompted a number of chemical modifications to effectively improve their properties for hybridization specificity and biological applications. PNAs and their analogues have also been finding numerous applications in materials science. While the structural role of the linear aeg backbone in modulating hybridization properties of PNAs is well documented, the role of the t-amide group that has restricted rotation around the N–CO bond linking the base with the aeg backbone has not been well studied. Preorganizing the aeg-PNA backbone into a “hybridization-competent conformation” imparts entropic gain in tuning its structure for selective or preferential binding to DNA/RNA. Furthermore, PNA can bind to DNA in either parallel or antiparallel orientations defined by the relative orientations of PNA/DNA strands (Figure 1). Several efforts from this laboratory and that of others on structural modifications of the aminoethylglycine (aeg) PNA backbone have led to a host of acyclic, cyclic, and chiral backbone-modified assortment of PNA analogues. An interesting rational modification of the aeg-PNA backbone is the introduction of cis-1,2-disubstituted cyclopentyl and cyclohexyl moieties to match the dihedral angle of the lone Cβ-Cγ bond in the ethylenediamine segment to 60° as found in the X-ray structure of the PNA:RNA duplex. This results in remarkable discrimination in binding of PNA with isosequential complementary DNA and RNA with preference for binding to RNA. Imparting chirality into the aeg-PNA backbone has also shown selectivity for binding DNA in either parallel or antiparallel modes depending on the nature of modifications. These approaches demonstrate that the presence of steric constrains and chirality may induce tuning of the PNA backbone to populate hybridization-competent conformations for prompting selectivity in binding with DNA/RNA. The introduction of cationic and hydrophilic side chains has led to improved solubility and enhanced cell uptake. A simpler way to impart steric constrains without introducing chirality is to incorporate gem-dimethyl substitution into the flexible aeg backbone of PNA. A number of α-isoaminobutyric acid (aib)-containing polypeptides occur naturally, and the gem-dimethyl substitution on α-carbon of glycine is well-known to promote helicity in peptides. This feature provides us a rationale for the design, synthesis, and study of α-gemdimethyl(gdm)-PNA (Figure 2a) having gdm in the glycine segment of aminoethylglycine PNA. The α-gdm-PNA-Tn homo-oligomer exhibits significant stabilization of the derived triplex (α-gdmPNA-Tn)2:dAn, and duplex α-gdm-PNA:DNA/RNA from the mixed-base sequence shows a higher Tm relative to that of analogous isosequential RNA duplexes. These results motivated us to explore the specific properties of gdm substitution at the β- and γ-sites on the aeg backbone in PNA-T monomers. Unsubstituted aeg-PNA monomers exhibited a mixture of two rotamers (major Z:minor E, 60:40) arising from the restricted rotation around the N–CO t-amide bond, with the Z-rotamer corresponding to δ-carbonyl pointing to glycine (C-terminus) and E-rotamer having δ-carbonyl toward the ethylenediamine (N-terminus) side (Figure 2b). In gdm-PNA T monomers, it was surprisingly found that the α-gdm substitution exhibited exclusively the Z-rotamer (Figure 2a), while the β-gdm substitution showed exclusively the E-rotamer (Figure 2c). The γ-gdm substitution (Figure 2d) gave a mixture of minor Z- and major E-rotamers, reverse of that seen in unsubstituted aeg-PNA monomers. The relative stabilities of Z- and E-rotamers (Figures 2e and f) were dictated by balance of repulsive steric clashes (red) and nonbonding attractive n → π* or C–H···O interactions (green). Thus, the Z-rotamer (Figure 2e/Z) was favored for the α-gdm-PNA monomer and the E-rotamer (Figure 2f/E) is favored for the β-gdm-PNA monomer. Although at the monomer level, aeg-PNA shows a mixture of Z- and E-rotamers, the crystal structure studies at the oligomer level and in duplexes show that the t-amide bond is always present as Z-rotamer in antiparallel duplexes. PNA analogues with substitutions (both cyclic and acyclic) at Cγ and Cα positions are well known, and examples of acyclic PNAs bearing substitution at the Cβ position are rare. With the β-gdm-PNA T monomer showing exclusively E-rotamers, it would be interesting to study its comparative hybridization and orientation specificity with complementary DNA/RNA at the oligomer level. The work reported in this article demonstrates that the nature of the Z/E-rotamer determines the orientation bias in binding to DNA/RNA and interestingly plays a role in preferentially stabilizing parallel or antiparallel duplexes. While α-gdm-PNA with exclusive Z-rotamer stabilized antiparallel duplexes, it is shown here that β-gdm-PNA having the E-rotamer predominantly stabilized parallel duplexes over the antiparallel duplexes, reverse of that found in unsubstituted aeg-PNA.
The β-gdm (Figure 2c) and γ-gdm (Figure 2d) aeg PNA monomers synthesized as reported before were used for the solid-phase synthesis of PNA oligomers on MBHA (4-methyl benzhydrylamine) resin using a Boc strategy starting from the C-terminus to the N-terminus. The orthogonally protected (Boc/Cbz)-l-lysine was first coupled to the resin at the C-terminus, and the synthesis was continued using protected monomers (c) or (d) for coupling. The in situ activation of the acid function was done with PyBOP. The completion of each coupling reaction was monitored by Kaiser’s test followed by deprotection of the N-t-Boc group using TFA. The Cγ- and Cβ-gdm-T monomers (Figure 2c,d) were incorporated at the T-sites in the aeg-PNA sequence H-GTAGATCACT-NH2 having three equally spaced thymine residues (Table 1). The same sequence was chosen as that studied earlier incorporating Cα-gdm substitution to enable the relative comparison of α-, β-, and γ-gdm substituent-dependent effects on duplex formation by the gdm-PNA oligomers. To study the sequence-dependent effects of Cβ/Cγ-gdm substitutions, PNA oligomers with single modification at the N-terminus (T2), middle (T6), and C-terminus (T10) along with double (T2, T6) and triple (T2, T6, T10) were synthesized (Table 1). After the synthesis, the PNA oligomers were cleaved from the resin by treatment with TFA/TFMSA, which also deprotected the bases, purified by reverse-phase HPLC, and characterized by MALDI-TOF mass spectral data (Supporting Information).
The antiparallel and parallel duplexes with complementary DNA were obtained from γ-gdm-PNA oligomers (PNA 2–PNA 5) and Cβ-gdm-PNA oligomers (PNA 6–PNA 9) by stoichiometric mixing of each PNA oligomer with DNA 1 (antiparallel) and DNA 2 (parallel), respectively (Figure 3). The Tm of each duplex was determined by temperature-dependent changes in the UV absorbance (UV-T plot), which exhibited typical sigmoidal behavior (Figure 4), suggesting the successful formation of duplexes in both parallel and antiparallel orientations. The thermal stabilities (Tm) of duplexes of all PNAs (PNA 1–PNA 9) with complementary antiparallel DNA 1 and parallel DNA 2 corresponding to the midpoint of biphasic transitions in UV–temperature plots (Figure 4) were obtained accurately from the maxima in their first derivative curves (Supporting Information) and are shown in Table 2.
The unmodified antiparallel duplex aeg-PNA 1:DNA 1 showed a Tm of 49.1 °C, and the Tm of γ-gdm-modified PNA:DNA duplexes showed a decrease as a function of the number and site of modifications (Figure 4A and Table 2). The destabilization of the duplexes compared to the unmodified PNA 1:DNA 1 duplex was as follows: single substitution at the N-terminus (t2, PNA 2:DNA 1) by 9.3 °C and substitution at the middle (t6, PNA 3:DNA 1) by 7.3 °C, (Table 2, entries 2 and 3), double substitution (t2,6, PNA 4:DNA 1) by 8.7 °C (Table 2, entry 4), and triple substitution (t2,6,10, PNA 5:DNA1) by 10.9 °C (Table 2, columns DNA 1, entries 1–5). In the case of parallel duplexes, (Figure 4B) the unmodified PNA 1:DNA 2 duplex showed a Tm of 41.1 °C destabilized by 8 °C compared to that of its antiparallel duplex PNA 1:DNA 1 as is known for PNA:DNA duplexes. However, in contrast to the antiparallel duplexes, the Cγ-gdm-substituted PNAs (PNA 2–PNA 5) actually enhanced the Tm of the derived parallel duplexes with DNA 2 compared to that of unsubstituted parallel duplex PNA 1:DNA 2. The stabilization increased as follows: N-terminus (t2, PNA 2:DNA 2; +4.3 °C), middle (t6, PNA 3:DNA 2; +1.9 °C), double (t2,6, PNA 4:DNA 2; +5.0 °C), and triple (t2,6,10, PNA 5:DNA 2; +7.2 °C) with enhancement in Tm amounting to ∼2.4 °C/modification (Table 2, columns DNA 2).
The UV–thermal stability melting curves for β-gdm PNA:DNA antiparallel and parallel duplexes are shown in Figure 4C,D, respectively, with the corresponding Tm values given in Table 2 (entries 6–9, DNA 1 column). The data indicated that successive substitutions of the β-gdm unit into the PNA oligomer destabilized the antiparallel duplexes with DNA similar to that seen with the corresponding γ-gdm PNA oligomers. Surprisingly, maximum destabilization was seen with single middle substitution (t6, PNA 7:DNA 1; −12.5 °C) compared to that at the N-terminus (t2, PNA 6:DNA 1; −3.0 °C) and double modifications (t2,6, PNA 8:DNA 1; −3.6 °C and t6,10, PNA 9:DNA 1; −10.5 °C) (Table 2, entries 6–9, DNA columns). In contrast, the parallel β-gdm PNA:DNA 2 duplexes became stabilized with successive β-gdm substitutions at t2 (PNA 6:DNA 2; +5.1 °C), t6 (PNA 7:DNA 2; +11.7 °C), t2,6 (PNA 8:DNA 2; +10.8 °C), and t6,10 (PNA 9:DNA 2; +2.8 °C) compared to the unsubstituted PNA 1:DNA 2 duplex (Table 2, entries 6–9, DNA 2 column). The overall data in Table 2 indicate that while the antiparallel duplex is more stable than the parallel duplex in unsubstituted PNA, with both γ- and β-gdm-PNA:DNA duplexes, a reverse trend is observed: the parallel duplexes are more stable than the analogous antiparallel duplexes. This is a notable result on the effect of position-dependent gdm substitution in the aeg-PNA backbone.
The comparative hybridization properties of γ/β-gdm-PNA with RNA 1 in the antiparallel orientation and RNA 2 in the parallel orientation (Table 2) were similarly examined by temperature-dependent UV absorbance (Figure 5). The sigmoidal behavior of UV-T plots typical of two-state transitions confirmed the formation of both parallel and antiparallel duplexes, and Tm values were obtained from the maxima in the first derivative plots (Supporting Information). Increasing the substitution of γ-gdm units destabilized the antiparallel duplexes with RNA 1 (Table 2, entries 2–5, RNA 1 column, −0.6 to −8.9 °C) compared to the unsubstituted aeg-PNA 1:RNA 1 duplex (entry 1). However, the corresponding parallel duplexes γ-gdm-PNA: RNA 2 did not show much variation (Table 2, entries 2–5, −1.2 to +2.0 °C; RNA 2 column). The antiparallel β-gdm-PNA:RNA 1 duplexes exhibited destabilization compared to the unsubstituted PNA:RNA 1 duplex (Table 2, entries 6–9, −2.2 to −19.4 °C, RNA 1 column,), much more than that of the analogous γ-gdm-PNA:RNA 1 antiparallel duplexes (Table 2, entries 2–5, −0.6 to −8.9 °C, RNA 1 column). The corresponding parallel duplexes of β-gdm-PNA with RNA 2 (Table 2, entries 6–9, −9.7 to +1.1 °C, RNA 2 column) were destabilized to a greater degree compared to the more stable γ-gdm-PNA:RNA 2 duplexes (Table 2, entries 2–5, −1.2 to +2.0 °C, RNA 2 column). Thus, in the case of γ/β-gdm PNAs, the antiparallel duplexes with RNA 1 were significantly destabilized and the equivalent parallel duplexes with RNA 2 were marginally destabilized compared to aeg-PNA:RNA duplexes. Overall, the γ/β-gdm PNA:RNA antiparallel duplexes were more stable than the parallel duplexes, which was the reverse of that seen for γ/β-gdm PNA:DNA duplexes. This clearly suggested the effect of the gdm substitutions on their relative hybridization stabilities with DNA and RNA, parallel DNA duplexes > antiparallel DNA duplexes and parallel RNA duplexes < antiparallel RNA duplexes (only for β).
CD spectroscopy shows the characteristic spectra for different PNA:DNA/RNA duplexes. Both γ- and β-gdm PNA:DNA antiparallel duplexes (Figure 6A/C) are characterized by major positive bands of comparable intensities at 220 and 260 nm, the latter band being broad with a shoulder at 275 nm. The negative bands are of high intensity at 210 nm, with weak to moderate intensity at 240 nm. In the case of β-gdm PNA:DNA antiparallel duplexes (Figure 6C), the band intensities are weak in monosubstituted oligomers and became prominent with disubstituted analogues. In comparison, all γ-gdm PNA:DNA parallel duplexes showed a high intensity positive band at 220 nm, followed by a much weaker positive band at 260 nm, without much effects of the degree of substitution (Figure 6B). The broad positive shoulder at 280 nm seen in antiparallel duplexes is noticeably weaker or absent in parallel duplexes and replaced by a weak negative band (Figure 6B). A similar pattern is seen for parallel β-gdm duplexes, with the intensities of bands at 220 nm increasing in disubstituted oligomers (Figure 6D). While parallel (Figure 6B,D) and antiparallel (Figure 6A,C) duplexes show distinct differences in the CD profile among them, within each type, the CD profiles are similar to those of unsubstituted aeg-PNA:DNA duplexes. This indicates that no major conformational differences are induced by γ/β-gdm substitutions compared to analogous aeg-PNA duplexes.
The temperature-dependent UV absorbance data (Table 2) showed variation in the thermal stabilities of γ/β-gdm-PNA:DNA antiparallel and parallel duplexes depending on the type and number of gdm modifications incorporated into PNA oligomers (Figure 7). As compared to the unmodified aeg-PNA:DNA duplexes, the γ/β-gdm-PNA:DNA parallel duplexes were stabilized (Figure 7, red bars) and antiparallel duplexes were destabilized (Figure 6, blue bars). Generally, β-gdm-PNA:DNA 1 parallel duplexes (Figure 7B) were more stable than γ-gdm-PNA:DNA 1 (Figure 7A), and the destabilization of antiparallel duplexes was dependent on the site of substitution, with gdm-PNAs having middle modifications being more destabilized. In comparison, γ/β-gdm-PNA:RNA duplexes (Figure 8) exhibited low or negligible stabilization of γ-gdm-PNA:RNA antiparallel duplexes (ΔTm, < 2.0 °C) decreasing with increasing substitution, while γ-gdm-PNA:RNA parallel duplexes became significantly destabilized compared to aeg-PNA:RNA duplex (Figure 8A). The β-gdm-PNA:RNA duplexes uniformly showed significant destabilization of both parallel and antiparallel duplexes relative to that of analogous aeg:PNA:RNA duplexes (Figure 8B). Inspite of destabilization with respect to unsubstituted aeg-PNA:RNA duplexes, the γ/β-gdm-PNA:RNA antiparallel duplexes showed higher stability than that of parallel duplexes, a reverse trend of that seen in γ/β-gdm-PNA:DNA duplexes. In contrast, α-gdm-PNAs showed an opposite preference of binding to RNA rather than to DNA. The triple-modified identical sequence H-GtAGAtCACt-LysNH2 (t = α-gdm) stabilized the antiparallel DNA duplex over the parallel duplex by 24.4 °C and antiparallel RNA duplex over the parallel duplex by 40.0 °C. The α-gdm-PNA:RNA duplex had higher stability than that of the α-gdm-PNA:DNA duplex by 3.0 °C for the parallel orientation, while it was almost similar for the antiparallel duplex.
The achiral α/β/γ-gdm-dimethyl groups in gdm-PNAs introduce steric rigidity into the somewhat conformationally flexible aeg-backbone of PNA. The nature and extent of conformational constrains in the gdm-backbone and its effects on the stability of duplexes with DNA/RNA are clearly dependent on the α/β/γ-site of substitution. The α-gdm-PNA with achiral gemdimethyl substitution on Cα in the glycine segment stabilized the derived (homo-PNA-Tn)2:(homo-An-DNA/RNA) triplexes compared to the unsubstituted aeg-PNA triplexes. The mixed-base PNA sequence incorporating α-gdm-T substitutions stabilized the resulting α-gdm-PNA:DNA/RNA duplexes compared to that with isosequential DNA/RNA, with the antiparallel duplexes exhibiting higher Tm values than those of aeg-PNA duplexes. In comparison, the β/γ-gdm PNAs reported here exhibited an interesting trend of (i) parallel DNA and RNA duplexes being more stable than aeg-PNA:DNA/RNA duplexes, (ii) Tm of the parallel DNA duplexes higher than Tm of antiparallel DNA duplexes, (iii) Tm of antiparallel RNA duplexes higher than that of parallel RNA duplexes, (iv) for antiparallel duplexes, RNA duplexes with higher melting than that of DNA duplexes, and (v) for parallel duplexes, DNA duplexes with higher melting than that of RNA duplexes. The relative preferences in the binding of α/β/γ-gdm-PNAs with DNA/RNA and in the parallel/antiparallel fashion may arise from the conformational preorganization of the PNA backbone to assume a favorable hybridization-competent conformation. An interesting attribute seen in α-gdm-PNA and β-gdm-PNA oligomers is that the former stabilizes the DNA duplex in the antiparallel orientation, while the latter stabilizes DNA duplexes in the parallel orientation. The conformational constrain in α/β/γ-gdm-PNA is caused briefly by the Thorpe–Ingold effect exerted by gemdimethyl groups, which may alter the backbone dihedral angles and thereby influence the PNA E/Z-rotamer population. The combined steric and electronic factors vary with the site of substitution, inducing different conformational stabilities of E/Z-rotamers as experimentally observed by us in 1H NMR of α-, β-, and γ-gdm-PNA-T monomers and by computational results. The β-gdm-PNA-T monomer shows exclusively E-rotamers in solution, while the γ-gdm-PNA-T monomer crystallized as E-rotamers (Figure 2). In crystal structures with the base-modified cyanuryl aeg-PNA monomer and in α-mono- or β/γ-disubstituted cyclic aeg-PNAs, the Z-rotamer has been exclusively observed. The major Z-rotamer seen for aeg-PNA at the monomer level leads to antiparallel PNA duplexes containing exclusively the Z-rotamer. Hence, if the exclusively seen E-rotamer for the β-gdm-PNA monomer also prevails in the corresponding PNA:DNA oligomers that show higher stability parallel duplexes, it is tempting to suggest that the E-rotamer may lead to preferential parallel duplexes, while the Z-rotamer results in antiparallel duplexes. Further structural and computational studies are needed at the oligomer level to verify the existence of such a correlation. It is relevant to point out that in olefinic OPA-PNA in which the tertiary amide group is replaced by C=C that restricts the structure to the E-type isomer, the parallel duplex is more stable than the antiparallel duplex. The observed higher stability of parallel duplexes of the mixed sequence over that of antiparallel duplexes in β/γ-gdm PNAs is so far unusual in the PNA literature. A comprehensive comparison of the relative thermal stabilities ΔTm (Tmp – Tmap) of parallel/antiparallel DNA/RNA duplexes of γ-gdm and β-gdm-PNA is depicted in Figure 9. In general, the results indicate that with γ- and β-gdm PNAs, (i) DNA parallel duplexes are more stable than antiparallel duplexes [blue bars, +(p-ap)], (ii) RNA antiparallel duplexes show higher stability compared to that of DNA duplexes [green bars, -(p-ap)], and (iii) increased substitutions enhance parallel duplex stability. As established in the literature, modifications in the center of the sequences exert maximum destabilization effects, which follows the order t6 > t2,6 > t10 > t6,10. The CD spectral profiles clearly show difference between parallel and antiparallel duplexes and confirm the thermal stability results.
We have demonstrated that the introduction of appropriate rigidity in the form of gemdimethyl substitution into the aeg backbone of PNA without imparting chirality improves not only the DNA/RNA binding properties of PNA but also its selectivity for DNA and prime preference for parallel duplex formation. The results presented here complement our earlier work that had demonstrated that chiral cyclohexanyl-PNAs favor binding with RNA compared with DNA and α-gdm-PNAs showing reversed preference for binding DNA to RNA. Although true validation should be derived from completely modified oligomers, significant selectivity is seen even in the partially substituted gdm-PNA oligomers. The present work advances the concept of engineering affinity and selectivity in PNA:DNA/RNA binding by structural tuning of the gemdimethyl substitution site on the backbone to yield specific Z- or E-rotamers. Such structural consequences of the sterically rigid achiral gemdimethyl groups in preorganizing the backbone conformation are favorable for complexation with DNA or RNA. The simple achiral gemdimethyl substitution to tune the PNA rotamer formation adds a new repertoire to the arsenal of PNA modifications. Recently, di-guaninyl PNA has attracted attention in terms of unusual light emitting properties due to a combination of base stacking and H-bonding interactions, facilitated by the Z-rotamer. It would be interesting to study such structure–light emitting property interactions induced by the E-rotamer through gemdimethyl substitutions, which may change the base stacking interactions. One can also extend the concept by simultaneous introduction of gemdimethyl substitutions at two sites on the backbone or structurally combining with the recently reported bimodal PNAs. The possibility of generating parallel duplexes with selectivity for DNA or RNA may also lead to novel platforms for biomolecular engineering of PNA, affording novel nanoassemblies. The introduction of gemdimethyl substituents also changes the overall hydrophobicity of PNA oligomers, thereby influencing their cell penetration properties, and may lead to newer motifs for molecular assemblies in biomedical applications, in view of the well-documented physicochemical, structural, and biological properties imparted by gem-dimethyl groups in medicinal chemistry.
The aeg-PNA, β-gdm-PNA-T, and γ-gdm-PNA-T monomers were synthesized by previously reported procedures and used for solid-phase synthesis. Mass spectra were obtained using an Applied Biosystems 4800 Plus MALDI-TOF/TOF mass spectrometer using TiO2 or 2,5-dihydroxybencoic acid (DHB), and the integrity of the PNA oligomer was checked on the same instrument using DHB or CHCA as the matrix. High-resolution mass spectra for final PNA monomers were recorded on a Synapt G2 high-definition mass spectrometer.
The aeg-PNA monomers and γ/β-gdm-PNA-T monomers were incorporated into the 10-mer PNA oligomer sequence H-GTAGATCACT-LysNH2 using standard solid-phase protocol on l-lysine-derivatized MBHA resin. The amine content on the resin was suitably lowered from 2 mmol/g to 0.35 mmol/g by partial acylation of amino groups using a calculated amount of acetic anhydride. The desired PNA monomers were coupled sequentially to assemble the PNA sequence using PyBOP and DIPEA in DMF/NMP as coupling reagents. The PNA oligomers were synthesized using repetitive cycles, each comprising the following steps: (i) deprotection of the N-t-Boc group using 50% TFA in DCM (3 × 15 min), (ii) washing of beads with DCM, DMF, and again DCM (thrice each), (iii) neutralization of the TFA salt of amine using 10% DIPEA in DCM to liberate free amine (3 × 10 min), (iv) washing of beads with DCM and DMF (thrice with each solvent), and (v) coupling of the free amine with activated carboxyl of the incoming monomer (3 equiv) using coupling agents. The coupling reactions were carried out in DMF/NMP with PyBOP as coupling reagents in the presence of DIPEA. Capping (when needed) of the unreacted amino groups was done using Ac2O in pyridine:DCM. After each coupling and deprotection step, Kaiser’s test was done for the confirmation of PNA chain elongation using the following reagents: (a) ninhydrin (5.0 g) dissolved in ethanol (100 mL), (b) phenol (80.0 mg) dissolved in ethanol (20 mL), and (c) potassium cyanide (2 mL, 0.001 M aq. solution) in pyridine (98 mL). After every coupling, few resin beads were taken in a test tube, washed with ethanol and 3–4 drops of each of the above-mentioned solutions were added. The test tube was heated for 1–2 min, and the emergence of color indicated the free uncoupled amine on the resin. In such a case, the coupling was continued with a second addition to complete the reaction.
The MBHA resin (10 mg) carrying the synthesized PNA oligomers was stirred with thioanisole (20 μL) and 1,2-ethanedithiol (8 μL) in an ice bath for 10 min. TFA (200 μL) was added under cooling and kept in an ice bath. TFMSA (16 μL) was added slowly with stirring to dissipate the generated heat. The reaction mixture was stirred for 1.5–2 h at room temperature. The resin was removed by filtration under reduced pressure and washed twice with TFA. The filtrate was combined and evaporated on a rotary evaporator at ambient temperature. The residue was transferred to an Eppendorf tube, and the peptide was precipitated by trituration with cold dry ether. The peptide was isolated by centrifugation, and the precipitate was dissolved in deionized water and used for purification.
PNA oligomers were purified on an Agilent HPLC system using a semipreparative BEH130 C18 (10 mm × 250 mm) column. The elution was done using water and CH3CN employing gradient elution: solvent A = 0.1% TFA in CH3CN:H2O (5:95); solvent B = 0.1% TFA in CH3CN:H2O (1:1), A to 100% B in 45 min; and flow rate of 3 mL/min with monitoring of eluants at 254 nm.
UV melting experiments were carried out on a Varian Cary 300 UV spectrophotometer equipped with a Peltier heating system. The samples were prepared in sodium cacodylate buffer (10 mM) and NaCl (10 mM); pH 7.2. Individual PNA oligomers and complementary DNA/RNA were mixed in a stoichiometric ratio (1:1, duplex) to achieve a final concentration of 2 μM for each strand. The samples were annealed by heating the mixture at 90 °C for 10 min and cooled slowly to room temperature over 8–10 h, followed by refrigeration for 24 h. The samples (500 μL) were transferred to a quartz cell and equilibrated at room temperature for 5 min. The absorbance at 260 nm was recorded in steps from 20 to 90 °C with a temperature increase of 0.5 °C. The absorbance plotted at 260 nm as a function of the temperature was fitted by a sigmoidal curve, with the R square value in range of 0.96–0.99. The data were processed using OriginPro 8.5. The Tm was determined from the first derivative of normalized absorbance with respect to the temperature and is accurate to ±1.0 °C. Each melting experiment was repeated thrice. The concentrations of all oligonucleotides were calculated on the basis of absorbance at 260 nm from the molar extinction coefficients of the corresponding nucleobases: T = 8.8 cm2/μmol; C = 6.6 cm2/μmol; G = 11.7 cm2/μmol, and A = 13.7 cm2/μmol as per the literature.
CD spectra were recorded on a JASCO J-815 spectropolarimeter. The PNA:DNA samples were prepared by stoichiometric mixing and annealing in the same way as for UV spectral studies. The CD spectra of bm-PNA:DNA complexes were recorded with samples in a 2 mm cell at a temperature of 10 °C, scanning from 300 to 200 nm using a resolution of 0.1 nm, bandwidth of 1 nm, sensitivity of 2 mdeg, response of 2 s, and scan speed of 50 nm/min. Final spectra are shown as the addition of three scans. |
||
PMC9647849 | Yura Jang,Thujitha Thuraisamy,Javier Redding‐Ochoa,Olga Pletnikova,Juan C. Troncoso,Liana S. Rosenthal,Ted M. Dawson,Alexander Y. Pantelyat,Chan Hyun Na | Mass spectrometry‐based proteomics analysis of human globus pallidus from progressive supranuclear palsy patients discovers multiple disease pathways | 10-11-2022 | electron transport chain complex,globus pallidus,mass spectrometry,progressive supranuclear palsy,proteomics,PSP | Abstract Background Progressive supranuclear palsy (PSP) is a neurodegenerative disorder clinically characterized by progressive postural instability, supranuclear gaze palsy, parkinsonism, and cognitive decline caused by degeneration in specific areas of the brain including globus pallidus (GP), substantia nigra, and subthalamic nucleus. However, the pathogenetic mechanism of PSP remains unclear to date.Unbiased global proteome analysis of patients' brain samples is an important step toward understanding PSP pathogenesis, as proteins serve as workhorses and building blocks of the cell. Methods In this study, we conducted unbiased mass spectrometry‐based global proteome analysis of GP samples from 15 PSP patients, 15 Parkinson disease (PD) patients, and 15 healthy control (HC) individuals. To analyze 45 samples, we conducted 5 batches of 11‐plex isobaric tandem mass tag (TMT)‐based multiplexing experiments. The identified proteins were subjected to statistical analysis, such as a permutation‐based statistical analysis in the significance analysis of microarray (SAM) method and bootstrap receiver operating characteristic curve (ROC)‐based statistical analysis. Subsequently, we conducted bioinformatics analyses using gene set enrichment analysis, Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) protein‐protein interaction (PPI) analysis, and weighted gene co‐expression network analysis (WGCNA). Results We have identified 10,231 proteins with ∼1,000 differentially expressed proteins. The gene set enrichment analysis results showed that the PD pathway was the most highly enriched, followed by pathways for oxidative phosphorylation, Alzheimer disease, Huntington disease, and non‐alcoholic fatty liver disease (NAFLD) when PSP was compared to HC or PD. Most of the proteins enriched in the gene set enrichment analysis were mitochondrial proteins such as cytochrome c oxidase, NADH dehydrogenase, acyl carrier protein, succinate dehydrogenase, ADP/ATP translocase, cytochrome b‐c1 complex, and/or ATP synthase. Strikingly, all of the enriched mitochondrial proteins in the PD pathway were downregulated in PSP compared to both HC and PD. The subsequent STRING PPI analysis and the WGCNA further supported that the mitochondrial proteins were the most highly enriched in PSP. Conclusion Our study showed that the mitochondrial respiratory electron transport chain complex was the key proteins that were dysregulated in GP of PSP, suggesting that the mitochondrial respiratory electron transport chain complex could potentially be involved in the pathogenesis of PSP. This is the first global proteome analysis of human GP from PSP patients, and this study paves the way to understanding the mechanistic pathogenesis of PSP. | Mass spectrometry‐based proteomics analysis of human globus pallidus from progressive supranuclear palsy patients discovers multiple disease pathways
Progressive supranuclear palsy (PSP) is a neurodegenerative disorder clinically characterized by progressive postural instability, supranuclear gaze palsy, parkinsonism, and cognitive decline caused by degeneration in specific areas of the brain including globus pallidus (GP), substantia nigra, and subthalamic nucleus. However, the pathogenetic mechanism of PSP remains unclear to date.Unbiased global proteome analysis of patients' brain samples is an important step toward understanding PSP pathogenesis, as proteins serve as workhorses and building blocks of the cell.
In this study, we conducted unbiased mass spectrometry‐based global proteome analysis of GP samples from 15 PSP patients, 15 Parkinson disease (PD) patients, and 15 healthy control (HC) individuals. To analyze 45 samples, we conducted 5 batches of 11‐plex isobaric tandem mass tag (TMT)‐based multiplexing experiments. The identified proteins were subjected to statistical analysis, such as a permutation‐based statistical analysis in the significance analysis of microarray (SAM) method and bootstrap receiver operating characteristic curve (ROC)‐based statistical analysis. Subsequently, we conducted bioinformatics analyses using gene set enrichment analysis, Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) protein‐protein interaction (PPI) analysis, and weighted gene co‐expression network analysis (WGCNA).
We have identified 10,231 proteins with ∼1,000 differentially expressed proteins. The gene set enrichment analysis results showed that the PD pathway was the most highly enriched, followed by pathways for oxidative phosphorylation, Alzheimer disease, Huntington disease, and non‐alcoholic fatty liver disease (NAFLD) when PSP was compared to HC or PD. Most of the proteins enriched in the gene set enrichment analysis were mitochondrial proteins such as cytochrome c oxidase, NADH dehydrogenase, acyl carrier protein, succinate dehydrogenase, ADP/ATP translocase, cytochrome b‐c1 complex, and/or ATP synthase. Strikingly, all of the enriched mitochondrial proteins in the PD pathway were downregulated in PSP compared to both HC and PD. The subsequent STRING PPI analysis and the WGCNA further supported that the mitochondrial proteins were the most highly enriched in PSP.
Our study showed that the mitochondrial respiratory electron transport chain complex was the key proteins that were dysregulated in GP of PSP, suggesting that the mitochondrial respiratory electron transport chain complex could potentially be involved in the pathogenesis of PSP. This is the first global proteome analysis of human GP from PSP patients, and this study paves the way to understanding the mechanistic pathogenesis of PSP.
Progressive supranuclear palsy (PSP) is a neurodegenerative disease clinically characterised by progressive parkinsonism, postural instability, vertical saccade slowing, supranuclear gaze palsy, and cognitive decline. PSP affects movement, gait, balance, speech, swallowing, vision, eye movement, mood, behaviour, and cognition. , The estimated prevalence of PSP is about five to six per 100,000 worldwide, and symptoms typically begin after the age of 60. , The disease is induced by the gradual degeneration of cells in specific areas of the brain, including the globus pallidus (GP), substantia nigra, subthalamic nucleus, and frontal lobes. , , The pathological hallmarks of PSP include the accumulation of four‐repeat tau proteins in neurofibrillary tangles, neuropil threads, and tau‐positive astrocytes. The underlying mechanisms of pathogenesis in PSP remain unclear. PSP is generally considered sporadic, but rare familial clusters have also been reported, and more than 10 genes with known mutations linked to PSP have been reported. The most studied gene in PSP is the microtubule‐associated protein tau (MAPT), which is expressed and regulated by alternative splicing in the human brain. MAPT H1 haplotype homozygosity significantly predisposes to PSP, and MAPT mutations cause familial PSP with monogenic autosomal dominant inheritance. Mutations in leucine‐rich repeat kinase 2 (LRRK2), which is known as one of the most common genetic causes of Parkinson's disease (PD), are also suggested as a cause of PSP, although more association studies are required. , , , Mutations of dynactin subunit 1 (DCTN1), which is one of the largest subunits in the dynactin family and is involved in cellular functions such as cell division and transport, were observed in patients with a clinical phenotype of PSP. Other genes with potential links to PSP include bassoon (BSN), chromosome 9 open reading frame 72 (C9orf72), eukaryotic translation initiation factor 2‐alpha kinase 3 (EIF2AK3), progranulin (GRN), myelin‐associated oligodendrocyte basic protein (MOBP), Niemann–Pick disease type C1 (NPC1), parkin (PARK2), Syntaxin 6 (STX6), TANK‐binding kinase 1 (TBK1), transactivation response element DNA‐binding protein (TARDBP), and several others. , Currently, there are no disease‐modifying treatments for PSP. , Symptoms of PSP are managed with medications for the treatment of other neurodegenerative diseases such as PD and AD, but effectiveness is limited. , , , To develop effective treatment of PSP, a deeper understanding of its pathogenetic mechanisms is essential. As such, it is crucial to identify proteins and relevant biological pathways involved in PSP pathogenesis. Since the advent of the mass spectrometry‐based proteomics approach, this method has been considered the gold standard for protein identification and measurement. Therefore, mass spectrometry‐based proteomic analysis of the human brain from PSP patients is essential to understand the pathogenesis of this disease. Nevertheless, no in‐depth global proteome data acquired from the brains of PSP patients is available to date. In this study, we conducted mass spectrometry‐based proteome analysis of GP from 15 PSP patients, 15 PD patients, and 15 healthy control (HC) individuals. To analyse and compare these 45 samples, we employed the 11‐plex isobaric tandem mass tag (TMT)‐based quantitative proteomics technology in which samples can be multiplexed up to 11 samples. To our knowledge, this study is the first in‐depth global proteome analysis of the GP from PSP patients. The proteins and relevant signalling pathways discovered in this study provide a foundation for unravelling the pathogenetic mechanisms of PSP.
We utilised GP samples from 15 PSP patients, 15 PD patients, and 15 HC individuals. GP was selected as a well‐defined basal ganglia region known to be affected by PSP pathology. The GP samples were collected from the Brain Resource Center at the Johns Hopkins University School of Medicine. The clinical information for the used samples is provided in Table 1 and Table S1. The inclusion criteria for PSP are patients with neuropathologic changes fulfilling PSP diagnostic criteria, age > 50 years, males and females, and any race. The exclusion criteria for PSP include patients with any significant neurodegenerative or vascular comorbidity.
All GP samples were prepared by sonicating (Branson sonifier 250, ultrasonics, Danbury, USA) in 8 M urea and 50 mM triethylammonium bicarbonate (TEAB) on ice. The protein amount of each sample was quantified by the bicinchoninic acid (BCA) protein assay (Pierce; Rockford, IL, USA). The 45 GP samples were divided into five batches to be analysed using 11‐plex TMT. Each batch included one master pool (MP) and one quality control (QC) sample. The MP and QC samples were prepared by combining an equal amount of proteins from all GP samples. The sample order for TMT labelling was randomised to minimise the effect of the TMT channel. The MP sample was added to the 11th channel of each TMT experimental batch to normalise the data from multiple TMT experimental batches. The QC samples for verification of technical and biological variations between the batches were divided and placed in a channel in each batch before reduction and alkylation. For the reduction and alkylation of the proteins, 10 mM tris (2‐carboxyethyl) phosphine hydrochloride (TCEP) and 40 mM chloroacetamide (CAA) were added to the samples and then incubated for 1 h at room temperature (RT, 22°C to 25°C). Proteins were digested by LysC (Lysyl endopeptidase mass spectrometry grade; Fujifilm Wako Pure Chemical Industries Co., Ltd., Osaka, Japan) in a ratio of 1:100 for 3 h at 37°C, and then further digested by trypsin (sequencing grade modified trypsin; Promega, Fitchburg, WI, USA) in a ratio of 1:50 at 37°C overnight (for over 18 h) after diluting the concentration of urea from 8 M to 2 M by adding three volumes of 50 mM TEAB. The samples were acidified with 1% trifluoroacetic acid (TFA) to the final concentration and desalted with C18 Stage‐Tips (3M EmporeTM; 3M, St. Paul, MN, USA). The eluted solution containing peptides was vacuum‐dried using a Savant SPD121P SpeedVac concentrator (Thermo Scientific, Waltham, MA, USA) and then stored at −80°C before use. ,
The digested peptides from GP samples were labelled using 11‐plex TMT reagents to perform TMT‐based quantitative mass spectrometry according to the manufacturer's instructions (Thermo Fisher Scientific, MA, USA). The MP sample was prepared in one tube and labelled by the 131C channel and split into five batches. The PSP, PD, HC, and QC samples were labelled with the rest of the channels. All TMT labelling reactions were performed for 1 h at RT and then quenched with 1/10 volume of 1 M Tris‐HCl (pH 8.0). The samples of each batch were pooled and subjected to prefractionation using basic pH reversed‐phase liquid chromatography (bRPLC) on an Agilent 1260 HPLC system (Agilent Technologies, Santa Clara, CA, USA). The TMT‐labelled peptides were reconstituted in solvent A (10 mM TEAB, pH 8.5) and loaded onto a C18 column (Agilent 300 Extend‐C18 column, 5 μm, 4.6 mm × 250 mm, Agilent Technologies). The loaded peptides were separated over the gradient of solvent B (10 mM TEAB in 90% acetonitrile (ACN), pH 8.5) at a flow rate of 0.3 mL/min. A total of 96 fractions collected over 97 min (the total run time of 150 min) were concatenated into 24 fractions. The concatenated samples were dried in a SpeedVac. , ,
The prepared peptide samples were trapped onto an Acclaim™ PepMap™ 100 LC C18 NanoViper trap column (100 μm × 2 cm, packed with 5‐μm C18 particles, Thermo Scientific) at a flow rate of 8 μL/min and resolved on an EASY‐Spray™ analytical column (75 μm × 50 cm, packed with 2‐μm C18 particles, Thermo Scientific) at a flow rate of 0.3 μL/min using an Ultimate3000 RSLCnano nanoflow liquid chromatography system (Thermo Fisher Scientific, MA, USA) that was coupled with an Orbitrap Fusion Lumos Tribrid Mass Spectrometer. The peptide separation was conducted by increasing the gradient of solvent B (0.1% FA in 95% ACN) from 8% to 28% over 90 min. An EASY‐Spray ion source was operated at 2.4 kV. The data acquisition for the peptides injected into the mass spectrometer was conducted in data‐dependent acquisition (DDA) mode. The MS1 scan range was set to m/z 300 to 1,800 with a 3‐sec per cycle of the “top speed” setting. The mass resolutions for MS1 and MS2 were 120,000 and 50,000 at an m/z of 200, respectively. Maximum ion injection times for MS1 and MS2 were 50 and 100 milliseconds, respectively. The automatic gain controls for MS1 and MS2 were 1 and 0.05 million ions, respectively. The higher‐energy collisional dissociation (HCD) value was set to 35%. The precursor isolation window was set to m/z 1.5 with an m/z 0.4 offset. Dynamic exclusion was set to 30 s with 7 ppm of the mass window. Single‐charged ions were rejected. Internal calibration was conducted using the lock mass option (m/z 445.1200025) from ambient air. , ,
The data analysis was conducted as described in Khan et al. with some modifications as follows : The version of Proteome Discoverer was 2.2.0.388. The UniProt database (released in May 2018) used in this study included both Swiss‐Prot and TrEMBL. The minimum peptide length was set to six amino acids. The MS order for the protein quantification was set to MS2. Reporter ion abundance was calculated based on the signal‐to‐noise (S/N) ratio. The average reporter ion S/N threshold and co‐isolation threshold were set to 50% and 30%, respectively.
The GP tissues were heated at 95°C for 5 min and sonicated in RIPA lysis buffer (150 mM NaCl, 1% NP‐40, 25 mM Tris‐HCl pH 7.6, 0.1% sodium dodecyl sulfate, and 1% sodium deoxycholate) supplemented with an EDTA‐free protease inhibitor cocktail (Roche, Basel, Switzerland). Subsequently, the lysed samples were centrifuged at 16,000 × g at 4°C for 5 min. Protein quantification of supernatant from each sample was performed using the BCA protein assay (Pierce; Rockford, IL, USA). The samples were added with 4X Laemmli buffer (BIO‐RAD, Hercules, CA, USA) containing 10% 2‐mercaptoethanol and heated at 70°C for 10 min. The proteins were then separated on Novex WedgeWell 4 to 20% Tris‐Glycine gels (ThermoFisher Scientific, MA, USA). Proteins were blotted onto a 0.2‐μm polyvinylidene difluoride (PVDF) membrane (BIO‐RAD) using wet transfer at 100 V for 1.5 h. Subsequently, the PVDF membranes were blocked in StartingBlock (PBS) Blocking Buffer (Thermo Scientific) at RT for 1 h. Blocking buffer was used to dilute the primary and secondary antibodies. The PVDF membranes were incubated at 4°C overnight with one of the following primary antibodies: anti‐NDUFB11 (1:500, Invitrogen, Waltham, MA, USA), anti‐UQCRH (1:200, Invitrogen), anti‐NDUFA4 (1:1,000, Thermo Scientific), and anti‐β‐actin (Invitrogen). The next day, the PVDF membranes were washed three times in Tris‐buffered saline with Tween 20 TBST (Cell Signaling Technology, Danvers, MA, USA). Each wash was performed at RT for 10 min. Subsequently, the PDVF membranes were incubated with anti‐rabbit (1:1,000, Cell Signaling Technology) IgG secondary antibody conjugated with horseradish peroxide (HRP) at RT for 1 h. Finally, the membranes were washed three times again under the same wash conditions mentioned above, followed by incubation of the membranes with SuperSignal West Pico PLUS substrate (Thermo Fisher Scientific, MA, USA) for chemiluminescent detection. The membranes were imaged using a western blot imaging system (Amersham Imager 600, GE Healthcare, Milwaukee, WI, USA). Densitometric analysis of the images was performed on ImageJ software (NIH) and t‐test statistical analysis was performed for the relative intensity of β‐actin using GraphPad Prism version 9.4.0 for Windows (San Diego, CA, USA).
The total number of GP samples used in this study was 45, composed of 15 PSP patients, 15 PD patients, and 15 HC individuals. We conducted sample size analysis using the pwr package in R. When we wanted to detect proteins with 1.5‐fold differences between groups, the required minimum sample size was 9.4 when the significance level was 0.0001, power was 0.8, sigma was 0.208, and delta was 0.585 ( = log2 1.5). This sigma value of 0.208 was derived from our in‐house TMT proteomics experiments. The significance level of 0.0001 was determined based on our previous studies. When we identified several thousand proteins, the majority of the proteins with a p value < 0.0001 showed a q‐value < 0.05. Based on this sample size analysis, we decided to use 15 samples per group. The statistical analysis of the mass spectrometry data was performed with the Perseus version 1.6.0.7 software package. The quality of mass spectrometry data was monitored by measuring coefficient variations (CV) of QC samples and the S/N ratios. The S/N ratios were calculated by dividing standard deviations (SD) of the samples by SDs of QCs. The protein abundance data from five batches of the TMT experiments was normalised by dividing the abundance data of the PSP/PD/HC samples and QCs by those of the MPs included in each batch, followed by dividing by the median values of each protein. The relative abundance values for each sample were log2‐transformed, followed by a z‐score transformation. , We removed proteins with one or more missing values across 45 samples. To further remove batch effects, an additional normalization was conducted with the ComBat package in R. Proteins with a q‐value of <0.05 were considered significant. The fold changes between the comparison groups were calculated by dividing the average abundance values of each protein from one group by the values of another group. According to our normality test using Shapiro–Wilk test in the dplyr package in R, the majority of the proteins showed normal distribution. Thus, p values between the comparison groups were calculated by the student's two‐sample t‐test. Since we are conducting multiple comparisons, we calculated a false discovery rate by comparing data with and without permutations between groups. The q‐values for the volcano plots were calculated by a permutation‐based FDR estimation in the significance analysis of microarrays (SAM) method, in which P values and fold‐changes were calculated before and after the permutation of samples from two groups. As an orthogonal method to increase the reliability of the selection for differentially expressed proteins between groups, we also used bootstrap receiver operating characteristic (ROC) curve‐based statistical analysis. , , , A bootstrap ROC analysis was carried out using the fbroc package in R. Sampling with replacement was repeated 1,000 times for the bootstrap ROC. The area under the curve (AUC) of a bootstrap ROC was computed for each sampling. The mean and SD values of AUCs from 1,000 bootstrap ROC were then calculated. , The q‐values of the bootstrap ROC‐based analysis data were calculated as follows: (1) The mean AUC values for non‐permuted and permuted data were sorted in descending order for proteins with mean AUCs > 0.5 and in ascending order for proteins with mean AUCs < 0.5; (2) the ratios of the protein numbers for the non‐permuted data to the protein numbers for the permuted data were calculated as lowering the cut‐off threshold, and the ratios were used as q‐values.
The differentially expressed proteins were used for the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis embedded in DAVID version 6.8. , Interactome analysis was carried out by the STRING PPI database version 11. , The weighted gene co‐expression network analysis (WGCNA) was conducted using the R software package. ,
The mass spectrometry data from this study have been deposited to the ProteomeXchange Consortium (https://www.proteomexchange.org) via PRIDE partner repository with the dataset identifier PXD031648 and project name “Mass spectrometry‐based proteomics analysis of human globus pallidus from progressive supranuclear palsy patients discovers multiple disease pathways.”
We conducted a quantitative proteome analysis of 45 GP samples from 15 PSP patients, 15 PD patients, and 15 HC individuals. For more accurate protein quantification, we exploited the 11‐plex TMT labelling method. For the analysis of 45 GP samples using 11‐plex TMT, we added MP to the 11th channel of each TMT experimental batch to normalise data from multiple TMT experimental batches. A QC was placed in one of the remaining 10 channels of each TMT experimental batch to evaluate technical variations and the S/N ratio, as shown in Figure 1. The extracted proteins from human GP samples were digested with Lys‐C and trypsin, followed by TMT labelling and bRPLC fractionation. The fractionated peptides were then analysed on an LC‐MS/MS. In total, 5,223,768 of the MS/MS spectra were acquired, and 1,278,010 spectra were assigned to peptides, leading to the identification of 120,671 peptides and 10,231 proteins (Data S1). The numbers of proteins identified from each batch and all five batches in common were ∼8,500 and ∼6,900, respectively (Figure 2A). To compare protein abundances from five different batches, protein intensity values were normalised by the intensity values of the MP sample in each batch. Because the batch effect estimation by PCA analysis showed a residual batch effect, we conducted an additional normalisation using the ComBat package in R, and we observed that most of the residual batch effect was removed (Figure 2B). Subsequently, we accessed technical variations and the S/N ratio using the QC samples. More than 70% of proteins showed CV of <30%, and ∼90% of proteins showed S/N ratios > 1 (Figure 2C). These results suggest that our mass spectrometry analysis was successfully conducted with high precision.
To identify proteins with differential expression in GP from PSP patients compared to PD or HC individuals, statistical analyses of the proteome data were conducted using two different approaches: SAM‐ and bootstrap ROC‐based analyses. In the SAM‐based approach, the numbers of differentially expressed proteins in PSP versus HC, PSP versus PD, and PD versus HC were 325, 934, and 18, respectively (Figure 3A; Tables S2–S4; and Data S2). HMGA1, ICAM1, EMILIN1, SQSTM1, NGFR, SPP1, IGHA2, SAA1, and so on were among the most upregulated proteins, and MT‐CO1, SLIRP, GABRB2, GJB6, APOA4, FXYD1, and so on were among the most downregulated proteins in PSP compared to HC. ICAM1, PSME2, FAM129A, SQSTM1, ANXA1, FCER1G, S100A6, CD44, GDA, IGHA2, and so on were among the most upregulated proteins, and COX6B1, SLC25A12, ACSL6, ADAM22, GJB6, LGI2, PVALB, EML2, SLC6A11, PDE10A, and so on were among the most downregulated proteins in PSP compared to PD. ACVR1, KLK6, SELENOP, SLC38A2, and so on were among the most upregulated proteins and DDC, TH, TPH2, SLC6A3, FCER1G, ADIRF, and so on were among the most downregulated proteins in PD compared to HC (Figure 3A). In the bootstrap ROC‐based approach, the numbers of differentially expressed proteins in PSP versus HC, PSP versus PD, and PD versus HC were 463, 1,066, and 55, respectively (Figure 4B; Tables S5–S7; and Data S2). HMGA1, SQSTM1, EMILIN1, ICAM1, RSU1, LSM8, RCOR3, ZC3H18, LSM4, and so on were among the most upregulated proteins, and MT‐CO1, ATAD1, GABRB2, EPDR1, CMC2, COX5A, UQCR10, MRPL16, and so on were among the most downregulated proteins in PSP compared to HC. FERMT3, FCGR1A, SQSTM1, ICAM1, STK4, S100A11, C1QB, MATN2, and so on were among the most upregulated proteins, and ACSL6, CMC2, COX6B1, AFG1L, CADPS, ADAM22, UQCRC1, MRPL12, LGI2, and so on were among the most downregulated proteins in PSP compared to PD. ADSL, FLAD1, PSMB2, SESN1, SLC38A2, NAAA, and so on were among the most upregulated proteins, and DDC, TH, TPH2, PYCR3, WNK2, PLXNC1, ZSCAN18, SLC6A3, RPS10, and so on were among the most downregulated proteins in PD compared to HC (Figure 3B). When the differentially expressed proteins identified from the SAM‐based analysis were compared with those identified with bootstrap ROC analysis, 225, 809, and 15 proteins overlapped in PSP versus HC, PSP versus PD, and PD versus HC, respectively (Figure 3C). To minimise the number of differentially expressed proteins selected by type I error, we decided to use the differential proteins common to both of our analytic approaches for further pathway analysis.
To uncover dysregulated signalling pathways in the GP of PSP, gene set enrichment analysis was conducted using the KEGG pathway database embedded in DAVID bioinformatics resources (Data S3). When PSP was compared to HC, the PD pathway was the most enriched one, followed by oxidative phosphorylation, Alzheimer's disease, Huntington's disease, and non‐alcoholic fatty liver disease (NAFLD) pathways (Table 2). For the PD pathway, 25 proteins were enriched, and strikingly, the majority of the proteins enriched in the pathway were proteins related to mitochondrial functions such as electron transport (COX4I1, COX5A, COX6B1, COX7A2, CYCS, and NDUFA4), NADH dehydrogenases (NDUFA9, NDUFB11, NDUFB3, NDUFB4, NDUFB7, NDUFB8, NDUFC1, NDUFS2, NDUFS5, NDUFS7, and NDUFS8), succinate dehydrogenase (SDHC), ADP/ATP translocase (SLC25A4 and SLC25A5), and cytochrome b‐c1 complex proteins (UQCR10, UQCR11, UQCRB, UQCRH, and UQCRQ) (Figure 4A). The majority of proteins enriched in the four other pathways were also related to mitochondrial functions (Table S4). All of the mitochondria‐related proteins were downregulated in PSP compared to HC. When PSP was compared to PD, the top five most enriched pathways were the same as the ones enriched in the comparison between PSP and HC, but the number of enriched proteins in each pathway was more than doubled (Table 3). For the PD pathway, 60 proteins were enriched and the majority of the proteins were related to mitochondrial functions, as was observed in the comparison between PSP and HC (Figure 4B). These mitochondrial proteins included ATP synthase (ATP5F1, ATP5H, ATP5J, and ATP5O) and acyl carrier protein (NDUFAB1) as well as the ones already observed in the comparison between PSP and HC, such as cytochrome c‐related proteins (COX4I1, COX5A, COX5B, COX6A1, COX6B1, COX6C, COX7A2, COX7A2L, COX7C, CYC1, CYCS, and NDUFA4), NADH dehydrogenases (NDUFA2, NDUFA3, NDUFA4, NDUFA7, NDUFA9, NDUFA10, NDUFA12, NDUFA13, NDUFAB1, NDUFB3, NDUFB4, NDUFB5, NDUFB6, NDUFB7, NDUFB8, NDUFB9, NDUFB10, NDUFB11, NDUFC1, NDUFC2, NDUFS1, NDUFS2, NDUFS3, NDUFS5, NDUFS7, NDUFS8, NDUFV1, and NDUFV3), succinate dehydrogenase (SDHA, SDHB, and SDHC), ADP/ATP translocase (SLC25A4, SLC25A5, and SLC25A6), and cytochrome b‐c1 complex proteins (UQCR11, UQCRB, UQCRC1, UQCRC2, UQCRFS1, UQCRH, and UQCRQ). The majority of proteins enriched in the four other pathways were also related to mitochondrial function, as observed in the comparison between PSP and HC (Table S4). All of the mitochondria‐related proteins were downregulated in PSP compared to HC. The gene set enrichment analysis results suggest that the downregulation of mitochondrial proteins is potentially linked to PSP pathogenesis.
The gene set enrichment analysis showed that most of the proteins enriched in the top five pathways were related to mitochondrial function. We reasoned that the analysis of differential proteins using an orthogonal approach would provide higher confidence in the relevance of mitochondrial proteins in PSP. For this, we conducted a PPI analysis of the differentially expressed proteins in the comparison groups using STRING PPI analysis. We used ‘Experiment’ alone as an active interaction source and a minimum required interaction score threshold of 0.9 (highest confidence). For the differentially expressed proteins in the comparison between PSP and HC, the STRING PPI analysis produced two highly connected clusters and one moderately connected cluster (Figure 5A). The most connected cluster was formed by mitochondrial ribosomal proteins (MRPs). The second and third most connected clusters were formed by the NADH dehydrogenases and cytochrome b‐c1 complex proteins. Reactome analysis embedded in STRING PPI also showed that all of the top four enriched pathways were related to mitochondrial translation (Table 4). For the differentially expressed proteins in the comparison between PSP and PD, STRING PPI analysis produced one highly connected and three moderately connected clusters (Figure 5B). All connected clusters were formed by proteins related to mitochondrial function, such as NADH dehydrogenase, ATP synthase, cytochrome b‐c1 complex with cytochrome c oxidase, and succinate dehydrogenase. Reactome analysis also showed that all of the top three enriched pathways were related to mitochondrial respiratory electron transport (Table 5). The interactome analysis results also suggest that mitochondrial proteins represent the main component of the differentially expressed proteins in PSP compared to HC and PD.
Both the gene set enrichment and the PPI analyses for the differentially expressed proteins in the GP of PSP patients suggested that mainly mitochondrial proteins were dysregulated. However, we still could not exclude the possibility that variables other than PSP pathology could contribute to the differential expression of mitochondrial proteins. For this, we conducted WGCNA, in which proteins with similar co‐expression patterns are identified, generating multiple modules that are composed of proteins with similar expression patterns. Subsequently, the correlations of the modules with various traits of the samples such as diagnosis, age, sex, and post‐mortem delay (PMD) are estimated (Table 1; Figure 6; and Figure S1). Since the PD, mitochondrial translation, and ribosome pathways were the ones most enriched in the gene set enrichment and the PPI analyses, we investigated whether the modules enriched with proteins belonging to these three pathways were correlated with variables other than disease diagnosis. First, we selected modules that showed correlations with the disease diagnosis trait and searched them for the modules that have proteins enriched with the three pathways among them. When PSP and HC data were analysed using WGCNA, the M5, M7, M8, and M15 modules showed positive correlations with p values of <0.05, demonstrating that the proteins in these clusters have a tendency to have increased expression levels in PSP compared to HC. On the other hand, the M9, M10, M11, and M12 modules showed negative correlations with p values of <0.05, indicating that the proteins in these clusters have a tendency to have decreased expression levels in PSP compared to HC (Figure 6A). In the M12 module, the top five pathways selected by the gene set enrichment analysis were the same as those observed for differential proteins between PSP and HC, with the PD pathway being the most enriched (Table S8; Table 2; and Data S4). In the M11 module, the ribosome pathway was the most enriched (Table S7 and Data S4). When PSP and PD data were analysed using WGCNA, the M10 and M19 modules showed positive correlations with p values of <0.05, demonstrating that the proteins in these clusters have a tendency to have increased expression levels in PSP compared to PD. On the other hand, the M1, M2, M3, and M6 modules showed negative correlations with p values of <0.05, indicating that the proteins in these clusters have a tendency to have decreased expression levels in PSP compared to PD (Figure 6B). In the M6 module, the top five pathways selected by the gene set enrichment analysis were the same as those observed for differential proteins between PSP and PD, with the PD pathway as the most enriched (Table S9; Table 3; and Data S4). Subsequently, we conducted the PPI analysis with proteins in the M12 and M11 modules generated by WGCNA of PSP and HC and the M6 module generated by WGCNA of PSP and PD. The M12 module from PSP and HC and the M6 module from PSP and PD showed highly connected clusters for mitochondrial respiratory electron transport chain proteins such as NADH dehydrogenase, ATP synthase, cytochrome c oxidase, cytochrome b‐c1 complex, and succinate dehydrogenase (Figures 7A,B), and they were enriched in the PD pathway. The M11 module from PSP and HC showed highly connected clusters for MRPs and they were enriched in mitochondrial translation and ribosome pathways (Figure 7C). These WGCNA results showed that the M12 and M11 modules from PSP and HC and the M6 module from PSP and PD were enriched for mitochondria‐related proteins. However, none of the modules exhibited high correlations with variables other than the diagnostic group. This suggests that the mitochondria‐related proteins in this study are highly linked to PSP pathogenesis, although the causality of the mitochondria‐related proteins for PSP pathogenesis remains to be further investigated.
Since the bioinformatic analysis of the proteomics data acquired in this study revealed that mitochondrial electron transport chain proteins were key molecules differentially regulated in the GP of the PSP, we conducted Western blot experiments for the validation of proteins in the mitochondrial electron transport chain in the GP from an independent cohort composed of 4 PSP, 4 PD, and 4 HC individuals (Table S1). We selected five proteins (SDHC, SLC25A5, NDUFB11, UQCRH, and NDUFA4) in the electron transport chain for the validation experiments, but only three proteins (NDUFB11, UQCRH, and NDUFA4) were detectable by Western blot. Only NDUFA4 showed a decrease in PSP with statistical significance (Figure S2). These results are likely due to sample preparation differences between proteomics and Western blot experiments. Although quantifications by MS and Western blot did not confirm all three proteins, it is well‐known that mass spectrometry‐based quantification data often shows a poor correlation with Western blot‐based quantification. ,
In this study, we conducted mass spectrometry‐based proteome analysis of human GP brain tissue samples from 15 PSP patients, 15 PD patients, and 15 HC individuals using a TMT‐based multiplexing method, in which we identified ∼10,000 proteins and ∼120,000 peptides. To our best knowledge, this is the first in‐depth proteome analysis of the human GP region from PSP patients. In this study, we used two different methods for the selection of differentially expressed proteins. We identified 325, 934 and 18 differentially expressed proteins in the comparisons between PSP and HC, between PSP and PD, and between PD and HC, respectively, using SAM‐based statistical analysis. On the other hand, we identified 463, 1,066, and 55 differentially expressed proteins in the comparisons between PSP and HC, between PSP and PD, and between PD and HC, respectively, by the bootstrap ROC‐based statistical analysis. Although both analysis methods rendered differentially expressed proteins with q‐values < 0.05, the number of overlapping proteins between the two analytical methods were ∼48%, ∼68%, and ∼25% for PSP versus HC, PSP versus PD, and PD versus HC, respectively. The proteins with low q‐values were selected as differentially expressed proteins by both methods, while proteins with higher q‐values were selected by only one method (Data S3). These results suggest that the application of multiple statistical analysis methods can increase the confidence of selection for differentially expressed proteins. In the gene set enrichment analysis, when PSP was compared to HC and PD, almost all proteins in the top five pathways were downregulated in PSP except for five proteins (SLC6A3, TH, UBE2L6, MAPT, and SOD2). Strikingly, most proteins enriched in the five pathways were mitochondrial proteins, and all of the mitochondrial proteins were downregulated except for SOD2. The upregulation of MAPT is expected since intracerebral accumulation of MAPT is a well‐known histopathologic feature of PSP. , SLC6A and TH, which were upregulated in PSP only when it was compared to PD, are proteins expressed in dopaminergic neurons. , We believe that these three proteins were identified from the dopaminergic neuronal axons projecting from substantia nigra to GP. The PPI analysis also demonstrated the clustering of mitochondrial proteins. When PSP was compared to HC, mitochondrial respiratory electron transport chain proteins and MRPs formed two main clusters. When PSP was compared to PD, only mitochondrial respiratory electron transport chain proteins formed the main cluster. Although the pathogenesis of PSP remains unclear, one of the known causes of parkinsonian disorders is neuronal cell death induced by inhibition of complex I in the mitochondrial respiratory electron chain. , , , We found many differentially expressed mitochondrial proteins included in complex I, such as NADH dehydrogenases (NDUFA4, NDUFA9, NDUFAF5, NDUFAF7, NDUFB11, NDUFB3, NDUFB4, NDUFB7, NDUFB8, NDUFC1, NDUFS2, NDUFS5, NDUFS7, NDUFS8, and so on) in this study. More strikingly, all the differentially expressed NADH dehydrogenase proteins were downregulated in PSP compared to both HC and PD. These proteins were included in all the top five pathways selected by the gene set enrichment analysis and formed the clusters in the PPI analysis. Other mitochondrial proteins such as complex II (succinate dehydrogenase), III (cytochrome b‐c1 complex), IV (cytochrome c oxidase), and V (ATP synthase) in the mitochondrial respiratory electron chain were also included in the differentially expressed proteins in the two comparisons between PSP versus HC and PD. When PSP was compared to HC, complex II, III, and IV proteins were enriched in all the top five pathways selected by the gene set enrichment analysis and complex III formed clusters by PPI analysis. When PSP was compared to PD, all of complex II, III, IV, and V were enriched in all of the top five pathways selected by the gene set enrichment analysis, and all of them also formed clusters in PPI analysis. Strikingly, all mitochondrial proteins related to complex I, II, III, IV, and V were downregulated in PSP compared to both HC and PD. This suggests that the mitochondrial dysfunction induced by the dysregulated mitochondrial respiratory electron transport chain complex could be a key component of the PSP pathogenesis accompanying MAPT aggregation. These results are in accordance with the previous reports that mitochondrial dysfunction is part of the etiopathogenesis of PSP. To analyse 45 samples, we conducted five batches of TMT experiments with the QC and the MP samples in each batch. Although the five batches of 11‐plex TMT‐based data were normalised by the MP samples, an obvious batch effect was observed, and further normalisation by the ComBat package removed it. This result suggests that simple normalisation by a common reference sample is often not enough to remove the batch effects when multiple batches of TMT experiments are conducted. Although we report that mitochondrial respiratory electron transport chain complex proteins were dysregulated in GP from PSP patients in this study, this outcome was derived from a mixture of multiple different cell types in GP. During the pathogenic process of neurodegenerative diseases, glial crosstalk is critical in the loss of cellular homeostasis and each cell type would show different responses to the inter‐ and intra‐cellular environment changes. , , Therefore, we need to deconvolute the proteome changes through cell‐type‐specific proteome analysis to understand the changes occurring in each cell type during the pathogenic process. To the best of our knowledge, this is the first study focusing on proteomic analysis of GP from PSP patients. Our discovery of the link between dysregulated mitochondrial respiratory electron transport chain complex proteins and PSP provides a foundation for further investigation of PSP pathogenesis.
We have no conflict of interest to declare.
Click here for additional data file. |
|
PMC9647850 | 36381110 | X Wang,HM Maeng,J Lee,C Xie | Therapeutic Implementation of Oncolytic Viruses for Cancer Immunotherapy: Review of Challenges and Current Clinical Trials | 20-10-2022 | Oncolytic virus,Cancer,vaccinia virus,Adenovirus,Herpes simplex virus,Vaccinia virus,Newcastle disease virus,Poliovirus,Tumor stroma | The development of cancer therapeutics has evolved from general targets with radiation and chemotherapy and shifted toward treatments with a more specific mechanism of action such as small molecule kinase inhibitors, monoclonal antibodies against tumor antigens, or checkpoint inhibitors. Recently, oncolytic viruses (OVs) have come to the forefront as a viable option for cancer immunotherapy, especially for “cold” tumors, which are known to inhabit an immunologically suppressive tumor microenvironment. Desired characteristics of viruses are selected through genetic attenuation of uncontrolled virulence, and some genes are replaced with ones that enhance conditional viral replication within tumor cells. Treatment with OVs must overcome various hurdles such as premature viral suppression by the host’s immune system and the dense stromal barrier. Currently, clinical studies investigate the efficacy of OVs in conjunction with various anti-cancer therapeutics, including radiotherapy, chemotherapy, immune checkpoint inhibitors, and monoclonal antibodies. Thus, future research should explore how cancer therapeutics work synergistically with certain OVs in order to create more effective combination therapies and improve patient outcomes. | Therapeutic Implementation of Oncolytic Viruses for Cancer Immunotherapy: Review of Challenges and Current Clinical Trials
The development of cancer therapeutics has evolved from general targets with radiation and chemotherapy and shifted toward treatments with a more specific mechanism of action such as small molecule kinase inhibitors, monoclonal antibodies against tumor antigens, or checkpoint inhibitors. Recently, oncolytic viruses (OVs) have come to the forefront as a viable option for cancer immunotherapy, especially for “cold” tumors, which are known to inhabit an immunologically suppressive tumor microenvironment. Desired characteristics of viruses are selected through genetic attenuation of uncontrolled virulence, and some genes are replaced with ones that enhance conditional viral replication within tumor cells. Treatment with OVs must overcome various hurdles such as premature viral suppression by the host’s immune system and the dense stromal barrier. Currently, clinical studies investigate the efficacy of OVs in conjunction with various anti-cancer therapeutics, including radiotherapy, chemotherapy, immune checkpoint inhibitors, and monoclonal antibodies. Thus, future research should explore how cancer therapeutics work synergistically with certain OVs in order to create more effective combination therapies and improve patient outcomes.
Cancer continues to be one of leading causes of death and remains a major threat to human health. The World Health Organization predicts the rate of cancer incidence and mortality will continue to rise in the next 20 years [1]. The effects of traditional therapeutic modalities such as surgical resection, chemotherapy, radiation therapy, and recently developed immunotherapy are not optimal despite recent improvements. Thus, there is a critical need for novel anti-tumoral strategies. Vaccines have become an important milestone in the development of the field of immunology and success in healthcare. The basic premise lies with inoculation of an attenuated form or noninfectious portion of the infectious organism into the body to elicit an immune response. Vaccines grant protective immunity to the body by program the immune system to recognize and target these foreign invaders. When exposed to the nonimpaired version of a virus, the immune system is then able to respond quickly and efficiently to subdue the virus, thus preventing major infection [2]. As a result, vaccine development has been critical in drastically reducing the number of deaths due to smallpox, yellow fever [3], measles, mumps, rubella, and varicella [4]. Certain viruses have been recognized for the ability to target tumor cells. These oncolytic viruses (OVs) tend to use live and infectious viruses and have become a topic of interest in the arena of cancer therapeutics because of their ability to induce selective cell death and specific anti-tumor immunity. In this review, we summarize and integrate what has been published in the literature in terms of the wide diversity of OVs, discuss the challenges in oncolytic viral therapy, and suggest how modification and implementation of OVs in conjunction with traditional cancer therapies may enhance the overall success of adjuvant treatments.
OVs are a type of cancer therapy where viruses are selected for their oncolytic capacity. Often, these viruses are attenuated through alterations in the viral genome that allow for reduced cytotoxicity toward non-cancerous cells and conditional replication in cancer cells. Alternatively, OVs may also be selected for through mutliple passages in tumor tissues. Key viral genes needed for virulence are substituted with genes that encode proteins to specifically target tumor cells. Thus, this strategy prevents viral targeting of nonmalignant tissues and restricts viral replication to only within tumor cells [5,6]. Moreover, use of engineered viruses in virotherapy often comes with a question of potential insertional mutagenesis where the viral genome integrates itself into the host’s genome [7]. However, OVs undergo multiple preclinical studies which assess for efficacy and safety before application in humans as demonstrated by Duerner et al.’s (2008) study of conditionally replication-competent murine leukaemia virus [8]. Despite the low possibility of genomic integration and detrimental impact in the clinical outcome, concrete long term safety data regarding administration of these engineered viruses in humans is needed as more therapeutics are implementing these viruses in both oncology and non-oncology clinics [7]. This selective elimination of cancer cells often depends on the viral strain, cancer type, tumor microenvironment (TME), and host immune system. OVs are intended to preferentially target cancer cells by exploiting unique extracellular surface markers on cancer cell, thus gaining entry into the cell. Commonly overexpressed surface markers in tumor cells are CD46, CD155, and integrin α2β1, which serve as receptors for measles virus, poliovirus, and echovirus respectively [9,10]. Cancer cells often have specific mutations (i.e., aberrations with in BCL-2, EGFR, PTEN, RAS, RB1, TP53, and WNT) that allow for unregulated cell proliferation. However, these mutations may also predispose the tumor cells to viral infection and subsequent cytotoxic elimination [11–13]. Nevertheless, normal cells often induce interferon (IFN) expression in response to viral infection. However, due to the inability of cancer cells to to induce type 1 IFN signaling [11], OVs are able to freely replicate within cancer cells, subsequently inducing oncolysis and release of viral progeny to continue the infection cycle. In addition, OVs may be armed to express immunostimulatory cytokines/chemokines (e.g., tumor necrotic factor (TNF), Interferon (IFN) α, and granulocyte-macrophage colony-stimulating factor (GM-CSF)), which allow the viruses to elicit a strong host immune response [3,4]. OV treatment begins with inoculation of the virus followed by viral replication which generates excessive virus-induced damage, compromising the integrity of cancer cells and results in oncolysis [11]. OV replication has also been found to promote strong anti-tumor immunity through the induction of immunogenic cell death (ICD), which releases tumor antigens (TA), damage-associated molecular patterns (DAMPs), OV-derived pathogen-associated molecular patterns (PAMPs), and inflammatory cytokines to activate and recruit both innate and adaptive immune cells [14,15]. PAMPs function to alert the immune system of the presence of pathogens [16], while DAMPs function to bring awareness to tissue trauma by binding to corresponding receptors on dendritic cells to induce T-cell activation and strongly influences the immune balance in the TME [17,18]. Furthermore, some OVs trigger the anti-tumor response without viral replication-mediated oncolysis. Binding of OVs to the tumor cell triggers activation of an antiviral immune response, where PAMPS trigger secretion of cytokine, and DAMPS to recruit immune cells to the area. Thus, this alternative pathway also promotes an anti-tumor immune response (Figure 1) [19,20]. Therefore, OVs are a feasible option to target notoriously non-immunogenic “cold” tumors, which are known to inhabit a TME that suppresses immune responses and T cell invasion by effectively stimulating both the innate and adaptive immune system. Many of these cold tumors are also non-reponsive to current available immune checkpoint inhibitors and demonstrates in crutial need improvement in the efficacy of cancer immunotherapy [21,22].
Various OVs with anti-tumor properties have been explored, including both DNA and RNA viruses (Table 1). It is important to note that not all DNA/RNA viruses are oncolytic viruses. The important factor that separates oncolytic viruses from other genetically altered viruses for treatment purposes is that OVs are able to replicate and induce cell lysis, hence their name. Genetically altered viruses such as certain adenovirus agents serve as viral vectors that simply deliver the gene(s) of interest, often tumor antigens, and are replication-defective (RD), which is a characteristic that aids in the safety of this treatment modality [23]. This is commonly implemented in vaccine development and has proven to provide effective protection with no serious adverse events such as clinical infection or shedding of the virus into the surrounding environment. This has been demonstrated in RD-recombinant chimpanzee adenovirus type 3-vectored ebolavirus vaccine (cAd3-EBO) and many other studies using adenoviral vectors as a vaccine platform [24]. From a biological perspective, DNA viruses demonstrate higher genome stability due to their high-fidelity DNA polymerases. Their larger genomes often allow for greater ability to incorporate larger transgene insertions without jeopardizing the capacity for viral infection and replication. Replication takes place in the nucleus. However, the large genome size impedes replication kinetics [25–27]. While DNA viruses can encode proteins that protect viral nucleic acid detection [27], these viruses are also able to a elicit strong antiviral responses which can aid in anti-tumor immunity. The caveat remains that high neutralizing antibodies (nAbs) may limit viral replication, thus hindering viral spread [28]; however there have been reports tjat oncolytic viruses are able to replicate efficiently even in the presense of nAbs that target the backbone virus [29]. On the other hand, RNA viruses such as Newcastle disease virus (NDV), poliovirus, and reovirus have limited genomic packaging capacity but can be more immunogenic. Some viruses may encode proteases that cleave RNA virus sensors, which inhibits the antiviral response [27]. Moreover, replication of RNA viruses takes place in the cytoplasm and demonstrate rapid proliferation. Their high mutation rates introduces new genentic variation due to the low-fidelity RNA polymerase [25,27]. This allows for rapid evolution toward a beneficital oncolytic phenotype but also may cause divergence from this desired characteristic. The issue of genetic stability has also been proposed as a possible advantage for “personalized” targeted therapy, where multiple optimized virus variants can promote tumor clearance even in the presence of antiviral immunity [30]. Thus, the use of RNA viruses can be a double-edged sword, thus calling for a judicious design in the construction of OVs and study designs. Common examples of oncolytic DNA viruse include vaccinia virus (VV), adenoviruses, and herpes simplex virus (HSV). VV is a double-stranded DNA (dsDNA) virus that infects and replicates within the cytoplasm of mammalian cells [31]. There have been various vacccina virus agents being studied. Pexa-Vec is an oncolytic VV with inactivated thymidine kinase (TK) gene that is replaced with a transgene that expresses human GM-CSF and β-galactosidase [32]. Pexa-Vec has been evaluated in the treatment of hepatocellular carcinoma (HCC) and colorectal cancer [31,33,34]. Moreover, GL-ONC1 (VV with Ruc-GFP, β-glucuronidase, and β-galactosidase transgene insertions), vvDD (VV with deletion of the vaccinia growth factor and TK genes), and TBio-6517 (VV that expresses Flt3 ligand, the cytokine IL-12, and an antibody targeting CTLA4) are under clinical investigation [35–37]. Adenovirus is a dsDNA virus. Onyx-015 (lontucirev) was the first recombinant adenovirus to be tested in humans and features viral attenuation and conditional replication due to deletion of the E1B locus, which encodes for 55 kD E1B protein [38]. Onyx-015 has been discontinued midway through phase III trials looking at head and neck cancer in China despite the possible success in meeting the primary endpoint based on the interm report of efficacy and safety [39]. Second-generation adenoviruses such as DNX-2401 (tasadenoturev) have demonstrated success in treating glioblastomas [40]. The 24 base pair deletion of the E1A gene prevents DNX-2401 from replicating in cells that maintain normal retinoblastoma (Rb) pathways and selectively targets cancer cells with Rb pathway abnormalities [41]. DNX-2401 has shown success in treating malignant gliomas and has been granted fast track orphan drug designation by the US Food and Drug Administration (FDA) in malignant glioma in 2014 [40]. Currently, combination with immune checkpoint inhibitors are being pursued. Additional examples of oncolytic adenoviruses under clinical investication include enadenotucirev (chinerix Ad11p/Ad3 oncolytic adenovirus with a 25 bp deletion of E4 and 2444 bp deletion in E3ORF), LOAd703 (a serotype 5 adenovirus with serotype 35 fiber and knob and encodes trimerized membrane-bound CD40L and 4–1BBL), and ONCOS-102 (a modified sertotype 5 adenovirus with a serotype 3 knob, insertion of the GM-CSF transgene, and a 24 bp deletion of the Rb binding site of the E1A gene) [42–45]. Herpes simplex virus (HSV), specifically HSV-1 and HSV-2, is a dsDNA virus that naturally infects humans [46,47]. HSV oncolytic therapy has been applied to the treatment of melanomas, gliomas, and colorectal cancer [48,49]. HSV1716, a mutant that lacks the ICP34.5 neurovirulence gene, selectively targets and replicates in human glioblastoma cells [50]. NV1020 is a mutant HSV with deletions of a 15-kb region at the UL/S junction including the UL56 gene and further attenuation by a 700-bp deltion encompassing the TK gene and the UL24 promotor [51, 52]. Reinsertion of viral HSV-1 TK gene enables control of NV1020 infection with TK-converted prodrugs like acyclovir. Weekly hepatic arterial infusion of NV1020 was noted to stabilize the liver metastasis in 50% of patients with heavily treated colorectal cancer at the optimal biological dose of 1×108 plaque-forming unit (PFU) [49]. Other HSV-based OVs that are under clinical trials include G207 (an HSV-1 strain with deletion of the neurovirulent γ134.5 gene and insertion of β-galactosidase to inactiave UL39 gene), ONCR-177 (an HSV-1 agent with a mutant UL37 gene, tissue-specific miRNA attenuation, and insertion of five transgenes for IL-12, FLT3LG, CCL2, and antagonists against PD-1 and CTLA-4), OH2 (genetically modified HSV-2 which expresses GM-CSF), and RP1 (an HSV-1 agent that expresses GM-CSF) [53–57]. NDV is a single-stranded RNA (ssRNA) virus that naturally infects avian hosts (poultry) [58,59]. One of the most studied strains of NDV is MTH-68/H, which has been applied to treatment of epithelial tumors as well as high-grade glioma [48, 60, 61]. Another NDV agent, LaSota, is a lentogenic strain of lower pathogenicity. LaSota has been studied in vitro using HPV E6/E7 expressing TC-1 cells that serves as a cervical cancer model and showed that the tumor cells had suppressed growth by OV induced apoptosis [61]. Poliovirus is a ssRNA virus that naturally targets neurons, which makes this an effective vehicle for glioma-targeted oncolytic therapies. Poliovirus infection is limited to human and old-world primates due to viral binding with the poliovirus receptor Nectin-like molecule 5 (Necl-5) or CD155 in order to enter host cells [62,63]. Recombinant virus PVSRIPO is an attenuated chimera created from non-pathogenic strains of rhinovirus and type 1 poliovirus vaccine and has been studied in malignent glioma and melanoma [64,65]. The poliovirus internal ribosomal entry site (IRES) has been replace with that of rhinovirus [66]. Deletion of the poliovirus IRES attenuates neurovirulence and selects for conditional replication in tumor cells, specifically binding to CD155 which has been found to be highly upregulated in many cancer types [63,67]. Respiratory enteric orphan virus (reovirus) is a nonenveloped, double-stranded RNA (dsRNA) virus that is able to infect a wide range of mammalian hosts [68], including bats, humans, minks, and pigs [69]. Reovirus is mostly nonpathogenic in humans and has demonstrated preferential replication within cancer cells that express a constitutively activated Ras pathway. However, the virus does not affect nonmalignent cells without Ras activation [70]. Pelareorep is a shortened form of reovirus that was given an orphan drug status in 2015 by the FDA and the European Medicine Agency (EMA) for the treatment of malignant gliomas, ovarian cancer, and pancreatic cancer, which are considered as Ras-activated tumors [71]. Since then, reovirus has also been used to treat melanomas, breast cancer, and head and neck squamous cell carcinoma [72–74].
The OV field is continuously gaining traction as a feasible option for immunotherapy, and intensive developmental piplines have led to the approval of four OVs throughout the world. The first registered OV was ECHO-7 (trade name Rigvir), which was approved in Latvia in 2004 [75]. ECHO-7 is a type 7, group IV, enteric cytopathogenic human orphan (ECHO) virus that has been repeatedly passaged in human tumor tissue cultures and selected for enhanced selective replication within tumor cells [75,76]. ECHO-7 was approved for local treatment of skin and subcutaneous melanoma metastases and delivered via intramuscular injections. However, it has been shown to be effective in a variety of cancer types other than melanoma, including colorectal, gastric, and small cell lung cancers [77,78]. Pumpure et al. (2020) documented the treatment of a female patient diagnosed with stage IVA primary malignant melanoma of the cervix. The patient reported no side effects or adverse reactions, and the patient had a survival of 67 months and progression-free survival (PFS) of 57 months at the time of publication [79]. However, the State Agency of Medicines of Latvia suspended marketing authorization of Rigvir in 2019 due to poor quality control [80]. In 2005, the Chinese State Food and Drug Administration approved H101 (trade name Oncorine) for treatment of head and neck cancer [81]. H101 is a type 5 recombinant human adenovirus with deletions of the gene that encodes the 55-kDa E1B protein and the E3 region gene segment. E1B works to bind and inactivate p53, thus deletion of this gene allows for proper p53 tetramer formation and cell cycle checkpoint regulation [82]. The E3 region contains seven expressed open reading frames that function to inhibit host immunity to enhance viral dissemination [83]. H101 has been tested on multiple types of solid tumors including gastric carcinoma, HCC, and lung cancer [84–86]. Zhang et al. (2021) evaluated H101 treatment with or without chemotherapy on 95 patients who were diagnosed with advanced gastric cancer. The study demonstrated that H101 combination therapy yielded a more effective response compared to single agent H101 or chemotherapy with a median overall survival (OS) of 29 months and a median PFS of 14.8 months [86]. In 2015, talimogene laherparepvex (tradename T-VEC) was approved as the first oncolytic virus by the FDA for local treatment of unresectable, cutaneous, subcutaneous, and nodal lesions of advanced melanoma or postoperative recurrent melanoma. T-VEC is a genetically modified herpes simplex 1 virus (HSV-1), where both copies of the gene that encodes infected cell protein 34.5 (ICP34.5), a peptide that enhances the virus’ neurovirulence [87], were deleted and replaced with a gene encoding GM-CSF. GM-CSF gene substitution induces secretion of the cytokine to recruit antigen presenting cells (APC) to the TME, and promote cytotoxic T lymphocytes (CD8+ T cells) responses to tumor-associated antigens (TAA). This modification is thought to improve viral replication in tumor cells that are defective in IFN pathways [88–91]. T-VEC has mainly been implemented in the treatment of melanomas. However, there has been some clinical trials focused on lymphomas as well [92,93]. Ramelyte et al. (2021) looked at intralesional T-VEC treatment of 13 patients with primary cutaneous B cell lymphomas (pCBCL). The patients reported mild side-effects such as flu-like symptoms, including chills, fever, and shivering, but no patients developed suspected HSV-associated systematic infection. T-VEC treatment demonstrated enhanced recruitment of an early innate immune response composed of activated natural killer (NK) cells and monocytes, followed by increased CD8+ T cell populations and reduced regulatory T cell (Treg) populations. Overall, T-VEC treatment was found to be effective in treating pCBCL (complete response (CR) = 46.2%, partial response (PR) = 38.4%, and progressive disease = 15.4%) [88]. A phase Ib trial investigated the T-VEC treatment in combination with Ipilimumab, a CTLA-4 inhibitor, in 19 patients with stage IIIB-IVM1c melalona that was not suitable for surgical resection. Puzanov et al. (2016) noted that the combination treatment was safe. A few patients developed grade 3/4 adverse events, but these findings did not lead to the discontinuation of T-VEC or ipilimumab. The treatment demonstrated promising results (CR = 22%, PR = 28%, stable disease (SD) = 22%). Probability of survival at 12 months and at 18 months was 72% and 67% respectfully [94]. Harrington et al. (2016) carried out a phase III OPTiM trial on the response rate of intratumoral injection of T-VEC compared to subcutaneous injection of GM-CSF in 249 patients with stage IIIB/C or IVM1a melanoma. OV treatment (Durable response rate (DRR) = 25.2%, overall response rate (ORR) = 40.5%) was determined to be more benefitial compared to GM-CSF treatment (DRR = 25.2%, ORR = 2.3%) Median OS of T-VEC versus GM-CSF treatment is 41.1 and 21.5 months respectively. Both therapeutic arms were well tolerated with patients reporting mild adverse events such as chills, fatigue, and influenza -like illness [95]. Thus, the data shows encouraging results suggesting that more in-depth research to confirm these results is warranted. In 2021, teserpaturev (G47Δ; trade name DELTACT) was conditionally approved for malignant glioma in Japan. Teserpaturev is an HSV-1 with deletion of the both copies of the γ34.5 gene, and deletion of the α47 gene with the US11 promoter. The lacZ gene as inserted to inactivate the ICP6 gene [96]. The γ34.5 gene functions to impede host cell-induced shutdown of protein synthesis in response to viral infection. Thus, deletion of this gene allows for viral replication in cancer cells as malignant cells often lack the ability to inactivate protein synthesis [97]. Deletion of the α47 gene removes viral inhibition of host cell transporters associated with antigen presentation, leading to enhanced anti-tumor immune activation [98]. Lastly, inactivation of the ICP6 gene induces selective viral replication in actively dividing cells since ICP6 encodes the large subunit of ribonucleotide reductase that is needed for viral DNA replication [99]. Uchihashi et al. (2021) investigated teserpaturev treatment of oral squamous cell carcinoma in a murine model. Teserpaturev was found to inhibit growth of primary lesions and prolonged the survival of athymic nude and immunocompetent mice that injected with tongue cancer cells. Injected teserpaturev was found to immediately disseminate into cervical lymph nodes to effectively suppress lymph node metastases [100].
Implementation of OVs requires careful evaluation as there are multiple factors that must be taken into account. Different methods of inoculation have benefits and drawbacks with viral therapy. Moreover, the tumor extracellular matrix (ECM) must be accounted for as an important factor as well as the tumor stroma. Cell populations including cancer-associated fibroblasts (CAFs) and tumor-associated macrophages (TAMs) can dramatically hinder oncolytic virotherapy efficacy.
OVs can be administered either through direct inoculation into the tumor bulk or systemic injection, which includes intravenous (IV) or intraarterial (IA) injections [101]. There are benefits and challenges with both methods of administration. Direct intratumoral (IT) inoculation has been the most successful, as shown with FDA approved fast tracking of T-VEC. Direct IT inoculation maximizes the concentration of virus at the site of the lesion and thus induces a strong immunological response. However, this method is limited by tumor accessibility. Deep-seated tumors or those that are located in sensitive locations restrict the applicability and feasibility of IT inoculation as such invasive procedures carry a risk of injuries and complications. Moreover, other limitations include poor intratumoral retention due to viral dissemination into the bloodstream, limited viral dispersion in tumor tissues, and adverse inflammatory responses [102]. In contrast, systemic therapy utilizes the body’s vascular system to circulate OVs throughout the body similar to the delivery of chemotherapy or other anti-cancer agents. Likewise, there are a few hypothesized disadvantages associated with indirect inoculation. The first area of concern lies with systemic toxicity, whether the dosage of OVs may result in unanticipated off-tumor tissue or organ damage. Another major concern is immune clearance or the neutralization of OVs by the B cell generated antibodies, which interferes with internalization of the virus and dramatically abates the viral titer that ultimately reaches the tumor site [101,103]. This brings up the issue with seropositivity to the backbone virus, which is especially important viruses that are highly prevalent in the community. For example, there are multiple reports confirming high prevalence of seropositivity against human adenovirus (hAdV) infections throughout the world, including the United States, Australia, Japan, and the Philiphines. Ye et al. (2018) looked at the prevalence of nAbs to HAdV type 4 and type 7 in a group of volunteers from Hunan Province, China. The seropositivity rates for HAdV4 and HAdV7 nAbs were 58.4 and 63.8% respectively [104]. Thus, it can be predicted that a large portion of worldwide populations in areas with a history of HAdV infection contain high seropositivity for HAdV nAbs. The issue remains that seropositivity limits viral replication [105]. Neutralizing antibodies would bind to the OVs and inhibit cellular receptor binding [103,106]. Thus, decision about the choice of viral strain and the mode of administration should be made with careful consideration to preexisting immune responses. Most of the literature agrees that suppression of humoral immunity is essential for systemic administrated oncolytic virotherapy [107]. The IFN pathway, specifically IFN-α, antagonizes OVs by reducing viral replication and stymying virus-mediated apoptosis [108]. Since cancers cells often lack a type 1 IFN response, these cell are more permissive to OV infection and replication [59]. Attempts have been made to protect OVs from the innate and adaptive immune system, specifically the humoral response with the use of IFN response inhibitors to enhance viral replication and efficacy of oncolysis. However, there have been safety concerns regarding the use of IFN antagonists. Saren et al. (2017) noted that treatment of glioblastoma bearing mice with Semliki Forest virus equipped with vaccinia virus-encoded type 1 IFN decoy receptor B18R controlled tumor growth but also induced severe neurotoxicity as the virus disseminated and replicated in healthy brain tissue [109]. Another method to protect OVs involves the use of genetically engineered protective coatings composed of chemical polymers, cell-derived nanovesicles, and liposomes that serve as a more direct method of overcoming the humoral immune response [110–112]. These protective coatings reduce immune recognition of the virus, thus limiting the production of nAbs against the OVs. The addition of tumor-targeting ligands can also help the OVs hone in on the tumor. The major concern with protective coatings is the practicality of the design. Protection of OVs increase the viral titer that reaches the tumor; however, the coatings may undermines the ligand-receptor interactions between OVs and tumor cell receptors resulting in reduced internalization of OVs. Moreover, additional drawbacks include issues with high production costs and limitations with large-scale transport of OVs [107]. Another feasible method is the use of carriers, either patient-derived cellular carriers (i.e., OV-infected cells that are injected back into the patient) or engineer carriers (i.e., nanoparticles). A wide range of cell types can be used as cellular carriers: endothelial cells, mesenchymal stromal cells, T-cells, and even tumor cells. However, there are safety concerns using certain cell types. Even though the patient’s own tumor cells are attractive from an immunologic standpoint, tumor cells or transformed cells should be studied with proper safety measurements. Furthermore, mesenchymal stem cells or neuronal stem cells demonstrate tumor tropism, allowing delivery of OVs throughout the body. However, such cell types are known to evade the immune system by allowing immune escape of tumor cells. The use of biodegradable nanoparticles is also gaining traction for compact delivery of viral antigens and the wide selection of nonmetal and metal-based compositions to maximize delivery of OVs [107]. Liposomal nanoparticles have demonstrated a high degree of biocompatibility with the host’s body and can be rapidly degraded by macrophages, making them a favorable candidate as a OV carrier [113].
Moreover, physical barriers such as the tumor stroma may prevent chemotherapy, tumor infiltrative effector cells, and OVs from effectively approaching tumor cells [114,115]. The tumor stroma is composed of non-tumor cells and structural components of the tumor tissue. Tumor cells are able to secrete cytokines to suppress certain anti-tumor functions of immune cells, while the stromal cells construct the desmoplastic stroma barrier, which physically impedes immune infiltration [116]. The stroma encapsulates the dense ECM, CAFs, TAMs, and tumor vasculature; all of which reinforce tumor resistance against the host’s immune system [117–119]. The ECM is generated by CAFs and poses the greatest barrier as it composes most of a tumor’s mass, creates an impenetrable barrier around the tumor, and undermines immune invasion and anti-tumor drug efficacy [119]. The denseness of the ECM also creates a paucity of oxygen and nutrients, which tumor cells exploit to induce activation of metabolic stress-related signaling pathways. Activation of these signaling pathways allows tumor cells to sculpt the TME to better suit their needs. For example, vascular endothelial cells (VECs) can dedifferentiate into tumor endothelial cells (TECs), which demonstrate enhanced proliferation, augmented migration capabilities, and facilitation of angiogenesis [120,121]. Another effect is the activation of drug efflux pumps and induction of senescence, both of which enhance tumor resistance against anti-cancer agents such as chemotherapy [119]. CAFs recruit myeloid-derived suppressor cells (MDSCs) and Tregs to create an immunosuppressive environment [122]. M2 TAMs have been shown to secrete TGF-β, which stimulates secretion and cross-linking of collagen, bolstering and fortifying the ECM [114,123]. Some studies have investigated methods to target the tumor stoma. For example, OVs expressing proteases such as matrix metallopeptidases (MMP)-9 can degrade ECM components. Sette et al. (2019) demonstrated that treatment of glioblastoma multiforme (GBM) with OV-derived HSV armed with MMP-9 increased viral invasion of GBM stem-like spheroids and improved survival of tumor-bearing nude mice [124]. In addition, OVs can be equipped with tissue inhibitor metalloproteinases 1–4 (TIMPs 1–4), which regulate proteolytic activity of MMPs and prevent rearrangement of the ECM [125]. Another method is to repolarize anti-inflammatory M2 TAMs into the pro-inflammatory M1 phenotype. M2 TAMs promote tumor proliferation through immune modulation and tolerance in addition to the recruitment of Tregs [126,127]. On the other hand, M1 TAMs secrete proinflammatory cytokines (e.g., IL-6, IL-12, and TNF-α) and reactive oxygen species (ROS) in order to enhance immune recruitment and function against malignant cells [128]. Rao et al. (2020) demonstrated the use of genetically engineered cell membrane-coated magnetic nanoparticles that triggers M2-M1 TAM repolarization demonstrated inhibition of tumor proliferation, reduced metastasis, and improved survival of mice with triple-negative breast cancer [129]. Overall, the complexity of the tumor stroma and the various components work in tandem to create an immunosuppressive environment and a physical barrier against not only tumor infiltrating cells but also anti-cancer agents. OVs can influence the TME to convert the pro-tumor TME into an anti-tumor environment, but there still is room for improvement for the strategies described above. Figure 2 shows an overview of This highlights the need for novel approaches in OV development and more research in targeting both the tumor and the surrounding stroma.
Given the compiled efforts in the field of oncolytic virotherapy, there are multiple ongoing clicical trials that investigate OVs in a variety of cancers, including breast, gastrointestinal, skin, and pancreatic cancers. The most common OV candidates include vaccinia virus, HSV, and adenovirus. A subset of ongoing clinical trials solely focus on determining patient response to OV single agent therapy, while the vast majority of trials use a combination approach (Figure 2), often pairing OV treatment with chemotherapy, monoclonal antibodies, or radiotherapy (Table 2, data was collected from clinicaltrials.gov in May 2022). However, there is a critical need for studies on accurate biomarkers to tailor and optimize therapeutic options that combine various treatments for specific patients as disease characteristic may differ across different patients. Table 2 summarized the growing body of research that focuses on immune checkpoint inhibition (ICI) as a means to eradicate tumor cells as a combination partner with OVs [11,48]. Several monoclonal antibodies are developed to target immune checkpoints such as cytotoxic T-lymphocyte antigen 4 (CTLA-4), programmed death protein 1 (PD-1), and programmed death protein ligand 1 (PD-L1). PD-1 is essential in maintaining exhausted T cells and blocking PD-1 after the development of exhausted T cells can boost the T cell immune effector functions, which can disrupt tumor cell immune evasion [130]. ICI has changed the landscape of cancer care since the first FDA approval for anti-CTLA antibody ipilimumab in 2011. However, the majority of the patients do not benefit from ICIs as the overall response rate remains around 20–40% in most of studied regimens so far. Thus, overcoming primary resistance to ICI by offering an opportunity to induce both novel tumor antigen specific immune responses and innate immune responses while shifting the TME toward a pro-inflammatory state is appealing. Ribas et al. (2017) studied the effect of oncolytic virotherapy, T-VEC, and pembrolizumab, an anti-PD-1 antibody, in patients with advanced melanoma. Previous studies have demonstrated that certain patients are resistant to PD-1 blockade due to the paucity of CD8+ T cells within the tumor lesion [131,132]. The use of T-VEC and anti-PD1 blockade combination elicited a strong immune response, increasing systemic circulation of CD4+ and CD8+ T cells, upregulated levels of T cell tumor infiltration, and reduced T cell exhaustion. Common T cell inhibitory markers include increased expression of CTLA4, PD-1, TIGIT, TIM3, and LAG3 [133]. The combination treatment demonstrated a reduction in tumor size with an overall response rate of 62%, and a CR of 33% in a phase 1b study (n=21) with low toxicity [22]. The phase 2 study (n=692) which was carried out in the same setting showed an acceptable safety profile but did not meet the PFS primary endpoint 14.3 (median; range = 10.3–22.1) months where the placebo and pembrolizumab arm showed PFS of 8.5 (median; range = 5.7–13.5, hazard ratio = 0.86; CI = 0.71–1.04, p = 0.13). The OS as a dual primary endpoint strategy is to be reported [134]. Overall, this strategy showed feasibility but requires further investigation into the most efficacious and synergistic combination regimen along with predictive biomarkers to better select the patients who will most benefit from the treatment with enhanced anti-tumor activity while minimizing unnecessary adverse events.
OVs have come to the forefront of immunotherapy, offering a wide range of viruses as a backbone which can be genetically engineered to selectively target and replicate within tumor cells, while leaving normal cells unscathed. Ultimately, cell lysis releases various factors that attract immune cells toward the tumor while viral progeny infects neighboring tumor cells to continue the oncolytic cycle. The conditional replication of OVs make them an appealing therapeutic option. However, the administration of OVs must overcome various barriers such as viral neutralization by the humoral immune response and the hostile TME, which requires further investigation. Some studies have looked into using protective coatings or cellular carriers to overcome viral neutralization and enhance delivery of OVs to the tumor site. Lastly, a review of clinical trial registries for ongoing clinical trials in oncolytic virotherapy reflects a profound interest in the involved biomedical community especially in combination approaches with conventional cancer treatments such as surgery, chemotherapy, radiation, as well as novel immune modulators. Future studies need to verify the long-term safety and efficacy of incorporating OV therapy. Additionally, further research is needed to develop a strategy that can target cancer heterogeneity while ensuring proper receptor binding for viral entry in the setting of rapidly evolving cancer cells which may need to involve precision medicine to offer a more personalized approach for patients. In summary, oncolytic virotherapy has secured its role to support cancer immunotherapy as the fourth pillar of cancer treatment, and research will continue to expand on the utilities of OV as an important element in multimodality approaches. |
PMC9647857 | Jiandong Liu,Nasha Zhang,Jiajia Zeng,Teng Wang,Yue Shen,Chi Ma,Ming Yang | N6‐methyladenosine‐modified lncRNA ARHGAP5‐AS1 stabilises CSDE1 and coordinates oncogenic RNA regulons in hepatocellular carcinoma | 10-11-2022 | ARHGAP5‐AS1,N 6‐methyladenosine,lncRNA,CSDE1,hepatocellular carcinoma | Abstract Background Hepatocellular carcinoma (HCC) ranks fourth among the malignancies leading to cancer‐related deaths all around the world. It is increasingly evident that long non‐coding RNAs (lncRNAs) are a key mode of hepatocarcinogenesis. As the most prevalent mRNA modification form, N 6‐methyladenosine (m6A) regulates gene expression by impacting multiple aspects of mRNA metabolism. However, there are still no reports on genome‐wide screening and functional annotation of m6A‐methylated lncRNAs in HCC. Methods The m6A modification and biologic functions of ARHGAP5‐AS1 in HCC were investigated through a series of biochemical assays. Clinical implications of ARHGAP5‐AS1 were examined in tissues from HCC patients. Results After systematically analysing the m6A‐seq data of HCC cells, we identified 22 candidate lncRNAs with evidently dysregulated m6A levels. Among these lncRNAs, we found that ARHGAP5‐AS1 is the lncRNA with the highest levels of m6A modification and significantly increased expression in HCC specimens. METTL14 acts as the m6A writer of ARHGAP5‐AS1 and IGF2BP2 stabilises the lncRNA as its m6A reader. ARHGAP5‐AS1 remarkably promotes malignant behaviours of HCC cells ex vivo and in vivo. We identified oncoprotein CSDE1 working as the interacting protein of the lncRNA and TRIM28 as the E3 ligase of CSDE1 in HCC. Interestingly, ARHGAP5‐AS1 could attenuate interactions between CSDE1 and TRIM28, which prevents the degradation of CSDE1 via the ubiquitin‐proteasome pathway. Elevated levels of CSDE1 coordinate oncogenic RNA regulons, promote translation of VIM and RAC1 and activate the ERK pathway, which contributes to HCC prognosis. Conclusions Our study reveals a new paradigm in m6A‐modified lncRNAs controlling CSDE1‐mediated oncogenic RNA regulons and highlights lncRNAs as potential targets for future therapeutics against HCC. | N6‐methyladenosine‐modified lncRNA ARHGAP5‐AS1 stabilises CSDE1 and coordinates oncogenic RNA regulons in hepatocellular carcinoma
Hepatocellular carcinoma (HCC) ranks fourth among the malignancies leading to cancer‐related deaths all around the world. It is increasingly evident that long non‐coding RNAs (lncRNAs) are a key mode of hepatocarcinogenesis. As the most prevalent mRNA modification form, N 6‐methyladenosine (m6A) regulates gene expression by impacting multiple aspects of mRNA metabolism. However, there are still no reports on genome‐wide screening and functional annotation of m6A‐methylated lncRNAs in HCC.
The m6A modification and biologic functions of ARHGAP5‐AS1 in HCC were investigated through a series of biochemical assays. Clinical implications of ARHGAP5‐AS1 were examined in tissues from HCC patients.
After systematically analysing the m6A‐seq data of HCC cells, we identified 22 candidate lncRNAs with evidently dysregulated m6A levels. Among these lncRNAs, we found that ARHGAP5‐AS1 is the lncRNA with the highest levels of m6A modification and significantly increased expression in HCC specimens. METTL14 acts as the m6A writer of ARHGAP5‐AS1 and IGF2BP2 stabilises the lncRNA as its m6A reader. ARHGAP5‐AS1 remarkably promotes malignant behaviours of HCC cells ex vivo and in vivo. We identified oncoprotein CSDE1 working as the interacting protein of the lncRNA and TRIM28 as the E3 ligase of CSDE1 in HCC. Interestingly, ARHGAP5‐AS1 could attenuate interactions between CSDE1 and TRIM28, which prevents the degradation of CSDE1 via the ubiquitin‐proteasome pathway. Elevated levels of CSDE1 coordinate oncogenic RNA regulons, promote translation of VIM and RAC1 and activate the ERK pathway, which contributes to HCC prognosis.
Our study reveals a new paradigm in m6A‐modified lncRNAs controlling CSDE1‐mediated oncogenic RNA regulons and highlights lncRNAs as potential targets for future therapeutics against HCC.
Hepatocellular carcinoma (HCC) is the sixth lethal malignancy and ranks fourth among neoplasms leading to cancer‐related deaths all around the world. , , Considerable global differences in the morbidity and mortality of HCC exist and about 85% of HCC patients are diagnosed in Eastern Asia as well as North Africa. , , Regrettably, the 5‐year survival rate of HCC patients is only 18%. , , Besides multiple well‐established risk factors, such as infection of hepatitis B virus (HBV) and/or hepatitis C virus (HCV), aflatoxin B1 intakes, heavy cigarette smoking and excessive alcohol consumption, , the importance of long non‐coding RNAs (lncRNAs) as a key, regulatory mode of hepatocarcinogenesis is increasingly evident. , , , , , Accumulated evidences demonstrated that N 6‐methyladenosine (m6A) plays important, wide‐ranging roles in various malignancies including HCC via post‐transcriptionally regulating gene expression. , , As a chemical derivative of adenosine in RNA, m6A shows a frequency of 0.15%‐0.6% of all adenosines across the mammal transcriptome. , , Typically, the m6A methylation is deposited onto transcripts of mRNAs, lncRNAs and primary microRNAs (pri‐miRNAs) by the METTL3/METTL14 methyltransferase complex co‐transcriptionally. , , In human cells, METTL14 interacts with METTL3 and acts the key methyltransferase to convert A to m6A in RNAs. Genetic knockout of Mettl14 is developmentally lethal in mice, indicating its crucial role in numerous physiological and pathophysiological processes via regulating m6A modification. , Although several mRNAs have been identified as targets of METTL14‐induced m6A modification, , , it is still largely unclear how m6A‐modified lncRNAs controlled by METTL14 contribute to HCC development. In this study, we firstly recognized 22 candidate lncRNAs with evidently dysregulated m6A levels after systematically analysing the m6A‐seq data of HCC cells with or without silencing of METTL14. Among these lncRNAs, we found that ARHGAP5‐AS1 is the lncRNA with the highest levels of m6A modification and increased expression in HCC specimens. METTL14 acts as the m6A writer of ARHGAP5‐AS1 and IGF2BP2 as its m6A reader to stabilise lncRNA ARHGAP5‐AS1 in HCC. LncRNA ARHGAP5‐AS1 remarkably promotes malignant behaviours of HCC cells ex vivo and in vivo. Interestingly, ARHGAP5‐AS1 attenuates interactions between the oncoprotein CSDE1 and its E3 ligase TRIM28, which prevents CSDE1 degradation via the proteasome. Particularly, elevated levels of CSDE1 promote the translation and expression of VIM and RAC1 genes and, thus, HCC cancerous traits.
In order to determine m6A‐modified lncRNAs in HCC, we thoroughly inspected the HepG2 cell m6A‐seq profiles after knocking‐down expression of METTL14 or not (GSE90642). Among 1,130 genes with dysregulated m6A modification levels (m6A fold change <0.667 or >1.5) after silencing of METTL14, there were 22 lncRNAs (ARHGAP5‐AS1, STX16‐NPEPL1, TUG1, ENTPD1‐AS1, FAM157A, MIR570HG, THAP9‐AS1, COX10‐AS1, CYTOR, ABALON, DHRS4‐AS1, LINC01146, MIR663AHG, C1QTNF1‐AS1, NDUFB2‐AS1, TSPEAR‐AS1, SLCO4A1‐AS1, DARS‐AS1, MZF1‐AS1, TEN1‐CDK3, MIR22HG and USP27X‐AS1) with remarkably differential m6A modification in HepG2 cells.
Human HCC HepG2 and SK‐HEP‐1 cells were obtained from the cell bank of type culture, Chinese Academy of Sciences (Shanghai). HEK293T cell line is a generous gift from Dr. Yunshan Wang working at Jinan Central Hospital (Shandong Province, China). DMEM medium (Gibco, C11995500BT) with 10% foetal bovine serum (FBS; Gibco, 1347575) was used for the cultivation of all cell lines. All cells were examined mycoplasma negative once in a while.
All RIP assays were operated with the Magna RIP RNA‐Binding Protein Immunoprecipitation Kit (Millipore, 17‐700) and the antibodies of IGF2BP1, IGF2BP2, IGF2BP3 or CSDE1 as well as IgG Isotype‐control (Table S1). The MeRIP assay was carried out using the same Kit (Millipore, 17‐700) with the m6A antibody or IgG Isotype‐control (Table S1). The target protein‐RNA complexes were then enriched with Dynabeads® Protein G (Invitrogen, 10003D). Levels of various lncRNAs in the protein‐RNA complexes were detected by quantitative reverse transcription PCR (RT‐qPCR).
Trizol reagent (Invitrogen, 94402) was used for the extraction of total RNAs. RNA samples were reverse‐transcribed into cDNAs with PrimeScriptTM RT Master Mix (TaKaRa, RR036A). The relative expression of eight lncRNAs (ARHGAP5‐AS1, LINC00152, C1QTNF1‐AS1, LINC00969, USP27X‐AS1, NDUFB2‐AS1, TEN1‐CDK3 and ABALON), METTL14, IGF2BP2, S14, U2, GAPDH and CSDE1 were detected at least in triplicate with indicated primers (Table S2). The melting‐curve analyses were done to confirm PCR product specificity.
The human ARHGAP5‐AS1 cDNA (NR_027263.1) with a tag sequence (5’‐GTCGTATCCAGTGCGAATACCTCGGACCCTGCACTGGATACGAC‐3’) at the RNA 3’‐end was synthesised and cloned into pcDNA3.1 by Genewiz (Suzhou, China), which was named as WT. Mutants 1, 2 and 3 are plasmids with the A‐to‐G mutation at the 876, 890 or 928 base of WT. The full‐length ARHGAP5‐AS1 cDNA was also cloned into pCDH‐CMV‐MCS‐EF1α‐Puro. As a result, the resultant plasmid was designated A‐AS1. The full‐length ARHGAP5‐AS1 cDNA with inserted T7 promoter upstream and downstream from the cloning site was also cloned into pcDNA3.1. The resultant plasmid was designated pcDNA‐A‐AS1. Two ARHGAP5‐AS1 shRNAs (shA‐AS1‐1 or shA‐AS1‐2, respectively) or the negative control shRNA (shNC) (Table S3) were synthesised and cloned into pLKO.1 by Genewiz (Table S3). These plasmids were named shA‐AS1‐1, shA‐AS1‐2 or shNC. The cDNA for the HA‐tagged CSDE1 (NM_007158.6) and truncated versions of HA‐tagged CSDE1 were cloned into pcDNA3.1 (Genewiz, China). To guarantee the orientation and integrity of plasmids, Sanger sequencing was performed.
Small interfering RNA (siRNA) duplexes for METTL14, IGF2BP2, CSDE1, TRIM28 or HERC5 and the negative control RNA duplex (NC) were all synthesised by Genepharma (Shanghai, China) and details are in Table S3. Transfection of all small RNAs was using INTERFERin reagent (Polyplus, 409–10), as reported previously. , The jetPRIME reagent (Polyplus, 114‐07) was used for transfection of all plasmids as reported previously.
Western blot was done with indicated antibodies (Table S1) as reported previously. , The ECL Western Blotting Substrate (Pierce, 32106) was used to visualise candidate proteins.
Eighty‐five HCC patients (Shandong cohort, n = 26, and Jiangsu cohort, n = 59) were recruited between April 2009 and December 2016 in this study. The demographics and clinical characteristics of all HCC cases were previously reported. , , All HCC patients were Han Chinese. This study was approved by the Institutional Review Board of Shandong Cancer Hospital and Institute. Before enrolling on this study, every patient agreed and signed the informed consent. All experimental methods comply with the Helsinki Declaration and are carried out according to the approved guidelines.
To prepare recombinant lentiviral particles, HEK293T cells were transiently transfected with the psPAX2 (Addgene, #12260) and pMD2.G (Addgene, #12259) plasmids plus the A‐AS1, shA‐AS1‐1 or shA‐AS1‐2 plasmid. At 48 and 72 h after transfection, cell culture supernatants with recombinant lentiviral particles were collected and filtered. Human HepG2 and SK‐HEP‐1 cells were infected with various viral supernatant supplemented with 5μg/mL polybrene and selected with 2 mg/mL puromycin. LncRNA ARHGAP5‐AS1 expression levels in these infected HCC cells were then detected.
For the stable ARHGAP5‐AS1‐OE or ARHGAP5‐AS1‐KD HepG2 or SK‐HEP‐1 cells, 3 × 104 cells per well were seeded in 12‐well plates, harvested, and counted at indicated time points after seeding. A total of 1 × 104 HepG2 or SK‐HEP‐1 cells were seeded and then transfected with 20 nmol/L CSDE1 siRNAs (siC1‐1 and siC1‐2) or NC RNA. Transiently transfected HCC cells were counted at indicated time points after transfection as reported previously.
A 6‐well plate was used to seed with a total of 1,000 stable ARHGAP5‐AS1‐OE or ARHGAP5‐AS1‐KD HepG2 or SK‐HEP‐1 cells per well. A total of 1,000 HepG2 or SK‐HEP‐1 cells per well were seeded into a 6‐well plate. After seeding, the cells were transfected with 20 nmol/L NC RNA, siC1‐1 or siC1‐2. After 14 days, HCC cell colonies in each well were dyed and counted.
To examine the effect of lncRNA ARHGAP5‐AS1 in vivo, a total of 1 × 107 stable ARHGAP5‐AS1‐KD (shNC, shA‐AS1‐1 or shA‐AS1‐2) SK‐HEP‐1 cells were subcutaneously inoculated into fossa axillaries of female nude BALB/c mice (five‐week‐old, Vital River Laboratory, Beijing, China). Tumour growth was monitored every 5 days when tumour volumes reached or were greater than 30 mm3. In in vivo metastasis assays, 2 × 106 SK‐HEP‐1 cells with stable firefly luciferase expression and ARHGAP5‐AS1‐KD (shNC, shA‐AS1‐1 or shA‐AS1‐2) were injected into female nude mice from tail vein (n = 3 per group). Distant metastases of HCC cells were visualised by the IVIS Spectrum In Vivo Imaging System from PerkinElmer. Processes during all mice assays were approved by the Animal Care Committee of Shandong Cancer Hospital and Institute.
When the cell layer of HCC cells was almost confluent, straight wounds of the same width were scratched with a 10μl pipette tip. The wound closure rate was then quantified at unified time points. The transwell assays were performed as reported previously. , After 36 or 24 h, HepG2 or SK‐HEP‐1 cells migrated to the lower wells were stained and the number of migrated cells was counted.
The nuclear/cytoplasmic Isolation Kit (Biovision, P0028) was applied to separately isolate the cytoplasm fractions and nuclear fractions of HCC cells in accordance with the manufacturer's specification.
The RNA pulldown experiment was performed as reported previously. The pcDNA‐A‐AS1 plasmid was used as the template for ex vivo synthesis of lncRNA ARHGAP5‐AS1. Sense and antisense ARHGAP5‐AS1 RNAs were biotinylated and incubated together with HepG2 cell extracts and Streptavidin magnetic beads (Thermo Fisher, 88816). The pull‐downed proteins were then screened by liquid chromatography‐tandem mass spectrometry (LS‐MS/MS) (Hoogen Biotech Co., Shanghai, China) and verified by Western Blot.
As reported previously, the turnover assays were performed. In brief, HepG2 and SK‐HEP‐1 cells with stable ARHGAP5‐AS1‐KD or ARHGAP5‐AS1‐OE were treated with cycloheximide (CHX) (MedChemExpress, HY‐12320/CS‐4985) to pause de novo protein synthesis. The protein levels of CSDE1 and GAPDH were then examined in HCC cells which were treated with CHX.
As reported previously, the ubiquitination assays were performed in HepG2 and SK‐HEP‐1 cells transfected with pcDNA3.1‐HA‐ubiquitin (HA‐Ub). Proteins in HCC cells treated with MG132 were immunoprecipitated to isolate ubiquitinated CSDE1 and then measured using the anti‐HA antibody.
To identify the potential E3 ubiquitin ligase(s) of CSDE1, IP‐MS was performed using the antibody of CSDE1. Co‐IP was performed between CSDE1 and TRIM28 as reported previously. In brief, HCC cells were lysed and then incubated with antibodies of CSDE1, TRIM28 or IgG (Invitrogen) (Table S1) at 4°C overnight. On the next day, cell lysates were incubated with Dynabeads® Protein G beads (Invitrogen) and then washed with the NP‐40 lysis buffer, and then examined by LS‐MS/MS (Hoogen Biotech Co., Shanghai, China) or Western Blot.
The immunofluorescence assay was performed as previously reported. After permeabilisation and blockage, HepG2 cells were incubated with primary antibodies overnight. Cells were then stained with secondary antibodies (Table S1), washed with PBS and incubated with 4,6‐diamidino‐2‐phenylindole (DAPI). RNA FISH was performed to examine the co‐localization of lncRNA ARHGAP5‐AS1 and CSDE1 protein using the lncRNA FISH Kit (RiboBio) and immunofluorescence staining of CSDE1 antibody. In short, cells were permeabilised with 0.5% Triton X‐100 and hybridised with the FISH probes overnight at 37 °C in dark. LncRNA ARHGAP5‐AS1 signals were detected using Cy3 channels. CSDE1 was stained with its antibody and CoraLite488‐conjugated Goat Anti‐Rabbit IgG(H + L) (Table S1). A Zeiss LSM800 confocal microscope (Zeiss, Germany) was used to visualise images.
Student's t‐test was performed to calculate the difference between the two groups. Spearman's correlation was utilised to calculate the significance of expression association between different genes. Kaplan–Meier plots and the log‐rank test were applied to examine the impacts of lncRNA ARHGAP5‐AS1 expression on HCC patients’ survival. A p value of less than 0.05 was considered statistical significance. SPSS software package (Version 16.0, SPSS Inc.) or GraphPad Prism (Version 5, GraphPad Software, Inc.) was used for all analyses.
To identify lncRNAs modified by m6A in HCC progression, we systematically analysed the m6A‐seq data of HepG2 cells with or without silencing of METTL14 (Figure 1A). There were 22 lncRNAs with significantly differential m6A modification in HepG2 cells with or without silencing of METTL14. Among these lncRNAs, levels of eight lncRNAs (ARHGAP5‐AS1, LINC00152, C1QTNF1‐AS1, LINC00969, USP27X‐AS1, NDUFB2‐AS1, TEN1‐CDK3 and ABALON) are markedly associated with the prognosis of TCGA liver cancer (LIHC) patients (Table S4). We then validated the m6A modification levels of these candidate lncRNAs in HCC cells (Figure 1B). The m6A RIP assays indicated that ARHGAP5‐AS1 is the lncRNA with the highest levels of m6A modification in HCC cells. By using m6A‐seq in HepG2 and the SRAMP algorithm (http://www.cuilab.cn/sramp), we identified three potential m6A sites (876A, 890A and 928A) of ARHGAP5‐AS1 RNA (Figure 1C). Subsequent m6A‐specific RIP coupled RT‐qPCR analyses indicated that m6A levels of ARHGAP5‐AS1 RNA were significantly decreased in cells with ectopic expression of the ARHGAP5‐AS1 mutant 3 compared with cells with ectopic WT ARHGAP5‐AS1 expression (Figure 1D,E), suggesting that ARHGAP5‐AS1 928A is its key m6A site in HCC. We next investigated the m6A‐ARHGAP5‐AS1 RNA levels in our HCC patient cohorts and found that tumours had significantly higher m6A‐ARHGAP5‐AS1 RNA levels than the normal tissues (Figure 1F). After silencing of METTL14 (siM14‐1 or siM14‐2) in cells (Figure 1G), we observed evidently decreased m6A modification levels and expression levels of ARHGAP5‐AS1 (Figure 1H,I). In line with this, there are significant expression correlations between METTL14 and ARHGAP5‐AS1 in HCC tissues (LIHC tissues of TCGA, p = 6.1 × 10−7) and normal liver tissues (TCGA and GTEx, p = .001) (Figure 1J). It has been reported that the m6A readers IGF2BPs (IGF2BP1/2/3) stabilise their target RNAs in an m6A‐dependent way. Considering decreased m6A modification levels of ARHGAP5‐AS1 downregulates expression of the lncRNA, we examined whether IGF2BPs are readers of the m6A‐modified ARHGAP5‐AS1 in HCC cells. The RIP‐qPCR assays indicated that IGF2BP2 is the reader protein with the highest binding affinity with lncRNA ARHGAP5‐AS1 in HCC cells (both p < 0.01) (Figure 1K). Importantly, silencing of IGF2BP2 markedly downregulated endogenous levels of ARHGAP5‐AS1 in HCC cells (all p < 0.05) (Figure 1L,M). Taken together, these data elucidated that METTL14 acts as the m6A writer of ARHGAP5‐AS1 and IGF2BP2 as its m6A reader to stabilise lncRNA ARHGAP5‐AS1 in HCC cells.
To explore the involvement of ARHGAP5‐AS1 in hepatocarcinogenesis, we firstly detected its levels in HCC specimens and paired normal tissues of the Shandong cohort and Jiangsu cohort (Figure 2A). There was an obvious up‐regulation of lncRNA ARHGAP5‐AS1 in HCC tissues compared with that in normal liver specimens in the Shandong cohort or Jiangsu cohort (both p < 0.001) (Figure 2A and Table S5). In multiple independent HCC cohorts of Chinese (GSE84005 and GSE115018), Japanese (GSE17856) and Italian (GSE55092), lncRNA ARHGAP5‐AS1 levels expressed in cancerous tissues were consistently elevated, compared to normal specimens (all p < 0.05) (Figure S1A). , , Importantly, high ARHGAP5‐AS1 levels in HCC specimens were also correlated with shortened time of progression free survival (PFS) (Log‐rank p = 0.001) or overall survival (OS) (Log‐rank p = 0.005) (Figure 2B). Collectively, these findings demonstrated that lncRNA ARHGAP5‐AS1 may be a novel oncogene in HCC. To reveal the biological significance of ARHGAP5‐AS1 in HCC, we developed the stable ARHGAP5‐AS1‐KD HepG2 and SK‐HEP‐1 cells (shA‐AS1‐1 and shA‐AS1‐2) and the stably ARHGAP5‐AS1‐OE HCC cells (A‐AS1) (Figure S1B). As shown in Figure 2C, stable ARHGAP5‐AS1‐KD resulted in an obviously inhibited proliferation of HCC cell lines compared to controls (p < .001). Stable ARHGAP5‐AS1‐OE could significantly enhance HCC cell proliferation (p < 0.001) (Figure 2D). Colony formation results also supported the oncogene role of lncRNA ARHGAP5‐AS1 in HCC (Figure 2E and Figure S1C). We then examined the oncogenic functions of ARHGAP5‐AS1 in vivo. We found that the ARHGAP5‐AS1‐KD HCC xenografts grew markedly slow as compared with the control xenografts (both p < 0.05) (Figure 2F). There were also obviously decreased tumour weights in the ARHGAP5‐AS1‐KD group compared to the control group (Figure 2F), which is in support of the oncogenic role of ARHGAP5‐AS1 in HCC.
We then evaluated the effects of lncRNA ARHGAP5‐AS1 in metastatic behaviours of HCC cells ex vivo and in vivo. Interestingly, the stable silencing of ARHGAP5‐AS1 significantly impaired cell motility of HepG2 or SK‐HEP‐1 cells (both p < 0.001) (Figure 3A and Figure S2A). On the contrary, forced expression of ARHGAP5‐AS1 promoted migration of HepG2 or SK‐HEP‐1 cells (both p < 0.001) (Figure 3B and Figure S2B). The Matrigel invasion assays indicated that ARHGAP5‐AS1‐KD impaired the invasion of HCC cells (Figure 3B and Figure S2B). In contrast, overexpression of ARHGAP5‐AS1 accelerated the invasion of HCC cells (Figure 3B and Figure S2C,D). The in vivo HCC metastasis results revealed that silencing of ARHGAP5‐AS1 can significantly impair the distant metastasis of the lung and other organs of HCC cells after injection of malignant cells (Figure 3C). Collectively, these data demonstrated that lncRNA ARHGAP5‐AS1 could enhance motility and invasion of HCC cells ex vivo and in vivo.
Accumulating evidences indicated that lncRNAs could play their roles through interacting with various proteins during tumorigenesis. , , Therefore, we hypothesised that lncRNA ARHGAP5‐AS1 may act as scaffolds for binding certain protein(s) to promote HCC development. To verify this, we examined the cellular localization of ARHGAP5‐AS1 and found that ARHGAP5‐AS1 nearly equally exists in either the nucleus or the cytoplasm of HCC cells (Figure 4A). After examining the pulled‐down proteins by lncRNA ARHGAP5‐AS1 using mass spectrometry proteomics, we identified multiple cancer‐related proteins including CSDE1, ZC3HAV1, CCT8, CKAP4, PARP1, PEG10 and APEX1 in HepG2 (Table S6). Independent assays in HCC cells were performed and successfully validated CSDE1 among these candidate proteins (Figure 4B). In line with this, there was significant lncRNA ARHGAP5‐AS1 enrichment in the RNA‐CSDE1 complexes in HCC cell lines (both p < 0.01) (Figure 4C). During RIP, lncRNA HOTTIP was used as the negative control (Figure 4C). To explore the specific domains required for the interaction between lncRNA and CSDE1, we then constructed various truncated CSDE1 (Figure 4D) and found that the CSDE1 RNA binding motif 2 (aa450‐525) is required for the interaction between lncRNA ARHGAP5‐AS1 and the protein (Figure 4E). Intriguingly, silencing of ARHGAP5‐AS1 significantly suppressed CSDE1 protein levels in HCC cells (Figure 4F). Instead, the over‐expressed ARHGAP5‐AS1 markedly up‐regulated CSDE1 protein in HepG2 and SK‐HEP‐1 cells (Figure 4F). Treatment of the ARHGAP5‐AS1‐KD HCC cells with the 26S protostome inhibitor MG132 increased the expression of endogenous CSDE1 protein in comparison with the control HCC cells (Figure 4G). Conversely, MG132 abolished ARHGAP5‐AS1‐induced up‐regulation of CSDE1 protein in HCC cells (Figure 4G), elucidating that the lncRNA may regulate the proteasome degradation of CSDE1. To confirm this, we next detected CSDE1 expression in HepG2 and SK‐HEP‐1 cells treated with CHX, a protein synthesis inhibitor. The results of the Western Blot showed that the protein levels of CSDE1 declined much faster in the stable ARHGAP5‐AS1‐KD HCC cells than those in the control cells (Figure 4H). In contrast, treatment of HCC cells overexpressing ARHGAP5‐AS1 with CHX led to an obviously longer half‐life of CSDE1 protein than in control cells (Figure 4I). We then investigated whether lncRNA ARHGAP5‐AS1‐controlled degradation of CSDE1 was mediated by ubiquitination of CSDE1. After endogenous CSDE1 was immunoprecipitated in HA‐Ub‐transfected HepG2 or SK‐HEP‐1 cells, evidently increased ubiquitination levels of CSDE1 protein were observed in the stable ARHGAP5‐AS1‐KD HCC compared to controls (Figure 4J). In line with this, the ubiquitination of CSDE1 was decreased in cells overexpressing ARHGAP5‐AS1 compared to controls (Figure 4J). Taken together, these results elucidated that lncRNA ARHGAP5‐AS1 stabilise CSDE1 protein by inhibiting its proteasome degradation.
To disclose how ARHGAP5‐AS1 retards the proteasome degradation of CSDE1, we systematically evaluated proteins precipitated by CSDE1 in HepG2 cells through mass spectrometry. Among all proteins identified, there were only two E3 ligases (TRIM28 and HERC5) (Table S7). To confirm if TRIM28 or HERC5 is the E3 ligase of CSDE1, we firstly examined CSDE1 levels in HCC cells with silenced expression of TRIM28 or HERC5 (Figure 5A,B). After the knocking‐down of TRIM28 expression, elevated CSDE1 protein levels could be detected in HCC cells in comparison with the control cells (Figure 5A). However, no such expression changes were observed after silencing of HERC5 expression in HCC cells (Figure 5B). Importantly, endogenous TRIM28 can be immunoprecipitated with CSDE1 in HepG2 or SK‐HEP‐1 cells (Figure 5C). Endogenous CSDE1 could also be precipitated with TRIM28 in HepG2 or SK‐HEP‐1 cells (Figure 5D). Immunofluorescence assays revealed that TRIM28 and CSDE1 exhibited evident co‐localization in HCC cells (Figure 5E). Similarly, RNA FISH assays indicated the co‐localization of lncRNA ARHGAP5‐AS1 and CSDE1 protein in cells (Figure S3). These data indicated that TRIM28 might be the potential E3 ligase of CSDE1 in HCC. We next investigated if lncRNA ARHGAP5‐AS1 influences the binding of CSDE1 with TRIM28 in HCC cells. More TRIM28 protein could be precipitated with CSDE1 in the stable ARHGAP5‐AS1‐KD HepG2 or SK‐HEP‐1 cells compared to controls (Figure 5F). Conversely, there was less TRIM28 protein precipitated with CSDE1 in the stably ARHGAP5‐AS1‐OE HCC cells in comparison with the control cells (Figure 5G). Taking together, these findings suggested that lncRNA ARHGAP5‐AS1 promotes CSDE1 stabilization by attenuating its interactions with the E3 ligase TRIM28.
Multiple lines of evidence demonstrated that the RNA‐binding protein CSDE1 acts as an oncogene in cancers and regulates the translation and stability of mRNAs at the post‐transcriptional level. , , Indeed, silencing of CSDE1 profoundly suppressed proliferation and clonogenicity of HCC cells (Figure 6A–C and Figure S4A). Moreover, the transwell assays indicated that siRNAs of CSDE1 could markedly inhibit the invasion capability of HCC cells (Figure 6D and Figure S4B). In line with these data, remarkably elevated CSDE1 expression in HCC tissues was detected in comparison with the normal specimens in both cohorts (p < 0.001). Aberrantly high expression of CSDE1 in the TCGA LIHC cohort was associated with evidently shortened OS of patients (Figure 6F), indicating the oncogenic nature of CSDE1 in HCC. Rescue assays indicated that silencing of CSDE1 with siRNAs significantly inhibited the proliferation of HCC cells with stably overexpressed ARHGAP5‐AS1 (both p < 0.05) (Figure S5). It has been found that the regulation of VIM and RAC1 mRNA translation by CSDE1 contributes to melanoma metastasis. As shown in Figure 6G, we observed evidently decreased protein levels of VIM (Vimentin) and RAC1 in HCC cells after silencing of CSDE1. Likewise, there is a downregulated expression of VIM and RAC1 in the stable ARHGAP5‐AS1‐KD HepG2 or SK‐HEP‐1 cells compared to controls (Figure 6H); whereas, ectopic ARHGAP5‐AS1 expression markedly elevated expression of VIM and RAC1 (Figure 6I). Interestingly, the knocking‐down of ARHGAP5‐AS1 or CSDE1 reduced the phosphorylation of ERK1/2 (Thr202/Tyr204) in cells (Figure 6G,H). Ectopic ARHGAP5‐AS1 obviously enhanced ERK1/2 phosphorylation in cells (Figure 6I). Together, these data demonstrated the key part of lncRNA ARHGAP5‐AS1 in stabilizing CSDE1 protein, promoting the translation of VIM and RAC1 as well as activating the ERK signalling in HCC (Figure 6J).
After the genome‐wide screening of lncRNAs in HCC via m6A‐seq and RNA‐seq, we successfully identified ARHGAP5‐AS1 as a novel m6A‐modified lncRNA. METTL14 is the m6A writer of ARHGAP5‐AS1 and IGF2BP2 acts as its m6A reader to stabilise the lncRNA. Increased ARHGAP5‐AS1 expression was detected in cancerous specimens and associated with evidently shortened survival time of HCC patients. Consistently, lncRNA ARHGAP5‐AS1 exhibited strong oncogenic potentials ex vivo and in vivo. In particular, ARHGAP5‐AS1 could interrupt interactions between CSDE1 and TRIM28, stabilise oncoprotein CSDE1, boost translation of VIM and RAC1 mRNAs, stimulate the ERK signalling and, thus, accelerate HCC progression. It is becoming more and more clear that the expression of certain lncRNAs is precisely regulated by their m6A modification levels during hepatocarcinogenesis. , , , , For example, METTL3‐mediated m6A modification of LINC00958 led to increased gene expression through stabilizing the lncRNA. Oncogenic LINC00958 is a lipogenesis‐related RNA which can sponge miR‐3619‐5p to elevate hepatoma‐derived growth factor (HDGF) expression and accelerated HCC development. Similarly, Jia et al. found that elevated expression of lncRNA LNCAROD was maintained by its increased m6A methylation. Enhancement of glycolysis, which is mediated by pyruvate kinase isoform M2 (PKM2), is vital for tumourigenesis. LNCAROD interacts with SRSF3 to induce switching from PKM to PKM2 and preserves expression levels of PKM2 by sponging miR‐145‐5p in HCC. As a result, LNCAROD could promote proliferation, invasion, and chemoresistance of HCC cells. However, there is still no systematical screening of m6A‐methylated lncRNAs in HCC cells up until now. After genome‐wide analyses of HCC m6A‐seq and RNA‐seq data, we successfully identified ARHGAP5‐AS1 as a novel m6A‐methylated lncRNA. Silencing of METTL14 results in evidently downregulated m6A and expression levels of ARHGAP5‐AS1. IGF2BP2, the m6A reader of ARHGAP5‐AS1, sustains its high levels in HCC. As the antisense non‐coding transcript of ARHGAP5, lncRNA ARHGAP5‐AS1 has been revealed to contribute to the development of breast cancer and gastric cancer. , After analysing the RNA‐seq data of breast cancer MDA‐MB‐231 cells and the highly metastatic derivative MDA‐MB‐231‐LM2 cells (LM2), Wang et al. found that ARHGAP5‐AS1 expression level was markedly reduced in LM2 cells. It has been reported that ARHGAP5‐AS1 could suppress cell migration through downregulating SMAD7 expression in breast cancer cells. On the contrary, ARHGAP5‐AS1 is upregulated in chemoresistant gastric cancer cells and the knocking down of ARHGAP5‐AS1 can effectively reverse chemoresistance. High ARHGAP5‐AS1 expression was significantly associated with the poor prognosis of patients who suffered from gastric cancer. Consistently, we also observed that lncRNA ARHGAP5‐AS1 acts as an oncogenic modulator of HCC progression. Overexpressed CSDE1 protein has been suggested as a vital component during tumourigenesis including melanoma and glioma. , , CSDE1 is an RNA‐binding protein and coordinates oncogenic RNA regulons, such as VIM and RAC1 genes. Through enhancing translation elongation of VIM and RAC1 mRNAs, CSDE1 upregulates the expression of VIM and RAC1, and, thus, promotes HCC development. , In line with these findings, we found that CSDE1 also plays its role as an oncogene in HCC and lncRNA ARHGAP5‐AS1 could interrupt the binding of CSDE1 with its E3 ligase TRIM28, stabilise CSDE1 protein, elevate expression of VIM and RAC1, and stimulate ERK signalling. This may underline mechanisms of how ARHGAP5‐AS1 contributes to HCC proliferation and metastasis ex vivo and in vivo.
In conclusion, we identified an oncogenic lncRNA, ARHGAP5‐AS1, by comprehensively analysing m6A‐modified RNAs and profiling lncRNA expression in HCC cells. Significantly elevated expression of lncRNA ARHGAP5‐AS1 due to its high m6A methylation levels accelerates the translation of VIM and RAC1 and stimulation of the ERK signalling pathway, which promotes cell proliferation and metastasis. Our current study reveals a novel paradigm in m6A‐modified lncRNAs controlling CSDE1‐mediated oncogenic RNA regulons and highlights lncRNAs as potential targets for future therapeutics against HCC.
The authors declare no competing financial interests.
Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. |
|
PMC9647896 | Ramla Shabbir,Muhammad Nasir Hayat Malik,Maryam Zaib,Alamgeer,Shah Jahan,Muhammad Tariq Khan | Amino Acid Conjugates of 2-Mercaptobenzimidazole Ameliorates High-Fat Diet-Induced Hyperlipidemia in Rats via Attenuation of HMGCR, APOB, and PCSK9 | 27-10-2022 | Purpose: This study was designed to explore the antihyperlipidemic effects of amino acid derivatives of 2-mercaptobenzimidazole (4J and 4K) in high-fat diet (HFD)-fed rats. Methods: Male Sprague-Dawley rats were divided into nine groups which received either standard diet or HFD for 28 days. Blood samples were taken on 27th day from HFD-fed rats to ensure hyperlipidemia. HFD-induced hyperlipidemic rats later received daily dosing of either vehicle or simvastatin (SIM; 20 mg/kg) or 4J/4K compounds (10, 20, and 30 mg/kg) for 12 consecutive days. On 40th day, animals were sacrificed, and blood samples were collected for the determination of serum lipid profile and liver function parameters. Liver samples were harvested for histopathological, antioxidant, and qPCR analyses. Molecular docking of tested compounds with HMGCR was also performed to assess the binding affinities. Results: 4J and 4K dose dependently decreased serum total cholesterol, triglycerides, low-density lipoprotein, very low-density lipoproteins, alanine transaminase (ALT), and aspartate aminotransferase (AST) levels while significantly alleviated high-density lipoproteins. However, SIM failed to reduce AST and ALT levels. Moreover, tested compounds displayed antioxidant effects by inducing superoxide dismutase and glutathione levels. Histopathology data also displayed protective effects of 4J and 4K against HFD-induced fatty changes and hepatic damage. In addition, 4J and 4K downregulated transcript levels of HMGCR, APOB, PCSK9, and VCAM1, and molecular docking analysis also supported the experimental data. Conclusion: It is conceivable from this study that 4J and 4K exert their antihyperlipidemic effects by modulating multiple targets regulating lipid levels. | Amino Acid Conjugates of 2-Mercaptobenzimidazole Ameliorates High-Fat Diet-Induced Hyperlipidemia in Rats via Attenuation of HMGCR, APOB, and PCSK9
Purpose: This study was designed to explore the antihyperlipidemic effects of amino acid derivatives of 2-mercaptobenzimidazole (4J and 4K) in high-fat diet (HFD)-fed rats. Methods: Male Sprague-Dawley rats were divided into nine groups which received either standard diet or HFD for 28 days. Blood samples were taken on 27th day from HFD-fed rats to ensure hyperlipidemia. HFD-induced hyperlipidemic rats later received daily dosing of either vehicle or simvastatin (SIM; 20 mg/kg) or 4J/4K compounds (10, 20, and 30 mg/kg) for 12 consecutive days. On 40th day, animals were sacrificed, and blood samples were collected for the determination of serum lipid profile and liver function parameters. Liver samples were harvested for histopathological, antioxidant, and qPCR analyses. Molecular docking of tested compounds with HMGCR was also performed to assess the binding affinities. Results: 4J and 4K dose dependently decreased serum total cholesterol, triglycerides, low-density lipoprotein, very low-density lipoproteins, alanine transaminase (ALT), and aspartate aminotransferase (AST) levels while significantly alleviated high-density lipoproteins. However, SIM failed to reduce AST and ALT levels. Moreover, tested compounds displayed antioxidant effects by inducing superoxide dismutase and glutathione levels. Histopathology data also displayed protective effects of 4J and 4K against HFD-induced fatty changes and hepatic damage. In addition, 4J and 4K downregulated transcript levels of HMGCR, APOB, PCSK9, and VCAM1, and molecular docking analysis also supported the experimental data. Conclusion: It is conceivable from this study that 4J and 4K exert their antihyperlipidemic effects by modulating multiple targets regulating lipid levels.
Cardiovascular diseases (CVDs) account for one-third of deaths worldwide, and it is projected that CVD would be the leading cause of morbidity and mortality in the developing world. The risk factors such as hyperlipidemia, diabetes, and hypertension are prevalent in individuals who are affected by CVDs. People with hyperlipidemia are considered at high stake for contracting CVDs. Hyperlipidemia is a state of increased levels of cholesterol, triglycerides (TG), or both in the blood which is usually caused by a variety of genetic or acquired disorders. These lipids accumulate in the walls of the arteries and increase the chances of atherosclerosis that can lead to life-threatening CVDs. Moreover, plasma lipoproteins also play an important role in the transport of cholesterol and TG to the site of absorption, catabolism, or elimination. The five major classes of lipoproteins are the chylomicrons, very low-density lipoproteins (VLDLs), low-density lipoproteins (LDLs), intermediate-density lipoproteins, and high-density lipoproteins (HDLs). The specific biomarkers for hyperlipidemia include high levels of total cholesterol (TC), TG, and LDL and low levels of HDL. Moreover, certain apolipoproteins (APOs) which are the constituents of the various lipoprotein classes also regulate the metabolism of plasma lipoproteins. They play three major functions, that is, (I) stabilization of micellar structures of lipoprotein particles, (II) acting as cofactors or activators of various enzymes or lipid-transfer proteins that participate in “remodeling” of lipoproteins, and (III) serving as a ligand for cell surface lipoprotein receptors. Therefore, LDL, HDL, and APO have been the targets of therapy for improving the outcomes in hyperlipidemic patients. Although a large number of drugs therapies are available for the treatment of hyperlipidemia like niacin, fibrates, statins, and bile acid binding resins but they are associated with lots of side effects. Statin therapy has long been a mainstay in the treatment of hypercholesterolemia, but it also has adverse effects that usually lead to patient noncompliance. Hence, various medicines are being developed for the management of hyperlipidemia. Recently, Food and Drug Administration approved two medications “Evolocumab and Alirocumab” which target a novel pathway to reduce LDL. These are monoclonal antibodies that inactivate proprotein convertase subtilsin-kexin type 9 (PCSK9). PCSK9 regulates LDL receptor (LDLR) degradation and is therefore regarded as a potential target for modulating LDLR expression and consequently LDL levels. However, the safety profile of these monoclonal antibodies is similar to statins. Therefore, there is a need to develop safer and efficacious antihyperlipidemic agents. In recent years, benzimidazole derivatives have gained importance due to their diverse pharmacological activities which include antimicrobial, antiviral, anticancer, anti-inflammatory, antioxidant, antihypertensive, anticoagulant, immunomodulator, antihyperlipidemic and antidiabetic effects. Benzimidazole nucleus is now considered an indispensable pharmacophore for the development of new therapeutic agents. Keeping in view the wide range of biological effects, we also evaluated the antihyperlipidemic activity of amino acid derivatives of 2-mercaptobenzimidazole (4J and 4K) in rats.
Cholesterol and cholic acid were purchased from Sigma-Aldrich (St. Louis, Missouri, USA), and simvastatin was gifted by Medpak Pharmaceuticals (Lahore, Pakistan). Standard kits were purchased from BioLabs (Boston, USA) and Zokeyo (Wuhan, China). Coconut oil, banaspati ghee, and all other chemicals used in this study were of analytical grade.
Male Sprague-Dawley rats weighing around 180–200 g were purchased from University of Veterinary and Animal Sciences, Lahore, Pakistan. The animals were kept under standard conditions in the animal house of Faculty of Pharmacy, The University of Lahore and were acclimatized to the laboratory conditions prior to the start of experiments. All the experiments were performed according to the guidelines of organization for economic co-operation and development (OECD), and the experimental protocols were approved by Institutional Research Ethics Committee of Faculty of Pharmacy, The University of Lahore (approval no.: IREC-2020-42).
Hyperlipidemia was induced by feeding high-fat diet (HFD) to animals for 28 days. HFD was prepared by homogeneous mixing of cholesterol (2% w/w), cholic acid (1% w/w), banaspati ghee, coconut oil (3:2 w/w), and egg yolk powder (5% w/w) with standard rat chow. After 27 days of HFD, blood samples were taken for the analysis of serum TC, TG, HDL, LDL, and VLDL. Animals having TC levels greater than 280 g/dL were selected for further study. Hyperlipidemic rats were later treated once a day with either simvastatin (SIM; 20 mg/kg) or with various doses (10, 20, and 30 mg/kg) of MBIZ, that is, 4J and 4K for 12 consecutive days. On 40th day, animals were sacrificed, and blood samples were collected for subsequent analyses of abovementioned lipid levels and liver function tests (LFTs). Liver samples were harvested in Trizol and 10% buffered formalin for mRNA and histopathological analyses, respectively.
The experimental animals were divided into nine groups, each containing three rats (n = 3). The division of groups was in the following manner: 1. Control—normal diet 2. Disease group—HFD 3. SIM (20 mg/kg)—HFD followed by SIM 4. 4J (10 mg/kg)—HFD followed by 4J (10 mg/kg) 5. 4J (20 mg/kg)—HFD followed by 4J (20 mg/kg) 6. 4J (30 mg/kg)—HFD followed by 4J (30 mg/kg) 7. 4K (10 mg/kg)—HFD followed by 4K (10 mg/kg) 8. 4K (20 mg/kg)—HFD followed by 4K (20 mg/kg) 9. 4K (30 mg/kg)—HFD followed by 4K (30 mg/kg)
Formalin fixed liver samples were processed by successive dehydration with ethanol baths and later embedded in paraffin. Paraffin blocks were sectioned at 5 μm using a rotary microtome and stained with hematoxylin (H) and eosin (E) using standard procedures.
Liver samples from disease and treated groups were harvested in 1× PBS. After washing, samples were homogenized and stored overnight at −20 °C. The samples were freeze–thaw two times and centrifuged to obtain the supernatants. Supernatants were later processed according to manufacturer’s instructions to measure superoxide dismutase (SOD) and glutathione (GSH) levels using standard ELISA kits.
Total RNA was extracted from liver samples using the Trizol reagent and was reverse transcribed using the WizScript cDNA synthesis kit (Wizbio solutions, New Mexico, USA). The relative transcript levels of various genes were measured by the ddCT method using SYBR Green qPCR mix (Zokeyo, Wuhan, China). mRNA levels were measured using the following conditions: initial denaturation at 94 °C for 2 min followed by 40 cycles of denaturation at 94 °C for 1 min and annealing at 60 °C for 2 min. Hypoxanthine–guanine phosphoribosyltransferase mRNA was used as an internal control. Sequences of primers are given in Table S1.
The three-dimensional (3D) structure of HMGCR was accessed from the Protein Data Bank (PDB) (www.rcsb.org) with PDB ID of 1T02. The target protein was prepared for docking analysis using Autodock Tools program. The protein was energy minimized, and Gasteiger charges were added and saved in the pdbqt format. The hydrophobicity and Ramachandran graphs were generated by Discovery Studio 4.1 Client (2012). The protein architecture and statistical percentage values of helices, β-sheets, coils, and turns were accessed by VADAR 1.8. The compounds (SIM, 4J, and 4K) were drawn in Discovery Studio Client and saved in the pdb format as ligands after energy minimization. Autodock tools were used for the preparation of ligands in their most stable conformation. The ligands were saved in the pdbqt format after addition of the Kolman and Gasteiger charges. The molecular docking experiment was used for all the synthesized ligands against HMGCR by the PyRx virtual screening tool with the Auto Dock VINA Wizard approach. The grid box center values were as follows (center X = 15.861, center Y = −8.806, and center Z = −22.278), and size values were adjusted (X = 88, Y = 60, and Z = 96) for better conformational position in the active region of the target protein. Ligands were docked individually against target protein with default exhaustiveness value of 50. The predicted docked complexes were evaluated based on the lowest binding energy values (kcal/mol). The 2D and 3D graphical depictions of all the docked complexes were accomplished by Discovery Studio (Discovery Studio Visualizer Software, Version 4.0., 2012).
Results were expressed as the mean ± standard deviation (SD), and data were analyzed by one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test using GraphPad prism 5.1 (Graphpad Software, Inc., San Diego, USA). Probability values of less than 0.05 were considered as statistically significant using the following abbreviations. *** = <0.001, ** = <0.01, and * = <0.05.
Our findings demonstrated that HFD significantly induced TC, TG, LDL, and VLDL levels and reduced HDL levels in the disease group. 4J and 4K treatments showed beneficial effects on the lipid profile as they remarkably restored all the lipid parameters. In addition, they displayed a dose-dependent improvement in the lipid profile, and at higher dose (30 mg/kg), their effects were comparable to the SIM-treated group (Figure 1).
HFD is known to alter liver functions and can cause fatty liver disease; we therefore measured aspartate aminotransferase (AST) and alanine transaminase (ALT) levels. As expected, the disease group and SIM-treated group showed prominent increase in AST and ALT levels indicating liver dysfunction. Treatment with 4J and 4K exhibited a dose-dependent reduction in HFD-induced AST and ALT levels, suggesting their hepatoprotective effects as well (Figure 2).
Histopathology of liver samples showed that control group had normal hepatocytes, and there were no signs of vacuolization. Hepatic cords also displayed perfect architecture with normal portal triads and central veins. Disease group specimens exhibited severe fatty changes which included cytoplasmic vacuolization, derangement of hepatic cords, inflammation, necrosis, and extra-medullary hemopoieses. Treatment with SIM reduced most of the pathological changes and showed moderate degree of fatty changes. The cells around the central vein showed normal cytoplasm. 4J treated group displayed mild degree of cytoplasmic vacuolization in hepatocytes. However, the hepatic architecture (hepatic cords) was intact, and no inflammatory or necrotic changes were noticed. Similarly, the liver sections of 4K-treated groups showed mild fatty change. Most of the hepatocytes displayed normal parenchyma with minimal to no vacuolization. Central vein and portal area also showed normal histological appearance. The findings clearly indicate the protective effects of 4J and 4K against HFD-induced hyperlipidemia (Figure 3).
Prolong use of HFD induces oxidative stress; we therefore measured antioxidant levels in the homogenates of liver samples. As anticipated, HFD drastically reduced the levels of SOD and GSH. Treatments with SIM or tested compounds significantly restored the levels of antioxidants, indicating their antioxidant potentials as well (Figure 4).
In order to identify the possible molecular mechanism behind antihyperlipidemic effects of MBIZ, we measured the transcript levels of lipid-regulating genes (HMGCR, APOB, APOE, PCSK9, LDLR, and VCAM1). RT-qPCR data showed that HFD significantly induced HMGCR, APOB, and VCAM1 and reduced APOE levels which are normally associated with hyperlipidemia. In addition, HFD did not alter the normal levels of PCSK9 and LDLR. Treatment with MBIZ (4J and 4K) significantly reduced the levels of HMGCR, APOB, and PCSK9. SIM on the other hand failed to reduce PCSK9 levels, which is a well-known effect of HMGCR inhibitors. These findings indicate that the antihyperlipidemic effects of 4J and 4K are due to inhibition of multiple lipid-regulating targets (Figure 5).
Structural analysis of target protein showed that HMGCR (PDB ID: 1T02) consisted of 45% helices (338 residues), 24% β-sheets (180 residues), 30% coils (227 residues), 8% turns (60 residues), and a total of 805 amino acid residues. The R-value of selected protein appeared to be 0.251, and the resolution was 2.50 Å. Unit cell dimensions for the lengths were observed to be as follows: a = 228.5, b = 228.5, and c = 228.5 with 90° angle for α, β, and γ. The Ramachandran plot confirmed that 97% amino acids were in the allowed regions for the phi (φ) and psi (ψ) angles (Figure 6). The affinity among the target protein and the ligands was investigated using molecular docking. AutoDock Vina program was used for the docking analysis through the PyRx user interface. The E-value (kcal/mol) was used to assess the affinity of protein and best docked pose complex. It provided prediction of binding free energy and binding constant for docked ligands. The results obtained from the docking studies of compounds against HMGCR were in conjuction with their pharmacological activities. Binding affinities of 4J and 4k with HMGCR were −5.0 and −6.9 kcal/mol, respectively, which were comparable to SIM (−7.4 kcal/mol) (Table 1; Figures 7 and 8).
Hyperlipidemia is a metabolic disease with elevated levels of lipid and/or lipoproteins in the blood. The long-term consumption of HFD and unhealthy lifestyle have contributed to the prevalence of hyperlipidemia. Patient non-compliance for medications is another major contributing factor for its prevalence. Although the current therapies effectively reduce plasma lipid levels, most of these treatments cause unwanted adverse effects. Therefore, scientists are looking for better and safer alternatives for the treatment of hyperlipidemias. In this study, we evaluated the antihyperlipidemic potential of MBIZ (4J and 4K) against HFD-induced hyperlipidemia. Previous studies have shown that these MBIZ exhibited strong analgesic and anti-inflammatory activities with better safety profile. In our study, we also did not witness any signs of apparent toxicity or lack of physical activity in animals. Both 4J and 4K prominently reduced TC, TG, LDL, and VLDL levels, and they were equally effective in enhancing HDL levels. Moreover, LFT analysis also revealed a hepatoprotective effect of these compounds, and similar findings were observed in histopathological examinations of liver samples. 4J and 4K effectively prevented HFD-induced vacuolization, inflammation, and necrosis of hepatocytes, and all the specimens of MBIZ-treated groups displayed normal hepatic architecture. The tested compounds also effectively raised the levels of SOD and GSH, demonstrating significant antioxidant effects. In addition, 4J and 4K demonstrated prominent HMGCR, APOB, PCSK9, and VCAM1 inhibitory activities. HMGCR plays a pivotal role in the cholesterol biosynthetic pathway, and in fact, it is the rate-limiting step of cholesterol biosynthesis. The conversion of HMG-CoA to mevalonate is the final step in the synthesis of cholesterol, which is inhibited by HMGCR. The inhibition of HMGCR can reduce LDL-cholesterol by 20–35%, a decrease which has been linked to a reduction in the incidences of CVDs. The transcription of HMGCR gene and the rate of synthesis of this enzyme are controlled by the sterol regulatory element-binding proteins (SREBPs). The SREBPs also promote LDLR expression and thereby regulate the cellular uptake, transport, and utilization of cholesterol. PCSK9 is another modulator of the cellular LDLR and plasma cholesterol levels. Studies have shown that gain-of-function mutations of PCSK9 cause hyperlipidemia and early onset coronary heart disease, whereas loss-of-function mutations result in low plasma cholesterol levels and protection against coronary heart disease without apparent negative consequences. PCSK9 interacts with LDLR epidermal growth factor-like repeat A domain and induces its lysosomal degradation, thereby regulates plasma cholesterol levels. Therefore, inhibition of PCSK9 is being pursued as an approach to reduce plasma LDL cholesterol and TG levels. APOB is an essential structural and receptor-binding component of all atherogenic lipoproteins, including LDL. Atherogenic lipoproteins of hepatic origin carry one molecule of APOB100 per lipoprotein particle, whereas APOB48, a truncated form of APO B100, is found in chylomicrons synthesized in the intestine. Mouse and human genetic models have shown that inhibition of hepatic APOB production may be a therapeutic approach for the treatment of dyslipidemia. The effects of antisense inhibition of APOB synthesis on lipid metabolism have been extensively studied across a number of species. The species-specific antisense APOB inhibitors have been reported to reduce LDL, VLDL, and TC.VCAM1 is another well-recognized marker of atherosclerotic plaque vulnerability. Its overexpression has been observed over the complete course of plaque development.VCAM1 expression promotes the adhesion of leukocytes to the endothelial cells, accelerates the migration of adherent leukocytes along the endothelial surface, and facilitates the proliferation of smooth muscle cells. Therefore, VCAM1 is considered a key player in the pathogenesis of atherosclerosis. Studies on animal models have shown that increased LDL levels promote the expression of VCAM1 in endothelial cells, and hypercholesterolemia can cause atherosclerosis-related pathophysiological changes in the arteries. 4J and 4K have clearly demonstrated prominent inhibition of aforementioned hyperlipidemic factors indicating their strong lipid lowering effects. Moreover, these compounds also reduced liver toxicity which is normally observed with the chronic use of statins. Since they have suppressed VCAM1 levels as well, it is plausible that they might also be effective in reducing or preventing atherosclerosis.
Based on recent findings, it can be concluded that the antihyperlipidemic effects of 4J and 4K could be attributed to the inhibition of HMGCR, APOB, PCSK9, and VCAM1. However, further studies are still required to validate these findings in vitro and in vivo hyperlipidemic models. |
||
PMC9647929 | You Wu,Boju Sun,Xiaoyuan Guo,Lili Wu,Yaomu Hu,Lingling Qin,Tao Yang,Mei Li,Tianyu Qin,Miao Jiang,Tonghua Liu | Zishen Pill alleviates diabetes in Db/db mice via activation of PI3K/AKT pathway in the liver | 10-11-2022 | ZiShen Pill,Diabetes,Insulin resistance,Transcriptomics,Glucolipid metabolism | Background The rising global incidence of type 2 diabetes mellitus (T2DM) highlights a need for new therapies. The Zishen Pill (ZSP) is a traditional Chinese herbal decoction that has previously shown hypoglycemic effects in C57BL/KsJ-db/db mice, although the therapeutic mechanism remains unknown. This study aims to explore the underlying mechanisms of ZSP’s hypoglycemic effects using db/db mice. Methods Db/db mice were divided into two groups: model group and ZSP group, while wt/wt mice were used as a normal control. ZSP was given to mice by gavage for 40 days. During treatment, blood glucose level and body weight were monitored continuously. Oral glucose tolerance test (OGTT) was performed at day 35. Blood and tissue samples were collected at the end of treatment for further analyses. Mice liver samples were analyzed with mRNA transcriptomics using functional annotation and pathway enrichment to identify potential mechanisms that were then explored with qPCR and Western Blot techniques. Results ZSP treatment significantly reduced weight gain and glycemic severity in db/db mice. ZSP also partially restored the glucose homeostasis in db/db mice and increased the hepatic glycogen content. Transcriptomic analyses showed ZSP increased expression of genes involved in glycolysis including Hk2, Hk3, Gck and Pfkb1, and decreased expression of G6pase. Additionally, the gene and protein expression of phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT) pathway, and Csf1 and Flt3 mRNA expression were significantly upregulated in ZSP group. Conclusion ZSP treatment reduced the severity of diabetic symptoms in db/db mice. ZSP increased expression of genes associated with glycogen synthesis and glycolysis, and decreased gluconeogenesis via the enhancement of the PI3K/AKT signaling in the liver. Supplementary Information The online version contains supplementary material available at 10.1186/s13020-022-00683-8. | Zishen Pill alleviates diabetes in Db/db mice via activation of PI3K/AKT pathway in the liver
The rising global incidence of type 2 diabetes mellitus (T2DM) highlights a need for new therapies. The Zishen Pill (ZSP) is a traditional Chinese herbal decoction that has previously shown hypoglycemic effects in C57BL/KsJ-db/db mice, although the therapeutic mechanism remains unknown. This study aims to explore the underlying mechanisms of ZSP’s hypoglycemic effects using db/db mice.
Db/db mice were divided into two groups: model group and ZSP group, while wt/wt mice were used as a normal control. ZSP was given to mice by gavage for 40 days. During treatment, blood glucose level and body weight were monitored continuously. Oral glucose tolerance test (OGTT) was performed at day 35. Blood and tissue samples were collected at the end of treatment for further analyses. Mice liver samples were analyzed with mRNA transcriptomics using functional annotation and pathway enrichment to identify potential mechanisms that were then explored with qPCR and Western Blot techniques.
ZSP treatment significantly reduced weight gain and glycemic severity in db/db mice. ZSP also partially restored the glucose homeostasis in db/db mice and increased the hepatic glycogen content. Transcriptomic analyses showed ZSP increased expression of genes involved in glycolysis including Hk2, Hk3, Gck and Pfkb1, and decreased expression of G6pase. Additionally, the gene and protein expression of phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT) pathway, and Csf1 and Flt3 mRNA expression were significantly upregulated in ZSP group.
ZSP treatment reduced the severity of diabetic symptoms in db/db mice. ZSP increased expression of genes associated with glycogen synthesis and glycolysis, and decreased gluconeogenesis via the enhancement of the PI3K/AKT signaling in the liver.
The online version contains supplementary material available at 10.1186/s13020-022-00683-8.
The most recent report by the International Diabetes Federation, estimates the global prevalence of diabetes in adults to be over 10.5% in 2021, effecting hundreds of millions of people [1]. Approximately 90% of these global diagnoses are type 2 diabetes mellitus (T2DM), which is typically accompanied by obesity, cardiovascular disease and other metabolic stressors, causing substantial economic burden on global health systems [2–4]. The rising global incidence indicates that present treatments and preventions for diabetes are insufficient, and that new more potent therapies will be needed. One potential avenue for new treatments is the exploration of ethnopharmacological methods used to treat diabetes [5]. Traditional Chinese Medicine (TCM), has been practiced for more than 2000 years, and is an important complementary therapy in many countries [6, 7]. The ZiShen Pill (ZSP), also known as TongGuan Wan, is a TCM herbal formular reported to have anti-diabetic effects. ZSP is composed of three herbs: Rhizoma Anemarrhenae, Cortex Phellodendri and Cinnamomum cassia, and a previous study demonstrated anti-diabetic and anti-obesity functions in C57BL/KsJ-db/db mice [8]. Our previous study confirmed the hypoglycemic effect of ZSP application and found it effective in treating diabetic nephropathy by alleviating inflammation [9]. Despite continued interest the molecular mechanisms of ZSP’s therapeutic effects remain unclear. Insulin resistance is a hallmark of T2DM, defined by a decrease of insulin sensitivity in insulin-targeted tissues including the liver, skeletal muscle and adipose tissue [10]. Insulin resistance may lead to dysregulated glucolipid metabolism which manifests as an increased blood glucose level. The liver is one of the most important organs in energy metabolism and plays an important role in both glucose and lipid metabolic processes [11]. The liver regulates glucose homeostasis by controlling glycogen content and hepatic gluconeogenesis [11], and influences the lipid metabolism by de novo lipogenesis and fatty acid oxidation [12]. As such, liver tissue relies heavily on insulin signaling to maintain glucose levels, hepatic tissue is exposed to insulin via the portal vein, making the concentration of insulin in the liver two to three times higher than in the wider circulatory system. Insulin activity in the liver occurs via the phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT) signal cascade to exert various metabolism-regulating functions [13, 14]. PI3K/AKT signalling is essential for cellular physiology [15] with a well-established role in diabetes [16, 17]. In mammals, PI3K exists in 3 classes, with class I PI3K has a variety of functions closely related to metabolism. The catalytic subunit of class I PI3K is encoded by four genes: pik3ca, pik3cb, pik3cd and pik3cg. Likewise there are 3 AKT isoforms, AKT1 and AKT2 exist in the liver where they regulate glucose metabolism [14]. In this study, we explore the effects of ZSP on db/db mice, a common animal model for diabetes, and use transcriptomics to investigate the changes in gene expression in treated mice livers to investigate underlying mechanisms for ZSP’s anti-diabetic effects.
ZSP is composed of three herbs: Rhizoma Anemarrhenae, Cortex Phellodendri and Cinnamomum cassia. Extracts of Rhizoma Anemarrhenae, Cortex Phellodendri and Cortex Cinnamomi (extraction ratio: 10:1) were purchased from Shaanxi Zhongxin Biotechnology (Xi’an, Shaanxi, China). To produce the decoction, the three extracts were mixed at a ratio of 10:10:1, dissolved in distilled water to 0.468 g/mL and thoroughly stirred before gavage.
The components of ZSP were detected by LC–MS/MS analysis. LC–MS/MS analysis was conducted using an Agilent 1290 ultra-high performance liquid chromatography system (Agilent Technologies, Santa Clara, CA, US) with a UPCL BEH C18 column (1.7 μm*2.1 mm*100 mm, Waters, Milford, MA, US). 5 μL of ZSP sample was loaded onto the system and flow rate was set at 400 μL/min. The elution program is described in Additional file 1: Table S1. The MS and MS/MS data were obtained by a Q Exactive Focus mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). Mass ranged from 100 to 1500 in each cycle, and the top three of every cycle were screened for further MS/MS data acquisition. The MS/MS data was matched to the database provided by Shanghai BIOTREE biotech Co., Ltd to identify the materials.
The animal experiments in this study were approved by the Animal Care and Ethics Committee of Beijing University of Chinese Medicine (approval No. BUCM-4-2019031003-1089). A total number of 14 6 week-old male C57BL/KsJ-db/db mice and 7 C57BL/KsJ-wt/wt mice of the same age and gender were purchased from Nanjing Biomedical Research Institute of Nanjing University (Nanjing, China). Mice were kept in individually ventilated cages with free access to food (normal chow diet) and water. The cages were placed in a specific-pathogen-free environment with a standard 12/12 h artificial light/dark cycle. After 1 week of accommodation, db/db mice were randomly divided into two groups (ZSP group, receiving 3.3 g/kg ZSP by gavage; model group, receiving distilled water by gavage) according to their blood glucose levels and body weights. Wt/wt mice were used as normal group (receiving distilled water by gavage). The dosage of ZSP treatment is calculated according to the dosage of human by the body surface area estimation method determined according to our previous study [9]. The treatment lasted for 40 days and mice received normal chow diet throughout. The body weight and food intake were measured every 3 days. Overnight fasting glucose level was measured every 7 days using blood drawn from the tail vein. At the end of treatment, the body length of mice was measured and body mass index (BMI) calculated as follows:
OGTT was conducted on mice at day 35 after over-night fasting. All mice were orally administered 1 g/kg body weight of glucose (dissolved in distilled water to 10%w/v). Blood glucose levels were measured at administration and after 30, 60, and 120 min of glucose treatment. The area under curve (AUC) of OGTT was calculated as follows:(G0, G30, G60 and G120 represent blood glucose at 0, 30, 60, and 120 min, respectively).
After 40 days of treatment, mice were sacrificed after anesthesia with isoflurane. Blood was drawn from abdominal aorta and collected in EP tubes, then centrifuged at 3000 rpm for 15 min at 4 °C to obtain serum. Serum was stored at − 20 °C. Liver and adipose tissue were removed and weighed. Tissue samples were snap frozen with liquid nitrogen and stored at − 80 °C. A slice of each tissue was kept in a 4% paraformaldehyde-PBS solution (Cat.P1110, Solarbio Science & Technology, Beijing, China) for 24 h before being embedded in paraffin. Hematoxylin & eosin (H&E) staining was conducted for morphological observation, oil red O staining and periodic acid-Schiff (PAS) staining was performed to examine the lipid accumulation and glycogen content, respectively.
Serum lipids including total cholesterol (TC), triglyceride (TG), high-density-lipoprotein (HDL) and low-density-lipoprotein (LDL), as well as hepatic function indicator alanine transaminase (ALT) and aspartate aminotransferase (AST) levels were measured by a chemistry analyzer (AU480, Beckman Coulter, Brea, CA, US). Hepatic glycogen was tested by colorimetric analysis using a commercial kit (Cat. BC0345, Solarbio Science & Technology). Fasting serum insulin levels were measured by enzyme-linked immunosorbent assay (ELISA) using Mouse Ultrasensitive Insulin ELISA kit (Cat. 80-INSMSU-E01, ALPCO Diagnostics, Salem, NH, US). Homeostatic model assessment for insulin resistance (HOMA-IR) was calculated as follows:
Total RNA from liver were extracted from 6 random-chosen mice using an RNA isolation kit (Cat. AM1560, Ambion, Austin, TX, US) according to the manufacturer’s protocol of extracting total RNA. After quality and integrity checks, RNA was used for library construction with TruSeq Stranded mRNA Sample Prep Kit (Cat. RS-122-2101, Illumina, San Diego, CA, US). Raw reads of RNA were generated by the Illumina HiSeq™ X Ten platform. Reads containing ploy-N and low-quality reads were removed. Clean reads were mapped to NCBI database (GRCm39) using HISAT2 [18].
Fragments per kb per million reads (FPKM) value [19] of gene reads was calculated. DESeq2 were applied to standardize and analyze mRNA expression levels. Threshold for differentially expressed genes (DEGs) were defined as p < 0.05 and foldchange >1.5. Principal component analysis (PCA) was used to analyze the relationship between transcriptome of samples. For functional enrichment, Gene Ontology (GO) database and Kyoto Encyclopedia of Genes and Genomes (KEGG) database were used to annotate and analyze identified DEGs.
Total RNA was extracted using RNA Easy Fast Tissue/Cell Kit (Cat. DP451, Tiangen Biotech, Beijing, China) according to the manufacturer’s instructions. Tissue was homogenized with lysis buffer, the suspensions were filtered through a gDNA eraser column set and an RNase-free spin column. RNA solutions were obtained after column elution with RNase-free double-distilled water. Reverse transcription was performed using HiScript III RT SuperMix for qPCR (Cat. R323-01, Vazyme, Nanjing, China) with 0.5 ng RNA per reaction. RT-qPCR was performed on an Applied Biosystems 7500 Real-Time PCR instrument (Thermo Fisher Scientific) using GoTaq® qPCR Master Mix Kit (Promega Biotech, Madison, WI, US). Primers used in RT-qPCR are listed in Additional file 1: Table S2. The RNA expression levels were normalized to Gapdh and quantitated by 2−ΔΔCT method.
Glycogen synthase (GYS) expression levels were assessed by IHC staining. Liver slices were incubated with anti-GYS primary antibody (Cat. 3886, Cell Signaling Technology) at 4 °C overnight, then incubated with HRP-conjugated goat anti-rabbit IgG secondary antibody (Cat. PV-6000, ZSGB-Bio, Beijing, China) for 20 min and stained with 3,3 N-Diaminobenzidine tertrahydrochloride (DAB) and counterstained with hematoxylin.
Total protein was extracted from liver and adipose tissues using a Total Protein Extraction kit (Cat. KGP250, KeyGen Biotech, Nanjing, China). Protein samples were run in polyacrylamide gels and transferred onto polyvinylidene fluoride membranes (Cat. IPVH00010, Millipore, Bedford, MA, US). After 1 h of blocking at room temperature, membranes were incubated with primary antibodies: PI3K p85 (Cat. 4257, Cell Signaling Technology), p-PI3K (phosphorylation sites: p85 Tyr458/p55 Tyr199, Cat. 4228, Cell Signaling Technology), AKT (Cat. 4685, Cell Signaling Technology), p-AKT (phosphorylation site: Thr308, Cat. 13038, Cell Signaling Technology), and β-actin (Cat. 4970, Cell Signaling Technology) overnight at 4 °C. Membranes were then incubated with HRP-conjugated AffiniPure goat anti-rabbit IgG secondary antibody (Cat. SA00001-2, Proteintech Group, Rosemont, IL, US) for 1 h at room temperature. Blots were visualized using an ultra-sensitive enhanced chemiluminescence reagent (Cat. PE0010, Solarbio). The protein expression levels were analyzed by Image Lab software (Version 3.0, Bio-Rad, Hercules, CA, US) and normalized to β-actin.
All data are shown as means ± standard deviation (SD), where n represents the number of biological replicates. Statistical analysis was performed using IBM SPSS Statistics software (Version 23.0, SPSS, Chicago, IL, US). Graphs were composed by GraphPad Prism (Version 7.0, GraphPad Software, San Diego, CA, US). Normally distributed data were analyzed by one-way analysis of variation (ANOVA) with Fisher’s Least Significant Difference (LSD) tests or Dunnett’s test post hoc. Non-normal data were analyzed by nonparametric Kruskal–Wallis test. Comparisons with p value < 0.05 were considered statistically significant.
The LC–MS/MS analysis identified several characteristic chemicals of the ZSP decoction reported previously [9, 20, 21], including mangiferin, cinnamic acid, berberine and several terpenoids like Timosaponin A-III and Timosaponin B-II. Different chemicals were marked in the base peak chromatogram from positive ion mode (Fig. 1a) and negative ion mode (Fig. 1b). Details of the LC–MS/MS results were listed in Additional file 1: Table S3.
ZSP treatment did not show acute or chronic toxicity in mice. ZSP treatment had no significant impact on the food or water intake (Additional file 1: Fig. S1a) and did not affect the growth of db/db mice, as body length showed no significant difference between groups (Additional file 1: Fig. S1c). The serum AST and ALT levels were consistent at day 28 or day 40, suggesting ZSP at dose 3.3 g/kg did not negatively impact the liver function of mice (Additional file 1: Fig. S1d).
The ZSP treatment group showed significantly lower blood glucose after 28 days (Fig. 2a). ZSP did not alter the serum insulin levels in db/db mice, but did significantly lower the HOMA-IR, suggesting improved insulin sensitivity (Fig. 2b, c). The ZSP group also showed better performance in the OGTT experiment and showed a significant lower AUC value (Fig. 2d). PAS staining indicated that the ZSP treatment improved the glycogen content in the liver, but not the muscle tissue (Fig. 2e), which was corroborated by the glycogen content assay (Fig. 2f).
The db/db mice showed significantly higher body weight compared to the wt/wt mice. After 26 days of treatment, the ZSP treated group showed notable decrease in BMI (Fig. 3a, b). Mice in the model group had higher serum TC, TG, and LDL levels compared to those in normal group. The serum TC and TG values of mice in ZSP group were not significantly different to those in normal group suggesting that ZSP treatment did reduce serum TC content in db/db mice (Fig. 3c). ZSP treatment also significantly lowered the liver weight and epididymal fat weight of db/db mice (Fig. 3d). The H&E-stained images showed that ZSP partially restored liver steatosis, and oil red O staining suggested that hepatic lipid accumulation was decreased in the ZSP treatment group. Size of epididymal adipocytes in ZSP group were smaller than those in model group. (Fig. 3e).
One outlier from both the ZSP group and the model group, were removed due to their inconsistency with other samples from the same group (Fig. 4a). Transcriptomics identified 1801 and 3411 DEGs between the normal and model groups respectively, indicative of a substantial difference in hepatic gene transcription (Fig. 4b). GO enrichment of DEGs from liver transcriptomes revealed ZSP significantly changed glucokinase (GCK), hexokinase (HK) and fructokinase expression (Fig. 4c). HK and GCK catalyze the phosphorylation of hexoses including glucose to hexose-6-phosphate [22–24], the first step in glycolysis. Upregulation of these genes may indicate ZSP enhances glucose utilization through increasing glycolysis. KEGG enrichment analysis revealing a number of DEGs were involved in metabolic processes. Specifically, 21 DEGs were involved in lipid metabolism, 15 DEGs were enriched in carbohydrate metabolism and 8 DEGs were enriched in glycan biosynthesis (Fig. 4d). These results suggest ZSP administration has robust effects on the transcriptome of db/db hepatic cells.
RT-qPCR was used to further explore DEGs of interest. The expression of HK and GCK were significantly lower in model group compared with those of the normal group, ZSP treatment partially restored expression while upregulating the expression of phosphofructokinase (PFK) in treated mice, consistent with the transcriptomic outcomes (Fig. 5a). As hepatic glycogen content was also raised in ZSP group, IHC was used to examine the expression of GYS, a key regulator of glycogen synthesis [25], revealing increased expression in ZSP group (Fig. 5b). ZSP also significantly reduced the mRNA level of G6pase, but showed no effects on Pepck (Fig. 5c). In the liver, the expression of Gck and G6pase are regulated by PI3K/AKT signaling via promoting the translocation of transcriptional factor FOXO1. PI3K/AKT also affects the activity of GYS trough phosphorylating glycogen synthase kinase 3. Since many of the downstream effectors of PI3K/AKT were influenced by ZSP treatment, the levels of PI3K and AKT levels in mice hepatocytes were examined. ZSP increased all genes encoding the catalytic subunit of class I PI3K and the genes encoding both types of AKT (Fig. 5d). The protein levels of PI3K and AKT, as well as their activated forms were all upregulated in the ZSP treated group (Fig. 5e) indicating the improved hepatic glucose metabolism is a product of PI3K/AKT pathway activation.
Unexpectedly, ZSP treatment did not significantly change the mRNA levels of Insr, Irs1 or Irs2 (Fig. 6a), which are regarded as indicators of insulin activity [26]. Thus, we propose that ZSP may have acted on PI3K through other mechanisms. Comparison of the transcriptomes between db/db mice in the model group and the ZSP group indicate that ZSP had a broad-spectrum promoter effect on many growth factors (GFs) and receptor tyrosine kinases (RTKs) (Fig. 6b). The analysis suggested several significantly altered GFs, including Csf1, Hgf, Vegfd, Pdgfc and Angpt2, and a significantly upregulated RTK gene Flt3. RTKs induce the activation of PI3K, especially the IA subclass [13]. qPCR corroborated ZSP induced increases in the transcriptional expression of Csf1 and Flt3. The average level of Pdgfc and Angpt2 were also higher in the ZSP group, though the difference failed to reach statistical significance (Fig. 6c). This suggests that ZSP may induce the activation of PI3K/AKT via upregulation of GFs and RTKs, particularly Flt3 and Csf1.
In this study, we investigated the effects of ZSP, a traditional Chinese decoction, on a type 2 diabetic mice model. Consistent with previous studies, ZSP treatment was found to significantly attenuate hyperglycemia and obesity in db/db mice, without causing notable toxicity, and with no observed side effects on food or water intake, hepatic or kidney function [8, 9]. Continuing improvement in high-throughput sequencing tests, including the transcriptomics used here, enables robust analysis of disease symptoms on multiple biologically relevant pathways. Through our application of such technology, we were able to identify a range of genes whose expression was influenced by ZSP treatment providing a potential mechanism for ZSP’s effect. ZSP was found to significantly upregulate the PI3K/AKT pathway, it increased the transcription of four genes catalytic subunit of PI3K: pik3ca, pik3cb, pik3cd and pik3cg, as well as increasing the transcription of Akt1 and Akt2, leading to increased protein expression. As PI3K and AKT activation is mediated by phosphorylation of certain amino acids, we examined the phosphorylated PI3K and AKT proteins as well. The results showed that compared to model group, ZSP induced PI3K/AKT signaling in both mRNA and protein levels. However, it is worth noting that the mRNA expression of PI3K and AKT are inconsistent with the protein expression in the normal group, which may be the result of potential different post-transcriptional regulations between the wt/wt and db/db mice [27]. Another explain for this phenomenon is that the insulin level in db/db mice is relatively high, and constant stimulation of insulin may also contribute to the elevated PI3K and AKT expression. However, as this study is aiming to investigate the effects of ZSP in a diabetic model, the comparison between treated and untreated db/db mice is of greater relevance here. In db/db mice, the mRNA and protein expression showed coordinate tendencies. Comparison between the treated and untreated group suggests ZSP significantly induced PI3K/AKT activation. The glucose transported into the liver is metabolized mainly through two ways: glycogen synthesis and glycolysis. After activation, AKT acts on two substrates: glycogen synthase kinase 3 (GSK3) and transcription factor forkhead box O1 (FOXO1). AKT phosphorylates GSK3 and inhibits its activity, thus diminishing the inhibitory effects of GYS and resulting in an increase in glycogen synthesis [28]. AKT also phosphorylates FOXO1 and promotes its exclusion from the cell nucleus, preventing FOXO1 from further influencing gene transcription [29]. This study also revealed that ZSP could raise the rate of hepatic glycogen synthesis possibly by stimulating GYS activity through increased PI3K/AKT signaling. The increase in GYS activity alone is insufficient to raise the glycogen content, as the glycogen content is determined simultaneously by the both synthesis and catabolism. ZSP significantly decreased the transcription of G6pase, which is a downstream factor of FOXO1. Glucose-6-phosphatase (G6Pase) dephosphorylates glucose-6-phosphate and generates glucose, which is a rate-limiting step for both gluconeogenesis and glycogenolysis [30]. Therefore, ZSP may increase the hepatic glycogen content by upregulating GYS and downregulating G6Pase. Aside from being stored as glycogen in the liver, glucose may undergo glycolysis, a step-wise catabolism that provides chemical energy for cell proliferation [31]. ZSP also altered the expression of genes related to glycolysis, specifically, by increasing transcription of Hk2, Hk3, Gck and Pfkb1 in db/db mice. HKs and GCK catalyze the first step of glycolysis, by phosphorylating glucose to glucose 6-phosphate, whereas PFK metabolizes fructose 6-phosphate into fructose 1,6-bisphosphate. In patients with T2DM, GCK activity is repressed and is correlated with increased fasting glucose levels [32]. Since FOXO1 inhibits the transcription of Gck by interfering with corepressor SIN3A [33], and activation of PI3K/AKT suppresses FOXO1, it is possible that the upregulation of GCK by ZSP is mediated by the increased PI3K/AKT signal. We also identified RTKs as the upstream factor of ZSP’s effect on PI3K/AKT. The two subclasses of PI3K: class IA and class IB, are typically activated by different signaling protein families. Class IA are activated by RTKs where class IB are activated by G protein-coupled receptors [34]. There are approximately 60 different RTKs encoded by the human genome [35], responsible for communicating extracellular signals by binding with a range of possible ligands including insulin, GFs, and chemokines [36]. The transcriptomic analysis found that ZSP broadly upregulated several GFs and RTKs, and qPCR confirmed a significant increase on Flt3 expression. It was reported that Fms-like tyrosine kinase 3 (FLT3) mediates PI3K/AKT signaling [37] and treating with FLT3 ligand could significantly prevent the progression to diabetes in diabetic NOD mice [38]. Taken together, these results indicate that ZSP may also activate PI3K/AKT by upregulating GFs and RTKs (Fig. 7). Insulin resistance is a fundamental pathological change in a number of metabolic diseases, including metabolic syndrome [39], polycystic ovary syndrome [40], nonalcoholic fatty liver disease [41], and T2DM [42]. Clinical studies have demonstrated the link between insulin resistance and T2DM, and insulin-sensitizing managements such as dietary control and exercises are proved effective in dealing with metabolic disorders [42]. Insulin-sensitizing is also the paratheatrical mechanism for some T2DM medicines. Metformin enhances insulin sensitivity in the body by promoting insulin receptor activity, increasing glycogen synthesis and promoting the translocation and activity of glucose translator-4 [43, 44]. Glucagon-like peptide-1 receptor (GLP-1R) agonists, such as Liraglutide, Dulaglutide and Exenatide, act on the GLP-1R expressed on pancreatic islets to promote insulin secretion, and suppress macrophage inflammatory response [45, 46]. Dipeptidyl peptidase-4 (DPP-4) degrades GLP-1 and mediates insulin resistance by impairing the activation of AKT [47]. DDP-4 inhibitors, such as gemigliptin, sitagliptin, teneligliptin, and vildagliptin, have now passed clinical approval and are now actively used to treat T2DM [47, 48]. ZSP, and other traditional herbal remedies like it, show exciting potential as complimentary therapy to current treatment strategies for T2DM [49]. In this research we studied the anti-diabetic effects of ZSP on db/db mice, and explored the molecular mechanisms in the liver. This study did not investigate the dose–effect relationship of ZSP which is certainly worth further exploration. Future studies may also focus on the effects of ZSP in other organs, as diabetes has profound effects on the whole body. As ZSP is taken orally, the decoction effects on gut microbiota may also be worth investigating.
In conclusion, ZSP treatment reduced the diabetic symptoms in db/db mice, and the hypoglycemic effect may be due to its activation of PI3K/AKT pathway. By upregulating GFs and RTKs, ZSP enhances the PI3K/AKT signal transduction in the liver, increasing glycogen synthesis and glycolysis, and decreasing gluconeogenesis. Considering its effectiveness and lack of toxicity, ZSP provides an intriguing candidate for further development of new T2DM therapies.
Additional file 1 Table S1. Elution program used in the UPLC system. Table S2. Primers used in qPCR analysis. Table S3. Characteristic chemicals identified from LC-MS/MS analysis. figure S1. ZSP showed no severe toxic effects on mice. a Food intake and water intake of mice (n=7). b Body length of mice (n=7). c Assessment of liver function of mice (n=7). Normal group: wt/wt mice treated with distilled water. Model group: db/db mice treated with distilled water. ZSP group: db/db mice treated with ZSP at 3.3g/kg/day. All data presented as mean ± SD. *p < 0.05, **p < 0.01, ***p <0.001, compared to the model group. #p < 0.05, ##p < 0.01, ### p < 0.001, compared to the normal group. n s. non-significant. |
|
PMC9647946 | Ziyao Wu,Yuxiao Wang,Jiaqi Zeng,Yizhuang Zhou | Constructing metagenome-assembled genomes for almost all components in a real bacterial consortium for binning benchmarking | 10-11-2022 | Binning,Metagenomics,Composition,Abundance,Benchmark | Background So far, a lot of binning approaches have been intensively developed for untangling metagenome-assembled genomes (MAGs) and evaluated by two main strategies. The strategy by comparison to known genomes prevails over the other strategy by using single-copy genes. However, there is still no dataset with all known genomes for a real (not simulated) bacterial consortium yet. Results Here, we continue investigating the real bacterial consortium F1RT enriched and sequenced by us previously, considering the high possibility to unearth all MAGs, due to its low complexity. The improved F1RT metagenome reassembled by metaSPAdes here utilizes about 98.62% of reads, and a series of analyses for the remaining reads suggests that the possibility of containing other low-abundance organisms in F1RT is greatly low, demonstrating that almost all MAGs are successfully assembled. Then, 4 isolates are obtained and individually sequenced. Based on the 4 isolate genomes and the entire metagenome, an elaborate pipeline is then in-house developed to construct all F1RT MAGs. A series of assessments extensively prove the high reliability of the herein reconstruction. Next, our findings further show that this dataset harbors several properties challenging for binning and thus is suitable to compare advanced binning tools available now or benchmark novel binners. Using this dataset, 8 advanced binning algorithms are assessed, giving useful insights for developing novel approaches. In addition, compared with our previous study, two novel MAGs termed FC8 and FC9 are discovered here, and 7 MAGs are solidly unearthed for species without any available genomes. Conclusion To our knowledge, it is the first time to construct a dataset with almost all known MAGs for a not simulated consortium. We hope that this dataset will be used as a routine toolkit to complement mock datasets for evaluating binning methods to further facilitate binning and metagenomic studies in the future. Supplementary Information The online version contains supplementary material available at 10.1186/s12864-022-08967-x. | Constructing metagenome-assembled genomes for almost all components in a real bacterial consortium for binning benchmarking
So far, a lot of binning approaches have been intensively developed for untangling metagenome-assembled genomes (MAGs) and evaluated by two main strategies. The strategy by comparison to known genomes prevails over the other strategy by using single-copy genes. However, there is still no dataset with all known genomes for a real (not simulated) bacterial consortium yet.
Here, we continue investigating the real bacterial consortium F1RT enriched and sequenced by us previously, considering the high possibility to unearth all MAGs, due to its low complexity. The improved F1RT metagenome reassembled by metaSPAdes here utilizes about 98.62% of reads, and a series of analyses for the remaining reads suggests that the possibility of containing other low-abundance organisms in F1RT is greatly low, demonstrating that almost all MAGs are successfully assembled. Then, 4 isolates are obtained and individually sequenced. Based on the 4 isolate genomes and the entire metagenome, an elaborate pipeline is then in-house developed to construct all F1RT MAGs. A series of assessments extensively prove the high reliability of the herein reconstruction. Next, our findings further show that this dataset harbors several properties challenging for binning and thus is suitable to compare advanced binning tools available now or benchmark novel binners. Using this dataset, 8 advanced binning algorithms are assessed, giving useful insights for developing novel approaches. In addition, compared with our previous study, two novel MAGs termed FC8 and FC9 are discovered here, and 7 MAGs are solidly unearthed for species without any available genomes.
To our knowledge, it is the first time to construct a dataset with almost all known MAGs for a not simulated consortium. We hope that this dataset will be used as a routine toolkit to complement mock datasets for evaluating binning methods to further facilitate binning and metagenomic studies in the future.
The online version contains supplementary material available at 10.1186/s12864-022-08967-x.
Thanks to technological advancements, the shift from gene-centric [1, 2] to genome-centric studies [3–7] was possible, and this has facilitated deepening the understanding of the functional capacity, evolution trend, ecological role, and composition structure of the microbial communities. However, assembling individual genomes directly from metagenomes is extremely challenging, possibly due to (but not limited to) the coexistence of intragenomic or intergenomic repeats. To address this problem, binning, the process to aggregate unassembled reads or assemblies into individual or population genomes, has been proposed and exemplified by several documented studies [3–9]. Meanwhile, an increasing number of binning tools have been intensively developed [10]. For assessing binning performance regarding completeness and purity, two types of datasets are generally used. One is a mock dataset with all known genomes, which can be established through in silico simulation of sequencing reads [11–14], in silico combination of randomly-selected reads sequenced from isolate genomes [15], or ab initio sequencing of synthetic consortiums comprising genome-sequenced strains [16, 17]. The other is a dataset sequenced from a real (not simulated) consortium with a negligible fraction or even no known genomes. Correspondingly, there are two main strategies for evaluating binning performance, the former relying on reference to known genomes and the latter with reliance on the conserved or lineage-specific essential single-copy genes (SCGs), such as CheckM [18]. In general, the latter is less than ideal, compared with the former, due to both the uneven distribution of SCGs across a genome and their low number, typically accounting for < 10% of all genes [19]. Accordingly, the former is preferred for mock datasets, while the latter is indispensably used for the dataset of real consortiums. For example, Alneberg et al. assessed binning performance by using the former for mock datasets and the latter for real datasets [20]. If the dataset for a real consortium has all known genomes, the former can be leveraged to achieve a complete and accurate assessment of binning tools. However, to our knowledge, no datasets with all known genomes for all components in not simulated consortiums have been established so far. F1RT, a real (not simulated) bacterial consortium enriched in the laboratory and in-depth described previously by us, is dramatically simple [only 7 components (species) found in the previous study including both high- and low-abundance ones] [21], providing an opportunity to uncover all its metagenome-assemble genomes (MAGs) to construct the first dataset with all known genomes for a real consortium. If so, F1RT can function as a thoroughly independent, and in-depth benchmark of binning tools. In this context, F1RT becomes the focal in this study. Here, we isolated 4 components from F1RT and found that these isolates, fortunately, separated the remaining except for FC8-9 based on sequencing coverage. Therefore, an elaborate in-house pipeline was developed to obtain almost all its MAGs. A series of assessments further confirmed that these MAGs are outstandingly reliable. Subsequently, we found that F1RT is suitable to function as a benchmark, due to several properties suitable for assessing the mainstream binning tools available. Finally, as a test, we used these MAGs to benchmark 8 advanced stand-alone binning tools and found that almost all of them poorly bin a simple metagenome even like F1RT, indicating that there is room to improve binning. We envisage that this dataset will become a useful benchmark in complementing mock datasets for evaluating binning methods and further facilitate binning and metagenomic studies in the future.
The herein research object F1RT had been deeply sequenced by Illumina technology with a size of 12.8 Gb raw metagenomic reads. Such deep sequencing yielded a broad range of sequencing depth from ~ 1015× for FC1 to ~ 16× for FC7 (see Table 1 in [21]). As previous studies showed that metaSPAdes is the best among the well-known assemblers [22–24], metaSPAdes was utilized to de novo reassemble the F1RT metagenome. A total of 31.65 Mb was de novo reassembled into 3,852 scaffolds (≥ 500 bp) by using the assembler metaSPAdes (version v3.15.4; default parameters) [25], accounting for about 98.62% of reads, which was calculated via mapping against all edge connections in the output file “assembly_graph.fastg” of metaSPAdes. Further tracking found that all edges are included in the final 3,852 scaffolds, demonstrating that all 98.62% of mapped reads originate from the components assembled here rather than others unassembled. For the remaining reads, we performed the additional analyses as follows. First, our analysis by using the kmerfreq software [26] showed that approximately 97.11% of the unmapped reads were possibly sequencing errors, due to harboring 15-mers with ≤ 2 frequencies under such deep sequencing, leaving only 23,992 (~ 0.04% of total reads) unmapped reads with relatively high quality. Second, we assembled all unmapped reads to see whether there were other low-abundance organisms and found that only a total of 46,736 bp for 62 scaffolds (≥ 500 bp) was assembled, with a low N50 and maximum scaffold size of 731 bp and 2,015 bp respectively. It is worth pointing out that the complexity is greatly reduced from the total reads to the unmapped reads considering the complex nature of metagenomes with the coexistence of intragenomic or intergenomic repeats, which may generate some additional assemblies, some of which may be artificial due to excluding other reads. However, even under this simplified condition, the total size of assemblies is still greatly small, possibly indicating there are no low-abundance organisms in F1RT. Third, we mapped all the initially unmapped reads against the 62 scaffolds and found that 17,966 read pairs can be concordantly mapped back to their scaffolds. Further checking found that the 62 scaffolds used 8111 of the 23,992 high-quality (> 2 frequencies) reads as well as 9855 of the 806,187 low-quality (≤ 2 frequencies) reads, showing that the high-quality reads were used more significantly than the low-quality reads (P < 2.2e-16, the chi-squared test) for assembly, possibly indicating that the majority of low-quality initially unmapped reads are prone to sequencing errors and thereby cannot be successfully assembled even under this simplified condition. Fourth, although the 62 scaffolds may be with low credibility, we still performed alignment to explore their possible origins and found that 55 of them were successfully mapped to the F1RT metagenome reassembled here (Additional file 1), supporting that these scaffolds may originate from the components assembled here. For the remaining 7 scaffolds, we additionally BLASTed them against the nucleotide sequence database in the National Center for Biotechnology Information (NCBI) and found only 2 scaffolds were credibly (≥ 0.9 alignment fraction, AF) aligned, both of which are from the phylum Firmicutes. Further analysis showed that the mapping ratios of the initially unmapped reads against the references are only about 0.0272% and 0.0059% respectively (Additional file 2: Table S1), indicating that the 2 scaffolds are possibly not from the two references. For the remaining 5 unaligned scaffolds, they total only 3059 bp, possibly indicating there are not from the other low-abundance bacteria, as it is greatly impossible to assemble a such draft genome with a small size under so deep sequencing. Also, these 5 unaligned scaffolds may be greatly impossible from phages, as phages may have multiple copies in one cell and thereby should have high sequencing coverage. Therefore, all these results indicate that the possibility of containing other low-abundance organisms in F1RT is dramatically low, although we cannot completely exclude this possibility. However, all these results at least responsibly demonstrated that almost all F1RT MAGs were successfully assembled. As most of the binners use assemblies instead of reads for binning, directly using the entire F1RT metagenome reassembled here, without considering whether no low-abundance organisms exist, does not affect binning benchmarking below. Besides, all reads instead of just the mapped reads can be directly used to calculate sequencing coverage for all scaffolds via mapping them against the entire F1RT metagenome. However, we also deposited only all mapped reads at https://github.com/Yizhuangzhou/F1RT for users to accelerate mapping for convenience. Using MetaGeneMark (version 3.38) [27], 31,775 genes (≥ 100 bp) had been totally predicted. All high-quality reads and original assemblies reused in this study were downloaded from the GigaScience database (http://gigadb.org/dataset/100049 ). For more details, please refer to our published study [21].
For the application of the binning assessment by reference to known genome assignments, almost all MAGs of F1RT should be constructed. So, we used serial dilution, plating, and repetitive subculturing to isolate its components and then determined colonies as axenic cultures by 16 S rRNA amplification and sequencing. Fortunately, we isolated 4 components in this study. Results from 16 S rRNA sequencing confirmed that these isolates are FC2-3, FC5 and FC7 respectively. Then, we performed genome sequencing at a sequencing depth of > 32 for them (Additional file 2: Table S2), ensuring de novo assembling them into draft genomes with high completeness ( > ~ 99.36%). All basic information about the assembled isolate genomes is tabulated in Additional file 2: Table S2.
The general workflow is presented in Additional file 2: Fig. S1. It is noteworthy that all 4 isolates have > ~ 99.36% of assembly completeness (Additional file 2: Table S2), indicating that we could obtain their MAGs through individual genome alignment against the entire metagenome. By using the NUCmer tool (step 1, Additional file 2: Fig. S1) [28], we obtained alignment for each isolate (Additional files 3, 4, 5 and 6). It was observed that the vast majority of AFs are ≥ 90% (Additional file 2: Fig. S2), suggesting high credibility for their deduced relationships. Thus, scaffolds with AFs of ≥ 90% were preferentially untangled (step 2, Additional file 2: Fig. S1). All scaffolds were found to be individually mapped to one isolate genome and thus unambiguously deconvoluted, comprising the primary MAGs for the 4 isolates. Based on the primary MAGs, their sequencing coverage ranges at the whole-genome level were determined (Fig. 1 A and B). It is worth pointing out that one scaffold (indicated in red) harbors extremely high sequencing coverage (see Additional file 2: Fig. S10D below) possibly due to plasmid sequence (see Additional file 2: Table S5 below) and was thus excluded to determine the sequencing coverage upper bound for FC7. The unaligned scaffolds are nearly pure for uncultured components for the following two reasons. One is > ~ 99.36% of assembly completeness for the 4 isolates (Additional file 2: Table S2). The other is that genome is more readily assembled for isolate than for metagenome, due to the complex nature of metagenome. We analyzed the sequencing coverage for all unaligned scaffolds reassembled by metaSPAdes and unfortunately found that there was no distinct separation (Additional file 2: Fig. S3). However, when using the unaligned scaffolds assembled by SOAPdenovo in the previous study [21], there are 4 distinct peaks (Fig. 1C), except for FC6 and FC8-9 (as indicated by two components below) with a minute fraction of overlappings (range of 20–22). It was shown that the majority of unaligned scaffolds uniquely assembled by metaSPAdes have < 20 or > 760 sequencing coverage (Additional file 7), and only need to extend the coverage range of FC1 to 12,858 without changing the coverage range for other components. By this means, 4 sequencing coverage ranges at the whole-genome level were determined (Fig. 1C), although sequencing coverage variations within FC1 and FC4 seem relatively large (~ 10×). As sequencing coverage follows a Poisson distribution [29], deep sequencing of 12.8 Gb reads for only 9 strains may result in larger intragenomic sequencing coverage variations for high-abundance FC1 and FC4 than for low-abundance FC5-9, demonstrated by the similar ~ 10× of intragenomic sequencing coverage variations for isolates FC2-3. In summary, the sequencing coverage ranges for all components were determined (Fig. 1B). Unaligned scaffolds within only one sequencing coverage range were unambiguously untangled (step 3, Additional file 2: Fig. S1), comprising the primary MAGs for uncultured components. Once MAGs had been primed, models of all components could be trained. Based on the trained primary MAGs of FC6 and FC8-9, 2 scaffolds within the range of 20–22 were separated into FC6 by the Naïve Bayesian Classifier (4-nt motif) [30, 31], as tetranucleotide difference is generally larger between genomes than that within genomes [32]. Next, the scaffolds with < 90% AFs were deconvoluted (step 4, Additional file 2: Fig. S1). Scaffolds within only one sequencing coverage range were explicitly recovered into their MAGs, including 1 for FC4 and 3 for FC6, while the others were further untangled by using the naïve Bayesian classifier (4-nt motif) based on primary MAGs determined above [30]. A scaffold was flagged for a certain isolate if this isolate possessed the highest posterior probability, and meanwhile contained an alignment against this scaffold. Otherwise, scaffolds were determined to be from uncultivated species with a maximal posterior probability. Our further SCG analysis showed that FC8-9 have two components, indicated by 54 SCGs with 2 copies and 2 SCGs with 3 copies (Additional file 8). FC8-9 have overlapping sequencing coverage and thus cannot be separated based on sequencing coverage. However, we, fortunately, found that the scaffolds harboring SCGs with > 1 copies are separated by using the posterior probability cutoff of -1350 determined by using the Naïve Bayesian Classifier with trained FC2 MAG (Additional file 2: Fig. S4). Similar patterns were found by using the trained FC3, FC5, or FC7 MAG (data not shown). Then, the primed SCG scaffolds were used as models to separate all scaffolds of FC8-9 (step 5, Additional file 2: Fig. S1). Taken together, in this way (Additional file 2: Fig. S1), we reconstructed MAGs for almost all components. Statistics in terms of total numbers and total base pairs of untangled scaffolds in steps 1–4 are shown for both isolates (Additional file 2: Fig. S5) and uncultured components (Additional file 2: Fig. S6). Besides, it is worth stressing that two additional MAGs (hereafter termed FC8 and FC9) were strikingly discovered in this study, compared with the original study [21]. This finding explains the high redundancy of the original FC7 [21], as the original FC7 substantially contains three species (herein termed FC7, FC8, and FC9).
The vast majority of metagenomic scaffolds for 4 isolates were reliably untangled by comparison to isolate genomes with high credibility of ≥ 90% AFs (Additional file 2: Fig. S5), indeed requiring no additional validation. However, for comprehensive and solid validation on all uncultured MAGs even under the more complicated condition with isolates than without isolates, all MAGs were chosen to validate together. We first validated reconstruction based on the 107 SCGs [33], which were previously applied to assess binning completeness and purity [34, 35]. A total of 907 SCGs were identified in the entire metagenome and then aggregated into scaffold level to identify 300 SCG-bearing scaffolds. By using these scaffolds, reconstruction was validated from the three following aspects independently. First, validation based on sequencing coverage showed that SCG-bearing scaffolds of the isolates separate all counterparts of uncultured organisms (Fig. 2), supporting that the above deduction of sequencing coverage ranges is reliable (Fig. 1C). Besides, the sequencing coverages for these scaffolds are gradually decreased from FC1 to FC9, indicating the high authenticity of our reconstruction. Second, taxonomic analysis was used based on the logic that SCGs originating from the same genome should be taxonomically concordant [36]. A total of 21,124 selected references (for details, see Methods) were used to search for their closest relatives. The reference with the highest SCG-based average amino acid identity (AAI) (termed SCG AAI) was determined as the closest relative (Additional file 2: Table S3). Then, we performed alignment of SCGs between each MAG and all closest relatives and found that for almost all components except for FC1, the SCG AAI distance between the closest relative and the reference with the second highest SCG AAI is very large (> 10.98) (Additional file 2: Fig. S7), while the SCG AAI distance for FC1 is only 1.12. It was reported that FC1 is a strain possibly from Clostridium sp. FG4 (16 S rRNA identity 99.3%) or Clostridium thermosuccinogenes (16 S rRNA identity 99.1%) [21]. However, genomes for these two strains are unavailable yet. Therefore, both closest relatives for FC1 (Clostridium josui JCM 17,888) and FC2 (Clostridium sp. Bc-iso-3) (Additional file 2: Table S3) were jointly employed to count the number of best hits for FC1. We found that almost all (> 95.18%) SCGs are taxonomically concordant (Fig. 3A), manifested by the fact that almost all best hits are concurrently from one close relative for FC2-9 or the combination of C. josui JCM 17,888 and C. sp. Bc-iso-3 for FC1 (Additional file 2: Fig. S8). Further analysis showed that 100% of FC2 SCGs are concurrently and significantly more similar to SCGs of C. sp. Bc-iso-3 than those for FC1 SCGs (Fig. 3B), implying low or even no contamination for FC1 and FC2. Third, validation based on the single-copy characteristic showed that MAGs were reconstructed to the theoretical limit of contamination, as almost all SCGs with > 1 copy except for PF00162.12 and PF01025.12 for FC8-9 are true duplication rather than contamination (Fig. 3C and Additional file 9). As all components are from the phylum Firmicutes (Additional file 2: Table S3), we further analyzed all 27,565 Firmicutes genomes and showed that almost all SCGs with > 1 copies except for TIGR00442 are veritably frequent with > 1 copies (median 0.52 for the percentage of genomes with > 1 copies) (Additional file 2: Fig. S9A). However, it is worth pointing out that only 4.73% of analyzed Firmicutes genomes are complete (Additional file 2: Fig. S9B) and about half of them harbor < 107 SCGs (Additional file 2: Fig. S9C), meaning that the possibility to be multiple may be underestimated here. Taken together, an assessment based on SCGs strongly indicated that our reconstruction is very reliable.
As conserved essential SCGs are located in a restricted part of a genome, validation based on SCG scaffolds estimates the reconstruction roughly. For elaborate validation, we extended SCG-bearing scaffolds to the whole MAGs. Next, we verified the MAG reconstruction based on sequencing coverage at the whole genome level. To alleviate scaffold size-induced unevenness, the coverage used here was calculated by using a fixed window of 500 bp with 250 bp overlap, while the above analyses (Figs. 1 and 2) were based on the average coverage across a whole scaffold. Our logic was that an idealized MAG should have only one main peak for sequencing coverage and MAG failing this condition probably indicates somehow contamination. Our results showed that both isolates (Additional file 2: Fig. S10) and uncultivated components (Additional file 2: Fig. S11) individually have only one main peak, providing evidence that our reconstruction is very confident.
It has been extensively reported that intragenomic oligonucleotide composition is generally more homogenous than intergenomic one [32, 37, 38]. Consequently, oligonucleotide composition was applied for reconstruction validation. Here, tetranucleotide frequencies were used as oligonucleotide composition considering its great balance between distinguishing ability and computing cost [39]. Besides, we previously found that the TETRA method could represent other statistic methods for tetranucleotide frequency [39] and thus TETRA was employed in this study. Reconstruction reliability is assessed by classification accuracy by TETRA. We found that ~ 86.79% for FC1 to ~ 98.90% for FC8 were correctly classified back into their MAGs with a total high 93.72% of classification accuracy (Fig. 4), proving that a vast fraction of scaffolds was correctly reconstructed. For comparison, we further performed an identical analysis for their references (Additional file 2: Table S4), including two species-level references for FC2 and FC3 (Table 1, see below) and six close relatives based on SCG AAI for FC1 and FC4-9 without species-level references (Additional file 2: Table S3). We found that classification accuracies for MAGs are even larger than those for references for all except for two isolates FC2 and FC6 (Fig. 4), yielding a larger total classification accuracy for MAGs than that for references. These results demonstrated that our reconstruction is pretty authentic. Besides, it has been reported that the oligonucleotide composition of closely related organisms is more similar than that of distantly related organisms [40], namely that a larger portion of fragments may be inherently misclassified between closely related organisms than between distantly related species. Expectedly, detailed tracking showed that a large fraction of fragments is substantially misclassified into closely related organisms. For example, ~ 9.79% of FC1 fragments were incorrectly classified into FC2 (Additional file 2: Fig. S12A), which is most closely related to FC1 (Additional file 2: Fig. S13A). Similar results were obtained for corresponding references (Additional file 2: Fig. S12B and Fig. S13B). These results imply that reconstruction reliability may be underestimated here. In addition, it is worth pointing out that the classification accuracy for FC2 MAG is slightly lower than that for its reference, while although the classification accuracy for FC6 MAG is relatively lower than that for its reference, its accuracy is still high (~ 93.36%) (Fig. 4).
Genome completeness, which was roughly assessed as the number of SCGs divided by 107 owing to about half of Firmicutes having 107 SCGs (Additional file 2: Fig. S9C), ranges from 60.75 to 100% (Fig. 5A), supplying an opportunity to assess binning performance for samples with a broad range of assembly completeness. It is worth stressing that a binned genome with low completeness may result from relatively low abundance rather than bad binning, such as FC8-9 here. In this context, for binning evaluation, the strategy by comparison to known genomes outperforms the strategy with reliance on SCGs. Besides, the relative abundances vary substantially, including ~ 45.96% for dominant taxa FC1 and even < 1% for rare components FC7-9 (Fig. 5B), providing a possibility to benchmark binners for both high- and low-abundance organisms. Collectively, three main discriminative characteristics underlie the binning tools, including reference (known) genomes, sequencing coverage (abundance), and sequence composition. The relatively outdated binners leverage one of them alone [32, 41–44], while the state-of-the-art binners at present leverage two or more characteristics together [12, 20, 45–52]. Here, we independently investigated them to see whether this dataset can be well binned by using one of them alone. For known genomes, our results showed that only FC2 and FC3 possess species-level references (Table 1), according to the ~ 95% average nucleotide identity (ANI) threshold for species delineation [53]. As a result, it is impossible to deconvolute the whole metagenome by merely using reference genomes, in line with the condition for most metagenomic studies with a paucity of reference genomes. For sequencing coverage, we found that it was impossible to separate any MAGs (Fig. 5C), if not using any other information such as isolates in this study. In addition, we observed that there is a small portion of scaffolds with abnormally high sequencing coverage (Fig. 5C), greatly challenging for binners (For reasons, see Discussion below). In contrast, datasets generated through in silico simulating sequencing reads possibly have no such characteristics. For sequence composition, binning accuracy is low for short scaffolds (< 10 kb) [32, 54]. It is worth pointing out that short scaffolds are differently defined from other studies. For example, CONCOCT and MetaBAT2 defined < 1-2.5 kb sequences as short scaffolds [20, 47], as < 1 kb sequences are unable to accurately capture species-specific composition and coverage patterns. This difference is possibly because these binners use sequencing coverage together with composition to improve binning. We discovered that the vast majority of scaffolds are < 10 kb for FC8-9 (Fig. 5D), posing a substantial challenge to binners solely based on composition. Furthermore, we found that still a fraction of long scaffolds (≥ 10 kb) are misclassified (Fig. 5E), even under the condition using the whole above-reconstructed MAGs (for details, see Methods), which yield higher classification accuracy than under the condition by using individual scaffolds during the binning period when their whole MAGs are indeed unknown [54]. Besides, it was reported that the naïve Bayesian classifier is suitable to classify short scaffolds with even < 1000 bp [30], although it has not been successfully integrated into binning methods yet. Here, we explored whether short scaffolds can be correctly classified with the naïve Bayesian classifier based on the trained whole MAGs (for details, see Methods) and found that the binning accuracies for short scaffolds are only ~ 25.20-92.98% (Fig. 5F), also showing challenging for binning. Together, these findings indicated that F1RT albeit simple is amazingly challenging for binners based on one characteristic alone, requiring state-of-the-art binners which leverage several characteristics [12, 20, 45–52]. Thus, the dataset constructed here is a pretty good benchmark to compare all binners including mainstream binners available now or assess new binners leveraging several characteristics.
Ensemble binning tools can be separated into two classes: (1) the stand-alone binners such as MaxBin [12] and CONCOCT [20]; (2) the binners refining results of other binners, such as GraphBin [55], GraphBin2 [56], METAMVGL [57] and MetaWRAP [58]. Here, we used this constructed dataset to assess the 8 aforementioned first-class stand-alone binners (for details, see Methods). MetaBAT2 achieved the highest overall precision (99.63%), followed by MaxBin and MaxBin2 with comparably good precision (Fig. 6A). Nonetheless, MetaBAT2 showed relatively low sensitivity of 79.55% (Fig. 6B), while MaxBin and MaxBin2 produced a higher sensitivity of 96.51% and 92.86% respectively. Accordingly, MaxBin, followed by MaxBin2 and MyCC, generated the best aggregate binning performance, as indicated by F-scores (Fig. 6C). In addition, MaxBin, MaxBin2 and CONCOCT recovered almost all (> 99%) scaffolds (≥ 500 bp) (Fig. 6D), while MetaBAT2 recovered a relatively lower fraction (92.88%), implying that MetaBAT2 may only bin scaffolds with high accuracies to generate the highest precision (Fig. 6A), as MetaBAT2 does not bin < 2.5 kb scaffolds [47, 59]. It is worth pointing out that CONCOCT yielded the lowest precision to achieve the minimal F-score. However, it may be a fair comparison for CONCOCT if multiple (usually > 50) samples are used [20]. It is also worth noting that the performance of MetaBAT2 was severely impaired without multiple samples [47]. Also, BinSanity, SolidBin, and COCACOLA met the obstacle to yield an F-score of > 90%, although they were developed later than MaxBin. Besides, we found that only MaxBin and MyCC recovered the correct number of 9 bins (Fig. 7), while SolidBin only correctly yielded 7 bins with one additional incorrect bin. Other methods split MAGs into multiple bins to yield large bin numbers. Together, MaxBin performed best on the F1RT dataset constructed here.
Our results showed that some scaffolds harbor abnormally higher sequencing coverage than other scaffolds in their MAGs (Figs. 1 and 5C, and Additional file 2: Fig. S10 and S11). For explanation, we performed the following additional analysis and obtained some clues. First, we found no significant difference in mapping qualities for scaffolds with abnormally high sequencing coverage compared to scaffolds with normal sequencing coverage (Additional file 2: Fig. S14), indicating that mapping quality is not the main reason for F1RT. The decreased mapping quality may result from randomly low-quality sequencing, which may be nonselective for any scaffolds in the same MAG, explaining the above observation. As duplicated reads generated by PCR amplification before sequencing or during the period of next-generation sequencing like Illumina sequencing used here may generate high sequencing coverage, we next explored this possibility and expectedly found that some scaffolds with high sequencing coverage indeed have relatively higher duplication than normal ones (Additional file 2: Fig. S15), possibly accounting for their higher sequencing coverage. As plasmids may have multiple copies in cells to generate higher sequencing coverage than chromosome sequences, we then performed alignments between plasmid sequences and scaffolds with high sequencing coverage and found that some are possibly from plasmids (Additional file 2: Table S5). Finally, our results showed that some scaffolds with high sequencing coverage are shorter than normal ones (Additional file 2: Fig. S16), indicating that they may localize at intragenomic or intergenomic regions, as repeats are difficult to be assembled. However, the exact reasons underlying abnormally high sequencing coverage for some scaffolds remain unknown yet, and whether existing other reasons awaits further research.
Despite great progress in culture technologies [60], uncultivability for the overwhelming majority of bacteria remains a major challenge yet. However, some component species could be isolated elaborately in pure culture. Sequencing the obtained monospecific colony will provide a fraction of isolate genomes, which in return efficiently reduces the complexity of the original target metagenome. Here, we demonstrate an effective strategy, namely that subtracting isolate genomes from the entire genome could untangle genomes for uncultured organisms. However, although pure cultures maybe not be easily obtained in some conditions, simplified mixed cultures may be roughly obtained. Miniature microflora will provide mini-metagenome, which in return similarly reduces the complexity of the original target metagenome to facilitate binning [61]. Therefore, it is evident that a combination of culturing regardless of pure culture or simplified mixed culture and classical metagenomics will increase the quality and quantity of genomes recovered.
Combining multiple samples for binning increases the dimensionality of sequencing coverage to improve binning [35, 62]. However, it also possesses several limitations. First, components, which are unique to one sample [63], will not necessitate multiple-sample sequencing. Second, it incurs high sequencing costs. Third, it requires elevated computational resources to co-assemble reads from multiple samples. Forth, the co-assembly required for multi-sample algorithms diminishes the strain-level microdiversity [35], which may be indispensable for functional investigation [64]. Therefore, in these circumstances, single metagenomes necessitate maximizing usable information. As F1RT is a single metagenome, it can serve as a particularly helpful toolkit for the realistic assessment of performance on single-sample metagenomes for binning methods and exploring whether available methods make full use of leveraging information without requiring additional samples. Besides, we should point out that if a tool works well on F1RT, it could potentially work well on complex metagenomes (because this cannot be guaranteed every time, especially if the metagenome has closely related strains with similar abundance). However, a tool, which yields low performance on F1RT, is surely not good enough and requires further improvement.
Almost all mainstream binners available now use an ensemble of sequence composition and sequencing coverage (including both single and multiple samples) for binning. MaxBin2, albeit using the multiple-sample abundances, performed slightly less accuracy than MaxBin (Figs. 6C and 7), implying that MaxBin2 makes poor use of single sequencing coverage. Therefore, we reason those tools using multiple-sample abundances should primarily make full use of single-sample abundance. Besides, some tools developed recently apply some additional information. For example, COCACOLA uses pair-end read linkage [50]; SolidBin applies co-alignment information including must-link and cannot-link constraints [51]. However, our findings showed that SolidBin and COCACOLA obtained lower F-scores than MaxBin (Fig. 6C) and yielded incorrect bin numbers (Fig. 7), indicating that they make poor use of sequence composition and sequencing coverage or poorly combine additional information with the basis of sequence composition and sequencing coverage. Therefore, novel tools should first make full use of sequence composition and sequencing coverage as a base and then integrate other information such as co-alignment information, pair-end read linkage, SCGs, assembly graphs [65], and DNA methylation [66] to further improve binning, or assembly graphs [55, 56] and a combination of assembly and paired-end graphs [57] to further refine binning of initial binners such as MaxBin2 [56, 57].
When developing a novel algorithm, reiterative assessment regarding the strengths and limitations and accordingly improvement/optimization is required. Hence, a simple dataset with < 10 species is greatly beneficial and convenient, as it can reduce the runtime for each test. Thus, F1RT is an ideal investigative model, due to its embracing only 9 components. In addition, the dataset constructed here is the first dataset with all known genomes for a real consortium. Compared with simulated datasets, this dataset provides some distinct characteristics challenging for binning. For example, a minute fraction of scaffolds with abnormally high sequencing coverage and low completeness for some components like FC8-9 here were found in F1RT. Therefore, a dataset like F1RT is indispensable. Lots of advanced algorithms incorporate SCG information to improve binning, albeit with a slight usage difference. For example, COCACOLA [50], MaxBin [12], and SolidBin [51] use SCGs to estimate the initialized bin number; MyCC [48] and MetaWatt [41] use SCGs to determine which clusters should be merged or split after a round of clustering; Autometa uses SCGs to guide clustering [67]; MetaBinner uses SCGs for k-means initialization [68]. In this scenario, using SCGs for binning assessment may be unreasonable and unfair. As an alternative, F1RT could be used to evaluate these methods owing to known genomes for all components. Accordingly, F1RT can complement simulated datasets like CAMI datasets [13, 14] to complete a full assessment for all binning tools. According to the above discussion, F1RT can be used as a benchmark for binning tools like MaxBin based on a single metagenome. As tools based on multiple metagenomes have evolved from the tools based on single metagenome [35] and single sequencing coverage is the base for developing multiple-metagenome-based tools, F1RT can be used as a benchmark to assess whether they make full use of single metagenome. In all, as a thoroughly independent and in-depth benchmark, we envisage that this dataset will become a standard dataset for comparison and improvement of binning methods in the future.
Here, through isolating 4 components and then using the in-house developed pipeline to construct almost all genomes with 4 isolates and the whole metagenome, we present the first dataset with known genomes for almost all components in a real bacterial consortium F1RT. Besides, compared with the original study, two novel components termed FC8 and FC9 in this study are discovered and 7 reliable genomes for species without available genomes are obtained. This dataset has several features suitable for binning benchmarking and has wide applications including (i) benchmarking each round of improvement when developing novel tools owing to its low complexity; (ii) comparing state-of-the-art tools available as an independent benchmark; (iii) as a base for exploring whether tools make full use of single-sample sequencing coverage to develop robust tools and (iv) assessing tools with the integration of SCGs for binning. Besides, this study demonstrates an effective strategy via combining culturing and classical metagenomics for uncovering uncultured genomes. Finally, this study provides useful insights for developing more promising tools. In all, this dataset will become a standard dataset for binning assessment on a real consortium and facilitate metagenomics in the future.
Pure cultures were obtained from single colonies by iterative subcultivation on the plates and identified by analyzing 16 S rRNA gene sequences. FC2 was putatively determined to be Clostridium straminisolvens (16 S rRNA identity 99.6%) according to the 16 S rRNA identity threshold of 98.65% [69]. Thus, FC2 was isolated and cultured at 50 ℃ in DMSZ medium 122 in anaerobic conditions according to the previous study [70]. Besides, FC3 was isolated and cultivated in DMSZ medium 1 for DSM 6347 at 37 ℃ in aerobic conditions; FC5 and FC7 were isolated and cultivated in DMSZ medium 122 and DMSZ medium 1 for DSM 6347 respectively at 37 ℃ in anaerobic conditions.
DNA was extracted based on the CTAB method. Illumina DNA PCR-free libraries with an insert size of ~ 500 bp were constructed for all 4 isolates according to the manufacturer’s instructions. Besides, a library with an insert size of ~ 2000 bp was additionally prepared for FC2. DNA manipulations, including the preparation of single-molecule arrays, cluster growth, and paired-end sequencing, were performed using an Illumina HiSeq 2000 sequencer according to standard protocols. The Illumina base-calling pipeline (version HCS1.4/RTA1.12) was used to process the raw fluorescent images and call sequences. Raw reads of low quality (those with three consecutive bases with quality ≤ Q20) were discarded before assembly. High-quality reads were assembled using the genome assembler SOAPdenovo2 with default parameters [71].
The NUCmer tool (version 3.23) [72] was utilized to perform genomic alignment against the whole metagenome for each isolate genome (-maxmatch). Then, the utility delta-filter was employed to filter NUCmer output (-q –r –l 200), and the show-coords utility was used to generate the aligned coordinates (-c –l –r –T). Finally, a custom Perl script in-house was used to calculate AFs.
High-quality reads retrieved from the original report (http://gigadb.org/dataset/100049 ) were mapped back onto the F1RT metagenomic scaffolds to determine coverage with SOAPaligner (v. 2.21) [73] by allowing at most two mismatches in the first 35-bp seed region and 90% of identity over the whole read. Subsequently, base coverage was calculated by SOAPcoverage (v. 2.7.7). It was shown that the sequencing coverage at both ends of each scaffold is slightly lower than in other regions. To obviate the effect of scaffold ends, both ends of 100 bp were discarded. Then the average coverage across the whole scaffold was computed to represent the sequencing coverage of this scaffold. For drawing Fig. S10 and Fig. 11 in Additional file 2, the sequencing coverage was calculated from 5’ to 3’ end with a window of 500 bp and sliding at 250 bp.
For scaffolds with multiple possible origins during the MAG reconstruction period (Additional file 2: Fig. S1), the naïve Bayesian classifier (4-nt motif) [30] was applied. Briefly, the primed MAGs were trained. Then, each scaffold was classified into MAG with the highest posterior probability.
A set of 107 Hidden Markov Models for SCGs [33] deposited in either TIGRFAMs [74] or Pfam libraries [75] were searched against the protein sequences of F1RT using HMMER3 with the default settings, except the trusted cutoff was used (-cut_tc). When one gene had multiple SCG annotations, only the one with a minimum e-value was assigned.
All 83,075 prokaryotic genomes were downloaded from the NCBI database. To filter out low-coverage genomes, 37 draft genomes with a summed length < 0.5 megabase pairs were discarded. Then based on the List of Prokaryotic names with Standing in Nomenclature database, 68,261 genomes were determined for 5680 named species. The remaining 15,444 genomes were directly as reference genomes. For each named species, one type strain (if available) or the largest genome was selected as the reference. In total, 21,124 genomes were used as references for SCG taxonomic analysis.
SCGs were identified for the resulting 21,214 references. SCGs in MAGs were then BLASTPed against the SCG database with a maximum e-value cutoff of 1e-5 and the resulting identities were transferred into protein-length-weighted sequence identities. Then, SCG AAI was calculated for each MAG. To ensure high confidence, > 95% SCGs should be mapped between the compared genomes. The close relatives with the highest SCG AAIs were thus found (Additional file 2: Table S3). Best hits from close relatives were counted to validate reconstruction.
Our results showed that all components are from the phylum Firmicutes (Additional file 2: Table S3). Thus, all 27,565 Firmicutes genomes downloaded from NCBI were subject to SCG identification. Then, the number of genomes with > 1 copy of each SCG was independently counted. Also, the assembly statuses and the number of SCGs across a genome were counted. All were tabulated in Additional file 2: Fig. S9.
For each MAG, all its scaffolds were sorted decreasingly according to their sequencing coverage and then concatenated directly. Subsequently, all MAGs were shredded into consecutive 10-kb fragments with 5-kb overlap. For each fragment, 9 TETRA values against all MAGs were calculated according to the previous studies [32, 39, 54] and the MAG with the highest TETRA value was assigned for this fragment. The percentage of fragments correctly classified back into their own MAGs was counted to indicate the reconstruction performance. Also, an identical analysis was performed for their references listed in Additional file 2: Table S4.
The FRAGTE method [54] was used to sieve closely related genome pairs from 9 × 21,124 pairs for species delineation. Then we calculated ANIs for all sieved pairs by using the NUCmer tool (version 3.23) according to the previous study [53]. MAG with an ANI of > ~ 95% was delineated at the species level, according to the ANI cutoff determined previously [53].
Short (< 10 kb) scaffolds were classified by the naïve Bayesian classifier (4-nt motif) [30]. Briefly, the whole MAGs were trained. Then, each short scaffold was classified into MAG with the highest posterior probability. The percentage of scaffolds classified back into their own MAGs was counted to indicate the classification accuracy.
Hybrid methods, which at least jointly leverage both sequence composition and sequencing coverage, are the most advanced. F1RT has only 3,852 scaffolds, while Vamb requires > 50,000 sequences for binning [76]. Therefore, Vamb was not benchmarked in this study. Also considering that F1RT is a single metagenome, algorithms such as GroopM [49] and MetaBMF [77] requiring multiple samples were excluded. This rationale led us to focus on the 8 first-class stand-alone binning tools without human-augmented refining, including MaxBin (v.1.4.5) [12], MaxBin2 (v. 2.2.7) [46], MetaBAT2 (v. 2.12.1) [47], CONCOCT (v.0.4.0) [20], MyCC (v.1.0) [48], Binsanity (v.0.5.4) [52], COCACOLA (v. 1.0) [50] and SolidBin (v.1.2) [51]. Here, Binsanity used the mode of Binsanity-lc comprising of binsanity and binsanity-refine, and SolidBin used the mode of SolidBin-naive. All used the default parameters except for the minimal length of 500 if allowed.
Assume there are N genomes in the dataset, which were assigned into M bins, and Sij indicates the total sequences (in terms of base pairs) belonging to genome j appear in cluster i. If one bin has ≥ 2 assigned species, only the species with the largest number of sequences are kept for this bin and denoted as ; if one genome is assigned into ≥ 2 bins, only the bin with the largest number of sequences is kept for this genome and denoted as . Furthermore, if ≥ 2 bins are assigned to one common species, the bin with the largest number of sequences for this species is denoted as . Then only bins satisfying are considered correct bins, while others are considered incorrect bins. The overall precision and sensitivity, as in the previous studies [12, 48], were calculated as follows: F-score, which indicates the overall binning performance via weighting both overall precision and sensitivity, was calculated as follows:
Additional file 1. Coordinates for nucleotide-based alignments between scaffolds assembled by unmapped reads and F1RT metagenome scaffolds (F1RT) reassembled by metaSPAdes.Additional file 2.Figure S1. A schematic flow chart for untangling MAGs for almost all F1RT components. Steps including genome-wide alignment by NUCmer (step 1), determination of primary MAGs for isolates (step 2) and for uncultivated components (step 3), assignment of contigs with <90% AF to obtain final MAGs for all components except for FC8-9 (step 4) and separation of FC8-9 by using the Naïve Bayesian Classifier (4-nt motif) are shown. MAGs, metagenome-assembled genomes; AF, alignment fraction; PP, posteriori probability. Figure S2. Alignment fraction distribution of aligned scaffolds for each isolate. Dashed line, the alignment fraction threshold of 90% for scaffold assignment with high confidence. Figure S3. The distribution of scaffold-level sequencing coverage for all unaligned scaffolds assembled by metaSPAdes. Figure S4. The distribution of posteriori probabilities for FC8-9 SCG scaffolds based on FC2 MAG. Only the SCG scaffolds harboring SCGs with >1 copiesare shown here. Posteriori probability was calculated by using the Naïve Bayesian Classifier (4-nt motif). Figure S5. Reconstruction statistics for isolates. A and B, for FC2 in terms of scaffold number and base pair (bp) respectively; C and D, for FC3 in terms of scaffold number and bp respectively; E and F, for FC5 in terms of scaffold number and bp respectively; G and H, for FC7 in terms of scaffold number and bp respectively. Figure S6. Reconstruction statistics for uncultured components. A and B, for FC1 in terms of scaffold number and base pair (bp) respectively; C and D, for FC4 in terms of scaffold number and bp respectively; E and F, for FC6 interms of scaffold number and bp respectively; G and H, for FC8 in terms of scaffold number and bp respectively; I and J, for FC9 in terms of scaffold number and bp respectively. Figure S7. The average amino-acid identity distance between the closest relative and the reference with the second highest SCG AAI for each component. The solid line indicates the distance. For the closest relatives, please refer to Additional file 2: Table S3. Figure S8. The heatmap showing the amino-acid identities for SCGs of all components. The reference in red, the closest relative for all components or an additional second closest relative for FC1. For the closest relatives, please refer to Additional file 2: Table S3. Figure S9. Statistics of all Firmicutes SCGs. A, percentage of genomes with >1 copies; B, assembly status for all Firmicutes genomes; C, number distribution of SCGs for all Firmicutes genomes. The data were from all 27,565 Firmicutes genomes. Red, SCG with >1 copies in at least one F1RT MAG; dashed line, the median for percentage of genomes with >1 copies at 0.52. Figure S10. The sequencing coverage distribution for isolates. A, for FC2; B, for FC3; C, for FC5; D, for FC7. Sequencing coverage is calculated using a fixed window of 500 bp with 250 bp overlap. Figure S11. The sequencing coverage distribution for uncultivated components. A, for FC1; B, for FC4; C, for FC6; D, for FC8; E for FC9. Sequencing coverage is calculated using a fixed window of 500 bp with 250 bp overlap. Figure 12. Classification statistics of 10-kb fragmentsfor all F1RT MAGs or their reference genomes. A, for F1RT MAGs; B for F1RT reference genomes. The number in a cell is the fraction (%) of fragments classified into their corresponding organism and used as a basis for color intensity. The 10-kb fragments are produced via dividing (pre-concatenated) MAGs or reference genomes. For references, please refer to Additional file 2: Table S4. Figure 13. Phylogenetic relationships for F1RT MAGs and their reference genomes. A, for F1RT MAGs; B, for reference genomes. Phylogenetic relationships were determined on the basis of the average amino acid identity for SCGs. For references, please refer to Additional file 2: Table S4. Figure 14. Mapping qualities for scaffolds with abnormally high or normal sequencing coverage. A, for FC1; B, for FC2; C, for FC3; D, for FC4; E, for FC5; F, for FC6; G, for FC7; H, for FC8; I, for FC9. MAPQ, mapping quality. Figure 15. Statistics of duplicated reads mapped to scaffolds with abnormally high or normal sequencing coverage. A, for FC1; B, for FC2; C, for FC3; D, for FC4; E, for FC5; F, for FC6; G, for FC7; H, for FC8; I, for FC9. Duplication is calculated as the total alignments divided by the alignments after removing duplication by using “samtools markdup -r”. Figure 16. Size statistics of scaffolds with abnormally high or normal sequencing coverage. A, for FC1; B, for FC2; C, for FC3; D, for FC4; E, for FC5; F, for FC6; G, for FC7; H, for FC8; I, for FC9. Table S1. Mapping summary and the read mapping ratios for the references of the two scaffolds. Table S2. The genomic statistics of F1RT isolates. Table S3. The close relatives of all F1RT components. Table S4. Reference genomes of all F1RT components for TETRA analysis. Table S5. Alignments between scaffolds with abnormally high sequencing coverage and plasmid sequences.Additional file 3.Coordinates for nucleotide-based alignments between scaffolds from the FC2 isolate genome and scaffolds from the F1RT metagenome assembled by metaSPAdes.Additional file 4.Coordinates for nucleotide-based alignments between scaffolds from the FC3 isolate genome and scaffolds from the F1RT metagenome assembled by metaSPAdes.Additional file 5.Coordinates for nucleotide-based alignments between scaffolds from the FC5 isolate genome and scaffolds from the F1RT metagenome assembled by metaSPAdes.Additional file 6.Coordinates for nucleotide-based alignments between scaffolds from the FC7 isolate genome and scaffolds from the F1RT metagenome assembled by metaSPAdes.Additional file 7.Sequencing coverage for scaffolds uniquely assembled by metaSPAdes.Additional file 8.SCG scaffolds and their sequencing coverage for FC8-9.Additional file 9.SCG scaffolds, their origins and sequencing coverage. Number in brackets, sequencing coverage across a whole scaffold. |
|
PMC9647949 | Jessy A. Slota,Babu V. Sajesh,Kathy F. Frost,Sarah J. Medina,Stephanie A. Booth | Dysregulation of neuroprotective astrocytes, a spectrum of microglial activation states, and altered hippocampal neurogenesis are revealed by single-cell RNA sequencing in prion disease | 09-11-2022 | Prion disease,Neurodegeneration,Pathogenesis,Single-cell RNAseq,Systems biology,Microglia,Astrocytes,Neurons | Prion diseases are neurodegenerative disorders with long asymptomatic incubation periods, followed by a rapid progression of cognitive and functional decline culminating in death. The complexity of intercellular interactions in the brain is challenging to unravel and the basis of disease pathobiology remains poorly understood. In this study, we employed single cell RNA sequencing (scRNAseq) to produce an atlas of 147,536 single cell transcriptomes from cortex and hippocampus of mice infected with prions and showing clinical signs. We identified transcriptionally distinct populations and sub-populations of all the major brain cell-types. Disease-related transcription was highly specific to not only overarching cell-types, but also to sub-populations of glia and neurons. Most striking was an apparent decrease in relative frequency of astrocytes expressing genes that are required for brain homeostasis such as lipid synthesis, glutamate clearance, synaptic modulation and regulation of blood flow. Additionally, we described a spectrum of microglial activation states that suggest delineation of phagocytic and neuroinflammatory functions in different cell subsets. Differential responses of immature and mature neuron populations were also observed, alongside abnormal hippocampal neurogenesis. Our scRNAseq library provides a new layer of knowledge on single cell gene expression in prion disease, and is a basis for a more detailed understanding of cellular interplay that leads to neurodegeneration. Supplementary Information The online version contains supplementary material available at 10.1186/s40478-022-01450-4. | Dysregulation of neuroprotective astrocytes, a spectrum of microglial activation states, and altered hippocampal neurogenesis are revealed by single-cell RNA sequencing in prion disease
Prion diseases are neurodegenerative disorders with long asymptomatic incubation periods, followed by a rapid progression of cognitive and functional decline culminating in death. The complexity of intercellular interactions in the brain is challenging to unravel and the basis of disease pathobiology remains poorly understood. In this study, we employed single cell RNA sequencing (scRNAseq) to produce an atlas of 147,536 single cell transcriptomes from cortex and hippocampus of mice infected with prions and showing clinical signs. We identified transcriptionally distinct populations and sub-populations of all the major brain cell-types. Disease-related transcription was highly specific to not only overarching cell-types, but also to sub-populations of glia and neurons. Most striking was an apparent decrease in relative frequency of astrocytes expressing genes that are required for brain homeostasis such as lipid synthesis, glutamate clearance, synaptic modulation and regulation of blood flow. Additionally, we described a spectrum of microglial activation states that suggest delineation of phagocytic and neuroinflammatory functions in different cell subsets. Differential responses of immature and mature neuron populations were also observed, alongside abnormal hippocampal neurogenesis. Our scRNAseq library provides a new layer of knowledge on single cell gene expression in prion disease, and is a basis for a more detailed understanding of cellular interplay that leads to neurodegeneration.
The online version contains supplementary material available at 10.1186/s40478-022-01450-4.
Prion diseases are a rare group of fatal and infectious neurodegenerative disorders that afflict humans and animals. According to the ‘protein only hypothesis’, prion disease is caused by structural transformation of cellular prion proteins (PrPC) into a misfolded conformation (PrPSc) that can be transmitted between individuals [67]. Prions replicate by recruiting and converting PrPC into the disease-associated conformation, adding to growing amyloid fibrils that accumulate and spread throughout the brain [50, 76]. Prion accumulation is associated with brain pathogenesis that includes reactive micro- and astro-gliosis, neuronal vacuolation and synaptic dysfunction, and eventual neuronal loss, culminating in rapidly progressive neurocognitive decline [74]. While neuronal dysfunction and demise presumably cause the clinical signs and symptoms of disease, the links between prion accumulation and pathogenesis remain mysterious. A few possible explanations for neuronal susceptibility to prion infection are: (1) direct toxicity from PrPSc, (2) loss of functional PrPC, (3) inflammation/oxidative damage from reactive astrocytes and microglia, and (4) loss of homeostatic brain cells that normally protect neurons. The complexity of prion pathogenesis makes it difficult to discern the contribution of these different possibilities towards prion neurotoxicity in vivo. A deep understanding of molecular changes within the prion-infected brain may help identify disease-associated markers and predict the interplay between diverse brain cell subtypes that differentially respond during disease. Accordingly, transcriptional profiling of bulk brain tissues in mouse models of prion disease is a popular approach that has been widely used to describe prion pathogenesis [12, 30, 39, 48, 49, 73]. Our group has previously used microdissection of the hippocampal CA1 region, and other brain regions, combined with transcriptional profiling to associate pathological gene expression changes with precise brain regions [46, 47, 72]. Studies by Scheckel et al. and Kaczmarczyk et al. used translating ribosome affinity combined with RNAseq to profile brain-cell type specific changes in proteins actively being translated from mRNAs during prion disease [38, 70]. These studies readily identify onset of inflammatory gene expression attributed to reactive microglia and astrocytes, even at early pre-clinical stages of disease. Contrary to this, transcriptional changes associated with neuronal synaptic dysfunction are less obvious and not detected until the final clinical stages of disease. A limitation of previous transcriptional profiling approaches to prion infection is the inability to distinguish differential responses of brain cell subtypes, hampering identification of cell type specific markers associated with disease. Here, we employed single-cell RNA sequencing (scRNAseq) of live brain cells isolated from the cortex and hippocampus of mice at the clinical stages of disease following inoculation with Rocky Mountain Laboratory strain of mouse-adapted scrapie (RML). While many neurodegenerative studies employ single-nucleus RNA sequencing [86], live single cell RNA sequencing has the advantages of wider gene coverage, unbiased transcript profiling, improved power for discriminating cell types [4], and better detection of transcriptional changes within disease-associated microglia [82]. We chose the cortex and hippocampus because they are amongst the most sensitive regions to disease-associated histological changes (neuronal vacuolation, PrPSc immunoreactivity and reactive gliosis) in this model of RML scrapie [55]. Sequencing libraries were prepared from 4 mock-infected mice collected at 110–180 days post infection (dpi) and 7 RML infected mice that reached clinical endpoint at various timepoints from 152–173 dpi. Datasets were integrated to produce a “single-cell atlas” of cortical and hippocampal brain cells, primarily consisting of microglia, astrocytes, vascular cells, oligodendrocyte progenitor cells and neurons. We identified differentially expressed transcripts within individual cell clusters, and we found many cell clusters to differ in abundance between RML and mock infected mice. In association with RML disease, we noted global decreases in relative frequencies of homeostatic astrocytes and vascular cells, and increases in oligodendrocyte progenitor cells. Additionally, we distinguished between homeostatic and disease-associated microglial (DAM) populations and identified a spectrum of microglial activation states that were associated with disease. We also examined differentially affected populations of immature and mature neurons during disease and observed evidence of abnormal neurogenesis in RML infected mice, particularly in the hippocampus.
The Animal Care Committee of the Canadian Science Centre for Human and Animal Health approved procedures involving live animals under animal use document # H-20-024. CD1 mice were intraperitoneally inoculated with 200 µL of either RML or non-infectious 2% brain homogenate. The mice were monitored for onset of clinical signs that include dull ruffled coat, pinched abdomen and weight loss of up to 20%. Mice were sacrificed by isoflurane anesthesia followed by transcardial perfusion with PBS. Brains were immediately removed and immersed in ice-cold loading buffer (EBSS supplemented with 0.04% BSA, 0.6% glucose and 1 mM Kynurenic acid).
The cortex and hippocampus from both hemispheres of freshly collected brains was dissected on ice and transferred into 6-well culture dishes with dissociation solution (Hibernate e minus calcium with 20 units/mL papain and 0.005% DNase I). Tissue was crudely minced with a scalpel and dissociated for 40 min at 37 °C by swirling the culture dishes every 3–4 min. A single cell suspension was made by gently triturating the tissue with a serum-coated p1000 pipette tip 10 times. At this point, cell suspensions were kept on ice and all centrifugation steps were performed at 4 °C. Large debris was allowed to settle for 2 min before transferring supernatant to a new 15 mL falcon tube. Debris was further removed using Debris removal solution (Miltenyi Biotech) according to manufacturer’s instructions. Briefly, the cell pellet was re-suspended in 1550 µL of loading buffer, mixed with 450 µL of debris removal solution, overlayed with 2 mL of loading buffer, centrifuged at 3000×g for 10 min and the top 2 layers were removed. The remaining cells were washed once in 5 mL of loading buffer before removing myelin using Myelin Removal Beads II (Miltenyi Biotech). Briefly, the cell pellet was re-suspended in 270 µL of loading buffer, mixed with 30 µL Myelin Removal Beads II, incubated for 15 min on ice, washed in 2700 µL of loading buffer and bead-labelled cell suspensions were passed through the magnetic field of a MACS separator using LS columns (Miltenyi). The flow through (containing myelin-freed cells) was retained, passed through a 70 µm MACS strainer (Miltenyi) and cells were re-suspended in a final volume of 50–100 µL of loading buffer before counting via trypan blue staining with a hemocytometer.
Single cell sequencing libraries were prepared from 10,000 cells using the Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 (Dual Index) (10× Genomics) according to manufacturer’s instructions. Briefly, single cell suspension made from the cortex and hippocampus were diluted into cDNA reaction mixtures to sequence 10,000 cells using the provided table from 10× genomics. The reaction mixture and gel beads were loaded onto Chromium Next GEM Chip G (10× genomics) before partitioning on a Chromium Controller. The resulting GEMs were further processed into sequencing libraries according to the manufacturer’s protocol without modification, except 15 cycles were used for the final PCR amplification step. Libraries were dual indexed using Dual Index Plate TT Set A and unique indices were used for each library. Quality of the amplified cDNA and final sequencing libraries was assessed using Bioanalyzer High Sensitivity DNA kits (Agilent) on a Bioanalzyer2000 instrument prior to sequencing. Libraries were sequenced to a minimum depth of 30,000 reads per cell on an Illumina NextSeq 2000 instrument using P3 reagents (100 cycles) with the recommended read configuration from 10× genomics (Read 1–28 cycles; i7 Index—10 cycles; i5 Index—10 cycles; Read 2–90 cycles).
Illumina sequencing reads were first pre-processed using the cellranger pipeline from 10× genomics. Raw bcl files were de-multiplexed using cellranger mkfastq, and cellranger count was then used to align fastq files to the mouse mm10 reference genome and count known transcripts. Each dataset was then independently quality controlled by removing ambient RNA reads, filtering low quality cells and removing doublets. The R package SoupX [87] was used to remove ambient RNA reads directly from the cellranger output using the default method of automatically estimating the contaminating fraction of UMIs. The following filtering criteria was then applied to remove low quality cells: n genes > 1000, % mitochondrial genome reads < 20, % ribosomal protein reads > 1, percent hemoglobin reads < 20. Finally, doublets were removed using the R package DoubletFinder [52] with default settings (pN = 0.25 and pK = 0.09), including the estimation of 7.6% doublets for 10,000 cells as indicated by 10 × genomics. The QC’d datasets were then normalized and integrated using the Seurat (v4) [27, 78] R package. Datasets were first normalized independently using the SCTransform function of Seurat, employing the "glmGamPoi" Gamma-Poisson Generalized Linear Model and regressing out the percentage of mitochondrial reads. Integration anchors between datasets were then identified with Seurat using reciprocal PCA analysis. One Mock (Mock48CX) and one RML (RML145CX) dataset were used as the reference for rPCA. Seurat was then used to integrate all datasets clustering cells using the default graph-based clustering approach and default UMAP dimensionality reduction approach. Cell clusters were then classified by brain cell type using the SCType [31] R package with a customized database of brain cell marker genes. The top marker genes of each cell cluster was also determined using the FindAllMarkers() function of Seurat and were inspected manually to make a final cell-type determination for each cluster. Cell clusters were classified as either astrocytes, microglia, perivascular macrophages, oligodendrocyte progenitor cells, glutamatergic neurons, immature neurons, endothelial cells, pericytes, vascular smooth muscle cells, vascular leptomeningeal cells, lymphocytes, or ependymal cells.
Marker genes were identified using Seurat using the default method of identifying differentially expressed genes between two groups of cells using a Wilcoxon Rank Sum test. p values were adjusted with the bonferroni correction using all genes in the dataset. Differentially expressed genes were identified by the following criteria: FDR p value < 0.05, |log2FC|> 0.5 and pct.1-pct.2 > − 0.1 or < 0.1 for increased/decreased genes respectively. This was used to identify differentially expressed genes between cell clusters, and between RML and Mock cells within individual clusters. Non-parametric Mann–Whitney U tests were used to compare the proportions of each cell cluster in the cortex and hippocampus between RML and Mock infected mice. We had limited statistical power when comparing cell populations between RML and Mock infected mice (only two hippocampal cell populations from Mock infected mice), so we applied a relaxed cutoff of p value < 0.1 to distinguish cell-population changes that were associated with disease. Hierarchical clustered heatmaps were prepared using the ComplexHeatmap[23] R package using the default Pearson distance method. Gene ontologies enrichment analysis of provided gene lists was performed with Enrichr [14, 43]. Enrichr was also used to identify lists of specific transcripts that were driving enrichment of gene ontologies and were mentioned throughout the text.
To profile the response of brain cell sub-populations to prion disease, we performed single cell RNA sequencing of cortical and hippocampal cells isolated from 8 RML infected mice when they reached clinical endpoint criteria at time points ranging from 150–172 dpi (Fig. 1A). For comparison, we also sequenced cortical cells from 4 mock mice collected at 147, 168, 186 and 189 dpi and hippocampal cells from 5 mock mice at 110, 147, 168, 186 and 189 dpi. We provide information on the mouse number, brain region, treatment and number of days post infection for each sequencing library included in our analysis in Additional file 1: Table S1. We were unable to exactly match the ages of Mock and RML mice used because we processed cortical and hippocampal brain tissues from one mouse per day. This was done to ensure consistency in the preparation of single-cell suspensions from all mice used in the study. Reagent clogs in the microfluidics of the chromium controller during separation of single cells reduced the usable dataset in the case of the hippocampal samples to 7 RML and 2 mock mice. These two mock mice used were collected at 110 and 147 dpi. Following pre-processing and quality control, the resulting 21 single-cell RNAseq datasets were integrated to produce an “atlas” of brain cells during prion infection, consisting of 147,536 cells that were classified into 39 transcriptionally distinct clusters via Seurat’s graph based clustering approach (Fig. 1B). We used automated classification of brain cell types with SCType (based on custom reference markers; Additional file 4) combined with manual inspection of marker genes (Additional file 2) to assign a cell type identity for each cluster. We noted that 3 of the clusters (8, 15 and 35) had unusually high expression of genes associated with technical-artefacts (e.g. high mitochondrial gene expression, Malat1 etc.) [15, 32]. Consequently, we could not identify clear brain cell type specific markers, so we removed these clusters from the final atlas. In total, we identified populations of astrocytes, microglia, perivascular macrophages, oligodendrocyte progenitor cells, glutamatergic neurons, immature neurons, endothelial cells, pericytes, vascular smooth muscle cells, vascular leptomeningeal cells, lymphocytes, and ependymal cells. We verified the identities of these cell types by examining the expression of the canonical marker genes P2ry12 (microglia), Gfap (astrocytes), Rbfox3 (mature neurons) Cd163 (perivascular macrophages), Pdgfra (oligodendrocyte progenitors), Spag17 (ependymal cells), Dcx (immature neurons), Mki67 (neural progenitors) and Cldn5 (endothelial cells) (Fig. 1C). Mature oligodendrocytes were absent from the dataset as expected, given the use of myelin removal beads to minimize clogs in the microfluidics of the chromium controller during cell separation. We then used these clusters representative of cellular sub-types to characterize differences related to brain pathobiology during prion infection. Given its relevance to disease, we examined the expression of Prnp across the single cell atlas, and found it to be most highly expressed by astrocytes, neurons, and surprisingly, ependymal cells (Additional file 1: Fig. S1).
To characterize transcriptional responses to prion disease, we performed differential expression analysis between all cells isolated from prion-infected versus mock-infected mice within each cluster independently (Fig. 2). According to criteria of FDR p values < 0.05, average |log2 fold change|> 0.5 and %cell-expression differences > − 0.1 or < 0.1 for increased/decreased genes respectively, we identified differentially expressed transcripts within most of the cell-type-specific clusters (Fig. 2A, Additional file 8). Further examination via hierarchical clustering of log2 fold change values within each cluster revealed the majority of transcriptional changes in response to prion disease showed little overlap between the different clusters, and we concluded that most were specific to cell-types or sub-types. (Fig. 2B). In single cell RNAseq, cells are assigned to specific clusters entirely based on their transcriptomes. Therefore, transcriptional responses to prion disease might also be reflected by differences in the relative frequency of cell clusters between prion- and mock-infected mice. In other words, differences in the relative frequency of sub-clusters of a given cell-type might indicate transitions from one transcriptional state to another during disease. Therefore, we examined the relative proportion of cells assigned to each cluster within the cortex and hippocampus from either RML infected- or mock-infected mice (Fig. 3, Additional file 5). Altogether, there were some striking differences in the relative composition of various cell-types associated with RML disease. Differences in relative frequency can also be influenced by changes to absolute cell count that occur during disease. In the context of prion disease, this is expected for reactive microglia and astrocytes that are well known to proliferate and for vulnerable neurons that decline in number due to cell death [74]. Technical factors can also influence the observed relative frequency, such as difficulties in cell dissociation, cell death during preparation of single-cell suspensions, and liberation of individual cells from debris or cell-to-cell contacts. Therefore, it is challenging to interpret whether the relative frequencies we present for each cell cluster reflect transitions of transcriptional status, or absolute challenges in cell count. Nonetheless, we present these relative frequencies because in many cases, this metric provided clues into the response of brain cell-types to prion disease. Additionally, the low sample size of n = 2 and n = 7 respectively for the Mock-hippocampus and RML-hippocampus groups was a limitation for distinguishing statistically significant disease-associated differences in relative frequency of hippocampal cell transcriptomes. Despite this, many changes in relative frequency were common between the hippocampus and cortex and were more reliable.
The blood brain barrier is comprised of various vascular cells including endothelial cells, pericytes, smooth muscle cells and vascular leptomeningeal cells [17]—all of which were present in our single cell atlas (Fig. 1). Disruptions to the blood brain barrier are a common feature of aging and neurodegeneration [42]. Consistent with this, prion-altered vascular transcripts were enriched in ontologies related to cell migration, blood vessel morphogenesis and vascular transport (Fig. 2B, Additional file 8). In addition to these transcriptional changes within vascular cell clusters, we also noticed that most vascular cell clusters decreased in relative frequency in association with disease (Fig. 3). We did not observe evidence of cell-death related transcription by vascular cells during prion disease, and so we cannot conclude whether the decrease of vascular cells were related to blood brain barrier breakdown. It is possible that we observed decreased vascular cell frequencies because of microglial proliferation or other technical factors. Many of the prion altered vascular were overexpressed by clusters endo.2, endo.11 and peri.14. Specific examples of notable disease-altered transcripts that represent enriched gene ontologies are listed as follows: Cluster endo.2 overexpressed transcripts related to cell migration (Sema5a, Flt1, Lef1, Pecam1, Gcnt2, Rhoc, Dock1, Ptk2, Rab11a, Igf1r) and angiogenesis (Sema5a, Ramp2, Flt1, Rock1, Rhoj, Tek, Ism1). Cluster endo.11 overexpressed transcripts related to blood brain barrier transport (Slco1c1, Slc16a1, Slc2a1, Mfsd2a). Cluster peri.14 overexpressed transcripts related to actin filament/supramolecular fiber organization (Carmil1, Myo1b, Ubb, Col8a1, Myh11, Mfge8, Aldoa, Svil, Eps8). These signatures of abnormal transcription seem to hint of vascular dysfunction or remodeling that might occur during prion disease, with blood brain barrier transport possibly being altered. Inflammation is well known to cause disruptions to brain vascular cells [42], so we were not surprised to observe evidence of vascular dysfunction in our single cell atlas. However, our analysis cannot determine whether blood brain barrier permeability is altered at the phenotypic level during prion disease, and so more detailed studies are required to investigate this hypothesis.
Transcriptional alterations to oligodendrocytes are rarely the focus of investigations into prion pathogenesis because mature oligodendrocytes are considered relatively resistant to prion replication [66]. However, mature Olig2+ oligodendrocytes were recently shown to decrease at the advanced stages of disease in a murine model of Creutzfeldt-Jakob disease [3], implying a role in pathogenesis. In our dataset, populations of oligodendrocyte progenitors were increased in relative frequency during RML disease, particularly in the hippocampus (Fig. 3). This is consistent with a previous bulk RNAseq analysis, where we inferred increased oligodendrocyte progenitors through increased Pdgfra abundance during RML disease [72]. However, it is also possible that we observed the increase in oligodendrocyte progenitor frequency due to technical reasons such as resistance to cell death during isolation relative to other cell types in the condition of prion disease. Disease-altered oligodendrocyte progenitor transcripts were enriched in ontologies related to neuron differentiation, cAMP signaling and response to retinoic acid and included downregulation of canonical oligodendrocyte progenitor cell markers Pdgfra and Vcan (Fig. 2B, Additional file 8). These transcripts were disease-altered in cluster opc.19, but not opc.37. A few disease-altered transcripts were also shared between oligodendrocyte progenitor cells with neurons and astrocytes. Notable upregulated disease-altered oligodendrocyte progenitor transcripts were related to cell adhesion (Kirrel3, Ptprt, Tenm1, Unc5d, Lrrc4c, Dscaml1, Cdh8, Cdh18), synapse assembly (Gabrb3, Kirrel3, Dnm3, Gabrb2, Farp1, Ppfibp1, Lrrc4c, Ppfia2) and neurotransmission (Gria2, Neto1, Dlgap2, Gria3, Grin3a). Notable downregulated oligodendrocyte progenitor transcripts were related to gap junctions (Ptprd, Cntnap2, Agt) and nervous system development (Cntnap2, Vcan, Cntn4, Adgrl3). These findings suggest that transcriptional dysfunction oligodendrocyte progenitor cells is an underappreciated component of prion pathogenesis—an avenue that is worthy of further investigation.
Microglia made up the largest population of cells (99,756/147,536 = 67%) assigned in our library and are well described as particularly responsive to prion replication, taking on reactive phenotypes that can both exacerbate pathology through excess inflammatory signaling and can protect against disease through clearance of toxic PrPSc [53, 58, 63]. As expected, altered microglial transcripts were involved in cytokine signaling, phagocytosis and microglial activation (Fig. 2B, Additional file 8). Surprisingly few markers of reactive microglia were increased within individual microglia clusters, although there were a few such as Lyz2, Apoe, Tyrobp and Irf8. Some microglia-specific markers were decreased across some of the individual microglial clusters, such as P2ry12, Tmem119, Csf1r, and Cx3cr1. Homeostatic markers like P2ry12 and Tmem119 are generally reported to decrease in reactive microglia [41, 54]. Similar to microglia, prion altered transcripts within perivascular macrophages were related to inflammatory signaling through cytokines, chemokines and antigen receptors. There was little overlap between prion-altered transcripts of microglia and perivascular macrophages, implying a distinct response to prion disease. Unsurprisingly, some of the most drastic changes in cellular populations during RML disease were seen in microglia (Fig. 3). We noticed a few of the microglial clusters either decreased, or did not change in relative frequency, whereas many of the microglial clusters increased in abundance and we suspected that these corresponded to reactive, or “disease-associated” microglia that are well known to increase during disease [74]. Unlike microglia, we did not observe an expansion of perivascular macrophages in disease, and perivascular macrophages (pvm.22 and pvm.29) appeared to decrease in the hippocampus and cortex. This could possibly reflect clustering of activated perivascular macrophages with the disease associated microglia, or cellular migration to other brain regions.
Nearly 100,000 microglial transcriptomes were sequenced, making our single cell dataset particularly well suited to characterize the diversity of microglial activation states during prion disease. Furthermore, compared to single nucleus sequencing, our live single-cell sequencing approach can improve detection of transcriptional changes within activated microglia [82]. Therefore, to define transcriptional states of individual microglia subtypes, we subset the dataset to include only the 99,756 microglial transcriptomes (Fig. 4A). We retained the original microglial clusters from the full atlas, and did not perform further sub-clustering. When comparing microglia isolated from infected with mock- infected mice, we could see that some microglial clusters were nearly unique to disease (micro.9, micro.12, micro.17, micro.23, and micro.36) and were strongly associated with disease (Fig. 4A). We next functionally characterized the microglial subtypes by identifying marker genes highly expressed by each cluster, up to a maximum of 25 per cluster, resulting in 218 transcripts supplied for hierarchical clustering (Fig. 4B). K-means clustering of genes was used to classify the marker transcripts into 8 gene modules that were functionally annotated with representative enriched gene ontologies. Altogether, we found that reactive microglia take on a spectrum of transcriptional states characterized by expression of genes important for various aspects of glial functionality such as phagocytosis, or cytokine signaling. Based on expression of these gene modules (Fig. 4B), and the expression of specific microglial marker transcripts (Additional file 1: Fig. S2), we further classified microglia into 5 subtypes: (1) homeostatic, and the following reactive subtypes: (2) proliferating, (3) phagocytic, (4) type I interferon (IFN) responding, and (5) antigen presenting (MHC). These functional subtypes are similar to what has previously been reported in relation to Alzheimer’s disease [16]. We also classified some of the microglial clusters as representing intermediate transcriptional states between these subtypes. Within the subset microglial dataset, the homeostatic microglial clusters decreased in relative frequency, whereas reactive microglial clusters (proliferating, phagocytic, IFN and MHC subtypes) increased in relative frequency in association with disease (Fig. 5A, Additional file 9). This could indicate conversion of homeostatic microglia into reactive forms. Therefore, we performed a monocle trajectory analysis of the microglial cells to measure transcriptional status as a function of gene “pseudotime” by supplying cells from cluster micro.4 to serve as the “root” for the trajectory (Fig. 5B and C). We noted a circular transcriptional trajectory between homeostatic microglia, intermediately activated microglial states, and proliferating microglia with branches into phagocytic, antigen-presenting and interferon-responding microglial populations. Altogether, our interpretation was that microglia form a continuous spectrum of transcriptional states, where multiple possible transcriptional trajectories can allow homeostatic microglia to reach distinct disease-associated reactive states, thus mirroring the complexity of functionally distinct phenotypes observed in the brain.
Homeostatic microglia (micro.3 and micro.4) were marked by high expression of the canonical microglial markers P2ry12, Cx3cr1, Tmem119 in addition to Nav2, a marker of microglia under healthy conditions [75] (Additional file 1: Fig. S2). These microglia had high abundance of gene modules 6, 7 and 8 that were enriched in ontologies related to calcium transport (Cacnb2, Bcl2, Ank2), regulation of glial apoptosis (Prkca) and regulation of microglial migration (P2ry12, Cx3cr1) (Fig. 4A). Homeostatic microglia elicited a disease-associated decrease in relative frequency (Fig. 5A), although this does not necessarily indicate that the absolute cell count of homeostatic microglia decreases in the prion infected brain. Given the possible transcriptional trajectories that could allow homeostatic microglia to reach different reactive states (Fig. 5B and C), this decrease in relative frequency likely corresponds to conversion of homeostatic microglia into intermediate and eventually reactive transcriptional states during disease. Gene modules 4 and 5 were involved in cytokine signaling, regulation of inflammation and regulation of cell proliferation and were highly expressed by proliferating microglia (micro.0, micro.9 and micro.12) (Fig. 4A) that were demarked by high expression of Jun, Fos and Il1a (Additional file 1: Fig. S2). Together, Jun and Fos encode proteins that form the AP1 transcription factor that induces inflammatory gene expression in microglia [84]. Interestingly, cluster micro.12 was uniquely marked by very high expression of the cytokine Il12b (Additional file 1: Fig. S2). Furthermore, Cd14 was highly expressed by micro.9 and micro.12 and is a co-receptor for LPS that modulates inflammatory signaling, important for microglial responses to tissue damage-associated signals [35]. We also noted an expansion of proliferating microglial clusters micro.9 and micro.12 in association with disease (Fig. 5A). Altogether, these results suggest that proliferating Jun+Fos+ microglia might contribute towards inflammatory cytokine signaling during prion infection. Specific examples of cytokine signaling related transcripts expressed by these microglia include Cd86, Egr1, Pdgfb, Fos, Ptgs2, Cxcl2, F3, Nfkb1, Socs3, Bcl6, Il1b, Ccl4, Il12b, Tnfsf9, and Junb. Phagocytic microglia (micro.13) were marked by high expression of Aif1, Ftl1 and Fau (Additional file 1: Fig. S2) and highly expressed gene module 2 that was enriched in synapse pruning and microglial activation, containing many microglial activation markers (Fig. 4B). Specific examples of classical microglial activation markers expressed by these microglia include C1qa, C1qb, C1qc, Tyrobp, Trem2, Aif1, B2m, Prdx5, Fcer1g, Cstb, Ctsz, Cd63, and Cd68. Given that this corresponds to a classic signature of reactive microglia, we were not surprised to see that the relative frequency of micro.13 increased in association with disease (Fig. 5A). Interestingly, Aif1+Ftl1+ microglia corresponded morphologically dystrophic iron-accumulating microglia in an Alzheimer’s mouse model [40], possibly providing clues as to the role of phagocytic microglia in prion disease. Antigen presenting microglia (micro.17 and micro.36) highly expressed phagocytosis related genes (gene module 2, also highly expressed by the phagocytic microglia cluster micro.13) and genes important for antigen presentation (Fig. 4B). These microglia were marked by high expression of Cd74, H2-Aa, Cd52 and Ccl6 (Additional file 1: Fig. S2). The antigen presentation genes that were highly expressed by these microglia were Cd74, H2-Aa, H2-Eb1, H2-Ab1, H2-K1, and H2-D1. Clusters micro.17 and micro.36 showed some of the most dramatic increases in relative frequency in association with disease (Fig. 5A) and were nearly absent the Mock mice. In fact, micro.36 was not detected at all in the hippocampal cells isolated from Mock mice and was only detected in one cortical cell suspension of Mock mice. Therefore, we postulate that these antigen-presenting microglia subtypes represent highly activated reactive microglia that are strongly associated with prion disease. Cd74 is thought of as a marker of M1 microglial activation, and is expressed by highly activated microglia in the diseased-brain [37, 79], supporting this notion. Trim30a, Oasl2 and Cxcl10 were highly expressed by type I interferon responsive microglia (micro.23, Additional file 1: Fig. S2) that highly expressed gene module 1 (Fig. 4). Examples of type I interferon responsive transcripts expressed by these microglia include Ifitm3, Bst2, Rsad2, Isg15, Ifit1, Gbp2, Ifit3, Ifit2, and Cxcl10. Like the other reactive microglia subtypes, the type I interferon responsive microglia also show a strong disease-associated increase in relative frequency (Fig. 5A). Type I interferon signaling is often thought of as detrimental in the context of brain pathology, but a recent study has suggested that this pathway might protect neurons during prion infection [33]. As expected, microglia that were considered to represent intermediate transcriptional states (micro.1, micro.5, micro.6, micro.7, and micro.25) had varying expression of the different microglial gene modules (Fig. 4B). Interestingly, micro.6 was uniquely marked by very high expression of Serpine1 (Additional file 1: Fig. S2)—an inhibitor of tissue plasminogen activator (tPA) that promotes microglial migration and inhibits phagocytosis in vitro [36]. Intermediate microglial clusters micro.6 and micro.7 were both positively associated with disease through increases in relative frequency (Fig. 5A).
Astrocytes are one of the main cells types responsible for brain homeostasis through neurotransmitter uptake/recycling, potassium buffering, metabolism, and protection against oxidative stress among other neuroprotective functions [6]. In the context of prion disease however, astrocytes are one of the first cells to take on active phenotypes during disease that may have various beneficial or detrimental roles, concomitant with the earliest detectable deposits of PrPSc [81]. We observed a striking global decrease in relative frequencies of astrocyte populations associated with RML disease (Fig. 3) and to our surprise; we did not observe a clearly resolved cell cluster corresponding to reactive astrocytes. Disease-altered astrocyte transcripts were enriched in ontologies related to synapse organization, blood brain barrier transport and sulfur biosynthesis, hinting at modulation of astrocyte homeostasis functions (Fig. 2B, Additional file 8). Most of the disease-altered astrocyte transcripts were upregulated in cluster astro.10, and notable transcripts were the reactive astrocyte marker Gfap, and transcripts related to cell junction assembly (Kirrel3, Gpm6a, Farp1, Cdh2, Cdh20, Ctnnd2, Nrcam, Cdh19), sulfur metabolism (Bcan, Gstm1, Angpt1, Cspg5, Prelp, Chsy3, Gstm5) and cell projection organization (Ntrk2, Fut9, Magi2, Il1rapl1, Atp1b2, Prkd1). We also noticed a notable group of transcripts downregulated in cluster astro.20 that were related to axonogenesis (Robo2, Auts2, Nrxn3, Slit2). This signature of differential transcription indicates dysfunction of the homeostatic astrocytes that were captured by our live single cell approach. To better characterize the population of astrocytes isolated, we performed a sub-cluster analysis by combining all 7,813 astrocyte transcriptomes (from clusters astro.10 and astro.20) and re-clustering into 11 new astrocyte sub-clusters (Fig. 6A). We examined the relative frequency of these astrocyte sub-clusters among all astrocytes and classified them based on whether they were depleted (“disease-depleted”), unchanged, or increased (“disease-associated”) during disease (Fig. 6C, Additional file 6). The majority of the astrocyte sub-clusters decreased in the prion infected brains, but two (astrocyte sub-clusters 6 and 8) increased, and we suspected that these might correspond to a small population of reactive astrocytes. We examined the expression of astrocyte marker genes across the astrocyte sub-clusters (Fig. 6B and Additional file 1: Fig. S3) and noted that disease-associated astrocyte sub-cluster 8 had high expression of Gfap, Aqp4, Vim, and low expression of Nrxn3, consistent with reactive astrocytes [19]. To compare disease-depleted with disease-associated astrocyte subpopulations, we next performed hierarchical clustering of all transcripts that were differentially expressed between astrocyte sub-clusters (Fig. 7). K-means clustering was used to classify astrocyte transcripts into 7 gene modules that were differentially abundant across the astrocyte sub-clusters and were enriched in gene ontologies relevant to astrocyte homeostasis functions such as regulating vascular permeability, axonogenesis, synaptic membrane adhesion, removal of superoxide radicals and metabolism (Fig. 7). Disease-depleted astrocyte sub-clusters (0, 2, 3, 4, 5, and 7) varied in expression of these homeostasis-related transcripts indicating that they represent populations of neuroprotective astrocytes. The top transcriptional markers of astrocyte sub-cluster 8 were S100a6, S100a1, Prdx1, and Hopx and among these, S100a6 [7, 29], and Prdx1 [80] are known to be expressed by disease-associated astrocytes. Astrocyte sub-cluster 8 also had particularly low expression of the homeostatic/neuroprotective gene modules (Fig. 7, gene modules 3, 4, 5, 6, and 7), suggesting loss of neuroprotection by this disease-associated sub-cluster. The top transcriptional markers of astrocyte sub-cluster 6 were Glis3, Cadm1, Zbtb20 and Maml2 (Additional file 1: Fig. S3). Cadm1 mediates astrocyte-to-astrocyte adhesion [69] while Zbtb20 promotes astrocytogenesis [56]. However, we did not observe a clear transcriptional profile unique to the disease-associated astrocyte sub-clusters 6 and 8 (Fig. 7), indicating that they were at best, only at the very early stages of becoming reactive. To exclude the possibility of reactive astrocytes clustering together with reactive microglia, we examined canonical astrocyte makers in the sub-clustered microglia dataset and did not observe the presence of transcriptionally distinct reactive astrocytes (Additional file 1: Fig. S4). Populations of astrocytes are well known to be maintained throughout prion disease [74], and we have no reason to suspect decreased astrocyte cell counts within the brain. Therefore, we attribute the decrease in astrocyte relative frequency during prion disease (Fig. 3) to a lack of reactive astrocytes from our dataset. The lack of reactive astrocytes is most likely the result of technical limitations of our approach for preparing single-cell suspensions. For example, reactive astrocytes might have been closely associated with cell debris and removed from the single cell suspensions. Unfortunately, our live scRNASeq approach did not allow us to assess how different the transcriptomes of strongly activated reactive astrocytes are from those of neuroprotective astrocytes. However, we suspect that the decrease in relative frequency of neuroprotective astrocyte sub-clusters (0, 2, 3, 4, 5, and 7) might indicate that these neuroprotective astrocytes are being converted into a reactive form. The differential expression analysis of astrocyte clusters from the full single cell atlas (Fig. 2B) would suggest that the few astrocytes isolated from the RML infected mice were at the very early stages of becoming reactive with disrupted homeostatic/neuroprotective functions, likely corresponding to astrocyte sub-clusters 6 and 8 identified by the astrocyte sub-cluster analysis (Fig. 7). If this is true, the striking disease-associated depletion of astrocytes from our single-cell atlas (Fig. 3) would imply that nearly all homeostatic astrocytes are converted to a reactive form. However, the technical limitations of our approach prevent us from making this conclusion, and it is possible that single nucleus RNAseq is better suited towards characterizing reactive astrocytes. Therefore, further studies are required to determine how many neuroprotective astrocytes remain non-reactive during disease, and this requires identification of specific markers to delineate between neuroprotective and reactive astrocytes. This would be an interesting line of investigation because observations by others suggest that loss of astrocyte homeostatic functions during prion disease might contribute to neurotoxicity [5, 44]. Indeed, loss of astrocyte homeostasis is a common feature of neurodegeneration [10, 65] and can contribute to neurotoxicity through abnormal EGFR signaling [85], excessive glutamate release/defective glutamate uptake [28, 60, 61, 71], oxidative stress [1], and dysfunctional metabolism [2].
Transcriptional changes related to neuronal dysfunction and demise have been challenging to identify using bulk RNAseq data, so we were interested to see how scRNASeq could contribute to defining molecular pathways of cell damage and death in neurons. Firstly we noted that the number of neurons within our dataset was small (8,244/147,536 = 5.6%). This was not surprising as the cellular connectivity and extended processes of neurons may make them particularly vulnerable to cell disruption techniques. However, we identified disease-altered neuronal transcripts involved in synaptic signaling and axon guidance (Fig. 2B, Additional file 8), similar to what is seen in bulk RNAseq [12, 30, 39, 46–49, 72, 73]. There were also altered transcripts of immature neurons that were related to Cxcl2 production, response to magnesium and exosome biogenesis. As expected, we noticed a trend towards decreased frequencies of mature neurons, while some of the immature neuron populations were increased (Fig. 3). To more precisely characterize neuronal subpopulations that differentially respond to prion disease, we collapsed all mature and immature neuron transcriptomes (g.neu.16, g.neu.31, g.neu.33, g.neu.34, im.neu.18, im.neu.24 and im.neu.32) and re-clustered the 8,244 cells into 16 transcriptionally distinct clusters (Fig. 8A). We categorized these clusters based on the abundance of marker genes for neural progenitor cells (Mki67), immature differentiating neurons (Dcx), cajal-retzius neurons (Reln), mature neurons (Rbfox3), excitatory neurons (Slc17a7) and inhibitory neurons (Gad1) (Fig. 8A and Additional file 1: Fig. S5). Clusters m.neu.0 and m.neu.11 were categorized as mature neurons because they did not express clear markers of either excitatory or inhibitory neurons. We also examined the expression of Prnp and found it to be most highly expressed by excitatory neurons (Additional file 1: Fig. S5). We next compared the relative frequency of the cell populations within the sub-clustered neuronal dataset and observed several cellular composition differences associated with prion disease (Fig. 8B, Additional file 7). Furthermore, we compared neuronal gene expression profiles via hierarchical clustering of all identified marker genes for each neuronal sub-cluster (Fig. 9). K-means clustering was used to classify these transcriptional markers into 5 gene modules and each neuron subtype expressed gene modules enriched with relevant functional ontologies. Neural progenitor cells expressed cell cycle genes, differentiating neurons expressed axon guidance genes and mature neurons primarily expressed synaptic signaling genes (Fig. 9).
The relative frequency of neural progenitor cells increased in association with RML disease, especially in the hippocampus (Fig. 8B). This was consistent with a number of studies that have detected increased neurogenesis in the hippocampus [20, 22] which may protect against prion disease [34]. Relatively few transcripts of neural progenitors were differentially expressed (Fig. 2, cluster im.neu.24), although we noted the top overexpressed transcripts were Spp1 and Lpl, and the top underexpressed transcripts was Csmd3. We also identified 5 clusters of immature differentiating neurons, some of which appeared to increase or decrease in association with prion disease (Fig. 8B), although only cluster diff.neu.3 achieved statistical significance (increased in the cortex). Therefore, this analysis was unable to resolve whether any of the differentiating neuron sub-clusters were associated with disease. We did however notice a number of differentially expressed transcripts of differentiating neurons, both within the full single cell atlas (Fig. 2, cluster im.neu.18) and within individual sub-clusters from the sub-cluster analysis (Additional file 1: Fig. S6). Upregulated transcripts of differentiating neurons in the full single cell atlas (cluster im.neu.18) were related to axonogenesis (Hsp90aa1, Dab1, Auts2, Dcc, Nrxn1, Ntn4, Ncam1, Ank3, Cntn4), potassium transport (Slc24a2, Kcnt2, Kcnd3, Kcnh7, Kcnb2, Kcnq5, Kcnn2) and synapse organization (Cacnb4, Cdh2, Nrxn1, Ank3, Snca), together implicating possible modulation of neuronal differentiation. Other studies have shown that PrPSc can replicate in Dcx+ immature neurons, resulting in impaired differentiation [68] and that newborn neurons differentiate abnormally in prion infected mice [22]. Thus, the disease-associated transcription of differentiating neurons might represent dysfunction, possibly explaining why neurogenesis is ultimately unsuccessful at mitigating neuronal loss during prion disease. We were also surprised to observe increased relative frequencies of cajal-retzius neurons in the hippocampus of RML infected mice (Fig. 8B). One of the primary functions of cajal-retzius neurons is secretion of Reelin (Reln), a protein that regulates neuron migration during neurogenesis [13]. It is therefore possible that cajal-retzius neurons are involved in abnormal hippocampal neurogenesis during prion disease. Very few transcripts of cajal-retzius neurons were differentially expressed (Fig. 2, cluster im.neu.32), so we have no reason to suspect that these cells were dysfunctional. We are not aware of any studies that have examined cajal-retzius neurons during prion disease, but they are reportedly decreased in the hippocampus due to apoptosis in an Alzheimer’s disease mouse model [88]. This might indicate that increased numbers of cajal-retzius cells are distinguishing feature of prion disease. Of course, here we only present relative frequencies, and it is possible that the increase was due to technical reasons, such as resistance of cajal-retzius neurons to death during our cell isolation protocol.
In the sub-clustered neuronal dataset, many excitatory neurons were generally enriched in either the hippocampus or cortex, and were not strongly affected at the population level by RML disease (Fig. 8B). Clusters ex.neu.9 and ex.neu.15 even trended towards a slight increase in the hippocampus of RML infected mice. We also identified populations of inhibitory neurons that were generally decreased in the hippocampus of RML infected mice (Fig. 8B). The trends towards slightly increased excitatory neurons and decreased inhibitory neurons in the hippocampus is consistent with previous histopathological analyses showing inhibitory neurons to be more vulnerable to prion infection [8, 24–26]. We found that excitatory neuron populations sustained during disease (ex.neu.9 and ex.neu.15) had high expression of gene module 2, which was expressed at much lower levels in the inhibitory neuron populations that appeared more sensitive to RML disease (inh.neu.10 and inh.neu.13) (Fig. 9). Gene module 2 was enriched in ontologies related to neuron maintenance, such as disordered domain binding, regulation of phosphatase activity, response to metal ion and response to unfolded protein. We were also surprised to see that Prnp was among gene module 2, and was more highly expressed by the neurons that appeared more resistant to prion infection compared to those that appeared more sensitive. Sub-cluster analysis of the excitatory and inhibitory neurons did not reveal any distinct clusters of transcriptomes that were associated with disease, and so we instead examined disease-associated transcription through differential expression analysis. Within the full single cell atlas, the main cluster of mature neurons (g.neu.16) had the largest number of differentially expressed transcripts out of any cell cluster with 168 upregulated and 128 downregulated. These disease-altered transcripts seem to hint at synaptic dysfunction. Notable upregulated transcripts of cell cluster g.neu.16 were related to axonogenesis (Robo2, Hsp90aa1, Epha6, Dcc, Sema6d, Nrxn3, Unc5d, Ank3, Kif5c, Tubb3, Map1b, Fez1, Ncam1, Slit2, Dscaml1), regulation of cation transport (Camk2b, Cacnb4, Dlg2, Lrrc7, Camk2a, Kcnab1, Ank2, Ank3, Grin2b, Shisa6), and synaptic transmission (Gabra2, Dtna, Syt1, Grid1, Grik2, Grin2b, Shisa6, Nrgn, Cacnb4, Dlg2, Grm7, Npy, Slc17a6, Asic2, Erc2, Dlgap2). Similar types of neuronal transcripts were also downregulated in cluster g.neu.16, including those related to synaptic transmission (Gabrb3, Snap25, Gabrb2, Gabbr2, Grid2, Ptprn2, Kcnd2, Nrxn1, Lin7a, Grik1, Cdh8, Rims1, Dlgap1, Gria3, Gria4), neuronal projections (Epha4, Ntrk2, Cdh2, Fut9, Il1rapl1, Nptn, Ndnf, Ctnna2, Tox, Pak3) and axonogenesis (Epha4, Dab1, Lama1, Dok6, Nrxn1, Nptn, Nrcam, Cck, Ctnna2, Efna5, Pak3). Within the full single cell atlas, excitatory and inhibitory neuronal subtypes were not completely resolved. Therefore, we also performed a differential expression analysis between cells isolated RML and Mock infected mice within each cluster in the sub-clustered neuronal dataset (Additional file 1: Fig. S6). Many of the differentially expressed transcripts originated from excitatory neuron populations and were related to synaptic transmission, although there were also differentially expressed transcripts in inhibitory neuron and differentiating neuron populations. This raises the possibility of synaptic dysfunction within the excitatory neurons that were not depleted in our dataset. A few of these synaptic transcripts were also altered in inhibitory neurons, but we noted a group of disease altered transcripts of inhibitory neurons that was not altered in excitatory neurons. These transcripts were enriched in ontologies related to potassium transport (Ank2, Atp1b2, Atp1b1) and calcium homeostasis (Calm3, Tmtc2, Ank2, Atp1b1, Snca), implying a distinct response of inhibitory neurons. Given that we were unable to resolve genuine disease-associated clusters of neuronal transcriptomes, a larger single-cell dataset of neurons is warranted to further investigate selective vulnerability and to better define molecular disruptions to neurons during disease.
We report an extensive library of 147,536 single cell transcriptomes from matched tissue samples of prion infected and non-infected mice. Whilst previous studies rely on bulk RNAseq to measure average transcript abundances within a tissue [12, 30, 39, 46–49, 70, 72, 73], here we profiled transcript-level and cell-population level responses to prion disease. To minimize cell-composition complexity and to target brain regions particularly affected during prion disease, we chose the murine cortex and hippocampus for this study. The data provides further resolution of the pathobiology of disease and some of the more striking findings were the apparent dysfunction of homeostatic astrocytes and vascular cells, the diversity of reactive microglia and differential response of neuronal populations. This represents a new technological advance with huge potential for uncovering the molecular basis for pathological changes within the prion-infected brain at the cellular subtype level. Our single cell differential gene expression correlates in many respects with previous bulk RNAseq datasets of prion infection [12, 30, 39, 46–49, 70, 72, 73]. Glia mount the most prominent phenotypic response to infection, and this was readily observable in our dataset. Our analysis precisely tracked transcriptional changes within individual subpopulations of brain cells and distinguished brain cell subtypes that were associated with prion disease. Nonetheless some limitations exist with this approach including biased selection for certain brain cell types during preparation of single cell suspensions, as well as any transcriptional changes that might occur during the process of tissue dissociation. Thus, the cell populations profiled here likely do not fully reflect brain cells in their natural state during prion disease. Other approaches to analyze individual cell types, such as single nuclei sequencing, and ribosomal profiling have their own associated technical challenges [4, 9, 21, 86], and we believe multiple approaches are necessary to fully describe molecular and cellular changes in the prion infected brain. Given that we have identified a number of unique cell type clusters, the next steps will be to perform validation of these cell types within the brain and determine the interplay between the different sub-populations and replicating prions. We characterized 11 sub-clusters of transcriptionally distinct microglia that differentially express functional markers of homeostasis, inflammatory cytokine signaling, phagocytosis, and antigen presentation. Based on expression of these functional markers, and consistent with observations from single cell RNAseq studies of Alzheimer’s disease [16], we described 5 subtypes of microglia including: (1) homeostatic, (2) proliferating, (3) phagocytic, (4) type I interferon responding (IFN) and (5) antigen presenting (MHC). Phagocytic, proliferating, IFN and MHC microglia subtypes corresponded to reactive microglia and were associated with disease through increased relative frequency within the RML infected mice. The microglia from our study were similar to those seen in a recent single-cell RNAseq study of human Alzheimer’s disease patients [59], raising the possibility that they are relevant to disease in humans. Specific genes such as Il12b, Serpine1, Jun, Ftl1, Nav2, Cd14, Trim30a and Cd74 demarked some of the microglial subtypes—possibly serving as markers of functionally diverse microglial subsets. Therefore, the next steps will be to verify these microglial sub-populations throughout disease progression and to determine their relative contribution to the reported protective and detrimental properties of activated microglia during prion disease [53, 58, 63]. In contrast to previous bulk RNAseq analyses of prion infected brain tissue that primarily identify inflammatory gene expression [12, 30, 39, 46–49, 72, 73], disruptions to brain homeostasis were much more apparent in our single cell dataset. This includes the dysregulation of homeostatic/neuroprotective astrocyte gene expression, decrease in relative frequency of homeostatic microglia, transcriptional dysfunction of vascular cell populations that make up the blood brain barrier, modulation of oligodendrocyte progenitor cells and abnormal neurogenesis. These details provide additional context towards understanding neuronal dysfunction and demise in prion disease, given that loss of astrocyte neuroprotection in particular can result in neurotoxicity [1, 2, 28, 60, 61, 71, 85]. Moreover, single cell studies of Alzheimer’s disease have found disruptions to homeostatic astrocytes and vascular cells to be important components of pathogenesis [45, 51]. From this, it is apparent that restoring brain homeostasis will be an important consideration for developing therapeutics against prion disease, in addition to removal of the disease-causing agent and attenuation of excess inflammatory signaling. Interestingly, we observed increased relative frequency of proliferating cell populations in the prion-infected brain, including neural progenitor cells, oligodendrocyte progenitor cells and proliferating microglial subtypes. While the relative frequencies presented here do not equate to true measurements of absolute cell count, previous studies have implicated proliferation of microglia [74] and neural progenitor cells [20, 22] during prion disease. We speculate that a common means exists to promote cell proliferation during prion disease—possibly in an unsuccessful attempt to restore brain cells that are lost during disease. Indeed, activated microglia are known secrete factors that promote oligodendrocyte progenitor cell proliferation [83] and enhance neurogenesis [18, 57, 64]. Alternatively, PrPC can modulate both proliferation of oligodendrocyte progenitor cells [11] and neurogenesis [62, 77]—raising the possibility that lack of functional PrPC owing to prion replication could contribute to increased cell proliferation. Therefore, uncovering exactly how neural progenitor cells are modulated during prion disease would be an interesting approach that could help inform strategies to restore dying neurons and repair damage in the diseased brain. In conclusion, single-cell RNAseq represents a comprehensive approach to characterize transcript-level and cell-composition changes throughout prion disease. We identified numerous disease-associated cellular subpopulations that warrant further validation, particularly in the case of microglia. Our analysis highlights the complexity of the glial and neuronal reactome to prion replication and accumulation. Future applications of the data will be to identify specific transcriptional markers that distinguish pathological cell phenotypes and to gain further molecular insight into the disruptions that underlie neurodegenerative progression in prion diseases. This includes discriminating sup-populations of disease-responding cells as either neuroprotective or driving pathology. Also important will be defining commonalities and differences between disease processes of various degenerative diseases. Overall, this dataset, and others like it, provide higher resolution in the journey to unravel the complex dysregulation occurring in different brain cell types throughout neurological disease during prion infection.
Additional file 1: Supplementary tables and figures.Additional file 2: Globally distinguishing transcriptional markers of each cell cluster within the full single-cell atlas. Only positive markers were reported, identified via criteria of |log2FC| > 0.5, FDR p-value < 0.05 and %cell expression difference > 0.1.Additional file 3: Total number of cells isolated from RML or Mock infected mice that were classified into each of the cell clusters from the full single cell RNAseq atlas.Additional file 4: Cell type identity scores calculated using SCType for each cell cluster in the full single cell RNAseq atlas.Additional file 5: Mann-Whitney p-values corresponding to relative frequency differences between RML and mock infected mice for each cell cluster in the full single cell RNAseq atlas.Additional file 6: Mann-Whitney p-values corresponding to relative frequency differences between RML and mock infected mice for each cell sub-cluster in the subset astrocyte dataset.Additional file 7: Mann-Whitney p-values corresponding to relative frequency differences between RML and mock infected mice for each cell sub-cluster in the subset neuron dataset.Additional file 8: Differentially expressed transcripts between cells isolated from RML and Mock infected mice within each cell cluster in the full single cell RNAseq atlas. Differentially expressed transcripts were defined by criteria of FDR p-value < 0.05, |log2FC| > 0.5 and %cell-expression difference > -0.1 or < 0.1 for increased/decreased genes respectively.Additional file 9: Mann-Whitney p-values corresponding to relative frequency differences between RML and mock infected mice for each cell sub-cluster in the subset microglia dataset. |
|
PMC9647952 | Kimiaki Takagi,Azumi Naruse,Kazutoshi Akita,Yuka Muramatsu-Maekawa,Kota Kawase,Takuya Koie,Masanobu Horie,Arizumi Kikuchi | CALN1 hypomethylation as a biomarker for high-risk bladder cancer | 09-11-2022 | Bladder cancer,CALN1,Methylation analysis,Methylation-sensitive restriction enzyme (MSRE),Molecular diagnosis technique,Transurethral resection of bladder tumor (TURBT) | Background DNA methylation in cancer is considered a diagnostic and predictive biomarker. We investigated the usefulness of the methylation status of CALN1 as a biomarker for bladder cancer using methylation-sensitive restriction enzyme (MSRE)-quantitative polymerase chain reaction (qPCR). Methods Eighty-two bladder cancer fresh samples were collected via transurethral resection of bladder tumors. Genomic DNA was extracted from the samples, and MSRE-qPCR was performed to determine the CALN1 methylation percentage. Reverse transcription-qPCR was performed to assess the correlation between CALN1 methylation and mRNA expression. The association between CALN1 methylation percentage and clinicopathological variables of all cases and intravesical recurrence of non-muscle-invasive bladder cancer (non-MIBC) cases were analyzed. Results Of the 82 patients, nine had MIBC and 71 had non-MIBC who had not undergone total cystectomy. The median CALN1 methylation percentage was 79.5% (interquartile range: 51.1–92.6%). The CALN1 methylation percentage had a negative relationship with CALN1 mRNA expression (Spearman’s ρ = − 0.563 and P = 0.012). Hypomethylation of CALN1 was associated with advanced tumor stage (P = 0.0007) and histologically high grade (P = 0.018). Furthermore, multivariate analysis revealed that CALN1 hypomethylation was an independent risk factor for intravesical recurrence in non-MIBC patients (hazard ratio 3.83, 95% confidence interval; 1.14–13.0, P = 0.031). Conclusion Our findings suggest that CALN1 methylation percentage could be a useful molecular biomarker for bladder cancer. Supplementary Information The online version contains supplementary material available at 10.1186/s12894-022-01136-y. | CALN1 hypomethylation as a biomarker for high-risk bladder cancer
DNA methylation in cancer is considered a diagnostic and predictive biomarker. We investigated the usefulness of the methylation status of CALN1 as a biomarker for bladder cancer using methylation-sensitive restriction enzyme (MSRE)-quantitative polymerase chain reaction (qPCR).
Eighty-two bladder cancer fresh samples were collected via transurethral resection of bladder tumors. Genomic DNA was extracted from the samples, and MSRE-qPCR was performed to determine the CALN1 methylation percentage. Reverse transcription-qPCR was performed to assess the correlation between CALN1 methylation and mRNA expression. The association between CALN1 methylation percentage and clinicopathological variables of all cases and intravesical recurrence of non-muscle-invasive bladder cancer (non-MIBC) cases were analyzed.
Of the 82 patients, nine had MIBC and 71 had non-MIBC who had not undergone total cystectomy. The median CALN1 methylation percentage was 79.5% (interquartile range: 51.1–92.6%). The CALN1 methylation percentage had a negative relationship with CALN1 mRNA expression (Spearman’s ρ = − 0.563 and P = 0.012). Hypomethylation of CALN1 was associated with advanced tumor stage (P = 0.0007) and histologically high grade (P = 0.018). Furthermore, multivariate analysis revealed that CALN1 hypomethylation was an independent risk factor for intravesical recurrence in non-MIBC patients (hazard ratio 3.83, 95% confidence interval; 1.14–13.0, P = 0.031).
Our findings suggest that CALN1 methylation percentage could be a useful molecular biomarker for bladder cancer.
The online version contains supplementary material available at 10.1186/s12894-022-01136-y.
Bladder cancer is common worldwide. According to the GLOBOCAN 2018 estimates of cancer incidence and mortality, there were 549,000 new cases of bladder cancer and 200,000 associated deaths worldwide [1]. Generally, the 5-year survival rate of patients with non-muscle-invasive bladder cancer (non-MIBC) is 96%. However, if the patients have distant metastasis, the 5-year survival rate is 6% [2]. Even though non-MIBC has a relatively good prognosis, 31–78% patients with non-MIBC show recurrence and 1–45% patients show progression to MIBC within 5 years of diagnosis [3]. Cystoscopy is the most effective technique for diagnosing bladder cancer recurrence but is highly invasive. Urine cytopathology is currently widely used for diagnosis, but its sensitivity for detecting bladder cancer is low and reportedly depends on the skill of the cytopathologist [4]. Although other methods, using several biomarkers and nucleic acid probes such as bladder tumor antigen [5], nuclear matrix protein 22 [6], and UroVysion™ fluorescence in situ hybridization [7], have been developed, the robustness of these methods for the early detection of bladder cancer and risk stratification in clinical practice has not been established. Thus, there is an urgent need to establish new biomarkers. DNA methylation is one of the epigenetic mechanisms that regulate gene expression without changing the base sequence. In recent years, DNA methylation status in bladder cancer has been widely studied [8]. Inactivation of gene expression due to promoter methylation could be a useful biomarker for bladder cancer [9–11]. We previously conducted a preliminary experiment focused on calnuelon 1 (CALN1), using the Ion Ampliseq™ Methylation Panel for Cancer Research, and found that CALN1 is associated with the clinicopathological features of bladder cancer (unpublished data). CALN1 encodes a protein that is highly similar to the calcium-binding proteins of the calmodulin family [12]. Calcium signaling is an important regulator in various cellular processes and has been implicated in important activities related to cancer progression, such as proliferation and infiltration [13, 14]. We hypothesized that the regulation of calcium signal transduction through methylation of CALN1 is involved in the development and progression of bladder cancer. In this study, we investigated the usefulness of determining CALN1 methylation status as a biomarker for bladder cancer.
Eighty-two patients who underwent transurethral resection of bladder tumor (TURBT) between April 2019 and June 2021 at Diyukai Daiichi Hospital were enrolled in this study. Data on age; sex; presence or absence of hematuria at diagnosis; smoking status; Brinkman index; and tumor stage, grade, number, size, and type (primary/recurrent) were collected. The study was performed following approval from the Ethics Committee of the Shakai Iryo Hojin Daiyukai (approval no.2,019,002) and was conducted in accordance with the Declaration of Helsinki.
Tissues collected from the patients were washed with saline and stored immediately at − 80 °C. Genomic DNA was extracted using the High Pure PCR Template Preparation Kit (Roche Molecular Systems, Pleasanton, CA, USA) according to the instruction manual, and the eluate (100 µL of elution buffer) was used for further analysis.
The isolated DNA (100 ng gDNA) was treated with Hap II (Takara Bio, Shiga, Japan), a methylation-sensitive restriction enzyme, and/or Msp I (Takara Bio), a methylation-independent restriction enzyme, according to the manufacturer’s instruction. Hap II and Msp I are isoschizomers of each other. Hap II does not cleave the methylated recognition sequence, whereas Msp I cleaves regardless of the methylation status.
Following enzymatic treatment, a quantitative DNA methylation analysis was performed using qPCR. Primers were designed using the intron 2 sequence of CALN1 with the GenBank accession number NC_000007.14 (Fig. 1). The reaction was carried out in the format of a hydrolyzed probe using the following primers and probe: forward: 5′-TCACTCAGTGTTGAGCCACAG-3′, reverse: 5′-TCCTGTGTTGGGTAGAAGTGG-3′; Universal Probe Library Probes Number 20 (Roche Molecular Systems). Using a 4 µL restriction enzyme-treated gDNA solution, each primer and probe were added to 10 µL of Essential Probe Master Mix (Roche Molecular Systems) at 0.4 µM, and analysis was performed in a total volume of 20 µL. The cycling conditions included initial denaturation at 95 °C for 10 min, followed by cycles of 95 °C for 10 s, 4.4 °C/s, 60 °C for 30 s, 2.2 °C/s annealing. PCR was performed using the LightCycler 96 and data were analyzed using the LightCycler 96 software 1.1 (Roche Molecular Systems). The methylation percentage was calculated using the formula shown in Fig. 2. gDNA extracted from the T24 cell line was used as the unmethylated control (UMcontrol), and EpiScope Methylated HeLa cell gDNA (Takara Bio) was used as the methylated control (Mcontrol). The nucleic acid extraction solution was adjusted to concentrations of 0, 6.25, 12.5, 25, 50, and 100% and the reaction of the measurement system was confirmed. The methylation percentage was determined from the Cp value of each sample.
To investigate the correlation between CALN1 methylation and mRNA expression, we performed an RT-qPCR-based assessment for the objective quantification of CALN1 mRNA levels. Of the 82 patients, 19 who were quantitatively and qualitatively suitable for assays were used in this analysis. RNA was extracted from fresh frozen TURBT tissue using the High Pure RNA Isolation Kit (Roche Molecular Systems) according to the manufacturer’s instructions. cDNA synthesis was performed under the following reaction conditions: 25 °C for 10 min, 55 °C for 60 min, and 85 °C for 5 min. The reaction product was diluted 5-fold with TE buffer and used for subsequent reactions. Primer sequences for CALN1 and the internal reference gene, glyceraldehyde 3-phosphate dehydrogenase (GAPDH), are shown in Table 1. RT-qPCR was carried out using the LightCycler 96 (Roche Molecular Systems), and the average value of duplicate measurements was determined using the LightCycler 96 software 1.1 (Roche Molecular Systems). The comparative C(T) method in relation to GAPDH was used for CALN1 expression analysis, and the correlation between CALN1 expression and the methylation percentage was analyzed.
In our institute, cystoscopy is performed every three months after TURBT for the first two years, then every six months until five years. Intravesical recurrence of bladder cancer was defined as a tumor identified by cystoscopy and confirmed by pathological diagnosis. Intravesical BCG therapy after TURBT was performed at the discretion of the attending physician. Follow-up was conducted in November 2021. The time point of entry was defined as the date when TURBT data were obtained. The primary endpoint was the intravesical recurrence of bladder cancer.
Because the variables were non-normally distributed, they are expressed as median and interquartile range. Differences between groups were assessed using Mann–Whitney U test. Fisher’s exact test was used to analyze categorical variables. We used Cox proportional hazards regression to examine the predictive value of CALN1 methylation percentage for intravesical recurrence in patients with non-MIBC. The covariates included CALN1 methylation percentage, age, sex, BCG therapy, tumor stage, tumor grade, tumor number, tumor size, and sample type (primary/recurrent). Baseline variables (P < 0.05) in the univariate analysis were included in the multivariate models. A receiver operating characteristic (ROC) curve was generated, and the area under the curve was calculated to determine the appropriate cut-off level of CALN1 methylation percentage to maximize the predictive power for intravesical recurrence-free survival of patients with non-MIBC. The methylation percentage was grouped into low and high based on the cut-off value confirmed using the ROC curve analysis. Kaplan–Meier curves of estimated intravesical recurrence-free survival were generated, and comparisons between the groups were performed using a 2-sided log-rank test. To assess whether the accuracy of predicting intravesical recurrence would improve after the addition of CALN1 methylation percentage to established risk factors, including tumor stage, grade, number, size, and sample type, we calculated the C-index, net reclassification improvement, and integrated discrimination improvement. The CALN1 methylation percentage and value of mRNA expression were not normally distributed (assessed using the Shapiro–Wilk test); therefore, non-parametric correlation coefficients (Spearman’s [ρ]) were used to determine the association between CALN1 methylation percentage and mRNA expression. Statistical significance was set at P < 0.05 and all statistical tests were two-sided. Statistical analyses were performed using the R software version 4.0.3.
In total, 82 patients (MIBC, n = 9; non–MIBC, n = 73) were enrolled in this study. During the follow-up period (median, 11.5 months), 13 of the 82 patietns died. Twenty-five of the remaining 71 patients whose bladders were preserved showed intravesical recurrence within 1 year. Of these 25 patients, 6 patients died. Three of the six patients died of bladder cancer. The median CALN1 methylation percentage was 79.5% (interquartile range: 51.1–92.6). In the univariate Cox proportional hazards analysis, the CALN1 methylation percentage was a significant predictor of intravesical recurrence (hazard ratio (HR) 0.98, 95% confidence interval (CI) 0.97–1.00, P = 0.0010). After adjusting for other confounders, the CALN1 methylation percentage was an independent predictor of intravesical recurrence (HR 0.98, 95% CI 0.97–1.00, P = 0.018). The ROC analysis was performed to maximize the predictive power of CALN1 methylation percentage for intravesical recurrence, and an 87% cut-off value was obtained (area under the curve = 0.711). Based on the cut-off value, 82 patients fit into two groups. Fifty-one (62%) patients fit in the low group with a methylation percentage of less than 87%, and 31 (38%) fit in the high group with a methylation percentage greater than 87%. Patient characteristics according to CALN1 methylation percentage are shown in Table 2 (Additional file 1). Of the 82 patients, 73 had non-MIBC and 9 had MIBC. Total cystectomy was performed in two patients with non-MIBC that was difficult to cure by TURBT during the follow-up period. Patients in the low group were significantly older than those in the high group and had a higher proportion of females and non-smokers. In addition, the low group tended to have significantly advanced tumor stages and more histologically high-grade tumors than the high group. To identify the association between the CALN1 methylation percentage and intravesical recurrence, a Kaplan–Meier analysis was performed in 71 patients with non-MIBC whose bladders were preserved. There was a significant difference between the groups in terms of intravesical recurrence-free survival (P = 0.0084). At the one-year follow-up, the Kaplan–Meier survival rates of patients with intravesical recurrence were 48.2% and 86.3% in the low and high groups, respectively (Fig. 3). The results of the univariate and multivariate Cox regression analyses to explore the prognostic factors of intravesical recurrence are shown in Table 3. A low CALN1 methylation percentage remained an independent prognostic factor after adjusting for tumor size in the multivariate analysis. The C-index increased, but did not reach statistical significance (0.744, P = 0.27). However, the net reclassification improvement and integrated discrimination improvement for the intravesical recurrence rate significantly improved after adding the CALN1 methylation percentage to the baseline model with established risk factors (0.57 and 0.07, P = 0.021 and P = 0.025, respectively, Table 4). In the analysis of the correlation between CALN1 methylation and the mRNA expression level, a significant negative correlation was observed (Fig. 4, Additional file 2).
We analyzed the relationship between the CALN1 methylation percentage and clinicopathological data of patients with bladder cancer. We found that CALN1 hypomethylation was significantly associated with advanced tumor stage, more histologically higher-grade tumors, and an increased risk of intravesical recurrence. To the best of our knowledge, this is the first study to show that CALN1 methylation percentage is associated with the clinicopathological features and prognosis of bladder cancer. The association between DNA methylation and various biological phenomena such as carcinogenesis have been identified [15, 16]. Methylation analysis could provide information that cannot be obtained using conventional tests, such as prediction of drug sensitivity or prognosis [17, 18]. Cao et al. used microarray analysis to show that calcium signal transduction was associated with the development of bladder cancer via the mitogen-activated protein kinase pathway [19]. In addition, intron 2 of CALN1 is a DNase I hypersensitive site that is strongly associated with transcriptional activity [20]. Therefore, we suspected that CALN1 methylation was involved in the action of a DNase I hypersensitive site and, as a result, may affect the expression of CALN1. Regarding the relationship between bladder cancer and methylation, various analytical reports have centered on CpG sites [21, 22], and testing systems such as Bladder EpiCheck [23] have been established. Although various trials have been conducted regarding the diagnosis and treatment of bladder cancer, methylation analysis of CALN1 and its association with bladder cancer has not been probed before. Bisulfite sequencing is widely used for methylation analyses. In this study, we performed methylation analysis using methylation-sensitive restriction enzyme (MSRE)-qPCR. This technique enables the analysis of a small amount of sample obtained by TURBT without bisulfite treatment. Bisulfite treatment involves the process of incubating the DNA solution at 50–70 ℃. There is a problem that the yield of DNA is extremely low because the DNA is cleaved during the heating process. Recently, high-yield methods have been developed, but DNA fragmentation has not been avoided completely [24]. In addition, because bisulfite sequencing requires a large number of cells, it is not feasible for clinical specimens with a low amount of DNA such as cell-free DNA or circulating tumor cells. In contrast, one of the advantages of MSRE-qPCR is the side-by-side comparison between control and experimental samples, even for very low amounts of DNA. In addition, MSRE-qPCR can be completed in less time than other methods with the same level of accuracy [25]. Comprehensive analysis using next-generation sequencing is also useful but less practical owing to high costs. MSRE-qPCR is useful for targeted analysis owing to its simple workflow. Further investigation exploring this diagnostic method with high sensitivity and specificity in combination with other diagnostic markers is necessary and will contribute to the development of new diagnostic systems for bladder cancer. The current study has some limitations. First, there were no criteria for intravesical BCG immunotherapy, though there was no difference in BCG therapy between the low- and high-methylation groups. Second, required sample size was not calculated before the study. However, based on the results obtained, the required sample size for comparison of the survival curves between the groups was calculated to be 20 patients per group. The sample size of this study was sufficient to meet this requirement. Nevertheless, the sample size was small and the follow-up period was short. Therefore, the findings of this study need to be validated in a larger study.
We performed methylation analysis of intron 2 of CALN1 using gDNA extracted from samples collected by TURBT. We found that low CALN1 methylation percentage is consistent with the occurrence of advanced tumor stages, high-grade tumors, and higher intravesical recurrence rates. Therefore, we suggest that CALN1 methylation percentage may be an indicator of high-risk bladder cancer and could be considered a useful biomarker for accurately predicting intravesical recurrence of non-MIBC.
Additional file 1. Demographicaland clinical information of the subjectsparticipating in this study.Additional file 2. Raw dataof relationship between CALN1 mRNA expression and CALN1 methylation percentage. |
|
PMC9647955 | Haizhen Ma,Panpan Li,Ning Xiao,Tao Xia | Poly-γ-glutamic acid promoted maize root development by affecting auxin signaling pathway and the abundance and diversity of rhizosphere microbial community | 10-11-2022 | Maize,γ-PGA,Auxin signaling,Roots development,Plant growth promoting bacteria | Background The root systems of higher plants play an important role in plant growth and development. In our present study, it was found that poly-γ-glutamic acid (γ-PGA), an environmentally friendly biomacromolecule, significantly improved root development in maize. Results After treatment with γ-PGA for 7 days, the fresh weight of maize roots was significantly increased and the differences between γ-PGA treated group and control group were mainly caused by the number (higher by 71.87% compared to the control) and length of lateral roots. RNAseq and RT-PCR analyses showed that γ-PGA treatment upregulated the expression of genes related to the synthesis of auxins and auxin signal in maize roots. In addition, γ-PGA promoted the accumulation of plant growth-promoting bacteria, such as Azospirillum, Azohydromonas, Ramlibacter, and Sphingobium (Proteobacteria), Streptomyces (Actinobacteria), Parasegetibacter (Bacteroidetes), and Gemmatimonas (Gemmatimonadetes) in rhizosphere soil and the secretion of auxins. The results of this study deepened our understanding of the effects and mechanism of γ-PGA on maize root development, and as well as highlighted the possibility of using γ-PGA to improve crop growth and soil environment. Conclusions γ-PGA promotes early growth and development of maize roots by inducing the secretion and accumulation of auxin in roots and in rhizosphere soil, and increasing the abundance of plant growth promoting bacteria. Supplementary Information The online version contains supplementary material available at 10.1186/s12870-022-03908-y. | Poly-γ-glutamic acid promoted maize root development by affecting auxin signaling pathway and the abundance and diversity of rhizosphere microbial community
The root systems of higher plants play an important role in plant growth and development. In our present study, it was found that poly-γ-glutamic acid (γ-PGA), an environmentally friendly biomacromolecule, significantly improved root development in maize.
After treatment with γ-PGA for 7 days, the fresh weight of maize roots was significantly increased and the differences between γ-PGA treated group and control group were mainly caused by the number (higher by 71.87% compared to the control) and length of lateral roots. RNAseq and RT-PCR analyses showed that γ-PGA treatment upregulated the expression of genes related to the synthesis of auxins and auxin signal in maize roots. In addition, γ-PGA promoted the accumulation of plant growth-promoting bacteria, such as Azospirillum, Azohydromonas, Ramlibacter, and Sphingobium (Proteobacteria), Streptomyces (Actinobacteria), Parasegetibacter (Bacteroidetes), and Gemmatimonas (Gemmatimonadetes) in rhizosphere soil and the secretion of auxins. The results of this study deepened our understanding of the effects and mechanism of γ-PGA on maize root development, and as well as highlighted the possibility of using γ-PGA to improve crop growth and soil environment.
γ-PGA promotes early growth and development of maize roots by inducing the secretion and accumulation of auxin in roots and in rhizosphere soil, and increasing the abundance of plant growth promoting bacteria.
The online version contains supplementary material available at 10.1186/s12870-022-03908-y.
Maize (Zea mays L.) is an important crop, which is used as animal feed, human food and bioethanol production. The complex root system of maize facilitates the uptake of water and nutrients and the anchorage of maize in the soil, and deeply affect the growth and development of maize [1]. The root system of maize can be divided into embryonic root system and post-embryonic root system. Embryonic root system includes a single primary root and a variable number of seminal roots. Post-embryonic root system contains crown roots, brace roots and lateral roots emerged from all major root types [2]. In all maize root types, lateral roots are initiated from phloem pole pericycle and endodermis cells [3]. Root architecture encompasses the density of lateral roots (LR) and the root branching pattern, which comprise the lateral and adventitious roots. In maize, root branching is an important aspect for root structure development. Well-branched roots increase the surface area for the absorption of water and nutrients. Adventitious roots, including aerial roots and root cap formed by underground nodes, anchor the plant and facilitate the absorption of water and mineral elements at the mature stage of maize. The development of lateral roots can start from the columnar sheath cells of the primary roots, the seminal roots, and the underground nodes of roots [4, 5]. The formation of lateral roots is a critical element of the root system because it affects the absorption of deeper water and nutrients. Lateral roots arise from the xylem pole pericycle cells, which undergo a series of divisions to form LR primordium before developing into lateral roots [6]. Auxins regulate the lateral and vascular root development and the architecture of roots [7–9], which mediates a variety of physiological processes and has long been known to promote lateral root formation [10]. Auxin synthesis, transport, and signaling pathways are triggered during the formation of lateral root (LR) [11]. Auxins are primarily synthesized via the Indole-3-pyruvic acid (IPA) pathway. In this pathway, tryptophan (TRP) is firstly converted to IPA by transaminases of the TAA family, before IPA is converted to IAA by YUC enzyme. YUC are enzymes usually located on the membrane of endoplasmic reticulum [12]. Synthesis/function of auxins is mainly regulated by the Transport Inhibitor Response 1 (TIR1) protein, members of the Auxin F-Box (AFBs) family, AUXIN/INDOLE-3-ACETIC ACID (Aux/IAA) proteins, and Auxin Response Factor (ARF) proteins [13, 14]. The TIR1 is an auxin receptor and also part of the E3 ubiquitin ligase complex SCF (TIR1). At low intracellular concentrations of auxins, the transcription of the Aux/IAA proteins was repressed by interacting with ARF proteins [15]. High levels of auxin promote the binding of Aux/IAA proteins to auxin receptor protein TIR1, inducing the ubiquitination and degradation by the 26S proteasome [16–18]. ARF proteins are then released from Aux/IAA complex, and their activation promotes the transcription of auxin-responsive target genes [19]. Different TIR1/AFB-Aux/IAA combinations may induce different transcriptional responses [8]. Auxin promotes LR development through SLR4/ARF7-ARF9 signal module [20]. The rum1 gene encodes an Aux/IAA protein ZmIAA10 which was required for the initiation of embryonic seminal and post-embryonic lateral root initiation in primary roots of maize [21]. RUM1 could interacts with the transcriptional activators ZmARF25 and ZmARF34. The mutated rum1 protein cannot interact with SCFTIR1 E3 ubiquitin–ligase complexes which prevents its ubiquitin-mediated proteasomal degradation and resulting in the constitutive repression of downstream gene expression [3]. Poly-γ-glutamic acid (γ-PGA) is a nontoxic, water-soluble, biodegradable and environmentally friendly biopolymer, composed of D/L-glutamic acid monomers and produced through Bacillus subtilis-mediated fermentation [22, 23]. γ-PGA can be applied in food, medicine, cosmetics, and agricultural fields based on its variable molecular weight [24]. Recent studies have shown that γ-PGA plays an important role in plant growth and development and, thus, is a promising supplement in agricultural fertilizers. γ-PGA significantly increases the yield of several crops, including cucumber, Chinese cabbage, wheat, and rapeseed [25, 26]. These studies made an effort to reveal the promotional effect of γ-PGA on plant growth from the perspective of plant nitrogen metabolism [27]. The root is an integral tissue of the plant and plays the most important role in the absorption and utilization of nutrients [28]. A few studies have found that γ-PGA could improve root biomass [27]. However, how γ-PGA promotes root growth remains unclear. Herein, we investigated the effect of γ-PGA on the synthesis and signal transduction of auxins in maize roots and the rhizosphere soil microbial community, in order to comprehensively understand the effect and mechanism of γ-PGA promoting maize root development.
In order to analyze the effect of γ-PGA on the development of maize root system, the seedlings germinated after 5 days were treated with γ-PGA (Fig. 1A). Previously, the effects of different molecular weight γ-PGA on maize growth and drought resistance, were studied and found that low molecular weight γ-PGA could promote the growth and drought resistance of maize (Fig. S1). Therefore, γ-PGA of 10 kDa was used to study its effects on maize root development in this study. The results showed that there were obvious lateral roots on the primary root after γ-PGA treatment for 1 day (Fig. 1A). After γ-PGA treatment for 7 days, the number and length of roots of different types of maize were counted (Fig. 1B-I). The number of seminal roots and crown roots did not significantly change between γ-PGA treated group and control group (Fig. 1C, D), but the number of lateral roots in γ-PGA treated group was significantly higher than that of the control group by 71.87% (Fig. 1E). The total root length of γ-PGA treated group was 72.58% higher than that of the control group, the differences between γ-PGA treated group and control group were mainly caused by the number and length of lateral roots (Fig. 1H). In addition, the length of seminal roots was 24.19% longer than that of the control group, and the length of crown roots was 210% longer than that of control group (Fig. 1F, G). We also treated the roots used the same concentration of L-Glutamate (L-Glu) (Fig. 1A). The results showed that the primary root length of maize after L-Glu treatment decreased significantly (35.84% less than that of the control group and 65.22% less than that of γ-PGA treated group) (Fig. 1B). However, after L-Glu treatment, the number of seminal roots increased (29.80% more than the control group and slightly higher than the γ-PGA treated group, but the difference was not significant) after L-Glu treatment (Fig. 1C). The number of lateral roots of L-Glu treatment group was 51.08% more than that in control group, but 12.10% less than that in γ-PGA group (Fig. 1E). However, the length of seminal roots in L-Glu treatment group was 60.86% longer than that of the control group and 29.52% longer than that of γ-PGA treatment (Fig. 1F). The total root length of L-Glu group was 51.73% more than that of control group, but 12.08% less than that of γ-PGA group (Fig. 1H). Finally, the fresh weight of maize roots after 7 days‘treatment was calculated. The results showed that the fresh weight of maize roots after γ-PGA treatment was also significantly increased (Fig. 1I). Therefore, the promotion of the root development after γ-PGA treatment was caused principally by the number and length of lateral roots.
It’s reported that auxin plays an important role in the development of plant roots. In order to explain the mechanism of γ-PGA promoting maize root development, the contents of auxin in maize roots after 0 h, 2 h, 6 h, 12 h, 24 h, 48 h, and 96 h of γ-PGA treatment were measured. The results showed that γ-PGA could induce the synthesis and accumulation of auxin (increased by 12.51%) in roots just after 2 h of γ-PGA treatment. After 24 h and 48 h of γ-PGA treatment, the auxin content in maize roots increased by 77.09 and 51.70% (Fig. 2). The results indicated that exogenous γ-PGA could induce the accumulation of auxin in maize roots.
The results of RNAseq of roots treated with and without γ-PGA performed previously revealed that γ-PGA upregulated the expression of genes related to the synthesis of auxins (Supplementary Fig. S2). In tryptophan-dependent IPA auxin synthesis pathway, γ-PGA treatment upregulated the expression of 7 TAA1-encoding genes in maize roots. γ-PGA treatment dysregulated the expression of 9 genes encoding YUCCA protein, in which 7 were over-expressed and 2 were under-expressed. Also, γ-PGA treatment dysregulated the expression of 12 DEG genes in tryptamine synthesis pathway, in which 11 were upregulated, whereas only one gene was down-regulated. γ-PGA treatment also dysregulated the expression of 6 genes in indole-3-acetamide pathway, in which 3 were down-regulated, 3 were upregulated (Supplementary Fig. S2). Furthermore, we also analyzed the expression of these genes in maize roots of γ-PGA treatment for 0 h, 2 h, 6 h, 12 h, 24 h, 48 h and 96 h (Figs. 3 and 4). The results showed that the expression of these genes was induced by γ-PGA. In addition, 6 genes for ARF, 1 gene for TIR1, and 1 gene for SAUR71 were also induced by γ-PGA in auxin signaling pathway (Fig. 4). Overall, these results suggested that γ-PGA promoted root growth via the auxin synthesis and auxin signaling pathways (Figs. 3 and 4).
Since maize rhizosphere soil was in close contact with the roots, the urease, which is closely related to soil nitrogen transformation, and IAA contents of maize rhizosphere soil, was also detected. It was found that γ-PGA treatments increased the urease activity in the rhizosphere of maize by 41.12%, whereas the IAA contents in the soil after γ-PGA treatments increased by 16.83% (Table 1).
The effect of γ-PGA on the diversity and abundance of the bacterial community in the maize rhizosphere was analyzed using high-throughput sequencing of the 16S rDNA region. NMDS (stress = 0.000108) of the Bray-Curtis UniFrac distance ordinations were also performed (Supplementary Fig. S3A), and the results indicated that the bacterial community composition of rhizospheric soil with γ-PGA application shifts compared with that of the soil without γ-PGA. The communities in maize rhizospheric soil with γ-PGA were grouped together and significantly separated from those in soil without γ-PGA. The obtained high-quality sequences belonged to 36 phylum，among which the main phylum was Proteobacteria, followed by Actinobacteria, Chloroflexi, Bacteroidetes and Cyanobacteria. Although the abundance of bacterial community changed after the addition of γ-PGA，the predominant phylum were similar. There was no difference in species composition among the γ-PGA-treated and non-treated groups; Compared to the control, the relative abundance of Proteobacteria, Acidobacteria, Cyanobacteria and Gemmatimonadetes were higher in soil added γ-PGA (Fig. 5A). The LEfSe analysis (LDA ≥ 3.35) had been used to obtain the species with the most significant variation (Supplementary Fig. S3B). At phylum level, γ-PGA treatment increased the abundance of Proteobacteria, Acidobacteria, Cyanobacteria and Gemmatimonadetes, Deltaproteobacteria and Gammaproteobacteria at class level (Fig. S2B). At genus level, γ-PGA treatment increased the abundance of Azospirillum, Azohydromonas, Ramlibacter and Sphingobium (Proteobacteria), Subgroup_7, Mycobacterium, Kribbella and Streptomyces (Actinobacteria), Flavisolibacter and Parasegetibacter (Bacteroidetes), Gemmatimonas (Gemmatimonadetes), and Microcoleus_Es-Yyy1400 (Cyanobacteria) in the maize rhizosphere (Fig. 5B).
Our study showed that γ-PGA significantly improved the growth of maize, which is consistent with previous findings. In this project, we mainly focused on the effect of γ-PGA on root development and found that γ-PGA can promote the growth of maize lateral roots. At the same time, γ-PGA significantly increased the biomass of maize roots, promoted the growth of roots, enhanced the ability of maize to absorb nutrients and promoted the growth of plants. In order to further explain the mechanism of γ-PGA promoting root growth, we also determined the auxin content and the expression of the auxin related genes in maize roots. The results showed that exogenous γ-PGA could induce the secretion and accumulation of auxin in maize root, which had not been reported in the literature. Studies have shown that L-Glutamate (L-Glu) can not only promote plant growth as a nitrogen nutrient, but also serve as a plant signal molecule. Glutamate can be involved in calcium signaling and root development. For example, L-Glu can inhibit the growth of primary roots of Arabidopsis thaliana and stimulate the growth of lateral roots near the primary root tip. Other amino acids, such as Gln and D-Glu, have no similar effect [29]. In addition, MEKK1 is involved in glutamate signaling pathway, which is involved in inducing changes in the root structure of Arabidopsis [30]. Exogenous L-Glu can not only support the growth of rice seedlings as nitrogen, but also induce the expression of a series of genes to regulate the development of rice roots [31]. Since γ-PGA is easily degraded into L-Glu and D-Glu, in order to prove whether the promotion of maize root development is the effect of exogenous γ-PGA or its degradation product L-Glu, we treated maize roots with the same concentration of L-Glu and compared it with γ-PGA. The results showed that both γ-PGA and L-Glu could change the development of maize roots, but the changes of different types of maize roots were different. The effect of L-Glu on maize root was not as good as γ-PGA. Many studies have shown that γ-PGA can significantly promote the growth of plant roots. γ-PGA treatment can improve the physical and chemical properties of soil, increase soil microbial biomass and microbial activity, and promote the absorption of nutrients by plants. The results showed that although both treatments increased root biomass, the effects on different types of maize roots were different. Previous studies have shown that γ-PGA may act as an N source to promote plant growth, and its mechanism of promoting root development may be related to L-Glu. In this study, maize roots were directly treated with the same concentration of γ-PGA and L-Glu respectively, and the culture medium was changed every day to prevent the degradation of γ-PGA as much as possible. The results showed that the difference between γ-PGA treatment group and control group was mainly caused by the number and length of lateral roots. However, the root length of L-Glu treatment decreased significantly and the number of seed roots became variable, while the number of lateral roots also increased, but the increase range was less than that of γ-PGA treatment. The total root length of L-Glu treatment group was 51.73% higher than that of control group, but 12.08% lower than that of γ-PGA treatment group. The results indicated that the mechanism of γ-PGA promoting maize root development may be different from that of L-Glu.
As sessile, plants rely on hormones to coordinate different stages of development and increase the plasticity of plant development in a changing environment. For example, under abiotic stresses such as drought or nutrient deficiencies, plants will induce root elongation and root structure changes, so as to penetrate deeper into the soil and obtain nutrients and water. Plant hormones play an important role in the regulation of root structure [32]. Among them, auxin regulates the division and differentiation of root meristem cells, the growth of primary roots and the development of lateral roots. Plants absorb water and nutrients through roots, which support the growth and development of plants. Thus, the change in root morphology deeply affects the growth and development of plants. Auxins regulate plant lateral root and vascular development, thus regulating root architecture [7–9]. Indole-3-pyruvic acid (IPA) pathway is the main pathway for synthesizing auxins. In this pathway, tryptophan (TRP) is firstly converted to IPA, catalyzed by transaminase belonging to TAA family. IPA is then converted to IAA by YUC enzyme. YUC is a family of enzymes usually attached to the membrane of endoplasmic reticulum [12]. Studies have confirmed the presence of YUC protein in plant root tissue [33]. In this study, it was found that 7 DEGs encoding TAA protein were upregulated, 9 genes encoding YUC protein dysregulated in maize roots with γ-PGA treatment. Of these, the expression of 7 DEGs was upregulated, 2 DEGs was downregulated. In addition，γ-PGA treatment upregulated the expression of 11 genes related to the tryptamine (auxin) synthesis pathway and of 3 genes that regulate the activities of Indole-3-acetamide auxin pathway, which are all related to the synthesis of auxins (Fig. 3). Many studies have shown that auxin plays an important role in lateral roots (LR) formation, especially in LR initiation and primordial development [34, 35]. Auxin signaling pathway, mediated by the Aux/IAA and ARF transcription factors, is required during LR initiation process [4]. During this process, auxin signals are transmitted to peripheral cells to promote the degradation of Aux/IAAs (such as IAA14/SLR). The degradation of Aux/IAAs initiates the development of lateral roots through the SCFTIR1/AFBs complex and 26S proteasome. This activates the ARF7/ARF19 pathway, which promotes the expression of target genes such as LBD16/ASL18 and LBD29/ASL16 that initiates the development of LR [36]. In the present study, we also found that γ-PGA treatment induced the expression of 6 ARF, 1 TIR1, and 1 SAUR71 genes, all of which were related to the synthesis of auxins (Fig. 4). The results showed that γ-PGA treatment significantly up-regulated the expression of genes involved in auxin synthesis and signaling pathways within a few hours, and the increased auxin concentration rapidly induced the initiation of lateral roots [37], but auxin was always in homeostasis in plants. Plants maintain auxin homeostasis by coordinating the biosynthesis, transport, inactivation of auxin [38]. Therefore, γ-PGA may constantly induce the accumulation of auxin, and the up-regulated auxin concentration must induce other unknown pathways that negatively regulate auxin biosynthesis to keep the auxin homeostasis, which may be the reason for the decrease of auxin concentration after 96 h of γ-PGA treatment.
Plant development affects the interaction of rhizosphere microbial communities [39]. Metabolites secreted by plant roots will alter the soil microbial composition. Primary and secondary metabolites produced and exuded by plants can selectively promote or inhibit specific microbial communities [40, 41]. Conversely, microbial activities also influence plant development and response to environmental factors [42, 43]. Soil microorganisms can stimulate the growth of lateral roots and root hairs, and further improve water and nutrient uptake. In addition, soil microorganisms can also promote plant growth and regeneration. Soil microbes can affect both internal and external processes that supporting plant growth and development. Under stress conditions, well-developed root structures facilitate the recruitment of beneficial microorganisms from microbiome-rich topsoil. In turn, microorganisms can produce or alter phytohormone levels in the rhizosphere or plants, thereby influencing plant development and stress responses [44, 45]. Numerous hormones (IAA, ABA, CKs, Gas, and ET) have been isolated from the growth medium of soil microorganisms. Hormone producing microorganisms, such as plant growth-promoting bacteria (PGPBs) and plant growth-promoting fungi (PGPFs)，are usually non-pathogenic and even beneficial to plants [46]. Studies have shown that beneficial microorganisms can produce hormones or activate the synthesis of plant hormones, which changes the structure of plant roots to provide a habitat for the microorganisms. For example, PGPBs affect the division and differentiation of root cells, thus changing the root structure [47]. Bacillus megaterium promotes the development of root structure through plant CK signaling pathway, which is independent of ET and auxin pathways [48]. In this study, we found that γ-PGA increases the synthesis of IAA (auxin) in maize rhizosphere. PGPBs synthesize IAA in maize rhizosphere. Although γ-PGA had no effect on the composition of dominant bacteria species, it strongly influenced their relative abundance. γ-PGA increased the relative abundance of Proteobacteria at phylum level. And previous studies have shown that some of Proteobacteria were participated in nitrogen fixation [49]. At the genus level, exogenous γ-PGA significantly increased the abundance of Azospirillum, Azohydromonas, Ramlibacter, Sphingobium of Proteobacteria, Streptomyces of Actinobacteria, Parasegetibacter of Bacteroidetes and Gemmatimonas of Gemmatimonadetes, most of which are PGPBs [50–53]. Azospirillum, Azohydromonas and Sphingobium are nitrogen-fixing bacteria, which can promote nitrogen absorption and plant growth [51, 54]. It was also reported that Azospirillum Brasilense, a PGPR of Azospirillum, could secretes IAA, nitric oxide, carotenoid and several cell surface components that promote plant growth [55, 56]. Streptomyces enhance plant growth and inhibits the growth of phytopathogens [57, 58]. Parasegetibacter, also as a PGPR, had important potential in promoting plant growth [59]. In addition, we found that γ-PGA treatment increased the urease activity in maize rhizosphere soils (Fig. 1H). Urease in the soil catalyzes the conversion of nitrogen to NH3, NH4+ and CO32-, participates in nitrogen transformation by catalyzing the conversion of nitrogen to NH3, NH4+, and CO32− through urea hydrolysis and thus providing nutrients for the plants. These findings imply that γ-PGA altered the microbial diversity in soil rhizosphere and increased the abundance of bacteria promoting root development. However, the effects of γ-PGA treatment on maize rhizosphere microbial community under normal growth conditions and drought conditions are different (Fig. S4). γ-PGA treatment mainly enriched some drought-resistant plant growth promoting bacteria, such as Actinobacteria, Chloroflexi and Cyanobacteria under drought stress [60]. Actinobacteria and Chloroflexi were reported to be the most prominent phyla under drought conditions [61]. Under normal irrigation conditions, the application of γ-PGA significantly enriched Proteobacteria, Acidobacteria, Cyanobacteria and Gemmatimonadetes, which can promote nitrogen absorption and auxin secretion, thus promoting the development of plant roots. In the rhizosphere soil without γ-PGA treatment, the amounts of microorganisms decreased significantly after drought stress, which was consistent with the previous report that drought may be the abiotic stress with the greatest impact on soil biological community, usually resulting in a significant reduction of microbial biomass [62].
Exogenous application of γ-PGA could enhance the abundance of PGPR in maize rhizosphere soil, increase the content of auxin in maize root and rhizosphere soil and promote the development of maize roots (Fig. 6). This study provides a more comprehensive understanding of the role and mechanism of exogenous application of γ-PGA in improving root development, and is conducive to its application in agriculture.
KN5585 maize seeds were sown in a soil box (10 cm*10 cm*10 cm). When the seedlings grew to 1-leaves, the seedlings were watered with solution 50 mg/L γ-PGA of different molecular weight (0, 10, 100,700, 1000, and 2000 kDa), and grown in greenhouse (28 ± 2 °C under nature light, and 25 ± 2 °C at night). At three-leaf stage, all seedlings were exposed to drought stress treatment by stopping watering to select the most suitable molecular weight of γ-PGA. KN5585 maize seeds were sterilized using 75% ethanol and germinated for 4 days in the dark at 28 °C on a moisturized filter paper in a sterile culture dish (diameter: 12.5 cm). The germinated seeds were transferred into a flask 15 cm high and 7 cm wide. Culturing was performed at 28 °C /25 °C (16 h light/8 h dark). The seedlings germinated for 6 days were treated with γ-PGA solution (10 kDa, 50 mg/L). The γ-PGA solution was replaced every day to prevent it from degrading and was aerated with a mini air pump. The number and the length of the main types of the maize roots were determined after 6 days. Finally, the plants were also examined phenotypically, and the fresh weight of the roots were also determined.
The IAA content of roots was measured at 0 h, 2 h, 6 h, 12 h, 24 h, 48 h and 96 h. About 1.0 g of each sample was rapidly frozen in liquid nitrogen and homogenized to powder. IAA was extracted and quantified according to the manufacturer’s instructions (Wuhan Metware Biotechnology Co., Ltd., Wuhan, China). IAA was quantified using LC-MS/MS system. The content of IAA was determined by external standard method and three biological replicates were carried out.
The total RNA from the maize roots after treatment with γ-PGA after 0 h, 2 h, 6 h, 12 h, 24 h, 48 h, and 96 h was extracted using the HiPure RNA Kit (Magen, Guangzhou, China). The RNA (2 μg) was reverse transcribed into cDNA using a Reverse transcription kit (TAKARA). The genes of interest and the internal control (ZmTub) were amplified using the SYBR Green I Master Mix (Roche, Indianapolis, USA) in the LightCycler 480 (Roche, USA) platform. Each gene was amplified in triplicate. The amplification conditions included initial denaturation at 95 °C for 5 min, subsequent denaturation through 40 cycles at 95 °C for 10 s, annealing at 60 °C for 10 s, and elongation at 72 °C for 20 s. The relative amplification of genes was calculated using the 2-ΔΔCT method. Differences between different groups were analyzed using ANOVA. Data were analyzed using the SPSS software, version 20. The primers used in this research are shown in Supplementary Table S1.
The determination of the auxin content in maize rhizosphere was according to the previous method [63]. The content of IAA was determined with High-performance liquid chromatography (HPLC, L-2000, Hitachi), using the Extend-C18 column size (5 μm* 4.6 mm*150 mm), the absorbance of IAA was 280 nm. The urease activity was determined as previously described by Wang [64]. And five biological replicates were performed.
Maize (KN5585) seeds were sown in a cubit soil box measuring 10 cm. After germination, the seedlings were grown in a greenhouse at 28 ± 2 °C under natural light and at 25 ± 2 °C at night. Watering was performed using γ-PGA (10 kDa, 50 mg/L) solution. After growth for 30 days, the soils tightly bound to the roots (served as rhizosphere soils) were collected and analyzed for the composition of microbial community. This experiment was performed in triplicate. Amplification and high-throughput sequencing of 16 s rDNA of soil bacteria in the maize rhizosphere were performed as described by Wang et al. [64]. The primers for the V4 region of 16 s rDNA of the bacteria were 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). High-throughput sequencing was performed using the Illumina Hiseq 2000 (Illumina Inc., San Diego, USA) platform. The differences in abundance at different taxa, including the phylum, class, order, family, and genus, between groups were analyzed using Metastats. Nonmetric multidimensional scaling (NMDS) was performed on distance matrices. The 2D graphical outputs were then drawn using the coordinates. Significance difference in the relative abundance of bacteria at specific taxa was evaluated using the LEfSe analysis.
All experiments were performed in triplicates. Continuous normally distributed data were expressed as mean ± standard deviation (SD). The difference between groups was analyzed using T-test and Duncan’s tests of one-way ANOVAs. The data were analyzed using SPSS (version 22.0.0.0). Statistical significance was set at *p < 0.05 or **p < 0.01.
Additional file 1: Fig. S1. The effect of γ-PGA of different molecular weight on the maize growth and drought resistance.Additional file 2: Fig. S2. The DEGs involved in auxin synthesis pathway. Roots from maize treated with γ-PGA was collected for RNA sequencing. The absolute values of log2 (CK+ γ-PGA/CK) ≥ 1, and FDR < 0.001 were used as the criteria for DEGs. The color of the box represented up (red) and down (green)-regulated genes.Additional file 3: Fig. S3. The NMDS and LEfSe analysis. A, Non-metric multidimensional scaling (NMDS) for the grouping patterns of microbial communities based on the bray-curtis distance. Each colored dot represented a sample. B, LEfSe analysis (LDA ≥ 3.35) for the species in the rhizosphere soil.Additional file 4: Fig. S4. The NMDS and LEfSe analysis for the species in the rhizosphere soil of the different treatment. A, Non-metric multidimensional scaling (NMDS) for the grouping patterns of microbial communities based on the bray-curtis distance. Each colored dot represented a sample. B, LEfSe analysis (LDA ≥ 3.73) for the species in the rhizosphere soil of the control maize (CK) and the maize treated with γ-PGA (CK-γ-PGA) on the normal growth condition, and the control maize (CK-D) and the maize treated with γ-PGA (CK-γ-PGA-D) after drought treatment.Additional file 5: Table S1. The sequence of primers used in this study. |
|
PMC9647956 | Xingzhao Ji,Lichao Han,Weiying Zhang,Lina Sun,Shuai Xu,Xiaotong Qiu,Shihong Fan,Zhenjun Li | Molecular, cellular and neurological consequences of infection by the neglected human pathogen Nocardia | 09-11-2022 | Nocardia farcinica,Dual RNA-seq,Virulence factor,Parkinson’s disease,Microglia | Background Nocardia is a facultative intracellular pathogen that infects the lungs and brains of immunocompromised patients with consequences that can be fatal. The incidence of such infections is rising, immunocompetent individuals are also being infected, and there is a need to learn more about this neglected bacterial pathogen and the interaction with its human host. Results We have applied dual RNA-seq to assess the global transcriptome changes that occur simultaneously in Nocardia farcinica (N. farcinica) and infected human epithelial alveolar host cells, and have tested a series of mutants in this in vitro system to identify candidate determinants of virulence. Using a mouse model, we revealed the profiles of inflammation-related factors in the lung after intranasal infection and confirmed that nbtB and nbtS are key virulence genes for Nocardia infection in vivo. Regarding the host response to infection, we found that the expression of many histones was dysregulated during the infection of lung cells, indicating that epigenetic modification might play a crucial role in the host during Nocardia infection. In our mouse model, Nocardia infection led to neurological symptoms and we found that 15 of 22 Nocardia clinical strains tested could cause obvious PD-like symptoms. Further experiments indicated that Nocardia infection could activate microglia and drive M1 microglial polarization, promote iNOS and CXCL-10 production, and cause neuroinflammation in the substantia nigra, all of which may be involved in causing PD-like symptoms. Importantly, the deletion of nbtS in N. farcinica completely attenuated the neurological symptoms. Conclusions Our data contribute to an in-depth understanding of the characteristics of both the host and Nocardia during infection and provide valuable clues for future studies of this neglected human pathogen, especially those addressing the underlying causes of infection-related neurological symptoms. Supplementary Information The online version contains supplementary material available at 10.1186/s12915-022-01452-7. | Molecular, cellular and neurological consequences of infection by the neglected human pathogen Nocardia
Nocardia is a facultative intracellular pathogen that infects the lungs and brains of immunocompromised patients with consequences that can be fatal. The incidence of such infections is rising, immunocompetent individuals are also being infected, and there is a need to learn more about this neglected bacterial pathogen and the interaction with its human host.
We have applied dual RNA-seq to assess the global transcriptome changes that occur simultaneously in Nocardia farcinica (N. farcinica) and infected human epithelial alveolar host cells, and have tested a series of mutants in this in vitro system to identify candidate determinants of virulence. Using a mouse model, we revealed the profiles of inflammation-related factors in the lung after intranasal infection and confirmed that nbtB and nbtS are key virulence genes for Nocardia infection in vivo. Regarding the host response to infection, we found that the expression of many histones was dysregulated during the infection of lung cells, indicating that epigenetic modification might play a crucial role in the host during Nocardia infection. In our mouse model, Nocardia infection led to neurological symptoms and we found that 15 of 22 Nocardia clinical strains tested could cause obvious PD-like symptoms. Further experiments indicated that Nocardia infection could activate microglia and drive M1 microglial polarization, promote iNOS and CXCL-10 production, and cause neuroinflammation in the substantia nigra, all of which may be involved in causing PD-like symptoms. Importantly, the deletion of nbtS in N. farcinica completely attenuated the neurological symptoms.
Our data contribute to an in-depth understanding of the characteristics of both the host and Nocardia during infection and provide valuable clues for future studies of this neglected human pathogen, especially those addressing the underlying causes of infection-related neurological symptoms.
The online version contains supplementary material available at 10.1186/s12915-022-01452-7.
Nocardiosis is an infectious disease caused by the ubiquitous, intracellular gram-positive aerobic bacterium Nocardia spp. Nocardiosis has been treated as an “opportunistic” disease that is mainly involved in immunocompromised patients. However, an increasing number of nocardiosis cases in immunocompromised individuals are being reported [1]. It has been reported that approximately 500 to 1000 cases of nocardiosis infections occur every year in the USA [2]. In recent years, with the increasing number of HIV patients, organ transplant patients, and the aging population, the infection rate of Nocardia has gradually increased. Nocardia predominantly causes infection in the lung and brain, and it can disseminate via the blood to cause infection in almost all organs. It will be life threatening when it disseminates to the central nervous system (CNS), with mortality rates as high as 85% in immunocompromised individuals [2]. Immunocompetent nocardia patients will have to get antibiotic treatment for 6 to 12 months and immunocompromised patients or those with CNS dissemination should receive treatment for at least 12 months [3]. At present, trimethoprim–sulfamethoxazole is the preferred therapy for nocardial infections [4]. Due to the nonspecific symptoms of infection, the long culture period, and the lack of specific diagnostic reagents, Nocardia has been underrecognized, underdiagnosed, and neglected [5]. If it cannot be diagnosed and treated in time, especially for patients with immunodeficiency, it will be fatal. Current research on Nocardia is limited. Nocardia can invade and survive in host cells, such as epithelial cells and macrophages, and can resist the host immune response by producing a variety of virulence factors, such as superoxide and hemolysin [6, 7]. Previously, we reported that the mce, hbha, and nfa34810 genes are involved in adhesion and invasion as virulence factors in Nocardia [8, 9]. Data from the complete genome sequence showed that the genome of N. farcinica contains several putative virulence genes, such as catalases and nbt, which may play a crucial role during the infection process [10]. Unfortunately, the role of most of the above genes in Nocardia during infection has not yet been investigated. Nocardia can quickly traverse capillary endothelial cells to enter the brain parenchyma and cause brain infections. Richter et al. reported for the first time that a patient infected with Nocardia had neurological symptoms at 6 weeks post-infection, such as mask face, trembling movement, stiffened muscles, and irregular limb tremors [11]. They found that Nocardia preferentially invaded the substantia nigra and putamen without causing apparent inflammation in both mice and Macaca fasicularius, and further study showed that head shake symptoms could be stopped temporarily after treatment with L-dopa [12]. David et al. documented that the neurological symptoms caused by Nocardia infection may be related to causes such as inner ear pathology; however, whether Nocardia infection causes PD remains to be clarified [13]. At present, research on inflammation of the nervous system caused by Nocardia infection is scarce, and the mechanism of Nocardia infection causing neurological symptoms needs to be examined in the future. Dual RNA-seq can simultaneously analyze the transcriptional profile in both hosts and pathogens during infection. Rieza et al. used dual RNA-seq to simultaneously study the interaction of pneumococcus and lung epithelium and revealed novel cellular processes and metabolic rewiring during pneumococcal infection [14]. Buket et al. analyzed infection-linked transcriptome adaptation in Haemophilus influenzae and host cells and revealed regulatory responses and metabolism modulation via dual RNA-seq, thus providing key insights into Haemophilus influenzae pathogenesis and the development of prevention strategies [15]. However, RNA-seq data for Nocardia, as emerging or neglected pathogens, are lacking, which merits the application of transcriptome data through dual RNA-seq technology to fill the gaps in Nocardia-related research fields. Taking advantage of dual RNA-seq advances, we first describe the transcriptional profile in both Nocardia and host cells. We combined the immune profiles of host cells from sequencing data with an in vivo assay to describe the characteristics of cytokine secretion in hosts post-infection with Nocardia. A series of novel virulence factors from Nocardia were found, and further in vivo experiments confirmed that iron acquisition genes played a key role in Nocardia infection. For host cells, we found that epigenetic modification might play an important role in the response to Nocardia infection. In vivo, we confirmed the neurological symptoms of mice post-infection with N. farcinica and analyzed 22 clinical strains that may cause neurological symptoms in mice. Mechanistically, the M1 microglial polarization induced by N. farcinica might be involved in neuroinflammation, which was related to the loss of dopaminergic neurons in the substantia nigra and decreased dopamine content in the striatum. Furthermore, the deletion of nbtS in N. farcinica completely attenuated the neurological symptoms. Our data present the first transcriptome analysis of Nocardia during interaction with alveolar epithelial cells and provide novel insights into Nocardia pathogenesis. More importantly, our study will effectively fill the gap in the research field of Nocardia and provide a theoretical basis for further in-depth system understanding and research on emerging or neglected pathogens.
To date, no RNA-seq data related to Nocardia infection have been reported. To fully understand the pathogenic mechanism of Nocardia and the adaptive response mechanism of the host response to Nocardia infection, the dual RNA-seq approach was applied in this study. To characterize the interactions of N. farcinica with lung epithelial cells, an in vitro infection model was built using human alveolar epithelial cells. N. farcinica was used to infect A549 cells for 1, 3, and 6 h at an MOI of 10. Giemsa analysis showed that some N. farcinica adhered and invaded the cells (Fig. 1a). Dual RNA-seq was used to determine genome-scale expression events in both hosts and bacteria at 1, 3, and 6 h post-infection (hpi). The transcriptomic data for dual RNA-seq were generated by applying the paired-end 75-nucleotide sequence method. Approximately 120 to 200 million reads were obtained at each time point after depletion of rRNA in both cells and bacteria, which was sufficient for dual RNA profiling (Fig. 1b). Principal component analysis (PCA) was used to investigate the trends in sequence data, which showed that similar samples clustered together without obvious batch effects (Fig. 1c). Pearson correlation of between samples is shown in Additional file 1: Fig. S1. To simplify further analyses, we selected genes that were differentially expressed ≥2-fold with an adjusted p<0.05 for follow-up research (Fig. 1d, e). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis indicated that these differentially expressed genes (DEGs) were mainly involved in membrane transport, translation, amino acid, and carbohydrate metabolism for N. farcinica, transport and catabolism, cell growth and death, signal transduction, signaling molecules and interaction, immune diseases, substance dependence, metabolism, and immune system for host cells (Additional file 1: Fig. S2 and Fig. S3). To validate the dual RNA-seq data, we first applied the total RNA that was used for sequencing as a template and randomly selected DEGs at different time points of infection for verification. Then, we repeated the experiments to obtain total RNA as a template and randomly selected DEGs for verification (Additional file 1: Table S1). As shown in Additional file 1: Fig. S4, the RNA-seq and qRT–PCR data showed a relatively strong correlation: R2 = 0.95 for Nocardia and R2 = 0.86 for epithelial cells, which indicated the reliability of the dual RNA-seq results.
Inflammation is a hallmark of Nocardia infection, which is mainly mediated by cytokines and chemokines. Notably, KEGG enrichment analysis of the DEGs during infection showed an enrichment of multiple immune-related signaling pathways of the host, including the tumor necrosis factor (TNF) signaling pathway and mitogen-activated protein kinase (MAPK) signaling pathway. To further clarify the characteristics of lung inflammation caused by Nocardia infection in vivo, N. farcinica in the exponential phase was used to infect mice intranasally, and immune factors in the lung were detected at 1, 3, and 7 days post-infection (dpi). As shown in Fig. 2, CXCL2, CXCL10, GM-CSF, M-CSF, IL-1β, IL-6, IL-17, and TNF-α were upregulated during early N. farcinica infection. The early host response and bacterial clearance mainly rely on neutrophils during pulmonary nocardiosis. IL-17 is involved in neutrophil infiltration during pulmonary nocardiosis [16], and the expression of IL-17 is upregulated during Nocardia infection. CXCL2, which has potent neutrophil chemotactic activity, was rapidly and significantly upregulated in the lung within 24 h of infection, which indicated that neutrophils recruited by CXCL2 played an important role during pulmonary nocardiosis. However, many chemokines have not yet been reported in the context of Nocardia infection. The production of CXCL10, which has mononuclear cell chemotactic activity, was also upregulated during early infection in this study. Interestingly, the expression of CCL2 was downregulated during lung infection caused by N. farcinica, which supported the differential regulation and functions of chemokines during Nocardia infection. GM-CSF expression was increased in the lung. Interestingly, anti-GM-CSF autoantibodies have been reported in patients with a primary cerebral abscess caused by Nocardia infection, and these patients may be at risk for later development of pulmonary alveolar proteinosis or other opportunistic infections [17, 18]. The expression of IFN-γ was not upregulated significantly during infection, indicating that the adaptive immunity of the Th1 response was not involved in pulmonary nocardiosis.
(i) Bacterial transcriptional adaptation at intermediate-late infection An increased number of Nocardia invaded A549 cells during infection under the present experimental conditions. To gain insights into the intermediate-late transcriptome signatures after colonization, DEGs at 3 and 6 phi were analyzed and compared with those at 1 hpi (Additional file 1: Fig. S5 and Additional file 2). The biosynthesis pathway of siderophore group nonribosomal peptides was significantly enriched at both 3 and 6 hpi compared with 1 hpi. However, no studies have been conducted on the virulence of these genes in Nocardia. In the present experiment, the RS03880 (nbtG), RS03155 (nbtS), RS03900 (nbtC), RS03910 (nbtE), RS03160 (nbtT), RS03905 (nbtD), RS03915 (nbtF), RS27150 (NFA_54680), and RS03895 (nbtB) genes, which are related to the biosynthesis of siderophore group nonribosomal peptides, were upregulated significantly (Fig. 3a). In addition, the adaptation of pathogen metabolism to the nutrients available in the host is an important prerequisite for survival [19]. These fruA and fruB genes, which are important in the metabolism of fructose, were upregulated during Nocardia infection. Numerous amino acids, such as proline and arginine, are available in vivo and could serve as a source of energy under anaerobic conditions. The mftE and pruA genes were upregulated during Nocardia infection and participated in the metabolism of arginine and proline. The pyruvate metabolism pathway was also enriched during infection, which supported the adaptation of Nocardia metabolism to the energy stresses present under in vivo conditions (Additional file 1: Fig. S6). (ii) Potential key factors for bacterial survival during infection To discover potential virulence-related genes associated with persistent infection after colonization, we first analyzed the DEGs that were expressed at both 3 and 6 hpi compared with 1 hpi (Fig. 3b). It was found 66 common genes were differentially expressed at both 6 and 3 hpi. Then, we analyzed 66 common genes and found that they were mainly involved in metabolic and environmental information processing pathways (Fig. 3c). These common DEGs are related to the biosynthesis pathway of siderophore group nonribosomal peptides, ABC transporters, and microbial metabolism in diverse environments, the phosphotransferase system (PTS), and the biosynthesis of secondary metabolites, which is crucial for the ability of Nocardia to cause infection and survive in the internal environment. (iii) Confirmation of virulence genes during infection At present, there are few studies on Nocardia virulence factors, and many potential virulence factors have not yet been discovered. To clarify the potential virulence-related or potentially important genes that cause persistent infection, we constructed a series of genetic deletions of N. farcinica to confirm the functions of these genes during infection in the mouse infection model [20] (Additional file 1: Fig. S7A). We first detected lactate dehydrogenase (LDH) in the culture supernatant of A549 cells at 8 h after infection with N. farcinica and the mutants. As shown in Fig. 4a, the ΔRS22575 (narI) and ΔRS24110 (NFA_48610) mutants showed significantly higher cytotoxicity to A549 cells than the wild-type, which indicated that these genes might have potential protective effects on host cells. To further clarify the function of these possible key genes, we infected mice and analyzed the bacterial load in the lung tissues and the mortality of the mice after infection. As shown in Fig. 4b and c, after infection with the ΔnbtB, ΔnbtS, ΔRS00660 (NFA_1310), and ΔRS03870 (NFA_7590) strains, the bacterial load in the lungs of mice was significantly reduced, and the survival rate was significantly improved compared with that of the mice infected with wild-type. In particular, strains ΔnbtB and ΔnbtS did not cause death of the mice, and thus, they may be the key virulence factors for Nocardia during infection. Interestingly, the mice infected with N. farcinica showed significant behavioral changes, such as turning in circles and regressive and other neurological symptoms, but the ΔnbtB and ΔnbtS mutants did not cause behavioral changes, which revealed that nbtB and nbtS played a crucial role during Nocardia infection and that these genes were required for the virulence of N. farcinica. We also found that NFA_7590, which encodes siderophore-interacting protein, was upregulated, and deletion of this gene resulted in significantly impaired bacterial virulence. In addition, the ΔRS24125(NFA_48640) and the ΔNFA_1310 mutant showed attenuated virulence. Furthermore, we found that the ΔnarI, ΔNFA_48610, and ΔRS03935 (NFA_7720) mutants failed to significantly affect the survival rate of mice compared with the wild-type after infection.
(i) Host transcriptional response upon infection at intermediate-late infection We then mainly focused on characterizing the host cell response to Nocardia infection (Fig. 5b, Additional file 1: Fig. S8, and Additional file 3). There were 48 DEGs during the intermediate-late stage, and further heatmaps and protein–protein interaction networks showed that these genes were mainly involved in epigenetic modifications mediated by histones (Fig. 5a, c). The expression of most histones in this experiment was downregulated during infection. Histones, primarily known for the role of condensing chromosomal DNA of mammals, are also involved in innate immune responses to antipathogens and the regulation of gene expression. Extracellular histones can activate proinflammatory signaling via Toll-like receptors [21]. M. tuberculosis can secrete several factors to target histones during infection, which could contribute to sustained bacterial survival in the host via histone modification [22, 23]. Indeed, modification of epigenomic processes is of importance for bacterial pathogen infection, and regulating or inhibiting these processes via histones may alter the outcome of infection. Angiopoietin-like 4 (ANGPTL4) was upregulated significantly at 6 hpi compared with 1 hpi. ANGPTL4 expression was reported to be elevated and involved in lung damage during infection caused by numerous stimuli, such as influenza pneumonia [24, 25]. Pneumonia is a common clinical symptom caused by Nocardia infection, and elevated ANGPTL4 may be involved in the lung damage caused by Nocardia. As shown in Fig. 5d and e, the production of ANGPTL4 was elevated significantly at 6 h after infection in A549 cells. In addition, the ANGPTL4 protein in the lungs of mice intranasally infected by N. farcinica was significantly upregulated. However, the mechanism by which ANGPTL4 mediates pulmonary inflammation induced by Nocardia is unclear and requires further study. In addition, ANGPTL4 plays an important role in regulating the integrity of endothelial vascular junctions via integrin pathways and destroys claudin-5 clusters and intercellular VE-cadherin [26]. Liu et al. reported that ANGPTL4 was significantly upregulated in meningitis and induced an increased permeability of the blood–brain barrier (BBB) by elevating myosin light chain 5 (MYL5) expression through RhoA signaling pathway activation [27]. The first step of infection of the brain induced by Nocardia is BBB disruption, but the underlying mechanism is unclear. In this study, ANGPTL4, which is involved in BBB integrity, was significantly upregulated, which indicated that this gene might play a critical role in the BBB destruction induced by Nocardia infection. WNT-inducible signaling pathway protein-1 (WISP1), which plays an important role in lung injury, was significantly upregulated during Nocardia infection. Recently, WISP1 was shown to promote the inflammatory response via TLR4/CD14 pathways in sepsis-induced lung injury [28]. DNA damage-inducible transcript 4 (DDIT4), encoding Rtp801, is also involved in inflammation in the lung and can promote alveolar inflammation and apoptosis of alveolar cells by suppressing mTOR signaling pathways, leading to lung injury [29]. In our study, we found that the expression of DDIT4 was upregulated, which suggested that this gene might participate in lung inflammation induced by Nocardia infection. (ii) Neurodegenerative symptom analysis During the in-depth exploration and analysis of the KEGG pathway, we found that some DEGs related to the pathways of the nervous system and neurodegenerative diseases were dysregulated, which caught our attention. Nocardia can quickly cross the blood–brain barrier and enter the brain parenchyma, causing central nervous system infection. Under our experimental conditions, not only were PD-related genes differentially expressed, but genes related to other nervous system and neurodegenerative diseases, such as Alzheimer’s disease (AD), amyotrophic lateral sclerosis (ALS), and other related genes, were also differentially expressed, which indicated that Nocardia entered the brain parenchyma and might cause a variety of symptoms related to the nervous system or neurodegenerative diseases (Fig. 6a). As shown in Fig. 6b, the RT–PCR results from PC12 cells were consistent with the gene expression trend of the RNA sequencing results. To further verify the association between Nocardia infection and PD-like symptoms through in vivo experiments, we infected mice with N. farcinica intravenously and analyzed the symptoms after infection. We found that all mice showed neurological symptoms without death post-infection with 5 × 106 CFU N. farcinica. However, a higher infectious dose caused some mice to die, and a lower infectious dose resulted in insignificant neurological symptoms in some mice. Therefore, an infection dose of 5 × 106 CFU was used in the subsequent study. The pole test, a method for testing the motor performance in PD, was applied to evaluating the motor dysfunction caused by Nocardia infection. As shown in Fig. 6c, N. farcinica infection caused motor dysfunction as measured by prolonging in the Tturn and Ttotal time as compared to control group. We also found that the ΔnbtS mutant strain almost have no effect on Tturn and Ttotal time as compared to control group. Besides, the behavioral disorder was visible as follows: (a) head falling on one side (Additional file 4: video 1); (b) a tendency to turn in the same direction when lifted by the tail (Additional file 4: video 2); (c) body quiescent tremor and rhythmical and vertical head movements (Additional file 4: video 1); (d) stagnation and turning backward in the same direction in unfamiliar environments, with the hind limbs open and stride length altered (Additional file 4: video 3); (e) circling of some mice at 3 months after infection (Additional file 4: video 4). The above symptoms further confirmed that N. farcinica infection could cause a series of neurodegenerative-like disease symptoms, which indicated that Nocardia infection might be involved in the development of neurodegenerative diseases such as PD. However, can all Nocardia infections invade the brain and cause neurological symptoms? We analyzed the relationship between 22 clinical Nocardia infections (Additional file 1: Fig. S9) and neurodegenerative diseases. Mice were infected intravenously with the above Nocardia strains, and symptoms were observed after infection. We found that N. Africa, N. kruczakiae, N. amikacinitolerans, N. pseudobrasiliensis, N. mexicana, N. novocastrense, N. caishijiensis, N. wallacei, N. pneumoniae, N. brasiliensis, N. inohanensis, N. transvalensis, N. beijingensis, N. blacklockiae, and N. asiatica infection caused neurological behavioral disorder at different times after infection, and the other strains did not cause obvious behavioral disorder symptoms after infection under the conditions used in our experiment (Fig. 6d). These results indicated that not all Nocardia infections could cause neurological symptoms, and strains capable of inducing neurological infection should arouse attention in clinical work. (iii) Microglial activation mediates the development of PD-like symptoms induced by Nocardia Damage or loss of dopaminergic neurons in the substantia nigra and a decreased dopamine content in the striatum are typical pathological features of PD. As shown in Fig. 7a, after Nocardia infection, the number of dopaminergic neurons in the substantia nigra of the mouse brain was significantly reduced post-infection, and the shape of the dopaminergic neurons was irregular. To further validate whether Nocardia infection caused PD-like symptoms, we analyzed the changes in dopamine content in the striatum of the mouse brain. As shown in Fig. 7b, the tyrosine hydroxylase (TH) in the striatum of the mouse brain was significantly reduced after infection, further indicating that Nocardia infection could cause a decrease in striatal dopamine content, which in turn led to PD-like neurological symptoms. In addition, we also observed a decrease in dopamine content in the olfactory bulb (Fig. 7)b. It has been reported that neuroinflammation is associated with neurodegenerative diseases and that microglia play a key role in inflammatory responses in the central nervous system [30–32]. In the present study, we found that microglia in the substantia nigra region were significantly activated, while astrocytes were not activated (Fig. 7c). Activated microglia can be divided into two different types: M1-like microglia and M2-like microglia [33]. However, whether Nocardia infection can cause polarization of microglia remains unclear. Therefore, we detected the expression of markers of M1 (iNOS, CXCL-10, and CD86) and M2 (CD206 and ARG1) microglia after Nocardia infection using BV2 cells (Additional file 1: Fig. S7B). As shown in Fig. 7d, M1-type markers (iNOS, CXCL-10, and CD86) were significantly upregulated after infection, and the expression level of M2-type markers (CD206 and ARG1) did not change significantly. In addition, we also found that conditioned medium from N. farcinica-infected (CoN) RAW264.7 cells could significantly stimulate the expression of M1 markers (iNOS and CXCL-10) in microglia. This result indicated that Nocardia infection might cause PD-like symptoms via neuroinflammation mediated by polarized M1 microglia. In addition, both Nocardia-microglia and macrophage-microglia interactions played a crucial role in driving M1 microglia polarization. To further clarify the characteristics of neuroinflammation caused by Nocardia infection, we analyzed inflammatory factors in the brain tissue of mice after Nocardia infection. To analyze whether the inflammatory factors in the brain were secreted by cells in the brain or were derived from peripheral blood and to analyze the relationship between the inflammatory factors in peripheral blood and brain tissue, we also detected the cytokine content in peripheral blood during different infection periods. As shown in Additional file 1: Fig. S10, cytokines in the brain (TNF-α, CCL-2, CXCL-2, CXCL-10, IL-1β, IL-6, IL-17, and M-CSF) were upregulated after infection at 3 dpi, and the cytokine content gradually decreased with extension of the infection time. In particular, the upregulated expression of CCL-2, CXCL-2, and CXCL-10 was the most significant. It has been reported that CXCL10, secreted from M1-type activated microglia, plays a crucial role in stimulating Th1 cell infiltration by serving as the ligand for CXCR3 T cells [34]. Interestingly, we found that CoN could stimulate microglia to significantly upregulate CXCL-10 compared with the direct interaction of N. farcinica and microglia. This result suggested that the upregulated CXCL-10 in the brain was partly secreted by polarized microglia, which was mainly mediated through macrophage-microglia interactions (Fig. 7d). Although these cytokines were upregulated in peripheral blood, their concentrations were significantly lower than those in brain tissue, which indicated that these cytokines were secreted by immune cells in the brain. In addition, the expression of IL-1β was significantly elevated at 3 dpi and lasted until 7 dpi in the brain. It has been reported that microglia activated by lipopolysaccharide (LPS) cause dopaminergic neuron damage in an IL-1β-dependent manner, resulting in PD-like neurological behavioral disorder [35]. To further analyze the mechanism of Nocardia infection responsible for PD-like neurological symptoms, the ΔnbtS mutant strain, which does not cause neurological symptoms after infection, was used in the next study. The mutant strain stimulated BV2 cells to produce fewer inflammatory factors, such as iNOS and CXCL-10, than the wild-type strain after infection. In addition, conditioned medium from the wild-type strain promoted BV2 cells to express more inflammatory factors than medium from the ΔnbtS mutant strain, especially CXCL-10 (Fig. 7d). These results indicated that CXCL-10 or iNOS played an important role in the neuroinflammation-mediated nervous system symptoms caused by Nocardia infection. In vivo, we found that the morphology of microglia was significantly different in the brain after infection with the wild-type and ΔnbtS mutant strains. As shown in Fig. 7e, the main morphology of brain microglia after infection by the mutant strain had a more rod-like and amoeboid shape, while the microglia mainly had a ramified phenotype after infection with the ΔnbtS mutant strain. These results indicated that the activation of microglial status played a key role in the neurological symptoms induced by Nocardia infection. Continuous activation of microglia is linked to the progression of PD by inducing dopaminergic neuron degeneration [36]. The MAPK signaling pathway in microglia plays an important role in the progression of PD [37]. In the present study, we found that Nocardia infection could cause significant activation of microglia, so we further studied the activation status of inflammation-related signaling pathways in vivo and in vitro. As shown in Fig. 7f, the extracellular regulated protein kinases (ERK) and c-jun n-terminal kinase (JNK) pathways were activated in BV2 cells after infection with Nocardia. In addition, these signaling molecules were also phosphorylated in the striatum of mouse brains infected with Nocardia. These results indicated that Nocardia-induced neurological behavioral disorder by activating microglia through the MAPK signaling pathway.
With the improvement of diagnostic technology, an increasing number of cases of nocardiosis have been reported. However, research on Nocardia infection is still insufficient. Several studies have applied a dual RNA-seq approach to reveal simultaneous transcription adaptation in host cells and pathogens [14, 15, 38, 39]. Here, we applied dual RNA-seq to obtain transcriptome data at different time points post-infection, confirmed several novel virulence genes for Nocardia infection in vivo, and expounded on the possible mechanism by which Nocardia infection causes a series of neurological symptoms, such as PD-like symptoms. In our study, we found that the differential expression of genes related to iron uptake was the most significant. Using gene knockout approach, we verified nbtB and nbtS were essential for Nocardia infection. This is the first study to report and verify nbtB and nbtS as key virulence factors for Nocardia in vivo. Indeed, the genes involved in iron acquisition and transport in Nocardia are still unclear, and further analysis of these genes related to iron uptake is essential for elucidating the pathogenic mechanism of Nocardia. Histones play a critical role in regulating gene expression by binding to DNA. In addition, histones involved in inflammatory responses are an important component in the recruitment of neutrophils to kill bacteria [40]. It has been reported that H2A and H2B have the capacity to neutralize endotoxins and act as antimicrobials against Escherichia coli [41]. Bacteria have also evolved to resist killing by histones. Finegoldia magna can bind histones with surface proteins and degrade histones with secreted proteases [42]. In this study, we found that histones were downregulated significantly at both 6 and 3 hpi (Fig. 5a, c). However, the role of dysregulated histones in the host cell response to Nocardia infection has not been reported, and it is necessary to further clarify the functions of histones in nocardiosis. Though several studies have documented that Nocardia infection can lead to PD-like neurological symptoms, there is some controversy about the relationship of Nocardia infection to the occurrence and development of PD [13, 43, 44]. We confirmed that N. farcinica infection was able to induce PD-like symptoms (Additional file 4: video 1, 2, and 3). It has been reported that Nocardia can change to cell wall-deficient L-forms after invading macrophages, which may be the reason why Nocardia is difficult to culture [45]. The L-forms of Nocardia will further increase the difficulty of diagnosing nocardiosis. Therefore, it is a challenge to diagnose patients with clinical PD-like symptoms caused by Nocardia infection, thus weakening the potential association between Nocardia and PD or PD-like disease, which should attract the attention of clinicians. Corrales et al. found an inflammatory response in parts of the brain in which Nocardia was not detected post-infection [46]. Similar to this finding, we noted the occurrence of diffuse inflammation in the brain, even in areas that were not affected by Nocardia, and this inflammation persisted even if the bacteria could not be cultured. However, the mechanism is currently unclear and needs to be further resolved in the future. Next, we found that only a portion of the Nocardia were capable of causing neurological symptoms, which indicated the presence of differences in brain susceptibility between Nocardia species and that the mechanism mediating the differences in susceptibility requires further study. These findings further remind us to focus on clinical strains of Nocardia to which the brain is susceptible. Microglial activation-mediated neuroinflammation plays a key role in central nervous system diseases, such as PD. Here, we found that microglia but not astrocytes were activated after infection with N. farcinica (Fig. 7c), indicating the key role of microglial activation in neuroinflammation caused by N. farcinica. Further study showed that N. farcinica could drive M1 microglial polarization directly or indirectly through macrophages. Interestingly, we found that the neurological symptoms were completely abolished after infection with the ΔnbtS mutant strain, and the expression of inflammatory factors, such as iNOS and CXCL-10, in BV2 cells stimulated with this mutant strain was lower than that induced by the wild-type strain after infection. Finally, we found that N. farcinica could stimulate the MAPK pathway of innate immunity both in vitro and in vivo, which suggested that the MAPK pathway might be involved in the activation of microglia and lead to neurological behavioral disorder after Nocardia infection. Indeed, our study does have certain limitations. Is the neurological symptom caused by direct infection of neurons by the Nocardia or is it mainly mediated by neuroinflammation mediated by microglia? This issue needs to be further investigated. In addition, whether the virulence gene nbtS are secreted to exert virulence has not yet been studied, and the pathogenic mechanism is unclear. Finally, transcriptome analysis of Nocardia-infected brain tissue will helps to clarify the molecular mechanism of Nocardia-induced neurological symptoms.
In summary, we disclosed a series of novel virulence genes and metabolic pathways for Nocardia and clarified the relationship between Nocardia infection and neurological diseases, especially PD-like symptoms. Our findings provide insight into a deeper understanding of host-pathogen interactions for emerging or neglected Nocardia and will lay the foundation for future studies on the pathogenesis of Nocardia. In addition, our study may serve as a blueprint that can be applied to other bacterial pathogens in the future.
Anti-p-JNK (Cat# 4668), anti-JNK (Cat# 9252), anti-p-ERK 1/2 (Cat# 4370), anti-ERK (Cat# 4695), and anti-β-actin (Cat# 3700) were purchased from Cell Signaling Technology. Anti-tyrosine hydroxylase (Cat# 75875) was purchased from Abcam. Anti-ANGPTL4 (Cat# 67577) was purchased from Proteintech. Anti-Iba1 (Cat# GB13105-1) and anti-GFAP (Cat# GB11096) antibodies were purchased from Servicebio. Luminex Mouse Magnetic Assay (Cat# LXSAMSM-11) was purchased from R&D Systems. Cytotoxicity Assay (Cat# G1782) was from Promega.
Cells were grown in DMEM supplemented with 10% fetal calf serum (FCS; TransGen Biotech), 2 mM l-glutamine (Gibco), and 1 mM sodium pyruvate (Gibco) in T-75 flasks (Corning) in a 5% CO2 humidified atmosphere at 37 °C. Nocardia strains were grown in brain heart infusion medium (BHI, Oxoid) at 37 °C.
In vitro, 1.5×106 cells/well were seeded in six-well plates. The fresh bacteria were resuspended in complete DMEM to prepare the inoculum. Infections were performed at a multiplicity of infection (MOI) of 10. The inocula or medium-only controls were added to the apical surface of the cultures and incubated for 1 h in a 5% (v/v) CO2 humidified atmosphere at 37 °C. The cells were washed three times with PBS to remove the free-floating N. farcinica prior to RNA extraction.
Total RNA was isolated at 1, 3, and 6 h from N. farcinica-infected A549 cells and corresponding mock controls (three replicates per time point). Total RNA isolated from bacteria conditioned in infection medium at 37 °C in 5% CO2 for 1 h was used as a bacterial baseline control. Three technical replicates (individual wells) were pooled into one biological replicate. Three biological replicates were used. Before isolation, the wells were gently rinsed three times with PBS after 1 h of infection to remove nonadherent N. farcinica cells, and the total RNA was extracted according to the manufacturer’s recommended protocol. The total RNA sample concentrations, RIN, and size were detected using an Agilent 2100 Bioanalyzer (Agilent RNA 6000 Nano Kit), and the purity of the samples was tested using a NanoDropTM.
Host and bacterial ribosomal RNAs were simultaneously depleted by a 1:1 mixture of human/mouse/rat and gram-positive bacterial capture probes (Ribo-Zero rRNA Removal Kits, Illumina, USA). cDNA sequencing of the 15 samples was carried out on the Illumina HiSeq Xten platform in 75-bp paired-end mode at The Beijing Genomics Institute. For cDNA library preparation, 450 ng RNA of each sample was used. Ribodepleted RNA samples were fragmented using fragmentation reagent, and first-strand cDNA was generated using random primer reverse transcription, followed by second-strand cDNA synthesis.
Sequencing reads were filtered with SOAPnuke software (https://github.com/BGI-flexlab/SOAPnuke, version v1.5.2) to remove reads with adaptors, reads in which unknown bases (N) made up more than 10% and low-quality reads. For the transcriptome analysis, reads were aligned in paired-end mode to a human genome (hg19) and N. farcinica IFM10152 genome (NCBI RefSeq accession numbers: NC_006361.1, NC_006362.1, NC_006363.1) using HISAT (Hierarchical Indexing for Spliced Alignment of Transcripts, Version v2.0.4) [47] with default settings. To avoid cross mapping, before mapping to the N. farcinica IFM10152 genome, clean reads from infected samples and the mimic control group were aligned to the human genome, and unmapped pairs of reads were then used for alignment to the bacterial genome [15]. For mapping of the human genome, clean reads were aligned to the human genome, and those reads that were cross mapped with bacteria were discarded. Clean reads were mapped to references using Bowtie2(V2.2.5) [48], and then the FPKM (fragments per kilobase million) was calculated with RSEM (V1.2.12) [49]. Differential expression was evaluated by comparing the data from infected samples to that from mimic control samples. DEGs that were differentially expressed≧2-fold with an adjusted≦0.05 were detected with DEseq2 (V1.36.0) with default parameter as described previously [50]. Differentially expressed genes between the infected and control groups were identified using the following thresholds: |log2fold change| of ≥1 and adjusted p value of ≤0.05. The Kyoto Encyclopedia of Genes and Genomes (KEGG) and clustering analysis of DEGs were completed on Dr. Tom system (https://biosys.bgi.com).
To validate the RNA-seq data, total RNA from sequenced or repeated infection samples was used in qRT–PCR. Briefly, qPCR was performed with SYBR Premix Ex Taq II reagents (TaKaRa, Japan) on a 7500 Fast Real-Time PCR System (Applied Biosystems) using 2 μl of the diluted cDNA samples, 10 μl of Power SYBR master mix, and 1 μl of 10 M gene-specific primer. The housekeeping genes secA and β-actin for Nocardia and epithelial cells, respectively, were used to normalize the level of gene expression. Fold changes in gene expression were determined using the 2(−∆∆Ct) method [51].
To obtain the minimum infectious dose for neurological symptoms, female C57BL/6J mice were infected intravenously with N. farcinica at approximately 1× 108, 1× 107, 5× 106, 1× 106, and 5× 105 CFU in 100 μl PBS, and the number of mice with neurological symptoms was analyzed.
Mice were trained to climb a pole 3 days before the pole test. In test, mice were placed on the top of a rough-surfaced and vertical pole (1 mm in diameter, 60 cm in height) with head up and the time needed to turn downward (Tturn) and the time to reach the bottom (Ttotal) were recorded. Three times measurements was used as the result.
The gene in-frame deletion mutant was constructed via homologous recombination according to previously described methods [20]. Upstream DNA fragments and downstream gene fragments were amplified by PCR using primers (Additional file 1: Table S2). These two fragments were subsequently ligated to generate a gene deletion fragment, which was then cloned into the pK18mobsacB vector. N. farcinica in logarithmic phase was washed 3 times with ice-cold water and then resuspended in 10% ice-cold glycerol to generate competent bacteria. The recombinant pK18mobsacB plasmid was transformed into competent bacteria and incubated in BHI broth for 2 h at 37 °C. The positive colonies were first selected on BHI plates containing neomycin and then selected on BHI plates containing 20% sucrose. The deletion of genes were confirmed by PCR (Additional file 1: Table S2).
Mice were infected intravenously with wild-type or mutant Nocardia in logarithmic phase at a dose of 3×107 CFU in 100 μl PBS, and then the survival rate of mice was analyzed after infection. Mice were infected intravenously with wild-type or mutant Nocardia at a dose of 5×106 CFU in 100 μl PBS, and the lung tissue was separated and ground in 1 ml PBS. Then, the colonies in the lung were counted.
A549 cells were cultured in DMEM supplemented with 10% FBS at 37 °C for 16–18 h before infection. Wild-type or mutant Nocardia in logarithmic phase were used at a MOI of 10:1. The CytoTox 96® Non-Radioactive Cytotoxicity Assay (Promega, USA) was used to measure the cytotoxicity according to the manufacturer’s instructions at 8 h post-infection as previously described [52].
RAW264.7 cells were infected with N. farcinica or ΔnbtS strains at a MOI of 10. Cell culture supernatants were obtained and filtered through a 0.2-μm sterile filter 24 h after infection [53]. The conditioned medium from N. farcinica- or mutant strain-infected cells was termed CoN, and medium from uninfected cells was used as a control. N. farcinica or mutant strains at a MOI of 10 or conditioned medium at a dilution of 1:3 were used to prime microglial cells for 4 h, and then total RNA was isolated and analyzed.
BV-2 or A549 cells were seeded onto 6-well plates for 16–18 h and infected with N. farcinica (MOI=10). After infection, the cells were lysed with lysis buffer supplemented with phosphatase and protease inhibitors (CWBIO, China) on ice as previously described [54]. The lung, striatum, and olfactory bulb isolated from mice were added to lysis buffer and ground on ice to extract the total protein. In brief, the total protein were separated by SDS–PAGE and then transferred to polyvinylidene fluoride membranes (Millipore). The membranes were incubated with primary antibodies and then incubated with HRP-conjugated anti-rabbit IgG (Beyotime Biotechnology), and the bands were measured using a Western Lightning Plus ECL kit (PerkinElmer, USA).
Mice were infected intranasally with N. farcinica (1×107 CFU), and lung tissue was sampled at 1, 3, and 7 days post-infection and then ground on ice with protease inhibitors. Then, the tissue supernatant was obtained after centrifugation (10,000g, 10 min) and stored at −80 °C for detection. For serum and brain, mice were infected intravenously with N. farcinica (5×106 CFU). Then, brain tissue and blood were isolated at 3, 7, and 14 days after infection. The supernatant of brain tissue and serum was collected after centrifugation and stored at −80 °C until testing. Cytokines (CCL2, CXCCL2, CXCL10, GM-CSF, IFN-γ, IL-1β, IL-2, IL-6, IL-17, M-CSF, TNF-α) in tissue supernatant and serum were determined by a Multiplex® system (R&D, USA) according to the manufacturer’s recommendations.
Mice were sacrificed 14 days post-infection and perfused intracardially with PBS and 4% paraformaldehyde. Then, the brain was collected and preserved in 4% paraformaldehyde. The coronal sections of brain (4 μm) containing the SNpc region were dissected. The sections were then incubated with primary antibodies targeting tyrosine hydroxylase, GFAP and Iba-1. After washing, the sections were treated with Cy3- or Alexa Fluor 488-conjugated secondary antibodies. The nuclei were labeled via DAPI counterstaining. Images were captured with a laser scanning confocal microscope (LSM 700, Carl Zeiss). For cells, A549 cells were seeded into 24-well plates with glass coverslips and grown until confluence. Then, A549 cells were infected with N. farcinica at logarithmic phase at a MOI of 10 for 6 h. Anti-ANGPTL4 antibody was used as a primary antibody and incubated for 30 min at 37 °C. Subsequently, anti-rabbit IgG was added to the cells for 30 min at 37 °C after 3 washes with PBS. Finally, DAPI was added to the cells for 10 min and then imaged.
All animal research was performed in accordance with animal ethics guidelines and approved protocols. The animal experiments were approved by the Ethics Review Committee of Shandong Provincial Hospital.
All data were analyzed and presented using GraphPad Prism 8. Analyses were performed using Student’s two-tailed T test and reported as mean ± standard deviations (SDs). Each statistical test used for each figure is described in the legends.
Additional file 1: Figure S1. [Heat map of the Pearson correlation of gene expression between samples for RNA-seq data]. Figure S2. [KEGG pathway analysis of DEGs for N. farcinica and A549 cells]. Figure S3. [KEGG pathway analysis]. Figure S4. [Validation of RNA-seq via qRT–PCR]. Figure S5. [Volcano plot obtained from DESeq2 analysis of DEGs from Nocardia farcinica]. Figure S6. [KEGG pathway analysis of Nocardia at 3 and 6 hpi]. Figure S7. [Analysis of the N. farcinica mutant strains]. Figure S8. [Volcano plot obtained from DESeq2 analysis of DEGs from A549 cells]. Figure S9. [The colony status of the Nocardia strains on the blood plate]. Figure S10. [Analysis of inflammatory factors in serum and brain]. Figure S11. [Original gel images with indicated figures].Additional file 2. DEGs of N. farcinica at 3 and 6 phi.Additional file 3. DEGs of A549 cells at 3 and 6 phi.Additional file 4: Video 1. [head falling on one side, body quiescent tremor and rhythmical and vertical head movements]. Video 2. [a tendency to turn in the same direction when lifted by the tail]. Video 3. [stagnation and turning backward in the same direction in unfamiliar environments, with the hind limbs open and stride length altered]. Video 4. [mice circling after infection]. |
|
PMC9647968 | Chandra Prajapati,Jussi Koivumäki,Mari Pekkanen-Mattila,Katriina Aalto-Setälä | Sex differences in heart: from basics to clinics | 09-11-2022 | Sex differences,Electrocardiogram,Arrhythmias,Heart failure,Torsades de pointes,iPSC modeling,Animal experiments | Sex differences exist in the structure and function of human heart. The patterns of ventricular repolarization in normal electrocardiograms (ECG) differ in men and women: men ECG pattern displays higher T-wave amplitude and increased ST angle. Generally, women have longer QT duration because of reduced repolarization reserve, and thus, women are more susceptible for the occurrence of torsades de pointes associated with drugs prolonging ventricular repolarization. Sex differences are also observed in the prevalence, penetrance and symptom severity, and also in the prognosis of cardiovascular disease. Generally, women live longer, have less clinical symptoms of cardiac diseases, and later onset of symptoms than men. Sex hormones also play an important role in regulating ventricular repolarization, suggesting that hormones directly influence various cellular functions and adrenergic regulation. From the clinical perspective, sex-based differences in heart physiology are widely recognized, but in daily practice, cardiac diseases are often underdiagnosed and untreated in the women. The underlying mechanisms of sex differences are, however, poorly understood. Here, we summarize sex-dependent differences in normal cardiac physiology, role of sex hormones, and differences in drug responses. Furthermore, we also discuss the importance of human induced pluripotent stem cell-derived cardiomyocytes in further understanding the mechanism of differences in women and men. | Sex differences in heart: from basics to clinics
Sex differences exist in the structure and function of human heart. The patterns of ventricular repolarization in normal electrocardiograms (ECG) differ in men and women: men ECG pattern displays higher T-wave amplitude and increased ST angle. Generally, women have longer QT duration because of reduced repolarization reserve, and thus, women are more susceptible for the occurrence of torsades de pointes associated with drugs prolonging ventricular repolarization. Sex differences are also observed in the prevalence, penetrance and symptom severity, and also in the prognosis of cardiovascular disease. Generally, women live longer, have less clinical symptoms of cardiac diseases, and later onset of symptoms than men. Sex hormones also play an important role in regulating ventricular repolarization, suggesting that hormones directly influence various cellular functions and adrenergic regulation. From the clinical perspective, sex-based differences in heart physiology are widely recognized, but in daily practice, cardiac diseases are often underdiagnosed and untreated in the women. The underlying mechanisms of sex differences are, however, poorly understood. Here, we summarize sex-dependent differences in normal cardiac physiology, role of sex hormones, and differences in drug responses. Furthermore, we also discuss the importance of human induced pluripotent stem cell-derived cardiomyocytes in further understanding the mechanism of differences in women and men.
Men and women have similar genetic materials except the amount and presence of sex chromosomes, but they differ in cardiac anatomy and physiology. However, sex-specific differences extend from cardiac structure and function to the presentation and progression of cardiac diseases, as recently reviewed by Shufelt et al. [1]. Men are generally more susceptible and at earlier age to cardiovascular diseases, such as coronary artery disease than women [2]. Typically, men develop heart failure (HF) with reduced ejection fraction, whereas women develop HF with preserved ejection fraction [3], and, men develop atrial fibrillation approximately 10 years earlier than women [4]. Furthermore, right ventricular outflow tract tachycardia has sex-specific triggers; for instance, arrhythmia is initiated by hormonal fluxes in women and exercise and/or stress in men [5]. Even though the age at diagnosis is significantly lower in men, women are usually more symptomatic at the time of diagnosis [6]. The most severe clinical manifestations, i.e., ventricular fibrillation or sudden cardiac death are 8–10 times more prevalent in men than in women [7], indicating that women have better outcomes for cardiovascular major events [8]. Although endogenous sex hormones play an important role in cardiac physiology, detailed reports on the mechanisms behind these sex differences remain unavailable. From the electrophysiological point of view, women exhibit higher beating rates and slower repolarization than men. The underlying differences have been observed at the level of individual cardiomyocytes (CMs). This has led to the comprehensive research interest in identifying the cellular mechanisms responsible for sex differences in CM function. Furthermore, men undergo greater cardiac remodeling with aging than women. However, owing to limited human-based research material, most of the studies identifying the key mechanisms underlying the sex differences are conducted in animals. Although these animal studies have depicted numerous meaningful reasons behind sex differences, their outcomes cannot be extrapolated directly to humans because of fundamental differences in animal and human cardiac physiology and estrous cycle [9]. In addition, studies about sex differences have many confounding variables that might affect the results such as diet, exercise, and environment. Most of these problems can be solved with the recent progress in human induced pluripotent stem cell-derived CMs (hiPSC-CMs) [10], which are not only human-based but also enable conducting research on sex difference in a controlled environment. This review aims to provide an overview of the currently known sex-specific differences in the cardiac electrophysiology at the organ and cellular levels. Additionally, the effect of sex hormones on ion channels is discussed as well as sex-specific differences in cardiovascular drug responses.
The normal structure and physiology of the heart differ in men and women, and these differences reach their peak after puberty.
For men and women, the left ventricular (LV) mass values are similar during infancy and childhood, suggesting an almost the same initial number of CMs [11]. The clear sex difference in LV mass is observed only after puberty, and male CMs undergo greater hypertrophy than female CMs [11]. Men have 25–38% greater LV mass because of larger LV chamber dimension and wall thickness [11, 12]. In both sexes, the posterior wall thickness, septal wall thickness, LV mass, and LV mass index increase with age, but their values remain significantly smaller in women from 20 to 80 years of age [12]. The level of CM apoptosis also influences sex-specific differences; a male heart undergoes a higher percentage of apoptosis of ventricular CMs [13]. Thus, aging is associated with a significant decrease in the absolute number of CMs in a male heart, but the cellular volume is increased (i.e., hypertrophy) [14, 15]. In the left and right ventricles of males, binucleated cells progressively increase in number, whereas mononucleated cells progressively reduce with age [15]. Such phenomena have not been observed in a women heart [15]. The reason for the number of CMs to remain similar in an aging women heart but because of the higher regenerative capacity of the CMs in a women heart but is unlikely because of the absence of CM apoptosis. Females’ hearts are generally smaller but in proportion to their smaller body size [16].
LV ejection fraction (LVEF) is one of the central measures of LV systolic function. LVEF is the fraction of chamber volume ejected in systole (stroke volume) in relation to ventricular blood volume at the end of diastole (end-diastolic volume). Reports on whether LVEF differs between men and women are still conflicting [17]. The Framingham Heart Offspring Study showed that LVEF is not different between men and women [18], whereas the Dallas Heart Study reported that women had larger LVEF than men with the same age range [17]. Another large population study demonstrated that LVEF increased with age in both sexes, but was generally larger in women [12, 19]. After indexing ventricular volumes to body size, women have larger LVEF than men with similar age [20]. LV response to exercise also differs between sexes. Although the LVEF is smaller in men than in women at rest, it is reportedly larger in men than in age-matched women during exercise [21, 22]. Conversely, the increase in the end-diastolic volume index is greater in women during exercise but smaller during rest [22]. Additionally, men have a greater exercise capacity than women, most likely because they have a larger LV volume [23].
Electrocardiogram (ECG) is the most widely used clinical test to evaluate heart’s electrical activity. Atrial depolarization is seen as the P wave; subsequently, the right and left ventricles are depolarized, forming the QRS complex. The T-wave represents ventricular repolarization to the resting state. The normal QT interval depends on the heart rate, and it is typically corrected (QTC) using various correction formulas [24]. T-wave amplitude (TWA) is the highest peak of the T-wave. ST angle is the angle between the baseline and the first 80 ms of the ST segment. Similar to the QT interval, the JT interval is used to quantify the repolarization duration; it is defined as the time period of QT subtracted by QRS duration [25] (Fig. 1). In addition, QT dispersion is defined as the difference between the longest and shortest QT interval in different ECG leads. Fetal heart rate examination at 20–36 weeks of age by echocardiography has revealed no sex-related differences [26]. Neonatal ECG has demonstrated that female babies have higher heart rates and shorter absolute values of the QT interval with a shorter cycle length, leading to a QTC similar to that in male babies [27]. Studies with boys and girls younger than 15 years of age have shown no differences in heart rate and QTC intervals [28, 29]. After puberty, the QTC interval drops by approximately 20 ms in men, whereas it remains unchanged in women, leading to a longer QTC in women irrespective of the correction methods [24, 29]. The QTC gradually increases in men, and the correlation between age and QTC is greater in men than in women [24]. Therefore, the QTC differences between sexes are greater at younger ages, diminish with age, and disappear after 75 years of age [24, 28]. Furthermore, aging modulates the dispersion of ventricular repolarization; the older the individuals, the higher the QTC dispersion, which may contribute to cardiac mortality in old population [30]. Although QTC dispersion in young and middle-age shows no sex differences, it has been reported to be higher in older men than in older women (≥ 70 years old) [30, 31]. Sex differences are observed not only in cardiac repolarization, but also in ECG patterns (Fig. 1). Typical male and female patterns are readily recognizable in most ECGs without measurements. Male ECGs exhibit a higher J point, i.e., the level at which repolarization starts, a steeper ST segment (angle of the ST segment), and a greater TWA than female ECGs [25, 28]. Greater levels of J point and steepness of the ST angle segment in men implies a faster velocity of initial repolarization and typical early depolarization pattern [25]. During the period in which QTC is longer in women, the distribution of men and women ECG patterns also shows differences [25, 28]. All these repolarization variables correlate negatively with age in men but not in women [25]. However, increasing age is associated with a higher heart rate and a shorter mean RR interval only in women [31–33]; this association is caused by differences in ambient autonomic tone [23]. In addition, the slopes of the linear relationship between the absolute values of QT and RR intervals are steeper in women, indicating that women have longer QT intervals than men at a decreased heart rate and that sex difference in QTC is more marked at a longer cycle [33]. However, sex difference in QTC is absent at a shorter cycle [33], possibly because women shorten their QT interval more than men in response heart rate increase [34]. In conclusion, sex differences in ECG and its patterns are apparent only after puberty. Men exhibit a shorter and faster repolarization after puberty until approximately 75 years of age. In both sexes, QTC prolongs and ECG pattern changes with age, but these changes are more prominent in men than in women. Older men (> 70 years old) have a higher risk for ventricular arrhythmia at least partially due to higher QTC dispersion.
Generally, women have significantly longer QTC intervals because men begin to shorten their QTC interval after puberty [29]. The different testosterone levels may explain the differences in QT interval between sexes, since testosterone accelerates ventricular repolarization [35, 36]. In addition, free testosterone at physiological levels inversely correlate with QT interval in men but not in women [35, 36]. Castrated males had longer rate-corrected JT intervals (JTC) and less steep ST-segments than non-castrated males, further supporting the role of testosterone on the configuration and duration of ventricular repolarization [37]. Furthermore, the age-dependent changes in male ECG pattern tend to correlate with the rise in testosterone level during puberty and the decline in older males [28]. Unlike testosterone, estrogen does not have clear effects on the duration and pattern of cardiac repolarization in humans [36]. Bilateral oophorectomy does not induce any ECG changes in postmenopausal females, but it increases the mean duration and decreases the amplitude of the T-wave in premenopausal women [38]. An estrogen does not alter the heart rate, QTC or JTC in resting condition, but estrogen-deficient state increases the QT interval dispersion, which decreases after hormonal replacement therapy suggesting the protective role of estrogen against severe ventricular arrhythmias and sudden cardiac death [39–41]. However, hormone replacement therapy with estrogen increases the QT interval dispersion during peak exercise [41]. When combined with progesterone, the QTC interval does not increase, suggesting that progesterone reverses estrogen-induced QT prolongation [42]. Additionally, females with long QT syndrome (LQTS) have a lower risk for cardiac events during pregnancy, when the progesterone level increases [43]. However, the risk suddenly increases at postpartum, when the progesterone level decreases, also suggesting the protective role of progesterone [43]. In the menstrual cycle, the estrogen and progesterone levels change; estrogen levels gradually increase during the follicular phase, reach a peak after 11–13 days, and then decrease during the luteal phase, while progesterone levels do not increase until the luteal phase. Results about the effect of these hormonal changes on ECG and heart rate are still inconsistent [44–47]. In short, sex hormones, especially testosterone play an important role for the hormonal factor of the sex-specific differences mainly by affecting cardiac repolarization.
Several studies have revealed that sex differences exist already at the cellular level. However, due to limited availability of human samples, most of the experiments have been conducted with animal models such as rodent and canine models. With the help of hiPSC technology, information about sex-specific differences in humans at the CMs level can be obtained.
Cyclic changes in intracellular calcium (Ca2+) concentration regulate cardiac contractility. The Ca2+ influx through the L-type Ca2+ channels triggers the opening of ryanodine receptors (RyR2s), and subsequent release of Ca2+ from the sarcoplasmic reticulum (SR). This phenomenon gives rise to Ca2+ transients, referring to the process called Ca2+-induced Ca2+ release [48]. When the intracellular Ca2+ level increases, Ca2+ binds to myofilaments (troponin C), causing CM contraction. Relaxation occurs when the majority of Ca2+ is transferred back into the SR via the SR Ca2+-ATPase (SERCA), while a smaller amount of Ca2+ is extruded from the cell predominantly by the sodium (Na+)/Ca2+ exchanger (NCX) [48]. At the cellular level, sex differences are observed in excitation–contraction coupling (ECC). Table 1 compares the intracellular Ca2+ handling properties between female and male CMs of different species. Several experimental results have shown that the CMs from female heart samples exhibit a smaller contraction along with longer time both to maximal/peak shortening and to 50% relaxation [49–53]. CM contractility depends not only on the myofilaments properties, but also on the diastolic and systolic Ca2+levels. The smaller contraction in female CMs arises from the smaller Ca2+ transient amplitude and/or lower diastolic Ca2+ levels in female CMs [49–54]. Another characteristic of female CMs is the slower rate of rise and decay of Ca2+ transient, corresponding to a longer time to peak shortening and the slower time to 50% relaxation [49, 50]. Smaller and slower rise of Ca2+ transient in female CMs implies reduced RyR2 activity, while a slower decay is due to reduced SERCA and/or NCX activity. Ca2+ content in SR is the same in male and female CMs and thus this does not explain the smaller Ca2+ release in female CMs [50, 54]. Actually, SR Ca2+ content has even been found to be higher in female CMs in one study [55]. The protein levels of SERCA, phospholamban (PLB), and calsequestrin show no sex differences [56, 57]. However, the protein and mRNA levels of RyR2 and NCX are higher in female hearts [57]. Thus, the altered intracellular Ca2+ handling in female CMs is not caused by the SR Ca2+ content and Ca2+ handling protein levels. Individual SR Ca2+ release units known as Ca2+ sparks may partly explain for this disparity. Female CMs have smaller Ca2+ spark amplitudes and durations (time to peak and decay time) [50, 54]. Furthermore, the degree of amplification of Ca2+ influx to the resulting amount of Ca2+ released from the SR can be quantified as ECC gain, which indicates the ratio of the amount of SR Ca2+ released per unit of L-type Ca2+ current (ICaL). In female CMs, the EEC gain is weaker, demonstrating that the amount of Ca2+ release per unit ICaL is lower in female CMs [53, 55]. Therefore, the smaller Ca2+ transient amplitude in female CMs is more likely to be caused by reduced Ca2+ influx trigger with a consequent reduction in SR Ca2+ release [49]. Aging also affects the sex differences in ECC [51, 52]. The fractional shortening, ICaL, and Ca2+ transients are reduced only in the CMs of older male mice, whereas SR Ca2+ content is increased in the CMs of older female mice [51]. Moreover, the Ca2+ sensitivity of myofilaments declines with age in male CMs, causing smaller contractions in older male CMs [52]. Therefore, despite that younger female CMs demonstrate smaller contractions, Ca2+ transient amplitude, diastolic Ca2+, and ECC gain, such differences vanish as the age advances [51, 52]. In addition, intracellular signaling and/or sex hormones play an important role in the regulation of intracellular Ca2+ handling, which engender sex-specific differences, as described below. Taken together, the ECC in CMs depends on both sex and age, and the age-related alterations are more prominent in males than in females. Of note, smaller Ca2+ sparks and lower ECC gain are fundamental characteristics of female CMs. The marked reduction in ECC gain in female CMs may limit SR Ca2+ release under physiological conditions of stress such as exercise; this may explain why females are less capable of increasing LVEF under stress or exercise than males.
Sex differences have been observed in action potential (AP) parameters and ionic currents and are strongly dependent on the species and origin of CMs within the ventricular wall (Table 2). In some animal studies, AP duration (APD) is longer in female CMs than in male CMs [55, 58–61], but other studies found no sex difference in APD [50, 53, 62, 63]. Xiao and coworkers examined sex differences at three transmural levels in dogs and demonstrated that APD was only significantly longer in the mid-myocardium of female dogs; the epicardium and endocardium showed no sex differences [64]. The resting membrane potential (RMP) [50, 53, 58, 60, 61, 63, 64], maximal upstroke velocity (dV/dt) [63], and AP amplitude (APA) [55, 63] are sex independent. Fast Na+ current (INa) is responsible for the rapid upstroke phase of AP. Although the INa densities in dog mid-myocardium do not differ in sexes [65], those in female endo/epicardium were significantly lower than those in male endo/epicardium [65]. Following rapid depolarization, transient outward potassium (K+) current (Ito) starts the repolarization of AP. The Ito densities are significantly lower in female mouse CMs [58] and in female dog endocardium than in male counterparts [64]. The ICaL is responsible for the plateau phase of AP. Female CMs from dogs [64] and guinea pigs [55] have increased ICaL densities. In contrast, James et al. showed that ICaL densities are decreased in female guinea pig CMs [59]. Furthermore, Sims and coworkers demonstrated significantly larger ICaL densities in CMs from the base of adult female rabbit hearts but found no difference between apical CMs [66]. As the plateau phase moves toward a more negative membrane potential, two types of K+ currents, namely, rapid rectifier K+ current (IKr) and slow rectifier K+ current (IKs), are activated. According to Liu et al., female rabbit CMs have significantly lower IKr densities than male rabbit CMs [67]. Zhu et al. also showed significantly lower IKs densities in female rabbit CMs [60]. Conversely, Xiao et al. demonstrated higher IKs densities in female dog epicardium and endocardium [64]. The ultra-rapid delayed rectifier K+ current (IKur) is mainly responsible for repolarization in mouse CMs, and IKur densities are significantly lower in female mouse CMs than in male CMs [61]. The inward rectifier K+ current (IK1) is activated during and after the repolarization phase to ensure terminal repolarization and stable RMP. The IK1 densities are significantly smaller in female guinea pig [59] and rabbit CMs [67]. Moreover, NCX current (INCX) densities are similar in sexes in pig [68] and rabbit atrial CMs [69] and CMs from the apex of a rabbit heart [70], but INCX densities are higher in female CMs from the base of a rabbit heart [70]. Verkerk et al. obtained human CMs from failing hearts and showed that female CMs have significantly longer APD and greater susceptibility to early after depolarization (EAD) [71]. In addition, female CMs express higher ICaL densities but smaller Ito densities than male CMs, but the APA, dV/dt, and RMP show no sex differences [71]. Gaborit et al. obtained CMs from healthy human hearts and demonstrated that female CMs express lower levels of various genes (human ether-a-go-go-related gene, mink, Kir2.3, Kv1.4, KChIP2, SUR2, and Kir6) responsible for cardiac repolarization [72]. Overall, many animal and human experiments have shown that APDs from female CMs are longer than that from male CMs, consistent with the clinical observation that women have longer QTc intervals. The cardiac AP results from the complex interplay of time- and voltage-dependent inward and outward ionic currents during AP’s various phases. The difference in APD between male and female CMs results from the unique composition of the ionic currents governing the APD. Furthermore, female CMs are more susceptible to EADs in response to increased cycle length, supporting the clinical observation that female sex is an independent risk factor for TdP [71].
Adrenergic receptor activation is the primary mechanism that increases cardiac performances under stress. Upon beta-adrenergic activation, adrenergic receptors couple with G proteins, leading to adenylyl cyclase activation and secondary-messenger cyclic adenosine monophosphate (cAMP) production. Subsequently, cAMP activates protein kinase A (PKA), which promotes phosphorylation in various substrates, such as (i) L-type Ca2+ channel, which increases Ca2+ entry into cardiomyocytes; (ii) PLB, which accelerates Ca2+ sequestration into the SR and cardiac relaxation, and (iii) troponin I and C proteins, which reduce myofilament sensitivity to Ca2+ [73]. Male CMs exhibit higher beta-adrenergic receptor density and basal intracellular cAMP levels than female CMs; therefore, they have larger cAMP production upon beta-adrenergic stimulation [54, 74]. Furthermore, beta-adrenergic stimulation causes a larger augmentation of ICaL current densities, leading to increased Ca2+ release from SR, although basal SR Ca2+ content is similar between male and female CMs [49, 54, 56, 74]. The enhanced response to beta-adrenergic stimulation due to augmented intracellular signaling explains the larger positive inotropic effect in male CMs, as observed in both ventricular [74] and atrial CMs [75]. Additionally, the reduction in the decay time constant of the Ca2+ transient caused by beta-adrenergic stimulation is higher in the male heart than in the female heart, revealing that male hearts have a more pronounced lusitropic effect (i.e., relaxation) [76]. Even though the male and female hearts have similar Ca2+ transient durations after beta-adrenergic stimulation, APD reduction is less prominent in the female heart, as shown in the simultaneous optical mapping in ventricular AP and Ca2+ transient recording [76]. Voltage-clamp study revealed that beta-adrenergic stimulation induces smaller IKs in female CMs, suggesting the underlying mechanism behind the reduced capability of female CMs to further decrease in APD upon beta-adrenergic stimulation [60]. In conclusion, lower beta-adrenergic receptor density and/or lower intracellular cAMP level in female CMs attenuates the PKA phosphorylation of Ca2+ handling proteins and ion channels, which is the putative mechanism behind the limited positive inotropic effect under beta-adrenergic stimulation. The preserved beta-adrenergic regulation might be associated with reduced arrhythmic activity, explaining why women are less prone to severe arrhythmias [76].
The effect of sex hormones on cardiovascular physiology has been widely studied. CMs express sex hormone receptors, indicating that these hormones have direct cardiac effects [77–79]. At cellular level, sex hormones regulate various voltage-gated ion channels (Fig. 2) and also intracellular Ca2+ handling, thereby altering the cardiac repolarization [80].
The effects of sex hormones on ion channels depend mainly on whether the hormone is present acutely or chronically (Fig. 3). The acute exposure of testosterone within the physiological range rapidly reduces the open probability of single ICaL and lowers ICaL current density [81]. It also decreases the APD mainly by enhancing IKs and suppressing ICaL via nitric oxide synthase 3 activation and nitric oxide production through a nongenomic pathway [82]. In contrast, the chronic treatment of testosterone increases the single-channel activity of ICaL and ICaL current density [81] and also INa current density [65]. Furthermore, testosterone upregulates the expression of genes for ICaL and NCX [83]. In female guinea pig CMs, the acute application of progesterone at 100 nM/L reduces APD mainly by enhancing IKs and inhibiting ICaL via a nongenomic pathway [84]. However, the supraphysiological concentration of progesterone (1–30 μM) in Langendorff-perfused female rabbit hearts exhibits a biphasic effect; it prolongs monophasic APD at lower concentrations (1–3 μM) but shortens at higher concentrations (10–30 μM) [85]. Moreover, high estradiol concentrations (1–30 μM) prolong the APD in a concentration-dependent manner by affecting one or more of the ionic currents, including ICaL, IKr, IKs, Ito, and IK1 [85, 86]. Estradiol can directly interact with ion channels without involving the membrane-associated estrogen receptors [87]. Of note, high estradiol concentrations prolong the APD by inhibiting IKr and IKs in female guinea pig CMs [88], but shorten it by inhibiting ICaL and delaying the recovery time of ICaL in male guinea pig CMs [89]. Therefore, while studying the effect of hormones on CMs, the sex of the animal from where CMs are isolated, should be considered. The estrogen of 300 nM concentrations shortens the APD by enhancing IKs and suppressing ICaL [90], but its physiological concentration (1 nM) prolongs the APD by suppressing IKr, with little or no effect on IKs and ICaL [90]. Furthermore, the incubation of CMs with estradiol (1 nM) enhances NCX current [70]. The plasma estrogen concentration is one of the key players in determining outward K+ current density thus, it affects the ventricular repolarization, as confirmed by reduced total outward K+ current densities and downregulation of K+ channel transcript level in female CMs [91, 92]. The long-term deficiency of ovarian hormones after ovariectomy results in higher ICaL, creating a larger “window” current that facilitates the increased occurrence of arrhythmias with and without isoprenaline in female guinea pig CMs [93]. However, in same study, the estradiol replacement prevents arrhythmias in CMs from ovariectomized guinea pig [93]. In conclusion, sex hormones directly regulate ion channels and alter APD depending on their concentrations and whether the exposures are short or long. The long-term exposure of sex hormones acts via a genomic pathway, in which sex hormones bind to sex hormone receptors, translocate into nucleus and lead to the transcriptional regulation of ion channels [81, 94]. In addition, sex hormones can also directly targets the ion channels in a receptor-independent manner and referred as nongenomic regulation [81, 94]. Nongenomic actions can be distinguished from genomic effects by a more rapid onset (seconds to minutes), take place outside the cell nucleus via the activation of intracellular signaling including endothelial nitric oxide synthase and mitogen-activated protein kinase, and the fact that observed effects may not be blocked by sex hormone receptor antagonists [81, 94].
The sex hormones also modulate the intracellular Ca2+ handling and contractile properties of CMs depending on acute or chronic application of hormones (Fig. 3). Pretreatment of rat CMs with testosterone (100 nM) for 24–30 h increases the peak Ca2+ transients, frequency of Ca2+ sparks, and fractional shortening thus it improves the CM contractility without altering the SR Ca2+ load [81]. In contrast, the acute testosterone application decreases the frequency of Ca2+ sparks and reduces contractility in testosterone-pretreated rat CMs [81]. When testosterone (100 nM) is acutely applied to non-testosterone-pretreated neonatal rat CMs, the intracellular Ca2+ release from the intracellular storage is increased by elevating the inositol 1,4,5-trisphosphate level [95]. In isolated male rat ventricular CMs, 24-h exposure of testosterone (1 μM) increases the peak shortening and relaxation velocity and decreases the time to peak shortening; conversely, acute exposure has no effect on the contraction and relaxation properties [79]. Animal experiments on gonadal testosterone withdrawal (GDX) have been conducted to study the influence of endogenous gonadal testosterone on the regulation of the Ca2+ handling in the heart. Two weeks after GDX reduced Ca2+ transient amplitude and peak shortening of CMs, and slowed down the Ca2+ transient decay is observed in male rats; however, the effects were completely reversed by testosterone replacement [96]. The reduction of the mRNA levels for the genes of NCX and L-type Ca2+ channels is observed [97]. However, long-term testosterone deficiency (10 weeks after GDX) increases the NCX1 expression, but it does not change the expression of SERCA2a, CSQ2, or total PLB in male mice [98]. If male rats receive testosterone replacement 9 weeks after GDX, the contractility and Ca2+ transient amplitude increase as a result of high Ca2+ release from SR and more efficient Ca2+ reuptake through SERCA and removal via NCX [99]. Moreover, acute estradiol application at supraphysiological concentrations (10–30 μM) reduces the contraction and Ca2+ transient amplitude [89]; the estrogen-induced negative inotropic effect is not mediated via estrogen receptors in the membrane, but it is more on the direct inhibition of ICaL [87]. Ovariectomy in female animals results in higher Ca2+ transient amplitude, faster rise and decay of Ca2+ transient, higher SR Ca2+ content, greater ECC gain, and higher spark frequencies in isolated CMs [93, 100, 101]. Ovariectomy increases SR Ca2+ storage and promotes SR Ca2+ loading, thus resulting into the increased Ca2+ spark frequency in ovariectomized CMs [101]. Furthermore, CMs from an ovariectomized animal heart exhibit decreased peak shortening, reduced maximum shortening/relengthening velocity, and prolonged shortening/relengthening duration [102]. The slow intracellular Ca2+ clearing and elevated resting intracellular Ca2+ levels caused by SERCA/PLB protein downregulation may underlie the mechanism behind the estrogen deficiency-induced altered mechanical and intracellular Ca2+ homeostasis [102, 103]. The estradiol [93, 102, 104] or progesterone [103] replacement in ovariectomized female animals ameliorates the changes in contractile function and intracellular Ca2+ handling properties of CMs caused by ovariectomy. Estrogen deficiency promotes the intracellular Ca2+ dysregulation, reduces myofilament Ca2+ sensitivity, and alters the contractile function, causing the formation of a more proarrhythmic substrate in an aging female heart. Taken together, the sex hormones directly modulate the myocardial function which partially explain the sex-based differences in myocardial function and may help to determine the differential incidence and outcomes of cardiac disease conditions in men and women.
Women have a higher risk of developing a special type of ventricular tachycardia, TdP, associated with adverse effects of drugs that prolong ventricular repolarization time. This is due to several sex-specific differences in physiology, pharmacokinetics, and/or pharmacodynamics [105]. Typically, women not only have a longer QTC at baseline, but also have more prolonged QT intervals after taking drugs known to prolong cardiac repolarization (e.g., Class I and III antiarrhythmic drugs), placing them at a higher risk for TdP [105]. However, the detailed mechanism of TdP remains poorly understood. One possible explanation could be that drug concentrations are higher in women because of their smaller body sizes. However, quinidine (Class Ia antiarrhythmic drug) causes greater QTC prolongation in women than in men, although its serum concentrations and pharmacokinetics are not different between men and women [106, 107]. Furthermore, d,l-sotalol (class III antiarrhythmic drugs) increases the mean heart rate in women, who also face a threefold increased risk of developing TdP [108]. One study demonstrated that d,l-sotalol could induce extreme JTC prolongation in women regardless of the baseline JTC [109]. In another study investigating dofetilide (class III antiarrhythmic drug) administration and creatinine clearance in age-matched men and women, more than half of women discontinued the drug or underwent dose reduction because of significant QTC prolongation [110]. The Digitalis Investigation Group evaluated the efficacy of digoxin therapy in patients with heart failure (HF) and found that women taking digoxin had a higher mortality rate, although women with HF, but without digoxin medication had a lower mortality risk than men [111]. The endogenous sex hormones also have a role in drug-induced QT prolongation. After ibutilide (class III antiarrhythmic drug) infusion, the QTC prolongation and TdP occurrence was higher in women than in men [112, 113]. However, the progesterone level inversely correlates with the mean ibutilide-induced QTC prolongation and ibutilide-induced QTC prolongation is shortest during the luteal phase compared with that during the menstrual and ovulation phases, suggesting the possible protective effect of progesterone [113]. Furthermore, drug response in genetic cardiac diseases also differs with sex. During beta-blocker treatment of LQT type 1 patients with similar QTC intervals, men have shorter QTC intervals than women [114]. Moreover, beta-blocker therapy reduces the cardiac event rates in women with LQT type 3 syndrome, but its efficacy in men is still inconclusive because of the low number of prior cardiac events [115]. Patients with Na+ loss-of-function condition exhibit cardiac conduction disturbances and could provoke Brugada syndrome. Women with Brugada syndrome show greater conduction intervals and QTC in response to Na+ channel blockers, even though their baseline ECG parameters are similar to those of men [116]. Several in vitro studies have been performed to understand sex-specific differences in drug responses, and the results are similar to those of in vivo studies. Animal studies showed that a female heart is more prone to TdP development subsequent to QT prolongation in response to E-4031 (IKr blocker) and 4-aminopyridine (Ito blocker) [66, 117]. Dofetilide (IKr blocker) administration induces greater APD prolongation, EAD incidence, and repolarization dispersion in CMs from female hearts [63]. In addition, CMs from a gonadectomized male heart have higher dofetilide-induced APD prolongation and the presence of EADs than those from a control male heart. Testosterone replacement in gonadectomized animals diminishes the effect of IKr blockers on QT and APD, whereas estrogen replacement increases blockers’ adverse effects such as a higher magnitude of drug-induced prolongation and EAD incidences [63, 92, 118]. In conclusion, women have larger drug-induced QTC prolongation than men, and sex hormones play an important role in cardiac repolarization. All these results suggest that female sex is an independent risk factor for increased cardiac side effects of various drugs, and women have especially an increased risk of developing TdP. Therefore, the QT interval in female patients receiving drugs with potential effects on cardiac repolarization should be monitored closely.
Considering the limited human heart material for research, most of our knowledge about sex-specific differences relies on animal experiments. However, cardiac ion channels, Ca2+ handling components, and estrous cycle differ among species. Although animal experiments reveal interesting results, direct extrapolation of animal data to humans is not always possible because of several uncertainties. Insufficient human-based data extensively limit our understanding of sex-specific electrophysiology, and further studies with human-based experiments are urgently required. The recent methodological and technical advancements with iPSC technology offer a robust and human-based platform for studying human cardiomyocyte physiology and pathophysiology [119, 120] as well as for screening pharmacological compounds [121]. HiPSCs provide an unlimited source of human CMs, offering an accessible and robust platform to study sex-specific drug responses. One of their advantages is that they carry donor-specific genetic information, enabling researchers to study the mechanisms of sex differences in different genetic cardiac diseases. One limitation of currently used hiPSC-CMs is that their phenotype resembles embryonic CMs [122] warranting caution in interpretation of the results. However, hiPSC-CMs have been successfully used to investigate the influence of genetics and sex hormones on the gene expression of cardiac ion channels and function of CMs. Papp and coworkers were the first to study the effect of sex hormones in hiPSC-CMs; they found that estradiol increases the ICaL and NCX current densities in hiPSC-CMs obtained from a female donor, and it slightly increases the ICaL in male-origin hiPSC-CMs but does not affect current NCX densities [123]. In clinical studies, women are more vulnerable to drug-induced QT prolongation, and similarly, hiPSC-CMs from women are more sensitive to rate-corrected field potential duration (FPDc) prolongation and arrhythmia incidence induced by IKr blockers [124, 125]. Similar to the result obtained from healthy human hearts [72], male-origin hiPSC-CMs displayed KCNE1 upregulation [125], resulting in a greater response against IKs blockers [124, 125]. In addition, estradiol administration increases FPD in female-origin hiPSC-CMs, whereas testosterone shortens FPD/APD in male-origin hiPSC-CMs [124, 126], consistent with clinical observations in which the QTC interval is shortened in males after puberty [29]. In conclusion, the intrinsic properties based on sex reveal fundamentally different responses in male and female-derived cells, making the hiPSC-CMs a useful model to study the mechanisms of sex differences and they help in predicting drug-induced arrhythmias in men and women. Furthermore, hiPSC-CMs can be maintained in a culture for a long time, making hiPSC-CMs a suitable model to study both the acute and the chronic effects of sex hormones. Sex-specific hiPSC-CMs enable us to control the environmental variables to study the genetic, epigenetic, and/or hormonal variables. Additionally, experimental variables such as the timing and dosage of hormonal application can also be optimized. Therefore, the development of male and female cardiac models helps achieve the goal of personalized medicine based on innovations in human stem cell technology without ethical constraints.
Both animal and human studies demonstrate that sex differences exist and range from gene expression to cardiac physiology. The presentation and progression of different cardiac diseases also differ in sex; men and women with cardiac diseases having the same gene mutations often have different clinical outcomes. Arrhythmia incidence is also different between men and women, and female sex itself is an independent risk factor a special ventricular arrhythmia, TdP, due to both genetic and acquired LQTS. Sex differences are an important factor to be considered when analyzing symptoms, searching for diagnosis, and optimizing the treatment for cardiac diseases. Sex-specific differences in CMs and cardiac functions are most prominent after puberty demonstrating that sex hormones play an important role for causing such differences. However, detailed mechanisms underlying these sex differences are still unclear. In many studies, sex is not considered at all, causing misinterpretation of the actual results. Increasing the number of female participants in all types of studies, including animal experiments as well as clinical trials, and incorporating the sex as a factor in the analysis are needed to broaden our understanding of sex differences. Furthermore, endogenous sex hormone levels also change with age in both men and women. Therefore, experimental studies should incorporate not only sex, but also the sex hormone levels. Currently, the data on sex-based electrophysiological and pharmacological responses from human-based models are limited and most of our understanding is still based on animal models. Although animal studies have delivered important information about sex differences and drug responses, such results cannot be directly translated in humans, and many drug therapies have still poorer outcomes in clinical studies than what was expected in animal studies [9]. One of the main reasons for this discrepancy is that cardiac physiology in animals differs from that in humans. Studies about sex-specific differences of new drugs should be performed in the human-based models with iPSC-derived male and female CMs. In addition, the sex differences should be consolidated into all phases of drug development to introduce safe and efficient drug treatment. Studies involving humans are expensive and can pose participants at an increased risk for side effects. Human-based heart biopsies are also limited, yielding a narrower human-based platform for research. Recent advances in hiPSC-CMs offer a robust human-based platform not only for a large drug screening, but also for studying the mechanisms of sex differences at cellular and tissue levels. |
|
PMC9647969 | Dengfang Guo,Qingling Wang,Jiancheng Huang,Zhanglin Hu,Chun Chen,Chun Zhang,Feng Lin | Downregulation of miR-451 in cholangiocarcinoma help the diagnsosi and promotes tumor progression | 09-11-2022 | miR-451,Cholangiocarcinoma,Early detection,Prognosis prediction,Biological function | Background Cholangiocarcinoma is a kind of invasive malignant tumor followed by hepatocellular carcinoma. miR-451 was suggested to function as regulator in various human tumors, but its role in mediating tumor progression and predicting the prognosis of cholangiocarcinoma remains unknown. The clinical significance and biological function of miR-451 in cholangiocarcinoma were assessed in this study. Results The tissue and serum expression of miR-451 was decreased in cholangiocarcinoma compared with corresponding normal samples. The downregulation of miR-451 was associated with the progressive TNM stage and positive lymph node metastasis of patients. miR-451 was identified to be an indicator of the diagnosis and prognosis of cholangiocarcinoma distinguishing cholangiocarcinoma patients from healthy volunteers and predicting the poor outcome of patients. miR-451 also served as a tumor suppressor negatively regulating the cellular processes of cholangiocarcinoma. Conclusions miR-451 played a vital role in the early detection and risk prediction of cholangiocarcinoma. miR-451 also suppressed the progression of cholangiocarcinoma, which provides a potential therapeutical target for cholangiocarcinoma treatment. Supplementary Information The online version contains supplementary material available at 10.1186/s12860-022-00445-2. | Downregulation of miR-451 in cholangiocarcinoma help the diagnsosi and promotes tumor progression
Cholangiocarcinoma is a kind of invasive malignant tumor followed by hepatocellular carcinoma. miR-451 was suggested to function as regulator in various human tumors, but its role in mediating tumor progression and predicting the prognosis of cholangiocarcinoma remains unknown. The clinical significance and biological function of miR-451 in cholangiocarcinoma were assessed in this study.
The tissue and serum expression of miR-451 was decreased in cholangiocarcinoma compared with corresponding normal samples. The downregulation of miR-451 was associated with the progressive TNM stage and positive lymph node metastasis of patients. miR-451 was identified to be an indicator of the diagnosis and prognosis of cholangiocarcinoma distinguishing cholangiocarcinoma patients from healthy volunteers and predicting the poor outcome of patients. miR-451 also served as a tumor suppressor negatively regulating the cellular processes of cholangiocarcinoma.
miR-451 played a vital role in the early detection and risk prediction of cholangiocarcinoma. miR-451 also suppressed the progression of cholangiocarcinoma, which provides a potential therapeutical target for cholangiocarcinoma treatment.
The online version contains supplementary material available at 10.1186/s12860-022-00445-2.
Cholangiocarcinoma is a malignant tumor in the hepatic second to hepatocellular carcinoma. Cholangiocarcinoma is a group of epithelial cancers mentioning the intrahepatic, perihilar, and distal biliary tree [1]. Owing to the invasive characteristics of cholangiocarcinoma, the disease development is uncontrolled, and there was a lack of obvious clinical characteristics and risk factors, which makes patients always diagnosed at an advanced stage [2]. Although the diagnosis and therapy technology have been developed, the incidence and mortality of cholangiocarcinoma are still increasing [3]. Identifying effective biomarkers to diagnose cholangiocarcinoma at an early stage and predict disease development could ameliorate patients’ clinical outcomes and improve the cure rate of cholangiocarcinoma. microRNAs (miRNAs) have been demonstrated to serve as indicators in the diagnosis, prognosis, and progression of human cancers [4]. Binding with the 3’UTR of relevant mRNAs is the major characteristic of miRNA, by which miRNAs mediate the cycle progression, apoptosis, and growth of cancer cells, and therefore participate in the occurrence and development of tumors [5]. The dysregulation of different miRNAs always implies their functional role in human diseases. For example, increased miR-25 was correlated with malignant development and poor survival of cholangiocarcinoma patients [6]. miR-186 was disclosed to be downregulated and suppress the proliferation, migration, and invasion of cholangiocarcinoma cells [7]. In the previous identification of differently expressed miRNAs which were considered candidate biomarkers of cholangiocarcinoma progression, miR-451 was found to be downregulated [8]. It has been reported that miR-451 not only regulated the biological function of tumor cells, but also regulated the physiological and pathological processes of humans, and it was also considered a novel therapeutic target of human cancers [9, 10]. miR-451 also shows significant diagnostic value in ischemic stroke and papillary thyroid carcinoma [11, 12]. In colorectal cancer, miR-451 inhibited cell growth and metastasis via targeting MIF [13]. While the specific function of miR-451 remains unclear. This study aimed to validate the expression of miRR-451 in cholangiocarcinoma and disclose its potential in clinical diagnosis and risk prediction of cholangiocarcinoma.
The expression of miR-451 was significantly lower in the serum of cholangiocarcinoma than that in the serum of healthy volunteers (P < 0.001, Fig. 1A). In the collected tissues, miR-451 was significantly downregulated in tumor tissues in comparison with the matched normal tissues (P < 0.001, Fig. 1B). Consistently in cholangiocarcinoma cell lines, the downregulation of miR-451 was also observed and showed a dramatic difference with normal cells (P < 0.001, Fig. 1C). Patients were partitioned into a high miR-451 group and a low miR-451 group based on the average expression level of miR-451 in serum and tissues of cholangiocarcinoma. The relatively low expression of miR-451 in tissues showed a significant association with the TNM stage (P = 0.014) and lymph node metastasis status (P = 0.015) of patients (Table 1). Consistently, a close association was also found between the serum miR-451 expression and the TNM stage (P = 0.022) and lymph node metastasis status (P = 0.042) of patients (Table 1).
miR-451 could distinguish cholangiocarcinoma patients from healthy volunteers with the AUC value of 0.864 of the ROC curve (sensitivity = 0.859, specificity = 0.774, Fig. 2A). Additionally, in cholangiocarcinoma patients, the downregulation of mir-451 was associated with the worse survival of patients (log-rank P = 0.021, Fig. 2B). Moreover, Cox regression analysis further demonstrated the prognostic value of miR-451. miR-451 and the TNM stage served as independent prognostic indicators of patients with HR values of 2.651 and 2.277, respectively (Table 2).
Due to the relatively high sensitivity of CCLP1 and HuCCT1 cells to the downregulation of miR-451, these two cells were selected for the following in vitro cell experiments. miR-451 was overexpressed by the transfection of miR-451 mimic and silenced by the transfection of miR-451 inhibitor in CCLP1 and HuCCT1 cell (P < 0.001, Fig. 3A). In transfected cells, miR-451 overexpression markedly suppressed cell proliferation, and miR-451 knockdown notably promoted CCLP1 and HuCCT1 cell proliferation (P < 0.05, P < 0.01, Fig. 3B). Additionally, the migration of CCLP1 and HuCCT1 cells was also inhibited by miR-451 overexpression and accelerated by the silencing of miR-451 (P < 0.001, Fig. 3C, Fig. S1). Similarly, the overexpression of miR-451 repressed the invasion of CCLP1 and HuCCT1 cells and the miR-451 knockdown showed a dramatically enhanced effect on cell invasion of cholangiocarcinoma (P < 0.001, Fig. 3D, Fig. S1).
ATF2 was predicted to bind with miR-451 with several binding sites, and the luciferase of ATF2 was suppressed by the overexpression of miR-451 and enhanced by miR-451 knockdown (Fig. 4A). While the expression of ATF2 was also negatively regulated by miR-451 (Fig. 4B).
Significant dysregulation of miRNAs in tumors always insinuates their potential functional roles in human diseases. miR-451 has been widely reported to possess abnormal expression and participate in the progression of human diseases. For example, miR-451 was identified as the most strongly downregulated miRNA in non-small cell lung cancer (NSCLC) and showed significant association with poor differentiation, advanced clinical stage, and positive lymph node metastasis of patients [14]. The abnormal expression of miR-451 was observed in colorectal cancer, gastric cancer, and bladder carcinoma [15–17]. miR-451 has been revealed to be downregulated in hepatocellular carcinoma (HCC) and was involved in the tumor progression and disease development of patients [18]. Both HCC and cholangiocarcinoma are derived from the substance of the hepatic parenchyma and are known as primary liver cancer [19]. In a previous study, miR-451 was demonstrated as a downregulated miRNA in cholangiocarcinoma [8]. miR-451 was speculated to be involved in the pathogenesis and development of cholangiocarcinoma, which lacked available data. The consistent downregulation of miR-451 in the cholangiocarcinoma was observed in the present study, and its significant association with TNM stage and lymph node metastasis status of patients, two major indicators of cholangiocarcinoma progression, was also dugout, suggesting its involvement in cholangiocarcinoma development. miR-451 was also demonstrated to participate in the development of many other cancers for its close relationship with the clinicopathological characteristics of patients. For instance, miR-451 was significantly correlated with the FIGO stage and lymph node metastasis of ovarian cancer patients, and it also predicted patients’ poor prognosis, indicating its significance in cancer progression and prognosis [20]. The significant association between miR-451 and lymph node metastasis was also observed in thyroid cancer and miR-451 was remarkably upregulated in lymph node metastasis tissues compared with tissues without lymph node metastasis [21]. The diagnostic value of miR-451 has been illustrated in various human solid tumors in previous studies, such as gastric cancer, breast cancer, and renal cell carcinoma [22–24]. miR-451 was also revealed to predict the recurrence of colorectal cancer and gastric cancer [25, 26]. Here, the downregulation of miR-451 could also differentiate cholangiocarcinoma patients from healthy volunteers, indicating that miR-451 could as serve as a diagnostic biomarker of cholangiocarcinoma. While the function of miR-451 in the prediction of cholangiocarcinoma recurrence needs further representative samples to estimate. Previously, miR-451 was disclosed to induce G0/G1 phase arrest and the apoptosis of glioblastoma, but the molecular mechanism was controversial [27, 28]. The inhibitory effect of miR-451 on cellular processes of osteosarcoma was revealed [29]. In vitro, the dysregulation of miR-451 affected the proliferation, migration, and invasion of cholangiocarcinoma cells. Specifically, the miR-451 overexpression inhibited cell growth, migration, and invasion, whereas the knockdown of miR-451 promoted the cellular processes of cholangiocarcinoma. These results leaked out that miR-451 functioned as a tumor suppressor during the progression of cholangiocarcinoma. Although the clinical significance and biological function of mir-451 has been revealed, the concrete mechanism underlying these functional roles is also an important part. Several molecules have been demonstrated as the direct targets of miR-451 during its biological function in many other tumors. For example, PGE2 has been reported to mediate the inhibition of osteosarcoma cellular processes by miR-451, and it was found to promote the development of cholangiocarcinoma [30, 31]. ATF2 was found to be negatively regulated by miR-451 through the results of luciferase reporter and expression validation, which is consistent with previous studies [32]. Therefore. ATF2 was speculated to mediate the suppressor role of miR-451 in cholangiocarcinoma. However, the identification of a single miRNA biomarker neglects the potential of other miRNAs with high scores. Recently, the establishment of miRNA signatures has become a research hot point in cancer research. Therefore, future studies would focus on the significance of miR-451 combining with other miRNAs to establish potential signatures.
In conclusion, downregulated miR-451 in cholangiocarcinoma showed a close association with the disease development and clinical prognosis. Additionally, miR-451 could distinguish cholangiocarcinoma patients from healthy volunteers with high specificity and sensitivity and it also acted as a tumor suppressor that negatively regulated the proliferation, migration, and invasion of cholangiocarcinoma cells.
This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Mindong Hospital Affiliated to Fujian Medical University. One hundred and fifty-nine patients diagnosed with cholangiocarcinoma and sixty-four healthy volunteers who received routine physical examinations at Mindong Hospital of Ningde City were included in this study during 2013–2015. The serum samples, tumor tissues, and matched normal tissues were collected after receiving informed consent from every participator. While only the serum samples were collected from healthy individuals. The cholangiocarcinoma patients were followed up for five years to obtain their survival status after surgery.
It is a two-step process of miR-451 expression assessment. Total RNA was isolated and used to synthesize cDNA with the TaqMan Advanced miRNA cDNA Synthesis Kit (Thermo Fisher Scientific, USA). cDNA was diluted and mixed with the SYBR Green master reagent and primer mix. The PCR process was performed with an ABI 7500 system (Applied Biosystems, USA). The 2−ΔΔCt method was used to calculate the relative expression of miR-451 with GAPDH as the internal reference.
Cholangiocarcinoma cell lines (CCLP1, HuCCT1, SNU1196, and KKU-100 cells, ATCC) and normal cholangiocyte H69 cells (ATCC) were cultured in a DMEM culture medium. Cell culture was conducted in a constant temperature incubator at 37°C with 5% CO2. Cells reached the logarithmic period were transfected with miR-451 mimic (5’-AAACCGUUACCAUUACUGAGUU-3’), miR-451 inhibitor (5’-AACUCAGUAAUGGUAACGGUUU-3’), or corresponding negative controls (mimic NC and inhibitor NC) with the help of Lipofectamine 2000 (Invitrogen, USA).
Cells (1× 105 cells/well) were seeded into 96-well plates and incubated with DMEM culture medium for a specific period. Then, the CCK8 reagent was added to each well and incubated with the mixture for 1 h. OD450 of each well was detected with the employment of a microplate reader (Thermo Fisher Scientific, USA). The experiments were performed three times to obtain the mean values.
A total of 2× 104 cells/well were seeded into the upper chamber of the 24-well transwell chambers with a pore size of 8 µm (Corning, USA). The upper chamber was supplied with a serum-free culture medium, while the FBS-containing medium was placed in the bottom chamber. The chambers were incubated at 37°C for 24 h, and the migrated and invaded cells on the lower surface were fixed and stained. The number of cells was counted with the help of a microscope (Olympus, Japan).
The wild-type vector was established by cloning the binding sites between miR-451 and ATF2, while the mutant-type vector was constructed with the point mutations. The vectors were co-transfected with miR-451 mimic, inhibitor, or negative controls into the CCLP1 cell, and the relative luciferase activity of ATF2 was detected after 48 h of transfection using the Dual-luciferase repoter Assay System (Promega, USA).
All data were represented as mean value ± standard deviation obtained from at least three independent experiments. The difference between groups was analyzed by the student’s t-test and one-way ANOVA. The difference in the expression of miR-451 between healthy volunteers and cholangiocarcinoma was used to evaluate the diagnostic value of miR-451 with the help of the ROC curve. While the prognostic value of miR-451 was assessed with the Kaplan-Meier and Cox regression analysis. P < 0.05 was considered to be statistically significant.
Additional file 1: Figure S1. Representative images of Transwell assay inevaluating cell migration and invasion. |
|
PMC9647978 | Sara A. Knaack,Daniel Conde,Sanhita Chakraborty,Kelly M. Balmant,Thomas B. Irving,Lucas Gontijo Silva Maia,Paolo M. Triozzi,Christopher Dervinis,Wendell J. Pereira,Junko Maeda,Henry W. Schmidt,Jean-Michel Ané,Matias Kirst,Sushmita Roy | Temporal change in chromatin accessibility predicts regulators of nodulation in Medicago truncatula | 09-11-2022 | Nitrogen fixation,Nodulation,Symbiosis,Chromatin accessibility,Transcriptome and chromatin dynamics,Gene regulatory network,Cis-regulatory elements,Machine learning,Medicago | Background Symbiotic associations between bacteria and leguminous plants lead to the formation of root nodules that fix nitrogen needed for sustainable agricultural systems. Symbiosis triggers extensive genome and transcriptome remodeling in the plant, yet an integrated understanding of the extent of chromatin changes and transcriptional networks that functionally regulate gene expression associated with symbiosis remains poorly understood. In particular, analyses of early temporal events driving this symbiosis have only captured correlative relationships between regulators and targets at mRNA level. Here, we characterize changes in transcriptome and chromatin accessibility in the model legume Medicago truncatula, in response to rhizobial signals that trigger the formation of root nodules. Results We profiled the temporal chromatin accessibility (ATAC-seq) and transcriptome (RNA-seq) dynamics of M. truncatula roots treated with bacterial small molecules called lipo-chitooligosaccharides that trigger host symbiotic pathways of nodule development. Using a novel approach, dynamic regulatory module networks, we integrated ATAC-seq and RNA-seq time courses to predict cis-regulatory elements and transcription factors that most significantly contribute to transcriptomic changes associated with symbiosis. Regulators involved in auxin (IAA4-5, SHY2), ethylene (EIN3, ERF1), and abscisic acid (ABI5) hormone response, as well as histone and DNA methylation (IBM1), emerged among those most predictive of transcriptome dynamics. RNAi-based knockdown of EIN3 and ERF1 reduced nodule number in M. truncatula validating the role of these predicted regulators in symbiosis between legumes and rhizobia. Conclusions Our transcriptomic and chromatin accessibility datasets provide a valuable resource to understand the gene regulatory programs controlling the early stages of the dynamic process of symbiosis. The regulators identified provide potential targets for future experimental validation, and the engineering of nodulation in species is unable to establish that symbiosis naturally. Supplementary Information The online version contains supplementary material available at 10.1186/s12915-022-01450-9. | Temporal change in chromatin accessibility predicts regulators of nodulation in Medicago truncatula
Symbiotic associations between bacteria and leguminous plants lead to the formation of root nodules that fix nitrogen needed for sustainable agricultural systems. Symbiosis triggers extensive genome and transcriptome remodeling in the plant, yet an integrated understanding of the extent of chromatin changes and transcriptional networks that functionally regulate gene expression associated with symbiosis remains poorly understood. In particular, analyses of early temporal events driving this symbiosis have only captured correlative relationships between regulators and targets at mRNA level. Here, we characterize changes in transcriptome and chromatin accessibility in the model legume Medicago truncatula, in response to rhizobial signals that trigger the formation of root nodules.
We profiled the temporal chromatin accessibility (ATAC-seq) and transcriptome (RNA-seq) dynamics of M. truncatula roots treated with bacterial small molecules called lipo-chitooligosaccharides that trigger host symbiotic pathways of nodule development. Using a novel approach, dynamic regulatory module networks, we integrated ATAC-seq and RNA-seq time courses to predict cis-regulatory elements and transcription factors that most significantly contribute to transcriptomic changes associated with symbiosis. Regulators involved in auxin (IAA4-5, SHY2), ethylene (EIN3, ERF1), and abscisic acid (ABI5) hormone response, as well as histone and DNA methylation (IBM1), emerged among those most predictive of transcriptome dynamics. RNAi-based knockdown of EIN3 and ERF1 reduced nodule number in M. truncatula validating the role of these predicted regulators in symbiosis between legumes and rhizobia.
Our transcriptomic and chromatin accessibility datasets provide a valuable resource to understand the gene regulatory programs controlling the early stages of the dynamic process of symbiosis. The regulators identified provide potential targets for future experimental validation, and the engineering of nodulation in species is unable to establish that symbiosis naturally.
The online version contains supplementary material available at 10.1186/s12915-022-01450-9.
Legumes such as Medicago truncatula can establish a well-characterized mutualism with nitrogen-fixing rhizobia. Signal exchanges between the host plant and bacteria initiate intracellular infection of host cells, followed by the development and colonization of root nodules [1]. Nodules provide a unique niche for the bacteria and fix nitrogen. Nodulating plants can grow with little to no outside sources of nitrogen and even build soil nitrogen levels for subsequent crops [2]. Hence, understanding symbiotic processes between legumes and rhizobia is extremely valuable for the productivity and sustainability of agricultural systems worldwide. Symbiosis begins with compatible rhizobia detecting flavonoids and isoflavonoids produced by the legume host [3] and subsequent release of lipo-chitooligosaccharides (LCOs) by the bacteria. The host plant perceives LCOs with LysM domain receptor-like kinases heterodimers, such as Nod factor perception (NFP) and LysM domain receptor-like kinase 3 (LYK3) in M. truncatula [4, 5]. LCO perception activates a signaling cascade, involving the plasma membrane-localized LRR-type receptor kinase doesn’t make infections 2/nodulation receptor kinase (MtDMI2/MtNORK), the calcium-regulated calcium channel (MtDMI1), cyclic nucleotide-gated calcium channels, M. truncatula calcium ATPase 8 (MtMCA8), and including the components of the nuclear pore complex [6–8]. The cascade results in oscillations of nuclear calcium concentrations, detectable by the nucleus-localized calcium/calmodulin-dependent protein kinase (CCaMK, MtDMI3 in M. truncatula) [9]. CCaMK activates the transcription factor (TF) interacting protein of DMI3 (MtIPD3/CYCLOPS). Downstream, other TFs are activated, such as nodulation-signaling pathway 1 and 2 (NSP1 and NSP2), Nodule INception (NIN), ethylene response factor required for nodulation 1, 2, and 3 (ERN1, 2, and 3), and nuclear factor YA-1 and YB-1 (NF-YA1 and NF-YB-1) [10, 11]. The coordinated activity of these TFs triggers transcriptional changes [12] essential for infection of the root hair cells (in M. truncatula), nodule organogenesis, and infection of the nodule cortex [10]. These processes require changes in chromatin accessibility [13] on a continuum from closed to open, which are important for cell function [14]. Chromatin reorganization has been shown to regulate a number of processes in plants including photomorphogenesis and flowering [15, 16]. For example, active DNA demethylation by DEMETER (DME) is critical for gene expression reprogramming during nodule differentiation in M. truncatula and the acquisition of organ identity [13]. Also, in M. truncatula, the gene expression level of nodule-specific cysteine-rich genes (NCR) across root nodule zones are correlated with chromatin accessibility [17]. The extent of chromatin accessibility change and impact on transcriptional regulation in rhizobial infection, colonization, and nodule development, remains unknown. Thus, we measured temporal changes in the transcriptome (RNA-seq—ribonucleic acid sequencing) and genome-wide chromatin accessibility (ATAC-seq—assay for transposase-accessible chromatin using sequencing) in response to Sinorhizobium meliloti LCOs in M. truncatula roots (Fig. 1A). To characterize the role of chromatin accessibility and consequent impact on transcriptional dynamics, we applied a novel algorithm, dynamic regulatory module networks (DRMN) [18], to predict gene expression as a function of chromatin accessibility profiles of cis-regulatory features. DRMN results suggest that chromatin accessibility and specific TFs play a critical role in regulating the transcriptional dynamics in response to LCOs.
We profiled the global transcriptomic changes of rhizobium LCO signaling with RNA-seq in M. truncatula using the Jemalong A17 genotype, treated with LCOs purified from S. meliloti. An LCO concentration of 10−8 M was used, as in previous studies [19, 20]. Samples were analyzed for control (t = 0 h) and seven time-point conditions after treatment (15 and 30 min; 1, 2, 4, 8, and 24 h). Principal component analysis (PCA) showed clustering of biological replicates and time-dependent ordering, the first component explaining ~36% of variation (Fig. 1B, Additional file 1: Figure S1). Comparison of expression levels at each time point (relative to control, t = 0 h) revealed 12,839 differentially expressed (DE) genes with significant change in expression (adjusted-P < 0.05), including 7540 and 7051 upregulated and downregulated at one time point relative to control, respectively (Additional file 1: Figure S2A). When comparing any pair of time-points we identified 17,391 DE genes in total (Additional file 1: Figure S2A). Both the statistics (Additional file 1: Figure S2B) and heat-maps of DE genes (Additional file 1: Figure S2C, D) present clear patterns of temporal change. To corroborate these results with previous work on transcriptome dynamics of symbiosis, the identified DEGs were compared to DEGs identified from a published time course data of M. truncatula roots inoculated with rhizobium from Larrainzar et al. [12] (see Additional file 1: Figure E-G). Comparisons were made to DEGs in the following genotypes: Jemalong A17 wild type, LCO-insensitive nfp mutant, infection lyk3 mutant, and LCO-hypersensitive skl mutant. The highest similarity, measured by F-score, to our DEG set was for the mutant genotype most sensitive to LCOs, skl (0.53), and the wild-type (WT) strain (0.40). Marker genes for rhizobium-induced nodulation were upregulated (compared to t = 0 h), including NIN (nodule inception, induced after 15 min, with a maximum induction at t = 1 h), CRE1 (cytokinin response 1, at 4 h, 8 h, and 24 h), ENOD11 (early nodulin 11, highly induced at 8 and 24 h), RPG (rhizobium-directed polar growth, at 4 h), and ERN1 (ethylene responsive factor required for nodulation 1, induced as early as t = 15 min, Fig. 1D). The similarity was lowest for the lyk3 (0.26) and nfp (0.17) mutants (Fig. 1C, see Additional file 1: Figure S2E, F). Furthermore, when comparing individual time points, DE gene sets are most similar for the later time points (Additional file 1: Figure S2G). While our DEGs had the greatest overlap with the skl genotype DEGs, we detected more DEGs compared to Larrainzar et al, likely due to differences in growth conditions (aeroponics versus agar plates) and treatment (purified LCOs versus Sinorhizobium medicae), both inducing a strong LCO response. To examine more complex transcriptome dynamics beyond pairwise DE analysis associated with LCO response, we applied ESCAROLE, a probabilistic clustering algorithm designed for non-stationary time series [21]. The expression data were clustered into seven modules at each time point (very low, low, medium-low, medium, medium-high, high and very high expression, Fig. 2A). Seven modules maximized the log-likelihood and silhouette index (Additional file 1: Figure S3A, B). Next, 12,261 transitioning genes (those changing module assignment over time) were identified, including several implicated in symbiosis (Additional file 1: Figure S3C). Transitioning genes with similar dynamics were clustered using hierarchical clustering, identifying 112 clusters (> = 10 genes each) (Fig. 2B) including 11,612 genes (Methods). Among clusters representing downregulation of expression over time, several were enriched for Gene Ontology (GO) processes implicated in defense responses to bacterium (cluster 293, downregulated from 2–4 h), and the biosynthesis of plant hormones involved in the suppression of nodulation (Fig. 2C). For instance, cluster 299 (downregulated after 2 h) is enriched (hypergeometric test q < = 0.05) for jasmonic acid (JA) biosynthesis and JA response genes, including Coronatine insensitive 1 (COI1), which forms part of the JA co-receptor complex for the perception of the JA-signal [22]. Among the gene clusters upregulated over time, several are implicated in early stages of symbiosis and nodule development. For instance, cluster 186 (induced 2–4 h after LCO treatment; Fig. 2C) is enriched in genes implicated in the regulation of meristem growth, including an Arabidopsis trithorax 3 (ATX3) homolog (MtrunA17Chr4g0005621) and a lateral organ boundaries domain (LBD) transcription factor (MtrunA17Chr4g0043421). ATX3 encodes an H3K4 methyltransferase [23], and LBD proteins are characterized by a conserved lateral organ boundaries (LOB) domain and are critical regulators of plant organ development [24], including lateral roots and nodules [25]. This cluster also contains EPP1 and the cytokinin receptor CRE1, both positive regulators of early nodule symbiosis and development [26, 27]. Other essential regulators of LCO signaling are also found in clusters exhibiting induction under LCO treatment (Additional file 1: Figure S3D), such as DMI1 (cluster 197, Fig. 2C), NIN (cluster 205), NF-YA1 (cluster 177), and the marker of LCO perception ENOD11 (cluster 296). Together, the DE and ESCAROLE analysis showed that M. truncatula response to LCOs is characterized by complex expression dynamics recapitulating several known molecular features of this process.
To study chromatin accessibility changes in a genome-wide manner in response to LCOs, we performed ATAC-seq on samples at all time points matching our RNA-seq time course. Overall, 54–235 million paired-end reads were obtained for each sample, with 46–75% mappable to the (v5) reference genome (Additional file 1: Figure S4). Moreover fragment distributions in ± 1 kbp of the transcription start site (TSS) were examined (Additional file 1: Figure S5, see Methods) and favorably compared to previously published M. truncatula ATAC-seq data [28] (Additional file 1: Figure S6). We next evaluated aggregated chromatin accessibility in gene promoter regions, defined as ± 2 kbp around the TSS, across time. To quantify promoter accessibility, we obtained the mean per base pair (per-bp) read coverage within each region, for each time point. For each time-point, the log-ratio of per-bp read coverage in each promoter was taken relative to the global mean of per-bp coverage, quantile normalized across time points. High consistency was found between promoter signals between technical replicates from each time point based on Pearson’s correlation (Additional file 1: Figure S7, (Pearson correlation 0.965–0.990)) and PCA (Additional file 1: Figure S8A). We partitioned the resulting 51,007 gene promoter accessibility profiles into six characteristic patterns (clusters) using k-means clustering (Fig. 3A, Additional file 1: Figure S8B). Clusters 1 (14,338 genes) and 6 (13,083 genes) exhibit general patterns of decrease and increase in accessibility, respectively, whereas clusters 2–5 (5460–6377 genes) present more transient variation. The correlation of accessibility between time points suggests an overall reorganization of promoter accessibility 1–2 h after the treatment (Additional file 1: Figure S8C). The temporal change in accessibility is evident for the promoters of several nodulation genes, including CRE1, CYCLOPS, and EIN2 (Fig. 3B, Additional file 1: Figure S8D, E; prepared with the Integrative Genome Viewer—IGV) [29]. PCA of the promoter signals showed time-dependent variation (Fig. 3C, Additional file 1: Figure S8A), with the first component explaining > 50% of the variance. We called peaks for each time point using the Model-based Analysis of ChIP-Seq version 2 (MACS2) algorithm [30] (Additional file 1: Figure S9A) and merged peaks across time points with at least 90% overlap into universal peaks (Additional file 1: Figure S9B). Chromatin accessibility peaks showed a similar genomic distribution across time points, with 32.1% of peaks located within 2 kbp upstream and 100 bp downstream of a gene TSS (Additional file 1: Figure S9A-C) and spanning 50.4 Mbp (11.7%) of the M. truncatula (v5) genome. As with the promoter accessibility, clustering accessibility profiles of universal peaks identified distinct patterns of temporal change (Additional file 1: Figure S9D, E). Several of the clusters were associated with known TF motifs (Additional file 1: Figure S9F) and specific types of genomic regions. For example, clusters 1, 2, and 7 had higher proportions of intergenic peaks (hypergeometric test P < 0.05, Additional file 1: Figure S9G). Genes mapped to peaks associated with cluster 2 were enriched for photosynthesis and protein-chromophore linkage (hypergeometric test q < 0.05). Collectively, these results suggest that LCO treatment had a genome-wide impact on chromatin accessibility, prospectively associated with simultaneous change in gene expression.
We evaluated the relationship between gene expression and promoter chromatin accessibility ± 2 kbp around the TSS and universal peaks within 10 kbp upstream and 1 kbp downstream of a gene TSS. Correlating promoter accessibility and gene expression profiles identified 6429 genes with significant correlation (Fig. 4A, P < 0.05 relative to random permutation): 4777 with positive correlation and 1652 with negative correlation (Fig. 4B), representing 17.2% of the 37,356 genes analyzed. Among these were 36 genes with known roles in symbiosis (Additional file 1: Figure S8D), including ERN1, CRE1, LYK10/EPR3, SKL/EIN2, and IDP3/CYCLOPS with positive correlation, and LYK8, ERN2, CAMTA3, and CAMTA4 with negative correlation. We next examined significantly correlated genes (Fig. 4A) and visualized those expression and accessibility profiles as ordered by the promoter accessibility clusters (Fig. 3A), separately for positive and negative correlation (Fig 4B). This revealed robust patterns of consistency between promoter accessibility and expression. Correlating accessibility of universal peaks centered within 10 kbp upstream to 1 kbp downstream of gene TSSs identified 100,722 peak-gene mappings (out of a total 125,140) associated with 28,803 (of 37,536) expressed genes (Fig. 4C, Additional file 1: Figure S9C and G). Peak accessibility was significantly correlated with gene expression in 15.7% of these pairings (Fig. 4C), comparable to the 17.2% (6429) genes with significant correlation between expression and gene TSS accessibility. When considering each gene and only the most correlated peak (28,803 selected pairs), 34.4% (9912 genes) were significantly correlated, including 56 nodulation genes (Fig. 4D). Of these 9912 genes presenting significant correlation, 5735 (57.9%) do not present significant correlation with the corresponding promoter accessibility, indicating a prominent role for distal regulation (> 2 kbp of gene TSS) for these genes. Such peaks were in general more distal from TSS sites than those that presented significant correlation with corresponding TSS accessibility (Kolmogorov-Smirnov/KS test P < 0.05). Finally, the ESCAROLE-defined transitioning gene clusters exhibited coordinated trends between promoter accessibility and gene expression (Fig. 2B, Additional file 1: Figure S3D). Two thousand five hundred one of the 11,612 (21.5%) transitioning genes that could be clustered exhibited significant correlation between their profiles of expression and promoter chromatin accessibility. These results suggest that chromatin accessibility is an important regulatory mechanism in transcriptional response to LCOs.
To better understand how chromatin accessibility contributes to transcriptional changes in rhizobia-plant symbiosis, we applied dynamic regulatory module networks (DRMN) [18] to integrate the RNA-seq and ATAC-seq time course data. DRMN extends the ESCAROLE analysis (which examined only the transcriptome) by modeling the relationship between variation in accessibility and gene expression. DRMN predicts gene expression as a function of regulatory features [31] by first grouping genes into modules based on expression levels (similar to ESCAROLE) and then learning a regulatory program for each module. DRMN uses regularized regression and multi-task learning to incorporate the temporal nature of a data set [32] to simultaneously learn regression models for each module in each time point. We applied DRMN with seven expression modules using two types of features (Fig. 5A, Additional file 2: Tables S1-S4): (1) the aggregated signal of ATAC-seq reads in gene promoters (± 2 kbp of the TSS) and (2) the ATAC-seq signal in genomic coordinates of known motifs within − 10 kbp and + 1 kbp of the TSS. Both feature types represent chromatin accessibility, but the first is independent of the presence of known motifs, whereas the second captures the accessibility of motif sites. Motif features were based on the CisBP v1.2 database for M. truncatula [33] and curated motifs of several known regulators of root nodulation, including CYCLOPS, NSP1, NIN, and the nitrate response cis-element (NRE). Hyper-parameters for DRMN were selected using a grid search and quality of inferred modules (Additional file 1: Figure S10A). The DRMN modules represent statistically different expression levels (Additional file 1: Figure S10B, Kolmogorov-Smirnov test P < 10−300). To assess the extent to which DRMN captures variation in expression, we correlated predicted and measured expression levels (Fig. 5B, Additional file 1: Figure S10A, C). The mean Pearson correlation of predicted and measured values per module was 0.26–0.46 (Additional file 1: Figure S10C) across all modules and time points, the least expressed module being most difficult to predict. Comparing the genes in each module showed that the modules are more similar (F-score 0.88–0.94, Fig. 5C) before and after 2 h, than across this time point (F-score < 0.80), suggesting a significant module reorganization at ~2 h. This is consistent with the general reorganization of promoter accessibility ~1–2 h after the treatment and global expression correlation around 2 h (Additional file 1: Figure S8C). We additionally tested the modules for enrichment of known motifs (Additional file 1: Figure S11, Additional file 2: Table S3) and Gene Ontology (GO) processes (Additional file 1: Figure S12). Several regulators (e.g., KNOX and EDN transcription factor family members) and processes relevant to symbiosis were identified, including nodule morphogenesis, root-hair elongation, and the MAPK cascade, as well as others relating to gene regulation and chromatin organization. Finally, we used the DRMN module assignments to define transitioning gene sets (Fig. 5D, Additional file 1: Figure S13A), similar to those from ESCAROLE (Fig. 2B, Additional file 1: Figure S13B). We identified 79 transitioning gene clusters including 10,176 genes, of which 5332 (>50%) were differentially expressed with DESeq (hypergeometric-test overlap adjusted-P < 0.05), and (8398) 77% were identified in ESCAROLE, indicating consistency between the analyses. We used the DRMN results to prioritize regulators that shape transcriptional response to LCOs. Specifically, we identified regulators whose regression coefficient changed significantly (T-test P < 0.05) between 0–2 and 4–24 h, corresponding to the reorganization of expression modules (Fig. 5C). According to this criterion chromatin accessibility of gene promoters was an important predictor of expression for highly expressed genes (“Promoter ATAC-seq” for modules 5 and 6, Fig. 6A). We also identified the TFs IBM1 (increase in BONSAI methylation 1), ERF1 (ethylene response factor 1), EDN1-3 (ERF differentially regulated during nodulation 1, 2, and 3), EIN3 (ethylene insensitive 3), SHY2 (short hypocotyl 2), ABI4-5 (abscisic acid-insensitive 4 and 5), MTF1 (MAD-box transcription factor 1), and MtRRB15 (type-B response regulator 15), as well as several markers of meristem cells, KNOX and PLT (PLETHORA) protein families as important regulators (Fig. 6B, Additional file 1: Figure S11).
DRMN identified regulators of gene expression dynamics in response to LCOs. Next, we aimed to identify their gene targets. Expression-based network inference is commonly used to define regulator-gene relationships [34] but is challenging with only 8 time-points. To address this, we used the DRMN transitioning gene sets and regulatory motifs selected by a regularized regression method, multi-task group LASSO (MTG-LASSO, where LASSO stands for least absolute shrinkage and selection operator) to define the targets of a gene (Methods). This approach modeled the variation in expression of each of the 79 transitioning gene clusters using a structured sparsity approach, multi-task group LASSO (MTG-LASSO) (SLEP v4.1 package [35], Fig. 7A, Additional file 2: Table S4) to identify regulators (motifs/TFs) for each of the transitioning gene clusters. Here, the same feature data from the DRMN analysis was used. We determined MTG-LASSO parameter settings for all 79 transitioning gene sets, identifying 33 with significant regulatory motif associations (Additional file 1: Figure S14). This generated 122,245 regulatory edges connecting 126 regulatory motifs to 5978 genes (Fig. 7B). Several gene sets exhibit consistent downregulation of expression and corresponding reduction in accessibility of predicted regulatory motifs between 0–2 and 4–24 h (Fig. 7C). For example, gene set 214 (57 genes) shows downregulation of gene expression and reduced motif accessibility (after 4 h) for multiple TFs: MTF1 and BHLH (Fig. 7C). Similarly, gene set 182 was predicted to be regulated by EDN3, MTF1, EIN3, and NF-Box motif and exhibited correlated trends between gene expression and regulatory feature accessibility (Fig. 7C). We prioritized regulators based on the number of targets they were predicted to regulate and found several known and novel regulators in the top-ranking set), such as ERF1 (ethylene response factor 1), EDN1-3 (ERF differentially regulated during nodulation 1, 2, and 3), EIN3 (ethylene insensitive 3), SHY2 (short hypocotyl 2), and MTF1 (MAD-box transcription factor 1) (Fig. 7D).
To experimentally test the involvement of DRMN prioritized transcription factors in root nodule symbiosis, we selected three TFs, EIN3, ERF1, and IAA4-5 which were among the DRMN selected regulators (Fig. 7D). We knocked down the expression of the corresponding genes by RNAi and examined the nodulation phenotype in composite M. truncatula plants (Methods). Knockdown of MtrunA17Chr5g0440591 (EIN3) and MtrunA17Chr1g0186741 (ERF1) significantly lowered the number of nodules produced on the RNAi roots (Fig. 8A, Additional file 1: Figure S15A, P<0.05 from an ANOVA test followed by Tukey’s HSD test post hoc). Knockdown of MtrunA17Chr1g0166011 (IAA4-5) did not alter nodulation relative to the empty vector (EV) control (Additional file 1: Figure S15B, Additional file 2: Table S5). These nodules were all colonized by S. meliloti (Fig. 8B). Together, these results validate the role of MtrunA17Chr5g0440591 (EIN3) and MtrunA17Chr1g0186741 (ERF1) in rhizobium-legume symbiosis, as predicted by DRMN.
The enormous economic and environmental cost of plant nitrogen fertilization motivates efforts towards identifying molecular mechanisms underlying legume perception of nitrogen-fixing bacteria and nodule development. We dissected the gene regulatory network in M. truncatula roots in response to S. meliloti LCOs by jointly profiling the temporal changes in the transcriptome and chromatin accessibility and integrating these data computationally. Extensive changes in the transcriptome are known to occur in Medicago roots in response to rhizobia signals, and we show these changes are accompanied and facilitated by extensive chromatin remodeling. While the overall percentage of accessible chromatin regions remained similar across our time course experiment, regions of accessibility underwent a dramatic shift 1–2 h after treatment. This remodeling appears to anticipate the development of root nodules, which requires stringent temporal and spatial control of gene expression. Chromatin accessibility of gene promoters notably also emerged as a significant predictor of gene expression (Fig. 6). These changes in chromatin accessibility enable and enhance the transcriptional changes required for nodule development by providing regulators access to promoters that may be inactive in other stages of plant development. Correlation was additionally observed between gene expression and promoter chromatin accessibility profiles of several essential regulators of nodulation, including ERN1, CRE1, SKL/EIN2, IDP3/CYCLOPS, and ERN2. Close coordination between chromatin accessibility and gene expression in LCO response is likely essential for root nodule development. We applied novel methods for time-series analysis, ESCAROLE and DRMN [36], to model temporal changes in gene expression and chromatin accessibility. ESCAROLE enabled us to characterize the transcriptional dynamics beyond pairwise differential expression analysis, while DRMN allowed us to jointly analyze transcriptome and chromatin dynamics and predict which transcription factors (TFs) are most important for expression dynamics. Consistent with the theme of chromatin reorganization under LCO treatment response, DRMN identified IBM1 as a critical regulator. IBM1 encodes a JmjC domain-containing histone demethylase that catalyzes the removal of H3K9 methylation and di-methylation in Arabidopsis [37]. DRMN also identified regulatory genes involved in hormone responses in the early steps of symbiosis and nodule formation such as ethylene (ERF1, EDN1-3, and EIN3) and ABA (ABI4-5). EIN3 is a transcription factor mediating ethylene-regulated gene expression and morphological responses in Arabidopsis. The role of EIN3 in rhizobium-legume symbiosis or LCOs signaling remains uncharacterized, but sickle (skl) mutants for an EIN2 ortholog develop more infection threads and nodules and respond more to LCOs than wild-type plants, and ethylene treatment inhibits LCO signaling and nodule formation [38]. ABI4 and ABI5, basic leucine zipper transcription factors implicated in several plant functions, coordinate LCO and cytokinin signaling during nodulation in M. truncatula [39]. DRMN also identified regulators associated with the hormones involved in the nodule initiation, auxin (SHY2), and cytokinin (MtRRB15). SHY2, a member of the Aux/IAA family, plays a critical role in cell differentiation at root apical meristem and is activated by cytokinin [40, 41]. SHY2 was proposed as a candidate for nodule meristem regulation and differentiation after showing a very localized expression pattern in the nodule meristematic region [42]. Also related to nodule meristem initiation, KNOX TF-family members and PLT1-5 were predicted as regulators of gene expression in response to LCOs. MtPLT genes (MtPLT1-5) are part of the root developmental program recruited from root formation and control meristem formation and maintenance for root and nodule organogenesis [43]. We experimentally validated two of our regulators EIN3 and ERF1 using RNAi in M. truncatula and showed a significant effect in nodule formation. Prior work of Asamizu et al. [44] independently supports the observation of the ERF1 ortholog as an effector of nodule development in L. japonicus, where the number of nodules was likewise reduced in a similar RNAi experiment. Their findings suggest ERF1 is induced by rhizobium on a 3 to 24 h time scale, echoing the observed time scale of chromatin reorganization in M. truncatula in our work. Recent work of Reid et al. [45] emphasizes an early, positive role of ethylene in rhizobium-legume symbiosis in L. japonicus, which supports why we observe ethylene-related TFs having a positive impact on nodulation, unlike the ethylene insensitive skl mutation [38]. The exact mechanisms by which these genes regulate rhizobium-legume symbiosis can be explored in future research. Our analysis predicted genome-wide targets for transcription factors, including novel regulators identified by DRMN and previously known regulators of root nodulation, such as NIN, NF-YA1/NF-YB1, and CYCLOPS. For example, MTG-LASSO analysis predicted NIN as a direct target of SHY2 and MTF1, and FLOT4, required for infection thread formation, as a target of IBM1 [46]. Among known regulators, MTG-LASSO indicated that ARF16a and SPK1 are targets of NF-Y TFs. ARF16a and SPK1 control infection initiation and nodule formation [1]. Several NF-Y genes (NF-YA5 and NF-YB17) were identified as regulated by CYCLOPS. These predicted regulatory relationships can be tested with future validation experiments and uncover key mechanisms underlying the regulation of gene expression in LCO response
The regulatory mechanisms underlying plant-microbe symbiotic relationships remain poorly characterized. Here, we present a novel dataset that profiles the concurrent changes in transcriptome and chromatin accessibility in the model legume, Medicago truncatula, in response to rhizobia signal that trigger nodule formation. We have jointly modeled the chromatin and transcriptome time series data to predict the most critical regulators of the response to these signals and that underlie molecular pathways driving nodule formation. Our transcriptomic and accessibility datasets and computational framework to integrate these datasets provide a valuable resource for identifying key regulators for the establishment of root nodulation symbiosis in M. truncatula that could inform engineering of nodulation in species unable to establish that symbiosis naturally.
Seeds of wild-type Medicago truncatula Jemalong A17 strain (available through the USDA Germplasm Resources Information Network (GRIN)) were sterilized and germinated in 1% agar plates, including 1μM GA3. Plates were stored at 4 °C for 3 days in the dark and placed at room temperature overnight for germination. Seedlings were grown vertically for 5 days on a modified Fahraeus medium with no nitrogen [47], in a growth chamber (24 °C, 16 h light/8h dark cycle, 70 μmol m−2 s−1 photosynthetic photon flux). LCOs were purified from S. meliloti strain 2011 as described previously [48]. Next, seedling roots were immersed in a solution of purified LCOs (10−8 M) or 0.005% ethanol solution (control) for 1 h. Roots were cut and immediately used for nuclei extraction and generation of ATAC-seq libraries (see below) or snap-frozen in liquid nitrogen for posterior RNA isolation and sequencing. Roots were collected at 0 h (control), 15, 30 min, 1, 2, 4, 8, and 24 h after LCO treatment. Roots from seven plants were pooled for each of three biological replicates used in RNA sequencing, while roots from 15 plants were pooled for one replicate used in ATAC-seq, in each time point of the experiment.
For ATAC-seq library preparation, we followed the protocol described previously [49] with modifications. Before nuclei isolation, all materials were precooled to 4 °C. Briefly, roots were chopped for 2 min in 1 ml of pre-chilled lysis buffer (15 mM Tris-HCl pH7.5, 2mM EDTA, 20 mM NaCl, 80 mM KCl, 0.5 mM spermine, 15 mM 2-ME, 0.15 % TritonX-100) in a cold room. This step was repeated four times with a 1 min interval between repetitions. The homogenate was filtered through one layer of pre-wetted Miracloth, loaded on the surface of a 2 mL dense sucrose buffer (1.7 M sucrose, 10 mM Tris-HCl pH8.0, 2 mM MgCl2, 5 mM 2-ME, 1 mM EDTA, 0.15 % Triton X100), and centrifuged (2400 g, 20 min at 4 °C). The supernatant was removed, and the nuclei were resuspended in 500 μl of lysis buffer and then filtered in 70 μm and 40 μm filters consecutively. The nuclei were then collected by centrifuging the solution at 1000g for 5 min at 4 °C. After washing with 950 μl 1×TAPS buffer (10 mM TAPS-NaOH, pH8.0, 5 mM MgCl2), the samples were centrifuged again at 1000g for 5 min at 4 °C. The supernatant was removed, leaving the nuclei suspended in approximately 10 μl of solution. Next, 1.5 μl of Tn5 transposase (Illumina FC-121-1030), 15 μl of Tagmentation buffer, and 13.5 μl of ddH20 were added to the solution. The reaction was incubated at 37 °C for 30 min. The product was purified using a QIAGEN MinElute PCR Purification kit and then amplified using Phusion DNA polymerase. One microliter of the product was used in 10 μl qPCR cocktail with Sybr Green. Cycle number X was determined as the cycle were the ¼ of the maximum signal was reached. Then, we amplified the rest of the product in a Phusion (NEB) PCR system with X-2 cycles (10 to 15 cycles, 50 μl of reaction). Amplified libraries were purified with AMPure beads (Beckman Coulter), and library concentrations were determined using a Qubit. Sequencing was carried out in an Illumina HiSeqX (2 × 150 cycles) at the HudsonAlpha Institute for Biotechnology (Huntsville, AL, USA).
For each RNA extraction, roots from 7 plants were pooled and ground while keeping the sample frozen. RNA extraction was performed as described previously [50]. Libraries were prepared using 1 μg of RNA in the NEBNext® Ultra™ Directional RNA Library Prep Kit following the supplier’s instructions (New England Biolabs, Ipswich, MA, USA). Sequencing was carried out with an Illumina HiSeq3000 (2 × 100 cycles) at the Interdisciplinary Center for Biotechnology Research at the University of Florida (Gainesville, FL, USA).
Between 8.7 and 17.9 million, 2 × 100 bp reads were obtained after sequencing the 24 RNA-seq libraries. Reads were aligned with Kallisto [51] to the M. truncatula transcriptome (v5, [52], Additional file 1: Figure S1). The average of the alignment rates across time points was 87–95%. A total of 37,536 genes were detected with non-zero expression at any of the time points. The data were processed with SLEUTH [53] for further analysis. Finally, TPM expression values were quantile-normalized and log-transformed before being used as input for further analysis. Principle component analysis was applied to these data in MATLAB (Fig. 1A). For comparative purposes, transcriptome time course data related to root nodulation [12] obtained from the M. truncatula wild-type reference accession Jemalong A17 and three mutants (lyk3, nfp, and skl/ein2), were analyzed using the same Kallisto/SLEUTH approach. The 144 samples characterized in that experiment presented alignment rates of 91–96%, except four outliers with rates of 73–88%. Analysis of this data set detected 40,988 genes with non-zero expression, of which 36,298 were in common with the 37,536 identified in the present LCO-treatment experiment (Additional file 1: Figure S1).
DESeq [54] was applied to both the data generated in the present work and previously published data sets for four rhizobial treatment [12]. The expected count matrices of each data set were used as input to the DESeq algorithm, used in a default manner per the author recommendations. For each of the five time-course experiments, we assessed differential expression relative to control (time 0 h) for each later time point (Additional file 1: Figure S2A and S2E, left) as well as between pairs of time points (Additional file 1: Figure S2E). An adjusted P threshold of 0.05 was applied to select differentially expressed (DE) genes for each time point in each experiment. Statistics (Additional file 1: Figure S2B) and heat-maps for genes DE relative to control and between (all) time points (Additional file 1: Figure S2C and D) present clear patterns of temporal change. For the Larrainzar et al. data set [12], we also identified differentially expressed genes between the three mutants (lyk3, nfp, and skl/ein2) relative to the wild-type reference (Jemalong A17) for matched time points (Additional file 1: Fig S2E, right). As in the first analysis the union of genes identified at any time point defined the set of differentially expressed genes for the dataset. The union of differentially expressed genes across time points was used for comparisons between datasets. We quantified the degree of overlap between DE gene sets with an F-score, or harmonic mean, of the fraction of overlapping genes in each set using the union across all time points (Fig. 1C, Additional file 1: Figure S2F) as well as individual pairs of time points (Additional file 1: Figure S2G). For two sets of N1 and N2 genes, respectively, and NO in common between the two, the F-score is defined as:
We analyzed the LCO-treatment time course data with ESCAROLE [21] to characterize the temporal changes in the transcriptome. We included 37,536 genes with at least one non-zero count in at least one of the 24 experiments (3 replicates × 8 time points). Transcriptome data from each time point were grouped by k-means clustering and used as an input module assignment for the ESCAROLE algorithm (Fig. 3). The algorithm was run for 100 iterations with non-fixed covariance Gaussian mixture model (GMM) clustering, and k = 7 modules. The selection of k = 7 was determined by the mean silhouette index per time point and overall BIC-corrected likelihood score (Additional file 1: Figure S3A, B). From ESCAROLE, we obtain a module assignment for each gene at each time point and identified sets of genes that transition in their module assignment across the eight time points (Fig. 2B). We define transitioning gene sets from ESCAROLE results by grouping genes with a similar module transition profile with agglomerative hierarchical clustering (Fig. 3C, Additional file 1: Figure S3D). The pairwise distance between genes used for this clustering approach was the fraction of mismatches in the module assignment across the (8 point) time course. The distance threshold (to determine the cut on the dendrogram for the hierarchical clustering) and the minimum number of genes in a cluster were the input parameters to define the transitioning gene clusters in this approach. In choosing settings for these parameters, we tested different pairwise distance threshold values (those corresponding to 0-4 mismatches between module assignment profiles) and examined the resulting cluster sets for their size, overlap with differentially expressed genes, and enrichments of Gene Ontology (GO) and motif terms (see also the “Integrative analysis of RNA-seq and ATAC-seq time course data using the dynamic regulatory module networks algorithm” section). We chose a pairwise distance threshold of 0.26 in the hierarchical clustering analysis (corresponding to two mismatches across the 8-point time course) based on these results and used those clusters with 10 or more genes to define the 112 transitioning gene sets from the ESCAROLE results.
Each of the eight ATAC-seq libraries was paired-end sequenced twice, and 54 to 235 million reads were obtained from each sequencing library (Additional file 1: Figure S4A and B). The data were aligned with Bowtie 2 [55] to the M. truncatula v5 genome, with 46–75% of the data found mappable (alignable) to the reference genome. Properly paired fragments with a quality score of 3 or greater were then obtained with “samtools view -Sb -q3 -f2,” (Properly paired, Additional file 1: Figure S4A, B) and duplicate-removal was applied with “samtools rmdup” [56] to define the final library data sets utilized (Selected, Additional file 1: Fig S4A, B). Fragment length distributions of each time-point data set (Additional file 1: Figure S4C) present the expected ~10 bp DNA pitch but not nucleosome occupancy dependence first illustrated by Buenostro et al. [57]. This is consistent with previously published plant ATAC-seq data from Blajic et al. [58] (see Fig. 2A of that work). The absence of nucleosome occupancy dependence can be in part due to aspects of the ATAC-seq protocol implemented in plants versus mammals. Another explanation could be the large proportion of our reads mapping to promoter regions, which tend to be nucleosome depleted further explaining the diminished nucleosome pitch. Moreover, TSS-centric (±1kbp) distributions of selected fragments for each time point were analyzed using the ATACSeqQC [59] pipeline and the ChIPpeakAnno [60] toolset’s featureAlignedHeatmap function and found to be both favorable (Additional file 1: Figure S5A and B) and comparable to results from the Maher et al. M. truncatula data (Additional file 1: Figure S6A and B) analyzed in the same way. Peak calling was performed by applying MACS2 [30] to ATAC-seq data from each time point using the command: We mapped these peaks to genes if the center of a peak was within 10 kbp upstream and 1 kbp downstream of a gene transcription start site (TSS). Peaks called at each time point were merged across time points to generate a set of “universal peaks” using custom scripts [61] (Additional file 1: Figure S9A and B) based on two criteria: (1) peaks from two different time points had a Jaccard score overlap of 0.9 or higher, and (2) the peak from one time point was contained within the peak detected in another time point. FriP values of the peaks called in each time point were found to be favorable [62] (Additional file 1: Figure S9A), i.e., > 0.30 for all time points in accordance with ENCODE consortium standards for ATAC-seq peak-calling results. Annotations of the universal peak set (Fig. 3D, Additional file 1: Figure S9C and G) were generated in three steps: (1) annotating all peaks centered within 2 kbp upstream and 100 kbp downstream a gene TSS as “Promoter” peaks, (2) annotating any remaining peaks centered within 2–10 kbp upstream of a gene TSS as “Upstream”, and (3) using the results of the Homer annotatePeaks.pl tool [63] for all remaining peaks. The proportion of universal peaks mapped to “Promoter” regions under this definition is 32.1%.
To enable quantitative comparison of the chromatin accessibility profiles across time, we aggregated the ATAC-seq read counts in two sets of genomic regions: (1) gene promoter regions (defined as 2 kbp upstream to 2 kbp downstream of a given gene TSS) and (2) universal peaks described above using custom scripts [64]. Briefly, we first generated the per base pair (per-bp) coverage of fragments for each time point data set with Bedtools [65] using the command bedtools genomecov <bam file> -bp -pc. For each ± 2 kbp gene promoter region, the average coverage per-bp was estimated, and log-ratio transformed relative to the global genome-wide average per-bp coverage in the respective time point data set. The genome-wide average per-bp coverage was obtained by dividing the total coverage on any base pair by the length of the genome. The signals aggregated to the promoter were used for downstream principal component analysis (Fig. 3C). The Pearson correlation of aggregated promoter signals for replicate data sets was 0.965–0.990 across time-points (Additional file 1: Figure S7), indicating high consistency, also indicated by the similarity of PCA results for the replicate data sets (Additional file 1: Figure S8A). The signal for the universal peaks was similarly quantified by the log-ratio of the mean per-bp coverage of the respective peak region relative to the global average per-bp coverage. For both data sets, this was followed by quantile normalization across time points, providing a continuous measure of the accessibility of gene promoter and peak regions. To evaluate the relationship between gene expression and either promoter or universal peak accessibility, we first performed a zero-mean transformation of each gene’s expression profile and the corresponding accessibility profiles. Next, a Pearson’s correlation was estimated. To assess the significance of correlation, we generated a null distribution of correlations from 1000 random permutations of the time points. We computed a P-value that estimates the probability of observing a correlation in the permuted data more significant in magnitude than an observed correlation, treating positive and negative correlations separately. For the eight time points in this data set Pearson’s correlations were typically significant (P < = 0.05) when > 0.50 or < − 0.50. The zero-meaned promoter and universal peak accessibility profiles were clustered with k-means clustering, and the optimal settings for k were determined separately for each data set. In both cases, the silhouette index (computed with correlation distance metric) was used to select the optimal k. Here, k = 6 clusters were chosen for the promoter accessibility data (Fig. 3A, Additional file 1: Figure S8B). For the universal peak accessibility profile clusters, we additionally used enrichments for motifs within the clusters of peaks to determine the optimal setting of k = 9 clusters (Additional file 1: Figure S9D, E). The clusters were enriched for motif instances of several known regulators (Additional file 1: Figure S9F). Furthermore, the peaks in clusters 1 and 2 were more likely to be annotated as intergenic regions than peaks in any other cluster (Additional file 1: Figure S9G).
We applied a novel algorithm, dynamic regulatory module networks (DRMN) [18, 66, 67], to our RNA-seq and ATAC-seq time course data set to identify cis-regulatory elements and transcription factors associated with genes that exhibit dynamic behavior. The inputs to this algorithm are the RNA-seq time series data, the number of expression modules, and regulatory features for each time point derived from the ATAC-seq time course by examining the genomic region around a gene’s TSS. The algorithm outputs gene expression modules (states) for each time point and their associated regulatory programs comprising the cis-regulatory elements that best predict gene expression of a particular module. To obtain the cis-regulatory features for each gene, we used 333 M. truncatula motif position weight matrices from the CisBP v1.02 database [33] and seven curated motifs of interest (including those for NIN, CYCLOPS (CYC-RE), and NSP1 and other binding motifs). ATAC-seq activity was aggregated for those known motif instances in the manner described above for promoter and universal peak regions. Motif finding was done for each of the associated position weight matrices using the pwmmatch.exact.r script (from the PIQ pipeline [68]) using the default log-likelihood score threshold of 5. Motifs mapped to 10 kbp upstream to 1 kbp downstream of gene TSSs were assigned as potential features describing the corresponding gene’s expression. This distance cutoff was motivated by the experimental validation of the daphne mutation for the NIN (NODULE INCEPTION) gene in Lotus japonicus by Yoro et al. [69], which is an insertion in a regulatory site ~7 kbp from this gene and affects its expression. Moreover, Liu et al. [70] have likewise validated similar regulatory interactions between sites ~5 kbp upstream of the NIN gene in M. truncatula. For each gene, the accessibility of multiple instances of the same motif mapped to that gene was summed. Finally, the aggregated motif accessibility feature data were merged across the time course and quantile normalized [64]. The normalized accessibility data for ±2 kbp promoter regions were also included as a predictive feature of gene expression. The DRMN algorithm takes as input the number of modules, k and uses a regularized regression model, Fused Lasso [71], to learn regression models for each module, k, for all time points jointly. This has the following objective: Here, Xc,k is the nk X 1 vector of expression levels for nk genes in modules k for time point c, Yc, k is nk X p motif-accessibility feature matrix corresponding to the same genes, are the regression coefficients which represent the quantified association of gene expression with individual regulatory motif features. Here, Θk is the matrix of coefficients across time points. The sum over c, c′ represents the sum over pairs of consecutive time points. Specifically, here, ‖.‖1 is the l1 norm (sum of absolute values), ‖.‖2 is the l2 norm (square-root of the sum of each value), and ‖.‖2, 1 is the l1,2 norm, i.e., the sum of the l2 norm of the columns of the given matrix. Furthermore, ρ1, ρ2, and ρ3 are hyper-parameters of the model that need to be tuned for optimal training and inference of DRMNs. These parameters represent (1) a sparsity penalty, (2) enforcing similarity of features for consecutive time points, and (3) enforcing an overall similarity of feature selection across all time points. We used several criteria to determine these hyper-parameter settings. The most important is the Pearson correlation of actual and predicted expression in threefold cross-validation settings to assess the resulting predictive power of models inferred for varied settings of the hyperparameters. Additionally, the quality of the clustering (silhouette index scores), the BIC-corrected likelihood score, and stability of predictive power in threefold cross-validation (Additional file 1: Figure S10A, C) were considered. We first varied ρ1 (values of 1, 5–60 in increments of 5, and 75 and 100) and ρ2 (values of 0–60 in increments of 5, and 75 and 100) independently and assessed the resulting predictive power for all models inferred. Predictive power generally monotonically decreased with increasing values of either parameter for values of ρ1>10, while for ρ2< = 25, the clustering was unstable. A choice was made for ρ1 = 5 over ρ1 = 1, since predictive power correlation was marginally higher for ρ2 = 30–60. With the ρ1 parameter fixed to 5, a second independent scan of ρ2 and ρ3 was performed, with (1) ρ2 varied from 25–60 in increments of 5, 75, and 100, and (2) ρ3 scanned for values of 0–60 in increments of 5, 75, and 100. For settings of ρ3 = 5–20, there tended to be unstable predictive power of the least expressed module, recovering comparable but not greater performance compared to results for ρ3 = 0 or ρ3 > 20, indicating no advantage for setting ρ3 > 0. We considered the cross-validation predictive power, silhouette index of modules, and similarity to ESCAROLE modules, in determining a setting for ρ2 (Additional file 1: Figure S10A). Comparable performance was found for ρ2 = 30–60, but ρ2 = 45 and 50 maximized the mean threefold cross-validation performance. We selected ρ2 = 45, as it was the lower of the two settings to avoid unnecessarily high values for a hyperparameter. Based on this assessment results for the hyperparameter settings of ρ1 = 5, ρ2 = 45, and ρ3 = 0 were chosen. We ran DRMN on our time-course data set for k = 7 input modules, based on the optimal numbers of modules determined in the ESCAROLE analysis. Each module (Additional file 2: Table S1) was predicted to have multiple regulators based on DRMN’s fused regression model. To allow initial interpretation of the regulators, we filtered them as follows: (1) the magnitude of regulator-module edge-weights (Additional file 2: Table S2) in at least one time point being greater than 0.02 and (2) the regulatory motif being enriched in the module (FDR corrected q-value from hyper-geometric test, q < 0.05) for all time points (Additional file 1: Figure S11, Additional file 2: Table S3). The modules was also tested for enrichment of GO terms, using an FDR corrected hypergeometric test (q < 0.05) to define significant enrichment (Additional file 1: Figure S12, Additional file 2: Table S3). To identify module network edges that were significantly varying in time we first merged module network edge weights across time points per module and identified those edge weights that were significantly varying (t-test P < 0.05 as implemented in MATLAB with the ttest2() function) across the 0–1 and 2–24 h portions of the time course. The choice to compare across the 1–> 2 h time point transition was motivated by the observation of module reorganization at this time window (Fig. 5C). Those regulatory edges found to be significantly varying are likely important at the module level of organization (Fig. 6). To identify gene sets that transition in their expression due to changes in their predictive regulatory programs, we grouped genes that changed their DRMN inferred module assignment across time points using the same agglomerative hierarchical clustering approach applied in the ESCAROLE transitioning gene clustering analysis. We performed GO and motif enrichment on these gene sets as well to assess the optimal threshold for cutting the dendrogram (Additional file 1: Figure S13A). In total, we identified 79 gene sets spanning 10,176 genes. These gene sets were further analyzed using a regularized regression approach (described below) to identify regulators for each gene set.
We identified fine-grained regulator gene interactions by predicting regulators for individual genes in transitioning gene sets using a structured sparsity approach called multi-task group lasso (MTG-LASSO, Fig. 7A). MTG-LASSO is a type of multi-task learning framework, where one performs a regression for multiple tasks simultaneously to share information among the tasks. Here, each gene in the gene set is a task, and MTG-LASSO enables us to select the same regulator (motif) for all genes in the set but with different regression parameters. The regulator identity defines the “group” in MTG-LASSO which includes the regression weights for the regulator for all genes in the set. MTG-LASSO selects or unselects entire groups of regression weights. The MTG-LASSO objective for each gene set is: Here, Xg is the expression profile over time for gene g, and Ym, g is the vector of motif accessibility features for motif m and gene g over time. The parameters Θm, g are the regression coefficients for predicting the expression of g using the feature data for motif m. This second term denotes the ‖.‖1/2 norm defined as and is used for (1) penalizing the number selected motif features according to the l1 norm and (2) enforcing smoothness of the regression coefficients across genes according to the l2 norm. λ is the hyper-parameter for controlling the group structure. For each of the 79 transitioning gene sets, MTG-LASSO was applied (using the SLEP v4.1 package [35] in MATLAB [72]) to infer the most predictive regulatory features of gene expression over time from the same motif accessibility features used in the DRMN analysis. For each gene set, we applied MTG-LASSO in a leave-one-out testing mode (Additional file 1: Figure S14), where each of the eight time points was left out one at a time, a model was fit on the remaining seven, and predictive power (Pearson's correlation) was computed on the left-out time point. For each regulator, we calculated a P-value to assess the significance of the frequency with which a given regulator was selected relative to random. This was achieved by randomizing the data 40 times and estimating a null distribution for the rate with which that regulator was selected across folds. A Z-test P-value was then obtained for the result relative to random. We called a regulator significant if it was selected at least 6 of 8 time-point folds, and the number of times it was selected was significantly higher (t test P < 0.05) relative to random for the frequency of selection across folds. MTG-LASSO’s hyper-parameter, λ, was determined for each transitioning gene set from the range 0.20–0.99 (in intervals 0.10) based on (1) the mean Pearson’s correlation (predictive power) of the inferred regulatory features and (2) the number of regulators (5–15 for most gene sets) identified as significant such that the ratio of the number of identified regulators to number of target genes being close to 0.05 (Additional file 1: Figure S14). This approach identified 33 gene sets (of the original 79) with predicted regulators. For the remaining transitioning gene sets, significant regulators were not found either because the available predictive features were not good descriptions of the respective gene expression profiles or regulators were obtained for only one or two settings of λ, hindering an appropriate assessment of results. For each of the 33 gene sets for which we identified regulators using MTG-LASSO (Additional file 1: Figure S14), we created regulator-target predictions between the significant regulatory features and member genes, defining 122,245 regulatory edges spanning 126 motifs for 5978 target genes (from 10,176 genes aggregated among the 79 transitioning gene clusters). Of the 126 motifs, we mapped 53 motifs to 278 M. truncatula regulator genes, including 31 well-studied regulators (specifically with common names in the v5 genome annotations). The remaining 73 motifs were assigned to 261 M. truncatula genes in the v5 genome assembly that were additionally identified as transcription factors (TFs). The relatively high number of motif to gene name mappings is because TF names were provided in CisBP v1.2 as systematic gene names from the v3/v3.5 M. truncatula genome assemblies rather than v5. We used a 70% BLAST similarity score to define mappings from M. truncatula v3/v3.5 genome systematic gene names to v5 genome systematic gene names.
We used RNAi to validate three predicted regulators from our DRMN analysis, EIN3, ERF1, and IAA4-5. 104 bp region in the CDS specific to the gene of interest was amplified with 5′-CACC and inserted into pENTR™/D-TOPO® using directional TOPO® cloning and further recombined in vitro with the destination vector pK7GW1WG2(II)-RedRoot (https://gatewayvectors.vib.be/collection/pk7gwiwg2ii-redroot) using Gateway® LR Clonase® II enzyme mix using manufacturer’s instructions. To validate RNAi, total RNA was extracted from transformed roots of each genotype using Qiagen RNeasy® Plant Mini kit and genomic DNA removed using TURBO DNA-free™ Kit (Ambion). First-strand cDNA was synthesized using RevertAid RT Reverse Transcription Kit (Thermo Scientific™). Quantitative RT-PCR was performed using BIORAD SsoAdvanced Universal SYBR Green Supermix on BIORAD CFX96™ Real-time system; C1000 Touch™ Thermal cycler. The HEL and UBC9 genes were used as endogenous controls. Two (EIN3—MtrunA17Chr5g0440591) or three (ERF1—MtrunA17Chr1g0186741) technical replicates were used. A BLAST was performed for all primers against the M. truncatula v5 genome to ensure specificity. The primers chosen for the validation of RNAi do not overlap with the RNAi regions (utilized primers provided in Additional file 1: Table S1). The RNAi expression clones were introduced into Agrobacterium rhizogenes MSU440 with electroporation. Composite M. truncatula plants were generated as previously described [73]. Three weeks after transformation with A. rhizogenes MSU440, the roots were screened for red fluorescence of tdTomato, and the composite plants with red roots were transferred to growth pouches containing modified nodulation medium (MNM) [74]. The plants were acclimated for 4 days and inoculated with S. meliloti 1021 harboring pXLGD4 [75]. Two weeks post inoculation, live seedlings were stained for lacZ (5 mM potassium ferrocyanide, 5 mM potassium ferricyanide, and 0.08% X-gal in 0.1 M PIPES, pH 7) overnight at 37 °C. Roots were rinsed in distilled water, and nodules were visualized and counted under a Leica fluorescence stereomicroscope (Fig. 8B, Additional file 2: Table S5).
Additional file 1: Table S1. Primers used in the RNAi validation study. Figure S1. Analysis workflow. Figure S2. Detailed DE gene statistics summary. Figure S3. Supplementary ESCAROLE clustering results. Figure S4. ATAC-seq data alignment statistics and fragment length distributions. Figure S5. ATAC-seq activity heatmaps and line plots for ±1 kb TSS regions in LCO-treatment data. Figure S6. ATAC-seq activity heatmaps and line plots for ±1 kb TSS regions in the comparable Maher et al. Medicago root sample data. Figure S7. Correlation of aggregated ATAC-seq activity for ±2 kb promoter regions. Figure S8. Supplementary ATAC-seq promoter analysis plots. Figure S9. Supplementary ATAC-seq peak-calling analysis plots. Figure S10. DRMN hyper-parameter tuning summary. Figure S11. DRMN module network edge-weight summary. Figure S12. DRMN module GO enrichment summary. Figure S13. Summary of ESCAROLE and DRMN transitioning gene set statistics and comparison. Figure S14. Summary of MTG-LASSO results and parameter tuning. Figure S15. Supplementary RNAi validation information.Additional file 2: Table S1. DRMN module assignments (all genes, all time points). Table S2. Inferred module-network edge-weights from DRMN. Table S3. Module motif enrichments. Table S4. MTG-LASSO target predictions. Table S5. RNAi validation results. |
|
PMC9647983 | Nirmal Vadgama,Mohamed Ameen,Laksshman Sundaram,Sadhana Gaddam,,Casey Gifford,Jamal Nasir,Ioannis Karakikes | De novo and inherited variants in coding and regulatory regions in genetic cardiomyopathies | 10-11-2022 | Cardiomyopathy,Single-cell,Noncoding,De novo,Regulome,Oligogenic | Background Cardiomyopathies are a leading cause of progressive heart failure and sudden cardiac death; however, their genetic aetiology remains poorly understood. We hypothesised that variants in noncoding regulatory regions and oligogenic inheritance mechanisms may help close the diagnostic gap. Methods We first analysed whole-genome sequencing data of 143 parent–offspring trios from Genomics England 100,000 Genomes Project. We used gene panel testing and a phenotype-based, variant prioritisation framework called Exomiser to identify candidate genes in trios. To assess the contribution of noncoding DNVs to cardiomyopathies, we intersected DNVs with open chromatin sequences from single-cell ATAC-seq data of cardiomyocytes. We also performed a case–control analysis in an exome-negative cohort, including 843 probands and 19,467 controls, to assess the association between noncoding variants in known cardiomyopathy genes and disease. Results In the trio analysis, a definite or probable genetic diagnosis was identified in 21 probands according to the American College of Medical Genetics guidelines. We identified novel DNVs in diagnostic-grade genes (RYR2, TNNT2, PTPN11, MYH7, LZR1, NKX2-5), and five cases harbouring a combination of prioritised variants, suggesting that oligogenic inheritance and genetic modifiers contribute to cardiomyopathies. Phenotype-based ranking of candidate genes identified in noncoding DNV analysis revealed JPH2 as the top candidate. Moreover, a case–control analysis revealed an enrichment of rare noncoding variants in regulatory elements of cardiomyopathy genes (p = .035, OR = 1.43, 95% Cl = 1.095–1.767) versus controls. Of the 25 variants associated with disease (p< 0.5), 23 are novel and nine are predicted to disrupt transcription factor binding motifs. Conclusion Our results highlight complex genetic mechanisms in cardiomyopathies and reveal novel genes for future investigations. Supplementary Information The online version contains supplementary material available at 10.1186/s40246-022-00420-0. | De novo and inherited variants in coding and regulatory regions in genetic cardiomyopathies
Cardiomyopathies are a leading cause of progressive heart failure and sudden cardiac death; however, their genetic aetiology remains poorly understood. We hypothesised that variants in noncoding regulatory regions and oligogenic inheritance mechanisms may help close the diagnostic gap.
We first analysed whole-genome sequencing data of 143 parent–offspring trios from Genomics England 100,000 Genomes Project. We used gene panel testing and a phenotype-based, variant prioritisation framework called Exomiser to identify candidate genes in trios. To assess the contribution of noncoding DNVs to cardiomyopathies, we intersected DNVs with open chromatin sequences from single-cell ATAC-seq data of cardiomyocytes. We also performed a case–control analysis in an exome-negative cohort, including 843 probands and 19,467 controls, to assess the association between noncoding variants in known cardiomyopathy genes and disease.
In the trio analysis, a definite or probable genetic diagnosis was identified in 21 probands according to the American College of Medical Genetics guidelines. We identified novel DNVs in diagnostic-grade genes (RYR2, TNNT2, PTPN11, MYH7, LZR1, NKX2-5), and five cases harbouring a combination of prioritised variants, suggesting that oligogenic inheritance and genetic modifiers contribute to cardiomyopathies. Phenotype-based ranking of candidate genes identified in noncoding DNV analysis revealed JPH2 as the top candidate. Moreover, a case–control analysis revealed an enrichment of rare noncoding variants in regulatory elements of cardiomyopathy genes (p = .035, OR = 1.43, 95% Cl = 1.095–1.767) versus controls. Of the 25 variants associated with disease (p< 0.5), 23 are novel and nine are predicted to disrupt transcription factor binding motifs.
Our results highlight complex genetic mechanisms in cardiomyopathies and reveal novel genes for future investigations.
The online version contains supplementary material available at 10.1186/s40246-022-00420-0.
The cardiomyopathies, herein divided into dilated cardiomyopathy (DCM), hypertrophic cardiomyopathy (HCM), arrhythmogenic right ventricular cardiomyopathy (ARVC) and left ventricular non-compaction cardiomyopathy (LVNC), are leading causes of heart failure [1]. Although there is considerable overlap between different cardiomyopathy subtypes, each has a signature genetic aetiology. HCM and ARVC are largely explained by alterations in sarcomere or desmosome proteins, respectively. Around half of HCM cases are caused by mutations in the genes MYH7 and MYBPC3 [2]. However, the genetic architecture of DCM is far more complex. To date, more than 250 genes have been implicated in DCM causation or risk, including genes encoding for cytoskeletal, sarcolemmal, mitochondrial, calcium cycling, costameric and sarcomeric proteins [3]. Although the aetiological basis of cardiomyopathy is incomplete, recent genetic studies suggest that a large proportion of cases may be explained by alterations in the noncoding genome [4]. Data from the ENCODE project suggest that biochemical functionality could be assigned to 80 per cent of the human genome, affecting regulatory and tissue-specific expression patterns [5]. Furthermore, genome-wide association studies (GWAS) show that over 90% of disease-associated SNPs are in noncoding regions of the genome, including regulatory regions, such as promoters and enhancers [6]. Thus, a key priority in the cardiomyopathy gene discovery pipeline is the identification of regulatory elements controlling genes associated with these disorders. High-throughput epigenomic profiling methods such as ATAC-seq and ChIP-seq have enabled profiling of chromatin accessibility across samples in a tissue-wide manner, providing the opportunity to identify millions of context-specific regulatory elements. However, these bulk measurements of chromatin accessibility limit the precise understanding of how tissue heterogeneity and multiple cell types in the population contribute to overall disease aetiology [7]. Recent advances in single-cell ‘omics technologies have enabled an unbiased identification of cell-type populations and regulatory elements in a heterogeneous biological sample. By mapping the chromatin-regulatory landscape at a single-cell resolution, studies have demonstrated the potential to link regulatory elements to their target genes, and map regulatory dynamics during complex cellular differentiation processes [7–9]. We first performed parent–offspring trio analysis to assess the impact of rare inherited recessive and dominant variants, and of DNVs on cardiomyopathy. We hypothesised that variants in regulatory regions that are specifically active in the adult heart could provide an aetiological basis for cases with unexplained genetics. We further supported this hypothesis through systematic examination of noncoding regulatory elements of known disease-risk genes in a mutation-negative cardiomyopathy cohort. We identified noncoding variants predicted to disrupt cis-regulatory elements involved in cardiac gene regulation. By combining inherited and DNV analysis of a clinically well-defined cohort, this study provides novel insights into the complex genetics of cardiomyopathy subtypes, which may lead to improved diagnosis and therapies.
All participants were recruited to the 100,000 Genomes Project (100KGP) (protocol version 7, 2020), with written informed consent. The full protocol is available online at https://doi.org/10.6084/m9.figshare.4530893.v7. Probands of parent–offspring trios (n = 143) and singleton offspring (n = 843) were diagnosed with either ARVC, LVNC, HCM, DCM or DCM and conduction defects. They were included in the study if they had a clear diagnosis under 40 years of age. Patients were excluded if they had an unclear diagnosis or a history suggestive of a non-genetic cause. The control cohort included 19,467 age-, sex-, and ethnicity-matched participants with no known heart disease. The study adheres to the principles set out in the Declaration of Helsinki. Patients and relatives gave written informed consent for genetic testing. Ethical approval for the 100KGP was granted by the East of England—Cambridge South Research Ethics Committee (REC Ref 14/EE/1112).
Whole-genome sequencing (WGS) were processed by Illumina, and sequencing data were passed through 100KGP's bioinformatics pipeline for alignment, annotation, and variant calling. Variants were prioritised using prespecified virtual gene panels from PanelApp (https://www.genomicsengland.co.uk). To date, 208 genes are listed for the Cardiomyopathies—including childhood onset (v1.37) panel. Based on the Human Phenotype Ontology (HPO) terms entered for these patients, additional gene panels were applied where applicable, including: undiagnosed metabolic disorders (v1.95), mitochondrial disorders (v1.127), intellectual disability (v3.2), and RASopathies (v1.27). Inherited variants were restricted to panel genes, where the allelic state was required to match the curated mode of inheritance.
The variants were categorised according to gene annotation, population allele frequency, functional prediction, and clinical interpretation. The raw list of SNVs and indels was annotated using ANNOVAR [10]. Variants were classified based on their mutational characteristics, position in the genome, allele frequency, and functional role in cardiomyopathy. In silico prediction of pathogenicity was performed using CADD [11] and REVEL [12], and conservation of nucleotides was scored using GERP + + [13]. Population allele frequencies were obtained from 100KGP, 1000Genomes, and gnomAD [14]. Highest priority was given to protein truncating (frameshift, stop gain, stop loss, splice acceptor variant, or splice donor variant) or de novo (protein truncating, missense, or splice region) variants in a gene on the diagnostic-grade list in the virtual gene panel for cardiomyopathy or any additional gene panel relevant to the phenotype to the patient [15]. Inherited protein-altering variants, such as missense and splice region variants, in diagnostic-grade genes were also considered. Variants were retained if they had a minor allele frequency (MAF) < 0.001; the allelic state matches the known mode of inheritance for the gene and disorder and segregates with disease (where applicable). Rare variants were ranked according to a REVEL score above the default threshold of 0.5, a CADD score greater than 20, and GERP + + score greater than 2. Variants with clinical significance as benign or likely benign according to the ClinVar dataset were removed. In parallel, we used Exomiser (v12.1.0) [16], a phenotype-driven variant prioritisation framework. Exomiser uses computational filters for variant frequency and predicted pathogenicity, protein interaction networks, patient phenotypes, cross-species phenotype comparisons, and pedigree information. A logistic regression model is used to combine the phenotype and variant scores to produce an overall Exomiser score. We considered the top three ranked variants that matched with our candidate-gene discovery analysis.
DNVs were identified by 100KGP's bioinformatics pipeline. Briefly, variants from WGS data were called using Platypus, and filtered for absence of the mutation in both parents, read depth (> 20), allele balance (> 0.3 and 0.7), and no overlap with segmental duplications, simple repeat regions, and patch regions. To analyse noncoding DNVs, we obtained single-cell ATAC and RNA data of human adult ventricle [14]. DNVs from 143 trios were intersected with the single-cell ATAC-seq peak sets using default parameters of bedtools v2.24.0. Peak sets were tested for an enrichment of DNVs in offspring as compared to a background peak set which contained peaks from all other cell types. We used a chi-squared test to compare the number of peaks with DNVs between the cardiomyocyte-specific peak set and the background peak set.
We used H3K27ac HiChIP to map active chromatin interactions genome wide on iPSC-derived cardiomyocytes (GSM3639703) [17]. HiChIP paired-end reads were aligned to GRCh38 genome using the HiC-Pro pipeline. Duplicate reads were removed and default HiC-pro settings were used to assign reads to MboI restriction fragments, filter for valid interactions, and generate binned interaction matrices. High-confidence contacts (FDR < 0.05) were called using the contact caller FitHiChIP with default settings at 10 kb resolution. These high-confidence contacts were used in visualisation.
We used a combination of methods to predict enhancer–gene interactions and interpret the functions of noncoding DNVs. Candidate enhancers were predicted using the recently developed activity-by-contact (ABC) model [18], which integrates H3K27ac ChIP-seq, HiChIP, and gene expression data with chromatin accessibility to predict enhancers and link them to their target genes. Using this method, we were able to identify sets of high-confidence putative enhancers for cardiomyocytes and their likely target genes. In addition, publicly available Hi-C data of human left and right ventricle tissue (sample GSM1419085 and GSM2322554, respectively) [19, 20] were analysed in the 3D-genome Interaction Viewer (3DIV) and database (http://kobic.kr/3div/)20. 3DIV was run using distance-normalised interaction frequency ≥ 2 to define significant enhancer–promoter interactions. Topologically associating domains (TADs) were identified using TopDom [22] with a window size of 20. DNVs that were within enhancers predicted by both the ABC model and 3DIV were considered for downstream analysis. Finally, we used a machine learning approach called FATHMM-MKL [23] to predict the functional impact of noncoding SNVs. This tool integrates functional annotations from ENCODE with nucleotide-based sequence conservation measures and provides predictions as p values in the range 0 to 1. We used the default score > 0.5 to indicate putative deleterious variants.
Promoters were defined as 2 kb upstream or 1 kb downstream of transcription start sites (TSSs) and determined based on the basic gene annotation file of release 33 from GENCODE [24]. Further, to detect distal promoter-interacting loci we used promoter capture Hi-C data generated from three different human cell/tissue-types, including cardiomyocytes (GSM2297135, GSM2297136, GSM2297137, GSM2297138, GSM2297139), left ventricle (GSM2297192, GSM2297193, GSM2297194, GSM2297195, GSM2297196, GSM2297197, GSM2297198, GSM2297199, GSM2297200, GSM3067218, GSM3067219), and right ventricle (GSM2297289, GSM2297290, GSM2297291, GSM2297292, GSM2297293, GSM2297294, GSM2297295, GSM2297296, GSM2297297) [21].
For noncoding DNVs, a functional enrichment analysis of the candidate genes was performed using the VarElect [25]. This tool uses the deep LifeMap Knowledgebase to infer the “direct” or “indirect” association of biological function between genes and the queried phenotype—i.e. “cardiomyopathy”. A direct association is determined if studies indicate that the gene in question directly affects disease development. An indirect association is based on factors such as shared pathways, protein–protein interaction networks, and mutual publications.
Independent to the trio analysis, we analysed 843 probands and 19,467 unrelated controls to identify high-risk noncoding variants in regulatory elements of 12 cardiomyopathy genes with definitive (BAG3, DES, FLNC, LMNA, MYH7, PLN, RBM20, SCN5A, TNNC1, TNNT2, TTN) or strong (DSP) evidence [26]. We focused on regulatory elements of diagnostic genes rather than the entire genome to avoid false-positive results related to genes with an unclear association with the disease. Regulatory elements for each gene were determined using the ABC model (described above). The functional impact of rare regulatory variants was assessed based on several tools, including RegulomeDB (https://regulomedb.org/), FunMotif (http://bioinf.icm.uu.se:3838/funmotifs/), and FATHMM-MKL. We mapped SNVs to these active regulatory regions of cardiomyopathy genes and defined them as high-risk if they were rare (MAF < 0.0001 in 100KGP and gnomAD population controls), predicted to alter transcription factor (TF) binding, and were enriched in cases versus controls (p < 0.05).
To compare variant burden between cases and unrelated controls for high-risk regulatory variants of cardiomyopathy genes, variant calls were required to have an MAF of ≤ 0.0001 in 100KGP controls and gnomAD. Controls were proportionally matched for age, sex and ethnicity. χ2 test, odds ratios (OR), and 95% confidence intervals (95% CIs) were calculated for regulatory regions of all genes by comparing the burden of rare variants. To evaluate the association between individual noncoding variants and the risk of cardiomyopathy, we performed a Fisher’s exact test as expected value were < 5. Statistical significance was considered at the 5% level (two-tailed). Statistical analyses were undertaken using R 4.2.0 and RStudio 2022.02.2.
In the trio analysis, 143 probands (85 males, 58 females), with severe or syndromic disease, together with their parents, were analysed. They were of different reported ethnicities across England (Fig. 1A). The age distribution of participants is shown in Fig. 1B. Participants recruited for HCM were more frequent in the 20 to 29 age group, and DCM between ages 20 to 39. LVNC was more common in younger people at enrolment. Of note, participants were enrolled in the 100KGP in their 40 s and 50 s as they still lacked a molecular diagnosis, despite all participants having an age of onset before 40 years of age. Figure 1C shows the top 20 HPO terms in all participants recruited, including intellectual disability, joint hypermobility, and skeletal myopathy, suggesting syndromic causes. After parent–offspring trio analysis and application of our stringent filtering criteria, each proband had an average of 69.7 DNVs (Fig. 2). Genetic findings and genotype–phenotype correlations are described in Tables 1, 2 and 3. We used Exomiser to help narrow down candidate variants. In addition, 843 exome-negative cardiomyopathy probands and 19,467 controls were incorporated into our case-control study from the 100KGP. The cases are singleton offspring whose parental WGS data were unavailable. Using the PanelApp software, which contains a crowd-sourced curation of genes with diagnostic-grade evidence, only cases lacking a molecular diagnosis were recruited in the case–control analysis. These cases included 61.8% HCM, 26.0% DCM, 8.5% ARVC, and 3.6% LVNC subtypes. Most participants were of European ancestry (70%), and 64% were male. De novo variants in diagnostic-grade genes. Using the American College of Medical Genetics (ACMG) guidelines, we identified deleterious DNVs in 11/143 trios (7.7%). These class 4 and 5 variants are defined as likely pathogenic or pathogenic, and are reported as consistent with or confirming a diagnosis, respectively (Table 1). Exomiser ranked the correct diagnosed variants as the top candidate in all these cases, and no parents were affected. Several novel DNVs were identified in syndromic and non-syndromic cases. In Fam659, the proband had a novel missense DNV c.568C > A (p.Arg190Ser) in the NKX2-5 gene. NKX2-5 (NK2 homeobox 5) encodes for a transcription factor that is important for the development of the myocardium [27]. Mutations in this gene are known to cause congenital heart disease, particularly atrial septal defect (with or without atrioventricular conduction defects), and ventricular septal defect. Consistent with this, the patient was diagnosed with LVNC, including atrial septal defect, atrioventricular block, and abnormal ventricular septum morphology. In the proband of Fam208, a novel missense DNV c.590 T > C (p.Leu197Pro) was identified in the LMNA gene (lamin A/C). LMNA encodes the A-type lamin proteins, lamin A and C, which are the major components of the nuclear membrane in mammals. Mutations in LMNA have been reported to cause a variety of clinical phenotypes, collectively known as laminopathies. These include cardiac disorders, premature ageing syndromes, and neuropathies [28]. In addition to DCM, the proband had a range of musculoskeletal-related abnormalities (Table 1). In the proband of Fam520, a missense DNV c.360C > A (p.His120Gln) was found in the LZTR1 gene. Mutations in LZTR1 (leucine-zipper-like transcriptional regulator 1) are associated with Noonan syndrome phenotypes and schwannomatosis [29]. As well as obstructive HCM, the proband had combined disorders of mitral, aortic and tricuspid valves, congestive heart failure, thyrotoxicosis with diffuse goitre, postprocedural hypothyroidism, and rheumatoid arthritis. According to the ACMG guidelines, the variant is classified as a variant of uncertain significance. It is predicated to be deleterious according to in silico algorithms and was not identified in public databases, including gnomAD and 1000G. We require further clinical information to determine a phenotype consistent with Noonan spectrum disorders. Dominant and recessive Noonan syndrome-causing mutations have been described near this variant, all of which within the highly conserved kelch domains [30, 31]. A novel frameshift DNV of high-impact c.6183del (p.Leu2062TrpfsTer25) in the ANKRD11 gene was observed in the proband of Fam451. Mutations in ANKRD11 (ankyrin repeat domain 11) are reported to be associated with KBG syndrome and intellectual disability [32]. The patient presented with LVNC and a variety of other phenotypes, including abnormal vena cava morphology, bicuspid aortic valve, abnormality of the face, delayed fine motor development, intellectual disability, and proportionate short stature (Table 1). In the proband of Fam478, a known pathogenic missense DNV c.782C > T (p.Pro261Leu) was found in the RAF1 gene. Diseases associated with RAF1 (Raf1 proto-oncogene, serine/threonine kinase) include Noonan and Leopard syndromes. These developmental disorders have overlapping features, including cardiac abnormalities, short stature, and facial dysmorphia [33]. The proband was diagnosed with HCM and had other phenotypes, including congenital malformation of cardiac septum, palpitations, anxiety disorder, and depression. This DNV has previously been reported in an individual affected with Noonan syndrome, including other individuals with clinical features of this disease [33–35]. Moreover, functional studies have shown that p.Pro261Leu leads to increased activity of the RAF1 protein [36]. In the proband of Fam828, we found a missense DNV c.836A > G (p.Tyr279Cys) in the PTPN11 gene (protein tyrosine phosphatase non-receptor type 11). Mutations in PTPN11 are well characterised in children with Noonan syndrome and juvenile myelomonocytic leukaemia [37]. In addition to HCM, the proband had other phenotypes, including intellectual disability, failure to thrive, right ventricular cardiomyopathy, and LVNC. Other known pathogenic DNVs were found consistent with a diagnosis of non-syndromic cardiomyopathy. Three participants with HCM had known missense variants in MYH7. Two variants (c.2156G > A (p.Arg719Gln) and c.1358G > A (p.Arg453His)) are reported as pathogenic on ClinVar, whereas variant c.2420G > C (p.Arg807Pro) is reported as likely pathogenic. The proband harbouring the latter MYH7 variant also had congenital malformations of the heart, congestive heart failure, arrhythmia, and died as an infant with sudden cardiac arrest. Pathogenic variants were found in other known genes, including DES, RYR2, TTN and TNNT2 (Table 1).
We also looked at DNVs in genes not included in PanelApp. In 27/143 probands (18.9%), 30 rare de novo coding variants were identified, which were considered deleterious based on in silico prediction tools (see methods). Using Exomiser, a phenotype-based prioritisation pipeline, 11 DNVs were ranked in the top three as the most likely cause. These include TUBA1B, KIRREL1, DAAM1, DOCK11, and KDM5B (Table 2). It is possible that some of the putative genes identified herein could be novel gene candidates or genetic modifiers. Interestingly, the proband of Fam231 harbouring a missense DNV c.49 T > C (p.Cys17Arg) in the DAAM1 gene also had a known pathogenic variant in TNNT2 (Table 1). Studies show that DAAM1 is required for cardiomyocyte maturation [38], and deletion of the gene is associated with congenital heart anomalies [39]. Multiple gene mutations occurring in cardiomyopathy families may result in a more severe clinical phenotype because of a compounding effect. Other examples of oligogenic inheritance are described below.
In 10/143 trios (7.0%), 14 rare inherited variants of potential clinical significance were identified in probands and their affected family members based on gene panel testing. These include variants with recessive, dominant, and compound heterozygous segregation patterns (Table 3). Two participants with HCM had missense variants in MYBPC3. In Fam599, the proband harbouring the c.1504C > T (p.Arg502Trp) variant presented with a range of diagnoses, including abnormal thumb, eye and oral morphology, intellectual disability, and mild microcephaly. The mother, also harbouring the variant, was diagnosed with atrial septal defect, short thumb, polycystic ovaries, and Raynaud syndrome. It is not clear why the proband presented with a more severe phenotype. In Fam411, the proband inherited the MYBPC3 splice donor variant c.25 + 1G > A from the affected mother. In addition to HCM, the mother presented with hypothyroidism, prolonged QT syndrome, severe depression, anxiety, and schizoid personality disorder. In Fam484, the proband inherited an autosomal dominant mutation c.2254 T > A in SAMD9 from the affected father. The proband was diagnosed with LVNC, ventricular septal defect, intellectual disability, joint hypermobility, and Wolff–Parkinson–White syndrome. Although the affected father and sibling carrying this variant presented with LVNC, neither had the comorbidities present in the proband. The father, however, presented with gastrointestinal haemorrhage, and diverticular disease of large intestine (without perforation or abscess). Mutations in SAMD9 have been described in patients with MIRAGE syndrome, a severe multisystem disorder [40]. This includes prominent gastrointestinal symptoms and intellectual disability. In the proband and affected mother of Fam957, a missense variant c.372C > G (p.Ile124Met) was identified in the ACTC1 gene. Mutations in ACTC1 (actin alpha cardiac muscle 1) are associated with atrial septal defect, DCM, and HCM [41, 42]. In addition to DCM, the proband had partial anomalous pulmonary venous return, dyspnoea, myocardial fibrosis, and oligospermia. The mother was diagnosed with DCM, secundum atrial septal defect, and bipolar affective disorder. Multilocus inheritance may explain the relatively low diagnostic yield for cardiomyopathy cases, or apparent phenotypic expansion. We found evidence for compound heterozygosity, digenic, and oligogenic inheritance in several families (described below). This highlights the importance of screening for additional genes even after a single mutation has been identified. We observed evidence for oligogenic inheritance in Fam919. The proband, born in 2018 and reported to be deceased, had a recessive mutation in POLR3A (c.1787C > T (p.Thr596Met)) and a compound heterozygous TTN mutation (c.92176C > T (p.Pro30726Ser)). Mutations in POLR3A are associated with a wide array of pathological phenotypes, some of which were present in the proband. In addition to DCM and congestive heart failure, the proband had multiple congenital anomalies, including microcephaly, endocardial fibroelastosis, hydrops fetalis, polymicrogyria, cortical dysplasia, and pedal oedema. The proband of Fam539 inherited five compound heterozygous TTN variants, one of which from the mother passed our filtering criteria for deleteriousness (c.20335A > T, p.Ser6779Cys) (see methods). The proband also inherited autosomal dominant variants of incomplete penetrance in COL6A1 and LZTR1. The LZTR1 stop gained variant c.1311G > A (p.Trp437Ter) was inherited from the mother, and the COL6A1 missense variant c.1712A > C (p.Lys571Thr) was inherited from the father. The proband was diagnosed with HCM, skeletal myopathy, and increased nuchal translucency. One out of two siblings are also affected with HCM; however, detailed medical notes or WGS data are not available. Although considered unaffected for cardiomyopathy, both parents were diagnosed with primary (essential) hypertension. In addition, the father has a family history of ischaemic heart disease, arrhythmia, syncope and collapse, and other ill-defined heart diseases. In the proband of Fam992, compound heterozygous mutations c.852_855del (p.Asn284LysfsTer4) and c.1038_1040del (p.Lys346del) in the DSG2 gene were identified. The mother of the proband had the frameshift variant c.852_855del, whereas the father had the inframe deletion c.1038_1040del. In addition to ARVC, the proband was diagnosed with disorders of magnesium metabolism, hypokalaemia, and congenital malformations of cardiac chambers and connections. Of note, a DNV in the enhancer of gene TUSC3 was also identified. TUSC3 constitutes a major component in cellular magnesium transport and homeostasis, and its function in regulation of embryonic development in vertebrates has been suggested [43, 44]. This may explain the disorders of magnesium metabolism and hypokalaemia as a secondary cause in the proband [45]. In Fam180, the proband and mother with HCM harbour a variant of unknown significance (class 3) in the MYH7 gene. In addition, two families with HCM had mutations in mitochondrial genes MT-CO1 and MT-ND6 (Table 3); both previously implicated in heart disease, although heteroplasmy proportions are yet to be determined in multiple tissue samples. These examples demonstrate that non-Mendelian inheritance may be an important factor in the cardiomyopathy cause-discovery pipeline. Other possibilities exist to help close the diagnostic gap, including noncoding mutations that affect regulatory elements.
We hypothesised that DNVs in human heart regulatory regions are more likely to perturb expression levels of genes that are essential for cardiac function. Annotation of ventricular cardiomyocyte peak set in genomic features shows enrichment in intronic and distal intergenic regions and in the flanking regions of TSSs, suggesting an enrichment of gene regulatory elements, such as enhancers. We intersected single-cell ATAC-seq peaks with publicly available H3K27ac ChIP-seq data (a marker for active enhancers) of eight healthy adult donors [46] and found significant overlap with our peaks (Permutation test, one-sided, p < 0.001).
To determine selective vulnerabilities across diverse cell types of the human heart, we intersected cell-type-specific ATAC-seq peaks with DNVs identified from parent–offspring trio analysis (see methods). Cardiomyocyte-specific peak sets were not significantly enriched for DNVs in offspring compared to a merged background peak set. A total of 288 DNVs from 143 trios intersected with a peak signal from ventricular cardiomyocytes. After filtering for parental affected status, H3K27ac overlap, and mapping regulatory regions to genes, 15 DNVs were within promoter regions, and 12 within predicted enhancers linked to their target genes. We used the tool FATHMM-MKL to predict the functional effects of noncoding variants. Additional file 1: Table S1 shows prioritised variants within ventricular cardiomyocyte open chromatin regions.
Using the ABC model [18], we predicted likely enhancers by integrating H3K27ac ChIP-seq, HiChIP, and gene expression data with chromatin accessibility. We identified sets of high-confidence putative enhancers for ventricular cardiomyocytes and their likely target genes. As a complementary approach, histone ChIP-seq experiments on Hi-C samples were analysed to provide epigenetic features using 3DIV. Annotation of enhancer/super-enhancers and histone ChIP-seq signals were provided for the following: H3K27ac, H3K27me3, H3K36me3, H3K4me1, H3K4me3. Genes with distance-normalised interaction frequency > 2 were retained. In addition, we used promoter capture Hi-C data to detect interactions with gene promoters. These data are summarised in Additional file 1: Table S1.
We applied another strategy to further prioritise the effect of DNVs on human cardiac regulome. We analysed the 62 genes associated with enhancers and promoters containing prioritised DNVs using VarElect to correlate their functions with different aspects of the clinical phenotype. Results suggest that 20 targets were directly related to cardiomyopathy, whereas 41 were indirectly related (Additional file 2: Table S2). One gene (MIR3143) was unrelated and therefore excluded from the analysis. Among the unified results, the top five genes with the highest score of correlation were JPH2, UTRN, H1-2, RHOD, and SAP30B. This score is an indication of the strength of the connection between the gene and the queried phenotypes. The score helps to rank and prioritise the list of queried genes by relevance to the disease. Interestingly, many of the top scoring genes were associated with the same DNV.
In the proband of Fam499, we identified a DNV within an enhancer of JPH2 (Fig. 3). This gene exhibited the highest phenotype association (VarElect score 36.89) (Additional file 2: Table S2). The proband, female and of British ethnicity, was diagnosed with HCM and reported to have died due to sudden cardiac arrest in the year 2020, at the age of 19. Both parents and natural sibling recruited in the study were unaffected. The junctophilin-2 gene (JPH2) is the major structural protein in cardiomyocytes for coupling of transverse tubule-associated L-type Ca2+ channels and type-2 ryanodine receptors on the sarcoplasmic reticulum within junctional membrane complexes (JMC) [47]. Signalling between these two Ca2+ channels is required for normal cardiac contractility. Disruption of the JMC is a common finding in failing hearts. Downregulation of JPH2 gene has been associated with heart failure, and mutations in this gene are associated with HCM [47, 48]. JPH2 was the only high-evidence gene found in our noncoding DNV analysis. We recognise that caution be exercised in the interpretation of variants in cardiomyopathy genes lacking robust evidence; however, the following preliminary results may help to explain the complex genetic architecture of cardiomyopathy. The proband of Fam126 was diagnosed with DCM and harbours a DNV within an enhancer region that regulates the genes UTRN, STX11, and SF3B5. Diseases associated with UTRN (utrophin) include muscular dystrophy, endothelial dysfunction, and DCM [49, 50]. The proband of Fam313 was diagnosed with DCM, including dysplastic tricuspid valve, right ventricular cardiomyopathy, tricuspid regurgitation, dyspnea, pulmonic stenosis, café au lait spot, and congenital heart disease. Although no DNVs were identified in enhancer regions, a deleterious coding DNV was identified in ADGRV1 (Table 2). In addition, we used pcHi-C to detect distal promoter-interacting regions and found that a DNV in this proband is associated with a large cluster of histone genes on human chromosome 6. Studies have shown that histone acetylation/deacetylation regulates cardiac morphogenesis, growth, and contractility [51]. Gene expression profiles of DCM patients have also shown that several histone family members are downregulated [52]. We hypothesise that the downregulation of these genes, which are responsible for higher order chromatin structure, may contribute to the clinical presentation in this proband via nucleosome formation blockade [53]. The proband of Fam334 was initially diagnosed with DCM, and later suspected to have aortopathy and early hypertension. Other diagnoses include lethargy, torsion of testis, headaches, and palpitations. The proband died in the accident and emergency department in the year 2018, at the age of 21. Neither parent or three natural siblings were affected. The DNV identified in this proband is (1) within the enhancer region of genes UNC13D, WBP2, SAP30BP and TRIM65; (2) overlapping the promoter regions of genes H3-3B, MIR4738 and UNK; and (3) interacting with distil gene promoters of TRIM56 and TMEM94. Diseases associated with TMEM94 include cardiac defects [54]. In Fam791, the proband was diagnosed with ARVC and DCM, in addition to arrhythmia, varicose veins of lower extremities with ulcer and inflammation, atherosclerotic heart disease, renal failure, gastrointestinal haemorrhage, and anal polyp. A DNV was identified within the enhancer region of genes GRK2 and RHOD, and overlapping the promoter region of RAD9A. G protein-coupled receptor kinase-2 (GRK2) regulates many cellular and physiological processes, including cardiac contractility, cell proliferation, cell cycle regulation, angiogenesis and vasodilatation. Inhibiting GRK2 can enhance cardiac contractility and protect from adverse heart remodelling in disorders related to cardiac dysfunction [55], suggesting its inhibition as a therapeutic strategy for heart failure. We hypothesise that this DNV may elevate levels and activity of this kinase, thus promoting cardiovascular disease. Moreover, phenome-wide associated loci in the proximity of RHOD is a likely causal gene for cardiomegaly and hematemesis [56], the latter of which may explain the gastrointestinal bleeding observed in this patient.
The contribution of disease-causing rare variants in noncoding regulatory regions remains elusive. The identification of candidate noncoding DNVs in our trio analysis led us to investigate high-risk regulatory variants associated with cardiomyopathy genes in a large cohort lacking a pathogenic mutation in gene panel testing. Overall, combining data from all genes, there was a significant difference in the proportion of cardiomyopathy cases and controls carrying one or more rare variant in regulatory elements of strong or definitive disease-risk genes (p = 0.035, OR = 1.43, 95% Cl = 1.095–1.767). Of the 843 probands lacking a molecular diagnosis, we performed variant-level analysis and identified 25 noncoding variants that were significantly associated in cases (p< 0.05). Of these, 9 predicted to effect TF binding motifs. Eight of the 12 genes investigated had one or more rare variant in regulatory regions, including DSP, RBM20, LMNA, TNNT1, TNNT2, BAG3, DES, and PLN. The highest-ranking regulatory elements for each gene are listed in Additional file 3: Table S3. Most of the significant variants (n = 23; 92%) were “private” to a single proband, with only two variants occurring in two unrelated probands, albeit with the same cardiomyopathy subtype. The private variants were not identified in control populations. Most variants (76%) occurred in HCM, 12% occurred in DCM, 8% occurred in ARVC, and 4% were observed in LVNC cases. These data are summarised in Table 4.
The pathogenesis of cardiomyopathy is largely unknown, and the diagnosis is challenging due to its clinical heterogeneity, involving incomplete penetrance and variable expression. The analysis of DNVs in clinically well-defined phenotypes is a powerful approach to delineate the aetiological basis of disease as it focuses on a relatively small number of variants that provide strong evidence of pathogenicity [57]. DNVs are responsible for the relatively high prevalence of complex disorders. The estimated rate for human germline de novo SNVs is (1.0 to 2.4) × 10−8 per base per generation [58, 59]. This translates to an average of 32 to 76.8 variants in the human genome, with one or two in exonic regions. We had an average of 69.7 DNVs per trio analysis, giving a mutation rate of 2.2 × 10−8 per base per generation. This is consistent with previous studies, thus showing the high quality of our data. We combined gene panel testing with Exomiser, a phenotype-based algorithmic framework, to prioritise inherited and DNVs. A definite or probable genetic diagnosis was identified in 21 probands according to the ACMG guidelines. Additional DNVs of potential clinical significance were identified in 30 genes, 11 of which were within the top three ranked by Exomiser, including TUBA1B, KIRREL1, DAAM1, DOCK11, and KDM5B. In addition, we integrated WGS and single-cell epigenomics to examine the role of regulatory DNVs in cardiomyopathies. Despite the genetic heterogeneity of cardiomyopathy, which stifles efforts to unequivocally demonstrate a causal role for individual noncoding variants, our results provide multiple lines of evidence to indicate the aetiological basis of functional regulatory variants in the human heart regulome. Notably, a DNV was identified within an enhancer of JPH2, a gene associated with HCM and the highest scored in our analysis. Interestingly, we found that more than one rare variant in different cardiomyopathy genes may be relevant for disease causation. Other studies have shown that cardiomyopathy can arise from co-inheritance of rare genetic variants that are benign on their own but harmful in combination [60]. The assumption that all or most patients will receive a single-gene diagnosis is now relegated to the margins. Investigating additional affected families does not necessarily lead to novel gene discovery, thus necessitating the exploration of non-Mendelian contributors to causation or risk. To further add weight to the hypothesis that noncoding variants are associated with cardiomyopathy, we performed a case-control analysis in a mutation-negative cohort and found an enrichment of high-impact regulatory SNVs in cases compared to controls. A variant-level association test showed that 25 SNVs were significantly associated with disease, of which 23 were not identified in control populations and nine are predicted to alter TF motifs. There were several limitations in this study. It is possible that probands in the trio analysis inherited variants within noncoding loci associated with disease, or inherited coding variants in genes beyond those listed in the applied panels. Moreover, in the case–control analysis, we only focused on genes that are strongly associated with cardiomyopathy. We also did not analyse structural variants, such as CNVs, inversions, balanced translocations, or complex rearrangements. Indeed, functional validation of the novel variants reported herein is warranted, a lack of which is acknowledged as a further limitation. Novel variants should not be considered causal merely because they are rare and predicted to be deleterious in silico [61]. Many of the disease-associated variants identified in this study are noncoding, which are in less-well understood regions of the genome. High-throughput assays with functional readout for putative regulatory elements would enable the identification of functional variants and the biological contexts in which they act. Massively parallel reporter assays (MPRAs) permit the high-throughput functional characterisation of noncoding genetic variation [62]. In Fig. 4, we offer a workflow to identify candidate noncoding variants associated with disease, and to assess the molecular consequences of their disruption experimentally. Adapting MPRAs for use in cardiomyocytes will be critical towards understanding cell-type-specific models of regulatory logic in contexts of greater clinical relevance. Our work brings together multiple ‘omics datasets to elucidate the role of pathogenic variants in coding and noncoding loci. This study should prompt extensive genetic analyses and variant-specific experimental modelling to elucidate the complex genetic mechanisms underlying cardiomyopathies.
Additional file 1: Table S1: Prioritised noncoding DNVs within the human ventricular cardiomyocyte regulome.Additional file 2: Table S2: All gene targets of noncoding DNVs directly or indirectly associated with cardiomyopathy.Additional file 3: Table S1: Activity-by-Contact (ABC) model predictions of known cardiomyopathy genes in ventricular cardiomyocytes. |
|
PMC9647984 | Nahideh Nazdikbin Yamchi,Mohammad Mojtaba Alizadeh Ashrafi,Hamed Abbasi,Farhad Amjadi,Mohammad Hossein Geranmayeh,Reza Shirazi,Amin Tamadon,Reza Rahbarghazi,Mahdi Mahdipour | Classical music restored fertility status in rat model of premature ovarian failure | 09-11-2022 | Premature ovarian failure,Music therapy,Sex-related hormones,Brain activity,Ovarian rejuvenation | Background: The restorative effect of classical music was assessed on the cyclophosphamide-induced animal model of premature ovarian failure (POF). Methods: Mozart’s piano classical music (K.448) was used for up to 4 and 8 weeks. Rats were exposed to music 6 h every day using a stereo system with a volume of 65–70 dB. Sera and ovarian tissue samples were collected for the evaluation of FSH, LH, and E2 and histopathological examination. At the same time points, samples were taken from the hypothalamus and hippocampus to monitor the expression of Ntrk2, Crh, and Pomc using real-time PCR. Mating trial was performed to evaluate the fertility status of POF rats. Results: Histopathological examination revealed a significant increase (p < 0.05) in the numbers of morphologically normal follicles at all the developmental stages in POF rats after music therapy compared to the POF group (p < 0.05). Music therapy decreased FSH and LH levels to near-to-normal levels conidied with elevation of E2 (p < 0.05). Ntrk2, Crh, and Pomc expressions were down-regulated in POF rats. Music therapy increasaed the expression of Ntrk2 in the hypothalamus of POF rats (p < 0.05). In contrast, Crh and Pomc failed to reach the detection limit before intervention and four weeks after the intervention however, these genes were expressed eight weeks after music therapy. Fertility status was increased (p < 0.05) in terms of litter size in POF rats after being exposed to music compared to the non-treated POF control group (p < 0.05). Conclusion: Results showed that music can exert therapeutic effects on POF rats via the alteration of sex-related hormones. Supplementary Information The online version contains supplementary material available at 10.1186/s12906-022-03759-y. | Classical music restored fertility status in rat model of premature ovarian failure
The restorative effect of classical music was assessed on the cyclophosphamide-induced animal model of premature ovarian failure (POF).
Mozart’s piano classical music (K.448) was used for up to 4 and 8 weeks. Rats were exposed to music 6 h every day using a stereo system with a volume of 65–70 dB. Sera and ovarian tissue samples were collected for the evaluation of FSH, LH, and E2 and histopathological examination. At the same time points, samples were taken from the hypothalamus and hippocampus to monitor the expression of Ntrk2, Crh, and Pomc using real-time PCR. Mating trial was performed to evaluate the fertility status of POF rats.
Histopathological examination revealed a significant increase (p < 0.05) in the numbers of morphologically normal follicles at all the developmental stages in POF rats after music therapy compared to the POF group (p < 0.05). Music therapy decreased FSH and LH levels to near-to-normal levels conidied with elevation of E2 (p < 0.05). Ntrk2, Crh, and Pomc expressions were down-regulated in POF rats. Music therapy increasaed the expression of Ntrk2 in the hypothalamus of POF rats (p < 0.05). In contrast, Crh and Pomc failed to reach the detection limit before intervention and four weeks after the intervention however, these genes were expressed eight weeks after music therapy. Fertility status was increased (p < 0.05) in terms of litter size in POF rats after being exposed to music compared to the non-treated POF control group (p < 0.05).
Results showed that music can exert therapeutic effects on POF rats via the alteration of sex-related hormones.
The online version contains supplementary material available at 10.1186/s12906-022-03759-y.
Premature ovarian failure (POF) is defined as an ovarian dysfunction, leading to the alteration of the development of ovarian follicles [1, 2]. Statistics have revealed that POF affects about 1% of women under 40 years old and about 0.1% of women under the age of 30 [3]. Both environmental and genetic factors are noted to cause POF, whereas chromosomal defects (fragile X syndrome), toxins (including chemotherapy and radiotherapy), autoimmune diseases, infections, and thyroid malfunction are reported to alleviate this disorder [3, 4]. In the POF patients, follicle-stimulating hormone (FSH) levels reach above 40 IU/mL while the content of anti-Mullerian hormone (AMH) declines below 1 ng/mL [5]. Different clinical symptoms such as amenorrhea, hypoestrogenism (estrogen reduction), and hyper-gonadotropism (increased levels of gonadotropins) can be manifested in POF women [6]. Until now, various pre-clinical strategies have been developed for the treatments of POF including hormone replacement therapy (HRT), gonadotropin-releasing hormone (GnRH) application, and modalities that are associated with the application of whole-cell, or cell-based products [7–14]. Despite the recent progress in the alleviation of POF, there is a great deal of emphasis on the use of complementary therapies. Among them, music therapy is at the center of attention. Music therapy has a long history and dates back to the writings of Plato, Pythagoras, and Aristotle, who were all aware of the power of prevention and treatment of music [15]. It has been indicated that this approach can lead to relaxation, accelerate the healing process of diseases, and improve mental function [16–18]. Music exerts its effects through the coordination of different rhythms of the body and regulates physiological responses in different ways [19]. Music can stimulate the pituitary gland to release several hormones into the nervous system and bloodstream [20]. Among the musical genres, the physiological and behavioral effects of classical music were studied in rats [21]. Besides, studies have been conducted to investigate the effect of music on anxiety and biological factors using workpieces of musicians such as Bach, Beethoven, Mozart, etc. [22, 23]. Classical music therapy has been practiced successfully on mice animal models for stress [24–26], diabetes mellitus [27], breast cancer [28], bone cancer [29], Alzheimer’s disease [20], autism [30], and schizophrenia [31]. Various signaling molecules are associated with POF including Ntrk2 in which loss of the Ntrk2/Kiss1r pathway in oocytes has been shown to cause POF conditions [32]. Previous studies have further shown this gene besides its unique influence on the development of the nervous system, is involved in controlling ovarian function. Thus, the ovaries of mice lacking Ntrk2 receptors show fewer primary follicles and causing deficiency in early follicular growth [33]. Also, the components of the corticotrophin-releasing factor (CRH) family, as a stress hormone receptor system, helps both initiate stress responses and restore systems to homeostasis after the removal of the stressor [34]. CRH can regulate steroidogenesis which is involved in follicular maturation, ovulation, and luteolysis [35]. Also, Proopiomelanocortin (POMC), is a precursor protein detected in the female reproductive system, from which peptides are synthesized in the ovary, and has been confirmed to play a significant role in ovarian function [36, 37]. Considering the positive effects of music therapy in various pathological conditions, in this study, we investigated the restorative effects of non-invasive music therapy on an experimentally induced rat model for POF. To this end, we explored the follicular counts, hormonal alterations as well as the expression of transcripts in the favor of regeneration of ovarian tissue and fertility preservation.
Here, 23 female Wistar rats (7–8 weeks old), weighing between 150 and 180 g, were purchased from Med Zist Company-Tehran. Rats were housed in the normal environment with a temperature of 22 ± 2 °C and 12 h of light/dark cycle and free access to standard pellet and water. Before the experiments, rats were kept untreated for 1 week for environmental adaptation. All experimental protocols were confirmed by the local ethics committee of Tabriz University of Medical Sciences (IR.TBZMED.VCR.REC.1398.361).
To induce the rat model of POF, 20 rats were subjected to the intraperitoneally (IP) administration of cyclophosphamide (CTX; Cat no: RHRI404, Supelco) at a dose of 200 mg/kg on day 1 and 8 mg/kg on days 2 to day 14. CTX is an active substance and can destroy follicles by the mechanism of apoptosis and tissue necrosis [38]. According to previous protocols, 21 days after the last injections rats can exhibit POF features [39]. To confirm POF status, 3 rats were randomly selected from both POF and the control groups and euthanized using an overdose of Ketamine and Xylazine. The left 18 rats were arbitrarily allocated into Control, POF, and POF plus music therapy. The rats in the experimental group were kept in the music box for 4 and 8 weeks (Fig. 1A).
To make an enclosed environment with an adequate rate of ventilation, illumination, and temperature, we designed an insulated acoustic box equipped with a control panel to adjust light, ventilation, and sound systems similar to room conditions (Fig. 1B). The sound speakers were connected from outside to an mp3 stereo playback system. Before playing the music, the volume was measured with a decibel meter. Finally, Mozart’s Piano Classical Music (K 448) was played for 6 h every day from 4 to 10 pm with a volume of 65–70 decibel (dB) up to 4 and 8 weeks. The developed box could hold two cages.
For euthanization, an overdose of Ketamine and Xylazine was administrated IP. Blood was taken directly from the heart to investigate the serum levels of hormones. For histopathological examination, tissues were collected, rinsed in phosphate-buffered saline (PBS) solution, and fixed in 10% formalin (Merck). For real-time PCR analysis, brain tissues (hypothalamus and hippocampus) were individually sampled in cryovials and stored at -80 until the analyses.
Formalin-fixed ovarian specimens were embedded in paraffin after dehydration in alcohol series and cut to a thickness of 5 μm using a microtome instrument (Leica). Hematoxylin and eosin (H&E) staining was performed to study the numbers and the quality of follicles at different developmental stages and corpus luteum (CL) [40]. Masson’s trichrome staining was also executed to evaluate collagen fiber deposition as a sign of tissue fibrosis [41]. After staining, follicular populations and the presence of collagen fibers were evaluated under an Olympus BX-51 light microscope and recorded using a digital camera.
To measure serum levels of FSH, LH, and E2, blood samples were clotted in glass tubes and serum was collected after centrifugation for 20 min at 400xg and stored at -80 °C. Enzyme-linked immunosorbent assay (ELISA) method was performed using a commercial kit for measuring the levels of FSH (0334 − 96, Monobind), LH (0234 − 96, Monobind) and E2 (4925-300 A, Monobind(.
To evaluate the expression of Ntrk2, Crh, and Pomc, hypothalamus and hippocampus samples were subjected to RNA isolation according to the protocol (Traysol: 0000124, MaxZol). Then, the RNA was reverse-transcribed to cDNA (cDNA synthesis kit; YT4500, Yekta Tajhiz Azma). Specific primer pairs were designed using online software (www.ncbi.nlm.nih.gov/tools/primer-blast/) by considering different variables for each gene (Table 1). Subsequently, a quantitative real-time polymerase chain reaction (qRT-PCR) was performed using cDNA and SYBR Green 2 × (5,000,850, Ampliqon) with the “Roche Light Cycler 96” system. The annealing temperature was detected using a gradient PCR. The PCR reaction program was performed in 45 cycles with denaturation, annealing, and extension (95, 60, and 72 °C respectively all last for 15 s). Finally, the specificity of each reaction was evaluated by analyzing the melting and propagation curves.
Finally, the remaining three rats from both POF + music and POF control groups were mated with fertility-proven male rats in a ratio of 2:1 to assess their reproductive status. After examining the vaginal plug, the rats were placed individually in separate cages for 3 weeks. Finally, the litter size/rat was registered.
All the results presented in mean ± SEM were examined by Graph Pad Prism 8 software. To evaluate the statistical significance between groups, data were analyzed with one-way ANOVA with a post-hoc test (Fisher’s least significant difference, LSD). The student’s t-test was incorporated to analyze the significant differences between the two groups. P-values less than 0.05 were considered significant.
Following CTX injection, H&E staining was performed. The general follicular atresia was noted in the POF rats compared to the control group. The number of morphologically healthy follicles at all developmental stages was significantly declined in the POF rats (p < 0.05). Our findings illustrated that the CTX has successfully induced morphological conditions similar to the POF features (Fig. 2A, B). Masson trichrome staining also revealed general collagen fiber deposition within the ovarian tissue of the POF rats (Fig. 2C).
We also examined the follicle population of morphologically healthy and atretic follicles after the intervention. Our findings showed that the total number of morphologically healthy follicles was increased significantly in the music group four and eight weeks after exposure to classical music compared to the control POF rats (p < 0.01and p < 0.001 respectively; Fig. 3B). Similar to what we observed before starting the intervention (Fig. 3A). In contrast, the total number of atretic follicles significantly declined after intervention (p < 0.01; Fig. 4A). Further looking at follicular development stages of primordial, primary, secondary, and antral, a general significant improvement in the numbers of morphologically healthy follicles was noted. According to our data, the number of atretic follicles was increased following the induction of POF in rat ovarian tissue. This pattern was reversed 4 and 8 weeks post-music therapy (Supplementary Fig. 1). Our results showed that POF rats had a relatively lower number of CL compared to the control healthy rats (p < 0.05; Fig. 3 C). After exposure to music, the number of CL was increased. However, the differences were not statistically significant (p > 0.05; Fig. 3 C). In terms of fibrotic changes, our results showed the reduction in collagen fiber deposition in ovarian tissue of POF rats 8 weeks after being exposed to music (Fig. 4B).
We examined the serum levels of FSH, LH, and E2 hormones in POF rats before and after music intervention. In POF rats, FSH level was elevated significantly in response to the CTX administration (p < 0.05). Along with these changes, LH and E2 levels were increased and decreased respectively, however, the differences were not statistically significant (p > 0.05). In POF rats exposed to music therapy, serum levels of FSH were statistically significant differences at week 8 (p < 0.05). A significant decrease pattern was also noted in terms of LH levels (p < 0.01) at week 4 post music intervention. Despite the increase of E2 in music-treated POF rats at both time points, the changes were not statistically significant (Fig. 5).
Real-time PCR results showed that Ntrk2, Crh, and Pomc genes were down-regulated in hypothalamus and hippocampus tissues following the induction of POF. Notably, the differences were only statistically significant for Ntrk2 (p < 0.05; Fig. 6). In the hypothalamus, Ntrk2 expression was interestingly up-regulated 4 weeks after music therapy (p < 0.01). In contrast, both Crh and Pomc genes did not reach the detection limit at this time point. Nevertheless, non-significant differences were noted regarding the expression of subjected genes in the hypothalamus 8 weeks after music therapy (Fig. 6 A). In hippocampus tissue, only the expression of the Ntrk2 gene was detected 4 weeks after therapy without significant changes compared to the POF rats (p > 0.05). Eight weeks after music therapy, Crh and Pomc expression were down-regulated with only significant changes for Pomc (p < 0.05) (Fig. 6B).
To evaluate the fertility status in rats, three remaining rats from the music group and three rats from the POF control group were mated to examine the number of offspring eight weeks after music therapy. In the music group, rats gave birth to 7, 6 and 9 healthy babies, however, only one rat gave birth to 5 babies which were in total statistically significant when the two groups were compared (p < 0.05; Fig. 6 C).
POF is one of the main causes of infertility in different countries with great clinical and economical concerns [42]. To date, routine therapeutics have been not effective enough to restore the function of ovarian tissue [43]. Therefore, various strategies have been practiced to regenerate ovarian tissue including cell and cell-product-based approaches mostly in animal setups [5]. The creation of chemotherapy-induced POF models has received a great deal of attention in recent years. Chemical compounds like Busulfan, Cisplatin, and CTX are shown to cause follicular atresia and depletion in ovarian tissue, mimicking POF-like conditions [44–46]. Here, we successfully induced a rat model for POF using CTX [47]. Biochemical analysis showed that the levels of FSH and LH hormones were significantly increased in the POF conditions. In contrast, the induction of ovarian insufficiency can lead to the reduction of E2 [48, 49]. According to changes in serum levels of sex-related hormones, effective treatment should focus on the regulation of these hormones. Based on our data, music therapy can reduce increased levels of FSH and LH in POF rats 8 weeks after treatment with music [50]. Statistically significant differences were notified between FSH, LH, and E2 levels in the music-treated POF rats compared to the non-treated POF group. These data showed that music can alter serum levels of sex-related hormones in the POF rats. Folliculogenesis is an important part of ovarian function and provides oocytes for reproduction [14]. In the POF conditions, the population of healthy follicles declines due to general atresia, leading to massive fibrosis. In the present study, healthy primordial, primary, secondary, and antral follicles were increased in POF rats four and eight weeks after being exposed to music. We also monitored the expression of Ntrk2, Crh, and Pomc in both the hypothalamus and hippocampus tissues. We noted that the POF condition can reduce the expression of these genes in both target tissues whereas even in some cases the expression level did not fall within the detection limit. In line with our findings, various studies have shown that Ntrk1, 2 are putative players in controlling ovarian function in addition to developing the nervous system [51]. Likewise, other researchers have reported that TrkA and B receptors encoded by Ntrk1, 2 facilitate follicle accumulation and early follicular growth in rat ovaries whereas ovaries of mice lacking the Ntrk gene had fewer primary and secondary follicles [33, 52]. Dorfman et al. stated that deletion of the Ntrk2 gene in mice oocytes resulted in POF conditions [53]. These findings show that the hypothalamic-pituitary-adrenocortical axis is altered shortly when animals failed to proceed with their normal oogenesis. We found that the expression of Ntrk2 was up-regulated 4 weeks after music therapy while the expression of Crh and Pomc were not quantifiable. After 8 weeks, however, the expression reached to close to the control POF group except Pomc gene in which significant reduction was noted (p < 0.05). Components of the corticotrophin-releasing factor (CRH) family, a stress hormone receptor system, help both initiate stress responses and restore systems to homeostasis after the removal of stressors [34]. This gene has also been identified in the reproductive system (ovary, endometrium, placenta, and testis). In the human ovary, receptors are detected in stromal cells and follicular fluid. CRH regulates the ovary in steroidogenesis and is involved in follicular maturation, ovulation, and luteolysis [35]. In a study by Calogero et al., the Crh gene was shown to be able to suppress estrogen production in mouse and human granulosa cells in vitro [54]. The results of our investigation also showed that the Crh gene is not expressed in the hypothalamus and hippocampus after the production of the POF model and also four weeks after receiving music, however not significant, expressed eight weeks after music therapy, probably due to the short timing exposure to the music. The expression of Pomc transcript expression has been confirmed in the ovary by various studies [37, 55]. In a study conducted by Galinelli and co-workers, the expression of Pomc was revealed in the ovaries of women of fertile age to be higher than women in post-menopausal states [37]. Chen et al. noted that the expression of the Pomc gene is regulated by gonadotropins in the ovaries, and experiments on rats showed that Pomc-derived peptides were more abundant during pregnancy than in immature rats [36]. In this experiment, Pomc expression was not quantifiable shortly after induction of POF model similar to Crh results. Eight weeks after music therapy, Pomc transcript was detected in both target samples in which significant downregulation was registered the hypothalamus tissue of music-treated POF group compared to the control POF rats (p < 0.05). This could be probably a sign of tissue rejuvenation as a result of therapy. Finally, the fertility of music-treated mice was assessed after eight weeks. According to the results of previous studies, we also showed that the number of offspring in POF rats exposed to music was more related to the non-treated POF group [9, 12, 56].
To the best of our knowledge, there is no enough data associated with the therapeutic effects of classical music on the restoration of POF consequences either in animal models or human counterparts. Our findings highlighted the positive effects of music on POF rats done via the improvement of ovarian function in terms of healthy follicles and hormonal activity. Music therapy can facilitate the restoration of fertility and diminish the possibility of tissue fibrosis via the changes in levels of sex-related hormones. It can be proposed that music therapy, as a non-invasive and complementary modality, can be considered as an alternative and/or combined approach for various fertility-related complications notably POF patients.
Below is the link to the electronic supplementary material. Supplementary Material 1 |
|
PMC9647995 | Jiyuan Hu,Linhui Wang,Luanfeng Li,Yutao Wang,Jianbin Bi | A novel focal adhesion-related risk model predicts prognosis of bladder cancer —— a bioinformatic study based on TCGA and GEO database | 10-11-2022 | Bladder cancer,Computational biology,Risk signature,Immune infiltration,Focal adhesion | Background Bladder cancer (BLCA) is the ninth most common cancer globally, as well as the fourth most common cancer in men, with an incidence of 7%. However, few effective prognostic biomarkers or models of BLCA are available at present. Methods The prognostic genes of BLCA were screened from one cohort of The Cancer Genome Atlas (TCGA) database through univariate Cox regression analysis and functionally annotated by Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. The intersecting genes of the BLCA gene set and focal adhesion-related gene were obtained and subjected to the least absolute shrinkage and selection operator regression (LASSO) to construct a prognostic model. Gene set enrichment analysis (GSEA) of high- and low-risk patients was performed to explore further the biological process related to focal adhesion genes. Univariate and multivariate Cox analysis, receiver operating characteristic (ROC) curve analysis, and Kaplan–Meier survival analysis (KM) were used to evaluate the prognostic model. DNA methylation analysis was presented to explore the relationship between prognosis and gene methylation. Furthermore, immune cell infiltration was assessed by CIBERSORT, ESTIMATE, and TIMER. The model was verified in an external GSE32894 cohort of the Gene Expression Omnibus (GEO) database, and the Prognoscan database presented further validation of genes. The HPA database validated the related protein level, and functional experiments verified significant risk factors in the model. Results VCL, COL6A1, RAC3, PDGFD, JUN, LAMA2, and ITGB6 were used to construct a prognostic model in the TCGA-BLCA cohort and validated in the GSE32894 cohort. The 7-gene model successfully stratified the patients into both cohorts’ high- and low-risk groups. The higher risk score was associated with a worse prognosis. Conclusions The 7-gene prognostic model can classify BLCA patients into high- and low-risk groups based on the risk score and predict the overall survival, which may aid clinical decision-making. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-022-10264-5. | A novel focal adhesion-related risk model predicts prognosis of bladder cancer —— a bioinformatic study based on TCGA and GEO database
Bladder cancer (BLCA) is the ninth most common cancer globally, as well as the fourth most common cancer in men, with an incidence of 7%. However, few effective prognostic biomarkers or models of BLCA are available at present.
The prognostic genes of BLCA were screened from one cohort of The Cancer Genome Atlas (TCGA) database through univariate Cox regression analysis and functionally annotated by Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. The intersecting genes of the BLCA gene set and focal adhesion-related gene were obtained and subjected to the least absolute shrinkage and selection operator regression (LASSO) to construct a prognostic model. Gene set enrichment analysis (GSEA) of high- and low-risk patients was performed to explore further the biological process related to focal adhesion genes. Univariate and multivariate Cox analysis, receiver operating characteristic (ROC) curve analysis, and Kaplan–Meier survival analysis (KM) were used to evaluate the prognostic model. DNA methylation analysis was presented to explore the relationship between prognosis and gene methylation. Furthermore, immune cell infiltration was assessed by CIBERSORT, ESTIMATE, and TIMER. The model was verified in an external GSE32894 cohort of the Gene Expression Omnibus (GEO) database, and the Prognoscan database presented further validation of genes. The HPA database validated the related protein level, and functional experiments verified significant risk factors in the model.
VCL, COL6A1, RAC3, PDGFD, JUN, LAMA2, and ITGB6 were used to construct a prognostic model in the TCGA-BLCA cohort and validated in the GSE32894 cohort. The 7-gene model successfully stratified the patients into both cohorts’ high- and low-risk groups. The higher risk score was associated with a worse prognosis.
The 7-gene prognostic model can classify BLCA patients into high- and low-risk groups based on the risk score and predict the overall survival, which may aid clinical decision-making.
The online version contains supplementary material available at 10.1186/s12885-022-10264-5.
According to Cancer Statistics 2021, published by American Cancer Society, bladder cancer (BLCA) is the ninth most commonly diagnosed cancer globally and is the fourth most common malignancy in men [1]. Over 570,000 new cases of BLCA and 210,000 deaths were recorded in 2020 alone, indicating poor prognosis [2]. Men are at four times the risk of developing BLCA than women [3]. The significant risk factors of BLCA are advanced age (between 70 and 84 years) and cigarette smoking. In fact, approximately 50% of BLCA patients are smokers [4]. Furthermore, almost 3/4th of the diagnosed cases are non-muscular invasive bladder cancer (NMIBC), often treated with transurethral resection of bladder tumors (TRUBT) and intravesical therapy. Muscular invasive bladder cancer (MIBC) is relatively rare and is generally treated by radical cystectomy and neoadjuvant chemotherapy [5, 6]. Although pathological biopsies and cystoscopies are routinely used to detect BLCA, these methods are invasive and inconvenient. Although several urine biomarkers of BLCA have been confirmed by the US Food and Drug Administration (FDA), they lack the diagnostic accuracy to replace cystoscopy [4]. Therefore, this study aimed to identify novel, effective diagnostic biomarkers of BLCA. Focal adhesion (FA) is a group of macromolecular proteins that connect the ends of specialized actin fibers to the extracellular matrix (ECM) and enable cell migration, which is critical to the process of tumor metastasis [7]. FAs are frequently downregulated during cancer metastasis, although some FA components are upregulated in some invasive tumors [8]. Thus, FAs are increasingly being considered therapeutic targets of cancer. Analysis of gene expression datasets from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) have helped identify prognostic gene signatures of various cancers. For instance, a predictive model consisting of hypoxia gene signatures was constructed for BLCA based on TCGA and GEO databases [9]. In addition, a risk score model of epithelial-mesenchymal transition (EMT)-related gene signature was also developed to predict BLCA prognosis based on the two databases [10]. A recent study established an 11-gene prognostic signature of BLCA based on five cohorts from TGCA and GEO [11]. However, the prognostic value of FAs has not been ascertained by any study so far. Therefore, this study aimed to explore the relationship between FAs and BLCA prognosis using the bioinformatics approach and establish a predictive model based on the risk score.
TCGA and GEO databases were screened for BLCA cohorts with a sample size > 150, including clinical data such as overall survival duration, survival status, gender, age, histological grade, pathological stage, TNM stage, and lymphatic stage metastasis. The GEO cohorts were further screened based on additional requirements for the verification set. The gene expression matrix dataset (HTSeq-FPKM) of bladder cancer (n = 430) was downloaded from TCGA on UCSC Xena, and the clinical data were obtained from cBioPortal (http://www.cbioportal.org). The external GSE32894 cohort (n = 308) with expression matrix and clinical data [12] was acquired from the GEO database.
TCGA-BLCA cohort was set as the training set, and prognostically relevant genes were screened using the univariate Cox analysis with a p-value < 0.01 as the criterion. R software package “limma” was used to identify DEGs between the BLCA and normal bladder samples in the same cohort from these selected genes [13]. The threshold was set as |log (fold change) |> 1, and the adjusted p-value < 0.01. The significant DEGs related to BLCA prognosis intersected with 199 FA-related genes obtained from the Molecular Signatures Database (MSigDB) of GSEA using keywords KEGG_FOCAL_ADHESION [14] using a Venn diagram.
A prognostic model was constructed with the intersecting genes identified as above by LASSO regression using the R software packages “glmnet” [15] and “survival”. The “CV.glmnet” function can randomly simulate 1000 times for k-fold cross-validation (k = 10). The dataset was automatically divided into 10 equal portions in the tenfold cross-validation. One random part was selected as the validation set, and the remaining 9 parts as training sets. The deviance of the 10 tests was used to evaluate the accuracy of the tenfold CV, and minimum deviance indicated the best performance of the model. The regression coefficients of individual genes were determined, and genes with regression coefficient approaching 0 with the increase in Lambda were excluded. The remaining candidate genes were used to construct the model, and the risk score of each patient in the TGCA-BLCA cohort was calculated as ∑( · ), where 7 is the number of candidate genes, is the gene expression value and is regression coefficient.
The latest KEGG Pathway gene annotations were obtained through KEGG rest API ( https://www.kegg.jp/kegg/rest/keggapi.html) in the KEGG official database [16]. KEGG pathway enrichment analysis was then performed using the R software package “clusterProfiler” through an online tool called Sangerbox (http://www.sangerbox.com/tool). The threshold for statistical significance were p < 0.05 and FDR of < 0.1.
GSEA software and predefined gene set files were downloaded from https://www.gsea-msigdb.org, and the samples were divided into high- and low-risk groups based on the risk score. The number of permutations was set as 1000.
The R software package "survival" was used for univariate and multivariate Cox regression analysis of the risk score and clinicopathological factors, including age, gender, pathological stage, T stage, histological grade, lymphatic metastasis, and angio-lymphatic invasion. Only the statistically significant factors in the univariate Cox analysis were included in the multivariate Cox model.
The R software package “pROC” was used for receiver operating characteristic (ROC) analysis. The area under the curve (AUC) was obtained, and the confidence interval was evaluated. The cut-off values for 1-, 3- and 5-year overall survival (OS) were calculated.
The R software package “Survival” was used to integrate the OS rate and duration with the gene expression data of both TCGA-BLCA and GSE32894 cohorts. The prognostic significance of each gene was evaluated by the Cox method. The patients were divided into the high- and low-risk groups using the cut-off value of 3-year OS. For the subgroups based on clinical variables and the expression levels of the 7 candidate genes, the best cut-off value for the risk score was calculated using the R software package “Maxstat”. The minimum sample size was set at > 25%, and the maximum sample size at < 75%, and the patients were divided into high- and low-risk groups.
It is believed that DNA methylation is responsible for influencing prognosis in cancer development. An online tool MethSurv (https://biit.cs.ut.ee/methsurv/) was used to explore the prognostic patterns of single CpG methylation of the 7 genes in bladder cancer [17]. Only the most significant prognostic p-values were selected (likelihood ratio (LR) test p-value).
The R packages “CIBERSORT” (used to calculate the cell composition as a function of gene expression profile) and “ESTIMATE” (used to calculate the fraction of stromal and immune cells according to gene expression level) were used to calculate the number of infiltrating immune cells, immune score, stromal score and tumor purity in each patient from TCGA-BLCA cohort [18, 19]. Twenty-two immune cell genotypes were obtained by combining CIBERSORT with LM22, a gene matrix downloaded from the CIBERSORT website (https://cibersort.stanford.edu/), within the R software. The differences between the risk groups were analyzed, and the immune score and risk score were combined for survival analysis. Besides, the TIMER platform (http://timer.cistrome.org/) was also used to verify the immune infiltration analysis completed by CIBERSORT. The “gene module” of immune association was presented to evaluate the correlation between immune cells and every 7 genes in the prognostic model [20].
The accuracy of the prognostic model was tested on the external GSE32894 dataset. Besides, the Prognoscan database (www.prognoscan.org) was also applied to validate further the correlation between gene expression and overall survival time [21], where GSE5287 and GSE13507 were utilized. The protein expression of individual genes in the model in cancer and normal tissues was also observed in the Human Protein Atlas (HPA) database (http://www.proteinatlas.org/), so as further to validate the genes in our model [22].
The human BLCA cell line T24 was used in this study, purchased from the Chinese Academy of Sciences cell bank. T24 was cultured in RPMI-1640 medium (Procell) with 10% fetal bovine serum. The sequence of siRNA targeting COL6A1 and LAMA2 purchased from JTSBIO Co., were listed in Supplementary Table 5.
Extraction of RNA was performed with RNAiso Plus (Takara). Prime Script RT Master Mix (Takara) was used to reverse transcription then cDNA was produced. The SYBR kit (Takara) was used to perform qRT-PCR. The relative expression of the gene was calculated by the 2−ΔΔCt method. The primer sequences targeting COL6A1, LAMA2 and GAPDH were listed in Supplementary Table 6.
The BLCA cells were seeded in a six-well plate. When the density reached more than 90%, a straight line was drawn with a 200-ul tip. Cultivation of cells was continued with a low-serum medium containing 3% serum. Photographs were taken at 0 h and 48 h, and then the speed of scratch healing was compared between the different groups.
Six hundred ul of medium containing 10% serum was added to the lower chamber of a 24-well plate. Each 200 ul BLCA cell suspension was inoculated in the upper chamber. Transwell chambers with 8-μm-pore were used for cell migration assay. Following incubation for 24 h, cells beneath the membrane were stained with crystal violet, and cells above the membrane were washed off and imaged by microscopy.
All the statistical analysis was completed by software R. The Logrank test was used to assess the significance of prognostic differences between different groups in the Kaplan–Meier analysis. The Kruskal–Wallis rank sum test was used in multiple groups comparisons of clinical sub-group analysis. Univariate analysis and multivariate analysis were performed using Cox regression analysis with the R package “survival”. The R package “limma” was used to identify DEGs between the tumor and normal samples in the same TCGA-BLCA cohort from these selected genes. The R package “glmnet” was used in LASSO regression to establish the predicting model. A p-value < 0.05 was considered as statistically significant.
The gene expression and clinical data of BLCA samples were retrieved from TCGA (n = 430) and GSE32894 (n = 308). After filtering the data, there were 403 cases in the TCGA-BLCA cohort and 224 cases in the GSE32894 cohort. The flow chart is shown in Fig. 1. Clinical data regarding age, gender, histological grade, WHO grade, pathological stage, T stage, lymphatic node metastasis, and angiolymphatic invasion of the two cohorts are summarized in Table 1.
Univariate Cox analysis of the TCGA-BLCA cohort identified 2461 genes (p < 0.01), of which 274 were differentially expressed between the tumor and non-tumor samples (Fig. 2a). Sixteen DEGs intersected with FA-related genes (Fig. 2b), and were functionally annotated by KEGG pathway enrichment analysis (Fig. 2c). The detailed information of these genes is listed in Supplementary Table 1. The above DEGs were subjected to Lasso regression, and 7 genes with the smallest deviance were included in the prognostic model (Figs. 3a, b). The coefficient values and other details of these genes are listed in Supplementary Table 2, and outcomes of univariate regression analysis are summarized in Supplementary Table 3. The risk score was calculated as VCL * 0.1452—ITGB6 * 0.0832 + COL6A1 * 0.0077 + RAC3 * 0.2404 + PDGFD * 0.0817 + JUN * 0.1192 + LAMA2 * 0.1927. Furthermore, we performed a matrix correlation analysis to determine any collinear relationship between these genes. As shown in Fig. 3c, apart from COL6A1 and LAMA2, the co-expression indices of the other gene pairs were all < 0.5.
The 403 samples in the TCGA-BLCA cohort were stratified into high- and low-risk groups based on the 7-gene risk score. GSEA further indicated that the high-risk group was significantly associated with biosynthesis of unsaturated fatty acids, tight junction, lysine degradation, and ubiquitin-mediated proteolysis (p < 0.005; Fig. 4).
The patients in the TCGA-BLCA cohort were stratified into high- and low-expression subgroups for each of the 7 prognostic genes and subjected to Kaplan–Meier analysis to determine their impact on survival. These seven genes include COL6A1 (Fig. 5a), ITGB6 (Fig. 5b), JUN (Fig. 5c), LAMA2 (Fig. 5d), PDGFD (Fig. 5e), RAC3 (Fig. 5f) and VCL (Fig. 5g). And the GSE32894 cohort was divided with the same criterion for validation. These seven genes include VCL (Fig. 6a), COL6A1 (Fig. 6b), ITGB6 (Fig. 6c), JUN (Fig. 6d), LAMA2 (Fig. 6e), PDGFD (Fig. 6f) and RAC3 (Fig. 6g). The heatmap demonstrated correlation between gene and survival (Fig. 6h), except for JUN and PDGFD in GSE32894; all genes comprising the prognostic model were significantly associated with the survival of BLCA patients (p < 0.05). The high expression of ITGB6 indicated a better prognosis in both the training and validation sets. In contrast, increased expression of RAC3, COL6A1, and LAMA2 was correlated with worse prognosis in both sets. Interestingly, high expression of VCL was associated with a worse prognosis in the training set but indicated a favorable prognosis in the validation set. Taken together, ITGB6, RAC3, COL6A1, and LAMA2 could accurately predict patient prognosis.
Univariate and multivariate Cox regression analyses identified age, angiolymphatic invasion, and the risk score as independent prognostic factors of BLCA (Table 2). Furthermore, ROC curve analysis of the three factors indicated that the AUC of the risk score was greater than that of the other two factors (Supplementary Fig. 1), suggesting more substantial predictive accuracy. Furthermore, higher risk scores correlated with more advanced angiolymphatic invasion, T stage, pathological stage, and lymphatic node metastasis. The patients were divided into subgroups based on these clinical factors, and the expression levels of the 7 prognostic genes were compared. As shown in Fig. 7, the expression levels of COL6A1 and LAMA2 were significantly different across all subgroups. Furthermore, the high- and low-risk groups had very different survival rates in subgroups demarcated by age, gender, histological grade, lymphatic node metastasis, and T stage (Fig. 8), indicating that the risk score can predict the prognosis in clinical sub-groups and may assist in clinical decision making.
We presented prognostic values of DNA methylation clustering the expression levels of each 7 genes of the prognostic model in the TCGA-BLCA cohort by KM curves (Supplementary Fig. 2). The CpG islands corresponding to the smallest likelihood ratio (LR) test p-value were chosen in all 7 genes to ensure the statistical significance. The specific CpG resource of each gene was depicted in the figure. Besides, the relationship between DNA methylation and the prognosis of ITGB6 demonstrated its protective effect, displaying the same trend as it does in the 7-gene prognostic model. prognosis of ITGB6 demonstrated its protective effect, displaying the same trend as in the 7-gene prognostic model.
The infiltration ratio of 22 immune cell types was analyzed in the TGCA-BLCA cohort using CIBERSORT (Fig. 9a) and compared between the high- and low-risk groups. As shown in Fig. 9b, the predominant infiltrating immune cells in the high-risk groups were activated CD4 memory T cells, resting dendritic cells (DCs), and activated mast cells, whereas the M1 macrophages and activated DCs showed higher infiltration in the low-risk group (p < 0.0001). Kaplan–Meier survival analysis further showed that high infiltration of resting DCs activated mast cells and activated CD4 memory T cells, along with a high-risk score, which portended the worst prognosis. In addition, low infiltration of M1 macrophages and activated DCs in the high-risk group was associated with the worst prognosis (Supplementary Fig. 3). The immune score and tumor purity in TCGA-BLCA cohort were evaluated using the “ESTIMATE” R package (Supplementary Table 4). The patients’ samples were divided into 4 clusters using the median risk and immune scores. As shown in Fig. 9c, patients with the lowest immune score and highest risk score had the worst prognosis. Besides, M0, M2 macrophages, and neutrophils were also statistically significant (p < 0.05), thus these immune cells were also selected for further analysis on the TIMER platform. After filtering associated immune cells, M2 macrophages were ultimately chosen to explore further the relationship with the 7 genes (Supplementary Fig. 4). Apart from ITGB6, which showed a negative correlation as a protective factor, the other genes were all positively correlated with Macrophages M2_CIBERSORT as risk factors, demonstrating the same trend as the prognostic model. COL6A1, RAC3, LAMA2, and VCL showed significant statistical meanings (p < 0.05).
The optimum cut-off value of the risk score for 3-year OS was -2.2174 and was used to divide the patients in the TCGA-BLCA cohort into the high- and low-risk groups. Likewise, the optimum cut-off value of the validation cohort (GSE32894) was calculated as -0.1290. As shown in Figs. 10a, b, patients in the high-risk group had significantly worse survival than the low-risk group in both the training and verification sets (p < 0.05). Furthermore, the AUC values of the risk score for 1-, 3- and 5-year OS were respectively 0.66, 0.68, and 0.69 in the training set, and 0.66, 0.73, and 0.72 in the validation set (Figs. 10c, d). We analyzed the relationship between the risk scores and OS duration and the changes in the expression of various genes in both cohorts (Figs. 10e, f). As expected, ITGB6 was identified as a protective factor in both cohorts. It was downregulated with the increasing risk score. Besides, limited by the sample size of the Prognoscan online tool, RAC3 and COL6A1 were the only risk factors that demonstrated a significant statistical correlation with the overall survival time of patients (Supplementary Figs. 5 and 6, p < 0.05) in external dataset GSE13507 and GSE5287, respectively. Though KM curves of ITGB6 in the GSE13507 cohort (Supplementary Fig. 7) still showed a tendency as a protective factor, the p-value was > 0.05, with no statistical significance. Taken together, the 7-gene risk model can successfully stratify BLCA patients into prognostic groups. Furthermore, In the validation of protein level on the HPA database, VCL, COL6A1, RAC3, PDGFD and JUN showed higher protein expression in BLCA tissue than in normal tissue. In comparison, ITGB6 showed higher protein expression in normal tissue than in BLCA tissue. In this database, LAMA2 expression was not detected in normal or cancerous tissues (Supplementary Fig. 8).
To further investigate the effect of COL6A1 and LAMA2 on the biological function of bladder cancer cells, a series of functional experiments were performed to verify its effect. After knocking down the expression of COL6A1 (Fig. 11a) and LAMA2 (Fig. 11b), the Transwell assay showed that the migration ability of bladder cancer cells was down-regulated (Figs. 11c, d). The result of the wound-healing assay is consistent with that of the Transwell assay (Figs. 11e, f). COL6A1 and LAMA2 were again validated to promote the migration of bladder cancer cells.
Bladder cancer is a common malignancy of the urinary system with unpredictive outcomes. Several bioinformatics studies in recent years have established prognostic models, including an immune genes-related model [23] and an 11-gene model based on 5 cohorts [11], for clinical decision-making. Furthermore, hypoxia-related risk factors [9] and immune landscapes [24] have also been associated with bladder cancer prognosis. In this study, we successfully established a reliable 7-gene focal adhesion-related prognostic model for BLCA using RNA-seq data from the TCGA-BLCA cohort. We verified it on the external GSE32894 dataset. Given the regional differences between the two datasets, the former being from Europe and the latter from North America, we can conclude that the model can be applied universally. The model comprises of 6 risk genes (VCL, COL6A1, RAC3, PDGFD, JUN, LAMA2) and 1 protective gene (ITGB6), all of which are closely related to focal adhesion. Among the 7 genes, COL6A1 and LAMA2 are the two most significant genes in either KM analysis for predicting prognosis or Kruskal–Wallis rank sum test for combined analysis of expression level and clinical factors, thus, deserve to be further explored. COL6A1, a gene encoding the collagen VI α1 chain, is widely present in the connective tissues of vertebrates [25]. Collagen VI is a major extracellular matrix protein commonly used to support cell structures. Some studies have shown that collagen VI can regulate cell migration, apoptosis, and tumor progression [26, 27]. Previous studies tend to focus on the function of collagen VI itself, due to its deficiency in myopathy and skeletal muscle diseases [28]. However, COL6A1 has been shown to stimulate proliferation and prevent apoptosis of cancer cells, which has also been found to be related to different types of cancers. For instance, it was proved to be a potential marker of cervical cancer progression in Kazobinka G et al.’s research. The over-expression of COL6A1 was correlated with cervical patients’ prognosis and cell biological functions [29]. Besides, the up-regulation of COL6A1 expression induces tumorigenesis in prostate cancer cells in vivo has also been reported in a study about castration-resistant prostate cancer [30] and enhanced probability of lung cancer cell metastasis in another research [31]. Some researchers also reported that the over-expression of COL6A1 contributed to poor prognosis of renal clear cell carcinoma and glioma patients and enhanced probability of lung cancer cell metastasis [32, 33]. In addition, Snipstad K et al. have reported an up-regulation of extracellular matrix proteins COL6A1 and LAMA4 in rectal cancer after radio‐chemotherapy [34]. All this evidence indicated that COL6A1 has a close relationship with tumor progression and was a novel biomarker of prognosis in different types of cancers, not a simple gene related to collagen anymore. As a member of the cell adhesion family, LAMA2 also encodes components of extracellular matrix protein called laminin, a glycoprotein in the connective tissue basement membrane, and promotes cell adhesion [35]. Laminin-α2, encoded by LAMA2, is abundant in skeletal muscle, motor nerves, and the brain. It is a composition of trimeric laminin-211 [36] and is an essential constituent of tumor stromal, which can be associated with the malignancy of the tumor. Since damage to the basement membrane of tumor cells plays a vital role in tumor invasion and transfer, many studies were conducted, and evidence has shown that laminin expression was related to tumor progression [37]. LAMA2 belongs to the laminin family. However, seldom researchers paid attention to the direct correlation between LAMA2 overexpression and tumor progression. Most studies focused on LAMA2 deficiency leading to muscular disease, and it is known that mutations in LAMA2 produce a particularly severe type of congenital muscular dystrophy, called LAMA2 chain deficient congenital muscular dystrophy (LAMA2-CMD) [38]. More importantly, seldom studies have explored the relationship between COL6A1, LAMA2, and bladder cancer, thus of great significance in our study. Besides, other genes in the prognostic model are also related to tumors. For instance, Cheng C et al. claimed RAC3 promoted proliferation, migration, and invasion through PYCR1/JAK/STAT signaling in bladder cancer [39]. VCL has been reported as an important prognostic biomarker in prostate cancer [40]. Satow R et al. reported that PDGFD promotes aggressiveness in prostate and colorectal cancer [41]. Previous research has reported that up-regulation of JUN is associated with the invasiveness of colorectal cancer cells [42]. Singh A et al. reported in 2009 that ITGB6 correlated with a well-differentiated K-Ras-driven cancer such as lung, pancreatic and colon cancer [43]. GSEA of the high- and low-risk groups indicated significant enrichment of biological processes including biosynthesis of unsaturated fatty acids, tight junction, lysine degradation and ubiquitin-mediated proteolysis. proteolysis. Focal adhesion and tight junction commonly belong to enriched cell adhesion/junction pathways [44], indicating that these biological processes can further explore the relationship between tumorigenesis in the future. The risk score also showed more substantial predictive power than multiple clinical factors. Furthermore, patients with advanced clinical features had higher risk scores. Survival analysis indicated that the prognosis of the high-risk group was worse in both the training set and validation set (p < 0.01), with respective 3-year AUC values of 0.68 and 0.73, suggesting that the risk score model was capable of predicting BLCA prognosis independent of the clinical factors. The risk score distribution was also similar in both sets, which indicated good consistency and universality of the risk model. DNA methylation analysis of each 7 genes demonstrated a strong correlation between the gene methylation and overall survival time in the TCGA-BLCA cohort. ITGB6 was a protective factor in our prognostic model and its protective effect was also provn by DNA methylation, as the lower methylation level indicated a better prognosis, which reflected the reliability of the 7-gene prognostic model to some extent. Anuraga G et al. and Wang Z et al. also reported DNA methylation analysis in their gene signature for predicting breast cancer and lung adenocarcinoma via MethSurv [45–47]. The validation of the HPA database on the protein level showed the same tendency as the 7 genes in the prognostic model. Borowczak J et al. reported CDK9 in bladder cancer via the HPA database [48]. The functional experiments successfully verified that COL6A1 promoted the migration of bladder cancer cells as a risk factor, and bladder cancer cells were down-regulated after knocking down the expression of COL6A1. This reflected the reliability of our bioinformatic predicting model. Previous studies have established tumor-infiltrating lymphocytes as one of the immune-related prognostic factors [49], and CD20 B cell has been identified as a long-term survival factor in BLCA [50]. In this study, we found that both resting and activated DCs were relevant to prognosis. High infiltration of DCs and a high-risk score indicate the worst prognosis. Thus, the infiltration ratio of DCs is a potential new prognostic factor that can be combined with the risk score for more accurate prediction. Furthermore, the correlation between every single gene in the model and Macrophage M2 replenished the evidence that immune cells can influence clinical outcomes, assist in specific immunotherapeutic responses, and help select suitable patients for immunotherapy combined with risk score. Previous studies have reported the relationship between gene and immune cells [51, 52]. However, our results are limited because we only analyzed data from TCGA and GEO databases. Though we validated the model on external cohorts, only the GSE32894 dataset showed an excellent result. GSE13507 and GSE5287 did not establish statistically significant relationships between the expression of each 7genes in the model and survival time on the Prognoscan database. Previous researchers have also drawn from such databases to determine the relationship between genes and prognosis [53, 54] Due to the lack of clinical cohorts, the model’s reliability cannot be verified clinically. And the best cut-off value was directly chosen as a 3-year OS value of ROC without calculating in a more accurate method. The two up-regulated key genes COL6A1 and LAMA2 were not confirmed for their significant roles on the basic experimental level.
The 7-gene FA-related prognostic model can accurately predict the prognosis of BLCA patients and aid in clinical decision-making. Further studies are needed to amend its accuracy and stability for clinical applications.
Additional file 1: Supplementary Figure 1. The ROC curves of (a) risk score, (b) angiolymphatic invasion and (c) age for 1-, 3- and 5-year OS and the corresponding AUC values.Additional file 2: Supplementary Figure 2. Kaplan–Meier survival curves of patients demarcated on the basis of high- and low-DNA methylation of each 7 genes in the model. (a) COL6A1, (b) ITGB6, (c) JUN, (d) LAMA2 (e) PDGFD, (f)RAC3, (g)VCL.Additional file 3: Supplementary Figure 3. Kaplan–Meier survival curves of high- and lowrisk patients demarcated on the basis of (a) activated CD4 memory T cell, (b) dendritic cell, (c) activated mast cells, (d) M1 macrophages and (e) dendritic cell infiltration.Additional file 4: Supplementary Figure 4. Correlation scatter plot of Macrophage M2 infiltration ration and expression level of each 7 genes in the model. (a) COL6A1, (b) ITGB6, (c) JUN, (d) LAMA2, (e) PDGFD, (f) RAC3, (g) VCL.Additional file 5: Supplementary Figure 5. Related validation plots on Prognoscan platform of the RAC3 expression. (a) expression level distribution plot, (b) expression level histogram plot, (c) p-value distribution plot, (d) K-M curves of patients with high- and low-expression of RAC3, (e) survival time distribution plot.Additional file 6: Supplementary Figure 6. Related validation plots on Prognoscan platform of the COL6A1 expression. (a) expression level distribution plot, (b) expression level histogram plot, (c) p-value distribution plot, (d) K-M curves of patients with high- and low-expression of COL6A1, (e) survival time distribution plot.Additional file 7: Supplementary Figure 7. Related validation plots on Prognoscan platform of the ITGB6 expression. (a) expression level distribution plot, (b) expression level histogram plot, (c) p-value distribution plot, (d) K-M curves of patients with high- and low-expression of ITGB6, (e) survival time distribution plot.Additional file 8: Supplementary Figure 8. Validation of the seven-mRNA prognostic signature in the HPA database. The deeper the color, the higher expression in the tissues. (a-b) COL6A1 expression was higher in tumor tissues. (c-d) ITGB6 expression was higher in normal tissues. (e-f) JUN expression was higher in tumor tissues. (g-h) LAMA2 expression differences was insignificant between normal and tumor tissues. (i-j) PDGFD expression was higher in tumor tissues. (k-l) RAC3 expression was higher in tumor tissues. (m-n) VCL expression was higher in tumor tissues.Additional file 9: Supplementary Table 1. Detailed genes table.Additional file 10: Supplementary Table 2. Detailed information of the 7 genes.Additional file 11: Supplementary Table 3. Univariate cox regression analysis of 7 genes (HR: hazard ratio; CI: confidence interval).Additional file 12: Supplementary Table 4. “ESTIMATE” results.Additional file 13: Supplementary Table 5. Sequences of siRNA for transfection.Additional file 14: Supplementary Table 6. Sequences of primer pair for qPCR. |
|
PMC9647996 | Zhuoying Chen,Meixiu Huang,Jiaying You,Yanhua Lin,Qiaoyun Huang,Caiping He | Circular RNA hsa_circ_0023404 promotes the proliferation, migration and invasion in endometrial cancer cells through regulating miR-217/MAPK1 axis | 09-11-2022 | hsa_circ_0023404,miR-217,MARK1,Circular RNA,Endometrial cancer | Background Emerging studies indicated that circular RNA hsa_circ_ 0023404 and its target miR-217/MARK1 axis play a critical role in cancer progression such as non-small cell lung cancer and cervical cancer. However, the role of hsa_circ_0023404/miR-217/MARK1 involved in endometrial cancer (EC) was not investigated yet. The aim of this study is to investigate the functions of hsa_circ_0023404 in endometrial cancer (EC) and the potential molecular mechanism. Methods We used RT-qPCR and Western blot approach to detect the expressed levels of related genes in EC cell lines. Transfected siRNAs were applied to knockdown the level of related mRNA in cells. Cell proliferation by CCK-8 assay and colony formation assay were applied to detect cell proliferation. Transwell migration and invasion assay was for detecting the migration and invasion of the cells. Results RT-qPCR showed that the levels of hsa_circ_0023404 and MARK1 mRNA were upregulated, but mirR-217 was decreased in three endometrial cancer cell lines. Knockdown of hsa_circ_0023404 by siRNA markedly increased the level of miR-217 and reduced the proliferation of the Ishikawa cells. It also inhibited the cell migration and invasion. Anti-miR-217 can reverse the promoted proliferation, migrations and invasion of Ishikawa cells mediated by si-circ_0023404. si-MARK1 restored the inhibited cell proliferation, migration and invasion of the co-transfected Ishikawa cells with si- circ_0023404 and anti-miR-217. Conclusion hsa_circ_0023404 exerts a tumor-promoting role in endometrial cancer by regulating miR-217/MARK1 axis. hsa_circ_0023404 inhibit miR-217 as sponge which inhibit endometrial cancer cell growth and metastasis. MARK1 is downstream target of miR217 and upregulated by hsa_circ_ 0023404/miR-217 axis and involved in the endometrial cancer progression. Supplementary Information The online version contains supplementary material available at 10.1186/s40001-022-00866-x. | Circular RNA hsa_circ_0023404 promotes the proliferation, migration and invasion in endometrial cancer cells through regulating miR-217/MAPK1 axis
Emerging studies indicated that circular RNA hsa_circ_ 0023404 and its target miR-217/MARK1 axis play a critical role in cancer progression such as non-small cell lung cancer and cervical cancer. However, the role of hsa_circ_0023404/miR-217/MARK1 involved in endometrial cancer (EC) was not investigated yet. The aim of this study is to investigate the functions of hsa_circ_0023404 in endometrial cancer (EC) and the potential molecular mechanism.
We used RT-qPCR and Western blot approach to detect the expressed levels of related genes in EC cell lines. Transfected siRNAs were applied to knockdown the level of related mRNA in cells. Cell proliferation by CCK-8 assay and colony formation assay were applied to detect cell proliferation. Transwell migration and invasion assay was for detecting the migration and invasion of the cells.
RT-qPCR showed that the levels of hsa_circ_0023404 and MARK1 mRNA were upregulated, but mirR-217 was decreased in three endometrial cancer cell lines. Knockdown of hsa_circ_0023404 by siRNA markedly increased the level of miR-217 and reduced the proliferation of the Ishikawa cells. It also inhibited the cell migration and invasion. Anti-miR-217 can reverse the promoted proliferation, migrations and invasion of Ishikawa cells mediated by si-circ_0023404. si-MARK1 restored the inhibited cell proliferation, migration and invasion of the co-transfected Ishikawa cells with si- circ_0023404 and anti-miR-217.
hsa_circ_0023404 exerts a tumor-promoting role in endometrial cancer by regulating miR-217/MARK1 axis. hsa_circ_0023404 inhibit miR-217 as sponge which inhibit endometrial cancer cell growth and metastasis. MARK1 is downstream target of miR217 and upregulated by hsa_circ_ 0023404/miR-217 axis and involved in the endometrial cancer progression.
The online version contains supplementary material available at 10.1186/s40001-022-00866-x.
Endometrial cancer (EC) is one of the most common types of gynecological cancer and the fourth most common cancer among women. Morbidity and mortality rates among patients with EC remain high globally [1]. Each year, approximately 140,000 women worldwide develop endometrial cancer and an estimated 40,000 women die of this cancer. Most cases of EC are diagnosed after menopause and the highest incidence rate is around 70 years old. Survival is usually determined by the stage and histology of the disease, and the prognosis of endometrial cancer varies greatly in different stages and histological types. The most common lesions (type I) are typically hormone-sensitive and in low-stage with good prognosis, while type II tumors have a high grade and are prone to relapse even in the early stages [2]. Recent large-scale genomic studies have shown that a large number of non-coding RNAs (such as microRNAs and long non-coding RNAs) are associated with the occurrence of gynecological diseases [3, 4].Circular RNAs (circRNAs) belongs to a new class of non-coding RNAs and are formed by a peculiar pre-mRNA with a covalently closed continuous loop. Due to its structures, circRNA are resistant to degradation by exonuclease activity and more stable than linear RNAs. circRNAs have been implicated in microRNA (miRNA) sequestration, modulation of protein–protein interactions and regulation of mRNA transcription. Among them, the most striking function is acting as a miRNA sponge and regulate the expression of their downstream genes [5, 6]. MicroRNAs (miRNAs or miRs) are a class of non-coding RNA molecules that negatively regulate the translation of messenger (m) RNAs by interacting with complementary sites in the 3' untranslated region (UTR) [7]. Many miRNAs act as tumor regulator genes by directly targeting oncogenes or tumor suppressor genes [8]. CircRNAs were implicated not only involved in cellular physiological functions, but also in various human pathologies including cancer. It was found that circRNAs are aberrantly modulated in human cancer tissues. Furthermore, research is currently focusing on understanding the possible implications of circRNAs in diagnostics, prognosis prediction, and eventually therapeutic intervention in human cancer [3]. CircRNA hsa_circ_0023404 (chr11: 71668272–71671937) is derived from mRNA of ring finger protein 121 (RNF121, NM_018320). Increasing evidence supported that hsa_circ_0023404 play a critical role in cancer progression. For example, it showed that hsa_circ_0023404 can promote the proliferation, migration and invasion of non-small cell lung cancer (NSCLC) by regulating miR-217/ZEB1 axis [9]. Compared with the miR-con group, overexpression of miR-217 reduced the relative luciferase activity of the pGL3-circ_0023404-WT reporter vector in vitro and strongly validated that hsa_circ_0023404 interacted with and sponged miR-217. Other studies demonstrated that hsa_circ_0023404 was involved in cervical cancer by regulating miR-5047 and miR-136/TFCP2 /YAP pathway [10]. Recently, mounting evidence showed that miR‑217 can regulated tumor biology depending on the cell type [11, 12]. It was observed that the WNT, mitogen‑activated protein kinase (MAPK), and PI3K/AKT signaling pathways were important molecular targets of miR-217 in different cancers and contributed to cancer progression [13]. MAPK1 was identified as a novel miR‑217 target and was a key component of RAS/RAF/MAPK pathway which was found activated in about 30% of all human cancer tissues. Activated MAPK1 translocated to the nucleus and catalyzed the phosphorylation of numerous nuclear transcription factors such as ETS (erythroblast transformation specific), ELK-1 (ETS Like-1 protein), c-Fos and activated variety target genes such as ErbB, VEGF, etc., which contributes to the progression of tumors [14, 15]. Inhibition of MAPK1 was shown to block tumor growth and metastasis in prostate cancer [16]. It was found that miR‑217 can suppress tumorigenicity of colorectal cancer targeting MAPK1 [17]. These indicated that hsa_circ_0023404 and its target miR-217/MARK1 axis play a critical role in cancer progression such as non-small cell lung cancer and cervical cancer, but the role of hsa_circ_0023404/miR-217/MARK1 involved in endometrial cancer was not investigated yet. In this study, we investigated the role of hsa_circ_0023404 in promoting endometrial cancer cells associated with miR-217/MAPK1 axis.
Human endometrial endothelial cell (HEEC) and human endometrial cancer cells (Ishikawa, RL95-2 and KLE) were purchased from the American Type Culture Collection (ATCC, USA) or National Infrastructure of Cell Line Resource (Beijing, China). Cells were incubated in DMEM (Gibco, USA) contained 10% fetal bovine serum (FBS; PAN biotech, Germany) and 1% penicillin/streptomycin (Solarbio, China) at 37 °C and 5% CO2.
Total RNA was extracted using Neurozol reagent (Macherey–Nagel, Germany) and cDNA was generated using reverse transcription reagent kit (PROMEGA, USA). Real-time PCR was performed using SYBR Green PCR kit (TaKara, China). U6 and GAPDH are internal controls. The qPCR analysis was then performed on an ABI 7500 Real-time PCR System (Applied Biosystems, Thermo Fisher Scientific, USA) according to the instructions supplied by the manufacturer. The relative expression levels of the genes were calculated by comparing to U6 or GAPDH using 2 − ΔΔCT method. The primers were used as follows: miR-217 FORWARD: CGCGTACTGCATCAGGAACTG; miR-217 REVERSE: AGTGCAGGGTCCGAGGTATT; miR-217-5p RT (anti-miR-217) Primer: GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACTCCAAT; U6 FORWARD: CTCGCTTCGGCAGCACA; U6 REVERSE: AACGCTTCACGAATTTGCGT; circ_0023404 FORWARD: ACCGTGGCCATGAAGCTATG; circ_0023404REVERSE: GGTCACCATATTGTAGGAGCGT; GAPDH FORWARD: AGAAGGCTGGGGCTCATTTG; GAPDH REVERSE: AGGGGCCATCCACAGTCTTC; MAPK1 FORWARD: CAGTTCTTGACCCCTGGTCC; MAPK1 REVERSE: GTACATACTGCCGCAGGTCA.
si-NC (negative control) sequence: UUCUCCGAACGUGUCACGUTT, si-circRNA (si-hsa_circ_0023404 #1–3; #3 sequence: GGUUCCUGCUAAUCUAUAATT, miR-217 or anti-miR-217 were synthesized by GenePharma (Shanghai, China). They were transfected in Ishikawa cells using Lipofectamine 3000 Reagent (Life Technologies, USA) and then culture in at 37 °C and 5% CO2 for 48–72 h.
Ishikawa cells were plated at 2 × 10E3 cells/well in 96-well plates and grown in medium containing 10% FBS for 24 h. After transfection with siRNA, 10 μl of cell count kit-8 (CCK-8, CK04, Dojindo, Japan) was added into each well and cells were incubated for 2 h in a 5% CO2 incubator at 37 °C. The absorbance of each well at 450 nm was read in GloMax™ 96 MICROPLATE (Promega, USA).
Ishikawa cells were transfected with siRNA for 48 h and trypsinized and dispensed into 6-well plates with a density of 800 cells/well. When the number of cells in a colony is more than 50, 10% formaldehyde was employed to fix colonies for 10 min and 0.5% crystal violet was adopted to stain colonies for 5 min. Images were photographed and the number of colonies was calculated by ImageJ.
For migration assay, transfected Ishikawa cells (1 × 10E5 cells) were suspended in 200 ul serum-free medium and then seeded on the top chamber. Medium contained 10% FBS was added into the lower chamber. After 24 h of incubation, cells on the lower surface of the lower chamber were fixed with 4% PFA and stained with 0.1% crystal. Cells were counted from five randomly selected microscopic fields. For invasion assay, Transwell inserts (Fisher Scientific, USA) were coated with Matrigel (BD, USA). After 24 h incubation, cells on the upper surface of the Transwell membrane were gently removed, and cells on the lower surface of the Transwell membrane were fixed and stained with crystal violet, counted from five randomly selected microscopic fields.
Cells were collected and lysed with RIPA buffer (Beyotime, China). Equal amount of protein was separated on SDS-PAGE and transferred to PVDF (Millipore, USA). Then, the membranes were incubated with the primary antibodies anti-MARK1 (Cat:125403, Novopro, China) and anti-actin (Sigma, USA). ECL substrates were used to visualize protein bands (Millipore, USA).
All experiments were replicated thrice and all data were expressed as mean ± standard deviation (SD). The software GraphPad 8.0 were used to carry out all statistical analyzes. Student's t-test and one-way ANOVA followed by Bonferroni 's post hoc test were utilized to analyze 2 or multiple groups, respectively. * means p < 0.05; ** means p < 0.01; *** means p < 0.001.
To examine the role of hsa_circ_0023404 and its target miR-217/MARK1 axis in endometrial cancer cell lines, The RT-qPCR was applied to determine the level of hsa_circ_0023404, miR-217 and MARK1 mRNA in human endometrial endothelial cell (HEEC) and three human endometrial cancer cells (RL95-2, KLE and Ishikawa). It was shown that hsa_circ_0023404 (Fig. 1A) and MARK1 (Fig. 1C) were upregulated in RL95-2, KLE and Ishikawa cell lines compared to HEEC. On the contrary, miR-217 (Fig. 1B) was downregulated in RL95-2, KLE and Ishikawa cell lines compared to HEEC. Among the three cell lines, the results in Ishikawa cell were most strikingly and we employed the Ishikawa cells in further study.
Since the level of hsa_circ_0023404 is upregulated in endometrial cancer cells, we investigate its biological role in endometrial cancer by knockdown of hsa_circ_0023404 with si-circ_0023404 in Ishikawa cells. It showed all three siRNAs targeted to hsa_circ_0023404 significantly reduced mRNA expression of the hsa_circ_002340 in Ishikawa cells compared to control siRNA (si-NC) analyzed by RT-qPCR (Fig. 1D). Among all siRNA, the siRNA#3 had the highest efficiency and was employed for subsequent experiments. We next examined the effect of si-circ_0023404 on miR-217 expression in Ishikawa cells and it showed the si-circ_0023404 #3 reduced the level of circ_ 0023404(Fig. 1E) while the level of miR-217 was upregulated (Fig. 1F). This is consistent with that the circ_0023404 inhibited the miR-217 expression acting as a miRNA sponge. Our data further showed that downregulation of hsa_circ_0023404 markedly decreased the proliferation of the Ishikawa cells detected by CCK-8 assay (Fig. 1G). Down-regulation of hsa_circ_0023404 also markedly decreased the capacity of colony formation of in Ishikawa cells compared to control (Fig. 1H, I). Consistently, knockdown of hsa_circ_0023404 inhibited the cell migration and invasion in Ishikawa cells analyzed by Transwell migrations assay (Fig. 1J, K) and Transwell invasion assay (Fig. 1J, L). These data indicated that hsa_circ_0023404 promoted cell proliferation, migration and invasion in endometrial cancer cells.
To investigate the role of miR-217 in endometrial cancer cells, Ishikawa cells were transfected with mimic NC and miR-217 mimic. The expression of transfected miR-217 mimic was confirmed by RT-qPCR (Fig. 2A). CCK-8 assay demonstrated that miR-217 mimic markedly decreased the proliferation of the Ishikawa cells (Fig. 2B). miR-217 mimic also markedly decreased the capacity of colony formation of in Ishikawa cells compared to mimic NC (Fig. 2C, D). In parallel, it showed that miR-217 reduced the cell migration and invasion in Ishikawa cells analyzed by Transwell migrations assay (Fig. 2E, F) and Transwell invasion assay (Fig. 2E, G). MARK1 is the target of miR-217 and WB analysis indicated that the miR-217 mimic transfection decreased the expression of MARK1 protein in Ishikawa cells (Fig. 2H, I). These data indicated that miR-217 played a critical role in inhibiting cell proliferation, migration and invasion in endometrial cancer cells and MARK1 protein is one downstream target of miR-217.
Studies indicated that miR-217 was one sponge target of hsa_circ_0023404 and we examined their interaction by co-transfection with si-circ_0023404 and anti-miR-217 (miR-217 inhibitor). Co-transfection showed that si-circ_0023404 attenuated the expression level of hsa_circ_0023404 while anti-miR-217 increased hsa_circ_0023404 (Fig. 3A); si-circ_0023404 increased the expression level of miR-217 while anti-miR-217 blocked the increased miR-217 (Fig. 3B). CCK-8 assay showed that downregulation of hsa_circ_0023404 decreased the proliferation of the Ishikawa cells but anti-miR-217 reversed the decrease (Fig. 3C, D). Transwell migrations and invasion assay (Fig. 3E, F) also indicated that anti-miR-217 blocked the promoted migrations and invasion of Ishikawa cells mediated by si-circ_0023404. Summary, these data showed that anti-miR-217 can block the promoted proliferation, migrations and invasion of Ishikawa cells by si-circ_0023404, consistent with that hsa_circ_0023404 acts as sponge of miR-217.
MARK1 is a potential target of miR-217 and our data showed that anti-miR-217 increased the MARK1 protein level which was blocked by co-transfection with si_circ_0023404 and anti-miR-217, supporting that MARK1 is downstream of the hsa_circ_0023404/mirR217 axis (Fig. 4A). Western blot showed that si-MARK1 can downregulate the induced MARK1 by co-transfection with si_circ_0023404 and anti-miR-217. si-MARK1 restored the inhibited cell proliferation of the co-transfected Ishikawa cells with si- circ_0023404 and anti-miR-217 analyzed by CCK8 assay (Fig. 4E) and colony formation assay (Fig. 4F, G). In parallel, it showed that si-MARK1 restored the inhibition in migration (Fig. 5A, B) and invasion (Fig. 5A, C) of the co-transfected Ishikawa cells with si-circ_0023404 and anti-miR-217 analyzed by Transwell migrations and invasion assay. These data supported that the MARK1 is downstream of hsa- circ_0023404/miR-217 axis and MARK1 knockdown by si-MARK1 can block the promotion of cancer biology mediated by si-circ_0023404/miR-217 axis.
Plenty of studies supported that circular RNA hsa_circ_0023404 is associated with tumorigenesis. In this study, we found that hsa_circ_0023404 was upregulated with decreased miR-217 in endometrial cancer cell lines. Knockdown of hsa_circ_0023404 lead to increased miR-217 and inhibited the proliferation and metastasis of endometrial cancer cells. Anti-miR-217 can reverse the imbibition by si-circ_0023404. These data indicated that hsa_circ_0023404 promoted the proliferation, migration and invasion in endometrial cancer cells by sponging miR-217. In further study, knockdown of MARK1 blocked the promotion of cancer biology mediated by si-circ_0023404/miR-217 axis, supporting that MARK1 is the target of miR-217 and involved in circ_0023404/miR-217-mediated endometrial cancer biology. In human cancer, circRNAs were implicated in the control of oncogenic activities, such as tumor cell proliferation, epithelial–mesenchymal transition, invasion, metastasis and chemoresistance. The most widely described mechanism of action of circRNAs is their ability to act as competing endogenous RNAs (ceRNAs) for miRNAs, lncRNAs and mRNAs, thus impacting along their axis [2, 18, 19]. Several studies revealed that circRNA hsa_circ_0023404 play critical role in tumorigenesis. For example, it enhances cervical cancer metastasis and chemoresistance through VEGFA and autophagy signaling by sponging miR-5047 [10]. hsa_circ_0023404 is also involved in cervical cancer progression through/miR-136/TFCP2/YAP axis. hsa_circ_0023404 promoted TFCP2 expression via inhibiting miR-136, leading to activation of YAP signaling pathway [20]. hsa_circ_0023404 was shown to interact with miR-217/ZEB1 axis to contribute to the growth, migration and invasion of NSCLC cells [9]. This study provided the strong evidence that hsa_circ_0023404 promoted the proliferation, migration and invasion in endometrial cancer cells through regulating miR-217/MARK1 axis. Dysregulated miRNA expression was involved in malignancies and miRNAs may serve as tumor suppressor or oncogene to participate in human cancer progression. As a miRNA, miR‑217 is closely linked to tumor progression and poor prognosis [21, 22]. Previous studies have reported that miR‑217 bound to its target mRNA to inhibit the formation and progression of tumors, including gastric cancer [22]. Bioinformatics identified MARK1 protein is the target of miR‑217 in cancer cells. There are two binding sequences for miR‑217 in MAPK1 3'UTRs which was confirmed by the luciferase activity assay [23]. Consistently, previous study showed that downregulated MAPK1 by miR-217 facilitated the metastasis and EMT process of HCC cells, indicating that miR-217 suppressed HCC via negatively modulating MAPK1 expression[24]. Mutiple evidence demonstrated that miR217-MAPK axis was involved in tumorgenesis. For example, it was uncovered that the apoptosis-inducing potential of miR-217-5p can induced apoptosis via blocking multiple target genes PRKCI, BAG3, ITGAV and MAPK1 in colorectal cancer cells [25]. It also is reported that circMAN2B2 acted as an onco-miRNA in HCC by sponging miR-217 to promote MAPK1 expression [26]. The MAPK pathway is effectively involved in the regulation of cancer cell proliferation, invasion and survival by activating target genes such as transcriptional factor ELK1, C-Fos and the ErbB, VEGF, which contributes to the progression of tumors [14, 15]. Previous studies have confirmed that increased MAPK1 expression could function as tumor promoter in human hepatocellular carcinoma (HCC) [27, 28], ovarian cancer [29] and cervical cancer [30]. It also showed that lncRNA RHPN1-AS1 activated ERK/MAPK pathway and promoted cell proliferation, migration and invasion of endometrial cancer [31]. Another study demonstrated that activation of MAPK and AKT by Type II transmembrane serine proteases 4TMPRSS4 were associated with the progression of endometrial cancer [32]. These data provided the evidence that MAPK can be regulated by non-coding RNA (ncRNA) including lnRNA and cirRNA, etc. In this study, our data showed that MARK1 is downstream target of hsa_circ_0023404/miR-217 axis and involved in the endometrial cancer progression. With the advancement of RNA sequencing technology and the rapid development of bioinformatics, a large number of circRNAs were discovered widely involved in a variety of cancer-related pathogenesis and drug resistance and in the diagnostic and prognostic biomarker and the therapeutic target in human cancer [33]. The powerful functions and unique properties of circRNAs have made them the focus of scientific and clinical research. Due to the structure of covalently closed continuous loop, circRNAs are relatively stable and exist stably at high levels in body fluids, including plasma, serum, exosomes and urine, etc. Therefore, circRNA potentially service as the liquid biopsy-based novel biomarkers for monitoring the development and progression of cancer including lung cancer [34], endometrial Cancer [35], bladder cancer [36], prostate cancer [37], etc. Downregulated circBNC2 and higher circSETDB1 levels were identified in patients with ovarian cancer [38]. hsa_circ_ 0109046 and hsa_circ_0002577 were found increase in the serum of patients with endometrial cancer [39]. The unique cellular stability and capacity of circRNA to sponge miRNA and protein may place circRNA as a promising vehicle for the delivery of cancer therapeutics [40]. In the current study, we demonstrated the molecular mechanism of hsa_circ_0023404/miR-217/MAPK involved in the endometrial cancer progression. However, there are several limitation. First, it was limit to draw the conclusion completely only dependent in vitro experiments, therefore, we would further carry out in vivo experiments on hsa_circ_0023404/miR-217/MAPK axis involved in the endometrial cancer. Second, the application of hsa_circ_0023404 on liquid biopsy was not perform on patients with endometrial cancer. We will collect the patients to investigate the level of hsa_circ_0023404/miR-217/MAPK in their blood sample and examine their potential as novel biomarker for endometrial cancer.
In this study, our data demonstrated that hsa_circ_0023404 exerts a tumor-promoting role in endometrial cancer by regulating miR-217/MARK1 axis. The hsa_circ_0023404 act as sponge for and inhibit miR-217 which inhibit endometrial cancer cell growth and metastasis. MARK1 is downstream target of miR217 and the induced MARK1 by hsa_circ_0023404 through miR217 inhibition contribute to the endometrial cancer progression (Fig. 5D). Targeting or knockdown of hsa_circ_0023404 by short hairpin RNA (shRNA) or CRISPR technique would be a potential therapeutic approach for endometrial cancer and will be investigated in the future.
Additional file 1: Figure S1. Uncropped Western blot images. |
|
PMC9648002 | Chunlin Xiao,Zhong Chen,Wanqiu Chen,Cory Padilla,Michael Colgan,Wenjun Wu,Li-Tai Fang,Tiantian Liu,Yibin Yang,Valerie Schneider,Charles Wang,Wenming Xiao | Personalized genome assembly for accurate cancer somatic mutation discovery using tumor-normal paired reference samples | 09-11-2022 | Background The use of a personalized haplotype-specific genome assembly, rather than an unrelated, mosaic genome like GRCh38, as a reference for detecting the full spectrum of somatic events from cancers has long been advocated but has never been explored in tumor-normal paired samples. Here, we provide the first demonstrated use of de novo assembled personalized genome as a reference for cancer mutation detection and quantifying the effects of the reference genomes on the accuracy of somatic mutation detection. Results We generate de novo assemblies of the first tumor-normal paired genomes, both nuclear and mitochondrial, derived from the same individual with triple negative breast cancer. The personalized genome was chromosomal scale, haplotype phased, and annotated. We demonstrate that it provides individual specific haplotypes for complex regions and medically relevant genes. We illustrate that the personalized genome reference not only improves read alignments for both short-read and long-read sequencing data but also ameliorates the detection accuracy of somatic SNVs and SVs. We identify the equivalent somatic mutation calls between two genome references and uncover novel somatic mutations only when personalized genome assembly is used as a reference. Conclusions Our findings demonstrate that use of a personalized genome with individual-specific haplotypes is essential for accurate detection of the full spectrum of somatic mutations in the paired tumor-normal samples. The unique resource and methodology established in this study will be beneficial to the development of precision oncology medicine not only for breast cancer, but also for other cancers. Supplementary Information The online version contains supplementary material available at 10.1186/s13059-022-02803-x. | Personalized genome assembly for accurate cancer somatic mutation discovery using tumor-normal paired reference samples
The use of a personalized haplotype-specific genome assembly, rather than an unrelated, mosaic genome like GRCh38, as a reference for detecting the full spectrum of somatic events from cancers has long been advocated but has never been explored in tumor-normal paired samples. Here, we provide the first demonstrated use of de novo assembled personalized genome as a reference for cancer mutation detection and quantifying the effects of the reference genomes on the accuracy of somatic mutation detection.
We generate de novo assemblies of the first tumor-normal paired genomes, both nuclear and mitochondrial, derived from the same individual with triple negative breast cancer. The personalized genome was chromosomal scale, haplotype phased, and annotated. We demonstrate that it provides individual specific haplotypes for complex regions and medically relevant genes. We illustrate that the personalized genome reference not only improves read alignments for both short-read and long-read sequencing data but also ameliorates the detection accuracy of somatic SNVs and SVs. We identify the equivalent somatic mutation calls between two genome references and uncover novel somatic mutations only when personalized genome assembly is used as a reference.
Our findings demonstrate that use of a personalized genome with individual-specific haplotypes is essential for accurate detection of the full spectrum of somatic mutations in the paired tumor-normal samples. The unique resource and methodology established in this study will be beneficial to the development of precision oncology medicine not only for breast cancer, but also for other cancers.
The online version contains supplementary material available at 10.1186/s13059-022-02803-x.
Accurately detecting somatic mutations and subsequently understanding genomic instability in cancer are critical for precision cancer therapies [1–4]. Many genomics studies, including tremendous efforts from the well-known TCGA and ICGC consortia, have greatly improved our understanding of genomic instability of cancer and cancer biology in general [3, 4]. Most recently, paired tumor-normal reference samples [5–7] and reference call sets were established by the Sequencing Quality Control-2 (SEQC-2) consortium for benchmarking somatic mutation detections using different sequencing platforms and bioinformatic analysis methods [5, 6]. These studies have provided a critically important resource, not just to cancer biology in general, but for assessing the accuracy and reproducibility of somatic mutation detection in cancer diagnostics, designing personalized cancer immunotherapies, and for analyzing potential off-target effects that may interfere with successful therapies based on gene editing. To date, discovering somatic events and defining high-confidence reference somatic call sets rely mainly on a standard human reference assembly (such as GRCh38) as a benchmark for sequence analysis. However, GRCh38 has its own limitations. Despite its high quality, it remains incomplete due to some unresolved assembly issues and persistent gaps, including those at centromeres, telomeres, and heterochromatic regions [8–11]. Incorrect or missing sequences in the GRCh38 reference assembly may lead to failed or spurious read mapping and unreliable subsequent analysis results, namely reference bias [12]. Moreover, the human reference assembly was constructed based on DNAs derived from multiple individuals, though approximately 70% of the GRCh38 sequences were contributed by a single African-European admixed male (RP11) [9]. Such mosaic haplotype representation in the reference assembly may complicate the identification of somatic variants from cancer samples. Therefore, use of a de novo assembly of a personalized genome rather than the standard reference assembly for confident cancer mutation discovery has been advocated [8, 10, 13], because there is more to be learned from direct comparison of the tumor genome to the normal genome from which it is derived, than to an unrelated, random, mosaic genome like GRCh38. Recent advancements in DNA sequencing technologies provide an extraordinary opportunity to generate a high-quality de novo assembly for an individual genome at affordable cost. Specifically, the breakthrough of long-range DNA sequencing from next-generation sequencing technologies now makes it possible to accurately assemble individual genomes to near completion, as has been done for several samples, including HX1 [14], AK1 [15], NA12878 [16], CHM13 [17], and HG002 [18]. These studies demonstrated the utilities of these recent advancements to genome assembly methods and subsequent germline variant detection. However, there has been no systematic investigation of the use of personalized genomes as references for somatic mutation detection, particularly in paired tumor-normal samples. The HCC1395 breast cancer cell line and a matched B lymphocyte cell line HCC1395BL [19], derived from the same individual, are one of the most important tumor-normal models for triple negative breast cancers (TNBC), and represent reference samples that have been characterized extensively by previous studies through the SEQC2 consortium [5–7]. Here we combined multiple sequencing technologies, including Illumina short reads, 10X Genomics linked reads, PacBio long reads, and Hi-C (high-throughput chromosome conformation capture) reads, to reconstruct what, to our knowledge, is the first direct comparison of paired tumor-normal genomes [10]. This study enables us to assess the qualities of the de novo assembled personalized genome, to evaluate the performance of somatic mutation detection comprehensively with respect to the underlying reference assemblies being used, and to interrogate the complete spectrum of genomic alterations more accurately, using a personal genome as the reference.
Sequencing data from five different platforms were used to assemble genome for the normal reference sample (HCC1395BL B Lymphocyte cell line), and data from three platforms were used for the tumor reference sample (HCC1395 breast cancer cell line from the same donor) (Fig. 1, top panel). Using data from multiple sequencing technologies, including short reads, linked reads, and long reads (Additional file 1: Table S1), we built a workflow to generate a de novo assembled personal genome (Fig. 1, middle panel), known as HCC1395BL_v1.0. We then used this assembled genome and GRCh38 as references for read mapping and somatic variant analyses (Fig. 1, bottom panel). In this workflow, we first generated two initial assemblies for HCC1395BL, using canu [20] with PacBio long reads and Supernova [21] with 10X Genomics linked reads, respectively. In contrast to the PacBio canu assembly, the Supernova assembly contained many small contigs (< 10 kb) (Additional file 1: Table S2). Although the N50s and the largest scaffold of the Supernova assembly were much larger than the PacBio contig assembly (Additional file 2: Figs. S1 and S2), the latter was more complete as measured via Benchmarking Universal Single-Copy Orthologue (BUSCO) genes (Additional file 2: Fig. S3). Additionally, a greater number of complete RefSeq protein-coding genes mapped to the PacBio assembly, and more base pairs from this assembly could be mapped to the GRCh38 reference (Additional file 1: Table S2). Taken together, these findings indicated that the overall quality of the PacBio canu assembly was higher than that of the Supernova assembly, particularly when gene content and completeness were the primary concerns. We selected the PacBio contig assembly of HCC1395BL for further scaffolding. In general, two steps of scaffolding, using 10X Genomics linked reads with ARCS followed by Hi-C reads with SALSA (PacBio_canu + ARCS + SALSA), produced a better scaffolded assembly than using one-step scaffolding only (either PacBio_canu + ARCS or PacBio_canu + SALSA). The final scaffold assembly (hereafter referred to as HCC1395BL_v1.0) was the one with the highest Top50 (see “Methods”) and scaffold N50 values, the largest scaffold size, and the greatest numbers of mapped complete BUSCOs and RefSeq transcripts (Table 1). HCC1395BL_v1.0 consisted of 1645 scaffolds totaling 2.9 Gb, of which 2.69 Gb (92.62%) were from Top50 scaffolds, with a scaffold N50 size of 69.97 Mb, in comparison to scaffold N50s of 67.79 Mb for GRCh38 [9] and 44.84 Mb for AK1, a recent assembly from a diploid sample [15], respectively. Both HCC1395BL_v1.0 and AK1 had a very similar assembly size (2.90 Gb), but HCC1395BL_v1.0 had fewer scaffolds (1,645 vs. 2,832), a smaller L50 (14 vs. 21), and a much greater Top50 (2.69 Gb vs. 2.26 Gb), N50 (69.97 Mb vs. 44.84 Mb), and the largest scaffold size (181.21Mb vs. 113.92 Mb). Moreover, the HCC1395BL_v1.0 assembly contained more complete RefSeq NM (protein-coding) transcripts (49,613 vs. 49,432) and RefSeq NR (non-protein-coding) transcripts (15,227 vs. 15,089). Consistency analysis (see “Methods”) with the GRCh38 primary assembly (alternate loci excluded) showed that five chromosomes (chr4, chr8, chr14, chr18, and chr20) were almost completely covered by single scaffolds. The largest HCC1395BL_v1.0 scaffold (Scaffold_1 181.21Mb) covered more than 95% of GRCh38 chromosome 4. Four chromosomes (chr2, chr3, chr12, and chr19) were broken only in centromeric regions. Several other chromosomal arms (chr1p, chr5p, chr6q, chr9p, chr10p, chr21q, and chrXq) were also covered by single HCC1395BL_v1.0 scaffolds (Fig. 2). Phasing analysis showed that 3.13 out of 3.17 million heterozygous sites were considered phased, and 6368 phased blocks accounted for 2.42 Gb of HCC1395BL_v1.0. The longest phased block was 6.37 Mb (Additional file 1: Table S3). Approximately 15-fold coverage of Nanopore long reads was used to further extend phasing to 2.54 Gb. The total number of phased blocks was subsequently decreased to 3204 from 6368 blocks, and the longest phased block was greatly improved, increasing from 6.37 to 20.45 Mb (Additional file 1: Table S3). With the phased assemblies (haplotype1 and haplotype2) for the HCC1395BL cell line, we were able to call 4,115,622 germline SNVs in diploid regions of autosomal chromosomes using dipcall [22]. For a comparison, we also generated de novo assemblies for the HCC1395 cancer cell line, using canu with PacBio long reads and Supernova with 10X Genomics’ linked reads, respectively. The resulting HCC1395 assembly was more fragmented than the HCC1395BL assembly (Additional file 2: Figs. S1 and S2; Additional file 1: Table S2), as demonstrated by the cytogenetic analysis of HCC1395 and HCC1395BL cell lines [6]. The N50s of HCC1395 assemblies were significantly smaller than that of the corresponding HCC1395BL assemblies (Additional file 2: Fig. S1; Additional file 1: Table S2). Moreover, we identified smaller numbers of BUSCO genes (Additional file 2: Fig. S3), RefSeq protein-coding and non-protein-coding transcripts (Additional file 1: Table S2) on both the PacBio canu, and 10X Genomic Supernova assemblies of HCC1395 tumor cell line, as compared to the corresponding assemblies from the HCC1395BL normal cell line. Similarly, the PacBio assembly of the HCC1395 cell line was also shown to be of better quality in terms of gene content and completeness, as compared with that of 10X Genomics’ Supernova assembly (Additional file 2: Fig. S3; Additional file 1: Table S2). A previous study [23] showed that mitochondrial genome assemblies in recently published de novo assemblies have either been absent or highly fragmented. In this study, we confirmed that we completely assembled the mitochondrial genome into a single contig using PacBio data in both the normal and tumor cell lines. Direct comparison of these two mitochondrial genomes (see “Methods”) revealed two nonsynonymous somatic mutations (T4813C and C4938A) in the MT-ND2 gene, and one nonsynonymous somatic mutation (G14249A) in the MT-ND6 gene (Additional file 2: Fig. S7).
We aligned an NCBI RefSeq transcript set (excluding all pseudogenes and genes from chromosome Y) to HCC1395BL_v1.0. In total, 19,303 of 19,325 (99.89%) RefSeq protein-coding genes could be mapped onto HCC1395BL_v1.0 successfully, with minimum 95% alignment identity and 50% alignment coverage, while 19,164 of 19,303 (99.27%) of these genes aligned with at least 95% coverage (Fig. 3A; Additional file 1: Table S4). Among RefSeq non-protein-coding genes, 10,049 of 10,061 (99.88%) could be aligned to HCC1395BL_v1.0 successfully with minimum 95% identity and 50% coverage; 9958 of 10,049 (99.09%) of those genes were covered at more than 95% in length (Fig. 3A; Additional file 1: Table S4). We next compared HLA gene family coverage in GRCh38 and HCC1395BL_v1.0, as the known variability in this region makes it likely that haplotypes found in this sample may differ from the haplotype represented in the chromosomes of the traditional reference genome GRCh38. HLA genes are located on the 6p region of chromosome 6, and previous cytogenetic analysis showed that this region from the HCC1395BL cell line was essentially haploid [6]. From the RefSeq gene set, 19 HLA protein-coding genes (25 protein-coding transcripts) are annotated on chromosome 6 of the GRCh38 primary assembly. We successfully identified all the HLA genes and corresponding transcripts in HCC1395BL_v1.0 and found that they are located on a single scaffold (Scaffold_30) aligning at minimum identity of 95% and the minimum alignment coverage of 95% with one exception, HLA-DQA1 (NM_002122.3) gene, aligned at 100% coverage to HCC1395BL_v1.0, but only 92.95% identity. No other mapped location was found for the HLA-DQA1 gene on HCC1395BL_v1.0. Notably, the order of the HLA genes on this scaffold was identical to that on GRCh38 (Fig. 3B). However, the haplotype of HLA-DRB genes between HLA-DRA and HLA-DQA1 is extremely divergent between HCC1395BL_v1.0 and GRCh38 [24]. The haplotype of HLA-DRB in the GRCh38 primary assembly is represented by the HLA-DRB1 and HLA-DRB5 genes (human HLA-DR51 haplotype group [25]), but the HLA-DRB haplotype in HCC1395BL_v1.0 consists of the HLA-DRB1 and HLA-DRB4 genes (human HLA-DR53 haplotype group), which is similar to the HLA-DRB haplotypes represented on the GRCh38.p13 ALT_REF_LOCI_4 scaffold NT_167246.2 and the GRCh38.p13 ALT_REF_LOCI_7 scaffold NT_167249.2. This demonstrated that the de novo assembly and scaffolding of HCC1395BL_v1.0 performed well on hypervariable/complex regions, such as those harboring the HLA genes. We also evaluated other clinically relevant genes whose only representations in GRCh38 are on alternate locus scaffolds, which are included in the reference to capture population diversity. HCC1395BL_v1.0 included GSTT1 (Glutathione S-transferase theta 1) and the KIR2DL5A (killer cell immunoglobulin like receptor, which has two Ig domains and long cytoplasmic tail 5A); these two genes are not included in the haplotypes represented on the chromosomes of the GRCh38 primary assembly. GSTT1, a gene previously localized to Chromosome 22 of the GRCh37 primary assembly, is found only on the alternate locus scaffold NT_187633.1 in GRCh38. Likewise, for the KIR2DL5 gene, the haplotype represented on the GRCh37 Chromosome 19 unlocalized scaffold NT_113949.1 included KIR2DL5A, but in the GRCh38 assembly, it is found only in the alternate locus, such as scaffold NT_113949.2. The changes in the localizations of these genes from GRCh37 to GRCh38 present analysis challenges when switching between different versions of traditional reference genome. In addition, because these genes are represented only on alternate loci and patch scaffolds in GRCh38, and most existing tool chains do not handle those alternate locus scaffolds, they are consequently more difficult to study. Their exclusion from analysis thus presents a heightened risk for misinterpretation of results. In contrast, as the individual-specific haplotypes for these clinically relevant genes are represented in haplotypes of the personalized assembly, no special handling of alternate loci or patches would be needed to assess them in the tumor genome if HCC1395BL_v1.0 were to be used as reference as opposed to GRCh38.
We compared HCC1395BL_v1.0 and GRCh38 as references for read mappings. While the mapping rates of short reads to the GRCh38 primary assembly (alternate loci scaffolds excluded) and HCC1395BL_v1.0 were very similar for all 12 WGS replicates from 6 sequencing centers for both HCC1395BL and HCC1395 cell lines, we observed overall improved read placements on HCC1395BL_v1.0 as opposed to GRCh38, with some variability in the extent of improvements across these replicates, possibly due to the differences in library preparations and sequencing coverages for paired normal and tumor samples when sequencing was performed in each of the sequencing centers (Fig. 4). For instance, only slightly higher percentages of properly paired reads for both normal and tumor samples were mapped onto HCC1395BL_v1.0 (Fig. 4A), but mappings for non-properly paired reads were reduced by as much as 41.4% for HCC1395BL, and up to 38.2% for HCC1395 (Fig. 4B). Notably, the mismatches for the mapped reads were decreased up to 18.2% for HCC1395BL and up to 16.6% for HCC1395 (Fig. 4C). In addition, read alignments with soft-clipping (without SA tags) were decreased up to 11.7% for HCC1395BL and up to 11.6% for HCC1395 (Additional file 2: Fig. S4A), while read alignments with hard-clipping were down by as much as 32.0% for HCC1395BL and as much as 28.7% for HCC1395 (Additional file 2: Fig. S4B). Moreover, read alignments with split reads (SA tags) were also reduced by up to 31.9% for HCC1395BL and up to 28.8% for HCC1395 (Fig. 4D). Interestingly, we observed that the library insert size standard deviations were 2.76 smaller on average for HCC1395BL and 2.83 smaller on average for HCC1395 when the personal genome HCC1395BL_v1.0 was used as reference (Additional file 2: Fig. S4C), indicating that paired reads were placed more consistently on HCC1395BL_v1.0 than GRCh38. In addition, we observed that the standard deviations of read coverages in alignments were much smaller on HCC1395BL_v1.0 than GRCh38 (Fig. 4E), demonstrating that reads were placed more uniformly on HCC1395BL_v1.0. We evaluated PacBio long-read mapping onto GRCh38 and HCC1395BL_v1.0 references using minimap2. We observed that the mapping rates of PacBio long reads were slightly higher (1.65% for normal and 2.98% tumor sample) on HCC1395BL_v1.0 than on GRCh38 (Additional file 2: Fig. S4D). The mismatches were approximately 1% lower for both normal and tumor samples on HCC1395BL_v1.0 than on GRCh38 (Additional file 2: Fig. S4D). The non-primary alignments and supplementary alignments were significantly lower on HCC1395BL_v1.0 for both the normal (6.73%, 14.5%) and tumor samples (1.8%, 10.39%) (Additional file 2: Fig. S4D). Furthermore, we also observed that the standard deviations of read coverages in alignments were much smaller on HCC1395BL_v1.0 than on GRCh38 (Fig. 4F), indicating PacBio long reads were also placed more uniformly on HCC1395BL_v1.0 than GRCh38. Taken together, these ameliorations in both short-read and long-read mappings provide the important signals indicating that alignment-based somatic mutation discovery will be improved when a personal genome is used as the reference for a paired tumor sample.
Based on a previous study [26] and recent SEQC2 reports [5, 6], two commonly used somatic mutation callers, Strelka2 [27] and MuTect2 [28], were selected to generate reports of somatic SNVs and small indels with the same settings based on the same set of Illumina short-read data using HCC1395BL_v1.0 and GRCh38 (alternate loci excluded) as reference genomes, respectively. We initially analyzed a pair of pooled sequencing data (FDN123 as normal and FDT123 as tumor, see “Methods”) from one sequencing center and found that more overlapping calls (1983 more somatic SNVs/indels) between Strelka2 and MuTect2 could be detected on HCC1395BL_v1.0 than GRCh38 (Additional file 1: Table S5). This trend was retained when we expanded to analyze all 12 paired WGS replicates from 6 sequencing centers [5, 6]. On average, 1689 more overlapping somatic SNVs (Fig. 5A) and 415 more overlapping somatic indels were seen on HCC1395BL_v1.0 than GRCh38 (Fig. 5B), due to the various improvements in short-read mappings on HCC1395BL_v1.0 reference as we have already demonstrated. Among 41,669 GRCh38-based somatic SNVs supported by both Strelka2 and MuTect2 callers (Additional file 1: Table S5), 40,768 SNVs (97.83%) were successfully mapped onto HCC1395BL_v1.0 with overlapping SNVs called by Strelka2/MuTect2 (Fig. 5C; Additional file 1: Table S6). An additional 682 SNVs (1.64%) were mapped on HCC1395BL_v1.0, but without overlapping Strelka2/MuTect2 calls, suggesting that these somatic SNVs might be questionable. Variant functional analysis using ANNOVAR [29] showed that 120 of these 682 SNVs were located within genes (Additional file 1: Table S6). Two hundred nineteen SNVs (0.53%) were considered as “not-mapped” on HCC1395BL_v1.0 due to the stringent mapping criteria we used (see “Methods”). Thus, inclusion of these questionable sites in mutation analysis would certainly cause misinterpretations. Moreover, 3995 SNVs (9.58%) were considered equivalent between GRCh38 and HCC1395BL_v1.0 (Additional file 1: Table S6), but with germline SNVs in their flanking sequences (Additional file 2: Fig. S5). For example, the same set of reads was found to align (with mapping quality 60) across corresponding intergenic SNV regions in HCC1395BL_v1.0 (scaffold_2:131886469-131886569 for SNV scaffold_2:131886519), and GRCh38 (chr1:177753949-177754049 for SNV chr1:177753999), but two additional homozygous germline SNVs were observed in flanking sequences in the latter (Additional file 2: Fig. S5A, B, G). Similar examples were found for an exonic SNV at scaffold_37: 17305121 or chr19:17555816 (causing amino acid change in gene COLGALT1) (Additional file 2: Fig. S5C, D) and an intronic SNV at scaffold_12:48083060 or chr10:114357477 (Additional file 2: Fig. S5E, F). Such discrepancies reflect the underlying genomic sequence differences between the personalized HCC1395BL_v1.0 and the common reference GRCh38, and illustrate the importance of using a personal genome for accurate somatic mutation discovery and subsequent analysis. For example, mismatches in allele-specific probes or primers would affect melting temperature and binding efficiency when they are used for validation of a SNP genotyping assay. Among the 43,285 somatic SNVs supported by both Strelka2 and MuTect2 on HCC1395BL_v1.0 (Additional file 1: Table S5), 2790 SNVs were identified that lacked equivalent GRCh38-based SNVs. Among them, 1017 sites were well-supported by more than 10 alternate allele reads with the percentage of alternate allele read coverage at least 50%. By co-locating these SNVs with RefSeq genes and transcripts mapped onto HCC1395BL_v1.0, 522 of 2790 SNVs were found within genes, while 177 of well-supported 1017 SNV subset were located in 71 gene regions. KEGG pathway enrichment analysis suggested some of these 71 genes were involved with important pathways (Fig. 5D). For example, GTF2H2 (general transcription factor IIH subunit 2) encodes the subunit of RNA polymerase II transcription initiation factor IIH, which is involved in both basal transcription and nucleotide excision repair (Additional file 1: Table S7). PTPN13 (protein tyrosine phosphatase non-receptor type 13) encodes a signaling molecule that belongs to the protein tyrosine phosphatase (PTP) family, which regulates a variety of cellular processes such as cell growth, differentiation, mitotic cycle, and oncogenic transformation (Additional file 1: Table S7). To demonstrate the validity of somatic mutations identified only on HCC1395BL_v1.0, we performed Sanger sequencing on a subset of SNVs in these 71 gene regions (177 SNVs). Eight of ten selected sites were confirmed as somatic SNVs (Additional file 2: Fig. S6; Additional file 1: Table S8), while one SNV site (scaffold_20:687304 with MAF=0.558) showed the mutation in both tumor and normal samples, and the other one (scaffold_19:2641776 with MAF=0.5) showed no point mutation (Additional file 1: Table S8). While a high validation rate (80%) was achieved by Sanger sequencing, the results may indicate that some artifacts might exist in novel SNVs identified only with HCC1395BL_v1.0.
Due to differences in underlying algorithms for predicting somatic SVs, reported SV events called by different tools can vary widely in terms of event numbers, event types, and event sizes [30, 31]. Thus, we initially limited our somatic SV analysis for short-read sequencing to data generated by one sequencing center (FD) only (see “Methods”). For short-read WGS sequencing data, we selected four somatic SV callers, including GRIDSS2/GRIPSS [32], Manta [33], Delly [34], and novoBreak [35] as representatives for different detection algorithms, to evaluate their relative performances with HCC1395BL_v1.0 reference as compared to GRCh38. Both GRIDSS2 and Manta use split-read, read-pair, and breakpoint assembly approaches for somatic SV detection, whereas Delly uses read-pair first and then split-read information for SV detection and refinement, and novoBreak uses local assembly of associated read pairs with tumor-specific k-mers for somatic SV identification. As expected, all callers predicted various numbers of somatic SVs on both HCC1395BL_v1.0 reference and GRCh38, but in general, all four callers reported more somatic calls on HCC1395BL_v1.0 reference, whether translocation calls (TRA) were included (Fig. 6A; Additional file 1: Table S9) or excluded (Additional file 2: Fig. S8A; Additional file 1: Table S9), as compared to GRCh38. The increases were observed in all SV types except DUP and INV calls by novoBreak (Additional file 2: Fig. S8B). We also observed that the total SV counts detected by both GRIDSS2 and Manta were higher than that by Delly and novoBreak, indicating that the callers with more sophisticated algorithms (such as GRIDSS2 and Manta that combined split-read, read-pair, and breakpoint assembly approaches) were likely more sensitive than relatively simpler callers (such as Delly and novoBreak) (Fig. 6A). Particularly, Delly reported far fewer DUP and TRA events than any of other callers, suggesting Delly may have lower sensitivity for detecting these two SV types (Additional file 2: Fig. S8B). Additionally, GRIDSS2 reported 2 insertions (scaffold_47:144105:INS:66bp and scaffold_49:9684521:INS:60bp called by GRIDSS/Manta/Delly, but not in the gene region), and Delly reported 3 insertions (scaffold_47:144105:INS:66bp and scaffold_49:9684521:INS:60bp called by GRIDSS/Manta/Delly, and scaffold_129:31804:INS:78bp called by Manta/Delly, not located in the gene region) on HCC1395BL_v1.0, but none on GRCh38, while novoBreak did not report any insertion event. Manta detected 20 insertions on HCC1395BL_v1.0 (one insertion scaffold_1:105582094:INS:52bp overlapping with a DUP scaffold_1:105582094-105582146 by GRIDSS/Manta, intronic region in LIN5 gene), but only 10 insertions on GRCh38 (Additional file 2: Fig. S8B), suggesting that the personalized HCC1395BL_v1.0 as reference may have slightly better sensitivity for somatic insertion detection with short-read data for this pair of samples due to the improvements in personalized genome reference and subsequent read mapping on HCC1395BL_v1.0. We used a consensus approach so as to define a high-quality somatic SV callset, comprised only of somatic SV calls detected by two or more callers (see “Methods”). With this consensus callset, all four callers still reported more somatic SVs on HCC1395BL_v1.0 reference (Fig. 6B; Additional file 1: Table S9). Breaking down the consensus callset by SV types, 28 (7.25%) more deletions (DEL), 3 (7.89%) more inversions (INV), and 21 (9.29%) more translocations (TRA) were seen on HCC1395BL_v1.0 as compared to GRCh38, while the SV counts for the combination of DUP and INS were largely the same (Additional file 2: Fig. S8C). However, when we mapped those GRCh38-based SVs supported by two or more callers (TRA excluded) onto the HCC1395BL_v1.0 reference, 617 of 646 (95%) SVs (TRA excluded) were localized with HCC1395BL_v1.0-based SVs, but 18 SVs (3%), including 10 DEL, 7 DUP, and 1 INV, were considered “unmapped,” while 11 SVs (2%), including 7 DEL and 4 DUP, were mapped but without matching SVs in the mapped locations on HCC1395BL_v1.0 reference (Fig. 6C). Inclusion of these 29 SVs (18 unmapped or 11 mapped-without-matching-SVs on HCC1395BL_v1.0) may cause misinterpretation in downstream analysis. Furthermore, we identified 59 HCC1395BL_v1.0-based SVs supported by two or more callers (including 41 DEL, 14 DUP/INS, and 4 INV) that lacked GRCh38-based SVs in corresponding locations on HCC1395BL_v1.0 with our current mapping criteria (Fig. 6D). By collocating these 59 SVs with RefSeq genes mapped onto HCC1395BL_v1.0, we found 17 SVs (11 DEL, 3 INV, and 3 DUP/INS) overlapped with 17 gene regions (including 7 exon-overlapping SVs). KEGG pathway enrichment analysis suggested that some of these 17 genes were involved in the pathways that may be related to tumor development, including TGF-beta signaling, cellular senescence, and Hippo signaling pathway (Fig. 6E). Manual inspection of read alignments in IGV revealed that 9 out of 11 somatic deletions discovered in short-read sequencing data could be confirmed with in-read deletions from PacBio long reads generated from the tumor cell line, but not from the normal cell line (Additional file 1: Table S11). For instance, a 311 base pair SINE/Alu heterozygous deletion (scaffold_17:32976348-32976659), which overlaps with CCDC91 (coiled-coil domain containing 91), a gene enabling identical protein binding activity, was detected with the HCC1395BL_v1.0 reference, but not with GRCh38 when using the same set of reads, as it lacks this Alu sequence in GRCh38 (Additional file 2: Fig. S9A). A 57 base pair homozygous deletion (scaffold_6:44613899-44613956) overlapping with MED12L (mediator complex subunit 12L), a gene involved in transcriptional coactivation of nearly all RNA polymerase II-dependent genes, was detected with the HCC1395BL_v1.0 reference, but analysis on GRCh38 using the same set of the reads finds only a 40 base pair deletion, as the HCC1395BL_v1.0 reference has 17 more adenine (A) nucleotides in this region (Additional file 2: Fig. S9B). These two examples illustrate the importance of using a personalized genome as reference to accurately detect somatic SVs with short-read sequencing data when assessing matched tumor-normal cell lines. We then extended our analysis to include all 12 tumor-normal paired WGS replicates from 6 sequencing centers [5, 6] so that we could look more deeply into how each of the four callers would be impacted by use of the HCC1395BL_v1.0 reference as compared to GRCh38. For this analysis, we required SVs to be called in at least two replicates for each of the four callers. Similar trends regarding the counts of the somatic calls were seen for each of four callers with GRCh38 and HCC1395BL_v1.0 as references (Fig. 7A; Additional file 1: Table S10; Additional file 2: Fig. S10). For example, the total counts of somatic SVs (particularly for DELs and TRAs) on HCC1395BL_v1.0 were all greater than that on GRCh38 for all callers, and Delly had the lowest counts of somatic SVs when compared with other callers (Fig. 7A; Additional file 2: Fig. S10). The inversion counts were largely similar on two references with just 1 or 2 more inversions on HCC1395BL_v1.0. For DUP, both GRIDSS2 and Manta report 10 (4.44%) and 11 (4.54%) more on HCC1395BL_v1.0, but the DUP counts detected by Delly were unchanged. With regard to insertions, we observed 2 novel ones by GRIDSS2 (scaffold_47:144105:INS:66bp and scaffold_49:9684521:INS:60bp called by GRIDSS/Manta/Delly, but neither located in gene regions), and 4 by Delly (scaffold_47:144105:INS:66bp and scaffold_49:9684521:INS:60bp by GRIDSS/Manta/Delly, and scaffold_129:31804:INS:78bp by Manta/Delly, scaffold_20:21223789:INS:50bp by Delly only, also not located in gene regions) on HCC1395BL_v1.0 reference only, but none by novoBreak. Manta reported 18 insertions (one insertion scaffold_1:105582094:INS:52bp overlapping with a DUP scaffold_1:105582094- 105582146 by GRIDSS/Manta, located in an intronic region in the LIN5 gene) on HCC1395BL_v1.0, but 14 insertions on GRCh38 (Additional file 2: Fig. S10). By mapping these GRCh38-based SVs having support from two or more replicates (TRA excluded) onto HCC1395BL_v1.0 reference, we found that 623, 673, 545, and 556 GRCh38-based SVs by GRIDSS2, Manta, Delly, and novoBreak, respectively, were mapped with HCC1395BL_v1.0-based SVs (Fig. 7B), while 24 SVs by GRIDSS2, 41 SVs by Manta, 43 SVs by Delly, and 28 SVs by novoBreak were considered as “unmapped” or “mapped but without matching SVs on HCC1395BL_v1.0” (Fig. 7C). Meanwhile, when using the HCC1395BL_v1.0 as reference, 61 SVs by GRIDSS2, 86 SVs by Manta, 61 SVs by Delly and 55 SVs by novoBreak with supports from two or more replicates lacked GRCh38-based SVs at corresponding locations on HCC1395BL_v1.0 reference under our mapping criteria (Fig. 7D).
Due to the lack of a somatic SV caller that uses both PacBio long reads and assembled contigs as inputs, somatic SVs were defined as SV calls in tumor cell line (HCC1395) that were without overlapping germline SV calls in normal (HCC1395BL) cell line in SV regions (see “Methods”). For each of the six calling methods, including four for PacBio long reads and two for assembled contigs, we observed that the total counts of somatic SVs were generally increased with the HCC1395BL_v1.0 reference as compared to GRCh38 (Additional file 1: Table S12). The initial merged consensus somatic SVs that had support from two or more calling methods contained an additional 194 SVs (138 DEL, 31 DUP/INS, 3 INV, and 22 TRA) by SV count when HCC1395BL_v1.0 was used as reference as compared to GRCh38 (Fig. 8A). By mapping the initial GRCh38-based consensus SVs having support from two or more calling methods (except TRA) to the HCC1395BL_v1.0 reference, we found 1144 of 1318 SVs (86.8%) were able to localize on HCC1395BL_v1.0, while 174 SVs (13.2%) were considered “unmapped” on HCC1395BL_v1.0 with our mapping criteria. Among those that were mapped, 814 SVs had matched SVs on HCC1395BL_v1.0, but 330 SVs were mapped without HCC1395BL_v1.0-based SVs. We observed that more than half of those “unmapped” and “mapped but without matching SVs” had support from only two calling methods. Thus, to further reduce potential noisy somatic SV calls from the consensus callset, we subsequently required somatic SVs to be supported by three or more calling methods. Applying such criteria, 660 of 744 SVs (89%) were mapped onto HCC1395BL_v1.0, and among them, 531 SVs were mapped with matched SVs on HCC1395BL_v1.0, whereas 129 SVs were mapped but without matching SVs. Eighty four of 744 SVs (11%) were considered “unmapped” on HCC1395BL_v1.0 (Fig. 8B). Functional analysis using ANNOVAR showed that 54 of 129 mapped but without matching SVs on HCC1395BL_v1.0 and 24 of 84 unmapped SVs overlapped with gene regions on GRCh38 (Additional file 1: Table S13). Therefore, inclusion of these questionable SVs in mutation analysis may cause misinterpretations. Furthermore, we identified 279 SVs (including 217 DEL, 61 DUP/INS, and 1 INV) on HCC1395BL_v1.0 lacking corresponding GRCh38-based SVs with our current mapping criteria. By collocating those SVs with RefSeq genes and transcripts mapped onto HCC1395BL_v1.0, we found 91 SVs (72 DEL and 19 DUP/INS) were mapped onto 86 gene regions. KEGG pathway enrichment analysis suggested that some of these 86 genes were involved with pathways related to cancer invasion and metastasis (e.g., CDH23, ST14), PI3K-Akt signaling pathway, and G protein-coupled receptor signaling pathway (e.g., GNG7) (Fig. 8C; Additional file 1: Table S14). An annotation of the sequences of these 72 deletions by RepeatMasker (https://www.repeatmasker.org/cgi-bin/WEBRepeatMasker) demonstrated that 32 deletions overlapped 10 classes of repeat families, among them 17 SINE/Alu, 8 simple repeats, and 3 retroposon/SVA (Fig. 8D). We manually curated several of the gene-overlapping deletions (DELs) in IGV and confirmed that these deletions were detected only on HC1395BL_v1.0, and not on GRCh38 with in-read deletions from PacBio long reads (Additional file 2: Fig. S12), demonstrating the importance of using a personalized genome as reference to accurately detect somatic SVs in tumor-normal paired samples. For example, CDH23 (cadherin related 23) belongs to the cadherin superfamily that encodes calcium dependent cell-cell adhesion glycoproteins. This gene has been reported to play a role in early stages of tumor metastasis through regulation of cell-cell adhesion, and upregulation of CDH23 gene may be associated with breast cancer [36, 37]. A 327 bp homozygous deletion (scaffold_12:20762508-20762835), which overlaps CDH23 gene, was uncovered in tumor cell line when HCC1395BL_v1.0 was used as reference, which includes a copy of SINE/AluY (284 bp), but this AluY sequence is not present in GRCh38. Thus, mapping the same set of tumor reads to GRCh38 would not identify this deletion (Additional file 2: Fig. S12A). Similarly, a 128 base pair homozygous deletion (scaffold_24:40364012-40364140) that overlapped an LTR/ERVL repeat was uncovered in tumor cell line when using HCC1395BL_v1.0 reference, but not when using GRCh38. This deletion was located in an intronic region (exon1 and exon2) of ST14 (ST14 transmembrane serine protease matriptase) (Additional file 2: Fig. S12B). Studies have associated the expression of this protease with breast, colon, prostate, and ovarian tumors. Additionally, an intronic 289 base pair homozygous AluY deletion (scaffold_20:26715781-26716070, overlapping with ACE gene) (Additional file 2: Fig. S12C), a 538 base pair homozygous deletion (scaffold_37:2324216-2324754, overlapping with GNG7 gene) (Additional file 2: Fig. S12D), and a 1672 base pair homozygous deletion (scaffold_8:42055418-42057090, overlapping with JAG2 gene) (Additional file 2: Fig. S12E) were detected in the tumor cell line when using HCC1395BL_v1.0 reference, but not when using GRCh38. Combining the aforementioned 6 somatic SV sets (generated by 6 calling methods using PacBio long-read data as well as assembled contigs from the tumor cell line) with 4 somatic SV sets (generated by 4 somatic SV callers using Illumina short-read data), we uncovered 1796 somatic SVs (TRAs excluded) supported by 2 or more calling methods when using the HCC1395BL_v1.0 as reference, which was 201 more somatic SVs detected than when using GRC38 reference (Additional file 2: Fig. S13A, B). Even when applying the stringent requirement that each of the somatic SVs is supported by 3 or more calling methods, 142 more somatic SVs (TRAs excluded) were detected with the HCC1395BL_v1.0 reference as compared to GRCh38 (Additional file 2: Fig. S13A, B). When requiring each of the somatic SVs to be supported by 2 or more calling methods and using the personalized assembly HCC1395BL_v1.0 as reference, 705 somatic SVs were supported by short-read sequencing data, 1381 somatic SVs were supported by PacBio long-read sequencing data, and 802 somatic SVs were supported by assembled contig sequences for the tumor cell line (Additional file 2: Fig. S13C). This observation indicates that each sequencing technology or data source has its own unique advantages or limitations. Together with all the evidence we have illustrated in this study, the benefits of using personalized genome assembly as reference are evident. The personalized genome not only comprises individual-specific haplotypes with better representations of the clinically important genes, but also enables better mappings for both short and long reads, and subsequently more accurate identification of somatic mutations in tumor-normal paired samples, when it is used as reference as compared to GRCh38 (Table 2).
In this study, we used a combination of multiple sequencing technologies, including sequencing data consisting of short reads, linked reads, and long reads, to construct the first de novo assemblies of a tumor-normal pair from the same individual with breast cancer. We subsequently used this well-assembled genome as a personal genome reference, in comparison to using the generic human reference GRCh38, for somatic variant detection and demonstrated the advantages of using a personalized genome as a reference. Our analyses of existing data for HCC1395BL demonstrated that we generated a high-quality assembly in terms of contiguity and gene content, i.e., 99.9% of RefSeq protein-coding genes were successfully mapped onto the personal genome reference with minimum 95% alignment identity and 50% alignment coverage, while the vast majority (99.27%) of these genes were aligned with at least 95% coverage. Complex genomic regions were well-assembled, as evidenced by our demonstration that the complete HLA region, representing an individualized haplotype, is found in a single scaffold for HCC1395BL. Additionally, we found that some clinically relevant genes such as GSTT1 and KIR2DL5 (KIR2DL5A), which are not represented in the chromosomes of the GRCh38 primary assembly, were also captured in our de novo HCC1395BL assembly. For the first time, we were able to identify cancer somatic mutations based on de novo assembly from the same person, instead of inferring them from the alignments to a mosaic standard reference benchmark such as GRCh38 [13]. Our analysis showed that the de novo assembly improved short-read mapping, resulting in a greater percentage of properly mapped mate-pair reads, reduced total numbers of mismatches and split reads, and many fewer reads with improper-pairing, soft-clipping, or hard-clipping, indicating that short-read mapping was improved with personalized reference genome. In particular, short reads were more uniformly placed on the personal genome reference than GRCh38 as shown by the smaller standard deviations for both read coverages and the library insert sizes. As a result, discovery of somatic SNVs and small indels by different calling algorithms with short reads was more consistent, and more overlapping calls between callers were observed with the personal reference. Mapping analysis of GRCh38-based somatic SNVs set with flanking regions to the de novo assembled personal reference revealed that only 88.25% somatic SNVs were completely identical to the personal reference-based SNVs, and 9.59% somatic SNVs may have the same reference/alternate alleles on the de novo assembly as on the GRCh38 reference genome, but their flanking sequences may be slightly different with some germline SNVs, highlighting the critical importance of personal genome assembly for individualized medical research. A small percentage (1.64%) of GRCh38-based SNVs had good mapping locations on the personal genome but did not have corresponding SNV calls, suggesting potential false positives exist. Our findings indicated that use of a personal genome as reference had impacts on SV discovery using short reads, but the extent of impact on SV calling depended on the SV callers, SV types, and the SV-calling algorithms. If the mappings of underlying supporting reads for potential SVs are improved with the personal reference, the respective SV calls should be improved, and such conspicuous improvements would be reflected in the SV results from these SV callers. Consistent with this assertion, we found with all tested callers that the somatic SV counts detected with use of the personal genome reference were generally higher than that when using traditional GRCh38 reference. In particular, even with short-read sequencing data, the personalized genome reference enabled us to identify additional somatic SVs that could not be detected on the GRCh38 reference due to the absence of certain personalized sequences (e.g., repeats) in the corresponding locations. The personal genome reference also impacted long-read mapping and contig assembly-to-assembly mapping, as well as subsequent SV detections. Our analysis showed that more reads were mapped with fewer mismatches, and mapped reads were more uniformly placed on the personal genome reference as shown by the smaller standard deviations of read coverages. As a consequence, large SVs detected were more accurate. Most strikingly, germline insertions were significantly reduced using the personal genome as reference, possibly due to systematic collapse of repeats (thus leading to genome-wide deletion bias) in GRCh38 [38]. Such bias would have impacts on germline SV detections with a tendency to call more insertions [17, 39]. Ultimately, this would affect somatic SV calling as well. As demonstrated in our analysis, somatic SV counts were generally greater when the de novo personal assembly was used as reference as compared to GRCh38. With PacBio long-read sequencing data and assembled contigs mapped onto the personalized genome reference, we uncovered additional somatic SVs only with HCC1395BL_v1.0 reference that could not be detected with GRCh38 reference due to the absence of the personalized sequences at these locations. Not surprisingly, some of these deletions could be easily confirmed with in-read deletions from PacBio reads using IGV and read mappings. Noticeably, some of these additional somatic SVs identified with the personalized genome reference using short- and long-read sequencing data were located in the regions of genes that are involved in pathways related to cancer development and metastasis. Our approach using a personal assembly (HCC1395BL_v1.0) as reference identified many additional personalized somatic SNVs and SVs which were missed using GRCh38 as reference. These more personalized SNVs/SVs provide additional target choices for patients, researchers or clinicians, and physicians to look into further for personalized patient care. They may reduce the pursuit of incorrect treatment options based on GRCh38-specific or other non-personalized reference somatic SNVs/SVs, and missed opportunities for interventional target-specific treatment options. As demonstrated in this study, use of a personalized genome as reference for somatic mutation calling in tumor-normal paired samples is promising, but the cost of creating such a personalized genome is still high as compared to using the generic reference. But the goal of scientific research is to find the truth, and this approach will assist in achieving that goal. Moreover, sequencing technologies have evolved rapidly in the last two decades, and the cost of sequencing has decreased substantially. Therefore, it is reasonable to expect that in the near future, the creation of a personalized genome for use as a reference will be cheaper than today. As sequencing technology continues to advance, longer read length and lower per-base error rate offer great opportunity to tackle many difficult genomic regions, such as telomeres, centromeres, and regions with unplaced/unlocalized sequences, and unfinished gaps as indicated in GRCh38 [9]. Those regions which were previously impossible to assemble, and whose biology is consequently poorly understood, are now within reach [17]. Such new developments should encourage the scientific community to continue improving the quality of the de novo personal assembly for this tumor-normal pair by applying PacBio’s HiFi reads and Oxford Nanopore’s ultra-long reads in the near future. Such advancement in genome assembly will also provide a better path forward for improving somatic variant identification using a personalized genome as reference. Ultimately, it will lead to discovery of vital genetic markers for cancer diagnosis and therapeutic monitoring, as well as more insights into molecular understanding of tumorigenesis, so that design of personalized immunotherapies, detection of potential off-target effects of gene editing, and other aspects of drug development will be improved. Recently, the Telomere-to-Telomere (T2T) consortium finished the first gapless telomere-to-telomere human genome assembly (T2T-CHM13) [40] and illustrated its advantages as a reference over GRCh38 for germline variant detection in population genetic analyses [41]. Theoretically, this new reference would improve somatic mutation detection as compared to GRCh38, but the extent of such improvements for somatic mutation discovery in tumor-normal samples has not yet been investigated. Some of the benefits we reported in this study may be impacted. For instance, read mappings to T2T-CHM13 are anticipated to be better than those to GRCh38, but a personal genome (especially a complete T2T personal genome) reference would still probably outperform T2T-CHM13. Although HCC1395 (https://www.atcc.org/products/crl-2324) is from a Caucasian sample and CHM13 is mostly of European origin [40], there are likely some T2T-CHM13-specific somatic mutations that should be avoided, as well as some personal genome-specific somatic mutations that we would like to use as additional choices for personalized patient care and precision oncology medicine. If the ultimate goal of our patient care is individualized or personalized, then use of a personalized assembly rather than GRCh38 or T2T-CHM13 as reference to identify the full spectrum of somatic mutations in tumor-normal samples is advocated.
We demonstrated that a personalized genome not only has individual-specific haplotypes that provide better representations of genomic regions in the sample, including clinically relevant genes, but it also enables better alignments for both short and long reads. Consequently, it allows for more accurate detection of somatic mutations, including somatic SNVs and SVs, in paired tumor-normal samples. In particular, novel somatic mutations (SNVs/SVs) were discovered only with a personalized genome as a reference, but not with traditional GRCh38. The unique resource we established in this study will be valuable to the development of precision oncology medicine not only for breast cancer, but also for other cancers.
A matched tumor/normal pair of cell lines, derived from a TNBC breast cancer (HCC1395) and from normal B cells from the same donor (HCC1395BL), was selected for whole genome sequencing with multiple platforms [5, 6]. We included about 175-fold of Illumina short reads from 3 replicates (FDN1, FDN2, and FDN3) as FDN123 sequenced by Fudan University, 161-fold of 10X Genomics (10X) linked reads, 53-fold of Pacific Bioscience (PacBio) long reads (full-pass subreads, average length 9089 bp), 71-fold of Hi-C reads, and 15-fold Oxford Nanopore technologies (ONT) reads in development of the HCC1395BL_v1.0 assembly (Additional file 1: Table S1). For HCC1395, we used about 170-fold of Illumina short reads from 3 replicates (FDT1, FDT2, and FDT3) as FDT123 sequenced by Fudan University, 160-fold of 10X Genomics linked reads, and 46-fold of Pacific Bioscience (PacBio) long reads (full-pass subreads, average length 8146 bp). We also added all 12 WGS tumor-normal paired replicates (Illumina short reads) from 6 sequencing centers, including FDT1/FDN1 (FDR1), FDT2/FDN2 (FDR2), FDT3/FDN3 (FDR3), ILT1/ILN1 (ILR1), ILT2/ILN2 (ILR2), ILT3/ILN3 (ILR3), NVT1/NVN1 (NVR1), NVT2/NVN2 (NVR2), NVT3/NVN3 (NVR3), NCT1/NCN1 (NCR1), EAT1/EAN1 (EAR1), and LLT1/LLN1 (LLR1), for short-read mapping and SNV analysis [5, 6]. The median insert sizes were 377/367 for FDT1/FDN1, 375/371 for FDT2/FDN2, 371/368 for FDT3/FDN3, 417/419 for ILT1/ILN1, 395/393 for ILT2/ILN2, 401/402 for ILT3/ILN3, 404/400 for NVT1/NVN1, 394/390 for NVT2/NVN2, 389/395 for NVT3/NVN3, 408/417 for NCT1/NCN1, 422/412 for EAT1/EAN1, and 372/377 for LLT1/LLN1 [5, 6]. Library preparations and sequencing for Illumina short reads, 10X Genomics linked reads, and PacBio long reads were described previously [5, 6]. Dovetail Hi-C library preparation and sequencing: The Dovetail Hi-C libraries were prepared as described previously (Erez Lieberman-Aiden et al., 2009). For each library, chromatin was fixed in place in the nucleus with a 1% formaldehyde solution and then extracted. Fixed chromatin was digested with DpnII, the 5′ overhangs were filled in with biotinylated nucleotides, and then free blunt ends were ligated. After ligation, crosslinks were reversed, and the DNA purified from protein. Purified DNA containing biotinylated free-ends was removed as it does not reflect proximity-ligated molecules. The DNA was then sheared to ~350 bp mean fragment size, and sequencing libraries were generated using NEBNext Ultra enzymes and Illumina-compatible adapters. Internal biotin-containing fragments were isolated using streptavidin beads before PCR enrichment of each library. The libraries were sequenced on an Illumina HiSeq X to a depth of ~200M read pairs per library. Oxford Nanopore technologies (ONT) MinION sequencing data: Genomic DNA from the HCC1395BL cell line was extracted using the QIAGEN MagAttract HMW DNA Kit (QIAGEN, Hilden, Germany). One microgram of freshly isolated genomic DNA without fragmentation was used for library construction using the SQK-LSK109 ligation sequencing kit (ONT, Oxford, UK). Libraries were prepared following ONT standard protocol. Each library was sequenced on an individual MinION FLO-MIN106D R9.4 flowcell. Prior to sequencing, flowcell pore counts were measured using the MinKNOW Platform QC script (Oxford Nanopore Technologies, Oxford, UK). About 300 ng of completed libraries was loaded as instructed by ONT. Raw sequence reads were called in real time by the MinION operating software MinKNOW (Guppy version 2.1.3). Sequence data passing quality parameters (qmean > 7) were converted to fastq format. Only the reads passed the QC were included in further analyses.
PacBio long reads data were first error-corrected and then assembled into primary contigs using the “canu” assembler (version 1.8) [20] with option by its developers. The contig sequences were then polished with Illumina paired-end reads using PILON (version 1.22) [42]. The polishing process was performed twice to achieve the best results. Scaffolding with linked reads was performed using ARCS (version 1.0.5) [43], while scaffolding with Hi-C data was completed using SALSA (https://github.com/marbl/SALSA) [44]. Linked reads from 10X Genomics were assembled using the “Supernova” assembler (version 2.0.0) [21] as instructed by its developer. Contig assembly with PacBio long reads and polishing with Illumina short reads: canu -p hcc1395bl -d hcc1395bl_out genomeSize=3.1g useGrid=false maxThreads=16 corConcurrency=4 corThreads=4 cormmapThreads=4 cormhapConcurrency=4 corovlConcurrency=4 maxMemory=360g correctedErrorRate=0.075 -pacbio-raw pacbio_reads.fa java -Xmx300G -jar pilon-1.22.jar --genome pacbio_contig.fa --frags ILMN_read.bam --diploid --fix bases --outdir pilon --output pilon --changes Scaffolding with 10X Genomics’ linked reads using ARCS: longranger-2.2.2/longranger align --id=FDN123 --fastqs=./lib1/,./lib2/,./lib3/,./lib4/,… --sample=FDN123 --reference=./refdata-pacbio_contig_pilon --localmem=200 --localcores=8 --jobmode=local arcs --file=./pacbio_contig_pilon.fasta --fofName=./aln_list.txt python ./arcs/Examples/makeTSVfile.py ./pacbio_contig_pilon.fa.scaff_s98_c5_l0_d0_e30000_r0.05_original.gv hccbl_pilon.fasta.scaff_s98_c5_l0_d0_e30000_r0.05.tigpair_checkpoint.tsv ./pacbio_contig_pilon.fa links_v1.8.6/LINKS -f ./ pacbio_contig_pilon.fa -s empty.fof -k 20 -b pacbio_contig_pilon.fa.scaff_s98_c5_l0_d0_e30000_r0.05 -l 5 -t 2 -a 0.3 mv pacbio_contig_pilon.fa.scaff_s98_c5_l0_d0_e30000_r0.05.scaffolds.fa pacbio_contig_pilon_arcs.scaffolds.fa Scaffolding with Hi-C reads using SALSA: SALSA/run_pipeline.py -a pacbio_contig_pilon.fa -l pacbio_contig_pilon.fa.fai -b reads.bam_k4sort.bed -e GATC -o scaffolds Supernova assembly for 10X Genomics linked reads: supernova-2.0.0/supernova run --id=FDN123 --fastqs=./lib1/,./lib2/,./lib3/,./lib4/,… --sample =FDN123 --localmem 360 After scaffolding with ARCS and SALSA, we mapped the unitig sequences, which were produced with Illumina short reads using fermikit (version r188) [45], to the scaffold assembly using BWA [46], and then used bcftools (version 1.6, https://samtools.github.io/bcftools/bcftools.html) to generate the final consensus assembly (HCC1395BL_v1.0). Scaffolds smaller than 10 kb were excluded from further analysis. fermi.kit/fermi2.pl unitig -s3g -t8 -l150 -p prefix "cat WGS_FDN123_R1.fq.gz WGS_FDN123_R2.fq.gz" > prefix.mak make -f prefix.mak bwa mem -t 8 scaffolds_FINAL.fasta prefix.mag.gz | samtools sort -@ 8 -o unitigFDN123_sorted.bam bwa index scaffolds_FINAL.fasta samtools mpileup -uf scaffolds_FINAL.fasta unitigFDN123_sorted.bam | bcftools call -mv -Oz -o calls.vcf.gz tabix calls.vcf.gz cat scaffolds_FINAL.fasta | bcftools consensus calls.vcf.gz > final_consensus.fa The Illumina short reads were aligned onto the HCC1395BL_v1.0 genome using BWA mem [46], and duplicated reads were marked with Picard MarkDuplicates. High-confidence heterozygous sites (QUAL ≥ 30) were identified using GATK4 (version gatk-4.0.3.0) [47]. Calls on chrX, chr6p, and chr16q regions were excluded. Phasing was performed with the identified high-confidence heterozygous sites and long reads from PacBio and ONT using WhatsHap (version 0.18) phasing tool [48]. Statistics of phasing was generated using “whatshap stats.” Two haplotypes of the assembly in FASTA format were also reconstructed with the phasing information. Assembly-based germline SNVs in diploid regions of autosomal chromosomes were called using dipcall (https://github.com/lh3/dipcall) with two haplotypes as inputs. whatshap phase --reference final_consensus.fa -o phased.vcf ILMN_gatk.vcf pacbio_minimap2_sorted.bam --ignore-read-groups --sample=HCC1395BL whatshap stats --gtf=phased.gtf phased.vcf 1>phased.vcf.gz_stats bgzip phased.vcf tabix phased.vcf.gz bcftools consensus -H 1 -f final_consensus.fa phased.vcf.gz > haplotype1.fasta bcftools consensus -H 2 -f final_consensus.fa phased.vcf.gz > haplotype2.fasta
QUAST (version 5.0.0) [49] and Benchmarking Universal Single-Copy Orthologue (BUSCO, version 3.0.0) [50] were used to assess the quality of each de novo assembly. BLAT (v36) was used for mapping all RefSeq mRNA transcripts (accession prefixed with NM_ and NR_) (https://ftp.ncbi.nlm.nih.gov/genomes/all/GCF/000/001/405/GCF_000001405.38_GRCh38.p12/GCF_000001405.38_GRCh38.p12_rna.fna.gz) that were previously annotated on the GRCh38 assembly to the new assembly with parameter minIdentity 92. quast assembly1.fa assembly2.fa assembly3.fa …. -m 1000 --no-icarus python run_BUSCO.py --in final_consensus.fa --out --lineage_path …lineage_files/mammalia_odb9 --mode genome -sp human For GRCh38 consistency analysis, each assembly was compared with the GRCh38 reference assembly (https://ftp-trace.ncbi.nlm.nih.gov/ReferenceSamples/seqc/Somatic_Mutation_WG/technical/reference_genome/GRCh38/GRCh38.d1.vd1.fa) using minmap2 [51]. Alignments with mapping quality 60 and alignment length 100Kb+ were considered as good links for the consistency plot by Circos (Krzywinski, M., et al., 2009). We also introduced a new parameter “Top50,” which is the summed length of the 50 longest scaffolds, to monitor the contiguity of a given assembly during the scaffolding process, as the long-read and Hi-C sequencing technologies could make it possible to have arm-scale or chromosomal scale assembly. For the human genome with a total of 48 chromosomal arms, Top50 might be a suitable indicator to reflect the contiguity of the scaffold assembly if each chromosomal arm forms a scaffold.
To better annotate the final assembly HCC1395BL_v1.0, BLAT (version 36) and AUGUSTUS (version 3.3.1) [52] were used to map the previously described RefSeq transcripts to the assembly (excluding all pseudogenes and genes from NC_000024 chromosome Y). Protein-coding transcripts with annotations containing “pseudogene” and non-protein-coding transcripts with annotations containing “pseudo=true” in their deflines were considered as “pseudogenes” in this analysis. For BLAT, the option “minIdentity” was set to 92. Transcripts with more than 95% alignment and 95% ungapped identity were considered mapped onto the HCC1395BL_v1.0 assembly. In case of multiple mapping locations, the best mapping location with the maximum number of matching bases for the transcript was selected. blat -minIdentity=92 -q=rna -out=psl final_consensus.fa GCF_000001405.38_GRCh38.p12_rna.fa cdna.out.psl augustus --species=human --hintsfile=hints.E.gff –extrinsicCfgFile extrinsic.ME.cfg --outfile=augustus.out final_consensus.fa
BWA [46] was used to align Illumina short reads from each of the 12 replicates onto the de novo (HCC1395BL_v1.0) and the GRCh38 primary assemblies, respectively. Duplicate reads were marked with Picard MarkDuplicates. For FDN123 and FDT123, 3 bams from 3 replicates of normal sample, and 3 bams from 3 replicates of tumor sample were merged separately using “samtools merge.” Mapping statistics such as reads mapped, reads unmapped, reads mapped and paired, reads properly paired, and mismatches were collected using samtools (version 1.11) with the “stats” option (http://www.htslib.org/doc/samtools-stats.html). We defined the numbers of non-properly (or improperly) paired reads as the subtraction of properly paired reads from the mapped-and-paired reads. PacBio long reads were aligned onto two references using minimap2 [51]. Standard deviations of read coverages were based on read alignments with minimum mapping quality 10. bwa mem -t 8 -R "@RG\tID:FDN1\tSM:HCC1395BL\tLB:FDN1\tPU:FDN1\tPL:illumina" GRCh38.d1.vd1.fa WGS_FDN1_R1.fq.gz WGS_FDN1_R2.fq.gz | samtools view -bS - > FDN1.bam samtools sort -T . FDN1.bam > FDN1_sorted.bam samtools index FDN1_sorted.bam java -jar picard.jar MarkDuplicates INPUT=FDN1_sorted.bam OUTPUT=FDN1_sorted_dupmarked.bam METRICS_FILE=metrics.txt samtools stats FDN1_sorted_dupmarked.bam > FDN1_sorted_dupmarked.bam_stats For all SNV/SV variant analysis, variants in VCF files with “PASS” filter were included. The chrX, chr6p (coordinates below 58,500,000), and chr16q (coordinates above 38,400,000) regions were not included for variant comparison, for consistency with the reference somatic set from the SEQC2 Somatic Mutation Working Group [5, 6]. Variant calls from chrY and unlocalized/unplaced sequences (names with chrUn_, _random, _decoy, etc.) were also excluded. Strelka2 (version 2.9.2) [27] and MuTect2 (version gatk-4.0.3.0/gatk Mutect2) [28] were used to identify somatic SNVs and indels. MuTect2 VCF output was filtered using “gatk FilterMutectCalls.” MuTect2 reported SNPs and InDels in a single VCF file, and Strelka2 reported SNVs and InDels in separate VCF files. SNPs and InDels from MuTect2 and Strelka2 calls were compared separately using “bcftools isec” followed with “rtg vcfeval” (https://github.com/RealTimeGenomics/rtg-tools; doi: 10.1101/023754) [53] to obtain common sites called by both callers. gatk Mutect2 --native-pair-hmm-threads 8 -R GRCh38.d1.vd1.fa -I FDT1_sorted_dupmarked.bam -tumor FDT1 -I FDN1_sorted_dupmarked.bam -normal FDN1 -O mutect2_snvs_indels.vcf.gz gatk FilterMutectCalls --variant mutect2_snvs_indels.vcf.gz --output mutect2_snvs_indels_filt.vcf strelka-2.9.2.centos6_x86_64/bin/configureStrelkaSomaticWorkflow.py --normalBam FDN1_sorted_dupmarked.bam --tumorBam FDT1_sorted_dupmarked.bam --referenceFasta GRCh38.d1.vd1.fa --runDir strelka2_fdt1 ./runWorkflow.py -m local -j 8 bcftools isec Strelka2_vcf.gz MuTect2_vcf.gz -p isec_Streka2_MuTet2 --collapse all rtg vcfeval --baseline 0000.vcf.gz --calls 0001.vcf.gz --template GRCh38/SDF --output rtg_results --sample HCC1395,HCC1395 --vcf-score-field INFO.TLOD --squash-ploidy Somatic SVs from Illumina short reads (WGS) were discovered using GRIDSS2/GRIPSS [32], Manta [33], Delly [34], and novoBreak [35] as suggested by their developers. Merged tumor and normal (FDT123/FDN123) bams from 3 tumor-normal paired replicates (FDT1/FDN1, FDT2/FDN2, and FDT3/FDN3) from one sequencing center (FD) were used for initial somatic SV discovery. Later on, Illumin short-read sequencing data from all 12 paired replicates from 6 sequencing centers were analyzed using all four short-read somatic callers. SVs with SVLEN smaller than 50bp were removed. Intra-chromosomal BNDs (if reported) were filtered out for comparison. SVs with quality score below 20 from novoBreak calls were excluded. gridss-2.13.2/gridss --jar gridss-2.13.2/gridss-2.13.2-gridss-jar-with-dependencies.jar –reference GRCh38.d1.vd1.fa --output gridss_output.vcf.gz --assembly assembly_n1t1.bam --thread 8 --workingdir ./gridss1 FDN1_sorted_dupmarked.bam FDT1_sorted_dupmarked.bam java -jar gripss/gripss.jar -sample FD_T1 -reference FD_N1 -ref_genome GRCh38.d1.vd1.fa -pon_sgl_file gridss1/pondir/gridss_pon_single_breakend.bed_sort -pon_sv_file gridss1/pondir/gridss_pon_breakpoint.bedpe_sort -vcf FD_T1_gripss.vcf.gz -output_dir gridss1/pondir R/4.1.2/bin/R --vanilla --slave < gridss/example/simple-event-annotation.R manta-1.6.0.centos6_x86_64/bin/configManta.py --normalBam FDN1_sorted_dupmarked.bam --tumorBam FDT1_sorted_dupmarked.bam --referenceFasta GRCh38.d1.vd1.fa --runDir ./T1N1 ./T1N1/runWorkflow.py delly call -x human.hg38.excl.tsv -q 20 -s 15 -o fd_t1.bcf -g GRCh38.d1.vd1.fa FDT1_sorted_dupmarked.bam FDN1_sorted_dupmarked.bam delly filter -f somatic -o fd_t1.pre.bcf -s ./samples.tsv fd_t1.bcf delly call -g GRCh38.d1.vd1.fa -v fd_t1.pre.bcf -o fd_geno.bcf -x human.hg38.excl.tsv FDT1_sorted_dupmarked.bam FDN1_sorted_dupmarked.bam delly filter -f somatic -o fd_t1.somatic.bcf -s samples.tsv fd_geno.bcf novoBreak_distribution_v1.1.3rc/run_novoBreak.sh novoBreak_distribution_v1.1.3rc GRCh38.d1.vd1.fa FDT1_sorted_dupmarked.bam FDN1_sorted_dupmarked.bam 8 novobreak_out Alignment-based structural variations from PacBio long-read data were identified using PBMM2/PBSV pipeline (version 2.4.0, https://github.com/PacificBiosciences/pbsv) and NGMLR/Sniffles2 pipeline [54] (version 2.0.5, https://github.com/fritzsedlazeck/Sniffles, https://github.com/philres/ngmlr). We noticed that there were certain differences in their alignments with two aligners (PBMM2 vs. NGMLR) using PacBio subreads, thus resulting SV calls (PBSV vs. Sniffles2) were different in some regions. Therefore, to minimize aligner and caller bias and maximize callers’ concordance in subsequent merged callset, in our PacBio long-read SV analysis, Sniffles2 was also applied to PacBio pbmm2 bams after the bams were processed with “samtools calmd -u,” and PBSV was applied to NGMLR bams. SVs were called jointly with both tumor and normal sample together. Therefore, for each assembly, four VCFs (PBMM2+PBSV, PBMM2+Sniffles2, NGMLR+PBSV, and NGMLR+Sniffles2) were generated for downstream analysis. Only calls with “PASS” were retained. Calls with “IMPRECISE,” “SHADOWED,” and SVLEN smaller than 50bp were removed. CNV calls and intra-chromosomal BND from PBSV were ignored. Two selection steps were applied for identifying somatic SVs based on the genotypes, alternate allele counts, and allele frequency in tumor and normal sample reported by PBSV and Sniffles2. Firstly, if the genotype of the site for normal sample was reported as “reference” in normal sample, the site was retained only if the genotype for tumor sample was reported as “1/1” or “0/1,” with minimum alternate allele count 5 and minimum allele frequency 0.2. Secondly, if the genotype of the site for normal sample was reported as “0/1” in normal sample, the site was retained only if the genotype for tumor sample was reported as “1/1” with minimum alternate allele count 10, minimum allele frequency 0.85, and allele frequency difference between tumor and normal sample 0.45 and above. pbsv discover -s Tumor HCC1395_pbmm2_merged.bam HCC1395_pacbio_B38.svsig.gz pbsv discover -s Normal HCC1395BL_pbmm2_merged.bam HCC1395BL_pacbio_B38.svsig.gz pbsv call -j 8 GRCh38.d1.vd1.fa HCC1395_pacbio_B38.svsig.gz HCC1395BL_pacbio_B38.svsig.gz HCC1395_pacbio_pbmm2_B38_pbsv_TN.vcf Sniffles2.0/bin/sniffles --input HCC1395_ngmlr_B38.bam --vcf HCC1395_pacbio_ngmlr_sniffles2_B38.vcf.gz --snf HCC1395_pacbio_ngmlr_sniffles2_B38.snf --threads 8 Sniffles2.0/bin/sniffles --input HCC1395BL_ngmlr.bam --vcf HCC1395BL_pacbio_ngmlr_sniffles2_B38.vcf.gz --snf HCC1395BL_pacbio_pbmm2_sniffles2_B38.snf --threads 8 Sniffles2.0/bin/sniffles --input HCC1395_pacbio_ngmlr_sniffles2_B38.snf HCC1395BL_pacbio_ngmlr_sniffles2_B38.snf --vcf HCC1395_pacbio_ngmlr_sniffles2_B38_TN.vcf Assembly-based SVs were generated from direct comparisons of the contigs of tumor cell line (HCC1395) with the HCC1395BL_v1.0 reference (as opposed to GRCh38 reference) using paftools [51] and Assemblytics [55] (https://github.com/MariaNattestad/assemblytics; http://assemblytics.com/) with procedures suggested by their developers. For Assemblytics, we prepared the delta input file by aligning contigs fasta to a reference using MUMmer/nucmer and run with the options of “10,000” for “Unique sequence length required,” “20,000” for “Maximum variant size,” and “50” for “Minimum variant size”. For paftools, only SVs with SVLEN equal to or greater than 50bp were included for analysis. The paftools tool reports only deletions and insertions, and it identified 7475 large deletions and 5215 large insertions based on HCC1395 contigs on GRCh38 reference, but only 3154 (57.8% less) large deletions and 2425 (53.5% less) insertions on HCC1395BL_v1.0 (Additional file 2: Fig. S11A). Assemblytics reports repeat/tandem contractions and expansions in addition to deletions and insertions. We observed that, with the exception of repeat_contraction category, Assemblytics identified fewer SVs on HCC1395BL_v1.0 (25.53% fewer deletions, 63.86% fewer insertions, 35.08% fewer repeat expansions, 32.49% fewer tandem contractions, and 71.29% fewer tandem expansions) (Additional file 2: Fig. S11B). To circumvent the different SV notations that were used by paftools and Assemblytics (e.g., a 668 bp Deletion at chr3:159539232-159539900 called by paftools vs. a 668 bp Repeat_contraction at chr3:159539232-159543981 called by Assemblytics), we combined deletions with repeat/tandem contractions (as “DEL”) and insertions with repeat/tandem expansions (as “INS”) from Assemblytics calls for comparison purposes. Somatic SVs were retained by removing calls that were overlapping with germline SV calls (requiring allele frequency 0.1 and above, and alternate allele count 5 or more) identified by Sniffles and PBSV in the normal sample with PacBio long reads. Due to the nature of such somatic SV set that were generated using similar approaches, for consensus somatic SVs analysis, we combined these 2 contig mapping-based somatic SV sets (by Assemblytics/paftools) along with 4 PacBio long-read-based somatic SV sets together (by Sniffles2/PBSV with PBMM2/ngmlr ) to evaluate how two genome references were affecting somatic SVs that were supported by at least two calling methods. minimap2 -cx asm5 -t8 --cs GRCh38.d1.vd1.fa HCC1395_pacbio_contigs_polished.fasta > asm.paf sort -k6,6 -k8,8n asm.paf > asm.srt.paf paftools.js call asm.srt.paf > asm.var.txt nucmer -maxmatch -l 100 -c 500 REFERENCE.fa contig.fa -prefix OUT gzip OUT.delta Consensus somatic SVs from multiple somatic SV callsets were generated using “merge” function of SURVIVOR (version: 1.0.7) with the parameters “max distance between breakpoints = 1000,” “Minimum number of supporting caller = 2,” and “Minimum size of SVs to be taken into account = 50.” SURVIVOR merge vcfs.list 1000 2 0 0 0 50 merged.vcf
To find the locations on the de novo assembly corresponding to the GRCh38-based somatic SNVs we identified, we used a two-step mapping approach. We extracted both the reference and alternate alleles of each SNV with their 50bp flanking sequences from GRCh38 and created a fasta file before mapping using BLAST (blast 2.10.1). The first step was to map all SNVs with more stringent criteria so that SNVs with identity ≥99% and alignment length ≥101 bp were selected. The unselected SNVs from Step1 were then mapped in the second step with lower thresholds (95% identity and 95 bp alignment) to select the SNVs that mapped best despite some mismatches and small indels (Additional file 2: Fig. S5). For both steps, both alleles of each SNV were required to map onto the same locations, with identical start and end positions on the de novo assembly. In addition, to be considered an equivalent SNV call between GRCh38 and the de novo assembly, the alternate allele was required to be at the center position. Manual inspections on IGV for some SNVs were also performed. Unselected SNVs from Step2 were considered to be unmapped. blastn -query query.fa -db blastdb/final_consensus -num_threads 8 -evalue 1e-10 -word_size 7 -num_alignments 10 -perc_identity 95 -qcov_hsp_perc 95 -dust no -soft_masking false -out blast.out -outfmt "6 qseqid sseqid qlen slen pident length mismatch gapopen qstart qend sstart send sstrand evalue bitscore" GRCh38-based SVs were mapped to HCC1395BL_v1.0 assembly using a similar approach as for mapping SNVs. Sequences of 100 base pairs from each SV’s flanking (for DEL/DUP/INS/INV only) were extracted as fasta format, and then were mapped using BLAST (blast 2.10.1). Only SV events with two flanking sequences being mapped in the same locations with the minimum 98% identity and 90 bp alignment were considered as “Mapped” onto HCC1395BL_v1.0. Otherwise, the SVs were considered as “Unmapped.”
Variant function analysis was performed using ANNOVAR (version: de74a7d59955d769c6cbb92a0d64d12c90c8eede, 2018-04-16) [29]. Pathway analysis was performed through the Enrichr web site (https://maayanlab.cloud/Enrichr/). annovar/annotate_variation.pl -build hg38 var.input humandb/ -dbtype ensGene Sequences of deletions (from predicted start to end) were extracted as fasta format, then were annotated using an online RepeatMasker tool (https://www.repeatmasker.org/cgi-bin/WEBRepeatMasker).
Contigs from HCC1395BL assembly and HCC1395 assembly that fully covered the mitochondrial sequences from GRCh38 (16,569 bp; https://www.ncbi.nlm.nih.gov/nuccore/NC_012920.1) were selected based on minimap2 mapping results. Since the mitochondrial genome is circular, the full mitochondrial sequences were extracted from each of the selected contigs based on BLAST mapping results. CLUSTAL (v1.2.4) was used to generate multiple sequence alignments for variant analysis. The variants were annotated with the MITOMAP human mitochondrial genome database (http://www.mitomap.org, 2019) and dbSNP (v153).
We randomly selected 12 SNVs (MAF ranges from 0.5 to 1) from those 177 SNVs that were discovered only using HCC1395BL_v1.0 as reference and designed primers for PCR validation using Sanger sequencing. To confirm the specific point mutations, primers flanking the mutations were designed with an online software- Primer3. The point mutation flanking regions were then amplified using either control or tumor DNA samples as a template with Phusion flash High-Fidelity PCR master mix (Thermo Fisher Scientific, Waltham, MA.). The PCR conditions were 98 °C for 30 s, followed by 35 cycles of denaturing at 98 °C for 1 s, annealing at 64 °C for 5 s and extension at 72 °C for 15 s. The PCR products were then purified with GeneJET PCR purification kit (Thermo Fisher Scientific, Waltham, MA) and sequenced at GENEWIZ (Genewiz, South Plainfield, NJ). We were not able to design primers to cover either the SNV scaffold_1:3482191 (MAF=1) or the SNV scaffold_8:61936447 (MAF=0.557) without overlapping due to the technical limitations for Sanger sequencing, thus these two SNVs were left out for further assessment.
Additional file 1. Included all the supplementary tables for this manuscript.Additional file 2. Included all the supplementary figures for this manuscript.Additional file 3. Review history. |
||
PMC9648004 | Yang Jiang,Junshuang Zhao,Rongqing Li,Yingliang Liu,Lin Zhou,Chengbin Wang,Caihong Lv,Liang Gao,Daming Cui | Correction: CircLRFN5 inhibits the progression of glioblastoma via PRRX2/GCH1 mediated ferroptosis | 10-11-2022 | Correction: CircLRFN5 inhibits the progression of glioblastoma via PRRX2/GCH1 mediated ferroptosis
Correction: J Exp Clin Cancer Res 41, 307 (2022) https://doi.org/10.1186/s13046-022-02518-8 Following publication of the original article [1], author identified an error in Fig. 4k. It was caused by errors in the publishing process. The correct figure is presented below: The correction does not have any effect on the results or conclusions of the paper. The original article has been corrected. |
|||
PMC9648005 | Raphaëlle Lesage,Mauricio N. Ferrao Blanco,Roberto Narcisi,Tim Welting,Gerjo J. V. M. van Osch,Liesbet Geris | An integrated in silico-in vitro approach for identifying therapeutic targets against osteoarthritis | 09-11-2022 | Network of signal transduction,Computational modeling,Drug targets,Osteoarthritis,Chondrocyte hypertrophy,In vitro validation,Regulatory network inference,Virtual cell | Background Without the availability of disease-modifying drugs, there is an unmet therapeutic need for osteoarthritic patients. During osteoarthritis, the homeostasis of articular chondrocytes is dysregulated and a phenotypical transition called hypertrophy occurs, leading to cartilage degeneration. Targeting this phenotypic transition has emerged as a potential therapeutic strategy. Chondrocyte phenotype maintenance and switch are controlled by an intricate network of intracellular factors, each influenced by a myriad of feedback mechanisms, making it challenging to intuitively predict treatment outcomes, while in silico modeling can help unravel that complexity. In this study, we aim to develop a virtual articular chondrocyte to guide experiments in order to rationalize the identification of potential drug targets via screening of combination therapies through computational modeling and simulations. Results We developed a signal transduction network model using knowledge-based and data-driven (machine learning) modeling technologies. The in silico high-throughput screening of (pairwise) perturbations operated with that network model highlighted conditions potentially affecting the hypertrophic switch. A selection of promising combinations was further tested in a murine cell line and primary human chondrocytes, which notably highlighted a previously unreported synergistic effect between the protein kinase A and the fibroblast growth factor receptor 1. Conclusions Here, we provide a virtual articular chondrocyte in the form of a signal transduction interactive knowledge base and of an executable computational model. Our in silico-in vitro strategy opens new routes for developing osteoarthritis targeting therapies by refining the early stages of drug target discovery. Graphical Abstract Supplementary Information The online version contains supplementary material available at 10.1186/s12915-022-01451-8. | An integrated in silico-in vitro approach for identifying therapeutic targets against osteoarthritis
Without the availability of disease-modifying drugs, there is an unmet therapeutic need for osteoarthritic patients. During osteoarthritis, the homeostasis of articular chondrocytes is dysregulated and a phenotypical transition called hypertrophy occurs, leading to cartilage degeneration. Targeting this phenotypic transition has emerged as a potential therapeutic strategy. Chondrocyte phenotype maintenance and switch are controlled by an intricate network of intracellular factors, each influenced by a myriad of feedback mechanisms, making it challenging to intuitively predict treatment outcomes, while in silico modeling can help unravel that complexity. In this study, we aim to develop a virtual articular chondrocyte to guide experiments in order to rationalize the identification of potential drug targets via screening of combination therapies through computational modeling and simulations.
We developed a signal transduction network model using knowledge-based and data-driven (machine learning) modeling technologies. The in silico high-throughput screening of (pairwise) perturbations operated with that network model highlighted conditions potentially affecting the hypertrophic switch. A selection of promising combinations was further tested in a murine cell line and primary human chondrocytes, which notably highlighted a previously unreported synergistic effect between the protein kinase A and the fibroblast growth factor receptor 1.
Here, we provide a virtual articular chondrocyte in the form of a signal transduction interactive knowledge base and of an executable computational model. Our in silico-in vitro strategy opens new routes for developing osteoarthritis targeting therapies by refining the early stages of drug target discovery.
The online version contains supplementary material available at 10.1186/s12915-022-01451-8.
Osteoarthritis (OA) is a degenerative disease of the joint increasingly prevalent due to the aging population. It is a major societal burden as no disease-modifying drugs are currently available on the market [1]. OA is characterized by cartilage damage, led by an overall increase of catabolic processes and disturbance of anabolic processes. The joint cartilage is composed of a unique cell type, the chondrocyte, which is responsible for maintaining the tissue homeostasis in an environment mainly composed of water and biomolecules such as proteoglycans and collagen fibers. Many factors, including inflammation, may influence the shift from stable healthy cartilage towards a diseased state [2]. Regardless of the exact inducing mechanisms, during that transition, some of the chondrocytes enter a maturation process called hypertrophy [3, 4], leading to extracellular matrix (ECM) degradation, mineralization, and bone formation. This pathological phenomenon resembles the hypertrophic changes observed in the course of endochondral ossification, during growth and development [2, 5–8]. Therefore, controlling chondrocyte phenotype to prevent hypertrophic maturation has emerged as a potential therapeutic strategy to treat OA patients [7, 9]. Crucial in this approach is the understanding of the process of articular chondrocyte hypertrophy for the identification of key regulators as potential drug targets. Several factors have been associated to the promotion of this phenotypic shift, such as Indian hedgehog (IHH) and inflammatory signaling pathways [10]. Routes downstream of various growth factors are thought to be important in the control or disruption of chondrocyte homeostasis, such as the WNT and Bone morphogenic protein (BMP) pathways, the parathyroid hormone-related peptide (PTHrP), as well as the insulin-like growth factor (IGF)-I, fibroblast growth factors (FGF) and transforming growth factors (TGF)-B [9, 11, 12]. However, the interplay of intracellular pathways is highly intricate with extensive feedback loops, non-linear pathways, redundancy, and intertwining [11, 13, 14]. This complicates the intuitive prediction of what will happen in case of perturbation of a specific target. For example, it was observed that the in vitro activation of the WNT pathway with the WNT3A ligand and the inhibition of that same pathway with Dickkopf1 (DKK1), both induced a reduction of glycosaminoglycan rich ECM in human articular chondrocytes [15]. The fact that an activator and an inhibitor of the canonical WNT pathway both lead to the same outcome is surprising and highlights the intricacy of the underlying mechanisms. Hence, the ability to predict the effect of external perturbations and potential therapies requires a systemic view on the process and a holistic approach [13, 14]. We propose to unravel the complexity of these regulatory pathways and to rationalize the identification of potential drug targets via screening of (pairwise) perturbations by using a classical engineering approach, namely that of computer modeling and simulation. Contrary to in vitro and in vivo approaches, having a systemic view of the process using an in silico model allows to study the system numerically, in a cost- and time-efficient way, and with less ethical concerns. In addition, it allows to prioritize experiments, thereby refining the traditional funnel of drug target identification in the drug discovery process. The in silico approach starts with collecting intracellular biological mechanisms. It is necessary to identify the important individual components of the system and to know how they interact and influence each other. A computational model built on this information should generate results consistent with current knowledge but also allows to investigate questions that would lead to new insights into yet unexplored situations and interactions [16]. Such computational mechanistic approaches were already used in the past to identify influential candidates for cancer therapeutics [17], study the control cell fate decision [18], including in cartilage [19, 20] or prioritize personalized combination therapies [21]. In this study, we developed an in silico model of the regulatory intracellular network capturing articular chondrocyte phenotypic changes during OA. This network is built by combining knowledge-based modeling and data-driven approaches to ensure the mechanistic accuracy of the network whilst taking advantage of current automatic network reconstruction technologies. We characterized the intracellular states of the articular chondrocyte model and investigated consistency with known physiological behaviors, through mathematical model implementation and computer simulations. We subsequently used the model as a “virtual chondrocyte” to perform an in silico high throughput screening for early predictions of the best potential therapeutic conditions to be tested in wet lab experiments. Finally, we investigated in vitro the role of a selection of predicted factors in the regulation of the phenotypic change, for both single and combinations of factors to guide future therapeutic strategy discovery.
In order to build an articular chondrocyte model, we gathered known biological mechanisms from the literature into an activity flow graph or network (Fig. 1). In this graphical influence network, or activity flow representation, the nodes represent biological components (i.e., proteins or genes, see Additional file 1 Table S1 for definitions) important for chondrocyte biology [9]. Directed edges linking two nodes represent interactions or activating/inhibitory influences between source and target proteins or between transcription factors (TFs) and target genes. Information was collected and adapted from a previously published model of growth plate chondrocytes [22, 23] as well as from additional deep literature and database curation (see the “Methods” section). The annotations, descriptions, and cross-references for each network node and its interactions can be consulted in two interactive subnetworks, a protein signaling one and a gene regulatory one. The former goes from the growth factors binding their respective receptors down to the TF entering the nucleus. The latter represents a network of transcription factors regulating the expression of their target genes, coding for the corresponding proteins in the signaling network (Fig. 1). The two subnetworks are interconnected as each biological component is represented by a gene in the gene regulatory network (GRN) and its corresponding protein in the protein signaling network. Both subnetworks, forming the full network, can be regarded as an interactive knowledge base on chondrocyte signaling pathways, which is available through the online platform Cell Collective [24]. We refer the reader to that interactive knowledge base for the literature support of our network model (see links in the “Availability of data and materials” section), while the main pathways that were included are listed below. It is noteworthy that some biological factors were not regulated in one of the subnetworks (unknown transcriptional regulators or absence of post-translational modifications) and/or did not have downstream targets, but they were included in the knowledge base anyway (possibly resulting in disconnected—but annotated—nodes in that subnetwork) since regulatory relationships were described in the other subnetwork (see Additional file 2 for more details). Unless stated otherwise, all components of the network and of the in silico model are referred to with names in upper cases that designate the numerical variables, which represent neither the protein nor the gene but a product of both. Therefore, neither the gene nor the protein official nomenclatures are used. A list of correspondence between the numerical variable’s names and actual mouse gene names is available in Additional file 3 Table S2 and reported in the Cell Collective interactive networks, for information. The regulatory pathways represented in the network include the canonical and non-canonical WNT BMP pathways, the PTHrP and IHH pathways, as well as the IGF-I, FGFs, and TGF-B pathways since they are all reported to play a role in chondrocyte fate decisions [9, 25]. The influence of pro-inflammatory cytokines, such as interleukin 1beta (IL-1B) or Tumor Necrosis Factor Alpha (TNF-A), was summarized through a node labeled ‘cytokines’, in the network (Fig. 1). That node can signal through a single receptor (labeled ‘Rinfl’) that can activate well known downstream pathways such as the phosphatidylinositol 3-kinase (PI3K)/AKT axis, the nuclear factor kappa B (NFKB) pathway and mitogen-activated protein kinases (MAPK) pathways. These MAPKs in question include extracellular signal-regulated kinase 1&2 (ERK1/2), c-Jun N-terminal kinase (JNK), and P38. For each of the introduced pathways, we represented the downstream signaling cascades and known transcription factors as well as their target genes in the nucleus. Examples of important transcription factors that were included are the IHH signal transducer GLI2, the signal transducer and activator of transcription (STAT1), the transcription factor 7 (TCF), the myocyte enhancer factor 2C (MEF2C) as well as SRY-Box Transcription Factor (SOX9), a marker of differentiated healthy chondrocytes, and runt-related transcription factor 2 (RUNX2), a hypertrophy marker. All pathways in the model are highly interconnected as shown by Fig. 1. In total, there are 60 biological components in the network, which are listed in Additional file 3, Table S2. Each component accounts for a different biochemical entity including 8 growth factors, 8 receptors, 20 transcription factors, 4 ECM proteins, and 20 signaling molecules of another type. Combining the signal transduction and the gene regulatory networks into one connected network leads to a total of 264 direct or indirect biochemical interactions. Each node has on average 7.2 direct neighbors (i.e., directly connected node) and the average shortest path (i.e., smaller number of edges) to connect two nodes is 3.01. Hence, all nodes may virtually influence others by means of a couple of intermediary components, supporting the need to represent and study that network mathematically to conclude on key controllers.
We complemented the knowledge-based GRN with data-driven network inference, allowing for identification of de novo regulatory links, introduction of previously unstudied or undiscovered interactions and the reduction of the bias related to human literature curation. To that end, we generated an informative cross-platform merged dataset on mouse osteoarthritic cartilage. It was composed of 109 samples coming from 6 microarray experiments from which we selected a subset corresponding to the expression profile of 41 genes (see the “Methods” section and full list in Additional file 4 Data S1). The selected genes were the ones present in or closely related to the biological factors from the mechanistic model. The purpose was to identify potential regulatory interactions among those genes of interest without adding new variables (see the “Methods” section). The sub-dataset was normalized and corrected for batch effects originating from the differences in microarray platform technologies, as described in the “Methods” section. Our data indicate that the gene expression distribution was correctly normalized among the different microarrays (Fig. 2A). The principal component analysis (PCA) showed the inter-array variance was strongly reduced after merging and correction, thereby highlighting successful removal of the batch effect. As a quality control, we applied an unsupervised clustering method to the merged dataset to evaluate whether the biological information was still maintained after merging and correction. The resulting heatmap features the clustering of the 109 samples as well as their prior annotations as “OA” and “WT” (for wild-type or control samples) (Fig. 2B). See the “Methods” section and Additional file 4 Data S1 for annotations definition. While splitting the clustering result in three groups, we could identify an OA-like group and a WT-like group as well as a WT-like sample clustering alone. Based on these categories, 81% of the OA-tagged samples were correctly identified in the OA group whereas 56% of the WT-tagged samples were correctly located in the WT group. These accuracy levels are like the ones achieved when clustering each of the individual microarray experiments separately, before batch effect removal (Additional file 5 Fig. S1). We concluded that, despite the diversity of the technical platforms used in the assembled dataset, most of the variance was not due the arrays’ underlying technology but rather due to the biologically meaningful information usable as input for GRN reconstruction. In order to complement the GRN with new data-derived hypothetical interactions, we inferred genetic interactions out of this newly assembled dataset. Directed edges between nodes were added in the GRN to account for newly inferred regulatory interactions, as represented in Fig. 2C. We inferred regulatory interactions directed from TFs towards target genes by using three different algorithms to avoid algorithm-specific bias on the inferred edges (see the “Methods” section) [27]. Only interactions predicted by all the three algorithms were implemented in the gene regulation layer of the mechanistic model. The matrix of inferred interactions is reported for each algorithm and for the consensus result in Additional file 6 Data S2. In addition, the sign of the Spearman correlation score allowed us to define whether the interactions were genetic activations or inhibitions (Fig. 2C). The inferred interactions are reported with the correlation factors in Fig. 2D. Note that the machine learning methods predicted some TF as regulator of a given target gene even for pairs that had not a high correlation score (Fig. 2D). This showed that the ensemble of algorithms we used went beyond the simple correlation for inferring genetic interactions, relying on the concept that some form of covariation is implied by a causal relationship. The inferred interactions were included in the GRN part of the model for all subsequent analyses presented in this study.
We translated the regulatory network into mathematical equations in order to develop an executable numerical model of the articular cartilage chondrocyte. We used a semi-quantitative additive modeling formalism with priority classes as it allows to study large networks without requiring much information on kinetics parameters. Each node takes on a continuous value in the interval [0,1], representing the global functional activity of that node, defined as the multiplication of the gene expression level and the protein activation potential. That way, a protein cannot exert its function on downstream targets unless it is both expressed at the genetic level and activated/not blocked at the post-translational level (global functional activity > 0) (see the “Methods” section, Additional file 2: supplementary computational method and Additional file 1 Table S1). With the above-described numerical chondrocyte model, we studied the system free of fixed external cues to identify mathematically stable states that may emerge naturally. These stable states are also called attractors and they equate potentially existing cell phenotypes (see definition of attractors in Additional file 1 Table S1). They were evaluated using methodologies like those used with logical models [28]. Any initial state inputted into the system of equations is like a set of external stimuli that would trigger signal transduction inside a cell, eventually leading to a specific cell state. By randomly initializing the in silico model (see the “Methods” section for Monte Carlo analysis), we were able to explore possible model outcomes and we observed three emerging attractors that were singleton stable states. No cyclic attractors were obtained. Each of those states had a unique global activity profile for the 60 components as reported in Fig. 3A; the details of the protein activation and gene expression level for each component are available in Additional file 8 Data S3. Two of the attractor states that we found were biologically relevant, meaning they were comparable to existing chondrocyte phenotypes identified based on known cell state biomarkers such as type II and type X collagen (COL-II and COL-X, respectively), RUNX2 or SOX9. We identified one of the states as a normal healthy articular chondrocyte since markers such as SOX9, NKX.3.2 and COL-II were strongly active and expressed while RUNX2, COL-X, and matrix metalloproteinase 13 (MMP13) were inactive or not expressed (Fig. 3A and [8, 29–31]). In addition, the inflammation-associated variables had low activity. That correlates with what is known of real chondrocytes during homeostasis [32, 33]. The second state corresponded to a hypertrophic-like chondrocyte since factors such as RUNX2, COL-X, MMP13, and IHH were present or active (i.e., global activities equaling to 1) while SOX9 was not and COL-II and the proteoglycans were degraded and/or not expressed (functional activities nearly zero) (Fig. 3A and [4, 30, 31, 34–37]). In addition, the WNT and inflammation-related pathways were active (e.g., WNT =1, DC=0, βcatenin = 1, cytokines =1, NFκB = 1, TAK1= 0.76) (Fig 3A and [12, 33, 38]). The third state we found had nearly all protein activities close to zero, consequently, we couldn’t associate it with any specific phenotype as it was more likely a trivial mathematical solution. We named it the ‘None’ state as it was neither healthy nor hypertrophic Fig. 3A. The number of random initializations reaching an attractor during the Monte Carlo simulation gives a sense of the probably of reaching that state (Fig. 3B). Most of the random initial states led to the ‘None’ final state, reflecting the fact that most of these random initial values might very well be nonsensical from a chondrocyte biology perspective. In addition, about 21% of the initializations led to the healthy state and about 2% to the hypertrophic state (Fig. 3B). These attractors are the spontaneously emerging states arising from the proposed map of biochemical interactions. We also sampled random initial states while fixing seven growth factors at values that were physiologically relevant in either a normal healthy or a disease environment (profile A and B of Fig. 3B, respectively). Interestingly, when imposing the profile A to the growth factors during the Monte Carlo, the system only settled into the healthy state and when imposing profile B it settled into the hypertrophic state, thereby abolishing the “None” state (Fig. 3B). As a first step to support the validity of our model, we established a tool enabling to test in silico scenarios and assess whether the model could successfully recapitulate relevant physiological behaviors. We used this tool to test specific scenarios for which the expected outcome was known from literature or hypothesized from clinical observations. For instance, inflammation in the knee is one of the symptoms of OA and has been shown to be one of the drivers of cartilage degradation, possibly by pushing chondrocytes to undergo hypertrophy and produce matrix-degrading enzymes [39, 40]. For that reason, inflammation-related targets are subject to several investigations for potential OA therapies. Interestingly, in silico experiments with our model showed that blocking important transducers of inflammation such as the TGF-β-activating kinase (TAK1) or NFKB concomitantly to activating the PTHrP-related pathway could push a hypertrophic-like chondrocyte into transitioning towards a more healthy or anabolic state (Additional file 9 Data S4). Moreover, other studies have reported that the TGF-β pathway had a protective effect against inflammation [41–44], a scenario we evaluated with our model too. In silico mimicking the presence of inflammation in a healthy chondrocyte by forcing several inflammation related pathways of the model to be set at their highest values led to 100% of transitions towards the diseased hypertrophic state (Fig. 4A). This effect was partially rescued by concomitantly forcing the presence of TGF-β since 5.3% ±1.2 of the perturbations failed to exit the healthy state, thus, confirming in silico that TGF-β could have a protective effect against inflammation through the mechanisms present in the model. Nevertheless, the role of TGFβ in OA has been reported to be dual as this growth factor transduces signals in chondrocytes mainly via two receptors, the type I TGFβ receptor (ALK5) and the type II TGFβ receptor ALK1 [45]. They are involved in different intra-signaling routes and depending on which receptor is activated, the downstream-activated signals would be rather anabolic (ALK5) or catabolic (ALK1) [45] and impact chondrocyte maturation differentially [46]. Clinical observations reported that the ALK5/ALK1 balance decreased with age and in OA patients [47–49]. In silico simulations with our model showed that, roughly, the rescue by TGFβ was lost when the ratio between the receptors was forced to be (Fig. 4 A and Additional file 10 Fig. S2). For higher values of ALK1, ALK1 activity could be as low as 0.86*ALK5 and still show the loss of the TGF-β protective effect (Additional file 10 Fig. S2). This is in line with what was previously modeled [41], demonstrating that the decrease in that balance could explain why TGF-β loses its protective effect against inflammation in OA patients. Together, these results show the ability of the articular chondrocyte in silico model to behave in a physiologically relevant way and predict emerging effects qualitatively, highlighting the pro- or anti-hypertrophic nature of biological components in specific conditions. The aforementioned tool was further implemented through an executable App (available in GitHub [50]) allowing users, such as biologists, to easily test hypotheses by performing in silico experiments on the virtual chondrocyte (see interface Additional file 11 Fig. S3).
We next decided to exploit further the model by studying the effect of all possible perturbations of each component. Starting either from the healthy or the hypertrophic-like state, variables were individually activated or inhibited in a systematic manner. Over-activation of the FGF receptor 1 (FGFR1) or NFKB was the most potent condition to trigger a transition from the healthy towards the hypertrophic state. To a lesser extent, activation of the variables for ERK1/2 kinases, the AKT member of the PI3K/Akt axis, the RUNX2 transcription factor or JNK kinases also promoted this diseased transition (Fig. 4B). On the other hand, transition from the hypertrophic towards the healthy state was mainly triggered by forced activation of the TGFβ intracellular effector SMAD3 and partially brought about by activation of the SOX9 transcription factor, IGF-I and members of the PTHrP pathway (i.e., the protein kinase A (PKA), the PTHrP receptor (PPR) and PTHrP). This constitutes a state transition study that can be summarized in the Markov Chain representation (Fig. 4C) with the system’s overall probability of transitioning from one state to another under random single-node perturbations. Interestingly, despite the low probability of reaching the hypertrophic state in the random canalization (Fig. 3B), the transition study shows that the total probability of transitioning out of that hypertrophic state under random single-node perturbations is 0.11. So, in 89% of the cases, a single node perturbation will not affect this state. This means that even if the hypertrophic state is difficult to reach in the articular cartilage system, it is particularly robust to small single perturbations, and once the numerical system has reached that state, it is unlikely to escape from it, via transitioning to any other state, with a single targeting strategy. For this reason, we also systematically investigated the effect of all possible combinatorial perturbations of two constituents (or pairwise perturbations). The full screening amounted to tested conditions per state. 94% of the variance was explained by the first two components in the PCA reporting the effect (percentage of transitions to destination attractors) of the combinatorial treatments on a hypertrophic-like chondrocyte (Fig. 4D). We searched for the most potent conditions to retrieve the healthy state from a hypertrophic chondrocyte (defined as conditions with more than 70% of perturbations transitioning to the healthy state and less than 5% to the “None” state), which could point towards potential drug therapies for OA. Based on those results, the most efficient way to transition from a hypertrophic towards a healthy chondrocyte in the in silico model was with the up-regulation of SMAD3 in combination with activation or inhibition of numerous other targets such as inhibition of the inflammatory mediator NFKB, inhibition of the DLX5 transcription factor or activation of SOCS, a blocker of pro-inflammatory signals transduction (see the complete list in Additional file 9 Data S4). Activation of PKA/PPR in combination with inhibition of various targets, such as WNT or FGFR1, also seemed to decrease hypertrophy successfully in the model with 100% of transitions towards the healthy state (Additional file 9 Data S4). Therefore, the PKA/PPR axis and SMAD3 seemed to be “enablers” that could “unblock” the system, facilitating the effect of other relevant targeting treatments, in silico. Moreover, some predictions among the ones triggering more than 70% of transitions towards the healthy state, in Fig. 4D, did not include the two aforementioned enablers. For example, the up-regulation of ALK5 in combination with the down-regulating ALK1, the two receptors of TGF-B in the model, gave more than 90% of transitions towards the healthy state. Additionally, inhibition of the WNT pathway while activating ALK5 as well as activation of IGF-1 while activating the destruction complex (DC) involved in the WNT pathway allowed between 70 and 90% of transitions towards the healthy state, to mention but a few (Additional file 9 Data S4). So, the in silico model and associated screening algorithms predicted pairwise targeting conditions with a potential role against chondrocyte hypertrophy. Those predictions can be used as an indicator to guide further validation experiments.
We used ATDC5s, a chondrogenic murine cell line able to undergo hypertrophy as well as human OA chondrocytes, in order to in vitro validate the in silico findings obtained from the model. We measured the activity of alkaline phosphatase (ALP), an enzyme typically secreted during hypertrophy and participating to ECM mineralization. It showed that there was a positive linear correlation between the level of ALP activity in the medium and the expression level of the hypertrophic gene Col10a1 during the ATDC5 differentiation (Fig. 5B). Hypertrophy was further increased when supplementing the differentiation medium with Ihh, as expected (Fig. 5B). Indeed, this growth factor is known for its pro-hypertrophic effect on ATDC5 [51]. These results allowed us to evaluate ALP activity in the medium as a proxy for the level of hypertrophy, significantly increasing the throughput of the experimental set-up for testing small molecule treatments. This semi-high throughput system was used for validating the in silico predictions. In vitro experiments evidenced that treatments with Forskolin, an activator of PKA activity or with an activator of the BMP/SMAD3 pathway, were sufficient to prevent the increase in the activity of secreted ALP and thereby were sufficient to block hypertrophy in ATDC5s (Fig. 5C). Additionally, Forskolin treatment decreased COL10A1 gene expression when applied to primary human OA chondrocytes (p < 0.01) (Fig. 5C). Together these results corroborated the in silico predictions that activation of PKA was sufficient to block hypertrophic differentiation in chondrocytes (Fig. 5C, D). Additionally, several combinatorial treatments predicted by the in silico screening were tested in ATDC5 and compared to the corresponding individual treatments, to assess their efficacy. Additional file 9 file Data S4 describes the potent combinatorial conditions and their predictions that were selected for subsequent validation (PKA activation + FGFR1 inhibition, WNT inhibition + PKA activation, ERK1/2 inhibition + PKA activation, IGFIR + PKA activation, BMP inhibition + IGFI, BMP inhibition + PKA activation, ALK5 + ERK1/2 inhibition, HDAC4 activation + PKA activation). The chemical compounds used for each target are described in Additional file 12 Table S4. The effect of those combinatorial treatments was assessed through evaluating ALP activity in ATDC5. ALP activity normalized to total DNA content is reported by means of z-score for all pairwise and single conditions in Fig. 5D. The ALP activity in the medium during hypertrophic differentiation was lower than the corresponding controls in 6 out of the 8 predicted combinations. The two conditions for which the strongest effect was measured were the inhibition of FGFR1 combined with the activation of PKA (“Forskolin + PD161570,” p = 0.032) and the inhibition of BMP combined with PKA activation (“LDN-193189 + Forsk,” p = 0.014). These two combinations seemed to generate an added effect compared to the single drugs. Contrary to the in silico model predictions, treatment with exogenous IGF-I combined with BMP inhibition (“IGF-I + LDN-193189,” p= 0.105) or inhibition of ERK1/2 combined with PKA activation (“PD0325901 + Forsk,” p = 0.264) did not show a significantly lower hypertrophic level, based on ALP activity (Fig. 5D). Elaborating one of the conditions that showed the strongest response, PKA activation combined with FGFR1 inhibition, we compared the combinatorial effect with the corresponding single drug treatments. The combinatorial effect was greater than the one for either of the single drug, for both tested concentrations ratios (Fig. 6A). This suggests that activating PKA (resp. inhibiting FGFR1) would potentiate or enable the effect of FGFR1 inhibition (resp. PKA activation) by blocking or unblocking key pathways and maintaining the necessary constraints on the system. Dose curve relationships need to be established to confirm that hypothesis. Screening various values of functional activities of PKA and FGFR1 with the virtual chondrocyte and looking at the percentage of transitions out of the hypertrophic state, showed that a minimal level of PKA activity (namely about 0.4 on a scale from 0 to1) was required to achieve any positive effect with this combinatorial treatment (Fig. 6B). In addition, a gradient effect was observed suggesting that the lower the PKA activity, the more we needed to block FGFR1 to achieve an equivalent positive effect (Fig. 6B). In vitro validation confirmed this dose effect since the overall gradient shape was very comparable between the in silico and in vitro situation (Fig. 6B). Comparing the diagonal (combinations of two drugs) to the single drug ranges also highlighted a likely synergistic effect in decreasing hypertrophy between the two drugs, at the tested 1:5 ratio. Taken together, these results support that targeting the regulatory mechanisms at multiple points might be necessary to maintain a physiologically healthy state for a chondrocyte experiencing hypertrophy-inducing cues.
We report the construction of a mechanistic model of chondrocyte phenotype control in articular cartilage by combining a knowledge-based approach with a data-driven approach. We have leveraged decades of knowledge and data about chondrocyte regulatory pathways and osteoarthritis by integrating that information in a numerical predictive model. This model recapitulated physiologically relevant observations and predicted conditional effects resulting from intricate intracellular signaling. It offers the possibility to screen a large amount of (combinatorial) treatments and prioritize subsequent in vitro experiments for the identification of molecular drivers of chondrocyte phenotype and anti-hypertrophic drug targets. The in silico target perturbation screening and the experimental validation of our findings show the potential of in silico experiments to guide in vitro experiments for target discovery. With further validation, these predictions might form the basis for successful OA treatment. In particular, our study points towards a possible synergistic effect of PKA and FGFR1 targeting strategies to regulate chondrocyte hypotrophy. Several of our insights have implications both for the network modeling community and for cell and cartilage biology. We have first built an interactive intracellular network as an online knowledge base and as a reference support for our numerical model. Most network-based models rely on prior mechanistic knowledge. Even current state-of-the-art computational tools meant to reconstruct numerical models automatically from (high-throughput) data often require or offer the possibility to introduce prior knowledge [52]. Therefore, there is a high need for integrating and curating originally isolated pieces of knowledge, in a comprehensive way. However, biochemical information specifically related to cartilage and osteochondral systems are scarce in public and private pathway databases, in which cancer-related cell types tend to be over-represented. The network we provide along with this study details the prior mechanistic knowledge we have put in the model and that was predominantly chondrocyte or osteochondral cell type-specific, with a focus on articular chondrocytes and osteoarthritis. In our opinion, it is valuable not only for cartilage and OA researchers but also for modelers as it can serve as a basis to derive other models (e.g., ODE-based models) and answer alternative questions. Combining knowledge and data in a comprehensive network resulted in a systemic view of chondrocyte intracellular regulation. Many independent pieces of information have accumulated over the past decades and many databases have made curated pathways available. Most of the time in literature, these pathway descriptions stop after the (in)activation of the downstream transcription factors while the identity of the target genes downstream of these pathways is left obscure. In this study, our strategy was to complement the knowledge-based network graph with automatically inferred transcriptional regulations from transcriptomic data using machine learning methods. This usually requires a large multi-perturbed dataset but in absence of such dataset for chondrocytes, we reconstructed one by merging various arrays. Our strategy is supported by a previous study, which reported that equivalent informative data could be successfully achieved by assembling naturally occurring and experimentally generated phenotypic variations of a given cell type [53]. Even though relatively few inferred gene regulatory interactions were integrated in the mechanistic model due to the stringency of our selection strategy, we limited the risk of integrating false positive predictions. This data-driven approach allowed us to take advantage of automatic network reconstruction technologies, which are becoming more and more the standard in the state of the art [54, 55]. We limited the inference to our genes of interest to gain knowledge on regulatory interactions without adding new nodes in the network, which would add more unknowns. Nevertheless, an alternative strategy could be to infer interactions between all genes from the dataset and integrate interactions involving our genes of interest, new nodes could be added if relevant feedback loops or non-linear paths were observed with some genes of interest. The translation of that network into mathematical rules was then further studied computationally and enabled the prediction of the overall effect of each network component on the virtual chondrocyte. Our perturbation screening approach helped to discover influencer nodes based on the mathematical dynamics, similarly to what had been previously proposed for other diseases [17]. Important to mention is that no information was put in the model that directly made one target prevail over the others in being pro- or anti-hypertrophic. The information that fed the mathematical model was about activating or inhibitory influences of one molecule on another molecule, a complex of molecules, or a pathway although some edges in the network represented indirect links for the sake of simplification. The advantage of the discrete semi-quantitative mathematical formalism we employed lies in its ability to reproduce qualitative dynamics using only the activating or inhibitory nature of interactions and additive rules, without any prerequisite about kinetics. This characteristic makes that method perfectly suitable for a large network such as the one of this study as the size of a model and quantity of unknow parameters should be subject to a tradeoff and is a function of the model purpose [56]. Many biostatistics and machine learning methods that make drug efficacy-related predictions solely based on omics data still lack predictability and interpretability [57]. In contrast, the numerical approach presented in this study has the advantage of providing mechanistic evidence supporting the predicted effects, thereby increasing the mechanistic interpretability. Generally, the validation of an in silico model to answer questions in a specific context may be achieved in two ways: either by showing its ability to predict non-linear effects that were already reported in literature or by confirming predictions with new experiments. Before carrying out new validation experiments, the relevance of our model to answer osteoarthritis-related questions was investigated by its ability to recapitulate earlier described behavior such as the changing role of TGFβ signaling in presence of inflammatory stimuli. A recent computational study reporting a quantitative (ODE-based) time-dependent model of cartilage breakdown [41] provided more detailed and complexed kinetics underlying the dual role of TGF-β. The fact that our model could lead to equivalent conclusions shows that the more simplifying semi-quantitative formalism that we employed could suffice to capture non-straightforward effects, thereby supporting its credibility for subsequent predictions. The Monte Carlo analysis of the system without constraints showed that the healthy state was more easily reached than the hypertrophic one. This corroborates with the situation for normal articular cartilage in which chondrocyte hypertrophy does not spontaneously occur unless the homeostasis is disturbed [58]. A great part of the state space was occupied by the “None” attractor that is most likely a trivial solution towards which the system converges when an initial state or a trajectory is too far away from a feasible biological state to meet all the constraints imposed by the equations. An analysis of the ensemble of initial states reaching the “None” state, could possibly spotlight initial states that are unlikely to happen in an in vivo physiological environment. Conversely, the analysis of canalization under imposed growth factor profiles has suggested that restricting ourselves to initializing growth factor’s values within biologically “feasible” ranges can decrease prevalence of the “None” state, although the boundaries of those feasible ranges could be further explored. Nevertheless, in the scope of finding potential therapeutic targets, we are less interested in the random canalization of the states than in their robustness to perturbations and capacity of transitioning. This study also suggests that chondrocyte phenotypes are not so sensitive to small, single-factor perturbations. Even though the resistance of the current in silico model to single factor perturbations might, in part, be due the omission of some parts of the real world system’s complexity. This could be subject to further experimental validation in chondrocytes. This trait of robustness to small environmental variations is often considered as a fundamental and ubiquitous trait of biological systems [59]. This trait allows systems to function in noisy environments [60] and systems biologists have theorized that disease may establish its own robustness, in some cases [61]. In line with that, the Markov chain in our computational model indicates that, for articular chondrocytes, the probability to change phenotype once it has been reached is rather low. Indeed, due to the very intricate interplay of molecules, it is likely that some pathways play redundant roles and that several factors should be targeted simultaneously to ‘unlock’ the system. Moreover, the in silico pairwise perturbation screening confirms that targeting at least two components concomitantly increases the chance of unlocking the hypertrophic commitment. Overall, as its in vivo counterpart, the in silico articular chondrocyte is unlikely to display hypertrophic signs in normal conditions or even sometimes under inducing treatment [10], but once the hypertrophic transition has been initiated, it is rather difficult to escape that fate. Another important outcome is the approach to validate certain predictions on previously unreported conditions. In practice, as reverting a hypertrophic chondrocyte back to a healthy state has never been observed experimentally, we hypothesize that the in silico predicted conditions are, at least, more likely to block hypertrophy. The in vitro results that we present here support this hypothesis and open new routes for further testing the suggested combinatorial conditions. Especially, the combination of PKA activation with FGFR1 inhibition is highlighted as good candidate treatment by our integrated in silico-in vitro approach. Future research should focus on the further validation of this result, for instance through an in vitro dose curve relationships study and in vivo testing. Activation of the PThrP pathway, to which PKA belongs, has already been reported to be rather anti-hypertrophic for growth plate chondrocytes [62]. Similarly, genetic inhibition of the Fgfr1 gene in mouse knee cartilage has been shown to attenuate the degeneration of articular cartilage in mice [63]. However, to our knowledge, this combination has never been investigated nor reported before for its synergistic potential against hypertrophy, cartilage degradation or OA. Some predictions (IGF-1 combined with BMP inhibition, and inhibition of ERK1/2 combined with PKA activation) could not be validated experimentally. This could be explained by several factors, such as the limitations of the in vitro model (ALP measurement in ATDC5), the omission of important regulatory mechanisms, or inaccurate assumptions in the in silico model, possibly pointing at necessary corrections in the regulatory network. Importantly, here we propose a strategy in which in silico predictions are used in an exploratory discovery phase. Hence, reducing the number of false positives is less important than increasing the true positives and reducing the false negatives - which hampers the discovery process. Validating in silico predictions for drug target discovery experimentally is a challenging task. This is especially true when numerical high-throughput screenings, enabled by high computing power, generate a large number of predictions. We leveraged the evaluation of secreted ALP activity in ATDC5 cell line as a semi-high throughput read-out for validating the in silico predictions. Indeed, it served as a convenient experimental system to assess hypertrophy modulation under many screened perturbations [64]. Nevertheless, it is important to note that this mouse cell line is cultured in isolation from a physiological environment. Therefore, we propose to use it within an in silico-in vitro pipeline that would act like a funnel with sequential filters for prioritization of candidate drug targets. As shown in our study the main hits can then be further evaluated in more detail using human OA chondrocytes. The next step could be to test the filtered conditions in vivo to evaluate their effect on the disease progression. We acknowledge that our study has some limitations. A drawback is the absence of a gold standard to assess the precision of the data-driven network inference. To mitigate this, we used a consensus approach by only integrating predictions made by three different machine learning methods. This reduced the number of interactions we would integrate but it alleviated each method’s weakness and reinforced the strengths of the predictions, as previously proved [27]. Secondly, the type of mathematical model we employed comprises almost no numerical parameters but the main one, the saturation constant, was assigned an arbitrary value based on a previous study [22]. This constant determines how fast a protein activity or gene expression can saturate to the maximal value depending on the amount of excess positive and negative upstream interactions. It intervenes in the weight of interactions and changing its value might slightly affect the influence of the network’s constituents on the system. Moreover, we did not experimentally verify the target specificity and dosing regimen for the small molecules employed for our in vitro validation since it was out of the scope of this study. However, those small molecules were selected based on their previously reported and well-known in vitro action on our targets of interest. Finally, the mechanisms integrated in the network model include a certain number of assumptions and regulatory relationships that represent the current state of the knowledge; however, those assumptions and regulations could still be updated and refined as knowledge grow.
This study is a proof-of-concept to showcase how an in silico-in vitro integrated approach can suggest single and combinatorial target perturbations affecting the hypertrophic transition and help to prioritize experiments for therapeutic target discovery in OA research. In that sense, our model offers the possibility to make hypotheses on the pro- or anti-hypertrophic nature of biochemical pathways and targets based on strict mathematical rules describing the intricate network connectivity. We are convinced that this type of approach can guide the process of therapy development from basic understanding to target selection early in the drug discovery pipeline while reducing time and cost of experiments as well as the use of animal models in early stages of drug discovery. Our study highlights targets - such the concomitant activation of PKA and inhibition of FGFR1 or BMP - that deserve additional investigation. With further validation, these conditions might form the basis of a successful OA treatment. Furthermore, investigating the effect of a new target that was not present in the current model should be possible by solely informing the model on how the target interacts with and connects to the rest of the network. Typical information on the nature of the upstream activators and inhibitors of the protein’s functionality, the nature of downstream proteins modulated by the target under scrutiny as well as information about its transcriptional regulators, from DNA binding assays for instance or reverse engineered from data would be needed. Ideally, this information should be as exhaustive as possible based on the current state-of-the-art knowledge, while hypothetical connections could be investigated and compared based on the simulated target’s effect. Finally, as scientific research is making progress in the identification OA disease subgroups based on molecular markers and clinical phenotypes [65, 66], we foresee data-informed mechanistic models can become more and more patient-type specific. For instance, such knowledge-based network model could serve as a prior and be further optimized with engineering approaches, adjusting the network topology and/or interaction weights, in order to fit chondrocyte baseline profiles of typical patient subgroups. The resulting network models and the effect of target perturbations could be compared across the different patient type-specific models. In conclusion, this work provides a virtual articular chondrocyte in the form of a signal transduction interactive knowledge base and of an executable computational model. The demonstrated in silico-in vitro strategy opens new routes for studying OA and discovering targeting therapies by refining the early stages of drug discovery.
The knowledge-derived networks were built by incorporating fine-grained mechanistic knowledge about signaling pathways and transcriptional regulations. A previously published model of chondrocyte differentiation in the growth plate [23] was used as a basis and was adapted and completed through literature curation of decades of knowledge about articular chondrocyte and osteoarthritis. Reference sources for the experimental evidence (binding assays, clinical observations, etc.) were mostly from mouse and human origin. References used chondrocytes (or related cell lines) and predominantly involved direct protein or promoter binding information, curated pathways, or observed phenomenon in cartilage during homeostasis and disease. All references and mechanism descriptions are available through the interactive networks in the online platform Cell Collective [24]. See the links for the chondrocyte knowledge base in the “Availability of data and materials” section (separated in two subnetworks). Additionally, some gene regulatory interactions were automatically identified with machine learning algorithms using gene expression multi-perturbed data as input. Such informative dataset was achieved by merging six published microarray datasets of mouse articular cartilage (GSE26475, GSE33754, GSE79239, GSE33656, GSE53857, GSE45793). Each dataset had a control or wild type group and an OA-like group, which was either using genetically modified mouse spontaneously developing OA or a DMM-induced OA mouse model. Importantly, the initial annotations provided by the authors were not necessarily “OA” or “WT” but rather experiment-specific annotations such as ‘DMM’ or ‘SHAM operated’. Therefore, the binary annotation as OA-like or WT-like samples was established by hand for the purpose of this study according to the original data annotations published on GEO, as reported Data S1. The datasets were merged applying an in-house developed pipeline adapted from previously published methods [67]. Briefly, the datasets were preprocessed first in a platform-specific way prior to assembly as a merged dataset. For the purpose of this study, a sub-dataset was created by restricting ourselves to the genetic profile of 74 genes of interest, being the ones present or closely related to the biological factors from the mechanistic model as listed in Additional file 4: Data S1. This allowed finding new regulatory relationships without adding new variables to limit the model’s size. 41 genes, out of the 5470 from the full dataset, matched our list of interest, based on the Ensembl IDs, and the associated genetic profiles constituted the final sub-dataset (41 genes (or variables) × 109 samples (or observations)). This dataset was quantile normalized and the batch effects were removed through the ComBat algorithm based on Bayes methods [68]. Inference was performed for regulatory interactions within the aforementioned subset of genes. This was achieved by employing three different algorithms, being ARACNE, TIGRESS, and GENIE3 [69–71]; the final retained network was a consensus network of the three algorithms. Typically, an interaction between a transcription factor and a gene was kept if it was present in the three methods’ results. An interaction was called present for a method if the interaction score was higher than a certain threshold for that method. This threshold was set to m − σ, where m is the average score for the given method and σ is the standard deviation of the scores. The validity of the inferred interactions was corroborated by searching the GeneCards database [26]. When one or several binding sites for the source gene (transcription factor) in the enhancer region of the target was found with GeneCards, one of the GeneHancer ID was reported (Fig. 2) and the full list with the exact gene IDs that were queried is in Additional file 7: Table S3. Unless specified, all the aforementioned microarray data analytics were accomplished using the R computational environment (v.3.2.2). Topological parameter analysis of the final network was carried out with the Network Analyzer plug-in of the Cytoscape software v3.7.2 (https://cytoscape.org/) [72].
The information contained in the network was translated into mathematical equations through an additive formalism with 2 priority classes to distinguish between fast and slow reactions, the importance of which was repeatedly highlighted [73–75]. This additive formalism resembles the Boolean threshold networks [76] and was previously implemented with priority classes as described by Kerkhofs and Geris [22]. In this formalism, the biological components were represented by variables evolving over pseudo-time steps. The model is semi-quantitative since the variables could take on a continuous activity value between 0 and 1. The evolution of all variables (proteins or genes) was defined by the sum of the upstream activating variables and the subtraction of the upstream inhibitory variables from the network (see definitions in Additional file 1 Table S1), in some exceptional cases (particular mechanisms) a product of several regulators was used (see Additional file 2, Supplementary computational method). Biological influences could happen at two time scales, reflecting the priority classes: reactions related to slow biological processes, such as gene expression, mRNA, or protein production, were referred to as slow reactions (lower priority), and those related to fast processes, such as protein activation or inhibition, were referred to as fast reactions (higher priority) (see Additional file 3 Table S2 for the definition of fast and slow reactions and variable). A formal description of the mathematical system underlying the model as well as the full list of equations is available in Additional file 2. The asymptotic solutions were evaluated with a Monte Carlo simulation procedure, like methods employed for logical models [28, 75]. When running a simulation (also see the section below on Monte Carlo analysis), an initial value in the interval [0,1] was assigned to each variable. Every simulation step, the sub-variables (see definition in Additional file 1 Table S1) were updated asynchronously according to the rules given in the equations and following the priority classes, in such a way that fast reactions were always updated before the slow reactions. The order in which variables were updated within a priority class was random, hereby recapitulating the stochasticity inherent to any biological system. See Additional file 13 Fig. S4 and Additional file 14 S5 for graphical explanations about the algorithm and simulation scheme on a reduced illustrative example network. A stable state (definition in Additional file 1 Table S1) was reached whenever the next iteration step did not bring any change for any of the variables up to a tolerance 10−2. In other words, when initializing the system at a random point, it was considered converged when the relations detailed by the system of equations were fulfilled up to a tolerance of 10−2. Thanks to the stochasticity of the model, the same initial input could lead to different types of stable states. Therefore, all computational results of this paper were computed 3 times and standard deviations were evaluated. All implementations and simulations were carried with the MathWorks® suite, MATLAB (2018b).
A Monte Carlo canalization estimated the nature of the attractors and their reachability, given the regulatory network provided in the equations. In short, all variables were initialized 10.000 times with random values in the interval [0,1] simulations converged towards several attractors (only singleton attractors were observed). We considered that two simulations reached the same unique stable state when the absolute difference between both final states was less than a tolerance set to 10-2 for all the 60 components since such a difference. The number of initializations reaching each final state was computed and reported in terms of percentage of initial states. This number gives a sense of the probability of reaching the state for the unperturbed system, i.e., without constraints [28, 75]. The number of initializations (10.000) was considered sufficient to estimate the state space of the system as higher numbers had little influence on the canalization results (see sensitivity analysis Additional file 15 Fig. S6). Then, we performed two other Monte Carlo canalizations in which all variables were randomly initialized, except for seven growth factors that were fixed at values meant to represent a normal healthy or a disease environment (profile A and B of Fig. 3B, respectively). The networkD3 package from R was used to produce the Sankey diagram for the visualization of the canalization results.
By essence, the attractors are stable, meaning that variables cannot evolve anymore. These states may however be escaped by forcing, computationally, the value of one or several variables to change. Such a perturbation was imposed for a fixed number of computational iterations, after which the system was left to evolve freely, thereby accounting for the fact that chemical treatments affect biological systems for a finite period of time. The duration of the perturbation was set to 1000 time steps as the perturbed state did not take more than 200 time-steps to be reached, on average and going further than 1000 time-steps would not induce further changes in the result. Imposing a perturbation on a stable state forces the system to evolve again, following rules imposed by the equations, and eventually settle down in the same initial or a new attractor (see convergence description above). A representative example of the pseudo-time evolution of representative variables (RUNX2, SOX9, MMPs, Collagen, etc.) after an input perturbation (PKA activation + FGFR1 inhibition) is provided in Additional file 16 Fig. S7 to illustrate the choice of the perturbation duration. The different in silico scenarios or treatment experiments that were tested amounted to perturbing one or several variables, from the healthy or the diseased hypertrophic state and assessing the effect of that perturbation on the state stability. Variables were perturbed by forcing their global activity value to be 0 or 1 for inhibition or activation respectively. Imposing intermediary values between 0 and 1 was also done for some specific questions in which extreme values would be unlikely, such as varying the ratio between different membrane receptors. Each perturbed condition was imposed starting from the relevant initial stable states (healthy or diseased) and the nature of the final state to which the perturbation led after simulation was documented. We considered that the tested perturbation triggered a state transition when the final state was different from the initial one. Given the stochastic nature of the model, the same perturbation could trigger a different outcome if simulated a second time, therefore the same perturbation was repeated 100 times and we reported the percentage of transitions towards each of the possible target states. Standard deviation in the percentage of transitions was assessed by repeating that experiment 3 times. Due to the computational cost associated with the systematic screening of all possible pairwise perturbations, (for each pair there are 4 possible pairwise conditions to be tested either from the healthy state or from the hypertrophic state. The independent simulations were run in parallel using high-performance computing infrastructure of the KU Leuven (Vlaams Supercomputer Centrum). The selection of potent conditions against hypertrophy that could be validated experimentally was done in two main steps. First, we automatically selected combinatorial conditions for which at least 70% of the perturbations triggered a transition from the hypertrophic state towards the healthy one. Among them, we focused on those conditions with more than 70% of transitions towards the healthy state but less than 5% towards the ‘None’ state as a first discriminatory factor. Conditions were classified by ranges of percentages of transitions towards the healthy state (100-90%, 89-80%, 79-70%), see Additional file 9 Data S4. Second, among those potent conditions, some were further selected for their druggability and their ease to be tested in a simple in vitro system. The selection criteria also involved additional elements such as the readily availability of the necessary small molecules to target the predicted component in the expected way, the availability of literature to define appropriate concentrations, and the variety of combinatorial conditions tested. For instance, conditions involving the modulation of transcription factors were not considered for the in vitro validation since no small molecule treatment could directly affect transcription factor activity, or for conditions that were too similar (e.g., PKA activation+ ERK1/2 inhibition vs. PPR activation + ERK1/2 inhibition, PPR being directly upstream of PKA in the signaling network) only the one of the two was selected.
The validation of our in silico predicted treatments required in vitro testing with small molecules. This was performed with ATDC5, a mouse chondroprogenitor cell line obtained from the Riken Biological Ressource Center. Cells were cultured in 2D in proliferation medium containing DMEM/F12 (ThermoFisher, UK), 5% Fetal Bovine Serum (Biowest, Belgium) and 1% antibiotic/antimycotic (Gibco, ThermoFisher Scientific). Chondrogenic differentiation was induced by plating the cells at 6,400 cell/cm2 in proliferation medium for 24h, followed by changing the medium to differentiation medium, being proliferation medium supplemented with 10 μg/ml insulin (Sigma-Aldrich), 10 μg/ml transferrin (Sigma-Aldrich) and 30 nM sodium selenite (Sigma-Aldrich). Cells were incubated in a humid environment at 37°C, 5% CO2 and differentiation medium was refreshed every other day for the first 10 days, and every day after the 10th day, for longer experiments (i.e., Fig. 5B). Supernatant medium was taken for ALP activity assay and cells were harvested for DNA quantification (0.05% Triton-X reagent). ALP activity was reported relatively to the total DNA quantity to alleviate potential variation in cell number. To study the correlation between Col10a1 expression and secreted ALP activity, cells were differentiated for 14 days with or without Ihh supplement (150ng/ml, R&D Systems Europe LTD). The cells were harvested for RNA isolation on days 0, 7, 9, 12, and 14 during differentiation (TRIzol reagent; Thermo Fisher Scientific), in addition to the ALP activity assay and DNA quantification. To assess the effect of small molecule and growth factor treatments on hypertrophic differentiation, cells were treated on day 8 of ATDC5 chondrogenic differentiation for readout at day 9. Cells were treated with one or a combination of the following compounds: Forskolin (1μM, Axon Medchem), Recombinant Human/Mouse/Rat Activin A Protein (100ng/ml, R&D Systems), Recombinant Mouse IGF-I/IGF-1 Protein (10ng/ml, R&D Systems), Transforming Growth Factor (TGF)β1 (10ng/ml, PreproTech), PD0325901 ( 1μM, Axon Medchem), PD161570 (1μM, Axon Medchem), ITSA1 (50μM, Chembridge), LDN-193189 (0.5μM, Axon Medchem), LY294002 (20μM, Axon Medchem) and IWP2 (2μM, Stem cell technology). ALP activity in treated conditions is expressed in terms of fold change with respect to the control medium with the appropriate amount of DMSO, which was used as a solvent for most small molecules. Four types of control media were used throughout this study due to sparse solubilities of the compounds. The control medium was with 0.02% DMSO (Medium1) for most treatments, with 0.1% DMSO (Medium2) for the ITSA1-related conditions, without DMSO) for the ActivinA treatment (Fig. 5), and with 0.0375% DMSO for the Forskolin/PD161570 synergy study and dose screening (Fig. 6).
Human articular cartilage was obtained with implicit consent as waste material from patients undergoing total knee replacement surgery. This protocol was approved by the medical ethical committee of the Erasmus MC, University Medical Center, Rotterdam, protocol number MEC-2004-322. To isolate chondrocytes, cartilage chips were subjected to protease (2 mg/ml, Sigma-Aldrich) for 2 h followed by overnight digestion with 1.5 mg/ml collagenase B (Roche Diagnostics, Switzerland) in Dulbecco’s modified Eagle’s medium (DMEM) high glucose supplemented with 10% fetal bovine serum. Single cell suspension was obtained by filtrating the cellular solution by a 100 μm filter. The isolated chondrocytes were expanded in monolayer at a seeding density of 7500 cells/cm2 in DMEM high glucose supplemented with 10% fetal bovine serum, 50 μg/ml gentamicin, and 1.5 μg/ml fungizone (Gibco, Grand Island, NY, USA). Upon approximately 80% confluency cells were trypsinized and reseeded at 7,500 cells/cm2. Cells were used for experiments after three passages. Redifferentiation of articular chondrocytes was performed using a 3D alginate bead culture model. For preparation of alginate beads, chondrocytes after three passages in monolayer were re-suspended in 1.2% (w/v) low viscosity alginate (Kelton LV alginate, Kelko Co, San Diego, CA, USA) in 0.9% NaCl (Sigma-Aldrich) at a concentration of 4 × 106 cells/mL. Beads were made by dripping the cell-alginate suspension in 105 mM CaCl2 (Sigma-Aldrich) through a 22-gauge needle. Beads were washed with 0.9% NaCl and DMEM low glucose. Beads with a size that deviated from the average after a visual inspection were not included in the experiment. Re-differentiation of chondrocytes was performed in a 24-well plate (BD Falcon) for two weeks in 100 μL/bead DMEM low glucose supplemented with 1% ITS fetal (Biosciences), 10 ng/ml transforming growth factor beta one (TGFβ1, recombinant human, R&D systems) 25 μg/mL l-ascorbic acid 2-phosphate (Sigma-Aldrich), 50 μg/ml gentamicin, and 1.5 μg/mL fungizone (both Gibco). After two weeks, TGFβ1 was no longer added to the medium and cells were cultured with and without 1μM of Forkosolin for 24h. Each experiment was performed with cells derived from 4 OA donors (3 Females, 1 Male, 65 ± 6 years), in triplicates.
Gene expression of Col10a1 in ATDC5 experiments was evaluated by RT-PCR. For RNA isolation, chloroform was added to the TRIzol samples (TRIzol 5: Chlororform 1), which were subsequently centrifuged for 15min at 15,000 rpm (i.e., RCF = 218849) and 4°C. RNA was isolated by collecting the aqueous phase and precipitated with isopropanol (aqueous phase 1: ispropanol 1) for 30min at -80°C. After centrifugation at 15,000 rpm (i.e., RCF = 218849) and 4°C for 30 min, supernatant was removed, and the resulting pellet was washed with 80% Ethanol. RNA pellets were dried for 10min in desiccator and dissolved in 15μl RNase-free water. Finally, RNA content and purity was determined with Nanodrop. RNA was converted to cDNA with the Revert Aid H Minus First strand cDNA synthesis kit (Thermo Scientific) according to the manufacturer’s protocols. Quantification of gene expression was done using Syber Select Master Mix (Applied Biosystems) adding 400nM forward and reverse oligonucleotides primers (primer sequences available in [77]). The StepOne Plus System (Applied Biosystems) was used for amplification using the following protocol: denaturation cycle at 95°C for 10min followed by 40 cycles of amplification (15 s 95°C and 1 min 60°C), followed by a melting curve. Expression levels were analyzed using the 2−ΔCt method and normalized for the expression of the reference gene Hprt. This housekeeping (HK) gene was determined after verification of multiple HK genes and selecting the one that remained most constant throughout the procedure.
Alginate beads were dissolved using citrate buffer, centrifuged at 200g and the pellet was resuspended in RLT (Qiagen, Hilden, Germany) buffer containing 1% beta-mercaptoethanol for RNA isolation. mRNA isolation was performed according to the manufacturer’s protocol utilizing the RNeasy Column system (Qiagen, Hilden, Germany). The RNA concentration was determined using a NanoDrop spectrophotometer (Isogen Life Science, Utrecht, the Netherlands). 0.5 μg RNA was used for cDNA synthesis following the protocol of the manufacturer of the RevertAid First Strand cDNA kit (Thermo Fisher Scientific, Waltham, MA, USA). qPCR was performed on a Bio-Rad CFX96 Real-Time PCR Detection System (Bio-Rad) to assess gene expression, Collagen type 10 (COL10A1), and Glyceraldehyde-3-phosphate dehydrogenase (GAPDH), which was found stable and therefore used as reference gene. Data were analyzed by the ΔΔCt method and normalized to the expression of GAPDH of each condition and compared to the corresponding gene expression in the control groups. Data and primers are available in [77].
Enzymatic activity of secreted ALP in the supernatant medium of ATDC5 cultures was determined in a colorimetric assay as previously described [78]. Briefly, ALP activity was determined in flat-bottom 96-well plates (Sigma-Aldrich, CAT M9410) containing assay buffer (1.5 M Tris-HCl, pH 9.0, 1 mM MgCl2; 7.5 mM p-nitrophenyl phosphate). The ALP activity was assessed as a function of formed nitrophenyl phosphate (pNp), the reaction-colored product, which was measured by spectrophotometry at 405nM after 30min of reaction. The reaction was stopped with 1M NaOH Stop Buffer. A calibration curve containing an increasing concentration of pNp served to determine the absolute amount of ALP-generated pNp. Sample values were normalized to total DNA amount, expressed as μmol of pNp/mg of DNA, and reported as fold change with respect to the relevant control medium.
After harvesting the cells with 350μl of 0.05% TritonX-100, samples were vortexed and frozen at -80°C for further processing. Samples were sonicated for 5 s, centrifuged for 1′ at 13,000 rpm (i.e. RCF= 164,380) and 200μl of the supernatant was harvested. Samples were diluted with a factor 1/10 with distilled water then DNA content was measured with the Qubit 3.0 fluorometer (Life Technologies). Qubit dsDNA HS (high sensitivity, 0.2 to 500ng) Assay Kit was used according to the manufacturer’s protocols; a sample volume of 5μl was added to 195μl of a Qubit working solution.
In general, one-tailed statistical tests were used to analyze in vitro results and relate them to expected outcomes from the in silico model’s predictions since the model predicts a directionality of the outcomes. Average effect of Forskolin and Activin treatments were compared to control in one representative experiment with 3 replicates thanks to a one-tailed unpaired t-test with Welch’s correction, in Fig. 5C. Effect of Forskolin treatment in OA chondrocytes from human donors was performed in triplicates for 4 donors. Comparison of the average effect with the control is done with a linear-mixed effect model to account for donor variability. Graphical visualization and statistical analyses for the semi-high throughput small molecule screening (Fig. 5D) were performed by modifying the BraDiPluS package from the Saez Lab [79]. More precisely, the probability that a treated condition resulted in a lower ALP activity z-score than the corresponding control in the ATDC5 screening was estimated with a one-tailed Wilcoxon rank-sum test (the in vitro screening dataset was not normally distributed based on the Shapiro-Wilk test), with Benjamini–Hochberg correction for multiple testing. Three independent experiments were performed for each condition, and treatment effects were assessed in triplicates, in each experiment (or run). Probabilities from the independent runs were combined with Fisher’s method using the combine.test function from the survcomp R package. A treated condition was considered lower than the control for p-value < 0.05 and z-score <0.
Additional file 1: Table S1. Definitions. Definitions of technical terms pertaining to biological signaling network modeling with a semi-quantitative additive formalism.Additional file 2. Supplementary computational method. Equations, mathematical framework and justification of deviation from the general rule in the equations.Additional file 3: Table S2. List of variables and mouse genes correspondence. All mathematical variables and the corresponding node in the network have names written in upper cases and do not reflect the official human or mouse nomenclatures. To relate those variables to actual genes more easily, we provide this table of correspondence. A related mouse gene name and NCBI ID is indicated for each variable. Nevertheless, this is not exhaustive since some variables represent a group of factors or a family of ligands rather than a single factor.Additional file 4: Data S1. Subset of genes for gene expression profiles and description of GEO datasets with manual binary annotations.Additional file 5: Fig. S1. Individual dataset clustering and heatmap. The microarray sub-datasets, used in the merged dataset, were also investigated individually. They were pre-processed (i.e. processing steps before merging and correcting for batch effect) in the same way as the merged dataset but the unsupervised clustering analysis was done on each sub-dataset separately in order to compare the biological or OA related information content before and after the data merging. The heatmaps show the expression profile of the same list of genes of interest than for the merged dataset (see Additional file 4, Data S1.). These expression datasets were submitted to unsupervised clustering with the Euclidean distance method and the Complete aggregation method in R thanks to the heatmap3 function from the Github repository https://github.com/obigriffith/biostar-tutorials/tree/master/Heatmaps . The headers indicate the GEO accession numbers of the 6 original datasets. Samples labeled as ‘WT’, for wild type, are in green, samples labeled as ‘OA’, for osteoarthritis, in red. The pre-labelling is the same as for the merged dataset (see Additional file 4, Data S1). When applicable, a grey scale indicate the time points (w stands for weeks, in the legend). For some datasets (e.g. GSE26475, GSE33656, GSE53857) the OA and the WT samples are well separated in different clusters, while for other the separation is not so clear. For instance, in GSE45793 the variance due the time point 6weeks is greater than the OA induced variance, while for weeks 1 and 2 OA and WT samples are well separated due to the OA condition.Additional file 6: Data S2. GRN inference and validation. (Microsoft Excel Worksheet). The GRN inference with the 3 algorithms and the consensus matrix are reported. We have set stringent rules and only interactions that were inferred by the 3 algorithms were included in the mechanistic model: those interactions are all reported in Fig. 2D. The validity of the inferred interactions was investigated by searching the GeneCard database. The relevant GeneHancer IDs are reported when supportive evidence was found.Additional file 7: Table S3. Validation of transcriptional data-inferred regulatory interactions. Transcriptional interactions were inferred from the merged OA dataset. Inference was run with three algorithms and only interactions that were present in the results of the three algorithms (GENIE3, ARACNE, TIGRESS) were considered as additions to the model. An interaction was considered present for one algorithm if it scored higher than a threshold defined as the difference between the mean and standard deviation of all scores (Additional file 6, Data S2 ). Inferred interactions were validated whenever possible by looking for binding sites of the source transcription factor in the enhancer region of the target gene with GeneCards. The complete list of GeneHancer IDs provided by GeneCards are reported in the last column. ‘NA’ indicate that no binding site was found. The exact gene names and gene IDs that were queried in GeneCards are reported in the first columns.Additional file 8: Data S3. Attractors complete protein and gene expression profiles. (Microsoft Excel Worksheet). The predicted profiles of the 3 attractors are reported for all variables. The dataset contains the values of the fast sub-variables (i.e. protein subnetwork), slow sub-variables (i.e. gene subnetwork) and the global functional activity.Additional file 9: Data S4. In silico screening predictions of combinatorial treatments. (Microsoft Excel Worksheet). This file reports the results of a screening of combinatorial perturbations. More particularly, it reports all conditions that lead to a transition towards a healthy chondrocyte (SOX9+) when starting from a hypertrophic-like chondrocyte (Runx2+). The first sheet reports conditions leading to such transition 100% of the time, between 99 and 90% of the time for the next sheet and so on. The TFs sheet references all the transcription factors of the model that could be retrieved in the result and that are less convenient to target experimentally (to define exclusion criteria for the experimental validation). The ‘Summary’ sheet, reports a list of conditions that might be interesting to test. It excludes conditions involving TFs.Additional file 10: Fig. S2. Effect of ALKs ratio in the influence of inflammation and TGFβ signaling on chondrocyte hypertrophy. Percentage of perturbations remaining in the healthy state (right) or transitioning towards the hypertrophic one (left) during inflammatory pathway activation with TGF-B treatment while changing the ratio between ALK1 and ALK5. The inflammatory and TGF-B profiles that were imposed are the same as in Fig. 4A except that the value imposed for ALK1 and ALK5 are varying between 0 and 1 with a 0.1 increment. ALK1=ALK5 on the red diagonals and ALK1>ALK5 in the upper right corner. Roughly, the rescue of the healthy state by TGF-B is lost when ALK1 is greater than ALK5. If ALK1 is high enough and that the difference between ALK1 and ALK5 is not greater than 10-20% then the protective effect of TGF-B is also disturbed for ALK1<ALK5.Additional file 11: Fig. S3. Screenshot of the user-friendly interface for the virtual chondrocytes App. The standalone Matlab-based applications can be launched and used without Matlab license, provided that the compiler Matlab Runtime is installed (https://nl.mathworks.com/products/compiler/matlab-runtime.html). The virtual chondrocyte initial state can be set as healthy or hypertrophic, allowing the user to test any scenarios. All the 60 components may be perturbed alone or in any sort of combination by forcing the variables to take a value in the interval [0:1], with a step of 0.1. The most left column indicate the value of the variable in the selected initial state, for information. Obviously, applying a perturbation that is equal to the initial value of the variable will not affect the system. Once the setting are done, the user can apply the experimental condition by pushing the button ‘Test condition’ and the percentage of transitions towards each of the possible basal stable states (i.e. ‘None’, ‘Healthy’ and ‘Hypertrophic’) is computed. If the ‘Compute statistics’ box is ticked, then the experiment is repeated 3 times and the average and standard deviations are displayed (variation occurs due to the stochastic nature of the model). The results may be exported and saved in an excel file via the ‘Save’ button. The application can be installed with the executable file on the GitHub repository, [https://github.com/Rapha-L/Insilico_chondro.git]. No Matlab license is required, however, the operating system should be able to support the Matlab software. For Linux users, a Windows virtual machine may be used.Additional file 12: Table S4. Targets and associated small molecules or growth factors for in vitro validation. The first row indicates the targets to be perturbed, [+] stands for activation while [-] stands for inhibition. The name (resp. cat number) of the small molecule or growth factor employed to achieve that effect is indicated in the row called ‘Molecule name’ (resp. ‘Cat n°’).Additional file 13: Fig. S4. Decision tree summarizing the variable updating scheme employed in algorithm to simulate the in silico chondrocyte. Each biological component is represented by the gene expression level (slow variable) and the protein activity potential (fast variable). Variables are updated based on the rules stored in the model’s equations. First, fast variables are updated in random order, when a pseudo-stable state is reached and that all fast variables have been updated, the next random chosen slow variable is updated. This goes on until a state that is stable both at the fast and slow level is reached. This is the final stable state. A state is considered stable if further variable updates do not bring further changes for any of the variables, with a predefined tolerance interval. The order in which variables are updated is random, thereby generating some stochasticity in the model. Within the fast (resp. slow) updating loops, variables are updated asynchronously (meaning the one after the others) according to the rules defined in the system of equations and in a random order. For some systems (i.e. set of equations) cyclic attractors may arise, meaning that the system never reaches a fixed stable state but oscillates between several states. This situation did not occur in the current study.Additional file 14: Fig. S5. Illustration of the algorithm for the asynchronous updating of variables with a simplified (3 nodes) example network. The network represents interactions happening in one of the subnetworks (protein= fast reactions or genetic = slow reactions). Inhibitions are represented in red and activations are in black. The mathematical rules corresponding to the network are displayed. If the rules result in a value lower than 0 (resp. higher than 1), the value is brought back to 0 (resp. 1). For this example, 3 different initial states are inputted and each of the three variables is updated asynchronously. The order in which variables are updated is random. The system reaches a stable state when the next update gives the same state as in the previous time step and that all variables were screened in the random ordering list. That state is a pseudo-stable state if the rules were describing fast reactions, in that case, a new slow variable can be updated (see Fig. S4.). However it is a final stable state if the rules were describing slow interactions since it would mean that the system had first reached a pseudo-stable state at the fast level and would now be stable at the slow level too. The example illustrates that different initial states may reach the same final state but also that the same initial state (e.g. [1 0 1] ) can reach different final states, depending on the order in which variables are updated, thereby introducing stochasticity in the system.Additional file 15: Fig. S6. Sensitivity analysis: impact of the number of initialization on the canalization results during the Monte Carlo. The percentage of random initialization reaching each attractor is displayed with the Sox9 positive state (healthy) in orange, the Runx2 positive state in blue and the None in grey. Data labels indicate the absolute amount of state reaching the attractors. None of initializations reached an alternative attractor, even for higher amount of random initializations The number of initialization has no significant impact on the basal canalization and 10.000 initializations were considered sufficient to screen the state space in the current study.Additional file 16: Fig. S7. Pseudo-time evolution of variables during simulations. The sequence of variable updating over each time steps after introducing a perturbation from the healthy state was saved as a timeseries and plotted. It shows the discrete behavior of the simulation and that a steady state is reached way before the duration of the perturbation (1000 time steps) is reached in such a way that maintain the perturbation longer would not change the output of the simulation once the perturbation is released. |
|
PMC9648007 | Ke Zhong,Yu Huang,Prince last Mudenda Zilundu,Yaqiong Wang,Yingying Zhou,Guangyin Yu,Rao Fu,Sookja Kim Chung,Yamei Tang,Xiao Cheng,Lihua Zhou | Motor neuron survival is associated with reduced neuroinflammation and increased autophagy after brachial plexus avulsion injury in aldose reductase-deficient mice | 09-11-2022 | Aldose reductase,Brachial plexus root avulsion,Motoneurons death,Autophagy,Neuroinflammation | Brachial plexus root avulsion (BPRA) is frequently caused by high-energy trauma including traffic accident and birth trauma, which will induces massive motoneurons (MNs) death as well as loss of motor and sensory function in the upper limb. The death of MNs is attributed to energy deficiency, neuroinflammation and oxidative stress at the injured ventral horn of spinal cord triggered by BPRA injury. It has been reported which aldose reductase (AR), an endogenous enzyme that catalyzes fructose synthesis, positively correlates with the poor prognosis following cerebral ischemic injury, diabetic retinopathy and diabetic peripheral neuropathy. However, the role of AR in BPRA remains unknown. Herein, we used a mouse model and found that in the spinal cord of BPRA mice, the upregulation of AR correlated significantly with (1) an inactivated SIRT1–AMPK–mTOR pathway and disrupted autophagy; (2) increased byproducts accumulation of lipid peroxidation metabolism and neuroinflammation; and (3) increased MNs death. Furthermore, our results demonstrated the role of AR in BPRA injury whereby the absence of AR (AR knockout mice, AR−/−) prevented the hyper-neuroinflammation and disrupted autophagy as well as motor neuron death caused by BPRA injury. Finally, we further demonstrate that AR inhibitor epalrestat is neuroprotective against BPRA injury by increasing autophagy level, alleviating neuroinflammation and rescuing MNs death in mice. Collectively, our data demonstrate that the AR upregulation in the spinal cord is an important factor contributing to autophagy disruption, neuroinflammation and MNs death following brachial plexus roots avulsion in mice. Our study also provides a promising therapy drug to assist re-implantation surgery for the treatment of BPRA. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-022-02632-6. | Motor neuron survival is associated with reduced neuroinflammation and increased autophagy after brachial plexus avulsion injury in aldose reductase-deficient mice
Brachial plexus root avulsion (BPRA) is frequently caused by high-energy trauma including traffic accident and birth trauma, which will induces massive motoneurons (MNs) death as well as loss of motor and sensory function in the upper limb. The death of MNs is attributed to energy deficiency, neuroinflammation and oxidative stress at the injured ventral horn of spinal cord triggered by BPRA injury. It has been reported which aldose reductase (AR), an endogenous enzyme that catalyzes fructose synthesis, positively correlates with the poor prognosis following cerebral ischemic injury, diabetic retinopathy and diabetic peripheral neuropathy. However, the role of AR in BPRA remains unknown. Herein, we used a mouse model and found that in the spinal cord of BPRA mice, the upregulation of AR correlated significantly with (1) an inactivated SIRT1–AMPK–mTOR pathway and disrupted autophagy; (2) increased byproducts accumulation of lipid peroxidation metabolism and neuroinflammation; and (3) increased MNs death. Furthermore, our results demonstrated the role of AR in BPRA injury whereby the absence of AR (AR knockout mice, AR−/−) prevented the hyper-neuroinflammation and disrupted autophagy as well as motor neuron death caused by BPRA injury. Finally, we further demonstrate that AR inhibitor epalrestat is neuroprotective against BPRA injury by increasing autophagy level, alleviating neuroinflammation and rescuing MNs death in mice. Collectively, our data demonstrate that the AR upregulation in the spinal cord is an important factor contributing to autophagy disruption, neuroinflammation and MNs death following brachial plexus roots avulsion in mice. Our study also provides a promising therapy drug to assist re-implantation surgery for the treatment of BPRA.
The online version contains supplementary material available at 10.1186/s12974-022-02632-6.
Brachial plexus roots avulsion (BPRA), mainly resulting from road traffic accidents or difficult childbirths, causes upper limb motor and sensory impairments which in turn seriously affects patient quality of life [1]. Motor neurons (MNs) tend to progressively degenerate and then die following brachial plexus root avulsion, thus leading to paralysis of the denervated muscles of the upper limbs [2]. Presently, numerous therapies including nerve root surgical reimplantation [3], local application of neurotrophic factors [4] and cell transplantation strategies [5] have been adopted. However, the neurological functional recovery remains unsatisfactory. Early and effective neuroprotective strategies are crucial to prolong the surgery therapeutic window and, thus, are essential to promote functional recovery following BPRA. Therefore, exploring novel medical approaches towards the treatment of BPRA is vital. The pathophysiological sequelae of primary avulsion injury in the spinal cord include ion homeostatic disequilibrium, heightened glutamate excitotoxicity, energy deficiency, disruption of normal mitochondrial function, and integrity of micro vessels, leading to organelle damage, toxic protein aggregation and progressive MNs loss [1]. Autophagy, a lysosome-dependent catabolic process, is responsible for degrading cytoplasm-based proteins, their aggregates and damaged organelles [6]. Previous studies revealed an aberrant expression of autophagic associated proteins in postmortem central nervous system samples obtained from animal models of amyotrophic lateral sclerosis and patients. These studies provided preliminary evidence that autophagy played a significant function in driving MNs death [7]. Besides, another important hallmark of BPRA is neuroinflammation in the injured spinal cord, which is one of the indispensable processes related to the aftermath of nervous tissue damage as well as regeneration [8]. In addition, neuroinflammation aggravated degeneration and death of MNs after different types of spinal cord injuries, including BPRA [9]. Inflammation following BPRA involves glial activation (activated astrocytes and microglia) which triggers excess release of many growth factors as well as mediators of oxidative stress and inflammation [10]. AR belongs to the aldo–keto reductase (AKR) superfamily. The NADPH-dependent AR is the sole enzyme that catalyzes the metabolism of excess glucose into sorbitol via the polyol glucose metabolism pathway [11]. Moreover, the NAD+-dependent enzyme sorbitol dehydrogenase oxidizes sorbitol into fructose. Therefore, AR will deplete cellular NADPH and elevate cytosolic NADH/NAD+ ratios, which will reduce the activity of NAD+-dependent deacetylase Sirtuin-1 (SIRT1) [12]. Furthermore, other studies have reported that AR was also involved in the reductions of lipid peroxidation-derived, higher affinity-aldehydes, especially the transformation of 4-hydroxy-2-nonenal (4-HNE) to glutathionyl-1,4-dihydroxynonanol (GS-DHN) [13]. Subsequently, the GS-DHN appears to be a mediator in inflammatory signaling [14]. However, there are few studies evaluating the role and the plausible molecular mechanisms of AR in the context of BPRA and MNs survival. In this study, we demonstrated elevated levels of AR concomitant with neuroinflammation, disrupted autophagy and MNs death in BPRA mice model. Additionally, we established a causal role of AR in MNs death using the AR knockout mice subjected to BPRA. Finally, we provided a proof of concept that pharmacological inhibition of AR is a potential therapeutic strategy to attenuate MNs death in BPRA.
AR gene knockout mice (AR−/−) in our stock were generated as described previously [15]. Age-matched normal C57BL/6 mice aged between 8 and 10 weeks and weighing 20–25 g were used as controls (wild-type mice, AR+/+). The wild-type mice were purchased from the Laboratory Animal Center of Sun Yat-sen University, Guangzhou city in China. The experimental animal use permit number is SYXK 2017-0081. The AR inhibitor epalrestat (EPAL, 40 mg/kg bodyweight) was administrated by daily (afternoon) oral gavage for 14 days to mice after receiving BPRA injury [16]. All the mice were housed under a 12-h light/dark cycle with unlimited access to standard mice chow and water available. The Chinese National Health and Medical Research Council animal ethics and ARRIVE guidelines were followed in executing surgical and animal care procedures. The Animal Care and Utilization Committee of Sun Yat-sen University approved all the experimental procedures.
The BPRA surgery was carried out aseptically as described in previous publications [17, 18]. Briefly, the avulsion groups mice were anesthetized with intraperitoneal injections of ketamine/xylazine (80/20 mg/kg) and maintained with 1% isoflurane. While in supine position, the right-side brachial plexus divisions, trunks and roots were carefully exposed, and the C5–T1 spinal nerve roots identified and then separated under a dissecting microscope (Cheng He Microsurgical Instruments Factory, Ning Bo, China). Micro-hemostatic forceps were used to grasp and pull out the dorsal and ventral rootlets of the region of interest. The spinal nerve region was cut immediately before the dorsal ganglion so that the proximal part bearing nerve rootlets was examined under a microscope to confirm whether or not the avulsion model succeeded. The muscles, fascia and skin were sutured in successively to close the surgical wound. The mice were then transferred into a pre-heated recovery chamber until fully awake and finally placed back to their home cages. In the sham control group, the laminectomy was performed to expose the right-side brachial plexus but avulsion was not carried out.
At several time points (1, 3, 7 and 14 days post-injury, dpi) following BPRA, the mice were anaesthetized using an intraperitoneal injection of 1% sodium pentobarbital (40 mg/kg, Sigma–Aldrich). An incision was made on the dorsal part of the next to expose the muscles and separate them. A bilateral laminectomy was then performed under a dissection microscope to expose the C5 to T1 spinal segments. The segments were rapidly dissected off and immediately frozen in liquid nitrogen for the ease of division into ipsilateral and contralateral halves. The cut spinal cord segments were then stored in liquid nitrogen until needed for PCR, WB and ELISA. For histology studies, the roots avulsed mice were killed as mentioned above at days 1, 3, 7 and 14 dpi (n = 5/group). Their blood was washed out by transcardial perfusion with 0.9% normal saline and then fixed with pre-cooled 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Bilateral laminectomy was carried out under a dissection microscope to expose the C7–C8 spinal segments. The segments of interest were carefully dissected out and post-fixed in 4% PFA for 3–4 h followed by overnight immersion in 30% (v/v) sucrose in PB solution in a fridge set at 4 °C. 30-μm transverse sections of the C7–C8 segments were cut on a freeze microtome and every third section was collected into 0.01 M PBS for use in staining described below.
MNs loss was determined by neutral red staining as previously reported [17]. 1% neutral red (N4638, Sigma–Aldrich, USA) in 0.1 M acetic acid (pH 4.8) was used to stain sections for 2 h and then followed by a series of dehydration in graded concentrations of ethanol. The images were captured Nikon light microscope using the bright field feature. The counting of motor neurons was conducted by 2 laboratory personnel blinded to the injury or treatment conditions of the mice. The MNs loss was determined by its survival rate (ratio of ipsilateral/contralateral MNs number). Only large soma neurons bearing a visible nucleolus and located in lamina IX of Rexed were enumerated.
A Nikon Eclipse 90i, fluorescence microscope (Nikon, Japan) was used for immunofluorescence evaluations. The spinal cord slices were rinsed thrice in PBS and then incubated in 0.3% Triton X-100/0.1% bovine serum albumin for half an hour at room temperature. The slices where then incubated with the following primary antibodies; mouse anti-NeuN (Abcam UK), rabbit anti-GFAP (Cell Signaling Technology Inc. USA), rabbit anti-IBA-1 (Wako Bioproducts, USA), rabbit anti-Cleaved Caspase-3 (Cell Signaling Technology Inc. USA), rabbit anti-Bcl-2 (Cell Signaling Technology Inc. USA), rabbit anti-AMPKα (Cell Signaling Technology Inc. USA), rabbit anti-Beclin1 (Cell Signaling Technology Inc. USA), rabbit anti-GAP43 (Cell Signaling Technology Inc. USA), rabbit anti-mTOR (Cell Signaling Technology Inc. USA), rabbit anti-p-mTOR (Cell Signaling Technology Inc. USA), rabbit anti-p-CREB (Cell Signaling Technology Inc. USA), rabbit anti-LC3B (Cell Signaling Technology Inc. USA), rabbit anti-SIRT1 (Abcam UK), rabbit anti-P62 (Abcam UK), rabbit anti-4-HNE (Abcam UK), rabbit anti-p-AMPKα (Abcam UK), mouse anti-Arginase1 (Santa Cruz Biotechnology Inc USA), goat anti-CD16/32 (R&D Systems Inc USA) and goat anti-ChAT (1:500, Millipore USA) These slices were incubated with primary antibodies at 4 °C overnight. The slices were then washed thrice with PBS, and then incubated in darkness with the following secondary antibodies for 1–2 h at room temperature; goat anti-mouse with Alexa Fluor-488 (Invitrogen USA), goat anti-rabbit with Alexa Fluor-555 (Invitrogen USA), donkey anti-goat with Alexa Fluor-488 (Invitrogen USA), and donkey anti-rabbit with Alexa Fluor-555 (Invitrogen USA). After the secondary antibody incubation, the slices were rinsed thrice in PBS and then incubated in Hoechst 33,342 (H3570, Life Technologies, USA). Finally, the slices were air-dried, cover-slipped and observed under a fluorescence microscope. Omitting of either a primary or secondary antibody was used as a as a negative control. We calculated the mean number of immunoreactive cells from the total 10-immunofluorescence sections serially obtained from each mice (Every third sections from each mice used for staining and counting). The immunoreactive cells were quantified under the 20 × magnification in six circular areas of 100 μm2 which is located in lamina IX of Rexed of spinal cord ipsilateral ventral horn. Two independent persons blinded to the sidedness of the groups performed cell counting, pooling of means and data analysis following previously published protocols [18].
Transmission electron microscopy (TEM) was used to evaluate mitochondrial and autophagosome morphology in the spinal cord motor neurons following brachial plexus roots avulsion (days 1, 3, 7, and 14 post-injury; n = 5 mice). The deeply anesthetized mice were transcardially perfused using 0.9% normal saline until the exudate was clear of blood stain and then fixed using a mixture of 2% PFA and 2.5% glutaraldehyde (Sigma–Aldrich, G6257, USA) in 0.1 M PBS. A 1 mm3 piece of ventral horn grey matter tissue located in lamina IX of Rexed per mouse was cut off and post-fixed in 2% glutaraldehyde for another 2 h at 4 °C. The glutaraldehyde-fixed tissue was then washed thrice in 0.1 M cacodylate buffer and further post-fixed in 1% osmium tetroxide for another 2 h. After three rinses in distilled water, the pieces were then dehydrated in a graded series of ethanol concentrations. A half-acetone and half-resin mixture was used to infiltrate the tiny tissue pieces overnight at 4 °C. The tissues were embedded in resin and then cured under the following settings: (1) 37 °C overnight, (2) 45 °C for 12 h, and (3) 60 °C for 24 h. Afterward, a vibratome was used to obtain 70-nm thin sections which were then stained with 3% uranyl acetate for 20 min followed by 0.5% lead citrate for 5 min. Ultrastructural changes in the motor neurons were evaluated under the transmission electron microscope (Philips CM 10, Eindhoven, Netherlands) operated at 100 kV.
Spinal cord tissue cytokines were assayed using ELISA kits for IL-1β (Mouse IL-1 beta/IL-1F2 Quantikine ELISA Kit, MLB00C), IL-6 (Mouse IL-6 Quantikine ELISA Kit, M6000B), ICAM1 (Mouse ICAM-1/CD54 Quantikine ELISA Kit, MIC100) and IL-10 Mouse IL-10 Quantikine ELISA Kit, M1000B), which were all from R&D Systems. The assays were performed according to the manufacturer’s instructions.
TRIzol Reagent (Invitrogen) was used for Total RNA extraction following recommended procedures laid down by the kit manufacturer. RNA concentration and purity (260/280 ratio) were measured using UV spectroscopy. The total RNA was reverse-transcribed using the Fast Quant RT Kit (with gDNase) (TIANGEN, China) following the manufacturer’s recommended steps. Then, real-time PCR was performed using SYBR Green assays (TIANGEN, China) according to the manufacturer’s instructions. All primers used in this study are listed in Table 1. The PCR settings on the CFX96 touch detection system (Bio–Rad, Hercules, CA, USA) were set as follows: a denaturation step-95 °C for 15 min followed by 40 × PCR cycles at 95 °C for 10 s and 60 °C for 30 s. For quantitative results, the expression of mRNA was represented as the fold change using the 2−ΔΔct method and was normalized to the control gene GAPDH. Messenger RNA expression differences between the avulsion groups and sham operated group were compared using Student’s t-test in SPSS (IBM SPSS Inc Version 22.0 USA). A p-value less than 0.05 was deemed statistically significant.
An electric homogenizer was used to process frozen ventral horn tissue samples in a whole-cell lysis buffer (Key GEN Biotech) mixed with a protease inhibitor and 1 mM PMSF (Sigma–Aldrich). The homogenates were left on ice for 1 h and then centrifuged at 12,000 rpm for 30 min at 4 °C. The resultant supernatants were pipetted out into new tubes. A small amount from each tube was used for protein concentrations using the BCA (Thermo Scientific) method. Equal amounts of protein from each sample were then run through the 10% SDS–PAGE and electro-transferred onto respective polyvinylidene fluoride (PVDF) membranes (Millipore, 0.45 μm) using Transblot Turbo (Bio–Rad, USA) at 300 mA for 1 h. The membranes bearing electro-transferred proteins were incubated with 5% nonfat milk/Tris buffer containing Tween-20 buffer (TBST; 10 mM Tris–HCl, pH 8.0; 150 mM NaCl; and 0.1% Tween 20) for 2 h at room temperature. The strips corresponding to target proteins’ molecular weights were then incubated in the primary antibodies, listed in the immunofluorescence method section, at 4 °C overnight. The membranes were then washed thrice with TBST (10 min each) and then probed with the corresponding secondary antibodies conjugated with horseradish peroxidase (Abcam, UK) at room temperature for 2 h. The membranes were then washed thrice for 10 min each using TBST remove unbound secondary antibodies and then visualized using enhanced chemiluminescence (Thermo Scientific Inc., USA). The densities of specific bands were measured using ImageJ software’s densitometry feature and normalized against a loading control (GAPDH).
All data are expressed as the mean ± standard error of the mean (SEM). The data were analyzed to one-way or two-way ANOVA, as appropriate, and then followed by a Bonferroni or Tukey post hoc test for statistically significant results. A p < 0.05 was considered to be statistically significant.
Generally, AR is hardly expressed in the normal tissue. However, there are upregulations of AR expression following stroke, non-alcoholic fatty liver disease, diabetic retinopathy as well as diabetic peripheral neuropathy [16, 19–21]. In our study, the expression of AR in the spinal cord of BPRA mice was compared with sham mice. AR was upregulated approximately 13-fold at the mRNA level and threefold at the protein level at 3 dpi (day post-injury) compared with the sham group (Fig. 1A–C). Immunofluorescence staining of spinal cord tissues also showed that the BPRA mice exhibit stronger AR expression than the sham mice at 3 dpi, and AR localization was predominantly cytoplasmic in neurons and microglia on the ipsilateral ventral horn of the spinal cord (Fig. 1D).
To investigate the effects of AR upregulation in BPRA, we sought to find out whether the deficiency of AR (examined using AR knockout mice, AR−/−) alleviated MNs death in the model of BPRA mice. Initially, we examined the effect of AR deficiency on MNs survival by staining neutral red, and the survival rate of injured MNs was assessed as the ratio of ipsilateral ventral horn neutral red-positive MNs to those on the contralateral ventral horn [22]. As shown in Fig. 2A and C, the survival ratio of MNs in AR−/− group was higher than WT group at 3, 7 and 14 dpi on the ipsilateral ventral horn of the spinal cord. Furthermore, AR−/− mice exhibited an increase in NeuN and anti-apoptosis gene Bcl-2 colocalization-positive neurons on the ipsilateral ventral horn of the spinal cord at 1–14 dpi (Fig. 2B, D). In addition, compared with the WT group, the AR−/− group showed more regeneration-related protein GAP43/NeuN colocalization-positive neurons at 7–14 dpi and higher GAP43 protein expression at 1–14 dpi on the ipsilateral ventral horn of the spinal cord (Fig. 2E, G, I, J). BPRA resulted in an increase in the apoptosis-related protein C-Caspase3/NeuN colocalization-positive neurons and C-Caspase3 protein expression level, which was attenuated in AR−/− mice at 1–14 dpi on the ipsilateral side of the spinal cord (Fig. 2F, H, I, K). Collectively, the results provided experimental evidence that genetic ablation of AR in mice prevented BPRA-mediated MNs death.
The dysregulation of autophagy is involved in motor neuron diseases, including spinal cord injury and amyotrophic lateral sclerosis. To evaluate the autophagy level after BPRA, we assessed the expression of the autophagy protein, LC3B in the injured spinal cord by immunofluorescence and western blotting. The conversion of LC3B-I to LC3B-II by adding phosphatidylethanolamine is important to autophagosome formation and this is considered as a marker of autophagosome formation and accumulation. Compared with BPRA WT mice, AR−/− mice had greatly increased LC3B/NeuN colocalization-positive neurons at 3–14 dpi on the ipsilateral ventral horn of the spinal cord and higher LC3B-II protein expression at 1–14 dpi on the ipsilateral side of the spinal cord (Fig. 3A, C, E, H), indicating the more accumulation of autophagosomes after injury in AR−/− mice. To elucidate the autophagy mechanisms following BPRA, we assessed the levels of proteins that would regulate and form autophagosomes. Increased accumulation of LC3-II is likely due to either increased formation or decreased degradation of autophagosomes. Ubiquitinated cargo is delivered to autophagosomes by the adapter protein P62 (SQSTM1). During this delivery, the adapter protein (p62) is also degraded by autophagy alongside its cargo. Therefore, the accumulation of p62 suggests that autophagic degradation has been disrupted. Compared with BPRA WT mice, AR−/− mice had greatly decreased P62/NeuN colocalization-positive neurons at 3–14 dpi on the ipsilateral ventral horn of the spinal cord and lower P62 protein expression at 1–14 dpi on the ipsilateral side of the spinal cord (Fig. 3B, D, E, G). Moreover, western blot analysis showed more expression of the autophagy regulatory protein Beclin1 in AR−/− mice on the ipsilateral side of the spinal cord at 7 and 14 dpi than in WT mice, suggesting that the genetic ablation of AR increased initiation of autophagy (Fig. 3E, F). Overall, these results suggested BPRA-mediated MNs death was suppressed in AR−/− mice, which may be associated with enhanced autophagy level.
We further used electron microscopy (EM) to evaluate autophagic vacuoles number of MNs in both WT and AR−/− mice following BPRA. The structures of typical autophagic vacuoles are shown under high-magnification (13,500 ×) EM (green arrow in Fig. 4A). Compared with WT mice, the number of autophagic vacuoles was remarkably increased at 1–14 dpi in AR−/− mice (Fig. 4E). These results provided definite evidence that autophagy initiation was induced in AR−/− mice following BPRA injury. Mitochondria accommodate most cell energetic demands by generating ATP. Moderate autophagy is responsible for degrading damaged mitochondria caused by BPRA injury. We analyzed mitochondrial structures and number in the BPRA model at 1–14 dpi in WT and AR−/− mice. We first examined mitochondrial morphology and size by transmission electron microscopy and observed that there were larger mitochondria with altered crista organization in MNs from WT mice when comparing with those in AR−/− mice following BPRA (colored arrowheads in Fig. 4A). We further analyzed mitochondrial alterations in depth and classified mitochondrial morphology into four categories namely [23]: class I: fairly dark mitochondria, with a uniform matrix filled with densely packed and regularly distributed cristae; class II: mitochondria with disrupted cristae and a loss of matrix density; class III: empty mitochondria with disorganized cristae or cristae on the periphery; and class IV: swollen mitochondria with disrupted membranes) (Fig. 4B). Quantifications of mitochondrial subclasses were then performed, revealing that the WT mice at 14 dpi exhibited 4.2% mitochondrial class I, 4.9% class II, 10.4% class III and 80.5% class IV (Fig. 4C). However, the AR−/− mice at 14 dpi displayed a drastic increase in “healthy” mitochondria class I (78.5%) and class II (12.4%), class III (6.1%), class IV (3%) (Fig. 4C). We also observed that AR−/− mice also displayed a significant increase in the mitochondrial number at 1–14 dpi (Fig. 4D) compared with the WT mice following BPRA injury. Taken together, these findings demonstrated that genetic ablation of AR could eliminate damaged mitochondria and enhance autophagy level in BPRA mice mode.
Elevated AR activity is known to deplete cellular NADPH and cause high cytosolic NADH/NAD+ ratio. This results to loss of NAD+-dependent deacetylase SIRT1 activity [12]. Compared with WT mice, AR−/− mice had greatly increased SIRT1/NeuN colocalization-positive neurons at 3–14 dpi on the ipsilateral ventral horn of the spinal cord and higher SIRT1 protein expression at 1–14 dpi on the ipsilateral side of the spinal cord (Fig. 5A, C, E, F). Our results indicate that BPRA decreased the expression of SIRT1 and that AR−/− mice significantly restored the BPRA-induced decrease in SIRT1 expression. A rising number of studies suggest that aberrant mTOR signaling impacts many pathways, such as glucose metabolism, energy production, mitochondrial function, cell growth and autophagy. AMPK and mTOR have a reciprocal relationship mediated by SIRT1, and the substrates of mTORC1 suppress autophagy [24]. We further measured the effect of AR ablation on BPRA-induced expression of AMPK and mTOR in motor neurons. Compared with WT mice, AR−/− mice had greatly increased p-AMPK/NeuN colocalization-positive neurons on the ipsilateral ventral horn of the spinal cord and higher p-AMPK protein expression at 3–14 dpi on the ipsilateral side of the spinal cord (Fig. 5B, D, E, G). Similarly, the BPRA-induced mTOR phosphorylation increase was also prevented in AR−/− mice at 1–14 dpi on the ipsilateral side of the spinal cord (Fig. 5H). It may infer that genetic ablation of AR potentially enhance SIRT1–AMPK–mTOR signaling and autophagy level following BPRA injury.
Following BPRA injury, the immunoreactive expression of Iba1 and GFAP, as markers of microglia and astrocytes, respectively, was used to assess neuroinflammation. Compared with the WT mice, the AR−/− mice had dramatically decreased Iba1 and GFAP average fluorescence intensity at 3–14 dpi on the ipsilateral ventral horn of the spinal cord (Fig. 6A–D). This evidence indicates that genetic ablation of AR attenuated ventral horn neuroinflammation in the injured spinal segment. Activated microglia can either become pro-inflammatory or anti-inflammatory phenotypes [25]. Compared with the WT group, the AR−/− group showed a dramatically decreased proportions of pro-inflammatory (CD16/32+/Iba-1+) microglia at 3–14 dpi (Fig. 7A, C) and an increased proportions of anti-inflammatory (Arginase1+/Iba-1+) microglia at 1–14 dpi on the ipsilateral ventral horn of the spinal cord (Fig. 7B, D). We further determined whether the AR-deficient mice also showed aberrant expression of cytokines that are functionally more important. Compared with WT mice, AR−/− mice had significantly reduced pro-inflammatory cytokines IL-1β and IL-6 levels at 1–14 dpi and inflammatory responses protein ICAM1 levels at 3–14 dpi, while increasing anti-inflammatory cytokine IL-10 levels at 1–14 dpi (Fig. 7E) on the ipsilateral side of the injured spinal cord. Moreover, compared with WT mice, AR−/− mice had significantly reduced pro-inflammatory cytokines IL-1β mRNA levels at 1–14 dpi and iNOS mRNA levels at 1–14 dpi, while increasing anti-inflammatory cytokines IL-10 mRNA levels at 3–14 dpi and Arg-1 mRNA levels at 3–14 dpi (Fig. 7F) on the ipsilateral side of the injured spinal cord. These results further indicate that genetic ablation of AR promotes microglia to switch from a pro-inflammatory to anti-inflammatory phenotype following BPRA injury.
Considering cyclic-AMP response binding protein (CREB) plays a key role in anti-inflammatory microglial phenotypes and AR can reduce 4-HNE into GS-DHN, a macrophage inflammatory mediator, we estimated whether genetic ablation of AR acts via 4-HNE–p-CREB signaling to exert its “phenotype switching” effect following BPRA injury [26, 27]. Therefore, we sought to find out the expression of 4-HNE–p-CREB signal axis following BPRA in WT and AR−/− mice. Compared with the WT mice, the AR−/− mice had dramatically increased number of 4-HNE+/Iba1+ microglia at 1–14 dpi and number of p-CREB+/Iba1+ microglia at 1–14 dpi on the ipsilateral ventral horn of the spinal cord after BPRA injury (Fig. 8A–D). These results suggest that genetic ablation of AR exerts its anti-inflammatory effects via the 4-HNE–p-CREB signaling pathway.
Epalrestat (EPAL), a pharmacological AR inhibitor with protective effects against diabetic complications in mouse models and human clinical trials [16], was used to evaluate the neuroprotective potential of AR suppression following BPRA injury. Significantly reduced AR, C-Caspase3 expression and increased ChAT, GAP43 expression on the ipsilateral ventral horn of the spinal cord, as evaluated using IF, were observed with pharmacological inhibition of AR by epalrestat at 14 dpi following BPRA injury (Fig. 9A). Furthermore, the administration of epalrestat significantly reduced the expression of P62, but greatly upregulated LC3B, SIRT1, p-AMPK expression on the ipsilateral ventral horn of the spinal cord at 14 dpi following BPRA injury (Fig. 9B). In addition, BPRA mice significantly reduced the expression of Iba1, GFAP and CD16/32 while greatly upregulating Arginase 1 expression on the ipsilateral ventral horn of the spinal cord at 14 dpi after treatment with epalrestat (Fig. 9C). Therefore, these results were similar to those of genetic knockout, suggesting that with use of epalrestat, pharmacological suppression of AR showed significant neuroprotective of MNs effects by attenuating BPRA-induced autophagy disruption and neuroinflammation (Fig. 10).
BPRA injury is mainly caused by motor vehicle accident and birth trauma. Severe BPRA can lead to partial or complete loss of motor and sensory functions of the upper limb because of massive MNs death, motor axon degeneration and de-innervation of targeted muscles. The novel finding of our study is that AR expression is markedly upregulated in the BPRA mice model. Importantly, the role of AR following BPRA is investigated by using AR gene knockout mice and inhibiting its activity pharmacologically with AR inhibitor epalrestat. Our date reveals that genetic knockout or epalrestat pharmacological suppression of AR contributed to better MNs survival by attenuating BPRA-induced autophagy disruption and neuroinflammation. These findings cast AR as a potential therapeutic target in BPRA injury treatments and that AR inhibitors could be a promising strategy for mitigating against BPRA caused MNs death. Autophagy is a lysosome-dependent intracellular degradation process that eliminates long-lived proteins as well as damaged organelles. An increasing number of evidence suggests that dysregulation of autophagy plays a pivotal role in the progression of neurodegenerative diseases including motor neuron disorders. Our data suggest that BPRA will cause MNs autophagy dysregulation and MNs death. Moreover, heightened levels of P62 and reduced levels of LC3II found in BPRA mice model were associated with MNs death. Notably, autophagy could prevent the accumulation of damaged macromolecules and organelles to maintain function and survival of MNs. Importantly, in AR-deficient mice (AR-/- or EPAL treatment), the ratio of LC3BII to LC3BI was significantly increased, while the level of P62 was significantly decreased, suggesting that AR suppression may enhance autophagy level and maintain MNs survival. In an earlier study, SIRT1 could protected cardiomyocytes against hypoxia-induced apoptotic death in a mechanism via autophagy and AMPK activation. AMPK, a serine/threonine kinase, senses changes in, and regulates cellular energy homeostasis. When activated, it promotes autophagy by inhibiting mTOR [28]. Both AMPK and SIRT1 are fuel-sensing molecules regulate each other. The AMPK activation are regulated by two upstream kinases, LKB1 (the primary AMPK kinase) and CaMKKβ (an AMPK kinase). Regulation of SIRT1 has been attributed to changes in NAD+ abundance and the NAD+/NADH ratio. The actual relationship between AMPK and SIRT1 need us examine the LKB1 activity and NAD+ abundance in the future study. At present, our findings implied that suppression of AR could enhance SIRT1–AMPK–mTOR signaling and autophagy level following BPRA injury. Our IHC studies come from ipsilateral ventral horn only, while WB studies were used to analyze the protein expression levels in all the cells (neurons, microglia, astrocytes and oligodendrocytes) from ventral horn and dorsal horn on the ipsilateral side of spinal cord. Although our results showed elevated autophagy level in AR−/− mice after BPRA injury, the exact mechanism of autophagy, its regions and cell-type specificity remained unknown. In a previous spinal cord injury study [29], the expressions of LC3 vary widely between different cell types. There would be high accumulation of LC3 in ventral horn motor neurons preferentially, which was similar in most of activated CD11b+ microglia according their cell morphology. In the white matter, LC3 accumulated in CC1-positive oligodendrocytes both in dorsal and ventral white area. However, unlike microglia and oligodendrocytes, the number of LC3-positive astrocytes remained very low in all regions at all time points examined. According to our previous and present study [9], BPRA injury causes microglia proliferation and activation in the injured spinal ventral horn and dorsal horn, with the peaks at 14 day post BPRA injury, which may be partly relevant with our western blot result. However, to understand the effect of AR on cell-type specificity autophagy function, further studies with cell-specific AR knockout mice are necessary. Inflammation is among the very crucial pathological processes in the secondary injury phase after BPRA injury. Resident microglia and hematogenous macrophages, invade the injury site shortly after the primary injury [30]. The inflammatory reactions that occur after spinal cord injury is a “double-edged sword”, which can cause both neuroprotective and neurotoxic effects depending on the dichotomous polarity of microglia/macrophages [31]. Cellular injury causes the release of pro-inflammatory cytokines and other mediators. Cytokine production and iNOS activation typifies the pro-inflammatory phenotypes macrophage and microglial in earlier phases of the spinal cord injury [32]. The anti-inflammatory phenotypes existence following spinal cord injuries have not been widely reported. In a contusion model of spinal cord injury in mice, the major type of microglia/macrophages are pro-inflammatory phenotype while the anti-inflammatory phenotype is transient and a minority. The overexpression of one of the classic markers of the anti-inflammatory phenotype, Arg1, appears to be transient and returns to normal expression levels by day 14 post-injury [32]. Our results also showed more pro-inflammatory phenotype microglia/macrophages in the WT mice following BPRA injury. However, the anti-inflammatory phenotype microglia/macrophages were predominant within the injured spinal cord of AR−/− mice. These findings demonstrate that AR plays an important role in the divergence of microglial/macrophage functions after BPRA injury in mice. Recently, several reports revealed that AR mediates LPS-induced inflammatory signals in macrophages [33]. Suppression of AR by several pharmacological inhibitors including sorbinil, tolrestat, and zopolrestat, attenuates LPS-induced production of TNF-α, IL-6, IL-1β, and IFN-γ and MCP-1 in murine peritoneal macrophages [34]. Results from these studies further showed that AR suppression or ablation averts macrophage transformation into the pro-inflammatory phenotype. In this study, we found that AR deficiency or suppression reduced the number of pro-inflammatory phenotype microglia/macrophages following BPRA. The reduction of 4-HNE to DHN, happens efficiently because of their much lower Km values (in the micromolar range of 10 to 30 μM) compared with that of glucose (50 to 100 mM) [35]. Therefore, suppression of AR averts macrophages polarization to the pro-inflammatory phenotype. This is possibly due to the reduced conversion of 4-HNE to DHN. Furthermore, inhibiting AR expression can avert LPS-induced knockdown of cAMP response element modulator (CREM), phosphorylation of CREB, and DNA binding of CREB in macrophages [36]. Therefore, we suggest that AR regulates this polarization switch. Specifically, 4-HNE would be reduced to DHN if there is sufficient AR, and it would cause the inflammation cytokines release thereby favoring microglial pro-inflammatory phenotype formation. However, in AR inhibition or knockout models, 4-HNE tends to accumulate in the cytoplasm where it activates CREB to favor the anti-inflammatory microglial phenotype formation (Additional file 1, Additional file 2). AR inhibitors have been under investigation for approximately four decades, although most of them are used in diabetic neuropathy and retinopathy [37]. Epalrestat, a post-market AR inhibitor (ARI) approved in Bangladesh, India, and China, is one of the most common ARIs for patients who suffer from diabetes mellitus [38]. Although clinical studies of epalrestat’s effects on diabetic neuropathy have been carried out, it is still controversial [39]. In this study, our data demonstrated that spinal AR upregulation and consequent elevation of neuroinflammation, aberrant autophagy caused motor neuron death after BPRA. Our study provides new insights regarding the mechanisms underlying the role of AR in BPRA pathogenesis. Moreover, our data imply that AR may be a novel therapeutic target in brachial plexus roots avulsion and provides a rationale for further testing the utility of epalrestat, for the treatment of BPRA. Exploring the roles of AR-mediated signaling pathways following BPRA and spinal cord injuries may be critical for increasing our understanding of efficacy and safety profiles of ARIs. These findings reveal that AR is a contributor to the BPRA pathogenesis, and AR deficiency is neuroprotective in mice subjected to BPRA. In the future, the cell type-specific AR knockouts could enhance our understanding of the actual function AR in those cells or tissues following BPRA or other neurodegeneration diseases. On the basis of results, we also consider epalrestat as a feasible pharmacological agent to attenuate MNs death after BPRA. Further biomedical studies and clinical trials are required to test the safety and efficacy of such AR inhibitors in BPRA patients.
Additional file 1: Figure S1. The AR−/− mice genotyping and the expression level of AR in AR−/− mice. (A) AR null allele were identified by northern blot analysis. The expected bands of ∼119 bp is observed in the wild-type mice while not seen in AR−/− mice. (B) Western blot analyze the AR expression in the AR−/− mice. (C) Immunofluorescence analyze the AR expression in the spinal cord of AR−/− mice at 3 day following BPRA injury.Additional file 2. Raw data of western blot. |
|
PMC9648016 | Edgar Gulavi,Fridah Mwendwa,David O. Atandi,Patricia O. Okiro,Michael Hall,Robert G. Beiko,Rodney D. Adam | Vaginal microbiota in women with spontaneous preterm labor versus those with term labor in Kenya: a case control study | 10-11-2022 | Preterm birth,Sub-Saharan Africa,Vaginal microbiota | Background Preterm birth is a global problem with about 12% of births in sub-Saharan Africa occurring before 37 weeks of gestation. Several studies have explored a potential association between vaginal microbiota and preterm birth, and some have found an association while others have not. We performed a study designed to determine whether there is an association with vaginal microbiota and/or placental microbiota and preterm birth in an African setting. Methods Women presenting to the study hospital in labor with a gestational age of 26 to 36 weeks plus six days were prospectively enrolled in a study of the microbiota in preterm labor along with controls matched for age and parity. A vaginal sample was collected at the time of presentation to the hospital in active labor. In addition, a placental sample was collected when available. Libraries were constructed using PCR primers to amplify the V6/V7/V8 variable regions of the 16S rRNA gene, followed by sequencing with an Illumina MiSeq machine and analysis using QIIME2 2022.2. Results Forty-nine women presenting with preterm labor and their controls were enrolled in the study of which 23 matched case–control pairs had sufficient sequence data for comparison. Lactobacillus was identified in all subjects, ranging in abundance from < 1% to > 99%, with Lactobacillus iners and Lactobacillus crispatus the most common species. Over half of the vaginal samples contained Gardnerella and/or Prevotella; both species were associated with preterm birth in previous studies. However, we found no significant difference in composition between mothers with preterm and those with full-term deliveries, with both groups showing roughly equal representation of different Lactobacillus species and dysbiosis-associated genera. Placental samples generally had poor DNA recovery, with a mix of probable sequencing artifacts, contamination, and bacteria acquired during passage through the birth canal. However, several placental samples showed strong evidence for the presence of Streptococcus species, which are known to infect the placenta. Conclusions The current study showed no association of preterm birth with composition of the vaginal community. It does provide important information on the range of sequence types in African women and supports other data suggesting that women of African ancestry have an increased frequency of non-Lactobacillus types, but without evidence of associated adverse outcomes. | Vaginal microbiota in women with spontaneous preterm labor versus those with term labor in Kenya: a case control study
Preterm birth is a global problem with about 12% of births in sub-Saharan Africa occurring before 37 weeks of gestation. Several studies have explored a potential association between vaginal microbiota and preterm birth, and some have found an association while others have not. We performed a study designed to determine whether there is an association with vaginal microbiota and/or placental microbiota and preterm birth in an African setting.
Women presenting to the study hospital in labor with a gestational age of 26 to 36 weeks plus six days were prospectively enrolled in a study of the microbiota in preterm labor along with controls matched for age and parity. A vaginal sample was collected at the time of presentation to the hospital in active labor. In addition, a placental sample was collected when available. Libraries were constructed using PCR primers to amplify the V6/V7/V8 variable regions of the 16S rRNA gene, followed by sequencing with an Illumina MiSeq machine and analysis using QIIME2 2022.2.
Forty-nine women presenting with preterm labor and their controls were enrolled in the study of which 23 matched case–control pairs had sufficient sequence data for comparison. Lactobacillus was identified in all subjects, ranging in abundance from < 1% to > 99%, with Lactobacillus iners and Lactobacillus crispatus the most common species. Over half of the vaginal samples contained Gardnerella and/or Prevotella; both species were associated with preterm birth in previous studies. However, we found no significant difference in composition between mothers with preterm and those with full-term deliveries, with both groups showing roughly equal representation of different Lactobacillus species and dysbiosis-associated genera. Placental samples generally had poor DNA recovery, with a mix of probable sequencing artifacts, contamination, and bacteria acquired during passage through the birth canal. However, several placental samples showed strong evidence for the presence of Streptococcus species, which are known to infect the placenta.
The current study showed no association of preterm birth with composition of the vaginal community. It does provide important information on the range of sequence types in African women and supports other data suggesting that women of African ancestry have an increased frequency of non-Lactobacillus types, but without evidence of associated adverse outcomes.
Preterm birth (PTB) is defined as birth before 37 completed weeks of gestation [1]. It is one of the leading causes of perinatal morbidity and mortality worldwide and about 15 million PTBs occur every year [2]. PTB is a global challenge affecting up to 12% of births in low-income countries and 9% of births in Western countries [3]. The majority of PTBs occur in sub-Saharan Africa and South Asia [4] with an estimate of a 12% PTB rate in sub-Saharan Africa [5]. Kenya has a 12% PTB rate with an estimated 190,000 babies born preterm every year [6]. Maternal–fetal factors and gene–environment interactions play roles in determining the length of gestation. Some of these factors include African ancestry (in the US and the UK), time of less than six months after a previous pregnancy, low prepartum maternal weight, previous preterm birth, multiplex pregnancy, and maternal infection or vaginal dysbiosis, as well as numerous other known or suspected risk factors [7, 8]. The vaginal microbiota is thought to play a role in pregnancy outcomes. In addition, vaginal dysbiosis has been associated with preterm labor [9]. Since African American women are at greater risk for vaginal dysbiosis and PTB than white women [9], it is important to understand any difference in vaginal microbiota of women of African vs. European ancestry. Ravel et al. [10] used 16S rRNA gene sequencing to analyze 98 vaginal swabs from European women and 104 vaginal swabs from African American women and classified the corresponding samples into five major groups termed Community State Types (CST). Four CSTs have predominantly Lactobacillus, including CST I (Lactobacillus crispatus), CST II (Lactobacillus gasseri), CST III (Lactobacillus iners), and CST V (Lactobacillus jensenii). CST IV comprises strict anaerobes that are often associated with bacterial vaginosis (BV) such as Prevotella, Gardnerella, Sneathia and Atopobium species. In that study, CST I was the most common CST among European women while CST IV was the most common in African American women [10]. Other studies have also shown that CST IV is more common in women of African descent than those of European descent [11, 12]. However, one study showed that the difference between white and black women disappeared when women with evidence of BV by Nugent’s criteria were excluded [13]. Attempts to determine associations between PTB and specific CST types or other designations of vaginal microbiota have also produced differing results. These studies of women with PTB have shown an association of PTB for Caucasian women with an increased Shannon Diversity Index [14], no correlation between CST and PTB in African American women [2], or an association of CST IV (Lactobacillus-poor) with PTB that was more pronounced with the presence of Gardnerella or Ureaplasma [15]. A recently reported meta-analysis using sequence data from five studies [2, 14–17] found that the vaginal microbiota from women with preterm delivery showed greater within-sample variation than those with term delivery and was found across racial groups [8]. They also found that three genera; Gardnerella, Lactobacillus, and Aerococcus were associated with third trimester preterm birth. Data available at the time of initiating the study suggested the presence of a distinct placental microbiota [18], and also raised the question of whether there was an association between placental microbiota and the occurrence of PTB. A better comprehension of the changes in vaginal microbiota during pregnancy could pave the way to predictive diagnostics and focused treatments of the complications associated with the intricate process of pregnancy, labor and birth. In the current study, we used a cohort study to determine whether there was a difference in the microbiota of women presenting with preterm labor compared to full term. In addition, we analyzed the placental microbiota to investigate any potential associations with preterm labor.
Aga Khan University Hospital (AKUH) is a 280-bed teaching hospital in Nairobi, Kenya that is accredited by the US-based Joint Commission International and has a full range of obstetric and neonatal services. Approximately 3600 deliveries per year are performed.
Pregnant women over the age of 18 years presenting in active labor or with preterm pre-labor rupture of membranes (PPROM) between 26 and 36 6/7 weeks gestation were recruited into the study from March 2018 to March 2019. Patients were excluded if they had medically indicated preterm delivery (for example preeclampsia, intrauterine growth restriction or congenital anomalies), antibiotics given more than 24 h prior to enrollment or within the last 4 weeks, cervical cerclage, progesterone supplementation, or HIV infection. A control group of mothers matching the study group as closely as possible for age and parity but presenting in labor at term (37 completed weeks) were enrolled in a 1:1 ratio. We considered a pregnancy to be normal if there were no obstetric or medical complications. Comparisons between the cases (preterm) and controls (term) were made using the Chi-square test and for nonparametric data, the Mann–Whitney test was used.
A physician or midwife collected the vaginal samples under direct visualization by swabbing the posterior vaginal fornix 3 to 5 times using sterile Snappable Polystyrene & Viscose Amies Swabs (Deltalab, Barcelona, Spain). Samples were stored at -80 ˚C until testing. Genomic DNA extraction was carried out using QIAamp DNA Mini Kit (Qiagen, Germany) as per manufacturer’s protocol. The placenta was collected into a clean ziplock bag after delivery and immediately transferred to a dedicated 4 °C refrigerator. Aseptically, a placental sample was cut from both fetal and maternal internal structures to minimize the risk of surface contamination. The samples were transferred to a -80 °C freezer for storage until DNA extraction. DNA extraction was carried out using Dneasy Blood & Tissue Kit (Qiagen, Germany) as per manufacturer’s protocol. Extracted DNA samples were shipped to the Dalhousie Integrated Microbiome Resource (IMR, Halifax, Nova Scotia, Canada) for sequencing. The protocol for sequencing is described at https://imr.bio/protocols.html#library; in brief, extracted DNA was amplified using PCR, targeting the conserved 16S ribosomal RNA gene. PCR primers amplified the V6/V7/V8 variable regions of the gene, providing over 400 nucleotides to use for species identification. Amplified DNA libraries were sequenced using an Illumina MiSeq machine. Sequencing runs were stored as FASTQ files; vaginal files with at least 2,000 associated reads were retained for subsequent analysis, while the minimum threshold for inclusion of placental samples was 100 reads.
Downstream analysis of DNA sequence data was performed using QIIME2 2022.2 [19]. Sequences were denoised using DADA2 [16] version 2022.2.0, with left and right truncation lengths of 280 and 270 nt, respectively. Primers were trimmed in both directions. Taxonomic assignment was performed as follows: reads were classified with the Naïve Bayes classifier using the SILVA version 138 reference database, with a minimum confidence score of 0.7 required to make a classification at a given taxonomic level. Taxonomic distributions were visualized using the “barplot” command of the “taxa” plugin. Community state assignments were based on the dominant Lactobacillus species for CST I (L. crispatus), II (L. gasseri), III (L. iners), and V (L. jensenii); samples that were dominated BV-associated taxa such as Prevotella, Gardnerella, Sneathia and Atopobium were assigned to CST IV. Tests for significant differences for the control vs. pre-term cohorts were performed using ALDEx2 [20], which addresses the issue of compositionality using the centered log-ratio transformation. Effect sizes and p-values were calculated using the QIIME2 “q2-aldex2” plugin’s “effect_plot” command, which computes both the Welch’s t-test and Wilcoxon test with Benjamini–Hochberg correction for multiple hypotheses. Alpha diversity values were computed for all samples using the Faith’s phylogenetic diversity and Shannon entropy measures, rarefaction curves were generated, and group differences between case and control samples were tested by the nonparametric Kruskal–Wallis test.
Ethics approval was obtained from the AKUH Ethics Review committee (2017/REC-86). Samples were collected only once during routine evaluation of women presenting with labor. For women in early labor, a written consent was followed by sample collection. For those in active labor, verbal assent was sought during labor for the sample collection; then written consent was sought after delivery. If the written consent was denied, the samples were discarded.
A total of 98 patients were recruited for the study between March 2018 and March 2019. Of these, 49 were patients with preterm labor who met the criteria and 49 were matched term controls (Fig. 1). The mean age of the participants was 32.2 with 24 years being the minimum age and 44 years as the maximum with no significant differences between the case and control groups (Table 1). There were 44 (89.8%) Africans, three (6.1%) Caucasians and two (4.1%) East Asians in the preterm group, and 48 (98%) Africans and one Caucasian in the control group. Most patients were nonvegetarian (n = 94, 95.9%) and did not have a history of prior PTB (n = 86, 87.7%). The majority of the controls (n = 45, 91.8%) were delivered vaginally, in comparison with only 18 (36.7%) in the preterm group. In addition, there was a significant difference in the gestational age by days between the two groups. The preterm group had a mean gestational age of 224.6 while the term group had a mean of 276.9 (Table 1).
A total of 100 vaginal and 71 placental samples were sequenced from the mothers in the case and control groups. The average read count was 181 and 26,798 per sample for placental and vaginal samples, respectively, after primer trimming, quality filtering, overlap assembly, and chimera removal. Vaginal samples with fewer than 2,000 reads were excluded from downstream analysis, leaving 74 vaginal samples (average of 35,932 reads per sample, total of 2,658,997 reads). Rarefaction curves of the vaginal samples suggest sequencing depth was sufficient to capture the majority of abundant taxa (data not shown), and a test of group differences revealed no significant difference in alpha diversity between case and control vaginal samples (Fig. 2). A total of 13 placental samples had read counts > 100 and were retained in the final data set.
Based on the SILVA taxonomic classification we observed three distinct species of Lactobacillus with an average abundance > 0.1% across all 74 high recovery vaginal samples: L. iners (43 individuals, 22.3% relative abundance), L. crispatus (50 individuals, 37.5% relative abundance), and L. jensenii (20 individuals, 0.3% relative abundance). An additional six named species were observed in lower abundance, most notably L. vaginalis which was found in 34 individuals but with an average abundance of only 0.16% (Fig. 3a and 4). Vaginal samples with L. crispatus tended to contain no other named species of Lactobacillus, while L. iners was found either alone or in association with L. jensenii as a minor component of the sample (Fig. 4). No amplicon sequence variants (ASVs) had a differential abundance that was significant between pre-term and term birth according to ALDEx2; the smallest Benjamini–Hochberg corrected p-value was 0.843. At the genus level, 73 out of 74 vaginal samples contained at least a small number of reads that were assigned to Lactobacillus, with a mean of 64.4% across all samples (Fig. 3b-c). Vaginal samples not dominated by Lactobacillus (such as subject 1036 with zero Lactobacillus reads; Fig. 4b) were generally dominated by genera commonly associated with BV (Fig. 4). Gardnerella and Prevotella were each found in 39 and 50 samples, respectively, with an average abundance of 11.1% and 8.8%. Other genera were found in relatively few samples, although often with high abundance: for example, Sneathia had a maximum abundance of 53.05% across 15 samples, while Pseudorhodobacter was present in only two samples but with an abundance of 65.49% in one sample. Conversely, several genera were found in many samples but at consistently low levels, including Dialister (33 samples; max abundance = 6.24%), Corynebacterium (25 samples, max abundance = 2.64%), and Atopobium (22 samples, max abundance = 8.77%). Streptococcus was found in 21 samples (10 case, 11 control) with an average abundance of 1.9% and a maximum of 82.75%. Of the 23 case–control pairs, all fell into three of the originally described CSTs [10], types I, III and IV (Table 2). There was no clear difference between the case and control groups in their CST assignment. When all 73 of the sequenced specimens were included (43 preterm and 30 term) whether or not they were matched, the results were similar with the same three CSTs dominating (Table 2). In this larger group, there were 28 individuals that could be considered as a part of CST I (L. crispatus), 18 pre-term birth cases and 10 controls; 17 individuals associated with CST III (L. iners), 12 pre-term birth cases and 5 controls; 1 individual associated with CST V (L. jensenii), a control; and 27 individuals associated with CST IV, 13 pre-term birth cases and 14 controls (Fig. 3). In this study’s cohort, L. crispatus was not associated with term birth and, conversely, a significant number of the cases and controls had a predominance of Gardnerella and/or Prevotella, but no association with preterm labor (demonstrated by the lack of significantly differentially abundant ASVs and the mixture of cases and controls in each.
Of the 71 placental samples (Fig. 5), only 13 (17.6%) had more than 100 reads that passed the quality threshold, and only two had more than 2,000. The 13 samples split nearly evenly between cases (5/13) and controls (8/13). Many samples with fewer than 100 sequences were dominated either by poorly classified reads that mapped only to “Bacteria” or “Phylum OD1”. Forty-four samples had at least one sequence that was classified at a lower taxonomic rank; twelve of these had classified reads that mapped only to Lactobacillus. Lactobacillus was not identified in an additional twelve samples. However, five placental samples showed evidence of Streptococcus with abundance between 4.4% and 80.5%; Streptococcus agalactiae was identified in a previous study as the only species that could be confidently recovered [21]. The sample with the highest percentage of Streptococcus (Case 1003) had a corresponding vaginal abundance of 0.78%; a rectal swab taken from the neonate immediately after birth yielded 99.6% Streptococcus, with the remaining sequence reads assigned to Enterobacteriaceae.
The present study is the first gene sequencing‐based vaginal microbiota study to date in Kenya with a case–control design comparing the vaginal microbiota of between women with spontaneous preterm labor with those who went to full term. The major objective was to identify whether there were any vaginal microbiota patterns associated with preterm labor in the Kenya population. We found no difference in CSTs between cases with preterm labor and controls with term labor. Similar to other studies of the vaginal microbiota, we observed vaginal microbial communities with a high incidence of species within the genus Lactobacillus; however, the number of distinct species groupings according to our ASV analysis was small, with the predominant species identified being L. iners and L. crispatus (Fig. 5). The ecological significance of these associations is unclear, and future metagenomic analysis may yield insights into the patterns we describe here. Although Lactobacillus was widely distributed across subjects as expected, a substantial number of both case and control samples had substantial counts of other genera, with 20/46 paired samples having a Lactobacillus relative abundance < 50%. These samples were dominated by bacteria such as Prevotella and Gardnerella that are frequently associated with BV. BV has long been associated with PTB and treatment with metronidazole has been used to prevent PTB [22]. However, treatment of asymptomatic BV did not reduce PTBs [23]. Thus, it is of interest to determine whether any of the five CSTs or individual organisms are associated with PTB, especially for CST IV or Gardnerella. Indeed, distinct taxa have been associated with PTB in a number of studies. In support of this possibility, a study of Indian women showed that L. iners, Megasphaera, G. vaginalis, and Sneathia sanguinegens were higher in women presenting in preterm labor, while L. gasseri was higher in those presenting at term [24]. In addition, L. crispatus has been protective in other studies [25] and the suggestion that L. crispatus is incompatible with G. vaginalis has supported the idea of a protective effect of L. crispatus [26]. A study of vaginal metabolites and preterm labor in the setting of a mostly white population of British women also suggested a protective effect of L. crispatus and an association of preterm labor with L. jensenii [27], while the Peruvian study noted above showed no association [28]. Some studies of women of African ancestry have found an association of PTB with certain taxa [29], while others have not [2, 17]. In addition, there is evidence for an increased frequency of non-Lactobacillus-related CSTs in women of African ancestry, but not necessarily associated with adverse outcomes [30]. Other microbial associations have also been described, including a study of Korean women that showed an association of Klebsiella in the vaginal microbiota and preterm labor [31]. Klebsiella is part of the Enterobacteriaceae class, which was not associated with preterm labor in our study. In the current study, nearly all the women fell into CST I, III, or IV, but the proportions of preterm and term did not show major differences for these three types. Our study also found no trend for an association with the presence of sequences from the genera Lactobacillus, Gardnerella, and Aerococcus that were associated with third semester PTB in a meta-analysis [8]. However, it is possible that certain relevant associations were missed in the current study in view of the relatively small number of matched cases and controls. In addition to the lack of difference of CST for preterm vs. term delivery, there was also no difference between the two groups for measures of alpha diversity. Our study also addressed the question of whether a placental microbiome is associated with PTB. This is especially important since some studies have suggested a placental microbiome or an association with certain outcomes [18], while other studies have found no evidence of a specific placental microbiome [32, 33]. In our study, the read counts and taxonomic affiliations of our placental samples were largely consistent with very low bacterial loads that have been reported elsewhere in the literature [34]. Many samples either returned no reads or were dominated by common vaginal flora (most notably Lactobacillus), unclassified sequences, or eukaryotic sequences and the Bradyrhizobium group that are likely contaminants. The very low read recovery in many placental samples may reflect difficulty in recovering viable DNA samples from placental matter. However, the presence of organisms including Lactobacillus and Veillonella suggests the more likely explanation that many “placental” samples are dominated by bacteria acquired during passage through the birth canal. These observations are consistent with reports that suggest placentally derived bacteria (i.e., a “placental microbiome”) are rare [32–36]. Thus, our observations are consistent with the evidence that there is not normally a separate placental microbiota. However, the few placental samples with the highest sequence recovery were often dominated by Streptococcus, which is interesting in light of the common association of S. agalactiae (Group B streptococci) with adverse maternal and neonatal outcomes [37].
In summary, these results contribute to the increasing data that shows that there is a spectrum of diversity in the vaginal microbiota without clear evidence of specific microbiota types that have a correlation with preterm labor. Therefore, an understanding of the variables associated with African ethnicity that contribute to this diverse microbiota has important implications regarding reproductive health outcomes. This study is a fundamental step towards gathering more information on the relationship of vaginal microbiota and PTB which would help us to establish a greater degree of accuracy on future implications of this portentous relationship. |
|
PMC9648022 | Jianwei Zhang,Silu Meng,Xiaoyan Wang,Jun Wang,Xinran Fan,Haiying Sun,Ruoqi Ning,Bing Xiao,Xiangqin Li,Yao Jia,Dongli Kong,Ruqi Chen,Changyu Wang,Ding Ma,Shuang Li | Sequential gene expression analysis of cervical malignant transformation identifies RFC4 as a novel diagnostic and prognostic biomarker | 09-11-2022 | Squamous intraepithelial lesions,Cervical cancer,Molecular changes,RFC4,Biomarkers | Background Cervical squamous cell carcinoma (SCC) is known to arise through increasingly higher-grade squamous intraepithelial lesions (SILs) or cervical intraepithelial neoplasias (CINs). This study aimed to describe sequential molecular changes and identify biomarkers in cervical malignant transformation. Methods Multidimensional data from five publicly available microarray and TCGA-CESC datasets were analyzed. Immunohistochemistry was carried out on 354 cervical tissues (42 normal, 62 CIN1, 26 CIN2, 47 CIN3, and 177 SCC) to determine the potential diagnostic and prognostic value of identified biomarkers. Results We demonstrated that normal epithelium and SILs presented higher molecular homogeneity than SCC. Genes in the region (e.g., 3q, 12q13) with copy number alteration or HPV integration were more likely to lose or gain expression. The IL-17 signaling pathway was enriched throughout disease progression with downregulation of IL17C and decreased Th17 cells at late stage. Furthermore, we identified AURKA, TOP2A, RFC4, and CEP55 as potential causative genes gradually upregulated during the normal-SILs-SCC transition. For detecting high-grade SIL (HSIL), TOP2A and RFC4 showed balanced sensitivity (both 88.2%) and specificity (87.1 and 90.1%), with high AUC (0.88 and 0.89). They had equivalent diagnostic performance alone to the combination of p16INK4a and Ki-67. Meanwhile, increased expression of RFC4 significantly and independently predicted favorable outcomes in multi-institutional cohorts of SCC patients. Conclusions Our comprehensive study of gene expression profiling has identified dysregulated genes and biological processes during cervical carcinogenesis. RFC4 is proposed as a novel surrogate biomarker for determining HSIL and HSIL+, and an independent prognostic biomarker for SCC. Supplementary Information The online version contains supplementary material available at 10.1186/s12916-022-02630-8. | Sequential gene expression analysis of cervical malignant transformation identifies RFC4 as a novel diagnostic and prognostic biomarker
Cervical squamous cell carcinoma (SCC) is known to arise through increasingly higher-grade squamous intraepithelial lesions (SILs) or cervical intraepithelial neoplasias (CINs). This study aimed to describe sequential molecular changes and identify biomarkers in cervical malignant transformation.
Multidimensional data from five publicly available microarray and TCGA-CESC datasets were analyzed. Immunohistochemistry was carried out on 354 cervical tissues (42 normal, 62 CIN1, 26 CIN2, 47 CIN3, and 177 SCC) to determine the potential diagnostic and prognostic value of identified biomarkers.
We demonstrated that normal epithelium and SILs presented higher molecular homogeneity than SCC. Genes in the region (e.g., 3q, 12q13) with copy number alteration or HPV integration were more likely to lose or gain expression. The IL-17 signaling pathway was enriched throughout disease progression with downregulation of IL17C and decreased Th17 cells at late stage. Furthermore, we identified AURKA, TOP2A, RFC4, and CEP55 as potential causative genes gradually upregulated during the normal-SILs-SCC transition. For detecting high-grade SIL (HSIL), TOP2A and RFC4 showed balanced sensitivity (both 88.2%) and specificity (87.1 and 90.1%), with high AUC (0.88 and 0.89). They had equivalent diagnostic performance alone to the combination of p16INK4a and Ki-67. Meanwhile, increased expression of RFC4 significantly and independently predicted favorable outcomes in multi-institutional cohorts of SCC patients.
Our comprehensive study of gene expression profiling has identified dysregulated genes and biological processes during cervical carcinogenesis. RFC4 is proposed as a novel surrogate biomarker for determining HSIL and HSIL+, and an independent prognostic biomarker for SCC.
The online version contains supplementary material available at 10.1186/s12916-022-02630-8.
Cervical cancer is the fourth most common cancer in females, with 604,127 new cases and 341,831 deaths estimated for 2020 worldwide [1]. Squamous cell carcinoma (SCC) is the predominant histological type of cervical cancer, with adenocarcinoma (AC) occurring less frequently [2]. Persistent high-risk human papillomavirus (HR-HPV) infection is associated with the development of cervical intraepithelial neoplasia (CIN), if untreated, which may progress to SCC over a period of 15 to 20 years [3]. Currently, a two-tier system of low- and high-grade squamous intraepithelial lesions (LSIL and HSIL) paralleling the terminology of the Bethesda System cytologic reports was recommended to replace the old CIN classification by World Health Organization (WHO) [4]. Cervical carcinogenesis is a complex process occurring as a consequence of multiple genomic alterations. Several expression microarray studies have been conducted investigating transcriptome changes in this process. Some research focused on specific dysregulated genes mediating the invasion of cervical cancer cells [5, 6]. Other research was designed to identify molecular changes that drive cervical cancer development [7–9]. Of note, studies based on next-generation sequencing are rare, probably due to ethical reasons and difficulties in obtaining tissue samples. Driven by the need for a comprehensive molecular characterization of the carcinogenic process, we performed a meta-analysis on publicly available gene expression profiles for an in-depth study. This study is also motivated by the clinical desire to develop novel biomarkers of cervical carcinogenesis. On the diagnostic front, early detection of HSIL and subsequent surgical intervention are necessary to prevent further progression [10]. However, the inter- and intra-observer reproducibility of SIL grade evaluation is often poor among different pathologists due to mimics of HSIL (e.g., atrophy, LSIL, and therapy changes) [11–13]. In routine pathology practice, p16INK4A and Ki-67 are the most commonly used biomarkers of HR-HPV infection and cell proliferation, respectively. It has been demonstrated that p16INK4a can distinguish HSIL from its mimics and improve the diagnostic consistency of precancerous lesions among pathologists [14, 15]. Nonetheless, p16INK4a has a certain positive rate in normal cervical tissue, cervicitis, and LSIL, which limits specificity for detecting HSIL [16–18]. On the prognostic front, although the incidence and mortality of cervical cancer are decreasing due to increased global vaccination and screening coverage, clinical outcomes of patients with advanced-stage or recurrence disease are still bleak and difficult to predict [19]. Driven by the need for effective biomarkers to improve the diagnosis of HSIL and the prognosis of SCC, we specifically focused on screening persistently altered genes involved in carcinogenesis.
The overall workflow of the present study is shown in Fig. 1. Briefly, the study was undertaken in two parts. First, an integrative bioinformatic analysis of gene expression microarray datasets was conducted to identify molecular changes and hub genes linked to SCC progression. Second, external datasets and multi-institutional cohorts were used to further validate the diagnostic and prognostic robustness of selected genes from the first step.
As of Jun 06, 2020, we performed a systematic search of the Gene Expression Omnibus (GEO) and ArrayExpress databases. The inclusion criteria were as follows: (i) the datasets of mRNA expression profile; (ii) human tissue samples containing at least three disease stages from normal, LSIL (CIN1), HSIL (CIN2-3), and SCC; (iii) at least 25 samples. Four microarray datasets were included, of which three Affymetrix-based datasets (GSE63514 [7, 20], GSE27678 [5, 21], and GSE7803 [6, 22]) were used as discovery datasets. The remaining Agilent-based dataset (GSE138080 [8, 23]) and another prospective study (GSE75132 [24, 25]) were selected as validation datasets. The characteristics of the microarray datasets are summarized in Additional file 1: Table S1. The series matrix files and detailed information of array experiments were downloaded. All gene expression data had already been normalized, and a log2-based transformation was applied if the data were not log2 transformed. Boxplots of normalized microarray data can be seen in Additional file 1: Fig. S1, which showed an essentially similar distribution of expression profiles among the samples in each dataset. Afterward, the probes were mapped to genes. Genes with multiple probes were represented by the probe with the highest mean expression level. The clinical and molecular data (including mRNA expression and copy number) of primary cervical cancer patients were retrieved from The Cancer Genome Atlas (TCGA) database through the R package TCGAbiolinks [26]. We downloaded gene expression quantitated as fragments per kilobase of transcript per million mapped reads upper quartile (FPKM-UQ) and Masked Copy Number Segment data generated by Affymetrix SNP 6.0 array. Moreover, we also downloaded the survival information from TCGA Pan-Cancer Clinical Data Resource [27].
The tissue specimens (n = 420) were obtained from Fanpu Biotech. Co., Ltd. (FBC; Guilin, China), Department of Gynecological Oncology of Tongji Hospital (TJH; Wuhan, China), and Outdo Biotech. Co., Ltd. (OBC; Shanghai, China) (Additional file 1: Table S2). This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee and Institutional Review Board of the three institutions above. Informed consent was obtained from all participants. Hematoxylin-eosin (H&E)-stained sections were reviewed by two independent pathologists (X.Y.W. and J.W.) blinded to the original diagnoses. A unanimous or majority diagnosis defined as agreement by at least two out of three diagnoses (original and the two new) was recognized as the final “gold standard” diagnosis. Finally, a total of 354 samples met the criteria for further study (Additional file 1: Fig. S2).
Differentially expressed gene (DEG) analysis was conducted between lesions and normal tissue (LSIL/HSIL/SCC vs. Normal, respectively) using the R package limma [28], with the criteria of Benjamini-Hochberg (BH) adjusted p-value < 0.05 and absolute fold change > 2. “Cross-study” longitudinal analysis was performed to obtain Gene Sets1, which referred to the union of upregulated genes and the union of downregulated genes per comparison after removing DEGs exhibiting inconsistent direction of change in any two discovery datasets (Fig. 1, Additional file 1: Table S3). The significance of copy number variations (CNVs) was assessed from the segmented data using GISTIC2.0 in GenePattern [29, 30]. Gene-level copy number values and frequency of CNVs were extracted for further analysis. The OmicCircos package in R [31] was utilized to visualize the expression patterns of the Sets1 genes according to a gradient of disease severity. The R package clusterProfiler [32] was used for the enrichment of Gene Sets1 and total DEGs. Cytoband enrichment was performed using positional gene sets (C1 collection available from MSigDB) with a q-value < 0.05. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were carried out to determine the biological significance with a q-value < 0.1. Selected pathway was visualized through the R package Pathview [33]. For GO enrichment in terms of biological process, we used the simplify function of clusterProfiler to remove redundant enriched terms. Then the R package GOSemSim [34] was used to estimate the pairwise semantic similarity between simplified terms for clustering. The abundance of Th17 cells was assessed with Immune Cell Abundance Identifier (ImmuCellAI) [35]. “Cross-stage” horizontal analysis was performed to obtain Gene Sets2, which referred to the intersection of DEGs and “stepwise genes” (showed gradually increasing/decreasing expression with progression of cervical lesions) per discovery dataset (Fig. 1, Additional file 1: Table S3). The protein-protein interaction (PPI) networks for Gene Sets2 were constructed with the STRING database and visualized with Cytoscape (version 3.7.1) [36, 37]. Then, we used the plug-in cytoHubba (version 0.1) [38] to rank and explore essential nodes in the interactome network. The top 10 nodes from each of the nine ranking methods (Betweenness, Bottleneck, Closeness, Degree, EPC, DMNC, MNC, Radiality, and Stress) were collected per study, and sub-PPI networks were established based on them. Any overlap in three sub-PPI networks was regarded as the “Hub Gene” of the present study.
Firstly, the tissue sections were baked at 65 °C for 30 min and then deparaffinized in xylene and passed through graded alcohol followed by antigen retrieval with 1 mM EDTA, pH 9.0 (G1203, Servicebio, Wuhan, China) in a microwave at 50 °C for 10 min, and then 30 °C for 10 min. The sections were incubated in 3% H2O2 for 25 min to quench endogenous peroxidase activity and then washed carefully in phosphate-buffered saline (PBS, pH 7.4) three times. 3% bovine serum albumin (G5001, Servicebio, Wuhan, China) was added onto the sections to cover the tissue evenly and incubated for 30 min at room temperature. The sections were subsequently incubated with the diluted antibodies (p16INK4a, Ki-67, AURKA, TOP2A, RFC4, CEP55) overnight at 4 °C. The details of antibodies are summarized in Additional file 1: Table S4. After carefully rinsing the sections with PBS, the sections were treated using the Pika general antibody (G1211, Servicebio, Wuhan, China; horseradish peroxidase-conjugated rabbit/mouse antibody) for 50 min, followed by diaminobenzidine (G1211, Servicebio, Wuhan, China) to detect expression under the microscope. Finally, the sections were counterstained with hematoxylin, dehydrated, and covered. Immunohistochemical interpretations were performed independent of the H&E diagnosis by the two pathologists mentioned above. Unqualified sections were firstly discarded, and the remaining sections were evaluated for positive or negative staining, stained cellular compartment (Additional file 1: Fig. S2). For noninvasive squamous epithelia, a summary of the immunohistochemical scoring system is given in Additional file 1: Table S5. Put simply, p16INK4a immunopositivity was determined following the modified version of the criteria described by Darragh et al. [39]. The cell-layer level of Ki-67 and TOP2A expression was evaluated (parabasal layer, 0; lower third of the epithelium, 1+; lower two thirds, 2+; more than lower two thirds up to full thickness, 3+). The TOP2A scores of 2+ and 3+ were grouped as 2+. AURKA and CEP55 expressions were evaluated for the staining intensity (no staining, 0; weak, 1+; moderate, 2+; strong, 3+). RFC4 expression was evaluated based on staining intensity and distribution. For SCC, samples with > 10% positive cancer cells were considered positive for all markers. Meanwhile, AURKA, TOP2A, and RFC4 staining in SCC were assessed using the semi-quantitative histologic score (HSCORE) system. The staining intensity (0, 1+, 2+, or 3+) of cells and percentage (0–100%) of cells at each staining intensity level were estimated. The HSCORE was assigned using the following formula: HSCORE = [1 × (% cells 1+) + 2 × (% cells 2+) + 3 × (% cells 3+)], with a ranking between 0 and 300. Only the staining intensity of CEP55 was estimated for its diffuse staining in SCC.
Three independent cohorts of SCC patients (TCGA, n = 252; TJH, n = 56; OBC, n =106) were included for survival analysis. The clinical and pathological characteristics of the cohorts are summarized in Additional file 1: Table S6. The analysis consists of three steps. Firstly, patients in the TCGA cohort were dichotomized according to the optimal cutoff value for FPKM-UQ of hub genes. The association of gene expression (mRNA) with clinical outcomes was evaluated. Secondly, patients in the TJH cohort were dichotomized according to the optimal cutoff value for HSCORE or staining intensity of hub genes to validate the association at the protein level. Finally, the OBC cohort was entered into the TJH cohort to construct an Extended cohort (n = 162) for more reliable validation of the selected gene. Different clinical outcome endpoints, overall survival (OS), progress-free interval (PFI), and disease-free survival (DFS), were defined in three cohorts. In the TCGA cohort, OS was defined as the time from initial diagnosis until death from any cause. PFI was defined as the time from initial diagnosis until recurrence of tumor, including locoregional recurrence, distant metastasis, new primary tumor, or death with tumor [27]. In the TJH and OBC cohorts, OS was defined as the time from primary surgery or the last day of therapy if no surgery until death from any cause. DFS was defined as the time from primary surgery or the last day of therapy if no surgery until recurrence of tumor or death from any cause. Patients who did not experience the event of interest were censored at the date of the last available follow-up or 5 years (whichever came first). Survival curves were plotted using the Kaplan-Meier method and compared using the log-rank test. Multivariate Cox regression analysis was performed to determine the prognostic value of the selected gene with considering clinical factors. The R packages survival, survminer, and timeROC [40] were utilized to perform the survival analysis and visualization.
All statistical analyses were performed in the R statistical computing environment (version 3.5.3). Either Pearson’s or Spearman’s correlation coefficients were calculated to ascertain bivariate correlations. Prior to comparison, data normality was evaluated by Shapiro-Wilk test, and homogeneity of variances was evaluated by Levene’s test. Student’s t test, Wilcoxon rank-sum test, one-way ANOVA test, and Kruskal-Wallis test were used for numerical variables. Chi-square test and McNemar’s test were used for independent and paired categorical variables, respectively. For analysis of the association between IHC findings of biomarkers and histological evaluation, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic (ROC) curve (AUC) were calculated. Cohen’s kappa coefficient (κ) was calculated to determine the agreement between tested IHC markers. Survival analysis has been described above. All p-values were two-sided, with p < 0.05 indicating statistical significance unless otherwise stated.
The principal component analysis (PCA) showed that normal and LSIL were not clearly discriminated in GSE63514. However, there was a distinct separation between normal and other disease stages in other datasets. The inconsistent result may be due to the different HPV statuses in normal tissue. LSIL showed higher similarity with normal epithelium with HR-HPV infection than those without HR-HPV infection (Fig. 2A, Additional file 1: Table S1). Pairwise correlation coefficients of gene expression profiles were calculated for cases at the corresponding stage to measure intralesional heterogeneity. In GSE63514, we observed significantly lower correlations for SCC as compared with HSIL, LSIL, and normal (mean Pearson’s r: 0.91 vs. 0.94 vs. 0.94 vs. 0.93, all p < 0.05), reflecting a molecular intralesional heterogeneity in SCC. Similar results were obtained from GSE7803 (Fig. 2B). Correlation heatmap of pairwise samples indicated a certain similarity between noninvasive squamous epithelium samples (Fig. 2C).
To further explore the transcriptional landscape during carcinogenesis, we performed differential expression analyses between LSIL, HSIL, SCC, and normal tissues (LSIL/HSIL/SCC vs. Normal: LN, HN, CN), respectively. The number of identified DEGs was limited in LN, followed by an increase in progression (Fig. 2D). On the one hand, we observed a reasonable consensus representation of DEGs across discovery datasets in HN and CN, indicating results reliability. On the other hand, GSE63514 displayed the highest proportion of study-specific DEGs (HN: 78.5%, CN: 84.0%), about a third of which could be explained by platform-specific genes (Fig. 2 E, F). Given the aforementioned reasons, we applied “cross-study” analysis to generate the union of up- and downregulated genes separately for each comparison (including LN_UP/DN, HN_UP/DN, and CN_UP/DN, termed Gene Sets1) to combine data from multiple independent microarray datasets. Chromosome mapping of Gene Sets1 revealed a genome-wide distribution, and regions with high-frequency chromosomal aberrations contained more specific DEGs (Fig. 2G). The deregulated genes were significantly clustered on 1q21, 1q32, 2q31, 3q, 8q13, 12q13, 17q21, 19q13, and 19p13, which correspond to previously reported HPV integration or CNV regions linked to cervical cancer. Interestingly, 1q21 and 19q13 previously described as amplification regions showed enrichment of downregulated genes. Enriched chromosome bands 3q13 and 19q13 were observed in HN and CN and only 17q21 in LN (Fig. 2H, Additional file 2: Table S7).
GO enrichment of Gene Sets1 and total DEGs revealed a similar enrichment pattern of HN and CN. Eight distinct clusters were determined, and each was assigned a unique enrichment signature (Fig. 3A). We observed that upregulated genes of HN and CN showed a strong enrichment for cell cycle and related terms (Clusters 1, 3, 4). Intriguingly, the positive regulation of cell cycle term was enriched with LN_DN, HN_UP/DN, and CN_UP genes, indicating a dynamic regulation of cell cycle control. Consistent with dysmaturation of keratinocytes throughout disease progression, downregulated genes of HN and CN showed significant enrichment for cornification term (Cluster 8). In addition, LN_UP genes were not enriched in any term, while LN_DN genes were mainly enriched in Clusters 1, 5–8 (Fig. 3A, Additional file 2: Table S8). KEGG enrichment of total DEGs showed a great agreement between HN and CN. While in the enrichment of downregulated genes, HN appears to be a transitional state since the enriched pathways of HN were partially overlapped with that of CN and LN (Additional file 1: Fig. S3). DNA repair pathways [homologous recombination (HR), base excision repair (BER), nucleotide excision repair (NER) and mismatch repair (MMR)], cell cycle-related pathways (cell cycle, DNA replication, and cellular senescence), and oncogenic p53 signaling pathways were enriched with upregulated genes of HN and CN (Fig. 3B, Additional file 2: Table S9). Downregulated genes of LN and HN were enriched in more cancer-related pathways (e.g., MAPK, ErbB, PI3K/AKT, TGF-β, and Wnt signaling pathways) than that of CN. Some pathways above, such as PI3K/AKT and Wnt pathways, could also be enriched in CN when considering total DEGs (Fig. 3B, Additional file 2: Table S9). Human T-cell leukemia virus 1 (HTLV-1) infection and immune-related IL-17 signaling pathway were enriched in all disease stages among DEGs (Fig. 3B). Many DEGs of HN and CN enriched in HTLV-1 infection signaling were cell cycle-related genes, and most of which are upregulated. Ten genes in HTLV-1 infection signaling were differentially expressed in LN, and all serum response factor (SRF) pathway genes (SRF, FOS, FOSL1, EGR1, and EGR2) were involved and downregulated (Additional file 1: Fig. S4, Additional file 2: Table S9). IL17 cytokine family comprises IL17A, IL17B, IL17C, IL17D, IL17E (also known as IL25), and IL17F. Among these cytokine genes, only IL17C was founded to be differentially expressed (downregulated) in HN and CN (Fig. 3C, Additional file 2: Table S9). IL17C was produced primarily by keratinocytes, gradually replaced by abnormal epithelial cells in malignant transformation. Although Th17-associated cytokine genes (IL17A and IL17F) were not affected, we still observed that Th17 cells showed significantly decreased abundance at the late stage of cancer progression (Fig. 3D), which is inconsistent with previous findings [41].
To identify pivotal DEGs and provide clues for early diagnosis and intervention of precancerous lesions, we applied “cross-stage” analysis to generate the intersection of DEGs and “stepwise genes” for each discovery dataset (termed Gene Sets2). Through PPI network analysis of Sets2 genes, four genes (AURKA, TOP2A, CEP55, and RFC4) occurring in at least two datasets were considered as “Hub Genes” (Fig. 4A, Additional file 1: Fig. S5). Hub gene expression increased significantly during progression from normal to SCC (all p < 0.05, Additional file 1: Fig. S6), which were also validated in GSE138080 (all p < 0.05, Fig. 4B). Furthermore, these genes had been proved to be involved in the malignant transformation of several different types of tumors, like Barrett’s adenocarcinoma, colorectal carcinoma, head and neck squamous cell carcinoma (HNSCC), etc. As expected, they tended to display a linear expression pattern with tumor progression (Additional file 1: Table S10). The observed disease stage-dependent expression changes could be induced by multi-factors such as genetic and epigenetic alterations, small RNAs, and HPV integrations. This study assessed associations between hub gene expression level and corresponding copy number alterations (CNAs) in AC and SCC. We found a strong positive correlation between expression level and CNAs for RFC4 (Spearman’s rho: 0.70 and 0.80, all p < 0.05), and a moderate correlation for AURKA (Spearman’s rho: 0.65 and 0.44, all p < 0.05) and CEP55 (Spearman’s rho: 0.40 and 0.42, all p < 0.05) in AC and SCC (Fig. 4C). Interestingly, CNAs of TOP2A have a tumor subtype-specific role in contributing to gene expression variability. AC shows a tighter correlation than SCC between those two parameters (Spearman’s rho: 0.50 vs. 0.17, Fig. 4C). Moreover, we examined whether the expression levels of hub genes were predictive for developing high-grade cervical lesions in a prospective cohort study (GSE75132). The study enrolled HPV-negative and persistently HPV16-infected women. HPV-infected women were divided into progressor (HPV-P) and sustainer (HPV-S) groups according to whether they progressed to CIN3+ or not during follow-up (Additional file 1: Table S1). We noted a trend toward a higher median expression level of RFC4 and TOP2A in HPV-P women than in HPV-S and HPV-negative women. Their expression differences between HPV-P and HPV-negative women were significant or marginally significant, respectively (p = 0.034 for RFC4 and p = 0.08 for TOP2A, Fig. 4D).
Hierarchical clustering of hub genes revealed a good separation between normal/LSIL and HSIL+ (HSIL and SCC) (Fig. 5A). Thus, we performed IHC for p16INK4a, Ki-67, and four hub genes to validate the identified transcriptomic changes at the protein level and explore their clinical utility in diagnosing HSIL and HSIL+. Corresponding to changes in mRNA, the protein expression of all tested markers gradually increased from normal to CIN3 (Fig. 5B, Additional file 1: Table S11). Because of different scoring systems, we could not compare marker expression between normal/CINs and SCC by IHC score. However, the neoplastic epithelial cells of SCC showed stronger staining (increased positive intensity and more diffuse expression) than those of CIN3 according to subjective visual estimation (Fig. 5B). Cellular markers were classified into three categories based on their biological function, including cell cycle regulation marker (p16INK4a, AURKA, RFC4), cell division marker (CEP55), and proliferation marker (Ki-67 and TOP2A). Then we developed an individual scoring system to evaluate staining intensity and extent (Additional file 1: Fig. S7, Table S5). The positive rates of all markers significantly increased with the severity of cervical lesion, especially in HSIL and SCC (Fig. 5C). For p16INK4a and Ki-67, our results were in good correspondence with previous studies, which improved confidence in the reliability of IHC and comparability of evaluation. Meanwhile, the previously reported frequency of positive TOP2A and ProExC (MCM2 and TOP2A) in cervical lesions were also collected (Additional file 1: Table S12). The concordance analysis demonstrated a strong agreement of 90.2% (κ = 0.8) in the whole stages and 97.1% (κ = 0.84) in HSIL for Ki-67 and TOP2A (Fig. 5D, Additional file 1: Table S13). For detecting HSIL, sensitivities, specificities, PPVs, NPVs, and AUCs of six markers are shown in Additional file 1: Table S14. Of note, CEP55 and AURKA with high sensitivity and low specificity were excluded for further comparative analysis due to their cytoplasmic staining, which made assessment challenging and interpretation difficult. To statistically compare the diagnostic utility of the remaining markers, sections with these markers evaluated simultaneously were analyzed (Additional file 1: Table S15, Table 1). Data showed the highest sensitivity of p16INK4a, but the specificity and PPV were low. In contrast to p16INK4a, Ki-67 and TOP2A provided a nearly equivalent sensitivity (92.6% vs. 91.2% vs. 88.2%, all p > 0.05) and a higher specificity (63.4% vs. 80.2% vs. 87.1%, all p < 0.05). The inclusion of Ki-67 could improve specificity (87.1% vs. 63.4%, p < 0.05), PPV (82.2% vs. 63%), and accuracy (AUC, 0.88 vs. 0.78) of p16INK4a (Fig. 5E), at the little expense of sensitivity (92.6 to 88.2%) and NPV (92.8 to 91.7%). TOP2A had the same performance as p16INK4a and Ki-67 combined. Additionally, RFC4 not only had same sensitivity (88.2% vs. 88.2%, p = 1) and relatively higher specificity (90.1 vs. 87.1%, p = 0.58), but also higher accuracy (AUC, 0.89 vs. 0.88; Fig. 5E) than the combination of p16INK4a and Ki-67. Meanwhile, the diagnostic efficacy of RFC4 and TOP2A improved for detecting HSIL+ as the AUC reached 0.91 and 0.89, respectively (Table 1). These results suggest RFC4 and TOP2A alone could be complementary surrogate markers to p16INK4a and Ki-67 for detecting HSIL and HSIL+. Moreover, serial and parallel interpretation of any marker pairs were compared to TOP2A and RFC4 alone (Additional file 1: Table S16). The combination of TOP2A and RFC4 in parallel interpretation with the highest accuracy (AUC, 0.90) showed significantly higher sensitivity (97.1 vs. 88.2%, p < 0.05) but lower specificity (82.2 vs. 90.1%, p < 0.05) compared with RFC4 for detecting HSIL (Additional file 1: Table S16a). None of the combinations presented certain advantages over RFC4 for detecting HSIL+ (Additional file 1: Table S16b).
To further investigate the clinical impact of hub genes in SCC progression, we examined the correlation between their expression and disease severity in SCC patients from the TCGA cohort. The analysis revealed that increased AURKA mRNA expression was significantly or marginal significantly associated with advanced FIGO stage (p = 0.039), higher histological grade (p = 0.01), and lymph nodes metastasis (p = 0.088; Additional file 1: Fig. S8). Next, we evaluated the effect of hub gene expression alterations on prognosis in three independent cohorts of SCC patients (see the “Methods” section). In the TCGA cohort (n = 252), high AURKA mRNA expression inversely correlated with OS (log-rank, p = 0.017; Fig. 6A). While higher RFC4 mRNA expression was significantly associated with better OS and PFI (log-rank, p = 6.8e−04 and p = 7.8e−03, respectively; Fig. 6A, Additional file 1: Fig. S9A). To validate the finding at the protein level, we performed immunostaining for these genes on SCC tissues from the TJH cohort (n = 56). Overexpression of RFC4 protein showed significant associations with increased OS and DFS (log-rank, p = 3.1e−03 and p = 1.5e−03, respectively; Fig. 6B, Additional file 1: Fig. S9B). In addition, higher TOP2A protein expression significantly correlated with better DFS (log-rank, p = 0.019; Additional file 1: Fig. S9B), which was not found at the mRNA level. Although not statistically significant, there was a tendency for higher CEP55 expression associated with increased PFI (log-rank, p = 0.1) in the TCGA cohort and increased DFS (log-rank, p = 0.072) in the TJH cohort (Fig. 6B, Additional file 1: Fig. S9B). Due to limited samples of the TJH cohort, we further investigated the RFC4 prognostic value in the combined TJH and OBC cohort (Extended cohort, n = 162; Additional file 1: Fig. S10). The effect of RFC4 protein expression on OS and DFS, as expected, remained significant (log-rank, p = 1.6e−04 and p = 2.2e−04, respectively; Fig. 6C, Additional file 1: Fig. S9C). The representative IHC staining images of hub genes in SCC are shown in Fig. 6D. Notably, multivariate Cox regression analysis revealed that RFC4 expression (mRNA and protein), after adjustment for age, FIGO stage, grade, and cohort, emerged as an independent predictor of clinical outcomes (OS, PFI, and DFS) for SCC patients in three cohorts (Fig. 6E, Additional file 1: Fig. S9D). The time-dependent AUC showed that the addition of RFC4 expression into the Cox proportional hazards model significantly increased the prognostic efficacy for 2- and 3-year OS (all p < 0.05; Fig. 6F), for 1-, 2-, and 3-year DFS (all p < 0.05; Additional file 1: Fig. S9E).
This meta-analysis based on previous studies comprehensively characterizes the transcriptomic profiles of cervical carcinogenesis and identifies four key genes (AURKA, TOP2A, RFC4, CEP55) associated with the initiation and progression of SCC. Then, we carefully assess their diagnostic performance in HSIL/HSIL+ and prognostic performance in SCC. To the best of our knowledge, our study is the first to evaluate and validate the diagnostic and prognostic value of RFC4 in cervical lesions. We found that the transcriptomes of normal epithelium and SILs were homogenous. However, the increased heterogeneity was observed upon progression to SCC. A study of the transcriptomic landscape of hepatocarcinogenesis presented homogeneity in dysplastic lesions and early carcinoma but heterogeneity in advanced liver cancer, somewhat similar to our results [42]. Due to the lack of FIGO staging and histological grading data, whether there is heterogeneity between early SCC and preinvasive or late SCC was unknown in our discovery datasets. In the PCA of GSE63514, HSIL was partially overlapped with normal/LSIL and SCC. Moreover, HSIL showed higher heterogeneity than normal in GSE27678. Akin to genetic alteration, we believe that some dysregulated genes common to SCC but only changed in a part of HSIL contributed to the potential heterogeneity of HSIL [43, 44]. Compared to enrichment with total DEGs, separate enrichment with up- and downregulated genes could detect more pathways associated with the phenotypic difference [45]. We used both strategies in this study. Although separate analysis consistently detected more terms and pathways, some pathways (e.g., Wnt signaling pathway in LN_DN and CN_DEG) enriched in different disease stages by two strategies respectively should not be ignored. Cell cycle, DNA repair, and oncogenic p53 pathways were activated in HSIL and SCC. The close association between these pathways and HPV has been evidenced. HR-HPV E6 and E7 oncoproteins interfere with p53 and pRB, leading to cell cycle disturbances and promoting DNA damage response (DDR) that has a known central role in cervical carcinogenesis [3, 46, 47]. Furthermore, we found inhibition of TGF-β and Hippo signaling pathways in LSIL, consistent with their tumor-suppressive properties in the early stage of carcinogenesis [48, 49]. Interestingly, the HTLV-1 infection and IL-17 signaling pathways were enriched in all disease stages. The deregulation of cell cycle is a common feature in cancer cells and HTLV-1-infected cells, which is why we believe that the HTLV-1 infection pathway was enriched in HN and CN [50, 51]. The HTLV-1 Tax oncoprotein interacts with SRF to activate the transcription of immediate early genes (FOS, FOSL1, EGR1, and EGR2) [52, 53]. However, these genes were downregulated in LN. The association between HPV and SRF in early stage of cervical carcinogenesis might be worth investigating. Of IL-17 cytokine genes, IL17C showed significantly lower expression in HSIL and SCC when compared to normal control. IL17C is an epithelial cell-derived cytokine that regulates innate epithelial immune responses [54], and its response to HPV infection has not been explicitly investigated. A previous study had reported that increased Th17 cells were associated with progression of SCC [41], which was inconsistent with our results. However, there were studies reporting that lymph nodes of premalignant lesion-bearing mouse contained more Th17 cells than HNSCC-bearing mouse lymph nodes [55, 56]. Reduced IL23 production and increased TGF-β production by HNSCC may lead to the decrease in Th17 by redirecting the immune phenotype toward Treg [56]. We found four hub genes through network analysis. Using IHC, the gradually increasing expression of hub genes along with the severity of lesions was validated. Notably, CEP55 was initially reported to be associated with the course of cervical lesions. The staining pattern of TOP2A was similar to that of Ki-67, and concordance between them was substantial. While a study comparing ProExC and Ki-67 expression in 197 cervical biopsies reported that 35% of cases showed discordant staining [57]. We then compared the diagnostic performance of p16INK4a, Ki-67, TOP2A, and RFC4 alone or in combination to detect HSIL/HSIL+. Among the four markers, p16INK4a routinely used in clinical practice showed the highest sensitivity but moderate specificity. Similar to previous reports, the combination of p16INK4a and Ki-67 in serial interpretation could improve specificity and accuracy for detecting HSIL [58]. RFC4 and TOP2A alone provided similar diagnostic performance to the combination of p16INK4a and Ki-67. Parallel interpretation of TOP2A and RFC4 produced the highest AUC, and parallel interpretation of Ki-67 and RFC4 produced the highest sensitivity and NPV for detecting HSIL. Importantly, RFC4 and TOP2A have additional advantages. The expression of RFC4 from 3q26 exhibited a high correlation with copy number gain, and 3q gain as a potential marker in the diagnosis of HSIL is frequently found in cervical cancer and its precancerous lesions [59, 60]. For TOP2A, its exclusive and clear nuclear staining is an advantage over nuclear and cytoplasmic staining of RFC4 and p16INK4a. Moreover, Shi et al. reported that TOP2A is more sensitive and specific than ProEXC for detecting HSIL [61]. Considering cost-effectiveness, a single biomarker with balanced sensitivity, specificity, and high accuracy is recommended. When meeting patients with suspected HSIL, we can choose parallel interpretation of Ki67/TOP2A and RFC4 with high sensitivity and NPV to safely exclude lesions. Furthermore, we explored the clinical and prognostic significance of identified genes in SCC. Compared with a continuous increase of hub gene expression in normal to SILs to SCC transitions, only AURKA mRNA expression significantly increased with advancing FIGO stage, increasing tumor differentiation and aggressiveness in SCC, as indicated by the poor OS. This is consistent with the findings observed previously [62], though the prognostic interest of AURKA could not be validated in the TJH cohort. A previous study demonstrated that high CEP55 protein expression correlates with better OS and recurrence-free survival (RFS) in SCC [63]. We found this trend in our research but not statistically significant. Several studies based on TCGA-CESC data have reported the relationship between TOP2A and RFC4 mRNA expression and the prognosis of cervical cancer [64, 65]. However, there was no other evidence to support their relationship, let alone the confirmation at the protein level. Here, we demonstrate for the first time that increased RFC4 and TOP2A protein expression correlates with a favorable outcome in patients with SCC, and RFC4 is an independent prognostic marker for SCC. Furthermore, preliminary investigations have also demonstrated the role of RFC4 in predicting the outcome of other neoplasia, such as non-small cell lung carcinoma, colorectal cancer, and breast tumor [66–68]. Of course, our research has some limitations. Firstly, diagnosis error cannot be excluded entirely because the histopathologic diagnosis of CIN is subject to substantial rates of discordance among pathologists. Due to the majority diagnosis from three expert gynecologic pathologists and the large sample size in our study, we considered this diagnosis bias only to influence results to a minor degree. Secondly, we focus on RFC4 dynamic expression and clinical application here, which could not clarify the cause-and-effect relationship between RFC4 overexpression and disease progression. Our laboratory has ongoing experimental studies of RFC4 in papillomavirus oncogenic cell transformation.
Collectively, our study has characterized the changes in gene expression and biological functions during cervical carcinogenesis, which contribute toward a better understanding of molecular mechanisms associated with disease progression. Furthermore, we have found that RFC4 and TOP2A alone could serve as potential surrogate markers for determining HSIL and HSIL+. Their potential clinical application in cytological specimens was foreseen. Finally, RFC4 was also confirmed as an independent prognostic biomarker for SCC, implicating its therapeutic targeting for the treatment of SCC.
Additional file 1: Figure S1. The quality control of selected datasets. Figure S2. The flow of sample selection and immunohistochemistry (IHC), related to Table S2. Figure S3. Pathway enrichment and comparison. Figure S4. Human T-cell leukemia virus 1 (HTLV-1) infection signaling pathway map. Figure S5. Differential expression of total genes in each comparison group of the discovery datasets. Figure S6. Correlation between hub gene expression and severity of cervical lesion in the discovery datasets. Figure S7. Schematic diagram of immune scoring criteria, related to Table S5. Figure S8. Assessment of the relationship between hub gene expression and clinical parameters in SCC patients from the TCGA cohort. Figure S9. Univariate and multivariate survival analysis for PFI/DFS in SCC patients, related to Fig. 6. Figure S10. The illustration of the tissue microarray (HUteS154Su01; TMA) from Outdo Biotech. Co., Ltd. Table S1. Microarray datasets analyzed in this study. Table S2. Histology information of all tissue specimens, related to Figure S2. Table S3. Summary of DEGs and Gene Sets1&2. Table S4. Antibodies used for immunohistochemical staining. Table S5. Immunohistochemical scoring system for noninvasive squamous epithelia. Table S6. Clinical characteristics of the collected cohorts for survival analysis. Table S10. Summary of the hub gene expression related to cancer progression. Table S11. The summary of immunohistochemical scoring results. Table S12. Characteristics of studies assessing p16INK4a, Ki-67, TOP2A and ProExC immunohistochemically. Table S13. Concordance analysis between IHC biomarkers. Table S14. Sensitivity, specificity, PPV, NPV, and AUC of six IHC biomarkers for detecting HSIL and HSIL+. Table S15. Comparison of positive rates of single and combined IHC biomarkers in different cervical lesions (based on sections with p16INK4a, Ki-67, TOP2A and RFC4 evaluated simultaneously). Table S16. Diagnostic performance of serial and parallel interpretation of IHC biomarker combinations for detecting HSIL and HSIL+ compared to TOP2A and RFC4 alone.Additional file 2: Table S7. Enrichment of chromosomal region. Table S8. Simplified results of GO enrichment. Table S9. KEGG enrichment results. |
|
PMC9648032 | Ye Eun Kim,Kyung Hyun Cho,Ina Bang,Chang Hee Kim,Young Shin Ryu,Yuchan Kim,Eun Mi Choi,Linh Khanh Nong,Donghyuk Kim,Sung Kuk Lee | Characterization of an Entner–Doudoroff pathway-activated Escherichia coli | 09-11-2022 | Escherichia coli,Glucose metabolism,Entner–Doudoroff pathway,Adaptive laboratory evolution,Phosphofructokinase,3-Hydroxypropionic acid | Background Escherichia coli have both the Embden–Meyerhof–Parnas pathway (EMPP) and Entner–Doudoroff pathway (EDP) for glucose breakdown, while the EDP primarily remains inactive for glucose metabolism. However, EDP is a more favorable route than EMPP for the production of certain products. Results EDP was activated by deleting the pfkAB genes in conjunction with subsequent adaptive laboratory evolution (ALE). The evolved strains acquired mutations in transcriptional regulatory genes for glycolytic process (crp, galR, and gntR) and in glycolysis-related genes (gnd, ptsG, and talB). The genotypic, transcriptomic and phenotypic analyses of those mutations deepen our understanding of their beneficial effects on cellulosic biomass bio-conversion. On top of these scientific understandings, we further engineered the strain to produce higher level of lycopene and 3-hydroxypropionic acid. Conclusions These results indicate that the E. coli strain has innate capability to use EDP in lieu of EMPP for glucose metabolism, and this versatility can be harnessed to further engineer E. coli for specific biotechnological applications. Supplementary Information The online version contains supplementary material available at 10.1186/s13068-022-02219-6. | Characterization of an Entner–Doudoroff pathway-activated Escherichia coli
Escherichia coli have both the Embden–Meyerhof–Parnas pathway (EMPP) and Entner–Doudoroff pathway (EDP) for glucose breakdown, while the EDP primarily remains inactive for glucose metabolism. However, EDP is a more favorable route than EMPP for the production of certain products.
EDP was activated by deleting the pfkAB genes in conjunction with subsequent adaptive laboratory evolution (ALE). The evolved strains acquired mutations in transcriptional regulatory genes for glycolytic process (crp, galR, and gntR) and in glycolysis-related genes (gnd, ptsG, and talB). The genotypic, transcriptomic and phenotypic analyses of those mutations deepen our understanding of their beneficial effects on cellulosic biomass bio-conversion. On top of these scientific understandings, we further engineered the strain to produce higher level of lycopene and 3-hydroxypropionic acid.
These results indicate that the E. coli strain has innate capability to use EDP in lieu of EMPP for glucose metabolism, and this versatility can be harnessed to further engineer E. coli for specific biotechnological applications.
The online version contains supplementary material available at 10.1186/s13068-022-02219-6.
Bacterial glucose metabolism is particularly diverse and involves the orchestrated actions of multiple pathways, including the Embden–Meyerhof–Parnas pathway (EMPP), Entner–Doudoroff pathway (EDP), pentose phosphate pathway (PPP), and oxidative pathway (via gluconic acid) [1, 2]. These pathways produce pyruvate, reducing power (NADH and/or NADPH), and energy (ATP), and play a central role in growth and bio-production [3, 4]. In contrast to Escherichia coli, which primarily uses the EMPP for glucose metabolism, Pseudomonas, and Zymomonas spp. employ the EDP, as they lack phosphofructokinase (Pfk), a pivotal EMPP enzyme. The PPP is a metabolic pathway that functions in parallel with either the EMPP or EDP and supplies the energy requirements for cell growth, NADPH regeneration, and replenishment of essential precursors for nucleotide and amino acid synthesis. The EMPP comprises 10 enzymatic steps, yielding two pyruvate, two net ATP, and two NADH molecules per one glucose molecule, whereas the EDP utilizes only five enzymes to have two pyruvates (one of which is produced from glyceraldehyde-3-phosphate (G3P) via the lower glycolysis), one net ATP, one NADH, and one NADPH molecule per one glucose molecule. The EDP has been reported to have certain advantages compared with the EMPP [5]. For example, it is the thermodynamically more favorable pathway, owing to lower ATP production, in turn, boosting glycolysis [6]. Considering the number of enzymatic steps involved, the costs of protein synthesis in EDP is lower than that of EMPP [7]. It is also characterized by less allosteric regulation due to the lack of Pfk catalysis, one of the committed steps in the EMPP, which is found to be subjected to intensive regulation at the transcription, translation, and even post-translation and allosteric levels via distal metabolites, as well as energetic precursors [8]. The EDP is also associated with higher NADPH production to provide high reducing power for biosynthesis, resistance to oxidative stress, and the optimal synthesis of both the carbon and redox precursors for terpenoid biosynthesis. For example, Z. mobilis that relies extensively on the EDP shows notably rapid glucose uptake and ethanol production rates [9], and Pseudomonas species that utilizes an incomplete EMPP shows high production of highly toxic chemicals associated with their innate high solvent tolerance. Besides, the absence of a Pfk step facilitates the cyclization of PPP, which regenerates excess NADPH [10, 11]. Nevertheless, although there are several distinct benefits of employing EDP, the industrial strains of E. coli do not use their inherent EDP for glucose metabolism. Several studies have been conducted to engineer E. coli in an effort to activate the intrinsic EDP and to block the EMPP to promote more efficient biosynthesis. Previously, blockage of the EMPP via deletion of the phosphoglucose isomerase gene (pgi) resulted in re-routing the glycolytic flux through the PPP and EDP. The Δpgi recombinant E. coli strains were characterized with enhanced production of polyhydroxybutyrate [12], isoprenoid [13], and hydrogen [14]. However, although the Δpgi mutants remain viable, given that the glycolytic flux is re-routed through the PPP and EDP, excess NADPH production was found to perturb a significant portion of the metabolic network, resulting in an approximately 80% reduction in cell growth in glucose minimal media when compared with the wild-type levels [15]. Disrupting the EMPP by knocking out the phosphofructokinase I (PfkA) gene also resulted in re-routing the glycolytic flux through the PPP (~ 60% of the glycolytic flux) and the native EDP (~ 14% of glycolytic flux) [16]. The ΔpfkA recombinant E. coli strains showed enhanced production of 1,3-diaminopropane [17], hydrogen and ethanol [18], lycopene [19], and methyl 3-hydroxybutyrate [20]. However, these strains showed a more significant decrease in growth rate compared with the Δpgi mutants owing to the partial cyclization of PPP, which may produce excess NADPH [21]. E. coli in which the EMPP was completely blocked by deleting pfkAB showed retarded growth on glucose minimal medium due to redox imbalance and reduced glucose uptake, suggesting the PPP and EDP were not sufficiently activated to alleviate the metabolic defect. In this study, we isolated and characterized a pfkAB-deleted E. coli mutant strains that can thrive on glucose minimal medium with adaptive laboratory evolution (ALE). Using these strains, we investigated the genotype–phenotype relationship to determine and classify the requirements for the full activation of the EDP in E. coli. Finally, as a proof of concept and based on the preliminary information obtained, we further engineered the evolved mutant to confer it with the ability to produce a higher level of 3-hydroxypropionic acid (3-HP) that requires more NADPH regeneration.
Given that E. coli has two Pfk isozymes, Pfk-I and Pfk-II encoded by pfkA and pfkB, respectively [22], a recombinant E. coli strain with complete inactivation of the EMPP was constructed by deleting both of pfkA and pfkB genes. The ∆pfkAB strain showed slower specific growth rate on minimal medium containing glucose as a sole carbon source (Fig. 1a). ALE experiments were performed to recover the reduced cell growth by activating the EDP. After 50 subcultures of the ∆pfkAB strain in M9 minimal medium supplemented with 5 g/L glucose, the strain thereafter grew more rapidly on glucose. The five evolved ∆pfkAB strains showing recovered cell growth were isolated and denoted as pfk_ALE-1, -2, -3, -4, and -5 strains (Additional file 1: Figure S1a). The ALE-1 exhibited 92% of cell mass accumulation of the wild-type MG1655 strain after culturing for 20 h. This is the first E. coli strain with a completely inactivated EMPP that can grow more rapidly than mutants with partial blockage of the EMPP, such as the ∆pfkA and ∆pgi mutants (Fig. 1a). Besides, the ALE-1 strain showed improved terpenoid biosynthesis via the methylerythritol phosphate pathway, which has been observed in EDP-activated E. coli strains, because the synthetic pathway starts with 1-deoxy-d-xylulose 5-phosphate (DXP) synthase reaction with the two substrates G3P and pyruvate that are the final products of EDP. In addition, the synthetic pathway uses NADPH as sources of reducing regent produced by EDP. Interestingly, the production culture conditions used in the present study were able to rescue cell growth of the ALE-1 strain, with a 115% higher cell mass accumulation than the wild type at 36 h of culturing (Additional file 1: Figure S2). We sequenced the whole genome of ALE-1 and compared the sequence with the parental strain, E. coli MG1655. Resequencing identified six mutations in ALE-1 (Table 1). To confirm enriched and major mutations during ALE, the mutations detected in the ALE-1 were analyzed in the rest of four evolved mutants by PCR amplification and DNA sequencing. Mutations were found in the gntR, galR, ptsG, and ugd-gnd-wbbL-2 sequences of all five pfk_ALE strains, whereas mutations in the talB and crp genes were detected only in one or two strains (Fig. 1b). Notably, galR showed three unique mutations during adaptive evolution, but all resulted in truncated and non-functional protein (Additional file 1: Figure S1b, c). This suggests that the galR mutation could be critical for the growth of the evolved strain. In contrast, mutation in the talB gene was only observed in the ALE-1 strain and crp mutation was detected in the ALE-1 and ALE-4 strains, probably indicating that these mutations might not considerably affect the physiological background during ALE [23–25]. To compare the effects of the above mutations on the growth of ALE-1 strain, single mutations or combinations of the multiple mutations were introduced into the background ΔpfkAB strain. The growth rates of the resulting strains were compared with that of the background and evolved strain. The single mutation in the gntR, galR, crp caused an increase in the specific growth rate of 1.2-fold, 1.1-fold, and 1.2-fold, respectively, relative to the background ΔpfkAB strain. The single mutation in talB and the ptsG deletion mutant of the background strain showed a slight decrease in the specific growth rate. The background strain carrying a deletion of the ugd-gnd-wbbL-2 region showed an even slower cell growth. When each of the six mutations was introduced independently, the restored cell growth levels of the resulting mutants were considerably lower than that of the evolved strain. Eventually, when the mutant was harboring all six mutations, the growth rate was comparable to ALE-1 (Fig. 1c). Based on these observations, we deduced that the enhanced growth differentiating ALE-1 from the background mutant was attributed not only to the occurrence of a single mutation in gntR, galR, ugd-gnd-wbbL-2, ptsG, crp, and talB, but also to their synergistic interaction. GntR is a DNA-binding transcriptional regulator that represses the expression of genes involved in gluconate metabolism in the absence of the effector molecule D-gluconate [26]. Gluconate is phosphorylated by gluconate kinase to produce 6PG, which is also produced from glucose. It is finally metabolized to G3P and pyruvate via the 6-phosphogluconate dehydratase (Edd) and 2-keto-3-deoxy-6-phosphogluconate aldolase (Eda) enzymes of EDP. Knockout of gntR has previously revealed that the expression of edd and eda genes can occur even in the absence of gluconate [27]. Therefore, we hypothesized that the loss-of-function mutation in the gntR gene de-represses the originally latent genes of the EDP, thereby providing a major route for glucose metabolism in ALE-1 with an inactive EMPP (ΔpfkAB) and an inactive oxidative PPP (Δgnd) (Table 1) [28]. In fact, ALE-1 Δedd strain could not grow on glucose, indicating that ALE-1 relies on the EDP on glucose as the sole carbon source (Additional file 1: Figure S3). GalR is a transcriptional regulator that down-regulates the genes associated with galactose metabolism when this sugar is unavailable [29]. The regulon of GalR harbors the galP gene encoding galactose permease, which also has glucose uptake activity [30]. Wild type E. coli mainly takes up external glucose by a phosphoenolpyruvate-dependent phosphotransferase system (PTS), which consists of enzyme I (PtsI), HPr (PtsH), and glucose-specific PTS EIIA (Crr) and EIICB (PtsG) components [31]. Thus, we anticipated that the evolved mutant that lost the PTS activity due to PtsG inactivation by insertion of the IS5 and SgrS regulation of sugar transporter mRNAs (ptsG and manXYZ) due to the accumulation of glucose-6-phosphate, and consequently, the strain can co-metabolize glucose and xylose, which was the case in the ptsG-deleted E. coli strains (Additional file 1: Figure S7d–f) [32–34]. It has also been reported that the loss of PtsG function results in retarded cell growth due to the limited capacity of glucose import, and non-PTS glucose transporters such as GalP can restore the glucose uptake rate in ptsG null mutants [35, 36]. Based on this assumption, we hypothesized that the expressed GalP, which is depressed by the inactivation of galR, rescues the glucose uptake capacity of ALE-1. Moreover, another possible reason for the expression of a non-PTS glucose permease GalP in EMPP-disrupted strains could be the low availability of phosphoenolpyruvate (PEP), which is essential for the PTS-dependent glucose uptake. Previous studies showed that the ∆pfkA mutant had reduced PEP concentration [15]. Besides, an increase in flux through the EDP can further lower the intracellular PEP pool [16]. It is supported by the stoichiometry of the EDP, where PEP generation per glucose is half of that of the PPP or EMPP [13]. Limited glucose importing capacity due to low PEP availability can be supplemented by the expression of PEP-independent glucose permease and glucokinase [35]. cAMP receptor protein (CRP) is a global transcription regulator that directly or indirectly regulates the expression of over 100 genes [37–39]. The mutation site of CRPA85S was very close to the cAMP binding site, suggesting the possible effect on the binding affinity. However, most of cAMP-independent CRP mutants contained mutations away from the binding site such as I112, T127, and A144 residues [40]. Therefore, it is hard to establish that the CRPA85S is a cAMP-independent mutant. Since no significant difference in the cell growth was observed by this mutation, the activity of this enzyme was not affected (Fig. 1c). However, the exact mechanism underlying the crp mutation-mediated growth rate enhancement in the evolved strain remains unclear. We discovered that crp mutation occurred in the later period of adaptation (during 30–50 repeats of culture) while ptsG mutation was detected in the earlier period of adaptation (during 0–30 repeats of culture). Although PTS inactivation by ptsG mutation is an efficient way to trigger carbon catabolite repression (CCR) release like galP expression, the modification of binding between CRP and cAMP might further contribute the release of CCR and provide a strong selective pressure during adaptation [41]. TalB is an enzyme involved in the non-oxidative PPP and provides a reversible reaction between the PPP and the upper EMPP [42]. The A247 residue is located in an α-helix and away from the active center of E. coli TalB. The B-factor value of the mutated residue was kept considerably lower, indicating that the mutation stably maintains its α-helix structure. In addition, the substrate binding affinity of the TalBA247T in E. coli W3110 strain was not modified in a previous study [43]. Moreover, the expression level of the talB gene in ALE-1 was not significantly changed (1.3-fold) by this mutation compared to that in MG1655 (Additional file 1: Figure S5). Since the ALE-1 strain lacks the gnd gene, a change in the non-oxidative PPP flux might be directed by a genetic mutation in talB during adaptation to supply precursors essential for growth.
Functional analysis of the mutations detected in the ALE-1 strain was conducted to elucidate the molecular mechanisms underlying the effects of these mutations. We performed transcriptome analysis with RNA-seq to investigate the transcriptomic effects of mutations involved in the glycolytic pathway and global transcriptional regulation. Transcriptomic analysis revealed the expression levels of 4497 genes on the chromosome of MG1655 and ALE-1 strains. Among those genes, we first focused on the glycolytic pathways including the EMPP, EDP and PPP. Complete deletion of the pfkA gene was confirmed with no pfkA expression. In ALE-1, the expression of the edd and eda genes, which are involved in the EDP, were 11.2- and 5.1-fold higher than in MG1655, respectively. This indicates that the latent EDP was activated during adaptation due to loss of function for GntR. Additionally, the expression level of the GntR regulon in ALE-1 was higher compared to MG1655. There was no expression of gnd, involved in the oxidative PPP, due to its complete deletion in ALE-1 (Fig. 2a). Taking these observations into consideration, the ALE-1 strain utilizes glucose via the EDP rather than the EMPP and oxidative PPP since ALE-1 lacks functional PfkA, PfkB, GntR, and Gnd. The RNA-seq analysis also showed lower expression level of ptsG in ALE-1 compared to MG1655, caused by degradation of ptsG mRNA. Instead, the expression level of the non-PTS glucose transporter, galP, in ALE-1 was 13.7-fold higher than MG1655. This phenomenon was also confirmed by quantitative RT-PCR, demonstrating 7.6-fold increased expression level (Additional file 1: Figure S5). Notably, we found that the expression levels of the galR regulon in ALE-1 was higher than that in MG1655 due to the inactivation of GalR (Fig. 2a). These indicate that the limited glucose uptake capacity was rescued by the GalP via the inactivation of GalR. However, only a small change was observed in the gntR and galR expression level between the MG1655 and ALE-1 strains, which suggests that the transcriptional levels of gntR and galR were not affected by the genetic mutation. Instead, mutations in gntR and galR might result in loss of protein function since the expression levels of their regulon were down-regulated. During evolution, ALE-1 activated the latent pathways, including the glyoxylate shunt and phosphoenolpyruvate carboxykinase (Pck) reaction. Since the glyoxylate shunt does not regenerate NADPH unlike the isocitrate dehydrogenase (Icd) reaction, the evolved strain might enhance the glyoxylate shunt flux rather than the Icd reaction to reduce the NADPH regeneration capacity. Overall, transcriptome analysis suggested that the glycolytic flux was re-routed through the activation of latent pathways and rescued the glucose uptake capacity in ALE-1. On the other hand, even though glucose can be utilized via mannose PTS [41], the expression levels of manX, manY, and manZ in ALE-1 were 2.9-, 2.0-, and 1.5-fold reduced compared to those in MG1655, respectively (Fig. 2a). To further confirm our hypotheses, we initially deleted the gntR, galR, and ptsG genes in the ALE-1 strain. We found that the deletion of either the gntR, galR or ptsG gene did not affect the growth of ALE-1 cells, whereas restoration of these mutations resulted in notable adverse effects on the growth of ALE-1 cells (Fig. 2b). These results imply that GntR and GalR mutants may be unable to repress the expression of their target genes, including edd, eda, and galP. As expected, the same phenotype was observed in the ALE-1 strain that had a transposon inserted in ptsG or had a deleted ptsG mutation, confirming that the ptsG gene with the IS insertion lost its activity. To verify the target genes of the two TFs, we expressed the plasmid-encoded endogenous edd-eda or galP gene in the gntR- and galR-restored ALE-1, respectively, resulting in increased growth from 36 and 55% up to 82% and 86% of ALE-1, respectively. These observations thus indicate that the plasmid-based expression of edd-eda and galP genes can functionally recover cell growth defects caused by the restored active GntR and GalR in ALE-1 (Fig. 2b and Additional file 1: Figure S4a). To verify the effect of the large deletion of the ugd-gnd-wbbL-2 region carrying the two functional genes ugd and gnd on ALE-1 cell growth, we re-introduced the deleted region into the chromosome of ALE-1. The restored strain showed reduced cell growth, and then the deletion of the gnd gene, but not the ugd gene, resulted in a substantial recovery of the growth defect caused by the restoration of the ugd-gnd-wbbL-2 region (Fig. 2b and Additional file 1: Figure 4a). According to the transcriptome analysis, we hypothesized that the ALE-1 uses EDP instead of the EMPP or oxidative PPP as the major glycolytic pathway. Since 2-keto-3-deoxy-6-phosphogluconate (KDPG) is an intermediate of EDP, intracellular KDPG level should be higher in ALE-1 compared to its wild-type strain. It was previously reported that the transcription of hexR, that originated from P. putida KT2440, was repressed in the presence of KDPG produced from glucose or gluconate [44]. To analyze the EDP flux, a KDPG-responsible biosensor using a transcriptional regulator (HexR) and GFP was constructed to sense the intracellular level of KDPG. Using an appropriate gate, M1, 85.8% of the population of ALE-1 and 9.5% of the wild-type MG1655 were detected (Fig. 2c). Previously, the glycolytic pathway of Δpgi ΔgntR double-knockout E. coli strain was redirected from EMPP to EDP [28]. In fact, 64.2% of the population of the Δpgi ΔgntR double-knockout cells were detected with M1 gate, which was higher than wild-type strain. This KDPG-responsible biosensor successfully supported our hypothesis that the EDP of ALE-1 was more active than that of the wild-type. In conclusion, the growth of the EMPP-deficient mutant was facilitated by a number of factors that includes, 1) the re-direction of the glucose flux from the oxidative PPP to EDP by the deletion of gnd, 2) the gain of the EDP activity by the lack of gntR, and 3) the recovery of glucose uptake by the inactivation of GalR (Fig. 2d and Additional file 1: Figure S4b).
Notably, the evolved strain had the gnd mutation resulting in the inactivation of oxidative PPP, an alternative glycolytic pathway. The ALE-1 strain thus employs only the EDP for glycolysis, indicating that an active PPP might harm the growth of bacterial cells using the EDP as the main glycolytic pathway. The availability of the ALE-1 strain with plasmid-based Gnd expression enabled us to assess the effect of glucose catabolism via both oxidative PPP and EDP on its cell growth (Fig. 3a, b). As expected, expression of Gnd conferring oxidative PPP activity together with the EDP had an adverse effect on the growth of ALE-1 cells, leading to a 23% decrease in the growth rate of the evolved strain. We further investigated the effect of the further increased glycolytic flux from the EDP to oxidative PPP on the growth of ALE-1 cells by expressing both gnd and gntR on a plasmid, where Gnd reactivates oxidative PPP and GntR represses the EDP (Fig. 3a, c). As a result, high glycolytic flux to oxidative PPP resulted in more severe growth defects, resulting in a 39% reduction in cell growth, but exhibiting a 120% higher NADPH/NADP+ ratio (Fig. 3d, e). Using HexR-based KDPG-responsible biosensor, we further examined the activity of EDP in ALE-1 with plasmid-based expression of the gnd and/or gntR genes to confirm that expression of those genes led to redirect glycolytic flux. As a result, the concentration of intracellular KDPG was decreased when only gnd was expressed and further decreased when both the gnd and gntR were expressed (Additional file 1: Figure S6). This observation led us to speculate about why the oxidative PPP function was lost instead of being conserved to handle the glycolytic flux together with the EDP. It has been suggested that one of the reasons why the E. coli Δpgi mutant, which is primarily dependent on the PPP for glucose catabolism, shows reduced glucose uptake rate is due to a redox imbalance and/or an excess of NADPH regenerated from the PPP [45]. When glucose is metabolized via the EDP, one less NADPH is regenerated than via the PPP in this mutant [46]. However, the PPP of the pfkAB knockout mutant can regenerate excess NADPH even more intensively due to its cyclized architecture compared with the pgi-null mutant [10]. Taking these observations into consideration, the effect of re-routing the glycolytic flux to the EDP on the reduction of the regeneration of NADPH would be more significant in the latter mutant. Despite partial relief afforded by the EDP, the NADPH formed via the cyclized PPP might exceed the demand for cell growth. Moreover, the EDP has been known to be thermodynamically favorable, even when compared with the native glycolytic route EMPP. This thermodynamically favorable property enables the EDP to sustain a glycolytic flux comparable to that of the EMPP [6, 7]. Indeed, Z. mobilis, which extensively employs the EDP for glycolysis, takes up glucose at a notably rapid rate [9]. This observation indicates that the capacity and turnover rate of the EDP would be substantial [47, 48]. Considering the foregoing observations, establishing the EDP as the sole glycolytic route, rather than sharing the flux with the PPP, would benefit the growth of the Pfk-deficient mutant in terms of both reducing NADPH production and expanding glycolytic capacity during selection. Consistent with this hypothesis, the soil bacterium P. putida KT2440, which lacks a functional EMPP, is known to be almost extensively dependent on the EDP for glycolysis, with less than 10% of the glucose flux entering the PPP [2].
To facilitate optimal growth, cells must stringently balance their intracellular redox. However, the microbial production of value-added chemicals often requires an abundant supply of reducing powers in the form of NADPH. Accordingly, there exists a fundamental difference in the optimal architecture of the core metabolic pathways to sustain normal growth and those required for metabolite overproduction. For the successful application of the evolved mutant described herein for bio-production, further manipulation of the core metabolic pathways would be necessary. For proof of concept, we introduced changes into the evolved strain ALE-1 machinery for it to attain an abundant supply of NADPH necessary for the production of 3-HP, an intermediate for many compounds. The bifunctionality of 3-HP enables it to be further transformed into a variety of high-value compounds [49–51]. From 1 mol of glucose, it is theoretically possible to generate 2 mol of 3-HP with a consumption of 4 mol of NADPH (Fig. 4a). However, its redox equivalent demand is far beyond the endogenous catabolic supply based on glucose catabolism via the EDP or the linear PPP, which regenerates 1 or 2 mol of NADPH per mol of glucose [13]. According to the stoichiometry of the EDP and PPP, restoration of cyclized PPP activity and the reduction of EDP activity might yield larger amounts of 3-HP [10, 13]. To address this issue, we demonstrated the conditional flux control towards the EDP and PPP at the 6PG node for the enhanced production of 3-HP. Before dealing with the redox issue, we attempted to eliminate a potential limiting factor, namely, the low availability of malonyl-CoA for 3-HP production. Efficient microbial production of chemicals such as flavonoids and fatty acids using malonyl-CoA as a precursor are often constrained by the low intracellular concentrations of this compound [52]. In the present study, we overexpressed acetyl-CoA carboxylase (ACC) to enhance the supply of malonyl-CoA, and in conjunction performed site-directed mutagenesis designed to increase the temperature sensitivity of endogenous enoyl-[acyl-carrier-protein] reductase (FabI), thereby reducing the malonyl-CoA consumption via fatty acid synthesis [53, 54] (Fig. 4a). At 48 h post-induction, the evolved strain ALE-1 produced 1.34 g/L of 3-HP, whereas the background strain produced 0.83 g/L. Although the evolved mutant produced a 1.6-fold higher 3-HP titer than the background strain under the same conditions, the fold-increase was subtle, considering there was a substantial difference in biomass accumulation between the two strains (Fig. 1a and Fig. 4b). To achieve conditional flux control towards the EDP and PPP, the E. coli gnd and gntR genes were expressed under the control of an IPTG-inducible promoter, and the resulting pGnd and pGnd-GntR plasmids were transformed into ALE-1 harboring the 3-HP-producing machinery. Induction of flux redistribution to both the EDP and PPP in the 3-HP-producing ALE-1 mutant harboring pGnd led to a considerable enhancement in the 3-HP titer (1.99 g/L) at 48 h post-induction. Furthermore, when the EDP was repressed by additional expression of its repressor GntR, 3-HP production in the evolved mutant increased further to 2.65 g/L, which is 3.2- and 3.3-fold higher than the titer obtained from the background mutant and wild-type E. coli strains, respectively, harboring the same 3-HP synthesis machinery (Fig. 4b and Additional file 1: Figure S2). These findings thus indicate that facilitating glucose flux via the PPP and reducing the flux through the EDP are critical for generating an abundant NADPH supply, resulting in higher 3-HP production.
In this study, we generated and characterized an engineered E. coli using EDP rather than EMPP as the major glycolytic pathway, whose pathway is not functional in the wild-type strain. Based on genetic and biochemical studies, we demonstrated that optimal cell growth of E. coli strain through the EDP requires three main metabolic changes: (1) de-repression of the EDP by inactivation of GntR, (2) inactivation of the oxidative PPP by deletion of Gnd, and (3) an increase in glucose uptake by the inactivation of GalR. Reverse engineering of the evolved strain was applied to enhance the production of 3-HP by increasing glycolytic flux from the EDP to PPP for the supply of excess NADPH during the production stage rather than the growth stage. The engineered strain and its derivatives have certain advantages with respect to specific bio-production purposes, such as 3-HP and lycopene production, which demand NADPH (Fig. 4b and Additional file 1: Figure S7a, b). Two EMPP-disrupted E. coli mutants, in which the pathway was blocked at the Pfk or Pgi step, resulted in distinct genotypic changes after ALE with glucose as the sole carbon source [18, 23, 55]. The evolved E. coli strain lacking pfkA showed a mutation in the promoter region of pfkB coding for a minor isozyme of Pfk. The expression of Pfk increased with time and this enabled the strain to restore glucose metabolism through the EMPP and could support cell growth. This observation indicates that, among three glycolytic pathways (EMPP, EDP, and PPP), the EMPP may be the most conducive for maximal cell growth in E. coli strains. The loss of Pgi led to a significant increase in relative flux through the PPP. None of the evolved Δpgi mutants could influence the absolute EDP flux. Instead, independent studies have revealed that mutations leading to enhanced activity of transhydrogenase have repeatedly been detected in adapted Δpgi mutants, suggesting that these mutations may reduce NADPH imbalance [23, 55–57]. In the present study, we found that the activation of the latent EDP by de-repressing downregulation was one of the key fitness-enhancing factors during the adaptation of the Pfk-null mutant. These findings imply that there are fundamental differences in the metabolic apparatus of the Pfk- and Pgi-null mutants, although both are widely used to redirect the glycolytic flux to the EMPP. One of them is that PPP can be cyclized and produce more NADPH in the Pfk-null mutant. Although slightly slower growth was found in ALE-1 compared with the wild-type strain that employs the EMPP as the main glycolytic pathway, the evolved strain generated in this study grew well on glucose minimal medium compared with pgi- or pfkA-deleted strains that are characterized by a partially inactive EMPP. An additional factor that may contribute to the suboptimal growth of the evolved strain is that an imbalance or excess of NADPH produced from EDP can reduce cell growth when glucose is used as the sole carbon source. Notably, E. coli expressing G6PDH is strictly NADP+-specific compared with that in other bacteria that primarily employ the EDP as the glycolytic pathway [58–60]. Previous studies have reported that the growth of the pgi mutant could not be restored fully to that of the wild type rate by overexpressing UdhA for the re-oxidation of NADPH, indicating that the transhydrogenase system is not sufficient to overcome an imbalance in NADPH [45]. When NADPH-consuming pathways (lycopene and 3-HP production) were engineered in ALE-1, we observed the recovery of cell growth; the lycopene- and 3-HP-producing ALE-1 cell showed higher cell growth than the wild type. These findings suggest that the evolved strain could be a suitable host cell for NADPH-dependent bio-production, as slightly reduced cell growth minimally affects production. The selection of a suitable host cell for the production of a biotechnology-based product is crucial to ensure economic viability. The bacterium, E. coli, is one of the best-characterized microorganisms from the genetic and metabolic perspectives and has a variety of tractable traits for biotechnological production. However, tolerance to adverse environmental stress, or unfavorable redox potentials may impede the use of E. coli as a platform host organism. The selection of host cells for production should be based on several factors, including the availability of genetic tools for expressing target products, efficient utilization of alternative carbon sources, high redox potential, and a tolerance to the metabolic products. Furthermore, unlike Pseudomonas and Zymomonas, ALE-1 can co-metabolize glucose and xylose, which was the case in the ptsG-deleted E. coli strains (Additional file 1: Figure S7d–f). ALE-1 strain even grows under anoxic conditions using nitrate as a final electron acceptor (Additional file 1: Figure S7c). In conclusion, we have demonstrated that the normally latent EDP in E. coli can be almost fully activated by enhancing glucose uptake through an inactivation of GalR, inactivating the EDP regulatory system, and blocking the glycolytic flux via the PPP. The resulting strain and its derivatives have advantageous traits that are found in microorganisms using the EDP and PPP as the major glycolytic route, and it is amenable to host strong redox reactions. The engineered strains could expand the utility of E. coli as a platform for biotechnological production.
Unless otherwise indicated, the reagents used in this study were purchased from Sigma–Aldrich (St. Louis, MO, USA). The bacterial strains and plasmids used in this study are listed in Additional file 1: Table S1. The wild-type E. coli MG1655 strain was used as a parent strain for the generation of mutant strains. The E. coli DH10B strain was used for bacterial transformation and plasmid amplification during the construction of the expression plasmids. Unless otherwise stated, as growth media, we used M9 minimal medium (BioShop Canada Inc., Burlington, Canada) supplemented with 2 mM MgSO4, 0.1 mM CaCl2, and glucose with appropriate antibiotics, if required, and Luria–Bertani (LB) medium. The LB medium was prepared using 5 g/L yeast extract, 10 g/L peptone and 10 g/L NaCl. Antibiotics were used at the following concentrations: ampicillin (100 mg/L), kanamycin (50 mg/L), and chloramphenicol (30 mg/L).
For growth measurements, a single colony from an agar plate was inoculated in 3 mL of LB medium at 37 °C and shaken at 200 rpm. Overnight-grown cells were 1:10 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose. After growing for 12 h, cells were inoculated into 25 mL of M9 medium in 250-mL shaking flask at 37 °C and shaken at 200 rpm to achieve an initial optical density at 600 nm (OD600) of approximately 0.1. For functional analysis, a single colony was cultivated in 3 mL of LB medium with appropriate antibiotics at 37 °C and shaken at 200 rpm overnight. A seeded culture was 1:10 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose and appropriate antibiotics. After 12 h in culture, cells were diluted into 25 mL of M9 medium in a 250 mL shaking flask to achieve an initial OD600 of 0.1. The expression of galP and edd-eda was induced by the addition of 6.25 µM of isopropyl-β-D-thiogalactopyranoside (IPTG) at the beginning of seeding, subculturing, and culturing. The expression of pRFP, pGnd, and pGnd-GntR plasmids was induced by the addition of 50 µM of IPTG after 3 h in culture. OD600 was measured using a Libra S22 spectrophotometer. For 3-HP production, a single colony from an agar plate was used to inoculate 3 mL of LB medium at 30 °C and shaken at 200 rpm. Overnight-grown cells were 1:20 inoculated into M9 medium supplemented with 20 g/L glucose, 1.0 g/L yeast extract, 2 mM MgSO4, 0.1 mM CaCl2, and trace elements. The composition of the trace element solution used was as follows: 2.4 g of FeCl3·6H2O, 0.3 g of CoCl2·6H2O, 0.15 g of CuCl2·2H2O, 0.3 g of ZnCl2, 0.3 g of Na2MO4·2H2O, 0.075 g of H3BO3, and 0.495 g of MnCl2·4H2O per liter [54]. After growing at 30 °C and shaken at 200 rpm for 9 h, 50 µM of IPTG and 100 nM of tetracycline were added to induce the expression of mcrC*-mcrN, and accDABC, after which the temperature of the culture was increased to 37 °C.
To isolate rapidly growing mutants, even those lacking a functional EMPP, a mutant in which both the pfkA and pfkB genes had been deleted was adapted by sequential transfer. Initially, an overnight culture of ∆pfkAB mutant cells in LB was transferred to M9 minimal medium supplemented with 2 mM MgSO4, 0.1 mM CaCl2 and 5 g/L glucose and grown at 37 °C and shaken at 200 rpm. At an OD600 of approximately 1.0, the cells were passaged to a 250-mL shaking flask containing 25 mL of the same fresh M9 medium. This procedure was repeated 50 times, after which, the adapted cells were plated onto an LB agar plate. Five evolved mutants showing the most rapid biomass accumulation were selected based on OD600 values monitored using a 96-well microplate reader and named pfk_ALE-1, -2, -3, -4, and -5 (henceforth referred to as ALE-1, -2, -3, -4, and -5).
Genomic DNA was isolated from the evolved ALE-1 strain using a DNA isolation kit (GeneAll Biotechnology Co., Korea), and subsequently sequenced via next-generation sequencing using an Illumina HiSeq 2000 sequencer (Macrogen Inc., Korea). The genomic sequence of E. coli MG1655 (NC_000913.3) was used as a reference. Read sequence data of wild-type and ALE-1 strains were deposited in the Sequence Read Archive (SRA) of the NCBI under BioProject accession number PRJNA818725, and BioSample accession number SAMN26878847 for wild-type strain, SAMN26878848 for ALE-1 strain. The occurrence of mutations was confirmed by DNA sequencing of the PCR amplicon of target regions in the background strain and evolved mutant. To analyze the mutation frequency, the presence of mutations that were either the same or equivalent to the mutations discovered in ALE-1 was analyzed in the other four ALE mutants by PCR amplification and DNA sequencing.
Gene knockout was conducted by homologous recombination via a λ RED recombination system, which was mediated by a kanamycin resistance (KmR) cassette flanked by Flp recognition target (FRT) knock-in and flippase-based curing [61, 62]. Knockout constructs were designed with 50-bp overhangs homologous to upstream and downstream sequences of the deletion site. Cells were screened for antibiotic sensitivity, and for those lacking antibiotic resistance, removal of the KmR gene was confirmed by colony PCR and sequencing. Point mutations were generated via oligonucleotide-mediated genome editing combined with a tetA-based dual selection system. The tetA cassette was inserted into the target site and replaced with oligos containing target mutations and short 50-bp overhangs homologous to sequences upstream and downstream of the insertion site. The recombinants were isolated by negative selection using 50 μM NiCl2. Cells were screened for antibiotic sensitivity, and for those lacking antibiotic resistance, the correct allelic replacement was confirmed by colony PCR and sequencing.
All plasmids constructed in this study were derived from BioBrick plasmids [63]. The primers used for plasmid cloning are listed in Additional file 1: Table S2. The primers were synthesized by Macrogen Inc. (Seoul, Korea). Plasmid construction was conducted using Pfu-X DNA Polymerase for the PCR reactions, and all restriction enzymes used were purchased from Fermentas (Vilnius, Lithuania). For plasmid isolation, we used a LaboPass plasmid miniprep kit (Cosmo Genentech, Korea). DNA purification (Cat. no.103-102) and gel extraction (Cat. no.102-102) kits were purchased from GeneAll Biotechnology (Seoul, Korea). For the construction of pEdd-Eda, the gene encoding the red fluorescent protein (RFP) in the pBbA6c-rfp vector was excised using restriction enzymes, EcoRI and BamHI. The edd-eda gene cluster was amplified from E. coli MG1655 genomic DNA using a forward primer containing an EcoRI site and a reverse primer containing a BamHI site. Following excision with EcoRI and BamHI, the purified edd-eda fragment was ligated with the gel-eluted pBbA6c backbone. For cloning the pGalP, the RFP-encoding gene in the pBbS6k-rfp vector was excised using restriction enzymes, BglII and XhoI. The E. coli galP gene was amplified from E. coli MG1655 genomic DNA using a forward primer containing a BglII site and a reverse primer containing an XhoI site. Following excision with restriction enzymes, BglII and XhoI, the purified galP fragment was ligated with the gel-eluted pBbS6k backbone. To construct pGnd, the RFP-encoding gene in the pRFP vector was excised using EcoRI and XhoI. The E. coli gnd gene was amplified from the E. coli MG1655 genomic DNA with a forward primer containing an EcoRI site and a reverse primer containing an XhoI site. Following excision with EcoRI and XhoI, the purified gnd fragment was ligated with the gel-eluted pRFP backbone. For the construction of pGnd-GntR, the E. coli gntR gene was amplified from E. coli MG1655 genomic DNA with a forward primer containing a BglII site and a reverse primer containing the XhoI site. Following excision with restriction enzymes, BglII and XhoI, the purified gntR fragment was ligated with the purified pGnd. For the construction of the pMCR plasmid, a mcrC*-mcrN fragment including genes encoding the dissected C-terminal domain carrying three mutations (N391V/K557W/S565N) and the N-terminal domain of malonyl-CoA reductase from Chloroflexus aurantiacus was extracted after cutting the pCDF-mcrC*-mcrN plasmid at EcoRI and XhoI sites. The pBbE2k-rfp plasmid was digested with restriction enzymes, EcoRI and XhoI to excise the gene encoding RFP, after which the DNA fragment and the gel-eluted pBbE2k backbone were ligated. For the construction of a pACC plasmid, a fragment harboring the accDABC gene cluster encoding the ACC complex was obtained by elution after cutting pBbA2k-accDABC at BglII and XhoI sites. The pBbA6c-rfp plasmid was digested with restriction enzymes, BglII and XhoI to excise the gene encoding RFP. After gel-elution, the pBbA6c backbone was ligated with the eluted accDABC fragment.
Cell culture samples were harvested at the mid-log growth phase (OD600 ~ 0.5 for MG1655 and ALE-1 strain). The RNA was stabilized using by RNAprotect® Bacteria Reagent (Qiagen #76506) and total RNA was isolated using RNeasy® Plus Mini Kit (Qiagen #74134). Ribosomal RNA (rRNA) was removed from isolated RNA samples through riboPooL Probes (Pan-prokaryote) siTOOLS BIOTECH and Dynabeads™ MyOne™ Streptavidin C1(Invitrogen #65001). For purification of RNA samples, RNA clean & concentrator (zymoresearch R1013) was used. RNA library was constructed from rRNA-depleted RNA using the KAPA stranded RNA-seq kit (Roche KK8400). The sequencing libraries were sequenced by Nextseq 550, NextSeq 500/550 High Output Kit v2.5 (024906). Using bowtie software, the sequence reads were mapped onto the wild-type reference genome (NC_000913.2) with the maximum insert size option of 1500 bp (for Additional file 1: Figure S8, 500-bp option was used) [64]. Fragments per kilobase of exon per million fragments (FPKM) and log2 fold change value were calculated by cufflinks [65]. All the datasets were visualized using Metascope and NimbleGen’s Signalmap software. The transcriptomic data of wild-type and ALE-1 strains were deposited in NCBI’s Gene Expression Omnibus (GEO) under GEO Series accession number GSE199153. On the other hand, the unexpected expression level of the pfkB gene in ALE-1 was affected by a remaining coding sequence of pfkB (Additional file 1: Figure S8). Considering that the active site of PfkB was deleted, we anticipated that the function for PfkB in the ALE-1 strain was lost.
For cultivation, a single colony was inoculated in 3 mL of LB medium at 37 °C and shaken at 200 rpm. Overnight-grown cells were 1:50 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose and 5 g/L yeast extract. After growing for 9 h, cells were collected and transferred to PBS. The KDPG-responsible biosensor was induced by the addition of 50 µM of IPTG at the beginning of seeding and culturing. The green fluorescent protein (GFP) fluorescence intensity of individual cells was measured by BD FACSCalibur Flow Cytometer (BD Bioscience, USA) with wavelength of 488 nm excitation. Nonbacterial particles were excluded by applying forward-scattered characteristics and side-scattered characteristics. Gate M1 was set to differentiate the population of cells based on fluorescence intensity. BD CellQuest Pro software was used to analyze the data and to create images.
Culture media were treated using previously reported methods with minor modifications [66]. After centrifuging for 30 min at 13,200 × g, the supernatant was heat-treated at 80 °C for 1 h. Thereafter, the supernatant obtained by centrifuging again for 30 min at 13,200 × g was used for analysis. The concentrations of glucose and 3-HP were measured by HPLC using a previously described method with slight modification [67]. After a 20-fold dilution, 20 µL of samples were injected into a 300 mm × 7.8 mm Aminex HPX-87H (Bio-Rad, USA) column at 35 °C using 5.41 mM H2SO4 as the mobile phase.
Additional file 1: Figure. S1 Cell growth and description of galR mutations detected in pfk_ALE-2, -3, and -5 strains. (a) Growth rate profiles (OD600) were compared for the four evolved strains isolated after the 50th subculture during ALE. Growth rate profiles of ΔpfkAB and ALE-1 strains are presented as controls. (b, c) DNA sequence of wild-type and mutant galR coding region in ALE-2, -3 and -5 strains were compared. The sequences include the start codon (lower case), emerging site of early stop codon (grey box), duplicated target sequence (underlined), duplication (red), and insertion (blue). Adenine in the start codon of the galR gene is designated as +1. (b) A nonsense mutation was caused by a frame shift due to a 7-bp duplication in the galR gene of the ALE-2, -3 strains, which resulted in truncation of the C-terminal 33 amino acids of the total 344 amino acids. (c) Insertion of 768 bp and duplication of 8 bp in the galR gene of ALE-5 caused insertion of 259 amino acids between the 3rd and 4th amino acids of the wild-type GalR protein. These GalR mutants are expected to be equivalent to the GalR mutant of ALE-1, which has low binding affinity to target sequence in the absence of an appropriate inducer, galactose. Figure. S2 Cell growth and 3-HP, glucose, and acetate concentration during 3-HP production. All strains have a fabI temperature-sensitive mutant (fabIts) and two plasmids pMCR and pACC. Profiles of 3-HP concentration (red circles), cell mass (yellow triangles), consumed glucose concentration (green squares) and acetate (blue diamonds) were compared for E. coli strains: (a) MG1655 (b) MG1655 ∆pfkA (c) MG1655 ∆pgi (d) ΔpfkAB (e) ALE-1 (f) ALE-1 with pGnd and (g) ALE-1 with pGnd-GntR. Cell mass was calculated from the OD600 values. 1 OD600 unit corresponds to 0.33 g CDW/L [1]. Error bars indicate standard deviations of three independent biological replicates. Figure. S3 The comparison of cell growth between pfk_ALE-1 and pfk_ALE-1 Δedd strains. Growth rate profiles (OD600) of ALE-1 Δedd strain was compared to that of ALE-1 strain on glucose as a sole carbon source. Cells were grown in 250 mL with 25 mL of M9 minimal medium containing 0.4% glucose. Error bars indicate standard deviations of three independent biological replicates. Figure .S4 Functional analysis of mutations involved in glucose uptake system in the evolved strains. (a) Target genes affected by the mutated TF (GalR) or ptsG deletion were identified and their role for improved cell growth in ALE-1 were determined. Cells were grown in M9 minimal medium with 0.4% glucose in shaking flasks. IPTG (6.25 µM) was added at the beginning of seeding, subculturing, and culturing to induce each plasmid. Error bars indicate standard deviations of three independent biological replicates. (b) The schematic diagram illustrates the altered route of glucose uptake in ALE-1 via a non-PTS glucose transporter GalP, the expression of which is negatively regulated by GalR. Figure. S5 Comparison of mRNA expression level of galP and talB in ALE-1 with MG1655. The mRNA expression level of galP and talB in MG1655 (gray bar) and ALE-1 (black bar) was analyzed by quantitative RT-PCR. Error bars indicate standard deviations of three independent biological replicates. Figure. S6 Flow cytometric analysis of pHexR-GFP with expression of gnd and/or gntR genes at 9 hours cultivation. Fluorescence histograms of pHexR-GFP ALE-1, ALE-1 pGnd, and ALE-1 pGnd-GntR strain were compared. Gate M1 was used to differentiate the population of wild-type and EMPP-deficient strains based on GFP intensity. Figure. S7 Diverse phenotypic characteristics of pfk_ALE-1. (a, b) Lycopene production was examined in ALE-1 and control strains all harboring the pAC-LYCO04 plasmid. (a) Lycopene titer of E. coli strains after 24 h of cultivation were compared. (b) Color of lycopene-producing strains after 24 h of cultivation. Two milliliters of lycopene-producing cells were harvested after 24 h of cultivation and re-suspended in 200 µL of distilled water. After transfer to 96-well plates, images of cells were taken using a 96-well microplate reader. (c) When nitrate was used as the final electron acceptor under anoxic conditions, OD600 of ALE-1 and MG1655 were measured. For OD600 of MG1655 with nitrate addition, there was no significant difference caused by nitrate addition (data not shown). (d–f) Residual sugar concentration and cell growth of ALE-1 and MG1655 strains when grown in M9 minimal media supplemented with 3 g/L of glucose and 2 g/L of xylose. Residual glucose and xylose concentration of (d) MG1655 and (e) ALE-1 strain. (f) The cell growth of both strains is presented. Figure. S8 Unexpected expression level of the pfkB gene of ALE-1. In the process of deletion of the pfkB gene, nucleotides of 157bp from the 5′ end and 177bp from the 3′ end were detected. Using bowtie software, the sequence reads were mapped onto the wild-type reference genome (NC_000913.2) with the maximum insert size option of 500bp. Remained coding sequence of the pfkB gene affected the expression level of the pfkB gene of ALE-1. Table S1. Strains and plasmid list. Table S2. Primer List. |
|
PMC9648045 | Pooja Ratre,Bulbul Jain,Roshani Kumari,Suresh Thareja,Rajnarayan Tiwari,Rupesh Kumar Srivastava,Irina Yu Goryacheva,Pradyumna Kumar Mishra | Bioanalytical Applications of Graphene Quantum Dots for Circulating Cell-Free Nucleic Acids: A Review | 26-10-2022 | Graphene quantum dots (GQDs) are carbonaceous nanodots that are natural crystalline semiconductors and range from 1 to 20 nm. The broad range of applications for GQDs is based on their unique physical and chemical properties. Compared to inorganic quantum dots, GQDs possess numerous advantages, including formidable biocompatibility, low intrinsic toxicity, excellent dispensability, hydrophilicity, and surface grating, thus making them promising materials for nanophotonic applications. Owing to their unique photonic compliant properties, such as superb solubility, robust chemical inertness, large specific surface area, superabundant surface conjugation sites, superior photostability, resistance to photobleaching, and nonblinking, GQDs have emerged as a novel class of probes for the detection of biomolecules and study of their molecular interactions. Here, we present a brief overview of GQDs, their advantages over quantum dots (QDs), various synthesis procedures, and different surface conjugation chemistries for detecting cell-free circulating nucleic acids (CNAs). With the prominent rise of liquid biopsy-based approaches for real-time detection of CNAs, GQDs-based strategies might be a step toward early diagnosis, prognosis, treatment monitoring, and outcome prediction of various non-communicable diseases, including cancers. | Bioanalytical Applications of Graphene Quantum Dots for Circulating Cell-Free Nucleic Acids: A Review
Graphene quantum dots (GQDs) are carbonaceous nanodots that are natural crystalline semiconductors and range from 1 to 20 nm. The broad range of applications for GQDs is based on their unique physical and chemical properties. Compared to inorganic quantum dots, GQDs possess numerous advantages, including formidable biocompatibility, low intrinsic toxicity, excellent dispensability, hydrophilicity, and surface grating, thus making them promising materials for nanophotonic applications. Owing to their unique photonic compliant properties, such as superb solubility, robust chemical inertness, large specific surface area, superabundant surface conjugation sites, superior photostability, resistance to photobleaching, and nonblinking, GQDs have emerged as a novel class of probes for the detection of biomolecules and study of their molecular interactions. Here, we present a brief overview of GQDs, their advantages over quantum dots (QDs), various synthesis procedures, and different surface conjugation chemistries for detecting cell-free circulating nucleic acids (CNAs). With the prominent rise of liquid biopsy-based approaches for real-time detection of CNAs, GQDs-based strategies might be a step toward early diagnosis, prognosis, treatment monitoring, and outcome prediction of various non-communicable diseases, including cancers.
The ultrasmall QD nanocrystals (1–10 nm), which are based on semiconductors, were revealed in 1983 and have since had great success in the domains of optics, electronics, and catalysis, bringing in a new era of nanotechnology. A heavy metal core surrounded by a bandgap semiconductor shell, such as CdTe, characterizes the vast majority of QDs. SiO2 surrounds PbSe, ZnSe, or CdS core materials, overcoming the surface deficit and boosting the quantum yield. Small particle size, customizable composition and properties, high quantum yield, great brightness, and intermittent light emission are just a few of the intriguing characteristics of QDs that have drawn them into a variety of applications. Metallic QDs like CdS, CdSe, and CdTe were previously the most researched QDs because of their superior optical and electrochemical properties. These QDs have been used for many studies; CdTe/CdSe QD-labeled oligonucleotides and hemin/G-quadruplex DNzyme-conjugated DNA assembly was used for the detection of lysozyme based on the analyte-induced rolling cycling amplification system. In another research study, following the structural and photonic transition of CdTe-QDs immobilized on paper, evoked by the silver ion (Ag+) separated from silver nanoparticles (AgNPs) with a cation exchange reaction, a concise, low-cost visual fluorescence immunosensor was created for disease-related biomarkers in biofluids. For the quantitative or qualitative measurement of prostate-specific antigen (PSA), a titanium carbide (Ti3C2) MXene QDs-encapsulated liposome with excellent photothermal activity was created using a near-infrared (NIR) photothermal immunoassay method. On the other hand, their biological applications are constrained by their cytotoxicity, high energy requirements, and nonrenewable chemical synthesis. To reduce these factors associated with heavy metals, cadmium-free QDs such as carbon, graphene, and silicon have been developed. These QDs exhibit equivalent optical properties. Carbon-based materials play an essential role in developing nanomaterials and fabricating biosensors for various biomedical applications, such as immunoassay of protein, fluorometric immunoassay for carcinogens (aflatoxin B 1), and other cancer biomarkers. One such study describes the development of a photoelectrochemical (PEC) biosensing device for the responsive and precise detection of thrombin utilizing glucose oxidase-encapsulated DNA nanoflowers (GOx-DFs) and graphene oxide-coated copper-doped zinc oxide quantum dots (Cu0.3Zn0.7O-GO QDs) as photoactive substances. In one report, the one-step process was used in the manufacture of carbon quantum dots (CQDs) for quick and accurate metal ion identification. An affordable and ecologically favorable precursor was ascorbic acid. Reactors with high temperatures and pressures were employed for this. In addition to fluorometric assay, these QDs are also employed in electrochemical detection of several analytes, including environmental pollutants. The exceptional and remarkable characteristics of these carbon-based materials have propelled current research and development. Researchers have expended significant effort in exploring carbon nanomaterials for diverse biomedical applications due to their unique physical properties, including optical, structural, and thermal properties. Due to its structure and the delocalization of electrons, graphene, a zero-bandgap semiconductor, cannot emit light (Figure 1). The high mechanical and thermal conductivity of the carbon allotrope graphene has led to the use of graphene-based materials in several biomedical applications, including sensors and bioimaging (Figure 2). Cutting graphene into nanometer-scale pieces and creating a gap therein is a viable strategy to get around this restriction because of the quantum confinement in graphene of any finite size due to its infinite exciton Bohr radius, known as the GQDs. GQDs, the latest zero-dimensional member of the carbon family, are composed of single or several layers of 10-nm-thick graphene sheets. GQDs have demonstrated novel charge transport and light absorption/emission phenomena and their bandgap energy can vary by up to 3 eV depending on their size. GQDs have several distinct advantages over their 2D counterparts, graphene sheets, including bandgap opening brought on by quantum confinement, excellent dispensability, more abundant active sites (edges, functional groups, dopants), better tunability in physicochemical properties, and a size comparable to biomolecules. These advantages enable new application possibilities. The excellent electrical capabilities of sp2 carbon nanoparticles, which have an enormous specific surface area and a lot of functional groups on the edges, are combined with the unique optical and structural advantages of GQDs to create a cutting-edge nanomaterial. However, there has always been a need to find low-cost, nontoxic, eco-friendly, safe, sustainable, and biocompatible materials to synthesize/fabricate. GQDs provide all these features and outshine other classes of QDs owing to their unique properties. Thus, GQDs, a new class of fluorescent materials derived from the carbon nanomaterial family, have perfect chemical and physical properties for usage in biological sensors. An outline of GQD synthesis techniques, including top-down and bottom-up methods, surface conjugation techniques, and their use in identifying circulating nucleic acids is provided here.
The graphene-based materials used to create GQDs cannot show luminescence properties, as graphene is a zero-band gap material. Therefore, manipulating the graphene band gap by various means has sparked considerable interest, including cutting graphene into GQDs to induce photoluminescence (PL) and alteration by various synthetic procedures. Additionally, it has been observed that the synthesis process and precursor material employed affect the characteristics of GQDs. Several synthetic techniques have obtained GQDs with ideal chemical, optical, and electrical properties. Thus, recent advances in GQD synthetic techniques have been fueled by the ease of synthesis and availability of reactants. The top-down and bottom-up approaches are two broad categories used to categorize the synthetic routes toward creating GQDs based on the reaction mechanism involved (Figure 3).
The top-down strategy is well-established and based on widely available bulk precursors such as graphitized materials like graphite and graphene oxide. This strategy aids in tuning various properties such as size and luminescence and obtains the desired GQDs; the bulk carbon materials are typically exfoliated chemically or physically. Therefore, cutting the sp2 or sp3 carbon allotropes of graphene, graphene oxide, carbon fibers, and carbon black using acidic oxidation, microfluidization, exfoliation, and electrochemical oxidation is referred to as top-down techniques (Figure 4). Such methods have several advantages, including an ample supply of raw materials, a simple process with fewer steps, and the large-scale development of bulk reagents and precursors. In addition, the GQDs prepared using this method primarily have the functional group oxygen on their surface and exhibit high water solubility and surface passivation due to the “edge-functionalized GQDs”. As a result, utilizing this type of strategy for controlling morphology appears to be crucial. Some of them are not feasible to implement on a large scale due to their high cost; scalability on a large scale is also tricky; and other issues include the use of acids, high temperatures, and the environment.
This technique primarily employs a higher-energy laser pulse to irradiate the surface target to reach them in a thermodynamic state, resulting in temperature and pressure elevation, as well as heat generation and evaporation, which converts them to a plasma state. Finally, the vapor is collected and crystallized into nanoparticles. The process can also be defined as removing materials from solid or liquid surfaces using laser beams. This method is widely used to modulate the size of GQDs, but it requires multiple steps and purification in a separate step. These GQDs have excellent fluorescent properties to detect different biomarkers. The fluorescence emission spectra of laser ablation-GQDs exhibit blue-shifted emission based on CO-GQDs, which are defined by the size of particles and surface characteristics. GQDs coupled with laser ablation (LA) have an average size of about 1.8 nm and a depth of one layer of graphene. One of the most potent methods for synthesizing unique nanostructures is pulsed laser ablation in liquids (PLAL). It is simple to make GQDs and GOQDs featuring tunable oxygen functional groups by only altering the laser wavelength for PLAL. By laser beam ablation in a fluid, applying pulsed graphene and adding ammonia–water to the graphene solution, nitrogen-doped GQDs were created. In a study, amino-based GQDs were made using polypyrrole (PPy) as an amine source and graphite as a carbon substrate using a one-step pulsed laser ablation (PLA) technique for the sensitive recognition of Fe3.
The action of creating a current between two electrodes (usually graphite rods) that causes them to vaporize is known as arc discharge. As a result, soot is produced, which may contain different graphene-based nanomaterials. Hoffman and Krastchmer used the arc-discharge method to create buckminsterfullerene for the first time in 1990. The arc-discharge technique produces pure, B-doped graphene in the form of soot collected in the electric oven during the arc-discharge process. This is an electrical breakdown of a gas that has developed an in-progress plasma discharge. It is the most suitable procedure for producing thermal plasma characterized by high-energy substances and the local thermal equilibrium state. Although this approach for synthesizing GQDs looks to be a simple conventional method, it is challenging to achieve a high yield and requires careful control of experimental conditions. In a study, a water arc discharge technique with a controlled degree of oxidation is used to manufacture blue-emitting GQDs with tunable PL emission.
Another possible method widely used to synthesize single-layer GQDs with a uniform size and high production yield is the electrochemical cleavage of carbon-based precursors. This technology has been thoroughly studied due to its low cost, high production and reproducibility, and simple operating principles. To convert functional GQDs with an average size of 3–5 nm from graphene film, Qu et al. established an electrochemical method. The synthesized GQDs showed green fluorescence and could last several months in an aqueous solution without losing stability. Electrochemical synthesis offers a route for the more accurate synthesis of GQDs by selectively oxidizing the precursor material by the applied electric potential. It can be further functionalized by altering the electrolyte solution. Due to the relative simplicity of the setup and the absence of powerful oxidizing agents, in the form of precipitation, graphene nanosheets are produced via the electrochemical exfoliation of graphite rods with oxidizing chemicals like KNO3.
Acoustic cavitation is a quick and efficient process for producing nanosized particles in which bubbles play a crucial role after a series of steps, such as production, nucleation, and rapid collapse of bubbles. In liquid, ultrasound can generate alternating low- and high-pressure waves, creating and bursting tiny vacuum bubbles. This cavitation generates high-speed impinging liquid jets and significant hydrodynamic shear forces. Therefore, ultrasonication can turn graphene sheets into GQDs by combining these properties. The one-phase approach used to make the GQDs required expensive equipment, and the environment was also unusual. Zhuo et al. proposed the new method, stating that ultrasonic equipment was used to explain graphene oxidation in concentrated nitric acid and sulfuric acid solutions at room temperature (RT) for 12 hours. The next step involved calcining the received mixture at 350 °C for 20 min to remove concentrated solutions (nitric acid and sulfuric acid). The next step involves filtration with a microporous membrane (0.22 m), resulting in a black suspension from a brown filtered solution. Finally, GQDs are obtained by dialyzing the obtained solution. Graphene oxide (GO) is oxidized ultrasonically to transform into nanometer-sized species, which are further chemically reduced and doped with nitrogen to form a novel catalyst, N-GQDs for electrochemical detection of 2,4,6-trinitrotoluene (TNT). An innovative and effective process was used to create high-quality GO and reduced graphene oxide (RGO) with a high level of durability, utilizing, an alternative probe that was quick, cheap, and environmentally friendly, which also used ultrasonic bath radiation. RGO1 and RGO2 were made in neutral and acidic media, respectively.
The bottom-up method employs small molecules as starting materials, which are then condensed to form a larger entity with a defined size, shape, and required properties for producing GQDs. These techniques include the controllable synthesis of Sp2 carbon from organic polymers and pyrolysis/carbonization processes that begin with organic molecules. Polycyclic aromatic hydrocarbon molecules (glutamic acid, glucose, and citric acid) are typically the most reliable precursors for forming high-quality GQDs. This method is suitable for modulating the size of GQDs, but it requires multiple steps and a separate purification step. The various synthesis methods reported reflect the diversity of carbon precursors used in the bottom-up synthesis (Figure 5).
Hydrothermal synthesis is one of the most used methods for producing GQDs. Carbon precursors, most commonly graphene sheets, are converted into GQDs via oxidation, followed by high-temperature treatment. Compared to other synthetic processes, hydrothermal exfoliation is a more simplistic method for producing GQDs. Hydrothermal processes require water and oxidizing agents such as strong acids or alkali, which are crucial to cutting carbon sources into GQDs. For the first time, Pan et al. reported a novel and simple hydrothermal technique and fabricated aqueous dispersible blue luminescent GQDs by the hydrothermal exfoliation of GO sheets. The synthesis steps involved the oxidation of graphite to GO before thermal treatment, which generated epoxy functional groups. These epoxy groups acted as cleavage points and were completely broken during the hydrothermal reaction, resulting in stable carbonyl groups responsible for GQD’s water solubility. According to Li et al., the hydrothermal method is simple, quick, and suitable for scaling up to develop GQDs in a short time (3 min) using the microwave irradiation (MA-GQDs) method. So, this method exhibited excellent fluorescence quantum with an efficiency of up to 35. Furthermore, the MA-GQDs application for ultrabright fluorescence and stable MA-GQDs is for fluorescence probe and phosphor to prepare white light-emitting diodes in the cell imaging area. Recently published studies using this route discovered the synthesis of high-quality, RGO from GO and KOH, as well as Ag–GO nanocomposites from GO, KOH, and AgNO3 in single fast steps, both of which have antibacterial properties. In a study, humidity sensors have been created using nanocomposites of GQDs and silver nanoparticles (AgNPs), which were produced using a hydrothermal process. In another work, the rapid and precise sensing of the S2-ion accomplished by creating a biosensor using a nanocomposite of fluorescent ionic liquid and GQDs (IL-GQDs) in a single process. The surface modification of GQDs by IL is done under hydrothermal conditions. One-step hydrothermal preparation of blue, fluorescent nitrogen-doped GQDs for the detection of human breast cancer cells (MCF-7 cells) using citric acid and diethylamine as precursors.
The solvothermal procedure, which employs organic solvents like dimethyl sulfoxide (DMSO), dimethylformamide (DMF), and benzene is another synthetic method that can produce GQDs. The solvent’s physicochemical nature directly impacts the final size and morphology of the product in this process. In a closed chamber, a chemical reaction occurs in solvents at temperatures greater than the solvent’s boiling temperature. This approach allows for exact control of particle size and shape distributions by altering the reaction conditions. Iron porphyrin (Fe-N-GQDs) is a new paramagnetic and fluorescent label synthesized that resembles GQDs in nature. The mixing of Fe, N, and C sources was used to make the Fe-N-GQD, which was then exposed to high-temperature pyrolysis before undergoing solvothermal preparation, which is basically used for structural changes in the Fe–N atoms in the graphene lattice. The technique was also used to create a TiO2/Sb2S3/GQDs nanocomposite to explore its antibacterial property.
The direct heating pyrolysis of small molecules has proven to be a simple bottom-up procedure and has given the highest yield without needing any special equipment. Specifically, the small organic-based precursor molecules are heated above their melting point, causing nucleation, condensation, and the subsequent fabrication of GQDs. This is a straightforward bottom-up process for preparing GQDs with sizes ranging from 15 nm to 0.5–2.0 nm in width and thickness. On the other hand, carbonization is an environmentally friendly and simple method for producing GQDs with a uniform size distribution. However, the structure and morphology of the GQDs are uncontrollable, and the yield is lower. Zhao et al. created a simple synthetic method for GQDs by carbonizing l-glutamic acid with a heating mantle device. In a recent work, a fluorescent probe, d-penicillamine (DPA) functionalized GQDs were used, which were made by pyrolyzing citric acid in the presence of DPA for ractopamine quantification in aqueous and plasma samples. The pyrolysis method is also used with hydrothermal for the preparation of N, S-GQDs@Au-polyaniline amperometric immunosensor to detect carcinoembryonic antigen.
Stepwise organic synthesis mediated GQDs fabrication is an efficient solution chemistry method that yields uniform and well-defined GQDs. Furthermore, the low throughput and aggregation of GQDs in solution due to interactions necessitated careful consideration for industrial production. Generally, the interaction of aliphatic side chains with aromatic molecules brings graphene sheets closer, triggering GQDs aggregation. Remarkably, the possibility of graphene wrapping into quasi 0-D fullerene GQDs provided a novel concept for producing well-ordered GQDs from fullerene via cage opening. In their study, Kaciulis et al. state that fullerene is added to a mixture of sodium nitrate, KMnO4, and concentrated H2SO4 to fabricate GQDs as fluorescent sensors. Lu et al. used the ruthenium-catalyzed cage opening technique of C60 to generate very small GQDs. The ruthenium surface develops strong contacts with the C60 molecules, resulting in a surface vacancy on the ruthenium and aiding the C60 molecules in becoming buried in the surface. Embedded molecules are fractured as the temperature rises, producing more carbon clusters that aggregate and diffuse to create GQDs. The shape or form of the GQDs can then be fixed by adjusting the annealing temperature.
CVD is widely used for creating nanoparticles with monolayer architectures and graphene sheets. This is a technique for laying down gaseous reactants on a substrate. Fan et al. created the CVD-grown GQDs first, using copper foil as a substrate and methane as a carbon source. According to the DLS analysis, the resulting GQDs had a broad size distribution (5–15 nm), whereas the height profiles (1–3 nm) suggested the formation of GQDs with a few layers. The sole difference between CVD and PVD (Physical Vapor Deposition) is that solid reactants are used instead of gaseous ones in CVD. Here, the carrier gases are combined in a reaction chamber that is kept at a specific temperature and pressure. Furthermore, the reaction occurs on the substrate, where the finished product, such as graphene, is deposited, and the byproducts are pumped away. The substrate is usually made of a transition metal (Ni/Cu) or ceramic-like glass. Finally, the substrate is chosen based on the graphene’s ability to be transferred to the required substance. Chemical vapor deposition is used to generate the 3D graphene, a new electrochemical process for producing high-quality GQDs from monolithic 3D graphene. Although the hydrothermal and ultrasound-aided methods are both quick and environmentally friendly, it is challenging to synthesize them on a large scale in industry. Before a hydrothermal reaction can take place, the raw materials must be treated with a potent oxidant. The reaction also requires high temperatures and high pressure, which might result in combustion or an explosion. Although H2SO4, HNO3, or other oxidants are required for the chemical oxidation process, which is now the most commonly used method, they may also result in corrosion or explosions. Even though the electrochemical oxidation process produces GQDs of uniform size, the pretreatment of raw materials and the output yield are both poor, making it challenging to carry out large-scale manufacturing. The microwave process involves filtering and purification, which makes it challenging to employ for large-scale manufacturing despite its quick reaction time. Although the pyrolysis process is an environmentally friendly way to produce GQDs, it is unable to regulate the size and structure of GQDs. Whereas electron beam irradiation, which is a quick and high-yield approach, it is not frequently employed since it necessitates pricey specialized equipment and poses a radiation danger to the user.
GQDs have demonstrated superior qualities in terms of chemical inertness, bioactivity, and viability, but there are still some challenges that prevent their use in bioimaging, such as relatively low luminescence quantum yields (the quantum yields of most GQDs are less than 10%), shifting fluorescence emissions, and an ambiguous luminescence process. As a result, a lot of work was done to use surface chemistry to boost the quantum yields (QY) and surface activity of GQDs (Figure 6).
The water-soluble GQDs surface can absorb some biomolecules. This mechanism is nonspecific and is influenced by various factors, including the molecule’s surface charge. Passive adsorption is a systematic and accessible approach for GQDs bioconjugation. Electrostatic adsorption occurs when species with opposing charges attract each other, producing a nonspecific interaction involving the NP and the biomolecule. Negatively charged GQDs typically interact with positively charged biomolecules to form noncovalent conjugates. GQDs were joined to several biomolecules using this bioconjugation technique, including proteins, porphyrins, lectins, polysaccharides, and nucleic acids. Hydrophilic NPs can also develop electrostatic interactions alongside polar molecules through their surface coatings. In general, proteins possess a hydrophilic surface and a hydrophobic core; therefore, they must shift configuration to associate with nonpolar particles. Since these structural changes might result in protein denaturation and a reduction of biological activity, electrostatic adsorption is usually the preferred strategy. Negatively charged semiconductor materials with positively charged biomolecules are a widely adopted approach for noncovalent GQD conjugation. In a report, GQD@MnO2 nanocomposites were created by the adsorption of MnO2 nanosheets to the edge of GQDs in order to detect internal glutathione-related tumors (GSH).
Several biomolecules and particles have been conjugated using the (strept)avidin–biotin combination. This tactic relies on avidin or streptavidin’s naturally high-affinity interactions with biotin, comparable to those between receptors and ligands or enzymes and substrates. The intensity of the (strept)avidin–biotin combination benefits bioconjugation chemistry. It is robust to pH, buffer salts, temperature variations, and process adjustments such as multiple washing steps. One biotin-binding site is present on each of the four identical subunits that make up the glycoprotein known as avidin. A tiny molecule called biotin, also referred to as vitamin H or vitamin B7, can be added to biomolecules or particles without changing their function or nature. As biotin has a carboxylic acid, it can be covalently conjugated to many species. Biomolecules can be chemically prepared with an amine, thiol, or carboxyl-reactive biotin reagent to biotinylate them, or they can be genetically altered to have a biotin acceptor. Therefore, the (strept)avidin–biotin binding is one of the most substantial noncovalent interactions, approaching the strength of a covalent bond. As a result, the procedure is frequently referred to as the “covalent conjugation method”. This conjugation process is frequently adopted. Using this method, an electrochemical biosensor was developed for direct detection of miRNA-21. However, it has a major drawback due to the vast size of its derived component structure, which is a protein. As a result of this constraint, this approach is rarely employed rather than the carbodiimide coupling technique.
As a result of the fact that no part of its chemical structure is incorporated into the final bond between conjugated molecules, this chemistry is referred to as zero-length (carboxyl to amine) cross-linking agents. EDC (1-Ethyl-3-(3-(Dimethylamino)propyl) Carbodiimide) combines with carboxylic acid to produce an intermediate, which then reacts with the amine to produce a conjugate containing an amide bond. EDC is frequently used in conjugation with an adjuvant, such as N-hydroxysuccinimide (NHS) or sulfo-NHS (an NHS water-soluble molecule), among the water-soluble carbodiimides. When the oxygen in the carboxylic acid combines with the carbodiimide, a very reactive intermediate is formed that can combine with amines to form an amide bond. When NHS is added, a secondary, more soluble, and more robust intermediate form then interacts with the amine to form the final product. Water-soluble carbodiimides are preferred for GQD conjugation because they enable the reaction to proceed in aqueous buffer solutions. Hou et al. used EDC-NHS for surface incubating GQDs to design an ultrasensitive electrochemiluminescence biosensor for specific detection of miRNA. Guitao et al. in 2019 developed a novel graphene QDs ECL biosensor using EDC-NHS to detect circulating DNA by a cycle amplification method. Kong et al. use graphene films on gold substrates, as working electrodes for electrochemical detection of nucleic acid (microRNA) miR-155 as a biomarker for the diagnosis of various diseases. They used EDC and NHS as coupling agents for the self-assembly monolayer (SAM)-modified gold substrate. The conjugation is also used to create cytometric-based nanobiosensing systems that directly quantify cell-free circulating (ccf) epigenomic signatures like methylated ccf-DNA, trimethylated histone H3 at lysine, and protein-bound argonaute 2 ccf-miRNAs.
The cycloaddition of azides and alkynes is a bio-orthogonal technique often utilized for GQD bioconjugation. The Huisgen cycloaddition, also known as the azide-terminal alkyne reaction, begins with a five-membered ring of triazole, a heterocyclic molecule containing three nitrogen atoms. Initially, this reaction was performed at maximum temperatures to enhance its yield. Some years later, it was established that this cycloaddition could be catalyzed by CuI and generate high yields of the heterocycle ring even at RT. For this reason, the CuI-catalyzed cycloaddition of azides and terminal alkynes was termed a “click-chemistry reaction”. Proteins, viruses, antibodies, and miRNAs have been coupled to QDs coated with azides or alkynes and suitably functionalized. Thiol–ene click reactions in a single step when compared to the conventional synthesis of GQDs. A unique technique for making GQDs from GO was applied, and it proved to be both inexpensive and effective with exceptional qualities, including their homogeneous nano size, robust green fluorescence, consistent stabilities, and great bioactivity.
The bioconjugation method involving the interaction of hydrazine derivatives with aldehydes or ketones is appealing from a biorthogonal perspective. Hydrazine derivatives respond quickly, particularly with aldehyde or ketone functional groups, making a hydrazone bond that is a sort of Schiff base. However, this reaction is faster with aldehydes than with ketones; the hydrazone bonds formed with ketones are steadier than those formed with aldehydes. Aldehyde-reactive chemical groups like hydrazides and alkoxyamines are frequently utilized in biomolecular probes to label and cross-link carbonyls (oxidized carbohydrates) on glycoproteins and other polysaccharides. At pH 5 to 7, hydrazides and aldehydes produced by periodate-oxidation of sugars in biological samples interact to form hydrazone bonds. The majority of protein-labeling applications can be completed using the hydrazone bond. The main advantage of this strategy is that biological systems mainly do not have the aldehyde and hydrazine groups. However, the biological system may consist of amines that react with aldehydes. The hydrazine–aldehyde coupling has been used to conjugate QDs to antibodies, synthetic peptides, and viruses. This approach can also be applied to oxime derivatives instead of hydrazines.
The planar graphitic domain’s extensive sp2 hybridization creates the possibility for functionalization in the absence of oxygen functionality or hydrogen bonds using π–π stacking or van der Waals forces. Unsaturated (poly)cyclic molecules establish a specific type of dispersion force from van der Waals forces known as the π–π interaction. Since graphene contains a hexaatomic ring of carbon atoms, it can spontaneously stack on aromatic biomolecules. Along with the hydrogen bonds between pairs of complementary nucleotides, these interactions significantly contribute to stabilizing DNA’s double helical structure. In addition to having less of an adverse effect on the structure of graphene materials, noncovalent functionalization based on the hydrophobic attraction, interaction, or van der Waals force between graphene materials and stabilizers also makes it possible to tailor their solubility and electronic properties. As an illustration, Green and colleagues investigated several functionalized pyrene derivatives and showed that these species could maintain single- and few-layered solvent-exfoliated graphene flakes in aqueous dispersions. Recent research has shown that small peptides assemble toward the planar surfaces or edges of GO through π–π interactions. Immobilized peptide-based GO materials have much promise for creating susceptible and versatile detection platforms. A non-single-stranded DNA (ssDNA) molecule was used as a biorecognition molecule in the first GO-based sensing mechanism study. According to a study, graphene-based nanostructures can interact with ssDNA molecules through π–π stacking interactions because they contain π-rich conjugation domains. Rafiei et al. created a GQDs-DNA nano assembly as a biosensor by using stacking to interact with ssDNA. Yew et al. also used GQDs for DNA detection, using a FAM-L probe that adsorbs onto the GQDs upon incubation via π–π stacking interactions.
Circulating nucleic acid (CNAs) includes various forms of nucleic acid like DNA, RNA, micro RNA, lncRNA, and mitochondrial DNA in plasma and serum. CNAs are released as nucleosomes when cells undergo apoptosis or necrosis. Apart from that, CNAs could be produced by the active metabolic release of DNA from cells. CNAs can be found in healthy and diseased bodies, with diseased ones having higher levels. Increased levels of circulating nucleic acids in the blood can signal some malignant and benign disorders. Although protein biomarkers have been identified, CNAs may be a better biomarker since they are more relevant, precise, and accurate than protein biomarkers. Their dysregulation is generally observed in tumors, which can be used as a diagnostic for malignancy. As a result, circulating nucleic acids are developing into a valuable resource for studying several chronic diseases, including cancer, and acting as biomarkers. The various CNAs types being examined include ccf-DNAs, ccf-RNAs, ccf-mtDNAs, ccf-miRNAs, ccf-lncRNAs, etc. The use of ccf-DNAs in clinical practice for a few diseases has already been made possible by developing newer technologies to isolate minute quantities of ccf-NAs and detect the unique signatures on these. It is crucial to determine the function of these ccf-NAs as epigenetic biomarkers in clinical settings because they are linked to various epigenetic modifications that exhibit disease-related variations. The field of noninvasive molecular diagnosis has undergone a revolution, with conventional screening and treatment techniques being replaced by epigenetic markers. The epigenetic markers for these ccf-NAs reflect the pattern unique to the tissue that produced them. Therefore, epigenetic biomarkers can aid in diagnosing a variety of diseases even before the appearance of actual symptoms, which will aid in better disease management. Numerous studies are being conducted to determine whether certain clinical condition-specific epigenetic marks exist on ccf-NAs. Despite the advancement of techniques for examining epigenetic changes, the application of epigenetic biomarkers discovered on ccf-NAs is limited due to their lower blood circulation levels. The detection and quantification of ccf-NAs, viz., RNA, fetal DNA, fetal RNA, mtDNA, mitochondrial RNA, and miRNA levels, in body fluids are of clinical importance. These ccf-NAs may serve as biomarkers for the diagnosis and prognosis of several diseases. Because of this, ccf-NAs are important in the pathogenesis and diagnosis of many diseases. Though the clinical utility of ccf-NAs is being widely recognized, in-depth characterization is warranted to ensure usage in point-of-care settings. ccf-NAs are well-known biomarkers used in prenatal diagnosis to screen for genetic abnormalities in fetuses. ccf-NAs (ccf-DNAs and cell-free noncoding RNAs) may be promising biomarkers in the diagnosis and prognosis of cancer, cardiovascular and neurological illnesses, and diabetes, according to growing data. Cell-free circulating tumor DNA, circulating tumor RNA, circulating tumor cells, and exosomes are all significant tumor carriers of genetic data in the blood. Because of its stability and ease of access, ccf-DNAs are an appealing alternative as a diagnostic, predictive, and prognostic biomarker for analyzing tumor genetic information utilizing GQDs (Table 1). Plasma ccf-DNAs levels have been linked to tumor size, invasion, cancer stage, survival, and treatment-related disease progression. Above all, microRNAs (also known as miRNAs or miRs) have drawn more attention because of their extensive involvement in regulating cellular processes. These quick 20–22-mer RNA sequences play a key role in the post-transcriptional precise control of several physiological cell functions, such as cell division, proliferation, and signaling. Only a percentage of tumor-derived DNA with diagnostically important mutations is present in ccf-DNAs, which is fragmented to an average length of 140 to 170 bp and expressed in low quantities per milliliter of peripheral blood. Several strategies have been introduced to detect low-level tumor-associated mutations in cancer patients’ ccf-DNAs. Finally, the small size of GQDs and properties like quantum confinement and edge effect are vital benefits for the further development of this diagnostic technique, given that one of the most desirable fields of application is their use as fluorescent tags and success in sensor research. Their great sensitivities and effectiveness, particularly when paired with additional methods like electrical and optical methods, make GQDs an effective tool in bioanalysis and the detection of biological targets.
Due to their remarkable physical and chemical characteristics, GQDs, the next generation of the graphene family, have been demonstrated to be the best sensing components for detecting circulating nucleic acids. These GQDs with various biomolecules can use optical, electrochemical, and chemiluminescent biosensors to selectively recognize and transform into a signal-specific ccf-NAs biomarker. Numerous studies have been conducted to ascertain how to alter electrode surfaces with nanosized materials with sizes between 1 and 100 nm, derived from organic or inorganic sources, to provide biosensors with increased reproducibility, selectivity, and sensitivity. Given their large surface-to-volume percentage and large specific surface area, nanomaterials have high adsorption of target analytes. Utilizing neodymium-doped BiOBr nanosheets (Nd-BiOBr) as a photoactive substrate, a photoelectrochemical bioassay for dopamine-loaded liposome-encoded magnetic beads is being developed to measure the amount of DNA associated with the human papillomavirus (HPV). Another recent study used CRISPR-Cas12a trans-cleaving the G-quadruplex for biorecognition/amplification and a hollow In2O3–In2S3-modified screen-printed electrode (In2O3–In2S3/SPE) as the photoactive material to identify the human papillomavirus-16 (HPV-16) on a foldable electrochemical detector. GQDs, the new class of fluorescent materials from the carbon nanomaterials family, possess ideal chemical and physical properties to be used and integrated into sensors for biological and medical applications. The optical characteristics of GQDs can be used in biosensing; this application uses the PL of GQDs and generally requires photon detection. GQDs, serve the purpose of detecting and indicating the presence of nucleic acid biomarkers in biosensing systems. GQD-based biosensors utilize the affinity between specific functional groups within GQDs and the analyte biomolecule. When a functional group conjugated onto the GQDs binds to the analyte, the association between the pair can provide different electronic states. A change in PL intensity can be used to measure the detection of an analyte by changing the electronic structure of the GQDs.
The π–π* shift of the CC bonds in GQDs makes them a popular choice for photon capturing in the shorter wavelength range. They exhibit more excellent optical absorption in the UV range of 260 to 320 nm, with a tip that continues into the visible spectrum. As a result, they become more effective at absorbing long wavelengths. Apart from that, these GQDs have a broad peak around 270 and 390 nm, indicating that they are involved in the n−π* transition of the CO bonds.
Since these characteristics are linked to the band gap of GQDs, the optical characteristics of GQDs are conceptually dependent on inherent variables like size, layer, shape, or edge orientation. Due to the −* transition of CC bonds, GQDs exhibit high optical absorption in the UV region at 230 nm. Additionally, a shoulder peak across the range of 270–390 nm caused by the n–* transition of C–O bonds was seen. The strong optical features of GQDs with distinct identification or dual emissions are susceptible to being embellished with additional distinct molecules. GQDs are superior PL sensors for detecting interesting analytes as compared to conventional organic dyes and semiconductor QDs probes because they offer great sensitivity, selectivity, stability, and security for biosensing systems.
The GQDs exhibit more QY than the bare CDs. This is because of the structure’s layering and the suitable crystalline property. The optical character of GQDs with a luminescence mechanism is also a challenge. The potent fluorescent GQDs, along with their layered structure, are used for confocal imaging of cancer cells with the help of different solvents. They showed well-designed PL emission of the GQDs in distinct liquid solutions, i.e., solvents, stating the strong PL emission of the GQDs due to various edge locations and functional groups linked to the GQDs. GQDs have a single-layer carbon core with chemical groups on the surface or edge. It has oxygen-based functional groups on the basal plane or at the edges. The states of the sp2 sites determine the fluorescent property. The fluorescence can be triggered by recombining electron–hole pairs in such sp2 clusters. The bandgaps of different sizes of sp2 cover a wide range in GO due to the vast size distribution of sp2 domains, resulting in a broad PL emission spectrum from visible to near-infrared. GQDs possess more defects, oxygen groups, and functional groups on the surface. The excitons in graphene have an infinite Bohr diameter. As a result, quantum confinement effects will be observed in graphene fragments of any size. As a result, GQDs have a nonzero bandgap and PL on excitation.
The working, counter, and standard electrodes are all included in an electrochemical biosensor. On the edge of the operating electrode, a chemical reaction occurs between the immobilized biomolecules and the relevant analyte, producing or consuming ions or electrons. These ions or electrons produce a voltage at the reference electrode and provide a signal that may be measured. The real-time, quick, sensitive, specific, and accurate detection and quantification of biomarkers at extremely low cutoff values for early diagnosis have been significantly improved by the combination of nanoparticles with electrochemical biosensing. In their research, they created an easy and sensitive electrochemical biosensor built on enzyme action using GQDs as a novel platform for immobilization for the efficient detection of miRNA-155.
Even with numerous advantages, there are some limitations to GQD uses, such as low yield and higher dispensability. Other drawbacks related to the method of preparation, like in the top-down approach, include the need for expensive equipment, setups, and materials. The situations are also critical and take longer than usual to rectify. The pathway for their preparation brings numerous drawbacks by using graphene, their oxide, and carbon fibers in the minor pieces as they are sometimes toxic. The other scheme has drawbacks like potent safety risks, environmental pollution, premium costs, and brutal methods for fabrication and after-processing methods. So, finding a method that favors the environment and should be eco-friendly while also originating from original greener precursors is challenging in this area. The main obstacle to using GQDs for the creation of sensing devices is the large-scale synthesis of high-quality and stable nanoparticles due to their specific size, shape, and charges, as these characteristics have a significant impact on the physicochemical properties of these nanomaterials and, consequently, on the performance of GQD-based sensors.
Of late, significant and rapid advancements in nanophotonics for prognosis and early diagnosis of various noncommunicable diseases and age-associated degenerative illnesses are of current interest. Semiconductor nanocrystals have provided an innovative milieu for qualitative and quantitative analyses of multiple analytes in the peripheral circulation. In this respect, QDs demonstrate significant potential in biomedical, bioimaging, photoluminescent, and fluorescent-based applications. GQDs have received considerable attention due to their unique properties, such as excellent solubility, robust chemical inertness, large specific surface area, superabundant surface conjugation sites, superior photostability, resistance to photobleaching, and nonblinking. In addition, their optical characteristics can be adjusted via size tunability, chemical doping, and surface functionalization for featured and specific applications. In this review, we have sought to showcase a comprehensive picture of the most recent advances in research, focusing on their biosensing applications. Following an in-depth discussion on the potential in vitro and in vivo bioimaging applications of GQDs, current progress in fabrication methodologies, including top-down and bottom-up, has been critically examined. In addition to these features, the review goes through the various surface conjugation approaches. In recent times, numerous reports have demonstrated that GQDs are developing into critical functional nanomaterials with applications in the medical, optoelectronic, and energy-related fields. However, the principle in several GQDs systems is currently unexplained, necessitating further research. The zero-dimensional GQDs showed great promise among the various nanosized substrates for detecting CNA biomarkers because of their exceptional electronic and optical properties, large surface area, and various active sites for chemical functionalization. GQDs can offer sophisticated sensing substrates for quick and accurate diagnosis at the point of care and the monitoring of therapeutic progress thanks to their capacity to create covalent connections with proteins that can identify numerous nucleic acid biomarkers for several chronic ailments, including cancer. Detection of ccf-NAs, DNAs, mtDNAs, mRNAs, miRNAs, or lncRNAs can help identify multiple cancers at the early stage. Often, these liquid biopsy methods utilize a state-of-the-art technology platform that facilitates the identification of ccf-NAs in peripheral circulation and localizes the tissue of origin. In this regard, GQDs facilitate real-time quantification of these molecules by conjugating the reporter to target entities, followed by detection by fluorescence excitation and acquisition of emission. Due to the inherent ability, GQDs-based conjugation strategies have helped enhance excitation and efficiently captured the photoemission in the presence of various noise signals. By engineering the spectral overlap, multiplexing strategies can be rationally designed to identify different target sequences of cell-free circulating nucleic acids in any test sample. With the prominent rise of liquid biopsy-based approaches, GQDs-based methods of detection might be a step toward early diagnosis, prognosis, treatment monitoring, and outcome prediction of various noncommunicable diseases, including cancers. |
||
PMC9648046 | Pei Zhang,Yang Du,Hua Bai,Zhijie Wang,Jianchun Duan,Xin Wang,Jia Zhong,Rui Wan,Jiachen Xu,Xiran He,Di Wang,Kailun Fei,Ruofei Yu,Jie Tian,Jie Wang | Optimized dose selective HDAC inhibitor tucidinostat overcomes anti-PD-L1 antibody resistance in experimental solid tumors | 09-11-2022 | Tucidinostat,Tumor microenvironment,PD-L1,CCL5,Solid tumor | Background Although immune checkpoint inhibitors (ICIs) have influenced the treatment paradigm for multiple solid tumors, increasing evidence suggests that primary and adaptive resistance may limit the long-term efficacy of ICIs. New therapeutic strategies with other drug combinations are hence warranted to enhance the antitumor efficacy of ICIs. As a novel tumor suppressor, histone deacetylase (HDAC) inhibitor tucidinostat has been successfully confirmed to act against hematological malignancies. However, the underlying mechanisms of action for tucidinostat and whether it can manipulate the tumor microenvironment (TME) in solid tumors remain unclear. Methods Three murine tumor models (4T1, LLC, and CT26) were developed to define the significant role of different doses of tucidinostat in TME. The immunotherapeutic effect of tucidinostat combined with anti-programmed cell death ligand 1 antibody (aPD-L1) was demonstrated. Furthermore, the effect of tucidinostat on phenotypic characteristics of peripheral blood mononuclear cells (PBMCs) from lung cancer patients was investigated. Results With an optimized dose, tucidinostat could alter TME and promote the migration and infiltration of CD8+ T cells into tumors, partially by increasing the activity of C-C motif chemokine ligand 5 (CCL5) via NF-κB signaling. Moreover, tucidinostat significantly promoted M1 polarization of macrophages and increased the in vivo antitumor efficacy of aPD-L1. Tucidinostat also enhanced the expression of the costimulatory molecules on human monocytes, suggesting a novel and improved antigen-presenting function. Conclusions A combination regimen of tucidinostat and aPD-L1 may work synergistically to reduce tumor burden in patients with cancer by enhancing the immune function and provided a promising treatment strategy to overcome ICI treatment resistance. Supplementary Information The online version contains supplementary material available at 10.1186/s12916-022-02598-5. | Optimized dose selective HDAC inhibitor tucidinostat overcomes anti-PD-L1 antibody resistance in experimental solid tumors
Although immune checkpoint inhibitors (ICIs) have influenced the treatment paradigm for multiple solid tumors, increasing evidence suggests that primary and adaptive resistance may limit the long-term efficacy of ICIs. New therapeutic strategies with other drug combinations are hence warranted to enhance the antitumor efficacy of ICIs. As a novel tumor suppressor, histone deacetylase (HDAC) inhibitor tucidinostat has been successfully confirmed to act against hematological malignancies. However, the underlying mechanisms of action for tucidinostat and whether it can manipulate the tumor microenvironment (TME) in solid tumors remain unclear.
Three murine tumor models (4T1, LLC, and CT26) were developed to define the significant role of different doses of tucidinostat in TME. The immunotherapeutic effect of tucidinostat combined with anti-programmed cell death ligand 1 antibody (aPD-L1) was demonstrated. Furthermore, the effect of tucidinostat on phenotypic characteristics of peripheral blood mononuclear cells (PBMCs) from lung cancer patients was investigated.
With an optimized dose, tucidinostat could alter TME and promote the migration and infiltration of CD8+ T cells into tumors, partially by increasing the activity of C-C motif chemokine ligand 5 (CCL5) via NF-κB signaling. Moreover, tucidinostat significantly promoted M1 polarization of macrophages and increased the in vivo antitumor efficacy of aPD-L1. Tucidinostat also enhanced the expression of the costimulatory molecules on human monocytes, suggesting a novel and improved antigen-presenting function.
A combination regimen of tucidinostat and aPD-L1 may work synergistically to reduce tumor burden in patients with cancer by enhancing the immune function and provided a promising treatment strategy to overcome ICI treatment resistance.
The online version contains supplementary material available at 10.1186/s12916-022-02598-5.
Programmed cell death protein 1 (PD-1) is an immune checkpoint receptor expressed on activated T cells that modulate tissue immune tolerance [1, 2]. Tumor cells frequently overexpress the ligand programmed death ligand-1 (PD-L1), thereby facilitating their escape from immune surveillance [3–5]. Monoclonal antibodies (mAbs) against PD-1 or PD-L1 have demonstrated remarkable clinical efficacy in patients with a variety of cancers [3]. However, accumulating evidence suggests that mAbs against PD-1 and PD-L1 are less effective in non-inflamed tumors, indicating that such tumors are resistant to immune attack. Indeed, tumors unresponsive to PD-1 or PD-L1 mAbs are characterized by poor lymphocyte infiltration, low PD-L1 expression, and increased immunosuppressive factor expression in the tumor microenvironment (TME). Combining PD-1 or PD-L1 mAbs with certain agents that can modulate the immunosuppressive state may overcome the primary and adaptive resistance [6, 7]. Cytotoxic chemotherapy or molecularly targeted therapy has been demonstrated to enhance the effect of PD-1/PD-L1 mAbs as immunotherapeutic drugs [8–10]. Several studies have suggested a bidirectional relationship between epigenetic modifications and antitumor immunity in TME [11–13]. Cancers can be caused not only due to a change in the genomic DNA sequence but also through two typical epigenetic modifications: DNA methylation and histone modification [14]. These epigenetic modifications remodel the chromatin structure, thereby altering the gene expression profile and cell phenotype and potentially resulting in cell cycle dysregulation and tumor development [15–17]. Conversely, the reversal of histone and non-histone protein acetylation by histone deacetylase (HDAC) inhibitors can induce cell cycle arrest, differentiation, and cancer cell death [18, 19]. Such antitumor effects of HDAC inhibitors have been proven in human hematological tumors, but not yet in solid tumors. Recently, preclinical studies have reported the efficacy of HDAC inhibitors combined with other therapeutic agents, including chemotherapy or targeted drugs, in treating solid tumors [20–22]. Trials investigating these combinations in patients with solid tumors are also ongoing, and the preliminary data obtained seem promising. However, the antitumor effects of HDAC inhibitors combined with immune checkpoint inhibitors (ICIs) have not been extensively studied [23, 24]. Challenges remain in harnessing the full potential of combined HDAC inhibition and immunotherapy and selecting optimal regimens for different solid tumors. Tucidinostat, an oral HDAC inhibitor belonging to the benzamide class and having specificity for HDAC1, HDAC2, HDAC3, and HDAC10 subtypes, has been approved for the treatment of relapsed or refractory peripheral T cell lymphoma and is under clinical development globally for various other neoplastic and non-neoplastic diseases [25–29]. A recent phase III trial reported that tucidinostat combined with endocrine blockade could be effective against advanced hormone receptor-positive HER2-negative breast cancer progression after endocrine therapy alone [30]. Additionally, grade 3 or 4 toxicities including neutropenia, leucopenia, and thrombocytopenia were more frequent in the tucidinostat group than in the placebo group, indicating that tucidinostat may induce immunosuppression, which is not conducive to effective immunotherapy. Therefore, assessing the effect of different doses of tucidinostat in the presence or absence of ICIs on solid tumor growth and TME immune status is warranted to discover an optimized combination therapy. Here, we analyzed the antitumor efficacy of different doses of tucidinostat alone and in combination with aPD-L1 in three murine solid tumor models that were unresponsive or transiently responsive to ICIs to explore the optimal strategy for combining tucidinostat and aPD-L1, as well as its underlying mechanisms. This may provide guidance to improve the clinical management of combined immunotherapy.
Tucidinostat (Cat# HY-109015), vorinostat (Cat# HY-10221), TMP-195 (Cat# HY-18361), and NF-κB inhibitor BAY11-7082 (Cat# HY-13453) were obtained from MCE (Monmouth Junction, NJ, USA). Dimethyl sulfoxide (DMSO) has been used as vehicle control for each drug.
4T1 breast cancer cells, Lewis lung cancer (LLC) cells, CT26 colorectal cancer cells, and Raw 264.7 cells were purchased from the Chinese Academy of Sciences (Beijing, China) and cultured in RPMI-1640 medium (Hyclone) containing 10% fetal bovine serum at 37°C in a 5% CO2 incubator.
For subcutaneous injections, mouse 4T1, LLC, and CT26 cells (5 × 105 cells) were injected into the right flank of BALB/c or C57 BL/6 mice. When established tumors were palpable 7 days after tumor cell inoculation, mice were treated with different doses of tucidinostat (MCE, Cat# HY-109015, 12.5, 25, 75 mg/kg, oral gavage, daily) and aPD-L1 (BioXcell, Cat# BE0101; 200 μg, intraperitoneal injection, every 3 days). The volume of tumor nodules was measured every 3 days and calculated as V = (a × b2)/2, where “a” and “b” are the long and short axis of the tumor nodule, respectively. Mice were monitored until their individual tumor volume reaches the approved protocol volume limit (2000mm3). At the treatment, the tumor-bearing mice were anesthetized and tissues were harvested for further analysis. Depletion of CD8+ T cells was performed by intraperitoneal injection of anti-mouse CD8a (aCD8, BioXcell, Cat# BP0117; 200 μg, every 3 days). After aCD8 treatment, the percentage of CD8+ T cells (CD3+CD4−CD8+ T cells) was significantly decreased in the tumor and spleen tissues. Depletion of macrophage was performed by intraperitoneal injection of clodronate liposomes (FormuMax, Cat# F70101C-A; 1.4 mg/20g body weight, every 3 days), respectively. After clodronate liposome treatment, the percentage of macrophages (CD11b+F4/80+ macrophages) was significantly decreased in the tumor and spleen tissues. Animal studies were conducted in accordance with the NIH animal use guidelines and approved by the Institutional Review Board of the National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College (Permit Number, NCC2020A167).
Total RNAs were extracted using the RNeasy Kit (Takara Bio). The qRT-PCR was carried out using SYBR Green Premix Ex TaqTM II (Takara Bio) on a ABI StepOnePlus Real-Time PCR Detection System (Thermo Fisher Scientific). Results were normalized to the housekeeping gene GAPDH. Relative gene expression levels from different groups were calculated with the 2-ΔΔCT method and compared with the expression level of appropriate control cells. Specific primer sequences for individual genes were as follows: CCL5 (forward: 5′-GTATTTCTACACCAGCAGCAAG-3′; reverse: 5′-TCTTGAACCCACTTCTTCTCTG-3′); CXCL9 (forward: 5′-AATCCCTCAAAGACCTCAAACA-3′; reverse: 5′-TCCCATTCTTTCATCAGCTTCT-3′); CXCL10 (forward: 5′-CAACTGCATCCATATCGATGAC-3′; reverse: 5′-GATTCCGGATTCAGACATCTCT-3′); PD-L1 (forward: 5′-TGAGCAAGTGATTCAGTTTGTG-3′; reverse: 5′-CATTTCCCTTCAAAAGCTGGTC-3′); iNOS (forward: 5′-GCCGAGTGCAAGCATGGAGAG-3′; reverse: 5′-GGCTGTGAGGTGAGGTTGAAGAAG-3′); CD86 (forward: 5′-ACGGAGTCAATGAAGATTTCCT-3′; reverse: 5′-GATTCGGCTTCTTGTGACATAC-3′); CD206 (forward: 5′-CCTATGAAAATTGGGCTTACGG-3′; reverse: 5′-CTGACAAATCCAGTTGTTGAGG-3′); Arg1 (forward: 5′-CATATCTGCCAAAGACATCGTG-3′; reverse: 5′-GACATCAAAGCTCAGGTGAATC-3′); and GAPDH (forward: 5′-GTATTTCTACACCAGCAGCAAG-3′; reverse: 5′-TCTTGAACCCACTTCTTCTCTG-3′).
The cell culture dish was placed on ice and add with ice-cold lysis buffer. The cell suspension was gently transferred into a pre-cooled microcentrifuge tube and centrifuged in a microcentrifuge at 4°C for 20 min at 12,000 rpm. The protein was collected and separated by 10% SDS-PAGE gel and transferred onto a polyvinylidene difluoride membrane (Millipore). Subsequently, the membrane was blocked and incubated overnight at 4°C with the primary antibody including anti-CCL5 mAb (1: 500, Abcam, Cat# ab7198) and anti-phospho-NF-κB p65 (1:1000, Cell Signaling Technology, Cat# 3039). The same membrane was probed for GAPDH (1:10,000, Abcam, Cat# ab8245) as the internal control. After washing with TBST solution, the membrane was incubated with the corresponding secondary Abs. Finally, the blots were developed in ECL reagent (Thermo Fisher Scientific Inc, Cat# 32209) and imaged using the ImageQuant LAS 500 system (GE Healthcare).
To assess the effect of different doses of tucidinostat on cell proliferation, 4T1, LLC, and CT26 cells were plated in 96-well plates at a started number of 3 × 103 cells/well and treated with different doses (2.5, 5, 7.5 μM) of tucidinostat for 24 h. The absorbance of each sample was measured using a Cell Counting Kit-8 (CCK-8) kit (Solarbio) on a microplate reader (Thermo Scientific) at 450 nm. The ratio of the optical density (OD) value of each cell group normalized to the cells without tucidinostat treatment is presented. The experiment was performed in triplicate.
To assess the effect of different doses of tucidinostat on cell apoptosis, 4T1, LLC, and CT26 cells were plated in 6-well plates at a started number of 3 × 105 cells/well and treated with different doses (2.5, 5, 7.5 μM) of tucidinostat for 6 h and then were collected and incubated with Annexin V-FITC and PI (Beyotime Biotechnology). After a 15-min incubation period at room temperature, the cells were analyzed by flow cytometric analysis (CytoFLEX, Beckman). Data was analyzed using the FlowJo software (Ashland, OR, USA). The experiment was performed in triplicate.
Tumors from the subcutaneous tumor model were harvested for single-cell suspensions using a tumor dissociation kit (Miltenyi Biotec GmbH). The drainage lymph nodes (dLNs) were harvested through mechanical dissociation. Dissociated cells were filtered through a 4μm strainer and suspended in phosphate-buffered saline (PBS) supplemented with 1% FBS. The cells were stained with the following Abs according to the manufacturer’s instructions: CD45 (Cat# 103114), CD3 (Cat# 100204), CD4 (Cat# 100414), CD8a (Cat# 100752), CD25 (Cat# 101923), CD44 (Cat# 103031), CD62L (Cat# 161203), F4/80 (Cat# 123128), CD11b (Cat# 101245), MHC-II (Cat# 107606), CD206 (Cat# 141708), Gr-1 (Cat# 108423), CD11c (Cat# 117329), CD86 (Cat# 105014), PD-1 (Cat# 135206), or PD-L1 (Cat# 124312) (2.5μl, all from BioLegend) were diluted in FACS buffer (Biolegend). Various immune cells were separated using a gating strategy based on the expression of known lineage markers for lymphocytes (CD45+), total T cells (CD45+CD3+), CD4+ T cells (CD45+CD3+CD4+), CD8+ T cells (CD45+CD3+CD8+), Treg cells (CD45+CD3+CD4+CD25+), central memory T cells (CD3+CD4+CD44+CD62L+), effective memory T cells (CD3+CD4+CD44+CD62L−), macrophages (CD45+CD11b+F4/80+), M1 macrophages (CD45+CD11b+F4/80+/MHC-II+), DC cells (CD45+CD11b−CD11c+), MDSCs (CD45+CD11b+Gr-1+), and NK cells (CD45+CD3−CD49b+). Data was performed on the flow cytometers (Cytek NL-CLC 3000, Cytek) and analyzed using the FlowJo software (Ashland, OR, USA).
Tumors were collected and fixed in 4% formalin. Sections of paraffin-embedded tissues (4 μm) were deparaffinized in xylene and rehydrated in a graded series of alcohol concentrations. 3% H2O2 was used to block endogenous peroxidase activity, and the slides were incubated in Tris-EDTA buffer for antigen retrieval. Subsequently, the sections were incubated overnight at 4°C with the primary antibody against CD3 (Cat# ab16669), CD4 (Cat# ab183685), or CD8 (Cat# ab217344) (1:200, all from Abcam) overnight. Sections incubated with normal mouse or rabbit IgG instead of primary antibodies were used as the negative control. For IHC, the sections were incubated with the HRP-linked secondary Ab and the cell nucleus was counterstained using hematoxylin. For immunofluorescence staining, the sections were incubated with the fluorophore-conjugated secondary Abs and the cell nucleus was counterstained using 4′,6-diamidino-2-phenylindole (DAPI).
Mouse CT26 cells (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cell inoculation, mice were treated with tucidinostat (25 mg/kg, gavage, daily, n=3) or DMSO as vehicle control (DMSO, n=3). To investigate the intrinsic mechanisms of tucidinostat on tumor immune microenvironment, tumor tissue from CT26 tumor-bearing mice on day 10 post-treatment initiation was harvested using TRIzol (Invitrogen) according to manufacturer’s instructions (Novogene co., Ltd). Total RNA was used as input material for the RNA sample preparations. PCR products were purified (AMPure XP system) and library quality was assessed on the Agilent Bioanalyzer 2100 system. After cluster generation, the library preparations were sequenced on an Illumina Novaseq platform and 150-bp paired-end reads were generated. Differential expression analysis of two groups was performed using the DESeq2 R package (1.20.0). Gene Ontology (GO) enrichment analysis of differentially expressed genes was implemented by the cluster Profiler R package, in which gene length bias was corrected. For tumor environment analysis, published mRNA signatures for T cells and other cell clusters were analyzed [31, 32].
To assess the effect of different doses of tucidinostat on hepatic/renal toxicity and cytokine secretion, peripheral blood was collected from the inner canthus of the experimental mice. Mouse white blood cell (WBC) count, red blood cell (RBC) count, platelet (PLT) count, and lymphocyte count were detected by a fully automatic hematology analyzer (BC-2800 Vet, Mindray). Peripheral blood was collected and then centrifuged at 3000 rpm for 10 min to isolate the serum. The serum alanine transaminase (ALT), alanine transaminase (ALT), and urea nitrogen (BUN) (all from Anoric Biotechnology) were measured using ELISA kits. The serum IL-10 (Cat# 431417), IFN-γ (Cat# 430807), TNF-α (Cat# 430907) (all from Biolegend), and CCL5 (R&D, Cat# DY478) were measured using ELISA kits. The absorbance of each sample was measured on a microplate reader (Thermo Scientific) at 450 nm.
To further confirm the chemotactic effect of CCL5 on CD8+ T cells, naïve CD8+ T cells were purified from mouse spleen and activated with Dynabeads containing mouse T-activator CD3 (Biolegend, Cat# 100301)/CD28 (Biolegend, Cat# 102101) and recombinant mouse IL-2 (Biolegend, Cat# 714604). And then, the activated CD8+ T cells were seeded in the upper chambers of transwell plates (BD Biosciences) and allowed to migrate for 24 h towards the lower chamber containing medium with different concentrations of CCL5 protein (Pepro Tech, Cat# 250-07).
Bone marrow cells were isolated from femurs and tibias of C57/BL6 mice. 5 × 106 cells per well in 24-well plates were cultured in RPMI-1640 medium containing 10% heated-inactivated fetal bovine serum at 37°C in a 5% CO2 incubator. Bone marrow-derived macrophages (BMDMs) were differentiated in the presence of recombinant cytokine M-CSF (20ng/ml, Pepro Tech, Cat# 315-02). Every 2 days, 50% of the medium were replaced with fresh culture medium. After 10 days, we harvested adherent cells and used them for BMDM experiments.
To determine the effect of different doses of tucidinostat in human leukocytes, the peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood samples obtained from non-small cell lung cancer (NSCLC) patients and rested in RPMI-1640 medium containing 10% heated-inactivated fetal bovine serum at 37°C in a 5% CO2 incubator for 6 h. Then, 5 × 105 cells per well in 24-well plates were cultured with different doses of tucidinostat as indicated for 24 h. The cells were stained with the following Abs according to the manufacturer’s instructions: CD14 (Cat# 325604), CD11b (Cat# 393112), CD3 (Cat# 344804), CD4 (Cat# 317428), CD8a (Cat# 344722), CD69 (Cat# 310906), CD86 (Cat# 374208), or HLA-DR (Cat# 307630) (2.5μl, all from BioLegend) were diluted in FACS buffer (Biolegend). Various immune cells were separated using a gating strategy based on the expression of known lineage markers for total peripheral blood monocytes (CD14+CD11b+), active peripheral blood monocytes (CD14+CD11b+HLA-DR+/CD14+CD11b+CD86+), total T cells (CD3+), CD4+ T cells (CD3+CD4+), CD8+ T cells (CD3+CD8+), active CD4+ T cells (CD3+CD4+CD69+), and active CD8+ T cells (CD3+CD8+CD69+). Data was collected and analyzed with the flow cytometers (BD Accuri® C6, BD). The study was approved by the Institutional Review Board of the National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College (Permit Number, NCC2020C072).
All data analysis was performed using GraphPad Prism software (version 5.0, GraphPad Software, Inc.). For the comparison among treatment groups in the in vitro and in vivo study, one-way ANOVA was performed. Survival time was defined from the day of tumor cell inoculation until the mice expired naturally or were euthanized. Survival curves were drawn using the Kaplan-Meier method and compared with the log-rank test. P<0.05 was considered statistically significant. In the figures, symbols were used as *P<0.05, **P<0.01, and ***P<0.001.
The antitumor effect of different doses (2.5, 5, 7.5 μM) of tucidinostat was first evaluated in three different cell lines—4T1 breast cancer cells, LLC lung cancer cells, and CT26 colorectal cancer cells—in vitro and the data revealed that the cell proliferation levels were significantly suppressed and the cell apoptosis levels were increased in the higher-dose groups (Additional file 1: Fig. S1a-b). To determine the optimized dose of tucidinostat, the in vivo activity of tucidinostat was assessed in the CT26 tumor-bearing mice. The mice were gavaged with tucidinostat daily at three different doses (12.5, 25, and 75 mg/kg) after tumor cell inoculation (Fig. 1a). The administration of the lower dose (12.5 mg/kg) and middle dose (25 mg/kg) of tucidinostat induced sustained and modest decrease of tumor growth with tolerable toxicity, and the higher dose (75 mg/kg) of tucidinostat induced significantly greater tumor growth suppression but with intolerable toxicities, such as rapid body weight loss, leucopenia, and lymphopenia (Fig. 1b–e). The higher dose (75 mg/kg) of tucidinostat also elevated the serum alanine transaminase (ALT) level, which indicated the impaired liver function. To evaluate kidney function, blood urea nitrogen (BUN) levels were measured; less damage was recorded in these groups (Additional file 1: Fig. S1c). Immunological changes occurring in TME after tucidinostat treatment were further assessed. Flow cytometry data demonstrated a significant increase in the number of CD8+ T cells infiltrating tumors at 10 days post-administration in the tucidinostat (25 mg/kg) group, indicating that this dose exerted a robust immune priming effect (Fig. 1f). Moreover, the CD8+ T cells in the drainage lymph nodes (dLNs) also increased, although the difference was not statistically significant. Interestingly, the number of CD3+ and CD4+ T cells increased following tucidinostat (25 mg/kg) treatment, suggesting that the treatment induced effective antitumor immune responses (Fig. 1f). Furthermore, immunofluorescence staining of tumor sections demonstrated that the proportion of CD8+ T cells was higher in this group (Fig. 1g). Considering the antitumor efficacy and safety profile, the optimized dose (25 mg/kg) of tucidinostat was selected for further investigations.
Next, potential mechanisms underlying the changes in TME were explored through gene expression profiling. Bulk mRNA-seq data comparing tucidinostat-treated (25 mg/kg) and untreated tumors from the CT26 murine model indicated that tucidinostat could markedly alter the TME, as indicated by the significantly higher immune microenvironment scores. A heatmap of differential expressed genes revealed a consistency between immune-related gene signatures and the changes observed in the T cell populations (Fig. 2a, b). Furthermore, the functional annotation of gene clustering indicated altered expression levels of a considerable amount of cytokines following tucidinostat treatment (Additional file 2: Fig. S2a). It is known that the best predictors of immunotherapy response are the number and phenotype of tumor-infiltrating CD8+ T cells recruited at the tumor site by the locally secreted chemokines [33, 34]. A large body of evidence exists to show that increased expression of CD8+ T cell-attracting chemokines, such as the C-C motif chemokine ligand 5 (CCL5) and C-X-C motif chemokine ligand 9 and 10 (CXCL9 and CXCL10), correlates with decreased levels of cancer metastasis and improved clinical outcome in patients with cancer [35–37]. Recently, a mechanistic link between epigenetic modification and the secretion of such cytokines in TME has been described [38, 39]. Here, we hypothesized that the immune modulation of tucidinostat may occur, at least partially, through tumor-derived cytokine secretion. qPCR was performed for the treated murine 4T1, LLC, and CT26 cancer cell lines, and the data showed that the total expression levels of CCL5, CXCL9, and CXCL10 were significantly increased after tucidinostat treatment (Fig. 2c). More importantly, the optimized dose of tucidinostat treatment also improved the total and cell surface expressions of PD-L1 (Fig. 2c, Additional file 2: Fig. S2b). In general, tucidinostat at the optimized dose could elevate the expression of PD-L1 and T cell-attracting chemokines, such as CCL5, CXCL9, and CXCL10. Considering that the stress-activated NF-κB pathway controls cytokine expression in multiple cell types and the key role of CCL5 in attracting CD8+ T cells [40, 41], we further hypothesized that the antitumor immune response induced by tucidinostat may be mediated through CCL5 upregulation via the NF-κB pathway. As expected, NF-κB pathway was activated along with an increased expression of CCL5 after the administration of tucidinostat treatment in CT26 cancer cell lines, and this upregulation of CCL5 expression was abrogated upon pharmacological NF-κB inhibition using BAY11-7082 (Fig. 2d, e). Previous data suggested that high intratumoral expression of CD8+ T cell-attracting chemokine CCL5 is correlated with better prognosis in several types of cancers [42, 43]. Moreover, CD8A expression is significantly correlated with CD8+ T cell infiltration and surface proteins that are critical to transduce antigen recognition into immune cell responses [44]. Therefore, the prognostic roles of CCL5 and CD8A were evaluated using public datasets from The Cancer Genome Atlas (TCGA). The analysis revealed that CCL5 expression is positively correlated with CD8A expression in three cancer types (breast invasive carcinoma, lung adenocarcinoma, and colon adenocarcinoma) (Additional file 2: Fig. S2c). Furthermore, both CCL5High and CCL5High CD8AHigh are negatively correlated with the risk of death or recurrence in breast cancer (Fig. 2f, Additional file 2: Fig. S2d), which is consistent with previous data. Ex vivo chemotaxis assays have also shown that higher CCL5 protein concentrations could enhance CD8+ T cell transwell migration (Fig. 2g), suggesting that CCL5 is necessary for T cell infiltration. Overall, these findings demonstrated a strong intrinsic anticancer activity of tucidinostat, which was mediated by the enhancement of CD8+ T cell recruitment through CCL5 upregulation via NF-κB signaling pathway activation.
Next, we examined whether tucidinostat could induce an innate antitumor immune response that leads to effective tumor regression. Several immune-related gene expressions were observed in Raw 264.7 cells and bone marrow-derived macrophages (BMDMs) 24h after the administration of tucidinostat treatment. mRNA expression levels of the M1-like macrophage marker inducible nitric oxide synthase (iNOS) and CD86 were increased (Fig. 3a). Furthermore, flow cytometry analysis revealed dose-dependent elevations of M1-like macrophage surface marker, MHC-II, on both Raw.264.7 cells and BMDMs (Fig. 3b). The tumor-conditioned medium collected from LLC cells after tucidinostat treatment also promoted M1 polarization of Raw.264.7 cells and BMDMs as assessed by qPCR and flow cytometry (Fig. 3c, d). To further investigate the influence of tucidinostat on macrophage polarization in vivo, the immunological changes observed in tumors after administering treatment to CT26 tumor-bearing mice were assessed. Compared with other groups, a significant decrease in the proportion of tumor-associated macrophages among the total viable cells and an increase in the ratio of M1 macrophage were observed in the tucidinostat-treated (25 mg/kg) group (Fig. 3e). Total macrophages in dLNs also distinctly decreased, although the downward trend did not reach statistical difference (Fig. 3e).
Because a robust antitumor immune response was induced by daily administration of 25 mg/kg tucidinostat, the treatment efficacy of tucidinostat combined with ICIs was further evaluated in three murine solid tumor models (4T1, LLC, and CT26). Accordingly, tumor-bearing mice were treated with tucidinostat (25 mg/kg) monotherapy through gavage, aPD-L1 (200 μg, every 3 days) monotherapy by intraperitoneal injection, tucidinostat plus aPD-L1, or vehicle control (Fig. 4a, Additional file 3: Fig. S3a). Tumor growth was significantly inhibited in the group treated with combination regimens compared to those treated with either single tucidinostat or vehicle control (Fig. 4b–d, Additional file 3: Fig. S3b-c). Notably, the combination of tucidinostat and aPD-L1 resulted in the significantly improved survival compared to vehicle control (Fig. 4e). To investigate the effect of tucidinostat combined with aPD-L1 on TME, the changes observed in the immune cell populations upon treatment were analyzed. In subcutaneous CT26 tumors, the total amount of infiltrated CD45+ lymphocytes was increased after treatment with the combination therapy (Fig. 5a). Furthermore, it induced a significant increase in the proportion of tumor-infiltrating CD4+ and CD8+ T cells (Fig. 5a). In addition, immunohistochemical assays demonstrated an increase in the number of CD8+ T cells in tumor tissues collected from the combination therapy group (Additional file 4: Fig. S4a). Interestingly, CD8+ and CD4+ T cells from the combination therapy group exhibited a noticeable reduction in the expression of the exhaustion marker PD-1 (Additional file 4: Fig. S4b). Furthermore, the reductions in the proportion of tumor-infiltrating macrophages and increases in that of M1 macrophages were observed in the combination therapy group (Fig. 5a). The fluorescence-activated cell sorting (FACS) data of dLNs also confirmed that the antitumor activity and immune function were improved following the combination therapy. The percentages of CD4+ T cells significantly increased, whereas those of macrophages declined after treatment with tucidinostat plus aPD-L1 (Fig. 5b). Other immune cells such as myeloid-derived suppressor cells (MDSCs), dendritic cells (DCs), and natural killer (NK) cells were not significantly influenced in tumor tissues or dLNs (Fig. 5a, b, Additional file 4: Fig. S4c). Furthermore, tucidinostat-induced changes in serum cytokine levels were also examined in CT26 tumor-bearing mice. In accordance with the changes observed in intratumoral expression of CD8+ T cells, serum CCL5 level was markedly increased after tucidinostat treatment alone as well as by the combination of tucidinostat plus aPD-L1 (Additional file 5: Fig. S5a-b). In addition, serum interferon-γ (IFN-γ) was enhanced in the combination therapy group compared with the vehicle control group. However, no changes were observed in serum interleukin-10 (IL-10) or tumor necrosis factor-α (TNF-α) (Additional file 5: Fig. S5c). In general, these findings indicate that tucidinostat might significantly augment the antitumor immune response of aPD-L1 through CD8+ T cell infiltration and M1 macrophage polarization in solid tumor-bearing mice.
As tucidinostat plus PD-L1 blockade resulted in tumor regression and flow cytometry profiling revealed enhanced CD8+ T cell infiltration, we hypothesized that the observed antitumor responses were mediated through CD8+ T cell populations. To further verify the contribution of CD8+ T cells to the enhanced antitumor efficacy of aPD-L1, we developed an in vivo CT26 mouse tumor model and utilized it for examining tumor responses following vehicle control, tucidinostat monotherapy ± CD8+ T cell depletion, and tucidinostat plus aPD-L1 combination therapy ± CD8+ T cell depletion. Upon pretreatment with an anti-CD8 antibody (aCD8), the degree of tumor shrinkage induced by both monotherapy and combination therapy was significantly reduced (Fig. 6a–d), supporting that tucidinostat potentiated the effects of PD-L1 blockade in vivo through CD8+ T cell-induced antitumor immune response. In addition, macrophage depletion using clodronate liposomes also reversed the beneficial effects of combination treatment, suggesting that macrophages also contribute to tumor regression. However, macrophage depletion failed to reverse the antitumor response induced by tucidinostat alone (Fig. 6a–d). Therefore, macrophages might be necessary but not sufficient for the tumor suppression effect of tucidinostat.
Previous studies have reported that tucidinostat could alter antigen-presenting cell (APC) function by regulating inflammatory cytokine production in patients with immune thrombocytopenia [28]. We therefore explored whether tucidinostat could alter the antigen-presenting function of monocytes (CD14+CD11b+) among the peripheral blood mononuclear cells (PBMCs) of patients with non-small cell lung cancer (NSCLC). After 24-h treatment with tucidinostat, the surface expression of CD86 and HLA-DR on the monocyte fraction were significantly upregulated (Fig. 7a). These findings demonstrated the modulatory effects of tucidinostat on human monocytes and suggested that tucidinostat promotes these phenotypic changes, conferring enhanced antigen presentation and costimulatory capabilities. However, tucidinostat did not appear to activate T cells directly as no upregulation of the CD69 expression was observed on the conventional CD8+ or CD4+ T cells among PBMCs of patients with NSCLC (Fig. 7b).
HDACs consist of a large family of proteins categorized into five groups—class I (HDAC 1, 2, 3, 8), class IIa (HDAC 4, 5, 7, 9), class IIb (HDAC 6, 10), class III (Sirtuins), and class IV (HDAC 11) [18]. Aberrant HDAC expression occurs in most solid tumors and hematological cancers. The dysregulation of histone acetylation can lead to aberrant gene expression, which can activate oncogenes, inactivate tumor suppressors, inhibit programmed cell death, and mediate immune evasion, ultimately resulting in tumor progression [4, 16, 18, 21]. To date, four HDAC inhibitors, vorinostat, romidepsin, panobinostat, and belinostat, have been approved by the US Food and Drug Administration and are used for treating hematological cancers [45]. At present, most of the HDAC inhibitors in the clinic are pan-HDAC inhibitors. This broad-spectrum activity may also produce undesirable side effects. Therefore, the development of selective HDAC inhibitors may be useful for better understanding the critical events related to their therapeutic effects and for providing a rational basis to exploit synergistic interactions with other clinically effective agents. Tucidinostat, a selective HDAC inhibitor having specificity for HDAC1, HDAC2, HDAC3, and HDAC10 subtypes, has been approved for the treatment of relapsed or refractory peripheral T cell lymphoma and is under clinical development globally for various other neoplastic and non-neoplastic diseases. Recently, tucidinostat has been approved by the National Medical Products Administration and is used for patients with advanced, hormone receptor-positive, HER2-negative breast cancer that progressed after previous endocrine therapy [30]. It might be the first HDAC inhibitor used for treating solid tumors in clinical settings. However, due to the grade 3 or 4 toxicities caused by it, tucidinostat might not effectively modulate TME. Therefore, the appropriate dose of tucidinostat should be defined for treating solid tumors. The mice were gavaged with tucidinostat daily at different dosages of tucidinostat after tumor cell inoculation. We demonstrated that tucidinostat at 25 mg/kg daily dosage could promote a rapid and sustained antitumor immune response through the preclinical mouse tumor model. Conversely, lower or opposite immunosuppressive effects were observed with the administration of lower or higher dosages. To date, few studies have reported regarding tucidinostat treatment optimization strategies for turning the TME of solid tumors from cold to hot. In this work, we mainly focused on exploring such optimization strategies and then delineating the possible underlying functional mechanisms both in vitro and in vivo. Next, we demonstrated that the optimized dose tucidinostat could promote the secretion of several CD8+ T cell-attracting chemokines, especially that of CCL5. Although the role of CCL5/CCR5 axis in carcinogenesis is controversial [46], increasing evidence has demonstrated that constitutive CCL5 expression enables tumor immune recognition and enhances immunotherapy response via increased infiltration of CD8+ T cells into tumors [47–50]. Besides, high intratumoral expression of CCL5 is correlated with better prognosis and strongly correlated with intratumoral CD8A expression across multiple cancer types according to our analysis of TCGA datasets. It was reported that DNA methylation negatively regulates CCL5 expression in lung and colon cancers [51]. Our findings indicated an additional epigenetic mechanism wherein the selective histone acetylation inhibitor tucidinostat could also induce CCL5 expression in tumors through the NF-κB signaling pathway, leading to CD8+ T cell infiltration into tumors. Because tucidinostat elevated the secretion of CCL5 and other T cell-attracting chemokines in TME, we sought to demonstrate that the optimized dose of tucidinostat can promote a rapid and sustained antitumor immune response when used in combination with aPD-L1 using multiple preclinical mouse tumor models. The response was dependent on enhanced CD8+ T cell infiltration in TME and was abrogated upon CD8+ T cell depletion. Thus, tucidinostat with an optimized dose could alter TME and promote the migration and infiltration of CD8+ T cells into tumors, partially by increasing the activity of chemokine CCL5 via NF-κB signaling. CCL5 might indeed be secreted by some other inflammatory cells in TME following tucidinostat treatment, which is required to be testified in further studies. Furthermore, we demonstrated that tucidinostat could also modulate M1 polarization of macrophages in solid tumors. It has been reported that the class IIa HDAC inhibitor TMP195 could alter the tumor microenvironment and reduce tumor burden and pulmonary metastases by modulating macrophage phenotype in a macrophage-dependent autochthonous mouse model of breast cancer. Furthermore, TMP195 induced the recruitment and differentiation of highly phagocytic and stimulatory macrophages within tumors [52]. In this study, it was seen that tucidinostat could directly promote M1 polarization of macrophages; moreover, the medium from tumor cells treated with tucidinostat could also induce M1 polarization of macrophages, suggesting that factors secreted by tumor cells in response to tucidinostat treatment could repolarize tumor-associated macrophages. The M1 macrophages, which have a role in mediating the destruction of tumor cells and facilitating the recruitment of Th1 cells, were also found to be highly sensitive to tucidinostat plus aPD-L1 treatment. Moreover, the antitumor effect was mostly abolished by macrophage depletion using clodronate liposomes. Therefore, tucidinostat could significantly promote M1 polarization of macrophages and increase the antitumor efficacy of aPD-L1 in vivo. In addition, significant upregulation of MHC class II molecules, CD86 and HLA-DR, was observed with phenotypic changes associated with increased APC priming. This observation was consistent with the known positive correlation between MHC class II expression and PD-1/PD-L1 inhibitor response [53, 54]. In fact, several studies have identified MHC class II expression as a potential biomarker for PD-1/PD-L1 therapeutic response. Therefore, tucidinostat may sensitize tumors against aPD-1/aPD-L1 blockade, at least partly, by modulating the expression of MHC class II molecules. However, tucidinostat failed to directly promote the transient activation of peripheral T lymphocytes, which is in agreement with the findings we obtained using mouse models wherein tucidinostat altered T cell function by upregulating T cell-attracting chemokines, such as CCL5. As we know, the success of pan-essential inhibitors suggests that targeting pan-essential genes will remain an important strategy for solid tumor therapeutics development. However, the broad requirement for HDAC activity in normal human tissues along with inhibitor polypharmacology made it likely that side effects should be limiting [55]. In this study, we chose the selective HDAC inhibitor tucidinostat which has been successfully used in clinical and found that the optimized dose of tucidinostat was seen to alter TME by promoting the infiltration of T cells via the activation of the NF-κB pathway and the subsequent release of immune-related cytokines such as CCL5. Moreover, the optimized dose of tucidinostat modulated M1 polarization of macrophages and dramatically potentiated the antitumor efficacy of PD-L1 blockade in solid tumors. Therefore, developing therapeutics that target pan-essential genes, such as HDACs, requires careful target prioritization and validation, dosing optimization, and combination strategies, which need in-depth research in the future.
Collectively, our study demonstrated that the combined use of tucidinostat at an optimized dose and PD-L1 blockade may work synergistically to reduce tumor burden by enhancing the immune function. The finding provides a strong rationale for conducting clinical trials to investigate this combination therapy for overcoming ICI treatment resistance and achieving better clinical outcomes for patients with solid tumors.
Additional file 1. The effect of tucidinostat on cell proliferation, apoptosis and hepatonephric function. a Comparison of cell proliferation of 4T1, LLC, and CT26 cells treated with different doses (2.5, 5, 7.5 μM) of tucidinostat for 24 h by CCK-8 assay. b Comparison of cell apoptosis of 4T1, LLC, and CT26 cells treated with different doses of tucidinostat (2.5, 5, 7.5 μM) for 6 h by Annexin V-FITC/PI assay. c Mouse CT26 cells (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cells inoculation, mice were treated with different doses (12.5, 25, 75 mg/kg, gavage, daily, n=5) of tucidinostat. Analysis on hepatonephric function on day 10 post treatment initiation. The serum ALT, AST, and BUN from peripheral blood were measured using ELISA kits. The error bars indicate mean ± SEM. *P<0.05, **P<0.01, ***P<0.001 by one-way ANOVA. ns: not significant. CON: control group; Tuc: tucidinostat; ALT: alanine transaminase; AST: aspartate aminotransferase; BUN: blood urea nitrogen.Additional file 2. The function of tucidinostat on tumor immunity. a Functional annotation clustering of genes regulated in tumor from CT26 tumor-bearing mice on day 10 post treatment with tucidinostat (25 mg/kg, gavage, daily, n=3) or DMSO as vehicle control (DMSO, n=3). b Representative cytograms (left) or summary histograms (right) for the cell surface PD-L1 expression in CT26 cells following different doses (2.5, 5, 7.5 μM) of vorinostat, tucidinostat, and TMP-195 treatment for 24h. c Scatter plots showing the range of associations (r) with 95% CI and proportionality of expression levels for CD8A and CCL5 in three solid tumor types (BRCA, LUAD, and COAD) using TCGA database. d Kaplan-Meier survival curves for overall survival in three solid tumor types as stratified by CCL5 expression status using TCGA database. TCGA: the genome cancer atlas; BRCA: breast invasive carcinoma; LUAD: lung adenocarcinoma; COAD: colon adenocarcinoma; CON: control group; Vor: vorinostat; Tuc: tucidinostat; TMP: TMP-195.Additional file 3. Tucidinostat plus checkpoint blockade induces improved therapeutic efficacy. a Schema of the experiment. For bioluminescence imaging (BLI) in vivo, mouse 4T1-luc (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cells inoculation, mice were treated with vehicle (DMSO, n=7), tucidinostat (25 mg/kg, gavage, daily, n=7), aPD-L1 (200 μg, i.p. injection, once every 3 days, n=7), or combination (n=7). Tumors volume were measured with calipers every three days. b Left: Comparison of tumor size from 4T1-Luc tumor-bearing mice on day 21 post treatment initiation. Middle: Luciferase imaging of living mice were measured using the Caliper IVIS Lumina III Live Imaging System. Right: Quantitative analysis of fluorescence intensity of the tumor from 4T1-Luc tumor-bearing mice on day 21 post treatment initiation. c HE staining of tumor from 4T1-Luc tumor-bearing mice on day 21 post treatment initiation. The error bars indicate mean ± SEM. *P<0.05, **P<0.01, ***P<0.001 by one-way ANOVA. ns: not significant. CON: control group; Vor: vorinostat; Tuc: tucidinostat; TMP: TMP-195; aPD-L1: anti-programmed cell death ligand 1 antibody; BLI: bioluminescence imaging.Additional file 4. Tucidinostat plus checkpoint blockade induces significant antitumor immunity. a Schema of the experiment. Mouse CT26 cells (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cells inoculation, mice were treated with vehicle (DMSO, n=7), tucidinostat (25 mg/kg, gavage, daily, n=7), aPD-L1 (200 μg, i.p. injection, once every 3 days, n=7), or combination (n=7). The proportion of intratumoral CD8+ T cells by IHC from CT26 tumor-bearing mice on day 21 post treatment initiation. b Flow cytometric quantification of PD-1 in CD4+ T cells (CD45+CD3+CD4+) and CD8+ T cells (CD45+CD3+CD8+) in tumor parenchyma and tumor drainage lymph nodes from CT26 tumor-bearing mice on day 21 post treatment initiation. c Flow cytometric quantification of MDSCs (CD45+CD11b+Gr-1+), active DC cells (CD45+CD11b-CD11c+CD86+), Effective Memory T cells (CD3+CD4+CD44+CD62L-), and NK cells (CD45+CD3-CD49b+) in tumor parenchyma and tumor drainage lymph nodes from CT26 tumor-bearing mice on day 21 post treatment initiation. The error bars indicate mean ± SEM. *P<0.05, **P<0.01, ***P<0.001 by one-way ANOVA. ns: not significant. CON: control group; Tuc: tucidinostat; aPD-L1: anti-programmed cell death ligand 1 antibody.Additional file 5. Tucidinostat plus checkpoint blockade induces CCL5 secretion. a Mouse CT26 cells (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cells inoculation, mice were treated with different doses (12.5, 25, 75 mg/kg, gavage, daily, n=5) of tucidinostat. The expression of CCL5 in peripheral blood serum from CT26 tumor-bearing mice on day 10 post treatment initiation. Mouse CT26 cells (5 × 105 cells) were engrafted into the flank of BALB/c mice. When established tumors were palpable 7 days after tumor cells inoculation, mice were treated with vehicle (DMSO, n=7), tucidinostat (25 mg/kg, gavage, daily, n=7), aPD-L1 (200 μg, i.p. injection, once every 3 days, n=7), or combination (n=7). The expression of CCL5 b, IFN-γ, TNF-α, and IL-10 c in peripheral blood serum from CT26 tumor-bearing mice on day 21 post treatment initiation. The error bars indicate mean ± SEM. *P<0.05, **P<0.01, ***P<0.001 by one-way ANOVA. ns: not significant. CON: control group; Tuc: tucidinostat; aPD-L1: anti-programmed cell death ligand 1 antibody.Additional file 6. The original blots. |
|
PMC9648055 | Qin Zhou,Shanshan Liu,Yuying Kou,Panpan Yang,Hongrui Liu,Tomoka Hasegawa,Rongjian Su,Guoxiong Zhu,Minqi Li | ATP Promotes Oral Squamous Cell Carcinoma Cell Invasion and Migration by Activating the PI3K/AKT Pathway via the P2Y2-Src-EGFR Axis | 26-10-2022 | Oral cancer is one of the most common malignancies of the head and neck, and approximately 90% of oral cancers are oral squamous cell carcinomas (OSCCs). The purinergic P2Y2 receptor is upregulated in breast cancer, pancreatic cancer, colorectal cancer, and liver cancer, but its role in OSCC is still unclear. Here, we examined the effects of P2Y2 on the invasion and migration of oral cancer cells (SCC15 and CAL27). The BALB/c mouse model was used to observe the involvement of P2Y2 with tumors in vivo. P2Y2, Src, and EGFR are highly expressed in OSCC tissues and cell lines. Stimulation with ATP significantly enhanced cell invasion and migration in oral cancer cells, and enhanced the activity of Src and EGFR protein kinases, which is mediated by the PI3K/AKT signaling pathway. P2Y2 knockdown attenuated the above ATP-driven events in vitro and in vivo. The PI3K/AKT signaling pathway was blocked by Src or EGFR inhibitor. Extracellular ATP activates the PI3K/AKT pathway through the P2Y2-Src-EGFR axis to promote OSCC invasion and migration, and thus, P2Y2 may be a potential novel target for antimetastasis therapy. | ATP Promotes Oral Squamous Cell Carcinoma Cell Invasion and Migration by Activating the PI3K/AKT Pathway via the P2Y2-Src-EGFR Axis
Oral cancer is one of the most common malignancies of the head and neck, and approximately 90% of oral cancers are oral squamous cell carcinomas (OSCCs). The purinergic P2Y2 receptor is upregulated in breast cancer, pancreatic cancer, colorectal cancer, and liver cancer, but its role in OSCC is still unclear. Here, we examined the effects of P2Y2 on the invasion and migration of oral cancer cells (SCC15 and CAL27). The BALB/c mouse model was used to observe the involvement of P2Y2 with tumors in vivo. P2Y2, Src, and EGFR are highly expressed in OSCC tissues and cell lines. Stimulation with ATP significantly enhanced cell invasion and migration in oral cancer cells, and enhanced the activity of Src and EGFR protein kinases, which is mediated by the PI3K/AKT signaling pathway. P2Y2 knockdown attenuated the above ATP-driven events in vitro and in vivo. The PI3K/AKT signaling pathway was blocked by Src or EGFR inhibitor. Extracellular ATP activates the PI3K/AKT pathway through the P2Y2-Src-EGFR axis to promote OSCC invasion and migration, and thus, P2Y2 may be a potential novel target for antimetastasis therapy.
Oral cancer is one of the most common malignancies of the head and neck. Approximately 90% of oral cancers are oral squamous cell carcinomas (OSCCs). According to the World Health Organization (WHO), more than 260,000 people are newly diagnosed with oral cancer every year, and the incidence in people aged over 65 years accounts for more than 50% of the total population. Despite recent advances in treatment, the rich blood supply and complex anatomical structure of the oral and maxillofacial region are conducive to recurrence and distant migration in approximately one-third of patients treated with conventional surgery or radiotherapy. Tumor invasion and metastasis are still the main causes of death in OSCC patients, and therefore, new antitumor methods are required for more effective clinical treatment of OSCC. Adenosine 5′-O-triphosphate (ATP) has long been considered as the body’s most direct source of energy. In 1972, Geoff Burnstock put forward the “purinergic hypothesis” of neurotransmission, and the concept of ATP as an extracellular signaling molecule was formalized as a scientific hypothesis. Studies have shown that ATP (P2 receptors) is an important transmitter of various biological effects mediated by purinergic receptors, including cell proliferation, differentiation, and death. ATP may be crucial in promoting or preventing malignant metastasis. Under normal conditions, the extracellular ATP concentration (mmol/L) is much lower than the intracellular concentration (3–5 mmol/L), and it remains balanced. In a tumor microenvironment, the ATP concentration (about 100 μmol/L) is higher than that in the normal extracellular environment. ATP invasive transfer activity was first reported in prostate cancer. However, the pathogenesis of OSCC is still unclear. Purinergic receptors are divided into P1 and P2 receptors; the natural ligand of the P1 receptor is adenosine. P2 receptors are divided further into the following two categories: P2X and P2Y receptors. The P2X receptor is a ligand-gated ion channel receptor, with seven currently known subtypes (P2X1-P2X7), activated by extracellular ATP to release cation flow. The P2Y receptor is a G-protein-coupled receptor (GPCR) that plays an important role in a variety of signaling pathways. Currently, eight functional mammalian P2Y receptors (P2Y1, P2Y2, P2Y4, P2Y6, P2Y11, P2Y12, P2Y13, and P2Y14) have been cloned and identified as GPCRs. P2Y2 is a functional receptor. Its first ligands are ATP and uridine triphosphate (UTP). P2Y2 can activate a variety of signaling pathways, with the typical path being a Gqα signaling path. Thus, when it is coupled with IP3 to promote the release of Ca2+ from the endoplasmic reticulum calcium store, it increases the intracellular Ca2+ concentration and activates protein kinase C(PKC). P2Y2 acts at the c-terminus of cells, and it activates the mitogen-activated protein kinase (MAPK) pathway by activating nonreceptor tyrosine protein kinase (Src). P2Y2 activates the matrix metalloproteinases ADAM10 and ADAM17, and the catalytic film binds to the growth factor to activate the epidermal growth factor receptor (EGFR). Although P2Y2 is upregulated in breast cancer, pancreatic cancer, colorectal cancer, and liver cancer, and is activated in cell proliferation, invasion, and migration, the role of P2Y2 in OSCC is still unclear and requires further research. In this study, we used CAL27 and SCC15 oral squamous cell lines to explore the reaction of P2Y2 to cellular and associated mechanisms of extracellular nucleotide induction.
P2Y2 was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). p-AKT, AKT, p-PI3K, PI3K, Src, p-Src, EGFR (D38B1), and phospho-EGFR Y1068 (D7A5) antibodies were purchased from Abcam (MA). Anti-GAPDH was purchased from Protein Tech (Wuhan, China). ATP was purchased from GLPBIO (Shanghai, China), and siRNA was purchased from RIBOBIO (Guangzhou, China). AG1478 and Dasatinib were purchased from MCE (Shanghai, China).
OSCC tissues were collected from patients (n = 6) who underwent radical surgery between January 2019 and January 2020 at Shandong University (Jinan, China) with informed consent obtained concerning the use of surgically resected specimens for research purposes. All of the patients agreed and signed the informed consent. All human tissue and sample experiments were approved by the Ethics Committee of the School of Stomatology, Shandong University (ref Med. No. 20210802; 10 August 2021). The experiments conformed to the guidelines set by the Declaration of Helsinki. The patients did not receive any form of adjuvant therapy before surgery.
Human OSCC cell lines (CAL27 and SCC15) were obtained from the Shanghai Cell Bank of the Chinese Academy of Sciences (Shanghai, China). The cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM-F12, Hyclone) containing 10% fetal bovine serum (Gibco, Grand Island, NY) and 1% penicillin-streptomycin, in a 37 °C, 95% humidified air and 5% CO2.
CAL27 and SCC15 cells were seeded in a 96-well culture plate at a density of 1 × 103 cells/well. CAL27 and SCC15 cells were treated with ATP, and after 24, 48, and 72 h of incubation, cell viability was determined by the cell counting kit assay (CCK8, Soleibao, Beijing, China). Subsequently, an enzyme-labeling instrument (iMark, Bio-Rad Laboratories, Inc.) was used to detect the absorbance to determine cell viability.
The sequence of a single small interfering RNA (siRNA)1 was CCCGTGCTCTACTTTGTCA; single siRNA2 was GTAGCGAGAACACTAAGGA, all siRNA single strands are synthesized in vitro, at Guangzhou Borui Co, Ltd. Division (Guangzhou, China). An appropriate number of cells were seeded onto six-well cell culture plates 24 h before transfection, allowing the cell density to reach 30–50%. A transfection complex solution was prepared in accordance with the manufacturer’s protocol. Subsequently, 50 nM siRNAs were mixed with the siRNA Transfection Reagent. CAL27 and SCC15 cells were then transfected for 48 h. Subsequently, quantitative real-time polymerase chain reaction (qPCR) and western blotting were used to confirm the transfection of cell lines, and the siP2Y2 with the highest transfection efficiency was selected. Transfected cells as the experimental group were treated with 100 μM ATP for 24 h, and the expression of related genes and proteins was detected.
The migration ability of CAL27 and SCC15 cells was tested via scratch assay. The two types of cells were inoculated into six-well plates at a density of 5 × 105 cells/well, adhere to 80% density, and then replace the serum-free α-DMEM-F12 cell culture medium for culture. Subsequently, the bottom of the plate was scraped vertically with the tip of a 200 μL liquid pipette and washed with PBS. Then, the cells were treated with 100 μM ATP for 24 and 48 h in serum-free medium. Finally, take pictures under an inverted microscope (BX53; Olympus, Japan) at ×100 magnification at 0, 24, and 48 h. Image-Pro Plus 6.0 software (Media Controlnetics, Inc., Rockville, Maryland) was used to calculate the width of the healing area in the cell monolayer Learn analysis.
The effect of P2Y2 on the invasion ability of CAL27 and SCC15 cells was evaluated by the transwell chamber test. A small chamber with 8 μM pore size was placed into a 24-well plate. The upper chamber was filled with 60–80 μl Matrigel (BD, Franklin Lakes, NJ), inoculated with 3.5 × 103 cells in 200 ml of culture medium in the upper chamber, experimental group cells were treated with 100 μM ATP, the lower chamber was filled with 750 mL of α-DMEM containing 10% serum after culturing the F12 medium at 37 °C for 24 h, and then the cells were removed. A cotton swab was used to gently remove the cells in the upper chamber. The cells were fixed with 4% methanol, stained with 0.1% crystal violet, photographed under an optical microscope, counted, and statistically analyzed.
CAL27 and SCC15 cells were seeded in a six-well plate at a density of 400 cells per well. The cells were treated with 100 μM ATP cultured with α-DMEM-F12 for 14 days, and when they grew into a colony of 50 cells, the cells were washed with PBS and fixed with 4% methanol. Then, the cells were stained with 0.1% crystal violet and scanned. The number of colonies greater than 50 cells was counted, and statistical analysis was performed.
CAL27 and SCC15 cell RNA was extracted from the cells using Trizol reagent (AG21102, Accurate Biotechnology Co., Ltd., China). cDNA was synthesized using the Evo M-MLV RT Reverse Transcription kit II (AG11711, Accurate Biotechnology). QPCR was performed using the SYBR Green Pro Taq HS premixed qPCR kit (AG11701, Accurate Biotechnology) in an RT fluorescence quantitative PCR system (Light Cycler 96 SW 1.1, Roche Ltd, Switzerland). The parameters required for denaturation, annealing, and extension were as follows: 95 °C for 30 s, 45 cycles at 95 °C for 5 s, and 60 °C for 20 s. The primer sequences are shown in Table 1. All data were normalized to GAPDH expression. Quantification of the qRT-PCR results was performed by the 2–ΔΔCT method.
CAL27 and SCC15 cells were washed three times with precooled PBS, RIPA lysate was added to lyse the cells and then collected, and the protein concentration was detected with the BCA protein detection kit. The same amount of total protein (10 μg) was separated by 10% sodium salt-polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred to a poly(vinylidene fluoride) (PVDF) membrane. After being blocked with 5% bovine serum albumin (BSA)/tris-buffered solution with Tween (TBST) for 1 h, the PVDF membrane was incubated with the P2Y2 antibody (concentration 1:2000), Src/p-Src antibody (concentration 1:2000), EGFR/p-EGFR antibody (concentration 1:2000), PI3K/p-PI3K antibody (concentration 1:500), AKT/p-AKT antibody (concentration 1:500), and GAPDH (concentration 1:10 000), at 4 °C overnight. Subsequently, the proteins were washed three times in TBST and then incubated in horseradish peroxidase-conjugated goat anti-rabbit IgG (1:2000) for 1 h. After washing in TBST, immune reaction zones were determined with the ECL detection system and then captured by the gel imaging system (Amersham Imager 600; General Electric Company).
Athymic nude BALB/c female mice (aged: 4 weeks, n = 10) were purchased from Jinan Peng Yue Laboratory Animal Breeding Co. Ltd. They were housed in a specific pathogen-free environment under the condition of a 12-h light/12-h dark cycle as well as free access to food and water. The mice were randomly divided into two groups (n = 5), and 2 × 106 CAL27 cells were subcutaneously injected into the back of the right upper limb of each mouse. First, 2 × 106 CAL27 cells and siRNA CAL27 cells were subcutaneously injected into the back of the right upper limb of each mouse. Tumor size was detected every 3 days using a slide caliper, and the tumor volume was calculated using the following formula: A × B2/2, where A is the length of the tumor and B is the width. After 30 days, the mice were euthanized and the tumors were isolated, weighed, photographed, and fixed immediately with 4% paraformaldehyde for subsequent analysis. The animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of Shandong University. Animal study and euthanasia were carried out following the recommendations of the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Department of School and Hospital of Stomatology, Shandong University (ref Med. No. 20210803; 10 August 2021).
GraphPad Prism 6.0 software (GraphPad Software, Inc.) was used for statistical analysis. Paired Student’s t-test was used for comparison between the two groups, and the significance level was adjusted according to the number of tests in multiple comparisons. The cells of the experimental group and control group were tested by independent sample t-test. The differences between three or more groups were tested by one-way analysis of variance (ANOVA). All experiments are repeated at least three times unless otherwise stated. All results are expressed as mean values ± standard deviation. p < 0.05 was considered statistically significant.
To investigate the expression and role of P2Y2 in OSCC, we collected three surgical specimens of OSCC. Western blotting and qRT-PCR showed that P2Y2 expression was significantly higher in tumor tissue than in adjacent noncancerous tissues (Figure 1A–C). P2Y2 was significantly expressed in all OSCC cells (Figure 1E–G). Furthermore, Src and EGFR were also highly expressed in OSCC (Figure 1A,B,D). To further study the role of P2Y2 receptors in human cancer and explore their clinical significance, we used TCAG database analysis to display that Src and EGFR were highly expressed in head and neck squamous cell carcinoma (HNSCC) (Figure 1H) and that the expression of P2Y2 was highly correlated with that of Src and EGFR (Figure 1I).
To determine whether OSCC cell lines express functional P2Y2, we used Cell Counting Kit-8 (CCK-8) to ascertain the optimal stimulating concentration of ATP. With increasing ATP concentration, cellular activity was also increased, but at a concentration of 200 μM, the cellular activity was significantly decreased (Figure 2A). Western blotting and qRT-PCR were used to verify the CCK8 results. According to these results in combination, the optimal ATP concentration for stimulation was determined to be 100 μM (Figure 2B–D). The Western blot analysis showed that ATP stimulated the phosphorylation of Src and EGFR in a dose- and time-dependent manner and reached peak activation at 100 μM ATP within 10–20 min (Figure 2E–H).
The OSCC cell lines were treated with the extracellular nucleotide ATP. In cell colony analysis, CAL27 and SCC15 cells were treated with 100 μM ATP for 14 days. The number of cell colonies was increased in comparison to that in the control group (Figure 3A). The scratch test and transwell test showed that CAL27 and SCC15 cell migration and invasion were significantly promoted compared to those in the control group (Figure 3B–D). To investigate whether extracellular ATP can enhance the invasion ability of OSCC cells, we used qRT-PCR assay to analyze the gene expression. In CAL27 and SCC15 cells stimulated with 100 μM ATP for 24 h, the expression of Zinc finger protein SNAI1 (Snail), matrix metalloproteinase (MMP)2, and MMP9 genes increased. At the same time, expression of E-cadherin and Vimentin decreased, and these results suggested that ATP induced the invasion and migration of OSCC cells and that P2Y2 receptor activation may play a major role in mediating the expression of genes related to invasion and migration (Figure 3E,F). To investigate whether the PI3K/AKT signaling pathway is involved in the regulation of ATP-induced invasion and migration of OSCC cells, we treated CAL27 and SCC15 cells with 100 μM ATP for 24 h. Compared with the control group, the expression of p-PI3K and p-AKT was significantly increased after 24 h of ATP treatment (Figure 3G).
To demonstrate the association between P2Y2 and the invasion and migration ability of OSCC cells, P2Y2 was knocked down by siRNA. CAL27 and SCC15 cells were transfected with 50 nM siRNA for 48 h, and the optimal siRNA was determined using real-time PCR and western blotting. The most effective siRNA of P2Y2-siRNA1 was selected and used in the experiment (Figure 4A,B). In the cell colony analysis, the cells were treated with ATP (100 μM) for 14 days, and after that, the number of cell colonies was lower compared with the P2Y2 knockdown group (Figure 4C). In the in vitro invasion test, P2Y2-siRNA cells were treated with 100 μM ATP for 24 h. The number of CAL27 and SCC15 cells in the P2Y2 knockdown group was significantly lower than that of the control group, indicating that P2Y2 may promote the invasion of CAL27 and SCC15 cells (Figure 4D). The wound-healing assay showed the same result in that after treatment with ATP (100 μM) for 24 h, the wound gap between CAL27 and SCC15 cells was significantly reduced compared with that in the control group, suggesting that P2Y2 is involved in ATP-promoted migration and invasion of OSCC cells (Figure 4E).
To study the downstream effect of invasion driven by ATP-P2Y2, we first focused on the genes related to invasion and migration. After P2Y2 was transfected, OSCC cells were stimulated with 100 μM ATP for 24 h. PCR detection showed that the ATP-mediated expression of Snail, MMP2, and MMP9 genes was inhibited, and the expression of E-cadherin and Vimentin was activated (Figure 5A–E). Nude mice were subcutaneously implanted with CAL27 cells from the NC and siRNA groups to study the anticancer effect of P2Y2 in vivo. The average tumor volume in the siRNA group was significantly smaller than that in the control group. The average tumor weight in the P2Y2-siRNA group was also significantly smaller than that in the control group (Figure 5F,G). These data support the hypothesis that P2Y2 receptors play an important role in ATP-mediated invasion and migration in vivo and in vitro.
To explore how P2Y2 regulates the Src and EGFR signaling pathways through extracellular ATP, we performed P2Y2 silencing. After P2Y2 silencing, CAL27 and SCC15 cells were stimulated with 100 μM ATP for 20 min. Western blotting showed that the phosphorylation of Src and EGFR was significantly reduced (Figure 6A,B). Following the application of Dasatinib and AG1478 inhibitors for 1 h, CAL27 and SCC15 cells were stimulated with 100 μM ATP for 20 min, and the phosphorylation of EGFR was found to be significantly reduced (Figure 6C,D).
To further investigate whether P2Y2 activates the PI3K/AKT signaling pathway through the Src-EGFR axis, CAL27 and SCC15 cells that have knocked down P2Y2 were treated with ATP (100 μM) for 24 h. Compared with the control group, the expression of p-PI3K and p-AKT was significantly decreased (Figure 7A,B). The cells were further pretreated with Dasatinib and AG1478 for 1 h and then stimulated with ATP. The expression of p-PI3K and p-AKT was decreased (Figure 7C,D). In the cell colony analysis, the cells were treated with PI3K/AKT inhibitors for 14 days, after which the number of cell colonies was lower compared with the ATP (100 μM) group (Figure 7E). In the in vitro invasion test, the cells were treated with PI3K/AKT inhibitors for 24 h. The number of CAL27 and SCC15 cells in the PI3K/AKT inhibitors group was significantly lower than that of the control group, indicating that PI3K/AKT may promote the invasion of CAL27 and SCC15 cells (Figure 7F). The wound-healing assay showed the same result; after treatment with PI3K/AKT inhibitors for 24 and 48 h, the wound gap between CAL27 and SCC15 cells was significantly reduced compared with that in the control group, suggesting that PI3K/AKT is involved in ATP-promoted migration and invasion of OSCC cells (Figure 7G,H).
The tumor microenvironment (TME) is a dynamic environment, and its biochemistry and cellular composition play a crucial role in the regulation of tumor cell metabolism, proliferation, and motility. The transdifferentiation of epithelial cells into motile mesenchymal cells, a process known since the 1980s as epithelial-mesenchymal transition (EMT), was first observed by Elizabeth Hay, who described epithelial to mesenchymal phenotype changes in the primitive streak of chick embryos. EMT is an indispensable part of the developmental process, and its underlying process is reactivated during wound healing, fibrosis, and cancer progression. During the development of cancer, the cytoplasmic damage caused by inflammation and hypoxia and the tissue destruction caused by tumor invasion result in an increased concentration of extracellular ATP, which plays a key role as an extracellular messenger. In 1980, P2Y2 suggested that specific plasma membrane receptors for extracellular ATP were expressed by inflammatory and cancer cells; P2Y2 is the first selected ligand of ATP. The role of ATP in the TME is multifaceted, as it regulates the permeability of cell connections by mobilizing intracellular Ca2+ storage, leading to tumor cell invasion and metastasis. In cancer, EMT is highly deregulated, and EMT-transcription factors exert important roles in all cancer stages, including initiation, primary tumor growth, invasion, dissemination, metastasis, colonization, and therapy resistance as well. Research findings in different models have demonstrated the participation of the P2Y2 receptor in inducing migration or the EMT process. In ovarian cancer, ATP induces EMT of ovarian cancer cells through the P2Y2 receptor-dependent activity of EGFR. In gastric cancer, purinergic P2Y2 and P2X4 receptors are involved in changes in the expression of EMT and related genes in gastric cancer cell lines. It is generally understood that ATP and other nucleotides, and their plasma membrane receptors play a central role in tumor cell proliferation and immune cell regulation. For this reason, an in-depth understanding of purinergic signals in the TME may provide new therapeutic prospects. Extracellular ATP acts on tumors via specific plasma membrane receptors. Almost all cancer and immune cells express P2 receptors and are sensitive to extracellular ATP. P2Y2, a member of the purine P2 receptor family, is a G-protein-coupled receptor (GPCR). P2Y2 was first cloned from mouse NG108-15 neuroblastoma, and it induces a variety of cancer cell responses through its unique structure, including cell proliferation, migration, and invasion. In pancreatic ductal cancer cells, P2Y2 activation induces cell proliferation dependent on the activation of platelet-derived growth factor receptor-β (PDGFR-β) and PI3K/AKT. P2Y2 is highly expressed in prostate cancer and promotes the invasion and migration of prostate cancer cells in vivo. The expression of P2Y2 in human hepatocellular carcinoma cells is higher than that in normal hepatocytes. ATP has been shown to promote the proliferation of gastric adenocarcinoma cells, which is blocked by specific purinergic antagonists, and ATP treatment can induce proliferation of different glioma cell lines after 24 and 48 h. ATP and UTP activation of P2Y2R can induce migration and proliferation of MDA-MB231 and MCF-7 breast cancer cells, and it is associated with inflammation cascade activation. ATP and UTP also support cancer cell growth in A-549 human lung cancer cells. However, the effect of P2Y2 on most other tumors is still unknown, and the association between P2Y2 and OSCC has been rarely studied. In our study, P2Y2 was found to be overexpressed in the OSCC cell line and to promote the growth, invasion, and migration of tumors. In clinical samples, the expression of P2Y2 in OSCC tissues was significantly higher than that in normal tissues. In addition, stimulation with the P2Y2 agonist ATP significantly increased P2Y2 expression in OSCC cells, increased the expression of Snail, MMP2, and MMP9 genes, and decreased the expression of E-cadherin and Vimentin; the invasion and migration ability of tumor cells was enhanced. After the knockdown of the P2Y2 receptor, the expression changes of genes related to intracellular invasion and migration regulated by ATP were also weakened, and the invasion and migration of tumor cells were weakened. These results are also consistent with other findings demonstrating that P2Y2 receptors have the potential for transformative use in both in vitro and in vivo. Extracellular ATP can activate many signaling pathways, and selectively modulate Src and EGFR. Based on analysis using the TCGA database data, it was shown that Src and EGFR are highly expressed in squamous cell carcinoma tissue and also that Src and EGRF are closely correlated with P2Y2. Src, a nonreceptor tyrosine kinase, can cause phosphorylation of tyrosine residues by substrate, serving as a signal transducer of the cell surface receptor. Studies have shown that Src is excessively expressed and highly activated in OSCC and is a cancer protein that drives OSCC progression. Furthermore, Src has been shown to have a close relationship with the progression, migration, and prognosis of solid tumors, The Src family protein kinase has been demonstrated to mediate epidermal growth factor receptor (EGFR)-dependent and nondependent passages, and can even act as an upstream activator of EGFR. EGFR is considered to be the main goal of a new treatment for OSCC, as it is overexpressed in the advanced stage of the disease and prognosis in OSCC patients. It has been confirmed that GPCR-induced cell migration requires the participation of EGFR. The mechanism may be that GPCR activates MMP to release heparin-binding epidermal growth factor (HB-EGF), which is originally bound to the cell surface or extracellular matrix, and subsequently interacts with EGFR. The P2Y receptor is activated by attracting nonreceptor tyrosine protein kinase Src phosphorylated EGFR. The phosphatidylinositol-3-kinase (PI3K)/AKT signaling pathway is involved in the regulation of various cell activities, and the activation of the PI3K/AKT signaling pathway can regulate the growth, proliferation, apoptosis, and energy metabolism of tumor cells. Many studies have confirmed that AKT can increase the glycolysis level of tumor cells and promote the production of ATP without affecting aerobic oxidation, thus providing sufficient substances for biosynthesis. Abnormal activation of PI3K/AKT signaling has been found in a variety of tumors. It was found that ATP promotes MCF-7 cell proliferation through the PI3K/AKT signaling pathway. The PI3K/AKT pathway also induces stem-cell-like properties in gastric cancer cells. The Src inhibitor PP2 has been shown to inhibit the PI3K activity of colonic cancer cells, and Src has been reported to play a role in the upstream modulation of PI3K. Therefore, investigations into blocking this signaling pathway as a potential therapeutic mechanism have been undertaken. The biological function of the PI3K/AKT signaling pathway in tumor progression has been well established, but the role of P2Y2 in its regulation of the PI3K/AKT pathway remains poorly understood. Our study found that Src and EGFR increase in a time-and dose-dependent manner with ATP. ATP upregulates Src and EGFR through P2Y2 expression and then activates the PI3K/AKT pathway. After silencing P2Y2, the expression of Src and EGFR was downregulated; the PI3K/AKT pathway was weakened; and tumor growth, invasion, and migration were significantly inhibited. After adding Src and EGFR inhibitors, the expression of EGFR and PI3K/AKT was significantly inhibited. By considering these results in the context of previous studies, we concluded that P2Y2 promotes the invasion and migration of OSCC by activating the PI3K/AKT signaling pathway through the Src-EGFR axis. P2Y2 is an active regulator in tumor progression. We found that ATP was involved in tumor metabolism through the P2Y2 receptor. Also, ATP promoted the invasion and migration of OSCC cells through the Src-EGFR axis activation of the PI3K/AKT signaling pathway. These findings provide important new insight into the occurrence and development of OSCC and deliver evidence that P2Y2 has the potential to be a new therapeutic target for OSCC. |
||
PMC9648062 | Julia Hankel,Saara Sander,Uthayakumar Muthukumarasamy,Till Strowig,Josef Kamphues,Klaus Jung,Christian Visscher | Microbiota of vaccinated and non-vaccinated clinically inconspicuous and conspicuous piglets under natural Lawsonia intracellularis infection | 27-10-2022 | intestinal pathogen,porcine proliferative enteropathy,oral vaccination,16S rRNA gene,microbiome,Prevotella,Collinsella | Lawsonia (L.) intracellularis is a widespread, economically important bacterium causing the porcine proliferative enteropathy (PPE). In this study, we evaluated intestinal microbiota of naturally exposed L. intracellularis-positive pigs under standardized conditions. To obtain three independent repetitions, 27 L. intracellularis-infected pigs (19.0 ± 1.50 kg body weight) from one farm were divided into three groups at an age of 7 to 8 weeks (nine pigs/group). Pigs were either vaccinated against L. intracellularis via oral drenching on their 21st day of life (attenuated live vaccine) or non-vaccinated and selected according to clinical findings (pigs without deviating fecal consistency or with moderate to soft fecal consistency). Comparison of the clinically inconspicuous piglets that differed regarding their vaccination status showed fewer significant differences in fecal microbiota composition. The vaccination led to an overall enrichment of bacterial species belonging to the order Clostridiales, while species of the genus Collinsella and Prevotella were decreased. Several bacterial species belonging to the order Bacteroidales, mainly of the family Prevotellacecae, often closely matching Prevotella copri differed significantly between non-vaccinated clinically inconspicuous and conspicuous piglets. Whether those bacterial species play a role in mitigating the severity of an L. intracellularis infection remains to be defined. | Microbiota of vaccinated and non-vaccinated clinically inconspicuous and conspicuous piglets under natural Lawsonia intracellularis infection
Lawsonia (L.) intracellularis is a widespread, economically important bacterium causing the porcine proliferative enteropathy (PPE). In this study, we evaluated intestinal microbiota of naturally exposed L. intracellularis-positive pigs under standardized conditions. To obtain three independent repetitions, 27 L. intracellularis-infected pigs (19.0 ± 1.50 kg body weight) from one farm were divided into three groups at an age of 7 to 8 weeks (nine pigs/group). Pigs were either vaccinated against L. intracellularis via oral drenching on their 21st day of life (attenuated live vaccine) or non-vaccinated and selected according to clinical findings (pigs without deviating fecal consistency or with moderate to soft fecal consistency). Comparison of the clinically inconspicuous piglets that differed regarding their vaccination status showed fewer significant differences in fecal microbiota composition. The vaccination led to an overall enrichment of bacterial species belonging to the order Clostridiales, while species of the genus Collinsella and Prevotella were decreased. Several bacterial species belonging to the order Bacteroidales, mainly of the family Prevotellacecae, often closely matching Prevotella copri differed significantly between non-vaccinated clinically inconspicuous and conspicuous piglets. Whether those bacterial species play a role in mitigating the severity of an L. intracellularis infection remains to be defined.
Lawsonia (L.) intracellularis is an economically important bacterium and of major concern to the pig industry worldwide (1–7). Also widespread in European countries, the bacterium was detected in fecal samples of 90.3% of all sampled pig herds (3). L. intracellularis is the cause of the porcine proliferative enteropathy (PPE) (8). The clinical presentation can be acute, chronic or subclinical (9). Clinical signs depend on the form of the disease, which may range from mild to severe diarrhea, decreased feed consumption, and poor growth in case of the chronic form, the porcine intestinal adenomatosis (PIA), to sudden death in case of the acute form, the proliferative hemorrhagic enteropathy (PHE) (9, 10). These two clinically distinct forms of the disease differ not only in severity and clinical symptoms, but also with regard to the occurrence at a particular age, while PIA is the most common form diagnosed usually between 6 and 20 weeks of age (9). Even though pigs with subclinical PPE have no detectable clinical signs, they show reduced weight gain during the growth and fattening period (9) and in all forms, a proliferation of intestinal epithelial cells containing intracellular L. intracellularis can be found (11). While L. intracellularis is the cause of disease, it is not possible to rule out the interaction of one or more additional bacteria required for disease (12). Contact to the pathogen must not necessarily always lead to clinical disease. Gnotobiotic pigs lacking a normal intestinal microbiota were not colonized by the organism and failed to develop lesions (13). Further studies indicate that the intestinal microbiota of pigs are influenced by the infection with L. intracellularis itself (12, 14, 15). A clear change in community structure was observed at 21 and 28 days after experimental infection in both the small and large intestine of pigs (12). In addition, serum concentrations of folate and cobalamin, which can be of dietary origin or supplied from biosynthesis by distal gut microbiota (16, 17), were lower in pigs with PIA compared to pigs with the subclinical form (18). Oral vaccination with an attenuated L. intracellularis strain (Enterisol® Ileitis) is helpful for counteracting the disease, but does not prevent infection or transmission of the bacterium (19–23). Enterisol® Ileitis is a licensed oral, live-attenuated vaccine that confers reduction of intestinal lesions caused by L. intracellularis, growth variability and loss of weight associated with the disease. Recent findings show that oral vaccination with Enterisol® Ileitis seems to additionally alter the intestinal microbiota (12, 24, 25). Finally, an even and diverse microbiota community seems to benefit pigs infected with L. intracellularis (26). Based on these findings, an interaction of L. intracellularis and the present intestinal microbiota or their metabolites can be assumed. The aim of the present evaluation was to compare fecal microbiota of vaccinated piglets, non-vaccinated piglets without clinical symptoms, and non-vaccinated piglets with moderate clinical symptoms to address two questions: First, whether vaccination alters intestinal microbiota under naturally occurring infection, and second, whether microbiota differ between non-vaccinated animals with and without clinical symptoms. Results from this evaluation may help to better understand potential interactions of the host microbiota and L. intracellularis and might identify bacteria that are potentially related to the protection against a clinical onset of the disease.
The experiment was approved by the Animal Welfare Officer of the University of Veterinary Medicine Hannover, Germany (reference: TiHo-T-2012-13).
The investigations took place in three independent repetitions (Rep 1–3), where in total, 27 piglets were reared under the same conditions. The piglets were obtained from one farm with 420 sows of Danish genetics (DK: Danish Landrace 50% × Yorkshire 50%). Sows were regularly vaccinated (Porcine Reproductive and Respiratory Syndrome Virus (PRRSV), Swine Influenza Virus, Porcine Parvovirus, Erysipelothrix rhusiopathiae, Escherichia coli + Clostridium perfringens dam vaccine) as were the piglets (DK x Pietrain) against Porcine Circovirus 2, Mycoplasma hyopneumoniae, and Glaesserella parasuis. This farm was system partner in a regional health-monitoring program including a regular monitoring scheme. Every 6 month a farm is sampled. Among others, blood samples are tested for PRRSV, Salmonella and porcine circovirus 2, and bulk samples of feces are formed and tested for Salmonella, Brachyspira hyodysenteriae, Brachyspira pilosicoli, and Lawsonia. In the context of such a regular and long-term screening program, the herd showed no signs of Salmonella and Brachyspira infections, but clinical symptoms of an L. intracellularis infection were a common finding in piglets. Confirmed by pathogen detection in feces, alterations in fecal consistency normally occurred at about an age of 7 to 9 weeks. In contrast to many other L. intracellularis infection studies in which pigs are often experimentally challenged with gut homogenates containing besides L. intracellularis other bacteria or viruses (12) that could affect microbiota composition, an experimental approach was chosen to best reflect natural infection examined under standardized conditions. At 2 week intervals, approximately 50 piglets per trial from a total of four to five complete litters were vaccinated via oral drenching on their 21st day of life with a commercially available inactivated L. intracellularis vaccine strain (Enterisol®Ileitis, Boehringer Ingelheim Vetmedica GmbH, Ingelheim/Rhine, Germany). Subsequently, the piglets were marked individually. At weaning, all 450 to 500 piglets were reared mixed (non-vaccinated and vaccinated animals) in groups of 12 to 30 animals. From weaning onwards, samples were taken on group basis regularly and tested for their L. intracellularis status by real-time PCR according to established methods (27). At the very moment when L. intracellularis was detected in one sample, a single-animal examination procedure took place and all piglets in a weaning group were clinically examined. Three conspicuous animals (pigs with moderate to soft fecal consistency, no siblings) were randomly taken from the appropriate weaning group (Non-vac/cs+). At the same time, three inconspicuous animals (without deviating fecal consistency, Non-vac/cs–) and three vaccinated animals (without deviating fecal consistency, Vac) of the same age and identical weight were randomly chosen in a balanced gender relationship (Figure 1). L. intracellularis was detected in fecal samples of all selected pigs by real-time PCR at this time. Thus, animals were selected for the study at an early stage of infection and not at a later stage when symptoms become more apparent and might also have an increasing impact on intestinal microbiota composition. All selected animals, 14 female and 13 castrated male pigs at an age of 7 to 8 weeks with a mean body weight of 19.0 ± 1.50 kg, were transported to the Institute for Animal Nutrition, University of Veterinary Medicine Hannover, Foundation, Hannover, Germany. At the Institute for Animal Nutrition, the pigs were fed the diet they had been accustomed to on the farm. The complete diet for all trial replications derived from one batch and was offered ad libitum. The nutrient composition was in accordance with the official recommendation for piglets in Germany. The diet consisted of wheat, barley, soybean meal, soybean oil, and of a mineral and vitamin supplement, and contained 20.1% crude protein on dry matter basis and 13.8 MJ ME per kg diet. Subsequent to a 3 day adaptation period, the total daily amount of feces was collected on 5 consecutive days. A bulk sample was formed from feces of the 5 day collection period, from which, subsequent to homogenization, an aliquot was taken for microbiota analysis, and determination of L. intracellularis genome equivalents (GE, shown in the logarithm of 10) via quantitative PCR according to established methods (27). Fecal shedding of L. intracellularis was found in all groups; in total, still 25 of all 27 piglets excreted L. intracellularis. The mean fecal excretion of L. intracellularis did not differ significantly, only numerically, between the chosen animals of the groups in the present study (22), thus making the groups comparable for comparisons to identify bacteria potentially contributing to the clinical outcome of the infection.
16S rRNA gene analyses were performed as described in Hankel (28). Samples were first purified (Kit: BS 365, BioBasic Inc., Ontario, Canada) before hypervariable region V4 of the 16S rRNA gene was amplified according to previously described protocols (29) using primer F515/R806. Amplicons were sequenced with the Illumina MiSeq platform (PE250) and Usearch8.1 software package (http://www.drive5.com/usearch/) was used to assemble, quality control and cluster obtained reads. The command “fastq_mergepairs” with argument “fastq_maxdiffs 30” was used to merge the reads. Chimeric sequences were identified and removed with the help of the command “cluster_otus” (-otu_radius_pct 3) and the Uchime command included in the Usearch8.1 workflow. Quality filtering was set up with “fastq_filter” (-fastq_maxee 1) accepting a minimum read length of 200 bp. Reads were clustered into 97% ID operational taxonomic units (OTUs) and the UPARSE algorithm (30) was used to determine the OTU clusters and representative sequences. Taxonomy assignment was performed with the help of Silva database v128 (31) and the Naïve Bayesian Classifier from the Ribosomal Database Project (RDP) (32) with a bootstrap confidence cutoff of 70%. OTUs that were not present in more than at least one sample were pruned and OTUs with an abundance < 0.02% were filtered. Finally, samples with fewer than 999 total reads were removed and reads assigned to chloroplast and mitochondria were filtered. After these filtering steps, all 27 samples could be included in the statistical analysis. The dataset contained 181,046 reads (mean number of reads: 6,705; range: 2,757–13,475) mapped to 124 OTUs.
Data visualization and statistical analyses of microbiota were performed with R (version 4.1.2, www.r-project.org) using the R-packages “phyloseq” (version 1.36.0) (33). Selected alpha diversity indices (Observed, Chao 1, and Shannon) were also calculated with “phyloseq.” Means of alpha diversity estimates were compared with the aim to evaluate the influence of the factor Group. Data were checked for normality by analyzing the model residuals with the Shapiro-Wilk normality test implemented in the package “rstatix” [version 0.7.0 (34)], before conducting multiple and pairwise comparisons. Total community structure and composition of samples taken during the whole experimental phase were assessed for changes in relation to the experimental repetition and the treatments by permutational multivariate analysis of variance using Bray-Curtis distance (PERMANOVA) via the adonis function of the “vegan” package (version 2.5.7) (35). Ordination was performed using the Bray–Curtis dissimilarity-based principal coordinate analysis (PCoA). Differentially abundant OTUs between the groups were identified with the help of the R-package “DESeq2” (version 1.32.0), which uses tests based on the negative binomial distribution (36). Raw p-values were adjusted using the method of Benjamini and Hochberg (37) to control a false discovery rate (FDR) of 5%. Additionally, a cutoff for the log2-fold change of ±1 was set. Volcano plots were used to visualize differentially abundant OTUs. Statements of statistical significance were based upon p-values < 0.05. Fecal L. intracellularis excretion (log10 GE) and relative abundance of Prevotella were additionally evaluated for association using Spearman's rank correlation.
Comparisons of measured species richness estimators, Observed Species, Chao 1 and Shannon index in feces of pigs revealed no statistically significant differences between the groups (Figures 2A–C). Both the experimental repetition as well as the grouping had a significant effect on bacterial composition of the samples, while separation was pronounced due to grouping (Table 1, Figure 2D).
The microbiota of feces were dominated at phylum level by Firmicutes, Bacteroidetes, and Actinobacteria. Relative abundances of families in fecal samples of animals within the three groups are shown in Figure 2E. Relative abundance of Prevotellaceae was highest in Non-vac/cs– (Non-vac/cs–: 17.2% ±7.0, Vac: 9.6% ±3.9, Non-vac/cs+: 4.2% ±2.3). Inversely, relative abundances of the family Coriobacteriaceae increased from vaccinated to non-vaccinated, clinically inconspicuous and non-vaccinated, clinically conspicuous animals (Vac: 0.7% ±0.5, Non-vac/cs–: 1.6% ±0.8, and Non-vac/cs+: 2.4% ±2.1). The number of differentially abundant OTUs in pairwise comparisons are shown in Table 2. Volcano plots visualize differentially abundant OTUs in Figure 3. Comparison of the clinically inconspicuous piglets that differed regarding their vaccination status (Vac vs. Non-vac/cs–) showed fewer significantly different OTUs. None met the selected criterion of an FDR-adjusted p-value < 0.05 and an absolute log2-fold change >1 (Figure 3A). Nevertheless, still 13 of 124 OTUs were significantly (p-values < 0.05 and absolute log2-fold change >1) different between Vac and Non-vac/cs–. Of a total of eight significantly enriched OTUs due to vaccination, sequences of seven OTUs belonged to the order Clostridiales (Supplementary Table 1). The only sequences of OTUs that decreased due to vaccination belonged to the genus Prevotella, closely matching Prevotella copri, and the family Coriobacteriaceae from which one OTU closely matched Collinsella aerofaciens (OTU_ID 4481613). At species level, four of 124 OTUs showed significantly (selected with a criterion of FDR-adjusted p-values < 0.05 and absolute log2-fold change >1) different abundant sequences between Vac and Non-vac/cs+ (Figure 3B, Supplementary Table 2). Two OTUs were enriched in fecal samples of non-vaccinated and clinically conspicuous animals: one bacterial member of the family Veillonellaceae (OTU 300829) and the second, again OTU 4481613 (genus Collinsella within the family Coriobacteriaceae, closely matching Collinsella aerofaciens). In contrast, sequences of one OTU belonging to the genus Prevotella, family Prevotellaceae, were enriched in vaccinated compared to non-vaccinated and clinically conspicuous animals. This OTU closely matched the species Prevotella copri (OTU_ID 180825). Comparison of both non-vaccinated groups of piglets that differed regarding their clinical outcome showed significantly different abundant sequences of six OTUs (selected with a criterion of FDR-adjusted p-values < 0.05 and absolute log2-fold change >1), of which five were enriched in clinically inconspicuous animals (Non-vac/cs–) compared to clinically conspicuous animals (Non-vac/cs+, Figure 3C, Supplementary Table 3). All five OTUs belonged to the order Bacteroidales, four of the five to the family Prevotellaceae and again three of the four OTUs closely matched the species Prevotella copri. The most enriched OTU that closely matched the species Prevotella copri was again OTU_ID 180825, already found to be enriched in vaccinated compared to non-vaccinated and clinically conspicuous animals (as mentioned above), while both other OTUs were decreased in Vac compared to Non-vac/cs–. Only one OTU (OTU_64384) belonging to the family Leuconostocaceae was enriched in samples of clinically conspicuous animals (Non-vac/cs+) compared to clinically inconspicuous animals (Non-vac/cs–). Spearman's rank correlation revealed no relation between fecal L. intracellularis excretion (log10 GE) and the relative abundance of the genus Prevotella in clinically inconspicuous animals (Non-vac/cs– and Vac; correlation coefficient: −0.127; p > 0.05), while a significant relation was found in animals of the non-vaccinated groups (Non-vac/cs– and Non-Vac/cs+; correlation coefficient: −0.490; p < 0.05).
Fecal bacterial diversity was not significantly different between the groups of the present study. Still, all measured richness estimators of non-vaccinated, clinically conspicuous pigs were numerically lower compared to both the vaccinated and the non-vaccinated clinically inconspicuous pigs. Muwonge (26) observed an increased diversity of ileal microbiota associated with a reduction in L. intracellularis shedding. It has to be taken into account that the sampling sites differed in both studies and it is quite possible that there were pronounced differences in microbiota diversity within the ileum of the pigs that could no longer be clearly visualized in feces. Significant differences in the bacterial composition of fecal samples between the three groups were observed in the present study (p = 0.001, R2 = 0.1432), while the distinction between group microbial communities was higher compared to the study of Leite et al. (12). At 3 weeks post-vaccination, a significant difference in community structure was found in feces of pigs due to oral live vaccine against L. intracellularis (Bray–Curtis dissimilarity PERMANOVA: p = 0.042, R2 = 0.041) (12). The comparison of the clinically inconspicuous piglets (vaccinated or not) with clinically conspicuous piglets repeatedly revealed differences concerning OTUs belonging to the genus Prevotella within the family Prevotellaceae (phylum Bacteroidetes). Sequences of OTU_180825 (closely matched the species Prevotella copri) showed an 8-fold decrease (log2-fold change of about 3) in abundance of sequences in clinically conspicuous animals compared to both vaccinated or non-vaccinated clinically inconspicuous animals. In addition, spearman's rank correlation revealed a relation between fecal L. intracellularis excretion (log10 GE) and the relative abundance of the genus Prevotella in animals of the non-vaccinated groups (Non-vac/cs– and Non-Vac/cs+; correlation coefficient: −0.490; p < 0.05). Two further OTUs that also closely matched this species, Prevotella copri, were found to be significantly enriched in non-vaccinated clinically inconspicuous compared to conspicuous pigs. However, as described above, both OTUs were decreased due to vaccination in comparison to vaccinated and non-vaccinated clinically inconspicuous animals. It should be highlighted that 16S rRNA gene sequencing do not provide a reliable taxonomic resolution at species level and especially in case of Prevotella copri, even being the most predominant species in the gastrointestinal tract of adult pigs, a genus level classification is insufficient, which prompts the need for characterizing the role of most common Prevotella species in pigs using metagenomic and culture dependent approaches (38). Within Bacteroidetes, members of the genus Prevotella emerged as an underexplored keystone species within the human and animal microbiome (38, 39). Prevotella is highly interactive with other bacterial species in the gut (40, 41), and those highly connected taxa have been termed as keystone taxa that drive the microbiome structure and function irrespective of their abundance (42). Prevotella was identified as the most predominant genus within the intestinal tract of pigs, being an important member of both the upper and lower gastrointestinal tract microbiota (40, 43, 44) and a central constituent in one of the two most common bacterial enterotypes of pigs after weaning (41, 45–47). Mucosal microbiota of the ileum in adult pigs harbor significantly greater abundance of Prevotella compared to luminal microbiota in the same site (44). Significant differences in abundance and dynamics at sub-OTU level within this genus were observed in pigs pre- and post-weaning, while Prevotella copri, likely to be vertically transmitted but suppressed during the suckling stage, dramatically increase in the gastrointestinal tract at the end of the nursery phase after the introduction of solid food and gradually decrease at subsequent stages (40). Prevotella (particularly Prevotella copri) might be a critical bacterial taxon stimulating the feed intake in pigs (45) and was significantly associated with fat accumulation (48) and growth performance of pigs after weaning (41, 46). There is currently no proposed or demonstrated mechanism through which Prevotella increases feed intake, and Prevotella enrichment may be a product of increased feed intake rather than its driver (38). Also in the present study, the average daily feed intake (Vac: 1.297 ± 0.116, Non-vac/cs–: 1.207 ± 0.119, Non-vac/cs+: 1.165 ± 0.148 kg dry matter/day) and average daily weight gain (Vac: 894a ± 73.3, Non-vac/cs−: 857ab ± 86.3, Non-vac/cs+: 785b ± 137 g/day) differed numerically between both non-vaccinated groups (22). In vitro studies have repeatedly demonstrated that gut microbiomes rich in Prevotella show higher complex polysaccharides utilizing capacity when compared to a Bacteroides-dominated population with significantly higher concentrations of SCFAs, especially propionate, which may in part explain why Prevotella-driven enterotype is believed to benefit host health (49). Feeding diets with high amylose-to-amylopectin ratio to finishing pigs with Prevotella-rich enterotype, rather than driven by Bacteroides, increases the number and activity of butyrate-producing bacteria and the concentration of total SCFA, which may benefit gut health due to potential decreased expression of mucosal inflammation associated genes (49). Members of the genus Prevotella were associated with positive outcomes in pig production, not only with regard to growth performance but also to immune response (38). Compared to the Ruminococcaceae enterotype, the Prevotella-dominant enterotype has been associated with higher production of secretory IgA in adult pigs (46), which serves as the first line of innate defense against invading pathogens (50) and also probably plays an important role in protecting the intestine against L. intracellularis invasion and intracellular proliferation (51). However, it is also possible that Prevotella does not induce an IgA response, but simply benefits from elevated levels of secretory IgA (38). In addition, Prevotella copri is a vitamin B producer and possesses a folate and cobalamin biosynthesis pathway (17). Serum concentrations of folate and cobalamin were found to be lower in pigs with PIA compared to those having the subclinical form (18). The authors discussed that this can have an effect on amino acid metabolism and nucleic acid synthesis (18). On the other hand, gut microbiota might modulate host immune function via B vitamins (17). Dietary vitamin B9 (folate) deficiency resulted in a reduction in the regulatory T cell population which plays an important role in the prevention of excessive immune responses, and mice fed a vitamin B9-deficient diet exhibit increased susceptibility to intestinal inflammation [reviewed in (17)]. Dietary vitamin B12 (cobalamin) deficiency decreases the number of CD8+ T cells and suppresses natural killer T-cell activity in mice, suggesting that vitamin B12 contributes to the immune response via CD8+ T cells and natural killer T cells [reviewed in (17)]. CD8+ and CD4+CD8+ lymphocytes in pigs produce IFNγ, which is an important cytokine for protection against intracellular pathogens and suspected of being a crucial factor mediating protection against L. intracellularis infection (52). It has been shown that piglets that harbor higher Prevotella may have better protection against diarrhea. Judging by the results of a study conducted by Dou et al. (53), the higher abundance of Prevotella may contribute to allowing healthy pigs to better adapt to post-weaning dietary conditions, thereby reducing the risk of developing post-weaning diarrhea. Similar findings were seen by Sun et al. (54) who reported greater abundance of gut Prevotella in non-diarrheic piglets compared to diarrheic piglets, while Prevotellaceae UCG-003 was the key bacterium in non-diarrheic microbiota of piglets. By using an ETEC (Enterotoxigenic Escherichia coli)-induced diarrheal model in piglets, it was shown that resistant piglets (challenged with ETEC but not suffering from diarrhea) demonstrated the highest percentage of Prevotella (6.7%) even compared to non-challenged control piglets (4.2%). In addition, Prevotella decreased as the piglets were transient from the pre-diarrheic state (1.7%) to the diarrheic state (0.2%) (55). Whether bacterial species of the order Bacteroidales (especially Prevotella) play a role in mitigating the severity of the disease or are more an indicator of unimpaired feed intake and “balanced” ileal starch and protein digestibility rates, which maintain continuous fermentation conditions in the large intestine, remains to be defined. Muwonge et al. (26) found the top three families Prevotellaceae, Ruminococcaceae, and Lactobacillaceae as the core microbiota in their study to be associated with changes in the threshold for shedding L. intracellularis, but not to have a major effect on the clinical outcome of the disease. Some Prevotella species were found to be decreased due to oral L. intracellularis vaccination in the present study as well as in investigations by Leite et al. (12, 24), who found one Prevotella species to be increased in the former study and two Prevotella species to be increased and two to be decreased in the latter study due to oral L. intracellularis vaccination. Prevotella is a genus with high genetic diversity within and between species, and certain strains may exhibit pathobiontic properties by promoting chronic inflammation in humans (56, 57). Similar to pigs, the controversial role played by Prevotella copri in human health is discussed as well (56). As the Prevotella genus is certainly not exclusively beneficial, further research into the characterization at species or even strain level is needed to clarify its preventative or promotive role in the pathogenesis of pig diarrhea (38).
To date, studies that evaluate whether oral immunization results in discernible alterations of the microbiota are rare (58, 59). Comparison of fecal microbiota between the vaccinated and non-vaccinated clinically inconspicuous piglets showed fewer differences compared to the other group comparisons in the present study. Nevertheless, the existing differences in terms of OTUs differing between the two groups (Collinsella aerofaciens, Prevotella copri, and Clostridiales) resemble previous investigations with experimental L. intracellularis challenge, that have also shown that oral vaccination against L. intracellularis shapes the gut microbiota in weaned piglets. Leite et al. (12, 24) who investigated the impact of an oral live vaccine against L. intracellularis (Enterisol® Ileitis) in pigs experimentally challenged with L. intracellularis were in line with results of the present study. Vaccination against L. intracellularis in dually challenged pigs (L. intracellularis and S. enterica serovar Typhimurium) also induced a significant decrease in abundance of an OTU that closely matched Collinsella aerofaciens and a significant increase in the abundance of Clostridium species (24). In a further investigation by Leite et al. (12), vaccination led again to a significant decrease in the abundance of Collinsella, Fusobacterium, and Campylobacter among other microbial changes compared with non-vaccinated and L. intracellularis-challenged animals. The authors suggest that the intestinal pathogen challenge induces dysbiosis resulting from an increase in the number of such pathobionts and vaccination might mitigate these effects, which could contribute to the development and severity of disease (12). However, it must be emphasized that both studies characterized the porcine intestinal microbiota response of pigs to an experimental L. intracellularis challenge. Challenge material originates from L. intracellularis grown in McCoy cells (24), or mucosal scrapings from the ileum of pigs with confirmed L. intracellularis infection and gross PPE lesions in which L. intracellularis was the predominant bacterium (12). In the latter, other bacteria (species of Campylobacter, Chlamydia trachomatis, Bacteroides fragilis, and Fusobacterium nucleatum) detected in challenge material, along with some viruses, were enriched in the intestine of pigs challenged with a gut homogenate and affected by oral vaccination (12). The question remains whether these bacteria also play a role under natural infectious conditions, when evaluating potential effects of vaccination on microbiota. Therefore, studies investigating microbiome changes under naturally occurring L. intracellularis infection are still needed (12). To the best of our knowledge, the present study investigated microbiota of naturally exposed L. intracellularis-positive pigs under standardized conditions for the first time and could show similar microbiota changes due to vaccination for Collinsella and Clostridiales species as found in experimental L. intracellularis challenge models. Compared to vaccinated piglets, the OTU 4481613 (family Coriobacteraceae, genus Collinsella, closely matching Collinsella aerofaciens) was especially enriched in non-vaccinated clinically conspicuous animals, but also in non-vaccinated clinically inconspicuous animals. Leite et al. (12, 24), who observed similar results refer to this bacterium as pathobiont. Pathobionts are commensal bacteria that are typically kept at low levels within a healthy gut and do not cause problems in immune-competent hosts, but have the potential to cause harm to the host leading to inflammation and pathology (60). In human intestinal epithelial cell lines, it could be shown that Collinsella increases gut permeability by reducing the expression of tight junction protein and induces expression of IL-17 network cytokines, suggesting that an expansion of Collinsella may cause an increase in pro-inflammatory conditions with a loss of gut epithelial integrity (61). Augmented gut permeability is believed to result in an escape of bacteria and endotoxins from the intestinal lumen to the mesenteric lymph nodes and in portal circulation (62). Comparison of vaccinated and non-vaccinated clinically conspicuous animals revealed besides Collinsella significant differences in sequences belonging to the family Veillonellaceae. Sequences of Veillonellaceae were enriched only in four of the nine animals belonging to the Non-vac/cs+ group (data not shown). The animal with the highest OTU sequence count belonging to Veillonellaceae had the highest fecal dry matter content in this group and at the same time within the range of clinically inconspicuous animals belonging to the other two groups (data not shown). For this reason, the enrichment in Veillonellaceae in Non-vac/cs+ does not appear to be related to disease severity. Prevotellaceae, Veillonellaceae as well as Ruminococcaceae and Lachnospiraceae very likely form a functional group in the lumen of healthy piglets, and the increased number of Prevotella within Bacteroidetes as well as of Veillonellaceae, Lachnospiraceae and Ruminococcaceae within the Firmicutes equip the large intestine with metabolic abilities that are indispensable for host survival; these families are positively correlated with gene functions related to amino acids, energy, cofactors and vitamins metabolism (63).
Intestinal microbiota of pigs changed similarly due to oral L. intracellularis vaccination in naturally occurring L. intracellularis infection compared with experimental L. intracellularis infection. In addition, intestinal microbiota differed between non-vaccinated animals with and without clinical symptoms presented as altered fecal quality. Sequences of several bacterial species belonging to the order Bacteroidales, mainly of the family Prevotellacecae, often closely matching Prevotella copri, were lower in non-vaccinated clinically conspicuous piglets compared to inconspicuous ones. Whether those bacterial species play a role in mitigating the severity of an L. intracellularis infection has yet to be determined.
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA862063.
The animal study was reviewed and approved by the Animal Welfare Officer of the University of Veterinary Medicine Hannover, Germany (reference: TiHo-T-2012-13).
JH, JK, and CV: conceptualization. JH, SS, TS, JK, KJ, and CV: methodology. JH, SS, JK, KJ, and CV: validation. JH and KJ: formal analysis. SS and CV: investigation and supervision. JK: resources. JH, UM, KJ, and CV: data curation. JH: writing-original draft preparation and visualization. JH, JK, TS, KJ, and CV: writing-review and editing. SS, JK, and CV: project administration. JK and CV: funding acquisition. All authors have read and agreed to the published version of the manuscript. All authors contributed to the article and approved the submitted version.
This Open Access publication was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - 491094227 Open Access Publication Costs and the University of Veterinary Medicine Hannover, Foundation.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648086 | Gulsah Evyapan,Umit Luleyap,Halil Mahir Kaplan,Ismail Oguz Kara | Ornidazole suppresses CD133+ melanoma stem cells via inhibiting hedgehog signaling pathway and inducing multiple death pathways in a mouse model | 01-10-2022 | Aim To evaluate the inhibitory effects of ornidazole on the proliferation and migration of metastatic melanoma cell line (B16F10) in vitro and its anti-cancer effects in vivo using a melanoma mouse model. Methods We investigated the effects of ornidazole on cell viability (Crystal Violet and MTT assay) and migration ability (wound-healing assay) of B16F10 melanoma cells, and its ability to trigger DNA damage (Comet assay) in vitro. We also sorted CD133+ and CD133- cells from B16F10 melanoma cell line and injected them subcutaneously into Swiss albino mice to induce tumor formation. Tumor-bearing mice were divided into control and treatment groups. Treatment group received intraperitoneal ornidazole injections. Tumors were resected. Real-time polymerase chain reaction was used to determine the expression of genes involved into Sonic hedgehog (Shh) signaling pathway, stemness, apoptosis, endoplasmic reticulum (ER) stress, ER stress-mediated apoptosis, and autophagy. Shh signaling pathway-related proteins and CD133 protein were analyzed by ELISA. Results Ornidazole effectively induced DNA damage in CD133+ melanoma cells and reduced their viability and migration ability in vitro. Moreover, it significantly suppressed tumor growth in melanoma mouse model seemingly by inhibiting the Shh signaling pathway and ER-stress mediated autophagy, as well as by activating multiple apoptosis pathways. Conclusions Our preclinical findings suggest the therapeutic potential of ornidazole in the treatment of metastatic melanoma. However, larger and more comprehensive studies are required to validate our results and to further explore the safety and clinical effectiveness of ornidazole. | Ornidazole suppresses CD133+ melanoma stem cells via inhibiting hedgehog signaling pathway and inducing multiple death pathways in a mouse model
To evaluate the inhibitory effects of ornidazole on the proliferation and migration of metastatic melanoma cell line (B16F10) in vitro and its anti-cancer effects in vivo using a melanoma mouse model.
We investigated the effects of ornidazole on cell viability (Crystal Violet and MTT assay) and migration ability (wound-healing assay) of B16F10 melanoma cells, and its ability to trigger DNA damage (Comet assay) in vitro. We also sorted CD133+ and CD133- cells from B16F10 melanoma cell line and injected them subcutaneously into Swiss albino mice to induce tumor formation. Tumor-bearing mice were divided into control and treatment groups. Treatment group received intraperitoneal ornidazole injections. Tumors were resected. Real-time polymerase chain reaction was used to determine the expression of genes involved into Sonic hedgehog (Shh) signaling pathway, stemness, apoptosis, endoplasmic reticulum (ER) stress, ER stress-mediated apoptosis, and autophagy. Shh signaling pathway-related proteins and CD133 protein were analyzed by ELISA.
Ornidazole effectively induced DNA damage in CD133+ melanoma cells and reduced their viability and migration ability in vitro. Moreover, it significantly suppressed tumor growth in melanoma mouse model seemingly by inhibiting the Shh signaling pathway and ER-stress mediated autophagy, as well as by activating multiple apoptosis pathways.
Our preclinical findings suggest the therapeutic potential of ornidazole in the treatment of metastatic melanoma. However, larger and more comprehensive studies are required to validate our results and to further explore the safety and clinical effectiveness of ornidazole.
Melanoma is a highly metastatic skin cancer developing as a result of malignant transformation of melanocytes and one of the fastest growing malignancies worldwide (1-4). Although the five-year survival rate of melanoma is considerably high (>95%) when diagnosed early, melanoma prognosis is extremely poor once the disease becomes metastatic (5,6). Conventional chemotherapies, generally aimed at inhibiting cell division, have shown little survival benefit. The most important cause of treatment failure is resistance to conventional chemotherapy and radiotherapy (7-9). Chemotherapy resistance in advanced-stage melanoma is associated with a subset of CD133+ melanoma-initiating cells (10), a type of cancer stem cells (CSCs) (11-13). Thus, eradication of CSCs is an important goal in therapeutic approaches because it could dramatically reduce the risk of metastatic dissemination and relapses, the major causes of mortality in oncology patients (14). Among CSC markers, CD133 (Prominin-1), a pentaspan membrane glycoprotein, has been considered as one of the most important surface markers for identification of melanoma stem cells (15,16). High levels of CD133 expression have been linked to the high tumorigenicity and metastatic potential of melanoma cells (17-19). Moreover, CD133 promotes metastasis via interaction with signaling pathways that regulate cell migration and polarity dynamics (17,20). Indeed, the crosstalk between several different signaling pathways, including Sonic hedgehog (Shh) (21,22), plays a role in the pathogenesis and progression of malignant melanoma (23,24). Therefore, anti-cancer treatments targeting the Shh pathway show promise in many human cancers, including melanoma, lung, breast, and prostate cancers (25-27). Over the last decade, novel treatment strategies, including immunotherapy and targeted therapy, targeting the key regulators of these signaling pathways (28-31) have resulted in some improvement in the setting of metastatic melanoma (32). Despite this, patients with melanoma often experience disease progression and relapse with acquired resistance after several months of monotherapy, and the five-year survival rate for the metastatic disease remains low (6,33). Development of new effective and selective therapeutic compounds that specifically target melanoma-specific CSCs and related signaling pathways is therefore needed (14). Ornidazole has been used for the treatment of infections caused by anaerobic bacteria and protozoa in humans through a mechanism that includes pre-activation by the reduction of the nitro groups and the production of toxic derivatives and radicals (34). Ornidazole enters the cell by passive diffusion under anaerobic conditions and inhibits DNA synthesis by breaking and destabilizing the DNA structure (35). The effects of ornidazole on melanoma cells and the mechanisms through which it may influence the tumor structure in melanoma have not been investigated. Given that tumor cells also create anaerobic environments, we hypothesized that ornidazole may be a promising compound for the treatment of malignant melanoma. To test this hypothesis, we evaluated the inhibitory effects of ornidazole on the proliferation and migration of metastatic melanoma cell line (B16F10) in vitro and its anti-cancer effects in vivo using a melanoma mouse model.
B16F10 mouse melanoma cancer cell line was purchased from ATCC (CRL-6475, Manassas, VA, USA). Ornidazole was obtained from Sigma (cat. no: 16773-42-5, Sigma, Roedermark, Germany) and was used after having been dissolved in Dulbecco's Modified Eagle Medium (DMEM) and distilled water at room temperature for in vitro and in vivo assays, respectively.
Cells were maintained in DMEM (cat. no: 11965084, ThermoFisher, Waltham, MA, USA) supplemented with 100 U/mL penicillin (cat. no: P4443, Sigma), 100 mg/mL streptomycin (cat. no: 85886, Sigma), and 10% fetal bovine serum (FBS) (cat. no: P2442, Sigma) in a humidified incubator at 37 °C under 5% CO2 atmosphere.
Crystal Violet staining was performed to observe the effects of ornidazole on the viability of B16F10 cells. Cells were seeded in a 12-well plate at a density of 70 000 cells/well and were treated with ornidazole at various concentrations (400, 800, and 1200 μg/mL) under humidified atmosphere containing 5% CO2 at 37 °C for 48 h. The treated cells were then stained with 0.5% Crystal Violet staining solution (cat. no: 548-62-9, Sigma) at room temperature for 30 minutes. After the plate was gently rinsed with water and dried, the absorbance values of each well were determined spectrophotometrically at 570 nm by using a plate reader (Biochrom, Cambourne, UK).
The effects of ornidazole on tumor cell growth were assessed with the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay (cat. no: CT01-5, Sigma). The MTT assay was performed as per the manufacturer's protocol. Shortly, the cells were seeded into a 96-well plate at a density of 10 000 cells/well and cultured for 24 h, 48 h, and 72 h, in the presence or absence of ornidazole at various concentrations (50, 100, 200, 800, 1200, 1600, 3600 μg/mL). The cells that were not treated with ornidazole were used as a control group, and an equal amount of DMEM medium was added to the control well plates. Following incubation at 37 °C for 4 h, 100 μL of solubilizing buffer was added to each well. After overnight incubation, the absorbance values of each well were determined with an ELISA plate reader at 590 nm to assess cell viability. In order to find the most effective dose of ornidazole (from the lowest dose to the lethal dose), more than one dose was selected at regular intervals taking into account the doses that act on various parasites. In addition, different drug concentrations from 0 to lethal dose (maximum concentration) were tested with increasing drug doses in the MTT assay, and the most suitable concentrations were selected for treatment.
The effect of ornidazole on cell migration was evaluated with wound-healing assay (36). In short, B16F10 cells were seeded into six-well plates with a density of 80 000 cells/well and allowed to grow to 80%-90% confluence. Next, the cells were washed with phosphate buffered saline (PBS) and cultured in DMEM containing 0.05% fetal bovine serum (FBS) for 16 h. A scratch/wound with clear edges was created across the width of the well with a 200-μL pipette tip and was washed twice with PBS to remove debris or detached cells. The cells then were exposed to different concentrations of ornidazole (0, 400, 800, 1200 μg/mL). Cell migration was photographed at 0, 8, and 24 h with a digital camera installed on the microscope (Leica, Wetzlar, Germany). The wound area was measured with the Image J software (NIH, Bethesda, MD, USA).
Ornidazole has been reported to efficiently induce DNA damage under the anaerobic conditions (37). Therefore, we reasoned that the hypoxic tumor microenvironment might facilitate the access of ornidazole into the tumor cells to create DNA damage. The Comet assay was performed to evaluate the double-strand breaks (DSBs) in the DNA of B16F10 melanoma cells. The cancer cells were treated with ornidazole at various concentrations (0, 400, 800, and 1200 μg/mL) for 24 h, 48 h, and 72 h. Treated cells were then mixed with low-melting-point agarose, and the mixture was spread on the frosted microscope slides precoated with a thin layer of normal-melting-point agarose. The slides were exposed to lysis buffer (2.5 M NaCl, 1% Triton X-100, 100 mM EDTA, 10 mM Tris-HCI, pH 10) at 4 °C for 1 h. Then, the electrophoresis chamber was filled with cold electrophoresis buffer (300 mM NaOH, 1 mM EDTA, pH: 13), and the slides were kept in the buffer for 40 min at 4 °C to allow DNA unwinding. After DNA unwinding, the slides were subjected to electrophoresis for 20 min at 25 V. Then, the slides were washed with neutralization buffer (0.4 M Tris-HCI, pH: 7.5) for 15 min and with 96% ethanol for 10 min. All preparatory steps were carried out in the dark to avoid additional DNA damage (38). After ethidium bromide staining, the slides were viewed at 40 × magnification, and the Comet tails were visualized with a UV microscope (Leica). The degree of DNA damage was assessed based on the tail length and distribution of DNA in the tail, which is known as the olive tail moment (OTM). The measurements were derived from 100 tumor cells per sample group. The data were analyzed with Comet Score software (CaspLab, Wroclaw, Poland).
CD133+ cells on B16F10 cell line were identified and isolated as previously described (39). In short, cells were detached using PBS, resuspended in PBS, and incubated with PE-conjugated mouse anti-human CD133 antibody (clone: 315-2C11, Biolegend, San Diego, CA, USA) at 1:100 dilution at room temperature for 30 minutes in the dark. Samples were acquired on FACSAria III flow cytometer (Becton Dickinson, Beckman Coulter, Inc., Brea, CA, United States), and CD133+ or CD133- cells were sorted into DMEM medium.
After sorting, cells were centrifuged, rinsed, and seeded in 6-well plates into serum-free medium supplemented with 20 μg/L basic fibroblast growth factor-basic (bFGF), 10 μg/L epidermal growth factor (EGF), penicillin G (100 U/mL), and streptomycin (100 μg/mL). The culture medium was changed every two days, and cell proliferation was observed under an inverted phase-contrast microscope, at 0 h, 24 h, 48 h, and 72 h.
Thirty-six male Swiss albino mice, 8 weeks of age, were purchased from the Cukurova University Health Sciences Experimental Application and Research Center and divided into three groups each consisting of 12 mice: 1) unsorted group (6 treatment and 6 control mice), 2) CD133+ group (6 treatment and 6 control mice), and 3) CD133- group (6 treatment and 6 control mice). All mice were kept under the following conditions: temperature 21-23 °C, humidity 70%, and 12 h light-dark cycle with food and water ad libitum. To minimize animal suffering, all mice were anesthetized with isoflurane (2%-2.5%) before euthanasia.
Cells were incubated in a humidified 37 °C incubator with 5% CO2. For melanoma cancer model, unsorted B16F10 cells (5x105), CD133+ cells (1x105), and CD133- cells (5x105) were resuspended in 100 μL of DMEM and subcutaneously injected into the Swiss albino mice (n = 36). Tumor length and width were measured three times a week with Vernier Calipers (40). The measurements were made between the longest longitudinal (length) and the longest transverse (width) sections. The long section was considered as the tumor length and the short section was considered as the tumor width. The tumor volume was calculated by the following formula: (width × width × length)/2 (40). After tumor volume reached ~ 100-150 mm3, intraperitoneal (IP) treatment with ornidazole (80 mg/kg body weight) was delivered to six mice (treatment arm) in each group daily for 12 days. The remaining six mice in each group did not receive ornidazole treatment.
We used RT-PCR to analyze whether ornidazole treatment affected the expression levels of hedgehog signaling pathway-related genes, including Shh, Smo, Gli1, Ptch1, and Bmi1 in tumor tissues. Next, we examined whether ornidazole altered the expression of significant CSC-related genes, including Prom1 (gene encoding CD133), Oct3/4, Nanog, and Sox2, in unsorted, CD133+ and CD133- tumor cells. As the BCL2 family plays a critical role in the execution of programmed cell death, we determined whether ornidazole-induced apoptosis was associated with changes in Bax, Casp3, Casp9 and Bcl2. We also explored whether ornidazole activated other apoptosis-related pathways in melanoma tumors, including Grp78 and Xbp1 as well as ER stress-mediated apoptosis markers, such as Chop and Casp12. As ER stress is also a potent inducer of autophagy process in the cells (41), we examined the expressions of autophagy-related genes, including Atg5, Atg12, Becn1, Map1lc3b, and Atf4 in unsorted, CD133+, and CD133- groups. RNA was prepared by using TRIzol (Invitrogen, Waltham, MA, USA). cDNA was prepared from 2 μg of total RNA in a 20-μL RT reaction by using the Applied Biosystems High-Capacity cDNA kit (ThermoFisher Inc., Waltham, MA, USA). β-actin was selected as the reference gene. The primers used are listed in Table 1. Polymerase chain reactions (PCR) were performed with Power SYBR Green PCR Master Mix (ThermoFisher Inc.) on a 96-well reaction plate, and Applied Biosystems StepOnePlus Real-Time PCR System (Applied Biosystems, Waltham, MA, USA) was used for quantification of gene expression. The quantitative PCR conditions were as follows: 50 °C for 2 min, 95 °C for 10 min, 40 cycles of 95 °C for 30 seconds, and 60 °C for 1 min). Each biological sample was run in triplicate on the same reaction plate, and the average Ct values obtained from technical replicates were used for final calculation.
Hedgehog signaling pathway-related proteins and CD133 protein were analyzed with ELISA kit (SunredBio Inc., Shanghai, China) according to the manufacturer's instructions. For total protein extraction, tumor tissues were treated with radioimmunoprecipitation assay and 0.3% (v/v) protease inhibitor (cat no: 8778, Sigma-Aldrich) and homogenized with an electric homogenizer on ice. Next, the lysates were centrifuged at 10 000 rpm for 10 min at 4 °C. The relative concentration of each protein in the supernatant was determined with appropriate ELISA kits.
The normality of distribution for continuous variables was confirmed with the Shapiro Wilk test. Differences between the groups in Crystal Violet and MTT assays results were assessed with a one-way ANOVA with multiple comparisons and a post-hoc Dunnet's test. Differences in wound-healing assay results were assessed with a one-way ANOVA with Bonferroni correction. Differences in Comet assay results, as well in tumor volumes in experimental melanoma models, were assessed with a two-way ANOVA with Bonferroni correction. The differences in gene expression were assessed with one-way ANOVA with Bonferroni correction. For data obtained by ELISA assay, two-way ANOVA with Bonferroni correction was performed. Different doses of ornidazole treatment were compared with the control group, and the data are presented as mean ± standard deviation. The level of significance was set at 0.05 (P < 0.05). Data analysis was performed with IBM SPSS, version 20.0 (IMB Corp., Armonk, NY, USA) and Graphpad Prism, version 5.0. (GraphPad Software Inc, San Diego, CA, USA)
The MTT and Crystal Violet assays confirmed the inhibitory effects of ornidazole on the viability and proliferation of B16F10 melanoma cancer cells as the survival percentage of cells treated with ornidazole was lower compared with that of control cells. Crystal Violet assay showed that melanoma cells treated with ornidazole could not survive in the medium for 48 h, as opposed to control cells (Figure 1A, P < 0.001). MTT assay showed that ornidazole inhibited B16F10 cell proliferation in a concentration-dependent and time-dependent manner, with concentrations >200 μg/mL markedly reducing cell viability (Figure 1B, 1C, and 1D, P < 0.001 for all). An increase in the concentration of ornidazole from 200 μg/mL to 3200 μg/mL raised the inhibition rate from 5% to 97% during 72-h treatment (Figure 1D).
Ornidazole dose-dependently and significantly inhibited the migration of B16F10 cells compared with the untreated control (Figure 2A). Treatment with 400, 800, and 1200 μg/mL ornidazole inhibited B16F10 cells migration by 15%, 60%, and 96%, respectively, after 24 h (Figure 2A and 2B, P < 0.05). These results may also suggest that ornidazole inhibits the invasion potential of melanoma cells.
The Comet assay showed that ornidazole increased DNA damage in B16F10 melanoma cancer cells in a time-dependent and dose-dependent manner (Figure 2C and 2D) compared with the control group. Moreover, increasing the concentration of ornidazole from 400 μg/mL to 1200 μg/mL increased the olive tail moment (OTM) rate from 5% to 17% and from 9% to 22.5% during 24-h and 72-h treatment, respectively (P < 0.001).
All mice developed tumors in the injection site. However, the rate of tumor growth and the tumor volume were significantly higher in the CD133+-injected mice than in the CD133- and unsorted groups (Figure 3F, 3H, and 3K). Furthermore, CD133+ cell-derived tumors created abnormally big metastasis around the injection site, whereas no metastasis was found in the mice who received CD133- or unsorted cell injections (Figure 3D, 3G, and 3I). Therefore, our results confirmed a greater in vivo tumorigenic potential of CD133+ cells compared with that of CD133- and unsorted B16F10 cells. Mean tumor volumes significantly decreased in all treatment groups (unsorted, CD133+, and CD133-) compared with the control mice in the same group. In the unsorted group, three mice from the treatment arm completely recovered (complete loss of tumor tissue) and the other three had a 96% reduction in tumor tissue, while six control mice had a 200% increase in tumor volume (Figure 3D, 3E and 3F). In the CD133+ group, six treated mice showed 50% reduction in tumor volume, while 6 control mice had an increase in the tumor volume by 337.5% (Figure 3G and 3H). In the CD133- group, in 6 treated mice tumor volume decreased, while in 6 control mice it significantly increased (150%) (Figure 3I, 3J, and 3K).
The expression levels of Shh, Smo, Gli1, Ptch1, and Bmi1 (Figure 4A, P < 0.001), as well as those of SHH, PTCH1, SMO, GLI1 (Figure 4b) and CD133 (Figure 4c) proteins were significantly reduced in the treatment group compared with the control group. Therefore, both RT-PCR (Figure 4A and Figure 5A) and ELISA showed that ornidazole selectively downregulated CD133 and hedgehog signaling pathway-related genes in melanoma tumors.
The expression levels of CD133 (P < 0.001), Oct3/4 (P < 0.001), Nanog (P < 0.001), and Sox2 (P < 0.001) significantly decreased in the ornidazole-treated mice in all three groups compared with the control mice (Figure 5A).
In all groups, ornidazole significantly increased Bax activation and decreased Bcl2 activation (Figure 5B P < 0.001). Considering that ornidazole also significantly decreased both mRNA and protein expression level of Gli1 (Figure 4A and 4B, P < 0.001), we believe that apoptosis induced by ornidazole in the melanoma tumor tissue might be partly mediated by Gli1/Bcl2/Bax-axis. Ornidazole increased the expression levels of cellular stress markers, including Grp78 and Xbp1 (Figure 6A, P < 0.001) as well as ER stress-mediated apoptosis markers, such as Chop and Casp12, in unsorted, CD133+, and CD133- groups (Figure 6B, P < 0.001). Autophagy-related genes, including Atg5, Atg12, Becn1, Map1lc3b, and Atf4, in unsorted, CD133+, and CD133- groups were significantly downregulated in ornidazole-treated cells compared with the control groups (Figure 6C, P < 0.001). Taken together, our results demonstrated that ornidazole might exert its anti-cancer effect in melanoma by inhibiting the autophagy machinery and by activating multiple apoptosis-related pathways.
To our knowledge, this is the first study to explore the therapeutic potential of ornidazole for the treatment of malignant melanoma. Our in vitro experiments demonstrated that ornidazole efficiently and significantly inhibited cell viability and proliferation, suppressed migration capacity, and induced DNA damage in B16F10 melanoma cells. Furthermore, in vivo data showed that ornidazole treatment dramatically reduced tumor volume, specifically targeted CD133+ CSCs, upregulated pathways associated with cellular and ER-related stress, inhibited the Shh signaling and ER-stress mediated autophagy process, and activated two different apoptosis pathways in melanoma tumors. Targeting the Shh signaling has been shown as a potential therapeutic approach for the treatment of some cancers, including melanoma (24,26,27,42). Interestingly, our study showed that the reduction in Smo and Bmi1 expressions in CD133+ melanoma cell-injected mice, although significantly decreased compared with the control mice, was lower compared with other Shh-related genes. In a recent study, Bmi1 induced an invasive signature that promoted metastasis and chemoresistance in melanoma (43). Smo has been shown to be involved in the pathogenesis and progression of many solid tumors, including breast, liver, pancreatic, and colon cancer. The overexpression of Smo has been linked to the tumor size, metastasis, invasiveness, and recurrence of disease, and SMO inhibitors have been used to suppress cancer formation, trigger apoptosis, and suppress cancer stem cell activity (44). Therefore, the downregulation of Smo and Bmi1, strong mediators of metastasis and recurrence, by ornidazole treatment highlights the therapeutic efficacy of ornidazole in melanoma setting. The overexpression of the Sox2, Oct4, and Nanog genes is a general hallmark of a variety of human malignancies. These genes are associated with tumor invasion/metastasis, tumor formation, drug resistance, and disease recurrence after chemo/radiotherapy (10,45). In fact, Sox2 contributes to melanoma cell invasion (46) and is a critical factor for self-renewal and tumorigenicity of melanoma-initiating cells (47). Targeting the molecular pathways in CSCs and downregulation of CD133, Nanog, Oct3/4, and Sox2 have been crucial steps to control the tumor progression (19,37,48). We observed a significant decrease in the expression levels of these genes in the treatment groups compared with controls. This suggests that ornidazole can effectively target the CSC population in melanoma tumors. In the last decades, developing treatment strategies that efficiently eliminate cancer cells and CSCs by apoptosis has become one of the major goals in cancer research. A limited number of anti-cancer agents directly target apoptotic pathways, and these small molecules are designed to inhibit anti-apoptotic BCL2 family members (49). Also, combining immunotherapy with agents that target the BCL2 antiapoptotic proteins may lead to a more effective treatment in melanoma (50). In this study, ornidazole treatment significantly downregulated Bax expression and upregulated Bcl2 expression compared with the control group. The increase in Bax expression and decrease in Bcl2 expression was again lower in the tumor tissues of CD133+-injected mice than in other two groups (unsorted and CD133), which might indicate the greater resistance of CD133+ melanoma cells to apoptosis. On the other hand, as ornidazole treatment also downregulated the expression of Gli1, it seems reasonable to argue that ornidazole activates the apoptosis through Gli1/Bcl2/Bax-axis in melanoma cells. We also observed significant differences in Casp3 expression between the ornidazole-treated groups. While Casp3 expression increased in mice that received CD133- melanoma cells, it decreased in unsorted and CD133+ cell-injected mice compared with their control groups. Normally, increased Casp3 activity is considered a sign of apoptosis and a positive indicator for efficient cancer treatment. However, growing evidence indicates that Casp3 promotes cancer cell growth, cellular migration, invasiveness, recurrence, and angiogenesis (51,52). We believe that this dual role of Casp3 may be related to its acting as a switch protein in important cellular pathways. However, investigating this issue was out of the scope of this study and it needs to be addressed in future studies. The unfolded protein response (UPR) is an intracellular signaling pathway activated by the accumulation of unfolded/misfolded proteins in the ER, and thus, a vital cytoprotective mechanism in response to ER stress (53). ER stress is involved in apoptotic mechanisms leading to melanoma cell death, and ER stress-related pathways have shown to play an important role in regulating tumor formation and resistance (54). Our results demonstrated a significant increase in the expression of the cellular stress markers Grp78 and Xbp1, and of ER stress-mediated apoptosis markers Chop and Casp12 in ornidazole-treated groups compared with the control group. These findings suggest that to induce cellular death in melanoma cancer cells, ornidazole triggers both the Gli1/Bcl2/Bax-dependent and ER stress-mediated apoptosis pathways. UPR signaling also activates autophagy, an evolutionarily conserved and lysosome-dependent degradation pathway in which cytoplasmic macromolecules, unfolded/misfolded proteins, damaged organelles, or pathogens are delivered to lysosomes, and are digested by lysosomal enzymes to generate ATP, nucleotides, amino acids, fatty acids, etc (55). In advanced stages of cancers, autophagy contributes to the survival and growth of the established tumors and facilitates metastasis (55). In addition, autophagy is involved in focal adhesion dynamics during cell migration and invasion. Inhibition of autophagy decreases tumor cell motility due to reduced focal adhesion turnover (56,57). In our study, ornidazole treatment significantly downregulated the expressions of ER stress-mediated autophagy markers, including Atg5, Atg12, Becn1, Map1lc3b, and Atf4, in all treatment groups compared with the control groups. Although these results suggest the potential of ornidazole to inhibit the autophagy mediated by ER stress in the metastatic melanoma setting, more detailed in vivo studies are required to analyze the influence of long-term autophagy arrest on tumor progression, metastasis, and survival. Our findings suggest the potential of ornidazole as a novel anti-cancer agent for the treatment of malignant melanoma. Its years-long availability on the market makes ornidazole a better therapeutic candidate compared with new drugs, which require costly development. Nevertheless, larger and comprehensive studies are required to explore the safety, feasibility, and clinical effectiveness of ornidazole treatment in melanoma. |
||
PMC9648094 | Julliane Tamara Araújo de Melo Campos,Matheus Sena de Oliveira,Luisa Pessoa Soares,Katarina Azevedo de Medeiros,Leonardo René dos Santos Campos,Josivan Gomes Lima | DNA repair-related genes and adipogenesis: Lessons from congenital lipodystrophies | 07-11-2022 | DNA repair,adipogenesis,genetic lipodystrophies,metabolism | Abstract Classical and progeroid congenital lipodystrophies are a collection of rare diseases displaying a large genetic heterogeneity. They occur due to pathogenic variants in genes associated with adipogenesis, DNA repair pathways, and genome stability. Subjects with lipodystrophy exhibit an impairment in the homeostasis of subcutaneous white adipose tissue (sWAT), resulting in low leptin and adiponectin levels, insulin resistance (IR), diabetes, dyslipidemia, ectopic fat deposition, inflammation, mitochondrial and endoplasmic reticulum commitments, among others. However, how pathogenic variants in adipogenesis-related genes modulate DNA repair in some classical congenital lipodystrophies has not been elucidated. In the same way, no data is clarifying how pathogenic variants in DNA repair genes result in sWAT loss in different types of progeroid lipodystrophies. This review will concentrate on the main molecular findings to understand the link between DNA damage/repair and adipogenesis in human and animal models of congenital lipodystrophies. We will focus on classical and progeroid congenital lipodystrophies directly or indirectly related to DNA repair pathways, highlighting the role of DNA repair-related proteins in maintaining sWAT homeostasis. | DNA repair-related genes and adipogenesis: Lessons from congenital lipodystrophies
Classical and progeroid congenital lipodystrophies are a collection of rare diseases displaying a large genetic heterogeneity. They occur due to pathogenic variants in genes associated with adipogenesis, DNA repair pathways, and genome stability. Subjects with lipodystrophy exhibit an impairment in the homeostasis of subcutaneous white adipose tissue (sWAT), resulting in low leptin and adiponectin levels, insulin resistance (IR), diabetes, dyslipidemia, ectopic fat deposition, inflammation, mitochondrial and endoplasmic reticulum commitments, among others. However, how pathogenic variants in adipogenesis-related genes modulate DNA repair in some classical congenital lipodystrophies has not been elucidated. In the same way, no data is clarifying how pathogenic variants in DNA repair genes result in sWAT loss in different types of progeroid lipodystrophies. This review will concentrate on the main molecular findings to understand the link between DNA damage/repair and adipogenesis in human and animal models of congenital lipodystrophies. We will focus on classical and progeroid congenital lipodystrophies directly or indirectly related to DNA repair pathways, highlighting the role of DNA repair-related proteins in maintaining sWAT homeostasis.
Nuclear and mitochondrial DNA are continuously exposed to damage induced by endogenous and exogenous sources (Evans et al., 2004; Bauer et al., 2015). Endogenous sources of DNA damage include reactive oxygen species (ROS) generated during normal cell metabolism, mainly by the mitochondria (Balaban et al., 2005), but also by the endoplasmic reticulum (ER), peroxisomes, and cell membrane (Bhattacharyya et al., 2014). Furthermore, exogenous DNA damage sources mainly include ultraviolet (UV) radiation, ionizing radiation (IRa), and alkylating agents (Evans et al., 2004). Cells have developed several DNA repair pathways to defend the genome against different types of damage, including the most deleterious lesions, such as oxidized DNA lesions, single strand breaks (SSBs), and double-strand breaks (DSBs) (Limpose et al., 2017). DNA repair pathways protect from frequent lesions resulting in DNA breaks. Oxidized DNA lesions and SSBs are usually repaired by the base excision repair (BER); DSBs are repaired by homologous recombination (HR) and non-homologous end joining (NHEJ). Although nucleotide excision repair (NER) is mainly responsible for repairing bulky DNA-distorting lesions induced by UV radiation, this pathway is also involved with the repair of oxidized DNA lesions together with BER (Dianov et al., 1999; Stevnsner et al., 2002; Tuo et al., 2002; D’Errico et al., 2006; Stevnsner et al., 2008; de Melo et al., 2016; Kumar et al., 2020). There are two NER sub-pathways, global genomic-NER (GG-NER) and transcription-coupled NER (TC-NER), which differ only in the initial step of DNA lesion recognition. Failure to repair DNA damage or misrepaired DNA lesions leads to genomic instability and changes in cellular homeostasis, resulting in cancer (Menck and Munford, 2014; Jeggo et al., 2016), neurodegenerative diseases (Weissman et al., 2007; Krasikova et al., 2021), aging (Schumacher et al., 2021), and progeroid diseases with loss of subcutaneous white adipose tissue (sWAT) (López-Otín et al., 2013; Araújo-Vilar and Santini, 2019; Araújo de Melo Campos et al., 2021). For example, in the progeroid Cockayne Syndrome (CS), defects in NER may lead to premature aging with loss of sWAT (László and Simon, 1986; Nance and Berry, 1992; Kamenisch et al., 2010). Aging is a process that disturbs most living cells and is related to the accretion of damage to the molecules, genomic instability, telomere dysfunction, heterochromatin loss, and loss of sWAT. Other hallmarks of aging include mitochondrial dysfunction, senescence, inflammation, deregulated nutrient sensing, and metabolic defects. Altogether, these changes lead to a failure in stem cell function, reducing their capabilities to regenerate tissue (Schosserer et al., 2018; Palmer et al., 2019; Smith et al., 2021). Over the past decade, a renewed interest in adipose tissue functions and genomic integrity has emerged. Accumulation of senescent white adipocytes occurs during aging, which is associated with hypertrophy of adipocytes, dyslipidemia, and IR (Unger, 2005; Smith et al., 2021; Von Bank et al., 2021). Extreme decrease of sWAT and senescence of adipocytes are hallmarks of an advanced age (Tchkonia et al., 2010; Liu et al., 2018). During aging, the reduced capacity of sWAT to store lipids may contribute to metabolic complications due to ectopic deposition of lipids (lipotoxicity) (Von Bank, et al., 2021). The mechanisms involved in adipose tissue aging were recently reviewed (Ou et al., 2022). The main hallmarks of senescent cells are a secretory phenotype, cell cycle arrest, and activation of a DNA damage response (DDR), with phosphorylated histone H2AX (γ-H2AX) and p53 expression as markers of senescent cells (Tchkonia et al., 2010; Liu et al., 2018). Further, a lower expression of the H2AX gene was found in sWAT of obese individuals (Rohde et al., 2020). However, the link between senescence, DNA damage, and loss of sWAT in congenital lipodystrophies is poorly understood. This review discusses recent molecular findings in the study of congenital lipodystrophies and the role of DNA repair in maintaining adipose tissue’s functions. We focused on human and animal models of congenital lipodystrophies to unravel the link between DNA damage/repair and sWAT homeostasis.
White adipose tissue (WAT) has been extensively studied due to the association between increased visceral WAT (vWAT) and metabolic and cardiovascular disturbs (Tchkonia et al., 2010; Item and Konrad, 2012). On the contrary, studies concerning sWAT and brown adipose tissue (BAT) have shown their beneficial effects in improving metabolism and insulin sensitivity. These findings highlight that distinct WAT depots have different roles related to metabolic health. While vWAT is found around visceral organs, such as gonadal, retroperitoneal, perirenal, omental, and mesenteric localization, depots of sWAT have restricted localization and functions, being found mainly under the skin (metabolically active sWAT) and in palms and soles (mechanic sWAT) (Wajchenberg, 2000; Choe et al., 2016; Schosserer et al., 2018). The primary interest of studies concerning WAT physiology was mainly directed to its role as an energy storage tissue. However, over the last years, WAT research has gained a lot of attention since WAT has an essential hormonal function and undergoes significant changes during aging (Ou et al., 2022). One of the proposed aging hallmarks is dysfunctional adipose tissue and the consequent metabolic defects, including a reduction in the levels of somatotrophic axis hormones, such as insulin-like growth factor 1 (IGF1) and growth hormone (GH), as well as steroid hormones (Carrero et al., 2016). Indeed, changes in redox homeostasis have been found in metabolic syndrome, obesity, type 2 diabetes mellitus (DM), and lipodystrophies. During aging, WAT suffers redistribution, BAT depots decrease, and adipose progenitor and stem cells (APSCs) decline. Further, dysfunctional smaller cells similar to adipocytes increase in aged WAT, which show reduced insulin sensitivity than fully differentiated adipocytes (Kirkland et al., 2002). Altogether, these age-related changes in adipose tissue result in decreased sWAT and increased vWAT depots, compromising body function. The pathophysiology of adipose tissue in lipodystrophies was remarkably discussed in recent reviews (Zammouri et al., 2021; Lim et al., 2021; Le Lay et al., 2022).
Genetic lipodystrophies are a group of rare, heterogeneous metabolic diseases caused by a lack of sWAT, which can be total or partial (Garg, 2011; Brown et al., 2016; Zammouri et al., 2021; Araújo de Melo Campos et al., 2021). As in aging, congenital lipodystrophies have been associated with adipose tissue redistribution, sWAT loss, increased vWAT, and ectopic fat deposition (Garg and Agarwal, 2009; Zammouri et al., 2021). The nearly complete lack of body fat at birth results in Congenital Generalized Lipodystrophy (CGL), the most severe form of lipodystrophy. Instead, Familial Partial Lipodystrophy (FPLD) is characterized by a deficiency of sWAT in the limbs and gluteus that emerges during childhood or puberty, associated with fatty tissue deposition in specific body regions, such as the face, neck, and intra-abdominal area. Progeroid syndromes are also a group of rare congenital diseases characterized by clinical features including aging, hair loss, cardiovascular commitments, comorbidities affecting the skeleton and muscle, lipodystrophy, metabolic changes, and others (Van Der Pluijm et al., 2007; Turaga et al., 2009; Carrero et al., 2016). Since generalized or partial lipodystrophy is an important clinical finding associated with numerous progeroid diseases, treatment strategies have been developed to fight metabolic and mitochondrial commitments found in these syndromes (Carrero et al., 2016; Zammouri et al., 2021). In this review, we will focus only on classical and progeroid lipodystrophies associated with senescence, DNA damage accumulation, and metabolic dysfunction, three hallmarks of aging (López-Otín et al., 2013). Table 1 shows the main classical and progeroid congenital syndromes.
The lack of sWAT in CGL causes a decrease in leptin levels and alters food intake, intensifying the appetite (Badman & Flier, 2007; Rodríguez et al., 2016). The blood circulating lipids result in hypertriglyceridemia (HTG), and their accumulation in ectopic sites, such as in the liver and skeletal muscle, can result in hepatic steatosis and weakness of respiratory muscle strength, respectively (Debray et al., 2013; Dantas De Medeiros et al., 2018; Araújo de Melo Campos et al., 2021). Severe IR causes hypertension, HTG, and difficulty in controlling diabetes. Liver fat deposition can result in cirrhosis. These comorbidities could explain the severity of CGL and its early mortality (Lima et al., 2018b). The most common pathogenic variants associated with CGLs are in AGPAT2 and BSCL2 genes, related to types 1 and 2 (CGL1 and CGL2), respectively (Magré et al., 2001; Agarwal et al., 2002; Craveiro Sarmento et al., 2019). Although CGL1 and CGL2 have similar metabolic abnormalities, the sWAT loss is less severe in CGL1 individuals, which have more mechanical sWAT, while CGL2 individuals display a significant reduction of both metabolically active and mechanic sWAT (Garg et al., 1992; Agarwal et al., 2003b; Simha and Garg, 2003). Regarding the AGPAT2 gene, it codifies to the 1-acylglycerol-3-phosphate o-acyltransferase (1-AGPAT 2) enzyme, which is associated with the synthesis of triacylglycerol (TG) and phospholipids in the ER (Agarwal and Garg, 2003). Recessive pathogenic variants in the BSCL2 gene, which codifies to the ER membrane-localized seipin, are the genetic cause of CGL2 (Magré et al., 2001). This protein acts to regulate the TG transport from the ER to lipid droplets (LDs) (Salo et al., 2019), converting nascent to mature LDs (Wang et al., 2016) and regulating ER-LDs contacts and cargo delivery (Salo et al., 2016). Seipin has essential functions related to adipose tissue homeostasis, such as coordinating 1-AGPAT2 function (Sim et al., 2020) and controlling Ca2+ (calcium) import and adipocyte metabolism at ER-mitochondria sites (Combot et al., 2022). Type 3 CGL (CGL3) occurs due to homozygous pathogenic variants in the CAV1 gene that codifies to caveolin-1 (Kim et al., 2008), whereas type 4 CGL (CGL4) occurs due to pathogenic variants in the CAVIN1 gene, which codifies to the cavin-1 protein (Hayashi et al., 2009; Rajab et al., 2010). Both cavin-1 and caveolin-1 are present in caveolae, which are cave-like structures located at the plasma membrane in most cells, mainly adipocytes. Caveolae are involved in cellular processes, such as cell metabolism, cholesterol homeostasis, cell proliferation, and senescence (Parton, 2018). However, the number of pathogenic variants in both genes is scarce relative to CGL1 and CGL2. Table 1 contains a summary of the the molecular basis and sWAT physiology of CGL syndromes. At the morphological level, CGL subjects present a typical phenotype, revealing acromegalic facies, prominent musculature, prognathism, phlebomegaly (prominent veins), umbilical protrusion, acanthosis nigricans, acrochordons, hirsutism, bone cysts, and others (Garg, 2000; Maldergem et al., 2002; Agarwal et al., 2003b; Garg and Agarwal, 2009; Vigouroux et al., 2011; Lima et al., 2016; Lima et al., 2017; Lima et al., 2018a). At metabolic and physiological levels, CGL subjects present dyslipidemia, hyperinsulinemia, IR, DM, low levels of leptin and adiponectin, decreased levels of high-density lipoprotein cholesterol (HDL-c), hepatosplenomegaly, and hypertrophic cardiomyopathy (Faria et al., 2009; Lima et al., 2016; de Azevedo Medeiros et al., 2017; Ponte et al., 2018; Dantas De Medeiros et al., 2018).
Concerning the FPLDs, eight subtypes were described, and the primary molecular causes of these heterogeneous diseases are genes related to the nuclear envelope and adipocyte homeostasis, such as LMNA and PPARγ (Patni and Garg, 2015; Araújo-Vilar and Santini, 2019; Fernández-Pombo et al., 2021). Type 1 FPLD (FPLD1, also named Köbberling syndrome) is probably a multigenic form of lipodystrophy (Patni and Garg, 2015; Araújo-Vilar and Santini, 2019). The most frequent FPLD is the Dunnigan syndrome, also referred to as type 2 FPLD (FPLD2), which occurs due to pathogenic variants in the LMNA gene. This gene encodes lamin-A and lamin-C (besides lamins CΔ10 and C2) which play a significant function in maintaining the stability of the cellular nucleus by physically supporting nuclear envelope components (Gonzalo et al., 2017). Over 400 pathogenic variants were described in the LMNA gene. In addition to FPLD2, they are related to more than a dozen degenerative diseases, such as neuropathies, muscular dystrophies, and premature aging (Broers et al., 2006; Bertrand et al., 2011; Gonzalo and Kreienkamp, 2015). Recent reviews discussed the association between LMNA variants and several diseases (Ho and Hegele, 2019; Lazarte and Hegele 2021). However, how different LMNA pathogenic variants result in a plethora of diseases has yet to be unraveled. FPLD2 phenotype was initially described in 1974 by Dunnigan and first associated with the LMNA gene in 1998 by Peters et al. (Dunnigan et al., 1974; Peters et al., 1998). This disease is characterized by loss of sWAT in the extremities and trunk, sparing the face and neck at puberty. Lamins A/C, encoded by the LMNA gene, are nuclear proteins, and specific pathogenic variants may lead to nuclear function disruption, resulting in premature adipocyte death (Garg, 2011). FPLD2 subjects show loss of sWAT mainly in the axial skeleton, such as in limbs, trunk, hips, and gluteus, but not in the appendicular skeleton (Garg et al., 2001; Chan et al., 2016). FPLD2 metabolic disturbances include HTG, low HDL-c levels, IR, hepatic steatosis, pancreatitis, and a high probability of developing cardiovascular diseases (Araújo-Vilar and Santini, 2019; Lazarte et al., 2021). Type 3 (FPLD3) is caused by pathogenic variants in the PPARγ gene. In 1999, three subjects were reported with severe IR harboring two different heterozygous pathogenic variants in the ligand-binding domain of peroxisome proliferator-activated receptor type γ (PPARγ) (Barroso et al., 1999). Later, these variants were associated with FPLD3 (Savage et al., 2003). As PPARγ is a critical transcription factor for adipogenesis, its dominant pathogenic variants may impair adipocyte differentiation (Garg, 2011). This type is characterized by loss of sWAT in the extremities, especially in distal regions (Araújo-Vilar and Santini 2019). Type 4 FPLD (FPLD4) was described and associated with two distinct heterozygous frameshift pathogenic variants in the PLIN1 gene (Gandotra et al., 2011). The PLIN1 gene encodes perilipin-1, an integral component of LDs, playing an essential role in lipid storage and hormone-regulated lipolysis (Garg, 2011). In this type, lipoatrophy is mainly evident in the gluteal regions and lower limbs, although the loss of subcutaneous fat has also been observed in the trunk and upper limbs. Type 5 FPLD (FPLD5) is caused by a homozygous truncating pathogenic variant in the CIDEC gene that was first reported in 2009 (Rubio-Cabezas et al., 2009). The clinical hallmarks are loss of sWAT in the lower limbs, prominent muscle mass, IR, diabetes, and decreased LD size in adipocytes. The CIDEC gene encodes the Cell Death Inducing DFFA Like Effector C (CIDEC) protein that is associated with LDs, inhibiting lipolysis and promoting the formation of unilocular LDs in adipocytes (Garg, 2011). Type 6 FPLD (FPLD6) is triggered by a homozygous pathogenic variant in the LIPE gene. The first to describe this disease and its association with this gene were Albert et al. (2014). The main clinical manifestations of this disease are progressive loss of sWAT in the legs that correlate with abnormal fat distribution, including fat accumulation in the neck, face, axilla, shoulders, back, abdomen, and pubic region. Furthermore, in some cases, myopathy, diabetes, HTG, low HDL-c, and hepatic steatosis may be observed (Zolotov et al., 2017). Pathogenic variants in the CAV1 gene, first related to CGL3, were also found in type 7 FPLD (FPLD7) individuals (Cao et al., 2008). However, heterozygous pathogenic variants in this gene are responsible for causing FPLD7 (Cao et al., 2008). This disease is characterized by loss of sWAT in different regions of the body, accompanied by metabolic complications such as IR, lipid abnormalities, and in some cases, cataracts and muscle spasticity (Garg et al., 2015). More studies are required to unravel the role of distinct CAV1 pathogenic variants in different types of congenital lipodystrophies, such as CGL3, FPLD7, and the neonatal onset of generalized lipodystrophy (Cao et al., 2008; Schrauwen et al., 2015; Garg et al., 2015). Table 1 summarizes the molecular basis and sWAT physiology of FPLD syndromes.
Monogenic, premature aging diseases are heterogeneous syndromes and present variable severity and overlapping phenotypes, making it difficult for the correct clinical diagnosis (Carrero et al., 2016). Molecular investigations are essential for deciphering the genetic causes of progeroid overlapping diseases. The hallmarks of progeroid syndromes include increased DNA damage accumulation, defective DNA repair, telomere dysfunction, aberrant nuclear architecture and chromatin structure, impaired cell cycle, senescence, disrupted epigenetics regulation, and lack of sWAT (Agarwal and Garg, 2006; Carrero et al., 2016; Niedernhofer et al., 2018). Cockayne Syndrome Cockayne Syndrome (CS) is a progressive rare autosomal recessive disorder, first described through the clinical study of two patients (Cockayne, 1936). This disease results in postnatal growth failure, and progressive neurologic dysfunction primarily due to demyelination, and photosensitivity (Nance and Berry, 1992). CS may manifest as delayed psychomotor development, behavioral and intellectual deterioration, microcephaly, increased or decreased muscle tone and reflexes, gait ataxia, tremor, incoordination, dysarthric speech, pigmentary degeneration of the retina, cataracts, optic atrophy or optic disk pallor, sensorineural hearing loss, dental complications, kidney complications, hyperinsulinemia or abnormal glucose tolerance, elevated serum cholesterol or lipoprotein levels, and very low levels of HDL-c (Nance and Berry, 1992). The aged appearance may come from the expression of thin hair, diminished subcutaneous tissue, scaly skin, erythematous dermatitis on the dorsum of the hands and wrists, on the legs, and on the face and ears, worsened after exposure to the sun, small faces with sunken eyes and prominent superior maxillae (Cockayne, 1936). Xeroderma Pigmentosum Xeroderma Pigmentosum (XP) was first documented in 1884 when three affected patients were clinically studied, presenting freckle-like pigment spots which appeared simultaneously upon the face, neck, back of forearms, hands, upper arms, and legs below the knees (Crocker, 1884). Later, other studies showed that such cutaneous symptoms had a median age of onset of between one and two years, and about forty-five percent of the patients had basal cell carcinoma or squamous cell carcinoma of the skin. Many of them also presented neurologic abnormalities, including progressive mental deterioration, hyporeflexia or areflexia, and progressive deafness, associated with dwarfism and immature sexual development (Cleaver, 1968; Kraemer et al., 1987). Next, James Cleaver discovered that fibroblasts obtained from XP patients displayed defective DNA repair after ultraviolet UV exposure (Cleaver, 1968). This condition has at least eight genetic groups, types A to G and a variant, which were identified through genetic complementation analysis (Tanaka, 1993). Cells from patients with the hereditary disease XP were expected to carry pathogenic variants in DNA repair genes. Their expression was either absent or much reduced compared to normal fibroblasts (Cleaver, 1968). This disorder presents over a 1,000-fold increased risk of skin cancer and a 10-fold increased risk of other tumors, along with progeroid symptoms. These symptoms were found in an XP patient, including an aged appearance, weight loss, epidermal atrophy, visual and hearing loss, ataxia, cerebral atrophy, hypertension, liver dysfunction, anemia, osteopenia, kyphosis, sarcopenia, and renal insufficiency (Niedernhofer et al., 2006). Néstor-Guillermo Progeria Syndrome Néstor-Guillermo Progeria Syndrome (NGPS) is a chronic progeroid disease characterized by aging phenotypes, including growth retardation, thin limbs, and loss of sWAT. NGPS is caused by a homozygous pathogenic variant in the BANF1 gene (c.34G>C; p.A12T), that encodes BANF1/BAF1 (barrier-to-autointegration factor 1) (Puente et al., 2011). Two unrelated Spanish families were clinically investigated by Puente et al. (2011). Both had the c.34G>A [p.Ala12Thr] pathogenic variant in the BANF1 gene. Skin fibroblasts from these patients exhibited deficient BANF1 levels and profound nuclear abnormalities, including blebs and aberrations. Concurrently, transfected mutant fibroblasts with an expression vector encoding an EGFP-BAF fusion protein, and confocal microscopy analysis, revealed that ectopic expression of EGFP-BAF in these progeroid fibroblasts rescued the nuclear abnormalities, confirming the causal role of the BAF p.Ala12Thr pathogenic variant (Puente et al., 2011). Later in the same year, Cabanillas et al. (2011) published a detailed clinical report of the two affected patients from the two unrelated families previously described. Affected patients showed partial phenocopy of Hutchinson Gilford Progeria Syndrome (HGPS) and Mandibuloacral dysplasia (MAD) but without cardiovascular alterations and metabolic abnormalities. They presented a collection of clinical outcomes that suggested a new progeroid syndrome. Such manifestations included: very severe osteolysis with intense bone resorption, a long lifespan relative to HGPS and MAD, presence of eyebrows and eyelashes, and persistence of scalp hair. They also observed a generalized loss of sWAT over the limbs and trophic facial subcutaneous fat pad, abdomen, neck, and head, and dry and atrophic skin with small light-brown spots over the thorax, scalp, and limbs. Low levels of 25-OH-vitamin D and leptin were also seen (Cabanillas et al., 2011; Puente et al., 2011). Werner and Bloom Syndromes Werner (WS) and Bloom (BS) syndromes are rare recessive autosomal diseases characterized by clinical features of premature aging that are caused by loss-of-function pathogenic variants in the WRN (RECQL2) and BLM (RECQL3) genes, respectively (Ellis and German, 1996; Yu et al., 1996; Hickson, 2003). WRN (WRN RecQ Like Helicase) and BLM (BLM RecQ Like Helicase) are ubiquitously expressed RECQ helicases that play major roles in a wide variety of DNA repair processes required for genomic integrity maintenance. WS was first described by Otto Werner in 1904, who presented the clinical WS phenotype as a “caricature of aging” (Werner 1985). WS patients exhibit metabolic complications including IR, DM, dyslipidemia, and fatty liver, as well as cataracts, cancer, and premature aging. Atherosclerosis is more frequent from the third decade onwards. At a molecular level, WS cells display a high rate of spontaneous mutations and karyotypic abnormalities, in addition to aberrant recombination, telomere defects, and hypersensitivity to DNA damage and/or cellular stress (Turaga et al., 2009). WS patients develop normally until the second decade of life, and the first clinical sign is the lack of peak pubertal growth. Between 20 and 30 years of age, patients begin to suffer from skin atrophy, gray hair, and hair loss. Soft tissue calcification is a feature often associated with ulcerations around the ankles (and occasionally elbows) that eventually may require lower limb amputation (Takemoto et al., 2013). Other complications include type 2 DM, osteoporosis, bilateral ocular cataract, premature and severe forms of arteriosclerosis, peripheral neuropathy, and multiple cancers mainly perceived in middle age (Lauper et al., 2013). These patients generally present a median age of death around 54 years, typically due to cancer or myocardial infarction (Goto, 1997; Goto et al., 2013; Martin et al., 2021). WRN protein has exonuclease and helicase activities that are important for genome integrity maintenance. This protein interacts physically and functionally with enzymes that play central roles in DNA replication and repair. It is remarkable that replication and recombination functions also appear to underlie the telomeres maintenance by RecQ helicases (Turaga et al., 2009). BS, also referred to as congenital telangiectatic erythema, was first described in 1954 (Bloom, 1954). This progeroid syndrome is caused by pathogenic variants in the BLM gene that results in errors in the DNA replication process, and a pronounced number of chromosomal breaks and rearrangements, leading to the symptoms and clinical feature of BS (Bloom, 1954; Hickson, 2003). BS patients generally demonstrate postnatal growth retardation, facial butterfly rash, often after exposure to sunlight, defective cellular and humoral immunity, and an increased risk of cancer, besides a high prevalence of DM, dyslipidemia, and hepatic steatosis. Both WS and BS syndromes show metabolically phenocopies of lipodystrophy (reduction in sWAT) and obesity (Epstein et al., 1966; Diaz et al., 2006; Goh et al., 2020). Hutchinson Gilford Progeria Syndrome HGPS is considered one of the most severe laminopathies, being included in the group of premature aging degenerative diseases. Patients live for an average of just 14.6 years, dying primarily due to myocardial infarction or strokes (Gordon et al., 2014). HGPS was first described in 1886 by the British physician Jonathan Hutchinson and, later, by Hastings Gilford in 1904 (Hutchinson, 1886; McKusick, 2005). The main clinical manifestations of HGPS patients are sWAT loss, alopecia, Ca2+ dysfunction, vascular stiffening, delayed dentition, heart infarction, and progressive arteriosclerosis (Goldman et al., 2004; Prokocimer et al., 2013). Molecularly, HGPS patient cells have nuclear shape abnormalities, telomere shortening, genomic instability, alterations in epigenetic regulation and gene expression, mitochondrial dysfunction, and premature senescence. HGPS occurs due to the heterozygous silent pathogenic variant c.G608G in the LMNA gene (Eriksson et al., 2003; De Sandre-Giovannoli et al., 2003). LMNA encodes the prelamin-A, which undergoes post-translational processing, leading to transient production of different intermediates, including farnesylated prelamin-A and carboxymethylated prelamin-A (Lattanzi et al., 2014). The zinc metalloproteinase STE24 homolog (ZMPSTE24) cleaves the prelamin-A in two independent steps: the first is the cleavage of the last three amino acids in the C-terminal region of farnesylated prelamin-A. This cleavage can also be performed by Ras converting CAAX endopeptidase 1 (RCE1). The second cleavage of farnesylated and carboxymethylated prelamin-A occurs at the leucine 647 (L647) and results in the removal of the last fifteen amino acids, producing the mature, unfarnesylated lamin-A (Lattanzi et al., 2014). The pathogenic variant c.G608G in the LMNA gene leads to the loss of the recognition site for the second cleavage of the farnesylated prelamin-A by ZMPSTE24 (Eriksson et al., 2003; De Sandre-Giovannoli et al., 2003). This change results in the accumulation of a permanently farnesylated and carboxymethylated dominant protein, referred to as progerin, disrupting the nuclear envelope (Broers et al., 2006; Bertrand et al., 2011; Bidault et al., 2020; Saxena and Kumar, 2020). Furthermore, the accumulation of farnesylated prelamin-A is related to nuclear enlargement, heterochromatin loss, euchromatin dispersion, and increased ROS production (Richards et al., 2011). Type A Mandibuloacral Dysplasia with Lipodystrophy Type A Mandibuloacral Dysplasia with Lipodystrophy (MADA) is a rare autosomal recessive disease in which the patients commonly present slow and progressive osteolysis of the mandible, terminal phalanges, and clavicles, resulting in mandibular hypoplasia, dental crowding, and clavicular resorption, as well as skin abnormalities, acanthosis nigricans, and partial lipodystrophy. However, there is an absence of neurodegeneration. This condition is associated with accelerated aging and is usually identified after 4 or 5 years after birth (Novelli et al., 2002). MADA patients express a partial lipodystrophy pattern of body fat distribution with degeneration of sWAT in the torso and limbs and accumulation in the face, neck, and trunks (Novelli et al., 2002). This syndrome may be associated with clinical features of metabolic syndromes, including IR, which was evidenced in the clinical study of three patients with MAD (Freidenberg et al., 1992), impaired glucose tolerance, DM, and lack of breast development with regular or irregular menstrual periods in female patients (Cenni et al., 2018). This disorder is caused by the accumulation of prelamin-A in MADA cells, leading to the restraint of cellular differentiation due to the impaired import of transcription factors required for adipogenic gene activation or stress response (Cenni et al., 2018). MAD was first reported by Young et al. (1971). Since then, other authors studied different cases of MAD in patients, such as Zina et al. (1981), Pallotta and Morgese (1984), and Tenconi et al. (1986), although the cause was still unknown. The official association between MADA and the LMNA gene was published in 2002, through the clinical and genetic investigation of five consanguineous Italian families, whose skin fibroblasts showed abnormal lamin nuclei (Novelli et al., 2002). Pathogenic variants in the LMNA gene, such as p.Arg471Cys, p.Arg527Cys, p.Arg527Leu, p.Arg527His, p.Ala529THR, p.Ala529Val, and p.Met540Ile (Marcelot et al., 2020), cause the accumulation of prelamin A (non farnesylated) to toxic levels, along with the mutated prelamin A (farnesylated), affecting the whole organization of the nuclear envelope. The most common pathogenic variant responsible for the MADA phenotype is the homozygous missense substitution of c.1580G > A mapping in the exon 9 of the LMNA gene, resulting in the p.Arg527His mutated protein. These variants in the LMNA gene cause loss of interaction between lamin-A and other proteins, impacting stress recovery mechanisms in MADA cells, which means that repeated stress stimuli and failure to properly manage this condition led to senescence. These cells show nuclear dysmorphism, loss of peripheral heterochromatin, and nuclear lamina thickening (Cenni et al., 2018). Type B Mandibuloacral Dysplasia with Lipodystrophy Type B Mandibuloacral Dysplasia with Lipodystrophy (MADB) is a rare autosomal recessive premature aging disease (Agarwal et al., 2003a). MADB is characterized by IR, metabolic comorbidities, atrophic skin, brittle hair, generalized loss of sWAT, skeletal abnormalities such as mandibular and clavicular hypoplasia, and acro-osteolysis of the distal phalanges (Hitzert et al., 2019). Although MADB and MADA have many similarities, MADB individuals develop early skeletal abnormalities (Agarwal et al., 2003a). ZMPSTE24 pathogenic variants are responsible for many different diseases, depending on the degree of prelamin-A processing impairment (Shackleton et al., 2005). MADB is caused by compound heterozygous or homozygous pathogenic variants in the ZMPSTE24 gene, resulting in reduced activity of the metalloprotease ZMPSTE24. Compound heterozygous variants in the ZMPSTE24 gene, such as p.Phe361fsX379/p.Trp340Arg (Agarwal et al., 2003a), p.Phe361fsX379/p.Asn265Ser (Shackleton et al., 2005; Agarwal et al., 2006), p.Gln41X/p.Pro248Leu (Miyoshi et al., 2008), p.Tyr70fs/p.Asn265Ser (Cunningham et al., 2010), and p.Pro248Leu/p.Trp450X (Ahmad et al., 2010), as well as the homozygous variants p.Leu94Pro (Yaou et al., 2011) and p.Tyr399Cys (Haye et al., 2016), can partially or totally affect the functions of the metalloprotease ZMPSTE24, resulting in the accumulation of farnesylated prelamin-A and progressive loss of sWAT. Zmpste24 -/- mice also displayed almost completed loss of sWAT due to the toxic accumulation of farnesylated prelamin-A (Bergo et al., 2002; Pendás et al., 2002). Wiedemann-Rautenstrauch Syndrome The POLR3A gene encodes the largest subunit of RNA polymerase III (Pol III), forming the catalytic core with POLR3B. Pol III is responsible for the transcription of different kinds of non-protein-coding RNAs, which regulate transcription, RNA processing, and translation (Sepehri and Hernandez 1997; Werner et al., 2009; Wu et al., 2021). This protein also acts in the proper function of the nucleolus, including ribosome assembly by enhancing 5S rRNA synthesis and protein translation, determining the metabolic state of the cell (Tiku and Antebi 2018; Báez-Becerra et al., 2020). Wiedemann-Rautenstrauch Syndrome (WRS) was first studied in 1977 (Rautenstrauch et al., 1977) and in 1979 (Wiedemann, 1979), both studies through clinical reports of patients with a progeroid syndrome, utilizing their lymphocytes and cultured skin fibroblasts. The relation between WRS and pathogenic variants in the POLR3A gene was confirmed by investigating DNA and RNA samples and fibroblast cultures of two affected Bulgarian families (Azmanov et al., 2016), showing that the POLR3A gene is the primary locus for the WRS phenotype. Since then, studies have presented POLR3A biallelic variants that alter splicing and/or truncate translation and are associated with WRS, such as c.1909þ18G>A and c.2617C>T (Jay et al., 2016), c.3337-5T>A, c.3337-11T>C, c.490+1G>A, c.2005C>T, c.760C>T, c.1572+1G>A, c.2617-1G>A, c.3G>T and c.*18C>T (Wambach et al., 2018), all found in clinical and genetic analysis of WRS patients. Accordingly, these POLR3A alterations are the cause of the WRS progeroid disease. WRS is sporadic and heterogeneous, characterized by intrauterine growth restriction (IUGR), poor postnatal weight gain, characteristic facial features, pseudohydrocephalus, generalized lipodystrophy, with an almost complete lack of subcutaneous fat and possible paradoxical caudal fat accumulation, premature alopecia, neonatal teeth, and teeth abnormalities (Rautenstrauch et al., 1977; Wiedemann, 1979). The progressive generalized lipodystrophy manifests with local fatty tissue accumulations, and cachectic appearance (Paolacci et al., 2017; Lessel and Kubisch, 2019). Ruijs-Aalfs Syndrome The SPRTN gene encodes to Spartan protein, a DNA-dependent metalloprotease associated with the replication machinery that repairs DNA-protein crosslinks (DPCs) through the SprT protease domain (Maskey et al., 2014, 2017). DPCs derive from proteins covalently and irreversibly bound to DNA, such as Topoisomerase 1 (Top1), and the SPRTN (SprT-Like N-Terminal Domain) proteolytic activity, which upon DNA and ubiquitin-binding and promotes cleavage of DPC substrates and itself (Lopez-Mosqueda et al., 2016; Li et al., 2019). Spartan malfunction, as a consequence of pathogenic variants such as c.721delA and c.350A>G (Lessel et al., 2014), is responsible for replication stress, which has been suggested to cause DSBs, translocation mosaicism, and genomic instability. Thus, pathogenic variants in the SPRTN gene have been linked to cancer and aging, more specifically to the Ruijs-Aalfs syndrome (RJALS), an autosomal recessive disorder firstly described by Ruijs et al. (2003). RJALS individuals display genome instability, short stature, cataract, progeria, low body weight, micrognathia, triangular face, muscular atrophy, lipodystrophy, and early-onset hepatocellular carcinoma (Ruijs et al., 2003; Lessel et al., 2014). The first association between SPRTN pathogenic variants, progeroid syndromes, and liver tumors was made in 2014, using Sprtn hypomorphic mice (Maskey et al., 2014) and in primary skin fibroblasts, liver tumor biopsies, and lymphoblastoid cells (LCLs) from three progeroid patients, as well as in U2OS, and HEK293T cell lines (Lessel et al., 2014). The pathogenic variants in the SPRTN gene, such as SPRTN-∆C and SPRTN-Y117C, and defects in DPC repair were shown in 2016 (Lopez-Mosqueda et al., 2016; Stingele et al., 2016; Vaz et al., 2016).
In the last years, a plethora of molecular findings unraveling the link between DNA damage/repair and adipogenesis in human and animal models has emerged. In this section, we will highlight the main findings concerning the role of genes related to DNA repair and genomic stability in progeroid syndromes with lipodystrophy. Table 2 summarizes the main findings of this section.
The link between changes in redox homeostasis, cell cycle, and senescence was investigated in fibroblasts from FPLD2 subjects carrying the pathogenic variants p.D47Y, p.L92F, p.L387V, p.R399H, p.L421P, and p.R482W in the LMNA genes (Caron et al., 2007). These pathogenic variants result in prelamin-A accumulation, the precursor of lamin-A, which was associated with the occurrence of mitochondrial dysfunction and higher levels of cytoplasmic ROS. Disturbances in the cell cycle and premature senescence were also found (Caron et al., 2007). Oxidative stress, inflammation, senescence, and calcification were also found in vascular smooth muscle cells (VSMCs) from FPLD2 subjects harboring R482W, D47Y, and R133L LMNA pathogenic variants (Afonso et al., 2016). This study only investigated DSBs accumulation by evaluating the amount of γH2AX foci. Unrepaired DSBs accumulation was also verified in human coronary artery endothelial cells (HCAECs) transduced with adenoviral vectors containing Flag-tagged p.R482W prelamin-A cDNA It was also verified that pravastatin treatment decreased the levels of γH2AX foci (Bidault et al., 2013). Further, prelamin-A accumulation was directly associated with accumulation of DSBs in VSMCs infected with prelamin-A adenovirus (Liu et al., 2013). The group performed microarray assays and found that DNA repair pathways responsible for the removal of DSBs were downregulated, suggesting that prelamin-A accumulation amplifies the DDR against DSBs. They also verified that the miRNA-141-3p levels were increased. This microRNA negatively regulates the ZMPSTE24, a prelamin-A maturation enzyme, which was considered a significant regulator of dysfunctional VSMCs from FPLD2 subjects. Although DNA repair pathways were not assessed in detail in this work, it is reasonable to suggest that the disrupted redox homeostasis found in those subjects could induce oxidized DNA damage and contribute to the pathophysiology of FPLD2. Indeed, Maynard and co-workers investigated the mechanism by which Lmna regulates the repair of oxidized DNA damage by the BER pathway in a mice model. They performed microarray gene expression and found that Lmna -/- MEFs (mouse embryonic fibroblasts) displayed an upregulation of genes related to the BER pathway and mitochondrial genome maintenance (Maynard et al., 2019). On the contrary, genes involved with metabolic processes and oxidative stress response mediated by NFE2L2 (nuclear factor erythroid 2-like 2; also termed NRF2) were downregulated. However, the authors did not explore the downregulated genes related to the metabolic process. Furthermore, they found that Lmna -/- MEFs were sensitive to DNA damage induced by hydrogen peroxide (H2O2) and menadione compared to Lmna +/+ MEFs. Besides, the levels of 7,8-dihydro-8-oxoguanine (8-oxoG), the most abundant oxidized DNA level mainly repaired by the 8-oxoG DNA glycosylase (OGG1) from BER (Cadet et al., 2003; Krokan and Bjørås, 2013), were higher in Lmna -/- MEFs relative to Lmna +/+ MEFs after H2O2-induced DNA damage. These data indicate that this lesion is less efficiently repaired in the absence of Lmna, corroborating with results obtained by Comet assay, which revealed the repair efficiency of oxidized DNA lesions, including 8-oxodG and FapyG, was decreased in Lmna -/- MEFs relative to Lmna +/+ MEFs. After H2O2-induced DNA damage, Lmna -/- MEFs also showed lower levels of Parp-1, Lig3, and Polβ mRNA expression as well as lower protein levels of PARP-1, LIG3, and Polβ. Interestingly, Lmna is required to APE1 and Polβ activities, which were PARP-1 dependent. Lmna depletion by siRNA also led to impaired BER in U2OS cells. Taken together, although these findings are very relevant to unravel the role of LMNA in the repair of oxidized DNA lesions, a link between BER and LMNA in the context of adipose tissue was not provided.
Recent evidence revealed that accumulation of progerin causes defects in the expression and recruitment of DNA repair components, in addition to the suppression of Poly-ADP-ribose polymerase 1 (PARP-1) (Liu et al., 2011; Zhang et al., 2014). Zhang and co-workers found PARP-1 suppression in smooth muscle cells (SMCs) obtained from HGPS at protein levels and by immunofluorescence. This result was confirmed in HGPS fibroblasts carrying the pathogenic variant c.1824 C>T (p.G608G). Co-expression of PARP-1/GFP in SMCs revealed that progerin induces a mislocalization of a PARP-1 fraction to the cytosol (Zhang et al., 2014). PARP-1 usually plays a role in suppressing the NHEJ DNA repair mechanism and protecting HR (Broers et al., 2006; Bertrand et al., 2011; Patel et al., 2011; Zhang et al., 2014). Besides, most SMCs from HGPS individuals activated the error-prone NHEJ repair during S-phase, while HR was deficient during S-phase, leading to mitotic disaster and cell death (Zhang et al., 2014). These data indicate the role of progerin in regulating PARP-1 expression and NHEJ activity in SMCs from HGPS individuals. The DDR to DSBs begins with the activation of ATM (Ataxia-Telangiectasia mutated) and ATR (ATM-and Rad3-related), which play central roles in DNA repair checkpoints. ATR is activated by broad DNA damage, whereas ATM is activated by DSBs. Activated ATM and ATR phosphorylate Chk-1 (Checkpoint kinase 1) and Chk-2 (Checkpoint kinase 2), initiating the signaling cascade that leads to p53 phosphorylation (Sancar et al., 2004; Li and Zou, 2005). Liu et al compared aged HGPS fibroblasts harboring the pathogenic variant c.1824 C>T and normal BJ fibroblasts to determine whether DNA damage pathway checkpoints were persistently activated. In this study, it was observed that progeroid cells showed more frequent DSBs, and persistent activation of ATM and ATR checkpoints, which led to higher levels of phosphorylated Chk-1 and Chk-2 and, consequently, higher levels of phosphorylated p53 (Liu et al., 2006). Another study observed that although some DNA repair proteins, such as ATM, ATR, Chk1, Chk2, and p53 were activated, Rad50 and Rad51 were not recruited to the DNA damage regions (Liu et al., 2008). Furthermore, surprisingly, XPA (Xeroderma pigmentosum complementation group A), a NER protein, was present in chromatin regions where DSBs had occurred in progeroid cells (Liu et al., 2008). The same was not observed in normal BJ fibroblasts, even when DSBs in DNA was induced by camptothecin (CPT). These findings suggest that the binding of XPA in DSBs regions prevents the recruitment of repair proteins such as Rad50 and Rad51 (Liu et al., 2008). In this way of thinking, XPA depletion was performed to verify whether the recruitment of repair proteins was restored. Indeed, a partial restoration of proteins such as Rad50, Rad51, and Ku70 was observed (Liu et al., 2008). Mitochondrial dysfunction and increased levels of ROS were also found in HGPS fibroblasts (Richards et al., 2011). Accumulation of misrepaired DSBs and increased sensitivity to DNA damage agents, such as H2O2, were observed in HGPS fibroblasts. The treatment with N-acetyl cysteine (NAC), a ROS scavenger, decreased DSBs and improved cell growth (Richards et al., 2011). Besides, Kubben and co-workers found that although NFE2L2 (NRF2) protein levels did not change in HGPS fibroblasts, progerin sequesters NFE2L2 (NRF2), reducing its transcriptional activity since the sequestered NRF2 is mislocated to the nuclear periphery (Kubben et al., 2016).
To investigate the role of the LMNA R527H pathogenic variant in the cell cycle control and DDR, Alessandra di Masi and co-workers analyzed the response of MADA fibroblasts to DNA damage induced by IRa (Di Masi et al., 2008). They found high levels of chromosome aberrations in G2-irradiated MADA fibroblasts, suggesting the occurrence of misrepaired DNA and that MADA cells are more sensitive to IRa than control fibroblasts. Basal levels of phosphorylated ATM (at S1981) were higher in MADA fibroblasts. Furthermore, increased phosphorylated ATM-S1981 foci were observed in almost 70% of MADA fibroblasts after X-ray treatment, suggesting accumulated DNA damage. Besides, as phosphorylation of γ-H2AX occurs around DSBs, being considered a marker for DSBs, immunofluorescence staining with the γ-H2AX antibody was performed. MADA cells presented a higher level of γ-H2AX after IRa treatment relative to control cells (Di Masi et al., 2008). Furthermore, p53 basal levels were 2-fold higher in MADA fibroblasts compared to control, suggesting that the prelamin-A accumulation in MADA cells can determine the persistence of misrepaired DNA damage.
The ZMPSTE24 gene contribution to genomic stability and aging was also studied in models of progeroid phenotypes. Using Zmpste24 -/- MEFs, Liu and co-workers discovered that the deficiency in Zmpste24 resulted in cell cycle arrest and senescence. These cells also presented chromosomal instability and quickly accumulated DNA damage relative to controls (Liu et al., 2005). Zmpste24 -/- MEFs had high 53BP1 foci and increased protein levels of γH2AX, a marker of DSBs, and phosphorylated chk1 (p-chk1), involved with DNA damage checkpoint response. They also found similar results in fibroblasts obtained from HGPS individuals. Zmpste24 -/- MEFs also were sensitive to DNA-damage agents, such as those inducing DSBs [mitomycin (MMC), methylmethanesulfonate (MMS), CPT, and etoposide] and UV. After γ-irradiation, the number of γH2AX/53BP1 co-localized foci were delayed in Zmpste24 -/- MEFs, suggesting that 53BP1 recruitment is affected. Besides, six and twelve hours after γ-irradiation, most of the 53BP1 foci disappeared in WT MEFs and fibroblasts. On the contrary, γH2AX/53BP1 co-localization was kept in Zmpste24 -/- MEFs and HGPS fibroblasts, suggesting misrepaired DSBs. Later, they investigated whether defective DNA repair is associated with ZMPSTE24 deficiency. Using comet assay, the authors showed that Zmpste24 -/- MEFs and HGPS fibroblasts had higher tail moment relative to controls, indicating that loss of Zmpste24 and progerin compromised DNA repair. It was also suggested that DNA repair deficiency in Zmpste24 -/- MEFs and HGPS fibroblasts may be due to decreased Rad51 foci formation. In another study, Varela and co-workers found that liver and heart from Zmpste24 -/- mice displayed an upregulation of p53 target genes, such as Gadd45a, p21 (Cdkn1a), and Atf3, as well as increased levels of γ-H2AX in the liver. Zmpste24 deficiency also resulted in a senescent phenotype (Varela et al., 2005). Taken together, the authors revealed that the accumulation of farnesylated prelamin-A due to Zmpste24 deficiency results in DNA damage accumulation, and the Rad51 recruitment is defective after γ-irradiation.
Progressive loss of sWAT was observed in a model of CS mice (Brace et al., 2013). CS is characterized by neurodegeneration, growth failure, and photosensitivity (Fousteri & Mullenders, 2008; Vessoni et al., 2020). Csa -/- /Xpa -/- (CX) mice showed more severe NER progeria, including small size and progressive loss of sWAT but not BAT. These mice also presented low levels of plasm triglycerides (TGs) and glucose. Therefore, the CX mice were a good model for studying human progeria. Later, the same group revealed changes in adiposity and lipid and glucose homeostasis in the CX mice model under chronic DNA damage induction, including IRa, crosslinking agent mitomycin (MMC), and ultraviolet (UV) radiation (Brace et al., 2016). They investigated how DNA damage affects energy metabolism and found that CX mice had a loss of sWAT and perigonadal WAT, as well as a decline in mature adipocyte size without inflammatory signals (crown-like - CL structures). Fasted CX mice had low glucose, insulin, HOMA-IR (homeostasis model assessment-estimated insulin resistance), and TGs in plasma compared to control mice. Circulating leptin levels were also decreased (Brace et al., 2016). Another study also investigated the mitochondrial fatty acid oxidation (FAO) rate in these CX mice models. They found increased oxygen consumption rate (OCR), reduced respiratory exchange ratio (RER), as well as an upregulation of FAO-related genes in muscle from fasted CX mice (Brace et al., 2016). They also verified the impact of DNA damage on FAO capacity. For this, they used mouse dermal fibroblasts (MDFs) isolated from tails of WT and CX mice, preadipocytes for CX mice, and human dermal fibroblasts (HDFs) from CSA and CSB patients. They confirmed an increase in FAO under UV-C treatment for the CX and CS models, as well as that MMC and IRa at high doses promoted a similar rise in FAO in CX MDFs, as they found for UV-C. These results suggested that increased FAO was a beneficial adaptive response to genotoxic stress induced by UV-C, MMC, and IRa and revealed a link between genotoxic stress and energy metabolism related to DNA damage. Furthermore, they showed that the ATP levels were decreased after UV-C or MMC treatments in WT MDFs and HDFs, which returned to normal levels almost 90 minutes later, indicating increased energy demands after the genotoxic stress induction. Interestingly, they also verified whether the ATP-reduced levels were linked to nicotinamide adenine dinucleotide (NAD+) depletion levels. NAD+ is a vital metabolite coenzyme for crucial metabolic pathways, such as glycolysis, TCA, and OXPHOS, as well as for ADP(ribosyl)ation reactions mediated by PARP-1 activity (Fouquerel and Sobol, 2014; Hurtado-Bagès et al., 2020). They found a reduction in NAD+ levels in WT MDFs after both UV-C and MMC treatments, which is in accordance with ATP low levels. They also assessed PARP-1 activation through PAR accumulation to better understand whether the PARP-1 activity is associated with ATP and NAD+ depletion in WT and PARP-1 KO MDFs under genotoxic stress. They confirmed the occurrence of an increased PARylation in WT MDFs after two different genotoxic stresses (UV-C and MMC), but not in PARP-1 KO MDFs. In addition, they found that phosphorylated adenosine monophosphate (AMP)-activated protein kinase (pAMPK), which regulates metabolic changes due to ATP depletion, was also increased in a PARP-1 dependent manner in MDFs, and this result was confirmed in MDFs obtained from AMPK KO mice. Besides, CX mice showed low levels of NAD+ and increased levels of pAMPK in the liver. Altogether, these findings revealed that NAD+/ATP depletion and AMPK activation in cells/tissues from CX mice are dependent on PARP-1 and link different types of genotoxic stresses (UV-C, MMC, and IRa) to increased FAO. These data also reveal that CX mice are a model of chronic genotoxic stress and lipodystrophy due to congenital DNA repair deficiency. However, adiponectin, an important hormone produced by adipose tissue that activates AMPK phosphorylation and is reduced in congenital lipodystrophy (Antuna-Puente et al., 2010; Lima et al., 2016; Craveiro Sarmento et al., 2020), was not investigated in this cell model. Loss of sWAT was also observed in Csb m/m /Xpa -/- mice that mimic the human progeroid CS syndrome (Van Der Pluijm et al., 2007). These mice presented increased levels of TGs and glycogen accumulation and low serum glucose and IGF. Moreover, GH/IGF1 growth axis reduction was not due to reduced GH levels or pituitary abnormalities. Using transcriptome analysis, the authors found an upregulation of Lepr and Pparg genes that codify to the leptin receptor and peroxisome proliferator-activated receptor gamma, respectively. Furthermore, upregulation of genes associated with fatty acids synthesis and genes encoding antioxidant enzymes in the liver from Csb m/m /Xpa -/- mice were found. In contrast, genes involved in glycolysis, TCA, OXPHOS, and controlling growth (Igf1) were downregulated. A similar loss of sWAT was similarly found in Csb m/m /Xpc -/- mice. The authors also compared the Csb m/m /Xpa -/- mice model with naturally aged mice. They found that the latter also presented accumulation of glycogen and TGs, and repression of genes related to oxidative metabolism and the IGF axis (Van Der Pluijm et al., 2007). Kamenisch and co-workers revealed that the presence of CSA and CSB proteins in mitochondria are essential for protecting against loss of sWAT (Kamenisch et al., 2010). After H2O2 treatment, oxidatively stressed WT fibroblasts had detectable levels of CSA and CSB within mitochondria. Further, they detected interactions between CSA or CSB and mitochondrial OGG1 (mtOGG1) and single-stranded DNA binding protein (mtSSBP1) only in H2O2-stressed WT cells. Cells from CSA and CSB patients and sWAT from Csb m/m and Csa -/- mice showed higher levels of mutations in mtDNA that was age-dependent. Fat tissue from 130-weak-old Csb m/m mice had a higher accumulation of mtDNA mutations. They also investigated whether the reduction of sWAT in Csb m/m mice was due to a reduction in the fat cell size or number. They found that sWAT from 130-weak-old Csb m/m mice had higher levels of macrophages containing granular lipofuscin in lysosomes, a phagocytosis marker, suggesting that the loss of sWAT in Csb m/m and Csa -/- mice is mediated on the fat number (Kamenisch et al., 2010). However, the authors did not investigate the metabolic parameters nor the levels of antioxidant adipokines, such as adiponectin, in Csb m/m and Csa -/- mice. It is known that mitochondrial function is crucial for adiponectin synthesis in adipocytes (Eun et al., 2007), adiponectin is downregulated in lipodystrophies (Antuna-Puente et al., 2010), and this adipose tissue-produced hormone induces antioxidant responses through NRF2 activation (Li et al., 2015; Ren et al., 2017). However, whether adiponectin is involved with the maintenance of mtDNA homeostasis in lipodystrophies remains to be shown.
Another association between DNA repair deficiency, absence of adipose tissue, and aging was also found (Niedernhofer et al., 2006). The authors used the Ercc1 -/- mice model as an accurate model of an XPF-ERCC1 (XFE) progeroid patient. They found that Ercc1 -/- mice presented weight loss, and the primary mouse embryonic fibroblasts isolated from these mice were sensitive to oxidative stress induced by treatment with H2O2 and paraquat. They showed premature aging in several organs and had liver failure. As in the Csb m/m /Xpa -/- mice model, a transcriptomic analysis from Ercc1 -/- mice liver revealed an upregulation of genes associated with fatty acids synthesis and genes encoding antioxidant enzymes. Furthermore, Lepr and Pparg genes were upregulated, and the Adipor2 (adiponectin receptor 2) was downregulated. On the contrary, low levels of glucose and IGF were also found in this cell model. Taken together, these findings show that both models of NER progeria are associated with loss of adipose tissue homeostasis, and this can be due to the accumulation of ROS and DNA damage accumulation. This results in the downregulation of GH/IGF1 hormonal axis in Ercc1 -/- mice to moderate the metabolism, indicating that IGF1 reduction may have beneficial effects in extending lifespan in mice. However, since DNA damage accumulates, degenerative processes will occur, such as loss of sWAT, resulting in aging. CS patients have been previously reported with low levels of IGF1 serum and decreased fat deposition (László and Simon, 1986; Park et al., 1994). As observed in Csb m/m /Xpa -/- mice model, the reduction of genes related to the GH/IGF1 growth axis in Ercc1-/- mice liver was also not due to reduced GH levels or pituitary abnormalities. In the same way, Karakasilioti and co-workers provided evidence for a causal link between persistent DNA damage and the gradual appearance of progressive lipodystrophy in NER progeria (Karakasilioti et al., 2013). To increase the understanding of the role of unrepaired DNA damage in adipose tissue degeneration, they found that DNA damage signaling resulted in fat depletion due to chronic inflammation in Ercc1 -/- fat depots from mice or in adipocytes (Karakasilioti et al., 2013). These mice presented a gradual reduction of epididymal WAT (eWAT), cervical, interscapular, and sWAT depots. To distinguish primary and secondary mechanisms related to fat depletion in Ercc1-deficient mice, the authors also created aP2-Ercc1 F/- mice, which present aP2 expression mainly but not exclusively in mature adipocytes (Shan et al., 2013), while Ercc1 is later deleted. This strategy aims to verify the effect of time-dependent accumulation of DNA damage only on adult AT depots. Progressive lipodystrophy was also found in eWAT, interscapular, and sWAT from aP2-Ercc1 F/- mice, which had high TGs and low levels of adiponectin. They also had decreased interscapular BAT depots. To further understand the role of ERCC1 in WAT, the authors analyzed the transcriptome of eWAT depots and found more than 2.000 differentially expressed genes. Genes related to response to DSBs (for ex. ATM signaling), response to stress (for ex. NRF2-related oxidative stress response), nuclear receptor (for ex. PPAR), and pro-inflammatory (TNF, NFκB) signaling were upregulated. Accumulation of γ-H2AX, phosphorylated ATM (pATM), RAD51, and FANCI was observed in adipocytes from aP2-Ercc1 F/- mice. Ablation of Ercc1 also triggered a gradual accumulation of persistent DNA damage, resulting in adipocytes’ necrosis.
Barrier-to-autointegration factor 1 (BANF1) is another protein related to severe premature aging and DNA damage/repair in NGPS (Bolderson et al., 2019; Rose et al., 2021). This protein is essential for controlling the DDR against oxidative stress by regulating PARP-1 activity (Bolderson et al., 2019). The authors found that skin fibroblasts from NGPS subjects harboring the c.34 G>A (p.A12T) pathogenic variant in the BANF1 gene had decreased PARP-1 poly-ADP-ribose activity and repair of oxidized DNA lesions induced by H2O2. Biochemical experiments in HEK293T cells revealed that the mutated BANF1 protein directly inhibits PARP-1 activity by binding to its NAD+ binding domain, maintaining the cellular levels of NAD+ after DNA damage induction. They concluded that the subcellular levels of the BANF1 protein are critical to reset PARP-1 activity under oxidative stress conditions, and the accumulation of oxidized DNA damage is associated with HGPS development. Figure 1 shows the main molecular findings concerning PARP-1 activity in different cellular models of progeroid lipodystrophy (HGPS, NGPS, and CS).
A multisystem disease characterized by mandibular hypoplasia, deafness, progeroid features, and lipodystrophy (MDPL) was associated with pathogenic variants in the POLD1 gene in seven patients (Shastry et al., 2010). Two MDPL patients from this work (named 300.4 and 500.4) were also described by Shastry and co-workers (named P3 and P4) (Weedon et al., 2013) (Shastry et al., 2010). Shastry and co-workers found a progressive loss of sWAT with partial lipodystrophy in four young adults, while generalized lipodystrophy was confirmed only in older patients. Weedon and co-workers found that, although the patients presented normal body weight and appearance at birth, they had a lack of sWAT in early childhood. Loss of sWAT in adulthood was observed in almost all sites, which contrasted with a remarkable increase of vWAT, resulting in a greater ratio of vWAT to sWAT (Weedon et al., 2013). They also presented IR, fibrosis of sWAT, and increased levels of fundamental extracellular matrix (ECM) genes, such as transforming growth factor (TGF)-β (TGFB1) and fibronectin (FN1) (Weedon et al., 2013). They identified an in-frame deletion c.1812-1814delCTC (p.Ser605del) in the POLD1 gene in two patients, which affects the polymerase’s active site. Assays for measuring the polymerase and exonuclease activities revealed that the heterozygous in-frame deletion affected the polymerase activity, which was not detectable, whereas the exonuclease activity was decreased. Another study reported a novel pathogenic variant in the exonuclease domain of the POLD1 gene (p.Arg507Cys). However, they did not perform functional experiments to characterize better how the activities of POLD1 are affected. In this case, the MDPL patient also had a loss of sWAT nearly in the entire body, except for mechanical adipose tissue (Pelosini et al., 2014). Reinier and co-workers also described a patient harboring the c.1812-1814delCTC (p.Ser605del) pathogenic variant in the POLD1 gene who had severe lipodystrophy and progeroid features (Reinier et al., 2015). The exact pathogenic variant was also found in Japanese subjects for two independent groups, suggesting that c.1812-1814delCTC (p.Ser605del) is a deletion hot spot variant associated with MDPL (Okada et al., 2017; Sasaki et al., 2018). Wang and co-workers reported the same family with subjects harboring two rare progeroid diseases, WS and MDPL (Wang et al., 2018). The proband had the hot spot c.1812-1814delCTC (p.Ser605del) pathogenic variant in the POLD1 gene. He presented a progressive loss of sWAT and progeroid features that started at 18 months. His three brothers who had WS showed a heterozygous frameshift pathogenic variant in the WRN gene (c.919_923delACTGA, p.Thr307ThrfsX5) (Wang et al., 2018). Another MDPL case due to the hot spot heterozygous in-frame deletion was also described in a Chinese patient who presented progressive loss of sWAT that started at the age of seven (Yu et al., 2021). Elouej and co-workers described a new heterozygous pathogenic variant affecting the zinc finger 2 (ZNF2) domain in the POLD1 gene (c.3209 T>A; p.Ile1070Asn) (Elouej et al., 2017). The patient developed lipodystrophy and progeroid facial features. Predictions using the PredictProtein server suggested that the substitution of isoleucine by asparagine at position 1070 can disrupt the Fe-S cluster within the CysB motif from the ZNF domain. Furthermore, Ajluni and co-workers also reported a new pathogenic variant affecting the ZNF2 domain (c.3199 G>A; p.Glu1067Lys). However, in this case, the two related subjects had reduced sWAT in the extremities but not around the neck, face, and abdominal wall. They presented IR, elevated CK levels, and proteinuria. They did not show progeroid features and deafness. In addition, while the MDPL patient had a high amount of nuclear atypia and disorganization in liver biopsy samples, these changes in the nuclear envelope integrity were lower when compared to patients harboring LMNA-pathogenic variants (p.R60G, p.R482Q, and p.R349W) (Ajluni et al., 2017). Mechanistically, two independent works found that the progeroid features of two MDPL patients harboring the in-frame heterozygous deletion p.Ser605del are related to impaired DNA repair capacity (Fiorillo et al., 2018; Murdocca et al., 2021). Fiorillo and coworkers found that an MDPL patient carrying the heterozygous single codon deletion c.1812-1814delCTC (p.Ser605del) in the POLD1 gene showed type 2 diabetes, hyperinsulinemia, and IR. HDFs obtained from this patient had nuclear envelope abnormalities, intranuclear accumulation of prelamin-A, high levels of micronuclei, cellular senescence, and growth decline. The authors studied the link between MDPL and DNA damage accumulation. After cisplatin-induced DSBs, they found high levels of γH2AX foci and a DNA repair recovery delay in HDFs compared with WT HDFs (Fiorillo et al., 2018). Similar results were found in HDFs obtained from a second MDPL patient (Murdocca et al., 2021). Although all these findings ratified the role of POLD1 in adipose tissue homeostasis, our understanding of how these pathogenic variants result in cellular defects in adipose tissue is scarce, and the mechanisms that link disrupted POLD1 activity to different diseases need to be further clarified.
WS and BS have been studied as a model for deciphering adipose tissue senescence. Using CRISPR/Cas9, Goh and co-workers generated WRN -/- and BLM -/- human pluripotent stem cells (hPSCs), which were differentiated in adipocyte precursors (APs) (Goh et al., 2020). They found that WRN -/- and BLM -/- APs displayed reduced cell proliferation, shorter telomeres, and senescence. The latter was confirmed by measuring the mRNA levels of the senescent biomarkers: p16, p21, Activin A, IL-6, and IL-8. These findings suggest that preadipocyte senescence may be the cause of metabolic complications in WS and BS. In another study, Turaga and co-workers transfected human diploid fibroblasts with a siRNA against WRN mRNA, which became senescent and presented a similar gene expression profile relative to fibroblasts established from old donor patients (Turaga et al., 2009). From 660 differentially expressed genes found in the microarray analysis, 542 (82%) were downregulated, whereas 118 genes (18%) were upregulated, revealing a repression scenario in cells with lower WRN levels. Western blotting was performed for fourteen proteins and they confirmed the downregulation of: CCNB1 (Cyclin B1), CDC2 (Cyclin-dependent kinase 1), FANCD2 (Fanconi anemia complementation group D2), FANCI (Fanconi anemia complementation group I), FANCJ (Fanconi anemia complementation group J), FAS (Fas cell surface death receptor), HUWE1 (E3 ubiquitin-protein ligase), MRE11A (Meiotic Recombination 11 homolog A), KIF4A (Kinesin family member 4A), LMNA (Lamin A/C), MAPK8 (Mitogen-activated protein kinase 8), POLD1 (DNA polymerase δ subunit 1), SAFB1 (Scaffold attachment factor B1), and TOP2A (Topoisomerase II alpha). The gene set enrichment analysis revealed that the genes related to adipocyte differentiation were downregulated in WRN-knockdown fibroblasts (Turaga et al., 2009). To confirm this observation, the authors also transfected the 3T3-L1 mice preadipocytes with a siRNA against Wrn mRNA. The expression of adipogenic markers, such as C/EBPβ (CCAAT/enhancer binding protein β) and fatty acid synthase (FASN), was decreased. These data link the role of WRN and BLM proteins in the maintenance of adipose tissue homeostasis.
The POLR3A gene is crucial for cell function and metabolism. Pathogenic variants can alter its ability to interact with DNA, causing drastic changes in its transcriptional function and RNA polymerase I and II regulation. This scenario is associated with an early senescent phenotype found in primary WRS fibroblasts carrying the pathogenic variant c.3772_3773delCT (p.Leu1258Glyfs*12) in the POLR3A gene. WRS fibroblasts presented increased expression levels of the mutant POLR3A protein in the nucleoplasm, which was not expressed in control fibroblasts. Senescence was revealed by the presence of higher beta-galactosidase-positive WRS cells and increased levels of p16 protein expression. Decreased telomere length, increased DNA damage, and variations in the morphology and number of nucleolus were also seen (Báez-Becerra et al., 2020). WRS fibroblasts exhibited strong phosphorylation levels of H2X in the Ser139 (termed γH2AX) and p53 (in the Ser15) relative to control cells, which were associated with increased nuclear staining. These results indicate that WRS fibroblasts show an increase in DNA damage that can induce DDR and, consequently, a p53-mediated cell senescence. Also, a pathway of POLR3-mediated p53 regulation is likely lost upon POLR3A pathogenic variants in WRS fibroblasts. Altogether, these results revealed a link between POLR3A variants and DDR in WRS fibroblasts. The SPRTN gene and RJALS Lessel et al. (2014), proposed a clinical study of three patients with early-onset hepatocellular carcinoma (HCC), genomic instability, and progeroid features. To analyze Spartan function in DNA damage, U2OS cells were depleted of endogenous SPRTN using siRNA. Later, these SPRTN knockdown cells were transfected with the WT SPRTN, the mutant p.Tyr117Cys SPRTN, or ∆C-TER SPRTN. The authors found that the WT and mutated p.Tyr117Cys SPRTN formed nuclear foci, but not the mutated ∆C-TER SPRTN. The histological and immunohistochemical investigation of the patients’ liver tumor biopsies showed an increased accumulation of γH2AX and 53BP1 after CPT treatment, a chemotherapeutic agent that induces DPCs, including Top1 cleavage complex (Top1ccs). This result was also confirmed in SPRTN-knockdown U2OS cells expressing the mutant p.Tyr117Cys SPRTN and ∆C-TER SPRTN. Severe growth defects were also observed in patient fibroblasts, which showed increased levels of DSBs when in the S-phase. Indeed, transfection of patient fibroblasts with WT SPRTN efficiently corrected the replication defects and reestablished cellular proliferation. These results revealed that cells expressing mutant SPRTN were unable to recover DNA replication fork progression, leading to DNA replication stress and replication-related DNA damage, especially DSBs (Lessel et al., 2014). In the same year, Maskey et al. (2014) demonstrated that γH2AX foci, a marker of DNA damage, were markedly increased in Sprtn F/- MEFs after 4-hydroxytamoxifen (4-OHT) treatments, and that Sprtn -/- MEFs had increased numbers of 53BP1 nuclear bodies, indicating incomplete DNA replication. To better characterize the molecular mechanism by which SPRTN contributes to genomic stability, Lopez-Mosqueda et al. (2016) verified the role of SPRTN in resolving DPCs. They found that SPRTN-KO MEFs were sensitive to agents that induce DPCs, such as formaldehyde, etoposide, and CPT. Also, B-II-1 lymphoblastoid cells derived from RJALS were sensitive to those DPCs-inductor agents. These cells also exhibited more γ-H2AX staining after formaldehyde and etoposide treatments (Lopez-Mosqueda et al., 2016). They also confirmed that SPRTN is a DNA binding protease involved with the removal of DPCs in vivo and in vitro. These data are consistent with accelerated aging phenotypes observed in the hypomorphic SPRTN mouse model, linking DPC repair deficiency to segmental progeroid syndrome (Lopez-Mosqueda et al., 2016). Vaz et al. (2016) confirmed that SPRTN protease is a protein specialized in the repair of DPCs, being essential for DNA replication progression and genome stability. They found that RJALS patient cells and SPRTN-depleted cells were hypersensitive to agents inducing DPCs. Besides, HeLa cells transfected with ∆-SPRTN showed a higher average number of 53BP1 foci relative to controls after CPT treatment. This was observed only in cyclin A-positive ∆-SPRTN HeLa cells, suggesting a role of SPRTN in preventing DSBs induced by DPCs during the S-phase. Thus, RJALS cells are unable to process DPCs during DNA replication, leading to DNA replication stress, one of the main causes of genome instability and cancer (Vaz et al., 2016). Maskey et al. (2017) used Sprtn hypomorphic MEFs, which express reduced levels of Spartan but have a normal cell-cycle distribution, to verify the role of Spartan in the repair of Top1ccs, a bulky CPT-induced DPC that blocks replication forks. They found that Sprtn hypomorphic MEFs exhibited high CPT sensitivity compared to control MEFs, suggesting that Spartan may play a role in Top1ccs repair. Furthermore, they studied the effects of DPCs in Sprtn hypomorphic mice, which recapitulate phenotypes observed on RJALS. They found an accumulation of Top1ccs in the liver, indicating an increased binding of Top1 to DNA (Maskey et al., 2017). Therefore, given that Spartan plays a significant role in DNA stability by being responsible for DPC repair throughout DNA replication, pathogenic variants in the SPRTN gene affect DNA repair and are associated with hepatocellular carcinoma and premature aging, such as in RJALS. Figure 2 shows a model depicting the occurrence of unrepaired DSBs and persistent γ-H2AX in some progeroid diseases with remarkable loss of sWAT. As reviewed here, activation of DDR in HGPS, MADA, MADB, WRS, RJALS, and MDPL was seen, revealing an association among DSBs’ accumulation, aging, and loss of sWAT. Indeed, the role of p53 in the maintenance of sWAT homeostasis during aging was confirmed by Liu and co-workers (Liu et al., 2018). Using adipocyte-specific MDM2-knockout mice (Adipo-MDM2-KO), the authors found that MDM2 mRNA and protein levels are selectively downregulated in sWAT and BAT, while p53 and p21 were induced in both AT depots. Adipose senescence and apoptosis were observed in aged adipose tissue, and adipocytes had an aberrant expression of pro-inflammatory cytokines, such as TNFα and IL-6, while the p21 senescent marker was increased. Furthermore, adipocytes from old Adipo-MDM2-KO showed remarkable and progressive loss of SWAT, eWAT, and BAT, and leptin and adiponectin levels were nearly undetectable, revealing an early onset of lipodystrophy in this mice model. These mice also had diabetes, fatty liver, and higher levels of TGs, insulin, and glucose in plasma. The role of p53 in adipocytes’ homeostasis was validated by the generation of a DKO mice model lacking p53. DKO mice showed a rescued phenotype of sWAT loss and improvement of the metabolic parameters, confirming that the p53 activation is related to the MDM2-null phenotypes. However, the contribution of DNA damage/repair to the MDM2-p53 axis in the Adipo-MDM2-KO mice model was not assessed.
The ER-localized seipin, an adipose tissue-related protein involved with LDs assembly (Wang et al., 2016), was associated with changes in redox homeostasis (Craveiro Sarmento et al., 2020). The authors verified that blood leukocytes from CGL2 individuals carrying the pathogenic variant c.325dupA (p.T109Nfs*5) in the BSCL2 gene displayed higher levels of serum oxidized glutathione and malondialdehyde, indicating the occurrence of oxidative stress and lipid peroxidation on blood from individuals presenting a paucity of sWAT since birth. Using LX-PCR to quantify the levels of mitochondrial DNA (mtDNA) damage, they found that the number of mtDNA lesions obtained from blood leukocytes from CGL2 subjects was higher relative to the control groups. Besides, the levels of mtDNA lesions were positively correlated with NFE2L2 (NRF2) mRNA levels, suggesting the activation of NRF2 antioxidant responses. A positive correlation was also found between NRF2 mRNA and serum adiponectin levels. Even in low levels in CGL2 subjects, this finding suggests that NRF2 activation occurred in an adiponectin-dependent manner. More studies are needed to unravel the relationship between NRF2 and adiponectin in the context of loss of sWAT. Moreover, mitochondrial bioinformatics predictions by Mitochondrial Disease Database (MITODB) (Scheibye-Knudsen et al., 2013), a software that determines whether a disease could be associated with mitochondrial commitments according to its phenotypes, revealed that CGL2 has a high probability (mito-score 92) of being related to mitochondrial disturbs since its clinical spectrum includes lipodystrophy, hepatomegaly, HTG, muscle hypertrophy, muscle hyperplasia, hypertrophic cardiomyopathy, and bone cysts (Lima et al., 2016). These findings are in accordance with recently published data (Combot et al., 2022), who found that seipin is localized at ER-mitochondria sites and has a role in the Ca2+ importation to mitochondria. However, how this protein regulates changes in redox homeostasis in CGL2 subjects needs more investigation. Since mtDNA lesions were higher and upregulation of NRF2 mRNA was found in CGL2 subjects, Craveiro-Sarmento et al. (2019) investigated whether the BER pathway could be regulated in blood leukocytes. These cells displayed higher mRNA levels of APEX1, OGG1, and OGG1α, and the latter is expressed both in the nucleus and mitochondria and has an essential role in the maintenance of mitochondrial functions (Lia et al., 2018). Table 2 summarizes the main findings of this topic.
Whole blood from a subject harboring the heterozygous pathogenic variant c.479_480delTT and c.51_52insGTC in the CAV1 was associated with a severe neonatal progeroid and lipodystrophy syndrome. The 3-year-old patient also presented a heterozygous variant c.51_52insGTC in the AGPAT2 gene. The contribution of the latter to the development of this lipodystrophic progeroid disease is unclear. The 3-year-old patient showed severe loss of sWAT, progeroid features, and high levels of TGs in infancy. Fibroblasts isolated from this subject displayed lower levels of the caveolin-1 protein relative to the controls. RNA-seq analysis suggested a downregulation of LMNA, ATM, RECQL4, and WRN genes in the whole blood cells from this subject. Furthermore, the Fanconi anemia pathway was also downregulated. However, experimental data were not conducted, and a list with all differentially expressed genes was not provided to confirm these findings. Table 2 summarizes the main findings of this topic.
The role of DNA repair enzymes in adipose tissue homeostasis was also studied in obesity, revealing the importance of DNA integrity for maintaining the functions of WAT. In this section, we will highlight the main findings concerning the role of NEIL1 (Nei like DNA glycosylase 1) and OGG1 DNA glycosylases, from the BER pathway; ATM, which is involved with the repair of DSBs; and XPV, the DNA polymerase eta that acts bypassing the UV-induced DNA lesions, being involved with damage tolerance by translesion synthesis (Menck and Munford, 2014). Table 3 shows the main findings of this section.
NEIL1 was one of the first BER enzymes associated with metabolic complications (Vartanian et al., 2006). Under chow diet ad libitum, Neil1 -/- mice displayed severe obesity, dyslipidemia, and hepatic steatosis. These mice exhibited hepatic steatosis, hyperleptinemia, and high levels of TGs and insulin in plasma. Besides, they found increased mitochondrial DNA (mtDNA) damage and deletions, especially in male Neil1-/- mice (Vartanian et al., 2006). In another study by the same group, Neil1 -/- mice under chronic oxidative stress induced by a high-fat diet (HFD) displayed increased body weight and body fat accumulation, HTG, and glucose intolerance (Sampath et al., 2011). They also observed an increased hepatic expression of inflammatory genes and a reduction in mitochondrial DNA. These data demonstrated the role of NEIL1 DNA glycosylase in adipose tissue accumulation and mitochondrial dysfunction.
The role of the OGG1 BER enzyme in metabolic homeostasis has also been investigated by the Lloyd and Sampath groups (Sampath et al., 2012; Vartanian et al., 2017; Komakula et al., 2018, 2021). They first found that Ogg1 -/- mice were more susceptible to obesity and metabolic dysfunction relative to control mice. Under a high-fat diet (HFD), they presented higher adiposity, developed hepatic steatosis, and showed higher levels of insulin and hepatic TGs. Analysis of microarray and qPCR revealed that genes related to the TCA cycle and FAO were downregulated in the liver of Ogg1 -/- mice, as well as the Ppargc1a and Ppargc1b genes that codify to the PPAR-gamma coactivator-1 alpha (Pgc1α) and PPAR-gamma coactivator-1 beta (Pgc1β), respectively (Sampath et al., 2012). Later, they verified that skeletal muscle from Ogg1 -/- mice show increased lipid deposition, which included TGs, cholesterol esters (CE), diacylglycerol (DAG), free fatty acids (FFAs), and phospholipids (PLs). Further, gene and protein expression of Drp1 and Fis1 proteins, which are associated with mitochondrial fission, were higher in muscle from Ogg1 -/- mice. Besides, the expression levels of genes regulating FAO and lipid uptake, as well as TCA, were increased relative to WT mice. No differences in 8-oxoG levels were found (Vartanian et al., 2017). The contribution of mitochondrial OGG1 to metabolic syndrome was also investigated. Using preadipocytes from transgenic mice targeting OGG1 to mitochondria (Ogg1 Tg mice), they found a protective role of OGG1 against diet-induced obesity, IR, and adipose tissue inflammation (Komakula et al., 2018). They observed a decreased body weight, fat body composition, and smaller adipocytes in eWAT in Ogg1 Tg mice under HFD. Furthermore, Ogg1 Tg mice displayed low levels of glucose, insulin, TGs, and cholesterol in plasma, as well as low levels of TGs and cholesterol in the liver, suggesting that the reduced fat mass observed in Ogg1 Tg mice does not result in lipodystrophic lipid accumulation in the liver. eWAT of Ogg1 Tg mice under HFD also exhibited high expression levels of Pgc1α, Sirt1, Tnfα, Ikkβ, and FAO genes, such as Cpt-1, Acox, Hsl, Atgl, and Pparα. Lower levels of leptin and higher levels of adiponectin were also found in Ogg1 Tg mice plasma. Since they previously found a downregulation in Pgc1α in Ogg1 -/- mice (Vartanian et al., 2017), the higher levels of this transcriptional co-activator from Ogg1 Tg mice indicate the role of OGG1 in promoting the mitochondrial metabolism in eWAT. Additionally, since SIRT1 regulates adiponectin levels (Qiang et al., 2007), and both are increased in eWAT of Ogg1 Tg mice, this work also demonstrated the importance of mtOGG1 for activating the SIRT1-adiponectin axis. They also investigated whether targeting OGG1 to mitochondria changes mitochondrial morphology. They found that mitochondrial are elongated in eWAT of Ogg1 Tg mice and these mice presented higher expression levels of mitochondrial fusion proteins, such as Mfn1, Mfn2, and Opa-1. Although 8-oxoG levels seem to be reduced in eWAT of Ogg1 Tg mice under HFD, no statical differences were observed relative to WT mice. Together, these data demonstrate the metabolic protective role of targeting OGG1 to mitochondria in eWAT. The role of OGG1 in adipogenesis and lipid accumulation was investigated (Komakula et al., 2021). Preadipocytes from Ogg1 -/- mice displayed increased expression of genes related to preadipocyte differentiation (Scd1, Pparγ, and c/ebpα) and enhanced lipid accumulation. On the contrary, mouse 3T3-L1 preadipocytes from Ogg1 Tg mice and 3T3-L1 cells expressing-MTS-hOGG1a showed attenuated expression of genes related to preadipocyte differentiation (Scd1, Pparγ, and c/ebpα) and reduced lipid accumulation. Since OGG1 activates PARP-1 (Noren Hooten et al., 2011), and PARylation inhibits adipogenesis (Devalaraja-Narashimha and Padanilam, 2010; Luo et al., 2017), they assessed the role of OGG1 on PARylation in mouse preadipocytes. While PARP-1 protein levels were higher before starting adipocytes differentiation, its levels decreased during adipogenesis induction in both 3T3-L1 cells (expressing-MTS-hOGG1a and GFP-controls), which in accordance with reduced PAR levels. However, MTS-hOGG1a cells exhibited higher PAR levels in all time points of adipocytes differentiation relative to control cells. Increased total protein PARylation was also verified in differentiated primary adipocytes and adipose tissue protein extracts from Ogg1 Tg mice, whereas primary adipocytes, adipose tissue extracts, liver, and BAT from Ogg1 -/- mice exhibited reduced levels of total protein PARylation. These findings reveal the role of OGG1 in promoting PARP-1 activity in mice. More data are needed to clarify the contribution of OGG1 in human adipogenesis.
The XP-V gene encodes polymerase η (Pol η), which plays a crucial role in preventing UV radiation-induced DNA damage (5). Defects in the gene encoding to pol η produce the variant form (V type) of the autosomal recessive disease Xeroderma Pigmentosum (XP-V) (Masutani et al., 1999). XP-V patients tend to have high sensitivity to UV radiation, which often leads them to develop skin cancer (Masutani et al., 1999). Chen and co-workers demonstrated that polymerase η deficiency in mice (polη -/- mice) causes obesity with visceral fat accumulation, hepatic steatosis, hyperleptinemia, hyperinsulinemia, and glucose intolerance. Hypertrophy of adipocytes, high levels of adipogenic regulator genes, such as SREBP1 and PPARγ, infiltration of macrophages, and the presence of CL structures were apparent in polη -/- mice. Comparisons between healthy and pol η-deficient mice showed that polη -/- mice had higher levels of DNA damage and greater DDR, due to upregulation and phosphorylation of ATM, H2AX, p21, and p53, as well as upregulation of NF-κB and PARP-1 (Chen et al., 2015). Further, polη -/- mice also displayed increased DSBs. It was also found that polη -/- mice under a high-fat diet, which induces oxidative stress, showed a DNA-damage mediated senescence. Besides, treatment with a p53 inhibitor, pifithrin-α (PFT-α), reduced adipocyte senescence and attenuated the metabolic abnormalities. (Chen et al., 2015). On the contrary, DNA damage attenuation induced by N-acetylcysteine (NAC) or metformin antioxidants ameliorated cellular senescence and metabolic abnormalities. These results indicate that high levels of DNA damage are responsible for promoting adipocyte senescence, playing a crucial role in the development of obesity and IR (Chen et al., 2015). These data revealed the involvement of the DNA lesion bypass polymerase Pol η to protect against metabolic comorbidities.
Ataxia-telangiectasia was first described in 1941 by Madam Louis-Bar as a disease characterized by progressive cerebellar ataxia followed by oculocutaneous telangiectasia. In 1957, Boder and Sedgwick reported the disease in seven patients, pointing to a family tendency and frequent pulmonary infection as less marked characteristics of the disease. In the same year, Wells and Shy founded an association between subcutaneous telangiectasia with progressive familial choreoathetosis. The disease caused a significant disorder in the central nervous system, which was initially overshadowed by pulmonary infections (Silberpfennig et al., 1941). Furthermore, ataxia-telangiectasia subjects display DM and IR (Bar et al., 1978; Blevins and Gebhart, 1996; Morio et al., 2009). The ataxia-telangiectasia mutated (ATM) gene encodes to the ATM protein, a kinase of 350 kDa that plays a crucial role in DNA repair and is necessary for genomic homeostasis maintenance (Mercer et al., 2010). DSBs activate ATM, which phosphorylates its substrates (or targets) downstream, promoting DNA repair. The main ATM targets are H2AX, cycle cell checkpoints kinases Chk-1 and Chk-2, and the p53 tumoral suppressor gene (Mercer et al., 2010; Takagi et al., 2015). Although ATM is better characterized as a DDR gene, recent studies point out that defective ATM causes atherosclerosis and metabolic abnormalities. Using an apolipoprotein/ATM heterozygous (Atm +/- /ApoE -/- ) mice, Mercer and co-workers revealed that Atm +/- /ApoE -/- mice displayed accelerated atherosclerosis and multiple phenotypes of metabolic syndrome (Mercer et al., 2010). Further, Atm +/- mice were fat, hypertensive, macrophage infiltration, and showed hyperlipidemia under HFD. Fat accumulation and macrophage infiltration were also verified in Atm+/-/ApoE-/- mice. VSMCs from Atm +/- mice showed higher DNA fragmentation induced by the prooxidant t-BHP, higher levels of p-ATM and γ-H2AX relative to Atm +/+ mice, and presented a delayed activation of Chk-2 and p53, but not Chk-1 (Mercer et al., 2010). Furthermore, increased levels of ROS and mtDNA damage in Atm +/- mice were found. Taken together, Mercer and co-workers observed that ATM haploinsufficiency results in DNA damage in cells that compose atherosclerotic plaques, in addition to accelerating atherosclerosis in vivo, and inducing several features of metabolic syndrome and mitochondrial dysfunction (Mercer et al., 2010). Therefore, defective ATM or its haploinsufficiency causes DNA damage, speeds up atherosclerosis and metabolic syndrome features, and may cause failure in DNA repair and p53 activation, resulting in the reduction of apoptosis and cycle cell interruption (Mercer et al., 2010). CCAAT/enhancer binding protein α (C/EBPα) and PPARγ are considered the central regulator for adipocyte differentiation. When PPARγ is activated by an agonist in fibroblasts, a complete differentiation program is stimulated, leading to morphological changes, accumulation of lipids, and the expression of almost all characteristic genes of adipocytes (Rosen and Spiegelman, 2000). Another study revealed that ATM is activated during adipogenesis, besides DNA damage and insulin stimulation, and controls this process via transcriptional regulation of C/EBPα and/or PPARγ, which are required for a complete adipocyte maturation (Takagi et al., 2015). Neither lipid accumulation nor adipocyte differentiation occurred in embryonic fibroblasts of Atm -/- knockout mice since there was a defective induction of C/EBPα and PPARγ ATM-dependent expression (Takagi et al., 2015). Besides, it was observed that Atm -/- mice were insulin resistant, presented lower levels of adiponectin and leptin, had less subcutaneous and interscapular adipose tissue, increased visceral fat level (similar to metabolic syndrome), and glucose intolerance when compared to normal Atm +/+ mice (Takagi et al., 2015). Finally, it is worth mentioning the importance of adipose tissue for glucose homeostasis, considering that adipokines such as adiponectin, leptin, visfatin, and omentin increase insulin sensitivity, while hypertrophic adipocytes secrete resistin and Tumor Necrosis Factor-alpha (TNFα), which decrease sensitivity to insulin (Rosen and Spiegelman, 2006). Therefore, ATM deficiency leads to impaired adipocyte differentiation, which impairs adipokine secretion, resulting in IR and glucose intolerance (Takagi et al., 2015). These data revealed the ATM in the regulation of fat metabolism. However, the contribution of DNA damage accumulation and repair in Atm -/- mice remains to be determined.
To better clarify the interplay between the altered DNA repair pathways reviewed here and the lipodystrophies’ cell models associated with these DNA repair changes, we performed some systems biology analysis. The interactions of the main proteins described in this review were analyzed using STRING database (Szklarczyk et al., 2017), Cytoscape desktop application (Shannon et al., 2003) and its plugins: Molecular Complex Detection (MCODE) (Bader and Hogue, 2003), CentiScaPe (Scardoni et al., 2009), Biological Networks Gene Ontology (BiNGO) (Maere et al., 2005), and iRegulon (Heberle et al., 2015), and InteractiVenn web tool (Janky et al., 2014). The network containing 49 proteins was firstly built using STRING, which collects and integrates physical (direct) and functional (indirect) interactions. Later, the network was analyzed using Cytoscape. CentiScaPe was used to identify centrality parameters, determining the network nodes that are experimentally and topologically relevant. The protein-protein interactions (PPI) from the network revealed 676 interactions between DNA repair and lipodystrophic proteins (Figure 3A). Two protein clusters (densely connected regions) were detected by MCODE: one cluster had 42 nodes and 580 interactions, and the gene ontology (GO) determined by BiNGO was DNA metabolic process (Figure 3B). The second cluster had 30 genes and 258 interactions, and the BiNGO-determined GO was fat cell differentiation (Figure 3C). CentiScaPe analysis showed that the most dynamic nodes of the network, referred to as hub-bottlenecks (in blue), include: LMNA, WRN, TP53, ATM, PARP1, PPARG, CEBPA, CDK2, SREBF1, and IGF1. InteractiVenn analysis revealed that 23 genes from the network are common to Cluster 1 and Cluster 2, ratifying the interplay of proteins from DNA repair and adipogenesis (Figure 4A). It is important to notice that since the STRING network was used as an input to Cytoscape, some experimental data reviewed here were not shown in STRING and, consequently, they were not depicted in the Cytoscape network, such as PARP1 with CEBPB, BSCL2 with OGG1, APEX1, and NFE2L2. However, even without these data, the network had a significant number of PPI. To scrutinize the regulators of the network, iRegulon was used to find the main transcription factors (TFs) regulating the genes of the network. The TFs controlling cluster 1 (DNA metabolic process) were: FOXM1, NF-YA, SIN3A, and E2F4 (Figure 4B). The role of FOXM1 in DNA repair, cell proliferation, and tissue homeostasis was previously described in different works (Tan et al., 2007; Kwok et al., 2010; Millour et al., 2011; Zhang et al., 2012; Monteiro et al., 2013; Khongkow et al., 2014; Zona et al., 2014). NF-YA role in DNA damage/repair was also verified (Jin et al., 2001; Lee et al., 2004; Lin et al., 2014). Besides, SIN3A is associated with genomic integrity, and DNA damage (McDonel et al., 2012), and the role of E2F4 in cell cycle progression was also shown (Ren et al., 2002). Furthermore, the TFs that regulate cluster 2 (fat cell differentiation) were: CEBPB, ATF4, JUN, and POLR2A (Figure 4C). The role of these TFs in adipogenesis was previously shown (Yu et al., 2014; Guo et al., 2015; Lee et al., 2016; Bradford et al., 2019; Ahmed et al., 2019; Ambele et al., 2020; Bléher et al., 2020). Data reviewed here and the interactomes shown in Figures 3 and 4 reveal a vigorous connection between DNA repair and adipose tissue-related genes. However, how this PPI affects the functions of these genes in the context of adipocyte differentiation has yet to be investigated. Further, the role of the abovementioned TFs in the regulation of this PPI remains to be elucidated. Therefore, lipodystrophies can be a useful model for studying the mechanisms that link genome instability, metabolic dysregulation, and aging.
Over recent years, advancements in our understanding concerning the genetics of congenital lipodystrophies led to a better knowledge of the onset and progression of these rare diseases. This review highlighted several findings showing the interplay between genes associated with DNA repair and adipogenesis. Based on the many results reviewed here, we concluded that the maintenance of genomic integrity and an effective DNA repair contribute to adipose tissue homeostasis. Therefore, the treatment strategies of congenital lipodystrophies should focus on the elimination/reduction of DNA damage accumulation, as well as on antioxidant therapies. Furthermore, some questions require more investigation. What is the link between genome stability and metabolism? How does DNA repair deficiency result in several forms of progeroid syndromes with lipodystrophy? How do lipodystrophies caused by pathogenic variants in adipose tissue-related genes result in DNA repair activation? To respond to these questions, it is crucial to scrutinize the DNA repair contributions in different adipose tissue depots obtained from adipose tissue-proficient and lipodystrophic cellular models. |
|
PMC9648119 | Matias Mendeville,Margaretha G. M. Roemer,G. Tjitske Los-de Vries,Martine E. D. Chamuleau,Daphne de Jong,Bauke Ylstra | The path towards consensus genome classification of diffuse large B-cell lymphoma for use in clinical practice | 27-10-2022 | diffuse large B-cell lymphoma (DLBCL),next generation sequencing (NGS),consensus classification,genomics,bioinformatics | Diffuse large B-cell lymphoma (DLBCL) is a widely heterogeneous disease in presentation, treatment response and outcome that results from a broad biological heterogeneity. Various stratification approaches have been proposed over time but failed to sufficiently capture the heterogeneous biology and behavior of the disease in a clinically relevant manner. The most recent DNA-based genomic subtyping studies are a major step forward by offering a level of refinement that could serve as a basis for exploration of personalized and targeted treatment for the years to come. To enable consistent trial designs and allow meaningful comparisons between studies, harmonization of the currently available knowledge into a single genomic classification widely applicable in daily practice is pivotal. In this review, we investigate potential avenues for harmonization of the presently available genomic subtypes of DLBCL inspired by consensus molecular classifications achieved for other malignancies. Finally, suggestions for laboratory techniques and infrastructure required for successful clinical implementation are described. | The path towards consensus genome classification of diffuse large B-cell lymphoma for use in clinical practice
Diffuse large B-cell lymphoma (DLBCL) is a widely heterogeneous disease in presentation, treatment response and outcome that results from a broad biological heterogeneity. Various stratification approaches have been proposed over time but failed to sufficiently capture the heterogeneous biology and behavior of the disease in a clinically relevant manner. The most recent DNA-based genomic subtyping studies are a major step forward by offering a level of refinement that could serve as a basis for exploration of personalized and targeted treatment for the years to come. To enable consistent trial designs and allow meaningful comparisons between studies, harmonization of the currently available knowledge into a single genomic classification widely applicable in daily practice is pivotal. In this review, we investigate potential avenues for harmonization of the presently available genomic subtypes of DLBCL inspired by consensus molecular classifications achieved for other malignancies. Finally, suggestions for laboratory techniques and infrastructure required for successful clinical implementation are described.
Molecular diagnostics of cancer has entered a new era, propelled by advances in omics- and bioinformatic technologies that provide a new layer of characteristics for tumor classification. In general, current state-of-the-art diagnostic pathology categorizes tumors using phenotypic macro- and microscopic and immunohistochemical (IHC) characteristics, combined with molecular assays for single or limited numbers of markers like PCR, and fluorescent in situ hybridization (FISH). Analyses of highly complex omics data by bioinformatic technologies have identified molecular patterns and pathways that underly biologically distinct, and thereby newly recognized categories. Vice versa, accepted diagnostics distinct categories may proof to be molecularly so closely related they may even be combined into a single entity. Diffuse large B-cell lymphoma (DLBCL), the most prevalent type of non-Hodgkin lymphoma and the focus of this review, is characterized by a complex, heterogeneous tumor biology that is reflected in clinical heterogeneity (1). This is evident from a wide outcome spectrum with cure for 60% of patients treated with standard immune-chemotherapy (R-CHOP) and disease progression for the other 40% of which the far majority eventually succumbs due to relapsing and/or refractory disease (2, 3). Since 2000, omics information started to contribute layers of comprehensive biological information to the diagnosis of DLBCL (4). At that time, RNA expression profiling by means of microarray analysis followed by unsupervised clustering revealed a relatively simple dichotomous distinction based on cell-of-origin (COO) (5). For universal application in daily clinical practice, this distinction was translated into various algorithms that relied on classic immunohistochemistry (IHC) assay data rather than complex RNA analytics. This undoubtedly aided to have DLBCL COO classification to be included in the updated 4th edition of the World Health Organization (WHO) Classification for Hematolymphoid Malignancies in 2016 (6). Nonetheless, it was never widely applied outside clinical trials, largely since the clinical implications ultimately proved to be limited (7–9). Almost 20 years after the RNA-based COO classification concept, several independent studies proposed DNA-based subtyping by next-generation sequencing (NGS) as an alternative means to capture the biological heterogeneity of DLBCL and to supersede or complement COO classification (10–13). The different DNA-subtyping studies bear significant similarities, but also differ in some a priori concepts, applied technologies, bioinformatical approaches and ultimately in part in recognized genomic subtypes (14, 15). These differences preclude uniform classification, which is a quintessential step towards clinical implementation and essential to perform meaningful clinical trials (16–18).
The more than 20-year-old RNA-based COO classification recognizes 2 major molecularly distinct classes considered to reflect different stages of B-cell differentiation; activated B-cell (ABC) and germinal center B-cell (GCB) while a small group of patients remains ‘unclassified’. Both in the primary discovery studies and various subsequent validation studies, patients with a GCB-type DLBCL consistently showed a better prognosis under guideline therapy than patients with an ABC-type DLBCL (4). The differential clinical outcomes coupled with distinctive underlying biology served as a justification for differential treatment. In the years that followed it became clear however that the complex and heterogeneous biology of DLBCL was not fully captured by this simple dichotomous classification (5). In particular, phase 2 and phase 3 clinical trials that either used COO as an inclusion parameter, or were post-hoc analyzed based on COO class, failed to demonstrate differential improvement of outcome for patients receiving experimental, targeted treatment alternatives (7, 8, 19). This does however not imply that RNA-based information would not provide essential information to dissect DLBCL biology, as specific host-immune response signatures could already be identified in the early 2000s (20). Most recently, deconvolution algorithms using known cell type specific RNA signatures to computationally infer cellular components from bulk RNA data have allowed to further dissect information on tumor features as well as non-malignant tumor immune microenvironment (TME) features. Thereby, the original GCB class was further divided into three to four differentiation phases (germinal center, dark zone, precursor memory B-cell, light zone) and ABC into two phases (pre-plasmablast, plasmablast/plasmacell). Hence, TME analysis from RNA expression data provided complementary signatures that could further and largely independently describe DLBCL biology in a clinically meaningful manner (21).
The first larger DNA-based next-generation sequencing (NGS) studies for DLBCL that were undertaken revealed a spectrum of mutations, numerical chromosomal copy number aberrations (CNAs) and translocations that were largely characteristic for either of the RNA expression-based COO classes (22–27). For example, mutations in the chromatin modifying genes CREBBP, KMT2D and EZH2, were described as characteristic of GCB-type DLBCL and chromosome 18q gain and MYD88 mutations characteristic of ABC-type DLBCL. Apart from these few COO-characteristic DNA alterations, the majority was shown to be only limitedly overrepresented in either class, explanatory for the extensive genetic heterogeneity of DLBCL. In 2018, research groups from the National Cancer Institute (NCI) and the Dana Farber Cancer Institute (DFCI) independently and practically simultaneously proposed DNA-based subtyping approaches based on whole exome sequencing (WES) (1, 10, 11). The NCI group made a first step towards harmonization of the two approaches by, like DFCI, also including CNAs to their classification which resulted in the LymphGen algorithm (12). The DFCI- and NCI studies included retrospectively collected patient cohorts and identified 5- and 7 genomic subtypes, respectively. Encouraging is that despite the different cohorts and bioinformatical approaches, both defining features and the resulting subtypes are largely overlapping ( Figure 1 and Box 1 ). Other groups, with other cohorts using overlapping bioinformatical approaches have been able to reproduce these subtypes by and large (13, 31–33), including unpublished results by the authors of this review. This all provides confidence that a DNA-based characterization of DLBCL has the potential to disentangle the biological heterogeneity that underlies DLBCL’s clinical heterogeneity.
The correspondence between the NCI’s LymphGen and DFCI’s NMF subtypes is 75% based on the 63.1% of patients classified by the LymphGen algorithm (12). If also the LymphGen unclassified samples are considered, the overall agreement between the two subtyping systems is around 50%. Approaches: Both studies performed comprehensive genomic profiling to detect somatic mutations, CNAs and translocations. Because of the lack of matched normal tissue for most samples, both studies applied custom computational pre-processing techniques to eliminate sequencing artifacts and distinguish somatic and germline variations. The DFCI group performed WES on a series of tissue biopsies of 304 patients with primary DLBCL. Samples were from 4 different trials and cohorts, of which 55% were derived from FFPE tissue, and 44% had matched-normal tissue availability (10). The NCI group performed WES on a series of fresh-frozen DLBCL tissue biopsies of 574 patients for which 96.5% were primary DLBCL tissues and the other 3.5% from relapsed or refractory, without matched-normal tissue (12). Below is a short summary of the most defining features which the NCI and DFCI proposed subtypes have in common. For a more comprehensive overview on details of their differences and commonalities we refer to a recent review by Crombie et al. (28). i. The C1 subtype recognized by DFCIs NMF algorithm finds its analogue in the BN2 subtype recognized by NCIs LymphGen algorithm. Combined, the two algorithms determined 21 defining genetic alterations, of which eight overlap. Overlapping genes include BCL6 translocations, alterations in NOTCH2 signaling genes and mutations targeting the NF-kB pathway. Furthermore, the C1/BN2 subtype is enriched for, but not restricted to ABC-type, and shows a favorable outcome. The C1/BN2 alterations form a genetic basis of immune evasion corresponding to mutations seen in marginal zone lymphoma. Non-overlapping genes include mutations of B2M, FAS, HLA-B and translocations of PD-1 ligands. ii. The NMF-C2 subtype is analogues to the LymphGen-A53 subtype. Both have characteristic TP53 inactivation, and a high degree of genome instability as reflected by the prominence of genome-wide CNAs. This subtype is not significantly enriched for either of the two COO types, which underpins that the original COO dichotomy was indeed an oversimplification of DLBCL biology. Overall survival of this C2/A53 subtype under R-CHOP treatment is unfavorable. A notable difference between the two subtypes is the high number of discordant subtype-defining features (36 from 41), including driver alterations such as chromosomal deletion of the CDKN2A locus (9p.21). iii. The NMF-C3 subtype is analogues to the LymphGen-EZB subtype, with a relatively high concordance of subtype-defining alterations (10 out of 18); including translocations of BCL2, and mutations in chromatin modifying genes. Discordant features include amplification of the REL locus (2p16.1) and mutations of FAS. The C3/EZB subtype represents classic GCB-type DLBCLs, and the genetic features are to a large extent alike follicular lymphoma (FL), which suggest that these DLBCLs represent transformed FL (29). Clinically, C3/EZB subtype tumors are considered of most high risk within the GCB-type of DLBCLs. Notably, also the RNA-based DHITSig is enriched in this subtype and used to further subdivide EZB. iv. The NMF-C4 subtype is analogous to the LymphGen-ST2 subtype. C4/ST2 subtype defining alterations affect BCR/PI3K signaling, the JAK/STAT pathway, and histone genes. Most of these DLBCLs belong to the GCB-type with favorable outcome. Few alterations linked to this subtype are concordant between the two classification systems (6 out of 24). The less defined nature of this subtype is further underpinned by a recent study suggesting that this subtype may be further subdivided into two subtypes with divergent biology: a TET2/SGK1 and a SOCS1/SGK1 subtype (13). v. The NMF-C5 subtype is analogues to the LymphGen-MCD subtype. Nine of the 24 characteristic alterations overlap which include mutations in genes associated with extranodal involvement (MYD88, CD79B, TBL1XR1). This C5/MCD subtype is highly enriched for ABC-type DLBCLs and is the subtype with the least favorable survival under R-CHOP treatment. Discordant alterations include other markers of immune evasion (mutations of HLA-B and translocations of PD-1 ligands) and copy number gains of chromosomal arms 3q and 18q. vi. Finally, the LymphGen classification describes the N1 subtype which is characterized by NOTCH1 mutations. This subtype occurs in less than 2% of DLBCLs Figure 1 and has the worst survival among the LymphGen subtypes. This subtype is not recognized by the NMF algorithm with the DFCI cohort. Also, when we extend the DFCI cohort with another 500 DLBCLs treated with R-CHOP, the NMF algorithm still does not recognize this class (authors unpublished results).
The DFCI group used unsupervised clustering combined with alteration-centric features ( Box 2 ). Driver alterations were discriminated from passengers, thereby reducing the genetic dataset to 158 features. Non-negative matrix factorization (NMF), an unsupervised clustering algorithm that detects patterns of co-occurring features and assigns a subtype to each included tumor, was used. The number of clusters to be identified was predefined between 4 and 10, which is actually an arbitrary choice. The NMF algorithm identified the optimal stability of clusters to be represented by 5 DLBCL groups of similar sizes, which the authors labelled as C1 to C5.
The two proposed DNA-based subtyping systems differ in their bioinformatic approaches for i) genomic feature definition, and ii) subtype identification (10, 12): i. To define genomic features a gene-centric approach can be applied that combines all DNA alterations that impact the same gene into 1 feature, independent of whether they are a mutation, translocation or CNA. For example, a point-mutation of CDKN2A and a deletion of the CDKN2A-locus 9p.21 would be recognized as 1 feature. Alternatively, an alteration-centric approach regards each DNA alteration type separately, independent of their location in the genome. In the example of CDKN2A, the mutation and 9p.21 deletion are regarded as two separate features. ii. Also machine learning algorithms for patient subtyping can generally be divided in 2 main approaches, supervised or unsupervised (30). The supervised approach uses predefined classes to construct a classification rule from the features. An unsupervised approach leaves it to the algorithm to identify a number of subtypes that are composed of feature characteristics prioritized by the algorithm. Semi-supervised learning would be where some prior knowledge on classes and or features is given.
The NCI group used semi-supervised clustering combined with gene-centric features ( Box 2 ). The prior knowledge given were four predefined classes, each composed of 1 or 2 specific DNA "seed" alterations: MCD (seed is co-mutation of CD79B and MYD88 L265P), BN2 (seed is NOTCH2 mutation or BCL6 translocation), N1 (seed is NOTCH1 mutation) and EZB (seed is EZH2 mutation or BCL2 translocation). Finally, the algorithm selected the additional genomic features that had the strongest association with the four classes through an iterative approach. All patient samples were included for classification with this 4-class algorithm, yet of the entire cohort, only 46% of cases could be assigned (11). In the remaining 54% of cases in the NCI cohort recurrent alterations of TP53 (25%), TET2 (10%) and SGK1 (6.9%) were identified. This prompted the NCI group to refine and extend the four classes with two additional classes: A53 (seed is mutation and/or CNA of TP53) and ST2 (seed is mutations of SGK1 and TET2), resulting in six seed classes (12). Subsequently, a Bayesian predictor model titled “LymphGen” was developed, which calculates for each individual tumor the subtype probabilities for each of the six classes based on its genetic alterations. Tumors designated as "core" tumors were defined as being attributed to one class with a probability score of >90%. Consequently, the Bayesian predictor allows tumors to be assigned to multiple classes. Those with a probability score greater than 90% for more than one class are the so called “genetically composite” tumors. Tumors with a probability score of 50%-90% for one single class were termed “extended” class members. Tumors with few subtype-specific genetic alterations were left unclassified. Thereby, the then 6-class LymphGen algorithm assigned 63.1% of cases of the NCI cohort (12). Later, the RNA expression-based MYC double-hit signature (DHITSig), previously developed by others (34), was added as a surrogate for MYC translocation status to split the EZB class in MYC positive and MYC negative cases.
Despite the different choices in feature identification and machine learning algorithms ( Box 2 ), the NCI and DFCI groups recognize a similar and extensive underlying biological heterogeneity of DLBCL. Some subtypes are already more similar than others. For example LymphGens MCD/NMF C5, LymphGen A53/NMF C2 and LymphGen EZB/NMF C3 are already relatively consistently defined. An important difference is that the LymphGen algorithm does only assign 63.1% of patients to any of their predefined subtypes, whereas the DFCIs NMF algorithm defines a number of subtypes to which 100% of the samples in the cohort are assigned. The N1 subtype is the rarest subtype and is only recognized by the NCI with the NOTCH1 mutation seed given to the LymphGen algorithm ( Box 1 ). A small fraction of DLBCL patients (<2%) carry NOTCH1 mutations which infers potential specific sensitivity to Ibrutinib, a Bruton’s tyrosine kinase (BTK) inhibitor. Due to its low frequency, the N1 subtype is not recognized using unsupervised techniques in relatively small series. The size of the currently studied cohorts has been too small, hence underpowered, to detect such rare genomic subtypes by unsupervised analysis. Unknown small genomic subtypes can only get recognized once the sample size is sufficiently large, as exemplified by Curtis et al. for breast cancer (35). Rare subtypes like N1 may be characterized by very specific biological characteristics that make them uniquely targetable with specific potent inhibitors and thereby highly relevant to be recognized. As an example from another cancer entity, in about 1% of metastatic colorectal cancers the ERBB2 oncogene on chromosome 17q is amplified, which can be effectively targeted by trastuzumab and neratinib and results in high response rates in these tumors (36–38). Likewise, 4-5% of non-small-cell lung cancers have a translocation of the ALK gene, which can be effectively targeted by the ALK inhibitor crizotinib (39). Not recognized by either NCI or DFCI are the actual high-grade B-cell lymphoma (HGBCL), B-cell lymphomas with MYC translocation together with either BCL2 and/or BCL6 translocation (double hit/triple hit). Unsupervised NMF clustering theoretically might be able to recognize this group as a subtype but, like the N1 subtype, it may have remained undetected as a result of the limited number of MYC-translocation positive DLBCLs in the DFCI dataset. The DHITSig signature used by the NCI is a surrogate marker to recognize a MYC subtype and troublesome for various reasons. First it is not DNA alteration derived and requires a different assay, namely RNA expression analysis. Second, the name of this signature is deceiving since it implies a genetic context of HGBCL, whereas only 64% of DHITSig-positive GCB-type DLBCLs actually carry a MYC translocation and 52% are actual double hit/triple hit DLBCLs (34). Third, also other lymphoma classes besides HGBCL double hit/triple hit such as Burkitt lymphoma score positive for DHITSig. This RNA DHIT signature is thus not specific for either MYC translocation or HGBCL (40, 41). Besides the choice of subtyping algorithms, the NCI gene-centric versus DFCI alteration-centric choices for genetic features deserve attention ( Box 2 ). The easiest solved are the focal chromosomal CNAs, aberrations smaller than 3Mb (42) which only encompass one or few genes, and can therefore be combined in a gene-centric fashion (43). The choice between alteration- or gene-centric is not obvious for the larger-scale chromosomal CNAs since they harbor hundreds of genes. Rather than rationalizing a choice between a gene-or alteration-centric approach, the machine learning algorithms can be offered data processed in either manner and side-by-side evaluated for best subtyping performance. Although the unsupervised clustering choice is an elegant data-driven approach to identify subtypes (17, 44, 45), in the end a classifier, like LymphGen, will need to be built to diagnose individual patients in daily clinical practice, which dictates another step towards harmonization.
The currently proposed DNA-based subtypes will be the basis for a unified biological classification that may require a two-step strategy (28). Step 1 would involve harmonization of the current DNA-subtyping systems into a single unified classification, Step 2 would be the development of a reproducible and widely applicable molecular diagnostic assay; certified, as well as cost- and time-effective to enable clinical implementation. This exposes various challenges, from the choice of laboratory technique, subtype-defining DNA alteration features and interpretation to classification algorithms and bioinformatic procedures.
Harmonization into a single classification is a first requirement for implementation in diagnostic routine. Objective, reproducible, and conclusive subtype definition for each patient sample, combined with a detailed understanding of the tumor biology of each defined DNA-class, will enable to explore clinical consequences of such classification, preferably in clinical trials (46). For various organ-specific malignancies molecular classifications for tumor families have now been standardized and integrated in the 5th edition series of the WHO Classifications and are starting to be implemented in the diagnostic workflow for those settings that have access to the technology (47–49). The road towards this level of applicability has been achieved with several research groups proposing their individual molecular classification as a starting point, at different moments in time and with different laboratory and bioinformatical techniques, as is exemplified by the classification of breast cancer, central nervous system (CNS) tumors and colorectal cancer (48, 49).
Probably breast cancer classification is one of the most successful early examples. An RNA-based classification for breast cancer found its way already into the 4th edition of the WHO Classifications of Breast Tumours in 2012, which was further expanded upon in 5th edition (47). it recognizes 5 molecular classes; each with different prognosis but also different treatment recommendations. The existing close transatlantic collaborations undoubtedly facilitated consensus formation, characterized as “organic” allowing different biological and bioinformatical perspectives to converge (46, 50, 51). Once a consensus classification was established and reproducible assays were developed, exploration of personalized and targeted treatment approaches could be effectively explored to identify bespoke treatment modalities, amongst others in the multi-armed I-SPY clinical trials (52). From the point of view of development of a molecular-based consensus classification, the present WHO classification for CNS tumors is an impressive result of intensive collaboration leading to a highly refined molecular classification. In 2014 a group of neuro-oncological pathologists, physically converged in 2014 in Haarlem (NLD) and prepared a clinically relevant histo-molecular diagnostic consensus classification, whilst reducing interobserver variability (53), which soon was implemented in the 4th edition of the WHO Classification of CNS Tumors (54). Subsequently, a largely novel approach was taken by means of genome-wide DNA methylation analysis where the large spectrum of CNS tumors were recognized by methylation profiles combined with a form of dimension reduction called t-distributed stochastic neighbor embedding (t-SNE) (55). The t-SNE methylation test alone allows for diagnoses of the large majority of CNS tumors, not seldomly more detailed and/or reliable compared to the histo-molecular diagnosis, resulting in redefinition of these entities. The collaborative effort with inclusion of samples and intellectual input from many research groups across the world as well as extensive discussions in the Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy (cIMPACT-NOW) (56) has helped a broad acceptance and indeed this molecular classification is now also included in the 5th edition of the WHO Classification of Central Nervous System Tumours (48, 48). To harmonize colorectal cancer (CRC) classification, the Colorectal Cancer Subtyping Consortium (CRCSC) was formed to integrate six independently published RNA-based classifications (49). As opposed to the CNS assembly consensus, a predefined mathematical harmonization path was taken with the aim to resolve inconsistencies between the various CRC classification systems. This approach culminated in four consensus molecular subtypes (CMSs) (49) to which each CRC sample adheres to a higher (core samples) or lesser (non-core) extend. Since the context in CRC classification is so very similar to the current status in DLBCL, we here provide a summary of this CMS approach where three generic methodological steps were involved ( Box 3 ).
Three generic methodological steps are involved in the path taken for consensus classification of colorectal cancer. i. Independent expert team subtyping prediction on normalized raw data sets: Eighteen RNA-based CRC gene expression data sets, derived from different continents and research groups were assembled from public resources (Gene Expression Omnibus and The Cancer Genome Atlas). The data sets were compiled from various genome-wide expression analysis techniques (arrays and RNA-sequencing), different sample types (formalin-fixed paraffin embedded and fresh-frozen tissue materials) and different study designs (retrospective and prospective series, including clinical trials). The first bioinformatics step concerned central pre-processing and normalization aimed to obtain expression profiles for each of the patients of the 18 gene-expression datasets, independent of cohort or technique. Next, each of the six initial participating research teams applied their original classification algorithm to each of the 18 data sets. Thus, resulting in six classifications, with a total of 27 different subtypes for all 3,962 patients. ii. Network analysis for consensus subtype identification: Using the six classification systems of the 3,962 patients, a network-based approach was applied to study the association between all the 27 subtypes. To detect robust clusters of recurrent molecular subtypes, an unsupervised Markov clustering approach was performed, resulting in the identification of four consensus molecular subtypes (CMSs). Of the 3962 samples, 3104 (78%) were identified as highly representative of a particular subtype and labelled as core consensus samples and the remaining n=858 as non-consensus samples. The core consensus samples were used to train the novel CMS classifier in the subsequent step. iii. CMS classifier construction and application: To allow classifications of individual cases, which is mandatory for diagnostic routine, a classification algorithm is required. Since the data sets were created using different RNA gene expression profiling techniques across the different studies, not all genes were included in all data sets. The CRCSC first converted all 18 separate data sets into a single data set. The genes that were commonly profiled by all separate data sets were selected to allow aggregation of all 18 data sets into a single data matrix. To construct the CMS classifier, the single data matrix, CMS classes and consensus sample set were used. The consensus samples were randomly split using two-third as training and one-third as validation set, and a random forest classifier was generated to calculate a prediction value for subtype assignment for each sample, by means of bootstrapping with 500 iterations. Application of the CMS classifier on the validation set demonstrated an overall accuracy of 90%. The CMS classifier was robust enough to allow assignment of 40% of the non-consensus samples, while the rest showed heterogeneous patterns of CMS subtypes and contained biological information of more than one class.
The process to come to a single, harmonized molecular classification for DLBCL may likely be the one taken for the development of colorectal cancer CMS. For DLBCL also, a similar issue in the underlying biology result in single class (core) tumors, unclassified samples and genetically composite tumors (12, 57). What should alleviate the consensus process is that for DLBCL two, rather than the six for colorectal cancer, existing DNA-classifications as a starting point while still various independent published and unpublished (authors of this review) datasets are available.
Any consensus classification for DLBCL will include a combination of mutations and structural chromosomal variations (CNAs and translocations) ( Box 1 ). Therefore, inclusion of this information into a single genome subtyping assay would be highly attractive. Various common laboratory and bioinformatics applications are available for mutation and CNA detection by NGS. Also NGS-based translocation detection is starting to become a cost-effective alternative for routinely used Fluorescent in situ hybridization (FISH) to determine translocations. ( Figure 2 ). FISH benefits from a choice of worldwide commercially available probes and assays but is labor-intensive with a certain level of technical variability and subjectivity in interpretation. Thereby, NGS outperforms FISH in several ways: it avoids interobserver variability, it can be performed with small and histologically compromised materials, and it is able to identify exact translocation breakpoints on nucleotide level. An additional advantage of some of the NGS approaches is that unknown translocation partners may be identified, that may be of clinical relevance for the biological and clinical interpretation of DLBCL patients with a MYC translocation (40). Various combinations of NGS and bioinformatics platforms have been successfully developed in this direction (58–61).
World-wide clinical implementation of any diagnostic routine requires relatively simple assays that are applicable to routine diagnostic tissue material, such as formalin-fixed paraffin embedded (FFPE) specimens. The elaborate laboratory- and informatics infrastructure needed for current NGS or array analysis may only be available in selected settings of large medical centers or commercial providers as exemplified for CNS tumors. Favorable aspects of commercial involvement are the wide availability, extensive standardization, quality control and rapid turnover time due to high case volumes. Downsides are amongst others worldwide availability, financial dependency and commercial goals, market dominance of individual commercial providers, lack of technical transparency and development, lack of flexibility to include most recent research developments and generally lack of integrated interpretation with other pathology parameters. Another option to bring a genome subtyping assay to implementation in daily practice is to “reduce” complex molecular information to simpler and widely applicable techniques. The DLBCL-COO classification alternative is a good example; genome-wide molecular classification with elaborate bioinformatics was translated into several simple immunohistochemistry (IHC) markers, of which the Hans classification is most widely used (62). All IHC-based COO assays show limited concordance with the gold standard of RNA expression-based assays (63). This prompted the development of a digital gene expression assay based on 20 key genes that can be applied on FFPE material (64). This Lymph2Cx assay, restricted to equipment from the company Nanostring (Seattle, USA), showed high concordance with the original RNA expression-based COO classification with a 2% error rate in COO assignment (65). These characteristics, together with a short turnaround time of less than 36 hours, allowed for rapid molecular characterization of patients, making this assay a suitable middle-ground alternative for employment in research and clinical trials (19). Similar assays have been commercialized by others (66). In view of the expected high-dimensional nature of a consensus molecular classifier for DLBCL, simple translation to an IHC is not likely. Current NGS techniques are already reliably applicable for FFPE biopsy samples offered by commercial providers. It may be expected that these companies will readily offer products for consensus molecular DLBCL classification once this would be developed.
A single genome subtyping assay that detects CNAs, mutations, and translocations in parallel would conceivably be most efficient in terms of labor, cost and tissue material. But is this also efficient in terms of turnaround time? A recent study showed that real-time molecular profiling of RNA-based COO determination of DLBCL is realistic to stratify patients in a timely manner, with a median turnaround time of 8 days (8). This would be a desirable timeframe for DNA-based DLBCL classification, such that based on tumor vulnerabilities, patients can be diverted after 1 or 2 cycles standard R-CHOP treatment, which is a successful approach facilitating rapid trial inclusion (67). A recent feasibility study in the Netherlands, which involves a WGS specialized non-profit organization, was performed to evaluate implementation of WGS into routine diagnostics (68). Meanwhile, they were able to optimize the turnaround time from biopsy to DNA report to 7 working days, demonstrating the potential of clinical implementation of NGS methods for these purposes.
Once validated, uniform and widely applicable, consensus molecular subtypes of DLBCL will be a sound basis to explore more effective, targeted treatment methods (1). The potential of DNA-based classification for precision medicine of DLBCL has been demonstrated in a recent retrospective analysis of a randomized phase-III trial (69). In this study, patients under 60 with two specific DNA subtypes (LymphGen’s MCD and N1) that received R-CHOP with Ibrutinib had significantly better survival (both subtypes 100% 3-year event-free survival) than patients that received R-CHOP alone (42.9% and 50%, respectively), clearly indicating the potential predictive value of the novel genomic subtypes. Next, prospective clinical trials may further explore associations with genomic subtypes and associations with targeted compounds, such as NFkB-inhibitors, PI3K inhibitors, P53-modulators and apoptosis modulators, as well as immunotherapy such as immune checkpoint inhibitors and CAR-T cell therapy. For this purpose, various dedicated next-generation designs are now proposed (70). It is obvious to further investigate to what extent the integration of the current DNA-based and RNA/microenvironmental-based subtyping methods for DLBCL would be of added value. Adding a layer of epigenetic information as for CNS (55) or even germline genetic characteristics might be considered (71). Also liquid biopsy strategies measuring circulating tumor DNA (ctDNA), will provide other lines of opportunities in diagnosis and disease monitoring of DLBCL patients (72–74) Future studies are required to investigate the potential integration of these approaches for the management of DLBCL patients.
The step forward to allow evaluation of new treatment modalities based on DLBCL genetics is now impeded by a discordancy between the 2 independently suggested genomic subtyping approaches, which dictates the challenge that lies ahead of us. Based on various other tumor entities we suggest a blueprint for harmonization of the proposed DNA subtypes, which may allow more widespread clinical implementation. Once this hurdle is taken, a diagnostic work up, applicable in a clinically relevant timeframe, will enable the design of next-generation prospective biomarker-based clinical trials. If successful, the precision medicine with targeted therapies that match dependencies of the molecular subtypes of DLBCL may be brought forward.
MM, MR, DJ, and BY contributed to conception and design of the review and wrote the first draft of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.
This work was supported by the Dutch Cancer Society grant KWF 2012-5711.
The authors like to thank dr. Erik van Dijk for critically reading the manuscript prior to submission and Prof. Dr. Pieter Wesseling for helpful discussions on CNS diagnostics, both affiliated to Amsterdam UMC, Vrije Universiteit Amsterdam, Department of Pathology, Amsterdam, The Netherlands.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648122 | Joydeep Chakraborty,Anis Ahmad Chaudhary,Salah-Ud-Din Khan,Hassan Ahmad Rudayni,Sayed Modinur Rahaman,Hironmoy Sarkar | CRISPR/Cas-Based Biosensor As a New Age Detection Method for Pathogenic Bacteria | 18-10-2022 | Methods enabling rapid and on-site detection of pathogenic bacteria are a prerequisite for public health assurance, medical diagnostics, ensuring food safety and security, and research. Many current bacteria detection technologies are inconvenient and time-consuming, making them unsuitable for field detection. New technology based on the CRISPR/Cas system has the potential to fill the existing gaps in detection. The clustered regularly interspaced short palindromic repeats (CRISPR) system is a part of the bacterial adaptive immune system to protect them from intruding bacteriophages. The immunological memory is saved by the CRISPR array of bacteria in the form of short DNA sequences (spacers) from invading viruses and incorporated with the CRISPR DNA repeats. Cas proteins are responsible for triggering and initiating the adaptive immune function of CRISPR/Cas systems. In advanced biological research, the CRISPR/Cas system has emerged as a significant tool from genome editing to pathogen detection. By considering its sensitivity and specificity, this system can become one of the leading detection methods for targeting DNA/RNA. This technique is well applied in virus detection like Dengue, ZIKA, SARS-CoV-2, etc., but for bacterial detection, this CRISPR/Cas system is limited to only a few organisms to date. In this review, we have discussed the different techniques based on the CRISPR/Cas system that have been developed for the detection of various pathogenic bacteria like L. monocytogenes, M. tuberculosis, Methicillin-resistant S. aureus, Salmonella, E. coli, P. aeruginosa, and A. baumannii. | CRISPR/Cas-Based Biosensor As a New Age Detection Method for Pathogenic Bacteria
Methods enabling rapid and on-site detection of pathogenic bacteria are a prerequisite for public health assurance, medical diagnostics, ensuring food safety and security, and research. Many current bacteria detection technologies are inconvenient and time-consuming, making them unsuitable for field detection. New technology based on the CRISPR/Cas system has the potential to fill the existing gaps in detection. The clustered regularly interspaced short palindromic repeats (CRISPR) system is a part of the bacterial adaptive immune system to protect them from intruding bacteriophages. The immunological memory is saved by the CRISPR array of bacteria in the form of short DNA sequences (spacers) from invading viruses and incorporated with the CRISPR DNA repeats. Cas proteins are responsible for triggering and initiating the adaptive immune function of CRISPR/Cas systems. In advanced biological research, the CRISPR/Cas system has emerged as a significant tool from genome editing to pathogen detection. By considering its sensitivity and specificity, this system can become one of the leading detection methods for targeting DNA/RNA. This technique is well applied in virus detection like Dengue, ZIKA, SARS-CoV-2, etc., but for bacterial detection, this CRISPR/Cas system is limited to only a few organisms to date. In this review, we have discussed the different techniques based on the CRISPR/Cas system that have been developed for the detection of various pathogenic bacteria like L. monocytogenes, M. tuberculosis, Methicillin-resistant S. aureus, Salmonella, E. coli, P. aeruginosa, and A. baumannii.
Rapid, sensitive, specific, and on-site detection of pathogenic bacteria is critical in clinical diagnosis, treatment, surveillance of foodborne disease, and biological research. It helps to gather clinical information in order to provide appropriate treatment and prevent the spread of disease. According to the World Health Organization’s standard, an ideal pathogen diagnostic test should be assured: affordable, sensitive, specific, easy-to-use, rapid, without large equipment, and delivered to the user. In order to detect nucleic acid signatures of pathogens, a vast array of detection methods have emerged based on PCR/qPCR, isothermal amplification-based detection assays, and next-generation sequencing. To improve sensitivity, affordability, simplicity, and rapidity there are various advanced nucleic acid detection techniques developed so far. One of the latest and advanced methods is clustered regularly interspaced short palindromic repeats (CRISPR) associated systems (CRISPR/Cas), which have recently gained great importance and attention in nucleic acid analysis and detection. In order to achieve higher sensitivity, the CRISPR/Cas system is frequently associated with polymerase chain reaction (PCR) and with isothermal nucleic acid amplification techniques like NASBA, RCA, SDA, LAMP, RPA, and EXPAR. The combination of CRISPR/Cas with advanced isothermal amplification technologies is promoting the development of novel optical and electrochemical biosensing devices. CRISPR/Cas systems provide adaptive protection to bacteria and archaea against invading foreign nucleic acids. The CRISPR/Cas system in bacteria recognizes and degrades foreign genetic elements generally from viruses. These systems are primarily guided by an RNA called guide-RNA (gRNA) or CRISPR RNA (crRNA), which recognizes the target and directs Cas proteins to locate and cleave invading DNA sequences. This system works through three steps: adaptation or spacer-acquisition, crRNA processing, and interference. In the spacer acquisition step, when new foreign DNA/RNA is introduced in the bacterial cell, a short piece of the DNA/RNA segment, called protospacer in the immediate upstream vicinity of a protospacer adjacent motif (PAM) present in the foreign DNA/RNA, is excised out by the help of the Cas1–Cas2 complex. This protospacer was then inserted as a new spacer into the bacterial genomic region of the CRISPR array (Figure 1a) where all the acquired spacer resides. The second step is crRNA biogenesis, which occurs when pre-CRISPR RNA (pre-crRNA) is transcribed by RNA polymerase (RNAP) from the CRISPR array region, then cleaved by specific endoribonucleases into small mature crRNA (Figure 1a). Each crRNA contains one complementary sequence of a spacer. The final step is interference, which entails sequence-specific targeting and cleavage of foreign DNA/RNA having a protospacer that is complementary to the spacer sequence in crRNA. To commence crRNA-mediated DNA binding, a protospacer adjacent motif (PAM) must be present in the immediate vicinity of a protospacer sequence. crRNAs recognize and produce complementary base pairs unique to foreign RNA or DNA, resulting in the cleavage of the crRNA-foreign nucleic acid complex (Figure 1a). CRISPR/Cas systems are classified according to their utilization of specific Cas enzymes and methods of interference. In accordance with recent publications, CRISPR/Cas systems can be categorized into two classes: class 1 and 2, six types: types I–VI, and numerous subtypes. Class 1 comprises types I, III, and IV; and Class 2 includes types II, V, and VI (Figure 1b). Each type is distinguished by discrete effector module configurations that include various signature proteins. The most widely used toolbox for nucleic acid detection belongs to the class 2 system that contains Cas9, Cas12, Cas13, and Cas14. Cas9 (type II) and Cas12 (type V) target DNA, while Cas13 (type VI) targets RNA and Cas14 targets ssDNA. CRISPR/Cas systems, specifically Cas9 (type II), have become a popular tool for transcription regulation, genome editing, and in situ DNA/RNA detection in recent years. Cas12 and Cas13 effectors have a unique property called “collateral cleavage”. In the presence of a target or reporter DNA/RNA, these Cas effectors are activated and can do collateral (nonspecific) cleavage on any single-stranded DNA/RNA present in the near vicinity. The advantage of this collateral cleavage is that it can easily be detected by fluorescence reporters tagged in single-stranded DNA/RNA. This has recently displayed remarkable potential in developing novel biosensing technologies for nucleic acid detection. This technology is widely harnessed for the detection of viral diseases, such as specific high-sensitivity enzymatic reporter unlocking (SHERLOCK) to detect Zika and Dengue and DNA endonuclease-targeted CRISPR trans reporter (DETECTR) for rapid and specific detection of HPV and SARS-CoV-2 in humans. Though the futuristic developments of CRISPR/Cas-based viral detection techniques are expanding rapidly in biosensing, this technique is limited to very few bacterial pathogens to date. This limited use for bacterial pathogens may be because there is a well established gold-standard detection technique for pathogenic bacteria and establishing these emerging techniques in practice will be time-consuming. Also, the viral rate of mutation is relatively much higher than bacteria, which provides more priority to develop new methods for viral detection and lesser focus on bacterial detection. In this review, we have discussed the different CRISPR/Cas systems as biosensors used for the detection of bacterial pathogens like L. monocytogenes, M. tuberculosis, Methicillin-resistant S. aureus, Salmonella, E. coli, P. aeruginosa, and A. baumannii.
Listeria monocytogenes is one of the most virulent foodborne pathogens and can be found in a variety of foods like milk, milk products, eggs, poultry, and meat. The FDA upholds a zero-tolerance policy for L. monocytogenes since it has a low infectious dose and high mortality rate. In healthy people, it can cause invasive listeriosis. In young, elderly, or immunocompromised people, it can cause septicemia, meningitis, and infections related to the central nervous system. Infections in pregnant women can be fatal and can result in spontaneous abortion or fetal death. The slow growth rate of L. monocytogenes is challenging for the conventional culture and plating-based detection methods, which can take up to 7 days to yield results. CRISPR/Cas9-triggered isothermal exponential amplification reaction (CAS-EXPAR) based detection against Listeria monocytogenes was developed by Huang et al. Here the hemolysin (hly) gene of L. monocytogenes was used as the target sequence. It utilizes the target-specific nicking activity of Cas9 and nicking endonuclease (NEase)-mediated amplification. From bacteria, RNA was isolated and cDNA was generated. cDNAs were cleaved by Cas9 with the help of specific sgRNA and PAMmers. These cleaved products are now subjected to EXPAR-mediated amplification by EXPAR templates and without exogenous primers. Finally, the amplified products were detected by fluorescence using SYBR green (Figure 2a). This method combines the benefits of Cas9/sgRNA site-specific cleavage and EXPAR fast amplification kinetics. This process is reported to be highly specific in discriminating single-nucleotide mismatch. Reprogrammable cleavage activity of Cas9/sgRNA is also beneficial for targeting various other pathogens. The merit of this method does not require exogenous primers for amplification. Therefore, the chances of nonspecific amplification followed by false positivity could be minimized. The sensitivity of this technique was reported to be 0.82 amol (Table 1) of synthetic ssDNA, but in bacteria, this technique was verified with 1.25 and 2.5 μg of total RNA. This method may have a problem to detect long targets, as EXPAR is not efficient for long DNA or RNA targets. Another method to detect L. monocytogenes was developed by Wang et al. based on CRISPR/Cas9 system integrated with lateral flow nuclic acid (CASLFA). Here also, the hly gene was used as a target. This technique is termed as the DNA unwinding-based hybridization assay with a lateral flow device for simple and easy detection by the naked eye. Genomic DNA from bacteria was subjected to amplification (by PCR or any isothermal reaction) with gene-specific biotinylated primers. Then biotinylated amplicons are incubated with target specific sgRNA and dCas9 to form a complex (Cas9/sgRNA-biotinylated amplicons) without cleaving the targets. This complex, when applied to a lateral flow device, bound with an AuNP-DNA probe (gold nanoparticle bound with complementary DNA sequence of target gene/sgRNA) and combined with immobilized streptavidin in the test line (T). Accumulation of AuNP-generated color band occurred on the test line (T). The excess unbound AuNP-DNA probe will form a control line (C) by binding with the precoated DNA probes with their control line hybridization region (Figure 2b). There are two kinds of DNA probes reported by the authors: DNA unwinding and sgRNA anchor-based. The DNA unwinding probe has a specific DNA sequence for individual target. Therefore, for every target, there is a need to generate a new probe. The sgRNA-based probe has a target sequence that has DNA sequences specific to the crRNA region. The advantage of the sgRNA-based probe is that it can detect multiple targets as this is specific to a portion of sgRNA but not the target DNA. This is a simple and rapid method for the detection of genetic targets by the naked eye. This detection limit was reported as low as 150 copies (Table 1) of bacterial targets. As reported, this method can detect the gene target with almost no background signal interference and can be completed based on a cheap and portable tool kit within 40 min in point of care.
Tuberculosis (TB) is the leading cause of death among infectious diseases. Due to the difficulty in its diagnosis, it was anticipated that 40% of the cases failed to be identified and reported. Zhang et al. have developed a novel and sensitive detection method to detect M. tuberculosis using CRISPR/dCas9. The Mtb 16S rRNA gene was used as a target sequence. In this method, the luciferase gene was split into N- and C-terminal halves (NFluc and CFluc) and fused each with separate dCas9 termed as a paired-dCas9 (PC) reporter system. Two guide RNAs (sgRNA) were also employed which are complementary to the upstream and downstream proximal segments (∼44bp) of a target DNA. Upstream and downstream sgRNAs are separately mixed with NFluc-Cas9 and CFluc-Cas9, respectively, to achieve higher efficiency and specificity. The two halves can initiate heterodimerization to rebuild the intact enzyme when target DNA containing the two segments in proximity is detected and bound by the corresponding sgRNAs followed by a pair of dCas9 (Figure 2c). Luminescence is generated from the catalytic activity of luciferase. The PC reporter system was reported to be sensitive up to 5 × 10–5 nmol/mL (Table 1) but also observed that when the target concentration was beyond 6 × 10–3 nmol/mL detection signals decreased due to inefficient pairing between two halves of luciferase. This method requires two target sites, for separate sgRNAs, spanning approximately 44bp, which could limit the selection of target region for other pathogens. The other method to detect M. tuberculosis by CRISPR-MTB was developed by Ai et al. Here MTB-specific insertion sequence IS6110 of ∼1.5 kb in length was used as the target sequence. In this method, optimization of the extraction process was done based on a combination strategy of bead beating, chemical lysis, and heating to ensure the higher efficiency of DNA extraction. As the amplification technique is RPA, there was no special need for any thermal-cycler. This method is a combination of RPA reaction with a CRISPR/Cas12a system (Figure 2d). Fluorescence signal can be detected by target activated reporter cleavage of Cas12a trans cleavage activity. The sensitivity of this method was reported to be 2–5 copies/μL or 50 CFU/mL (Table 1). It was reported that the extraction efficiency was found to be high and the extraction time was reduced with fewer centrifugation steps over the column-based traditional method. This method also was tested on a variety of sample types such as sputum, BALF, CSF, and pus. The drawback of this detection will be that it will fail to detect MTB strains that lack IS6110 genomic segments.
Methicillin-resistant Staphylococcus aureus (MRSA) is one of the most important multidrug-resistant human pathogens, causing severe life-threatening diseases. MRSA infections are four times more likely than methicillin-susceptible S. aureus (MSSA), and it causes serious morbidity and mortality worldwide. Traditional culture-based identification methods for MRSA are time-consuming, and conventional techniques like MALDI-TOF, RT-PCR,S. aureus protein A (spa) typing, multilocus sequence typing (MLST), and pulsed-field gel electrophoresis (PFGE) are labor intensive and require a high level of professional expertise. Therefore, MRSA detection requires simplified detection procedures that are faster, less labor-intensive, and highly specific. Kyeonghye Guk and colleagues recently introduced a CRISPR-mediated DNA-FISH. This CRISPR-mediated DNA-FISH was developed to detect methicillin-resistant Staphylococcus aureus (MRSA) by targeting the gene mecA. This technique involves dCas9 to specifically recognize the target gene without cleavage activity and SYBR Green as a fluorescent probe. The genomic DNA of the target organism was isolated and treated with dCas9/sgRNA for 15 min at room temperature to bind the dCas9 with the target. After hybridization, the dCas9/sgRNA complex was isolated using Ni-NTA magnet beads, and nontarget unbound DNA was removed by washing. SYBR green was added to detect the presence of bound DNA as a target (Figure 2e). In clinical isolates, this method can detect as low as 10 CFU/mL within 30 min (Table 1). This approach is both quick and sensitive. This method does not require any amplification which is an advantage that reduces the detection time and complexity. The combination of dCas9/sgRNA and SYBR green as a fluorescent probe makes for labeling a reasonably straightforward and inexpensive approach. This method of detection has great potential to be used easily in patient point of care. Suea-Ngam et al. have developed another amplification-free method for the detection of MRSA. Here also, the mecA gene was chosen as the target. In this method of detection, the silver metallization technology was combined with the CRISPR/Cas to create a novel silver-enhanced E-CRISPR biosensor (E-Si-CRISPR) for MRSA detection. In the presence of target DNA the Cas12a–gRNA complex cleaves the ssDNA at random sites, destroying the electrode’s ssDNA surface layer (Figure 2f). The trans-cleavage mechanism fails in the absence of the target. The degree of silver deposition during the succeeding silver metallization stage is proportional to the quantity of ssDNA left and thus proportional to the initial amount of target DNA. Square wave voltammetry was used to read the final electrochemical signal (Figure 2f). The detection and quantification limits were found to be 3.5 and 10 fM (Table 1). This new electrochemical CRISPR/Cas biosensor, based on silver metallization, was stated to be highly selective, sensitive, and without DNA amplification cycles. As reported, this amplification-free detection method could yield results within 1.5 h. This method is innovative in the aspect of its unique readout of results through electrochemical signals. One drawback can be perceived that, contrary to the other conventional methods, the positive signals are lower than the negative signals in this method, which could be inconvenient.
Food poisoning by Salmonella species is the second most prevalent cause of food poising followed by severe gastroenteritis and bacteremia worldwide. To date, traditional biochemical culture, immunological testing, and molecular biological approaches (PCR/real-time PCR) have been used to detect Salmonella. These procedures are time-consuming, have low specificity, and require expensive laboratory equipment. Detection of Salmonella enteritidis using a unique allosteric probe (AP) with a combination of CRISPR/Cas13a (APC-Cas)is developed by Shen et al., where whole bacteria were used as a target. The allosteric probe (AP) comprises of three functional domains (Figure 3a): aptamer domain for target pathogen identification (purple), primer binding site domain (blue), and T7 promoter domain (yellow). A phosphate group was added to the 3′ end of AP to prevent self-extension and make the DNA molecule resistant to enzymatic hydrolysis. The aptamer domain of AP can specifically recognize and engage with the target pathogen that is Salmonella enteritidis. The hairpin structure of AP will unfold and flip to its active configuration, allowing primers to anneal to the exposed primer binding site domain. The AP then acts as a template for the production of double-stranded DNA (dsDNA) with the help of DNA polymerase, followed by the displacement of the target pathogen for the next catalytic cycle (primary amplification) because of the polymerase extension reaction. T7 RNA polymerase is then utilized to identify the T7 promoter sequence on the created dsDNA and perform amplification via transcription to generate a large number of single-stranded RNAs (ssRNAs) (secondary amplification). Finally, the crRNA is intended to contain two areas, a guide sequence that is complementary to the transcripted ssRNA, and the repeat sequence that is required for crRNA to attach the Cas13a enzyme. When the above ssRNAs hybridize with Cas13a/crRNA, Cas13a/collateral crRNA’s cleavage capacity is activated, allowing numerous RNA reporter probes to be cleaved (tertiary amplification), resulting in the amplification of fluorescence signals (Figure 3a). This procedure does not require bacterial isolation, nucleic acid extraction, and a washing step. It is cost-effective, very sensitive up to 1 CFU (Table 1), and can be done in a relatively short period. Designing an allosteric probe for different bacteria could be a challenging task for this method of detection. In order to detect Salmonella, Ma et al. have developed gold-nanoparticles (AuNPs)-based method termed as CRISPR/Cas12a-powered dual-mode biosensor. The target DNA was Invasion gene A (invA), a virulence gene of Salmonella. This method involves DNA extraction as well as PCR amplification of target sequence. The designed biosensor is built on the trans-cleavage activity of the CRISPR/Cas12a. The AuNPs probe is coated with DNA, and a linker ssDNA hybridizes with AuNPs-DNA probe pairs. In the absence of target amplicons, the linker ssDNA remains intact, and aggregated AuNPs are maintained with a purple color (Figure 3b). Upon recognition of the target amplicons with designed crRNA, the trans-cleavage of CRISPR/Cas12a is activated and the linker ssDNA is cut off, and the AuNPs are dispersed in solution. The dispersed AuNP solution exhibits a red color, and the change can be detected by the naked eye or colorimetrically or photothermally (Figure 3b). The detection limit for this technique was reported as 1 CFU/mL (Table 1). This method was just used to detect the bacteria in milk samples. There is a need to explore with other food samples. This technique was the first to explore the gold-based nanoparticles as a probe.
Escherichia coli O157:H7 is one of the most common causes of hemorrhagic colitis.E. coli O157:H7 can be found in water as well as other food sources such as milk, juice, fruits, and vegetables. Infections that are severe enough can lead to hemorrhagic colitis, hemolytic uremic syndrome, and even death. A CRISPR/Cas9 triggered SDA–RCA method on the UiO66 platform was developed by Sun et al. to detect Escherichia coli O157:H7. The method employs the target sequence of gene hemolysin A (hlyA). Nanoparticle (UiO66) and Two amplification methods: strand displacement amplification (SDA) and rolling circle amplification (RCA) are used for this method. After isolation of DNA from the bacteria, the pair of CRISPR/Cas9 (by sgRNA1 and sgRNA2) recognized and cleaved the two proximal regions of the target DNA. Primary amplification by SDA synthesis and extending at the nicked position results in short–ssDNA indefinitely. This short–ssDNA was the template for secondary amplification by RCA, which generates long–ssDNA having repetitive sequences complementary to the fluorescence-labeled DNA probe (Figure 3c). This probe in bound form with long–ssDNA can be detected with 480 and 518 nm of excitation and emission wavelength. Unbound probes are absorbed in the UiO66 where fluorescence is quenched. In the presence of the target sequence, short (by SDA) followed by long ssDNA (by RCA) will be generated. The fluorescence probes will leave UiO66 and hybridize with the long–ssDNA, resulting in a fluorescence signal. As a result, the fluorescence intensity can be used to detect the target DNA quantitatively (Figure 3c). It is reported that this technique can detect low amounts of E. coli O157:H7 (40 CFU/mL) with high sensitivity and a wide detection range under mild response conditions (Table 1).
Pseudomonas aeruginosa is a multidrug-resistant, highly infectious opportunistic human pathogenic bacteria with a large and complex genome. Its widespread distribution in nature indicates a high level of genetic and physiological flexibility in response to environmental changes. In 2017, the World Health Organization designated P. aeruginosa as a critical pathogen that poses a serious threat to human health, necessitating the development of new treatments. Mukama et al. have developed a method to detect P. aeruginosa(37) based on CRISPR/Cas and loop mediated Isothermal Amplification (CIA). The acyltransferase gene from P. aeruginosa was used as a target. Here the samples were directly used for loop-mediated isothermal amplification (LAMP) for the target gene. Products of LAMP were incubated with CRISPR/Cas12, to activate the collateral cleavage of the biotinylated ssDNA reporter, followed by a run on a strip for final results (Figure 3d). In the absence of a target, the gold nanoparticle-streptavidin (AuNP-SA) complex binds with the biotin of the ssDNA reporter. Then the whole complex (AuNP-SA-ssDNA) binds with complementary DNA to the ssDNA reporter immobilized in the test (T) line which results in a visible colored band (Figure 3d). But, in the presence of a target, the reporter DNA was cleaved by CRISPR/Cas12, so there will not be any formation of the AuNP-SA-ssDNA complex, hence no visible band on the test line. In the control line (C), only AuNP-SA is bound with immobilized antibodies to streptavidin. That means in the presence of any P. aeruginosa, there will be only one band in the control and no visible band in the test line (Figure 3d). Whereas, if the sample is negative for P. aeruginosa then the visible band will be there in the T as well as in the C line in the strip. This method was reported to be a fast, accurate, robust, and inexpensive technique with a detection limit of 1 CFU/mL (Table 1). The best feature of this approach is that it allows for naked-eye detection. That means this detection technique has the potential to apply in the patient point of care. On the other hand, the method of detection is unconventional too. In general, we are accustomed with a positive sample with a positive band in the test line, but here the positive sample is associated with a negative band in the test line, which may have some inconvenience in the hand of technicians. Additionally, we cannot eliminate the possibility of false positivity due to various reasons like degradation of ssDNA reporter, AuNP-SA, or complementary DNA in the test line.
The rapid emergence of multidrug-resistant A. baumannii has posed a severe threat to worldwide public health. In humans, it can be an opportunistic pathogen that affects immunocompromised persons and is becoming more common as a hospital-borne (nosocomial) infection. Detection of A. baumannii based on the CRISPR/Cas system was developed by Wang et al. This method was integrated with multiplex PCR where simultaneously many genes of β-lactamase, responsible for antibiotic resistance, were detected. Extracted genomic DNA from bacteria was used for the multiplex PCR reaction. When the target gene is present, the system will amplify the target and then initiate Cas12a’s nonspecific ssDNA trans cleavage activity. The ssDNA reporter, conjugated with fluorophore and quencher, was cleaved after the Cas12a-crRNA-DNA assembly, resulting in an increase in fluorescence signals (Figure 3e). Different crRNAs were used to detect different genes. The detection limit was reported to be 50 CFU/mL (Table 1). Integration of multiplex PCR provided an added advantage where multiple targets can be detected at once, provided the individual target specific crRNA needs to be developed.
All the methods discussed above based on CRISPR/Cas-based approaches for bacterial detection are distinct in their own way, employing various Cas enzymes and techniques. According to the ASSURED standards of WHO, all the above-described methods may not fully qualify all the standards, whereas some might be more affordable and others might be more sensitive or have less bulky equipment, or be easy to use in point of care. Likewise, amplification free methods, such as DNA-FISH and E-Si-CRISPR, need less time and fewer components, making them relatively inexpensive. Methods like CASLFA, APC-Cas, CIA, and CRISPR-MTB can also be conducted utilizing the cost-effective isothermal amplification technique. The CASLFA, CIA, and CRISPR/Cas12a-Powered Dual-Mode Biosensor have naked eye readout capabilities requiring no expensive equipment. The APC-Cas technique also does not need bacterial isolation or DNA extraction, which contributes to its low cost. The turnaround times for the methods described above ranged from 30 to 140 min. The CRISPR-mediated DNA-FISH for Methicillin-resistant Staphylococcus aureus (MRSA) detection has the quickest turnaround time of 30 min as it is an amplification free method. In terms of the analytical sensitivity of the CRISPR/Cas assays, the limit of detection (LoD) was reported in different units: copies/mL, CFU/mL, moles, and molarity. The majority of methods reported the LoD in CFU/mL, with the lowest reported LoD being 1 CFU/mL for both CRISPR-Cas12a-powered dual-mode biosensor and CIA (Table 1). APC-Cas and CASLFA was reported to be 1 CFU and 150 copies (Table 1). Between dCas9 and E-Si-CRISPR, where both the LoD were reported in molar concentration, the lowest being for E-Si-CRISPR that is up to 3.5 fM (Table 1). The LoD of CAS-EXPAR was estimated to 0.82 amol. Expressing the sensitivity with different units is misleading; therefore, it is hard to compare among methods. In the case of the bacterial detection method, it would have been helpful for the readers, if the sensitivity could have been reported to the standard unit that is CFU/mL. The CRISPR-based methods reduced the need for large equipment, which is a notable feature that has significant possibilities for field implementation, particularly for controlling epidemic outbreaks in resource-limited areas. CRISPR/Cas systems make it easier to create a wide range of readout signals from fluorescence to naked eye detection. Methods based on LFA such as CASLFA and CIA might have higher utility at the patient point of care. Though CRISPR/Cas-based pathogen detection technologies have exceptional sensitivity and specificity, there are yet many scopes for future advancements. Due to the concerning inherent off-target impact of CRISPR-based detection, improved specificity is the utmost requirement for practical detection approaches. For example, outside of the PAM-proximal 5–12 bp seed areas, Cas9-mediated cleavage is very tolerant to mismatches, and dCas9 off-target binding is random, which can weaken the analytical specificity and sensitivity. In the past few years, Cas9 variants with reduced off-target cleavage, such as SpCas9-HF1, eSpCas9 (1.1), and HypaCas9, have been developed, thereby providing potential future solutions for the off-target effect of Cas9. Therefore, for Cas enzymes, future research should be concentrated on high-fidelity nucleic acid detection to minimize off-targeting. A more user-friendly one-step diagnostic that comprises pathogen nucleic acid release, preamplification, CRISPR/Cas-induced reaction, and signal readout should be developed in the future. For example, several simple formats, such as paper-based biosensors with visual readout (NASBACC, SHERLOCK, DETECTR), pathogen detection without nucleic acid extraction (HUDSON), and a single-tube test combining isothermal amplification and Cas-mediated reaction, have been developed, which might be combined for more simplicity, affordability, and user-friendliness. More equipment-free approaches for signal readout, such as lateral flow assay or naked-eye view under light, should be introduced, as they might be easier at the point of care. Currently, the CRISPR-based detection system must be stored and delivered in a cold chain, which is an inconvenience in many remote areas. The NASBACC detection system with freeze-dried reagents and the SHERLOCK system with freeze-dried and paper spotting reagents showed long-term storage and transport. Therefore, more storage and transportation strategies for the CRISPR-based reaction kit should be developed.
CRISPR/Cas-mediated detection system is a very powerful and advanced technique with high specificity and sensitivity. Therefore, this CRISPR/Cas technique could be of high potential for early diagnosis in the present emerging scenario of antibiotic resistance. In most of the techniques discussed above for pathogenic bacterial detection, there are different amplification techniques like PCR, LAMP, RCA, and SDA integrated along with the CRISPR/Cas system. Previously, only positive amplification was sufficient for detection, but due to the high rate of false positivity, reliance toward only amplification to detect pathogens accurately becomes untrustworthy. Therefore, target amplification followed by CRISPR/Cas as biosensor-mediated detection of pathogenic bacteria can make the process more robust, reliable, sensitive, and specific. |
||
PMC9648124 | Graciela M. Escandar,Alejandro C. Olivieri | A Critical Review on the Development of Optical Sensors for the Determination of Heavy Metals in Water Samples. The Case of Mercury(II) Ion | 27-10-2022 | Recent publications are reviewed concerning the development of sensors for the determination of mercury in drinking water, based on spectroscopic methodologies. A critical analysis is made of the specific details and figures of merit of the developed protocols. Special emphasis is directed to the validation and applicability to real samples in the usual concentration range of mercury, considering the maximum allowed limits in drinking water established by international regulations. It was found that while most publications describe in detail the synthesis, structure, and physicochemical properties of the sensing phases, they do not follow the state of the art in the analytical developments. Recommendations are provided regarding the proper method development and validation, including the setting of the calibration concentration range, the correct estimation of the limits of detection and quantitation, the concentration levels to be set for producing spiked water samples, the number of real samples for adequate validation, the comparison of the developed method with a reference technique, and other analytical features which should be followed. | A Critical Review on the Development of Optical Sensors for the Determination of Heavy Metals in Water Samples. The Case of Mercury(II) Ion
Recent publications are reviewed concerning the development of sensors for the determination of mercury in drinking water, based on spectroscopic methodologies. A critical analysis is made of the specific details and figures of merit of the developed protocols. Special emphasis is directed to the validation and applicability to real samples in the usual concentration range of mercury, considering the maximum allowed limits in drinking water established by international regulations. It was found that while most publications describe in detail the synthesis, structure, and physicochemical properties of the sensing phases, they do not follow the state of the art in the analytical developments. Recommendations are provided regarding the proper method development and validation, including the setting of the calibration concentration range, the correct estimation of the limits of detection and quantitation, the concentration levels to be set for producing spiked water samples, the number of real samples for adequate validation, the comparison of the developed method with a reference technique, and other analytical features which should be followed.
Safe and readily available water is important for public health. Drinking water may contain small amounts of heavy metals, some of which are known to be essential for human health. However, an excess of essential metal ions, or even small quantities of other ions may have a serious negative impact. The presence of heavy metals in water is due to several reasons: (1) the release into the environment by natural causes; (2) anthropogenic activities, such as industrial activities generating wastes (electroplating, metal smelting and chemical industries); or (3) poorly treated wastewaters. There is a growing number of publications devoted to the development of new spectroscopic platforms for the determination of inorganic ions in samples such as drinking and natural waters, foodstuff, etc., which are of high importance for consumer safety. The methods are based on either absorptimetric or emissive probes, which are developed in such a way that they are highly selective and proposed to be sufficiently sensitive to the presence of specific metal ions in the studied samples. The advantages of the proposed sensors include simplicity, low cost, and speed. Typically, the concentrations to be measured in the latter samples are small, requiring limits of detection which should be compatible with the maximum levels set by international regulating agencies. Therefore, efforts are directed to maximize the sensitivity of the analysis, by incorporating a series of clever approaches in the new developments. Due to the large number of papers published in recent years on the subject, the present report is concerned with a specific case: the determination of mercury in drinking water. This analyte was selected because of the challenges in measuring very low concentrations due to its toxicity. However, most of the considerations regarding chemical analysis at low concentrations also apply to other heavy metal ions, anions and organic pollutants in environmental samples, foodstuff, etc. The sources of mercury in water include the erosion of natural deposits, the discharge from refineries and factories, and the runoff from landfills and croplands. For metal ions and other contaminants, the US Environmental Protection Agency (EPA) sets two limits: (1) the Maximum Contaminant Level Goal (MCLG), which is the level of a contaminant in drinking water below which there is no known or expected risk to health; and (2) the Maximum Contaminant Level (MCL), which is the highest level of a contaminant that is allowed in drinking water and is an enforceable standard. MCLs are set as close to MCLGs as feasible using the best available treatment technology and taking cost into consideration. In the case of mercury, the MCLG and MCL are both 2 μg L–1 (2 ppb, 10 nM). Below this level of mercury in drinking water, there is no known or expected risk to health. Otherwise, serious problems may arise, the main of which is kidney damage. The analytical method recommended by the EPA for the determination of mercury in drinking water is cold vapor atomic absorption spectrometry. The concentration range of the method is 0.2–10 μg L–1 (i.e., the lowest calibration concentration is well below the MCL for Hg in this type of samples). It is recommended to estimate the method detection limit (MDL) as 3 times the standard deviation of 7 replicates of a sample prepared to contain twice the concentration corresponding to a rough estimation of the true LOD. However, modern IUPAC recommendations use a different concept for the LOD estimation, which will be discussed below in detail. It is apparent that both the LOD and LOQ must be low enough to detect and measure mercury concentrations at the required levels according to official regulations. If the LOQ of a new method is larger than the MCL, then it should probably be called a method to detect Hg in highly contaminated water samples and may require a substantial preconcentration step for analyzing drinking water. Nevertheless, even if the detection capability of a method is adequate, it should be appropriately validated to be considered reliable. In the present report, we review a number of selected publications on the development of spectroscopic methods for determining Hg in water samples, with emphasis on the requirements to be met for achieving the intended purpose.
Analytical reports should not only describe with particular detail the synthesis, structure, and physicochemical properties of a new chemosensor or probe for specific analytes. The analytical section should be as complete as possible, following the standard guidelines which include the figures of merit, the statistical analysis of the results, and the comparison with those provided by previously published methods. Below we summarize the main figures of merit that should be reported and recommendations for the correct statistical analysis and method comparison.
The sensitivity and analytical sensitivity should be reported, preferably including a comparison with the same parameters in previous analytical reports on the same analyte. The sensitivity is defined as the slope of the calibration graph; however, the analytical sensitivity is preferred for method comparison because it is independent of the measured signal. The latter is defined as the ratio of the sensitivity and the instrumental noise level and has inverse concentration units.
Most authors still employ the correlation coefficient (R2) of the calibration line as an indicator of the linear relationship between signal and concentration. IUPAC discourages this practice, recommending the estimation of an F value as the ratio of the residual variance of the calibration line to the variance of the instrumental noise. Comparison should be made with the critical F value, as detailed in a IUPAC report. Other relevant discussions can be found in the literature.
The definition of the LOD has been evolving with time, from the old concept based on three times the standard deviation for the blank, to the modern view considering both Type I and Type II errors, also called α and β, or false positives and false negatives. Furthermore, error propagation should consider the uncertainties in the measurement of the test sample and also those coming from the calibration phase. Specifically, the modern IUPAC recommendation first requires to define the limit of decision (LD), which only considers a risk of false positives (Figure 1, green-shaded area). The limit of detection is then defined as the level for which the risk of false negatives has a probability β, corresponding to the red-shaded area in Figure 1. According to this latter figure, the expression for the LOD iswhere tα,ν and tβ,ν are student coefficients with ν = N – 2 degrees of freedom and probabilities α and β respectively, N is the number of calibration samples, σc,0 and σc,LOD are the concentration standard errors at the blank and LOD levels, m is the slope of the calibration line, and sy/x is the residual standard deviation. Assuming σc,0 = σc,LOD, the 95% confidence level (α = β = 0.05) and a large number of degrees of freedom, the right-hand side of eq 1 is obtained, where h0 is the blank leverage value given bywhere c̅cal is the mean calibration concentration and cn represents the concentration of the analyte in the nth calibration sample. In some cases, authors refer to the IUPAC definition of the LOD but still use the old definition. The following important remarks by IUPAC should be considered: “It is accordingly recommended that...the approximate detection limit (is) calculated as 3S0. Note that with the recommended minimum number of degrees of freedom, this value is quite uncertain, and may easily be in error by a factor of 2. Where more rigorous estimates are required (e.g., to support decisions on based on detection or otherwise of a material), reference should be made to appropriate guidance (see, e.g., refs 22, 23).” In these latter references the new definitions of the LOD and LOQ are provided. Another relevant issue is the use of the log transformation for coping with nonlinear signal-concentration relationships. However, it is preferable to employ the nonlinear calibration line, estimating the LOD according to a recent procedure, adapted from the IUPAC recommendations for linear calibration lines. Due to the presence of additional terms in the square root of eq 1 beyond the classical value of 1, the modern LOD definition is stricter (i.e., larger) than the old one. It is important to notice that all the reviewed publications quote the incorrect detection capability based on the abandoned LOD definition. They are consequently lower (perhaps significantly) than the true LOD values. Even more important than the LOD is perhaps the limit of quantitation (LOQ) defined aswhere the factor 10 sets a relative prediction uncertainty of 10%. The relevance of the LOQ is that the linear range of a developed method for the determination of mercury ranges from the LOQ to the upper concentration limit where the linear signal-concentration relationship holds. To be able to quantitate Hg below the levels set by official agencies, a sufficiently low LOQ is required.
It is not uncommon to find reported analytical results with an excessive number of significant figures, which is unreasonable. In general, all results should be reported with a number of significant figures compatible with the associated standard error. In the case of uncertainty measures (standard errors, root-mean-square errors, relative prediction errors) and also in the case of parameters derived from uncertainties (detection limit, quantitation limit), they should be reported with one or at most two significant figures. A good rule of thumb is to use two significant figures when the first one is 1, or when the first is 2 and the second is smaller than 5; in the remaining cases, a single significant figure should be reported.
The number of studied samples is usually too small to gather reliable information on the applicability of the method. A new analytical development should include the validation of the protocol with a number of test samples (preferably certified reference materials or real samples, not artificially laboratory-prepared or spiked samples), comparing the determined concentrations of the analyte with those provided by a reference standard technique. The estimated recovered values should be compared with reference values. Statistical tests should be applied to assess whether a recovery is not statistically different than 100%, reporting both the experimental and critical t values. It should be noticed, however, that these tests assume certain conditions that the data should fulfill, such as constant variance.
When comparing the average analyte errors estimated by different methods, the conclusions are often drawn on a visual basis. This impression may be false, and an objective statistical technique is required to compare error values, to be able to conclude that one error is significantly smaller than a second one. One interesting alternative is the randomization test described by van der Voet. The latter paper provides an easily implementable computer code in its Appendix.
The conventional analytical approaches for Hg(II) analysis include atomic absorption spectroscopy, inductively coupled plasma mass spectrometry, and atomic fluorescence spectrometry. Because these methods are rather complex, expensive, nonportable, and require specialized laboratories, the current scientific efforts are directed to developing efficient chemical sensors as alternatives for the determination of mercury in environmental samples. In general, these sensors display analytical features that make them attractive and useful to be implemented both in the analytical laboratory and in the field. For the present discussion, we have selected and analyzed some reported optical mercury sensors where absorption, fluorescence, and/or surface enhanced Raman scattering are the measured signals. Most optical sensors for the detection of Hg(II) in aqueous media involve the use of noble metal nanostructures due to their unique physicochemical properties. For example, they have the ability to support surface plasmons which are generated by the coupling between the incident electromagnetic waves and the conduction electrons. In other words, when the electromagnetic radiation of a certain frequency interacts with the metal nanoparticles, their conduction electrons collectively oscillate on a metal/dielectric interface. This resonance is called surface plasmon resonance (SPR) or, more precisely, local SPR (LSPR) because nanoparticles are involved. Gold- and silver nanoparticles, either bare or with functionalized surfaces, are the two most widely used plasmonic nanomaterials, which have demonstrated sensitive LSPR signals used for the selective detection of Hg(II) ion. The LSPR spectra can be modified by tailoring the size, the shape (rod, cube, disc, etc.), the structures (shell, cage, tube, etc.), and the environment of the metal nanostructures. The LSPR sensing mechanism can be produced through the aggregation of the nanoparticles or by changes in their refractive index. The aggregation of nanoparticles, generally functionalized with oligonucleotides, oligopeptides, or chemical functional molecules, is promoted by the Hg(II) ion through its affinity with specific functional groups of the surface, producing a shift of the LSPR band as a result of the strong coupling between the localized SPRs of the nanostructures. On the other hand, refractive index sensors are based on the LSPR peak shift due to the variation in the local refractive index induced by a mercury amalgamation on the nanoparticle surface. In this case, the Hg(II) ion is reduced with an adequate reagent (NaBH4, citrate, etc.) and the generated Hg(0) is strongly bonded onto the surface of either Au- or Ag- based nanomaterials to form an amalgam, shifting the LSPR spectrum of the nanoparticles to shorter wavelengths. An alternative colorimetric assay is based on the inhibition of the peroxidase-like activity of noble metal nanostructures by the Hg(II) ion. The reaction between the chromogenic substrate 3,3′,5,5′-tetramethylbenzidine (TMB) and H2O2 is generally chosen for these probes. When the catalytic activity of the nanostructure decreases due to the presence of the Hg(II) ion, the concentration of the oxidation product decreases, leading to a color change of the system. In other cases, the presence of mercuric ion produces the opposite effect to that described above. For example, the formation of either the stable thymine complex (T-Hg2+-T) or amalgams with noble metals, can reverse the inhibitory effect that certain molecules or nanoparticles have on the oxidase mimetic activity of specific nanostructures. In the presence of Hg(II) ions, the oxidase-like activity of the indicated nanostructures is restored and the concentration of the oxidation product increases. Fluorescent sensors are constituted by nanomaterial-bound fluorescent units (the fluorophores). When the cation binds to the sensor, the fluorophore photophysical properties change due to different processes (photoinduced electron transfer or charge transfer, energy transfer, excimer formation, etc.), leading to either a fluorescence intensity enhancement or to a quenching effect. Another class of optical sensors for Hg(II) ion resort to surface enhanced Raman spectroscopy (SERS), combining laser spectroscopy with the optical properties of nanosized noble metal structures, resulting in significantly increased Raman signals. The SERS methods for the detection of Hg2+ are based on either the specific interaction between the ion and Raman active molecules exhibiting a strong and specific SERS spectrum (reporters) or the formation of the T-Hg2+-T complex. The detection of the Hg(II) ion is achieved by enhancing or turning off the SERS signals of the Raman reporters. The DNA-based T-Hg2+-T system relies on the coordination chemistry of the metal ion with the nucleotide and the formation of nanoparticles aggregates with the concomitant changes in the SERS intensity. Some research groups have proposed the mercury determination by simultaneously measuring two types of signals with the same sensor, one detection mode generally being superior to the other. However, in some of these dual methods one of the two types of signal results in an unfeasible LOD for the determination of low levels of Hg(II) ion. Thus, it would be more correct to characterize the method with the signal that allows achieving the analytical purpose rather than as a dual one.
As indicated above, numerous scientific articles report the determination of the mercury(II) ion in water samples. For the sake of brevity, we have randomly selected a number of publications on optical mercury sensors. The information from the literature search was compiled in Tables 1–4 following an increasing order of publication year and were organized according to the measured spectroscopic signals, as detailed in Section 3. Thus, whereas Tables 1, 2 and 3 describe sensors based on absorption, fluorescence, and SERS signals, respectively, Table 4 includes sensors displaying two different types of signals. The first and second columns of these tables show the operational basis of the sensors and the relationship between the measured signal and the mercury concentration. In the next column, the analytical efficiency of each of the proposed methods is briefly mentioned, with an emphasis on both the linear range and the attained limit of detection. Regarding the values of these latter parameters in the tables, it is important to point out that none of them was calculated using the modern IUPAC definition (see above). Nevertheless, to assess and compare the detection capabilities of the sensors, the reported LODs were used in our discussion. To facilitate the comparison among the different works, the concentration units were unified to micrograms per liter (parts-per-billion). The tables also include the type of studied water sample, the applied pretreatment (if any), the added mercury concentrations to the spiked samples and/or the original values found in real samples without addition of the analyte. Since in these latter samples the Hg(II) ion content is unknown, it is mandatory to compare the obtained results with those provided by a reference method. From the 42 selected methods shown in the tables, only two of them meet the correct analytical standards for the validation protocol in real samples. Qi et al. developed a colorimetric mercury sensor blocking with the Hg(II) ion the inhibition produced by oligonucleotides on the peroxidase-mimicking catalytic activity of graphene oxide/AuNPs in the oxidation of 3,3′,5,5′-tetramethylbenzidine (TMB) by H2O2. Under the optimized conditions, a calibration line was built with the absorbance at a proper wavelength as a function of the mercury ion concentration. Seven replicate calibration samples containing Hg(II) concentrations from 1.04 μg L–1 to 24 μg L–1 were employed. Although the limit of detection (LOD = 0.076 μg L–1) was calculated with the old LOD concept, the value was below the US EPA maximum level in drinking water of 2 μg L–1. The method was successfully applied in both spiked and real samples without externally added analyte, and the estimated concentrations were compared with those provided by a reference method. On the other hand, Zheng et al. quantified mercury in tap, pond, and river waters using a fluorescence sensor. The latter was based on the capability of the Hg (II) ion to remove the quenching effect AuNPs modified with thioglycolic acid on the fluorescence of rhodamine B by disrupting the absorption of the dye molecules on the surface of the nanoparticles. The calibration was performed under optimal experimental conditions, measuring the fluorescence intensity at the emission maximum as a function of analyte concentration, ranging from 0.2 to 6.2 μg L–1. The LOD calculated as the ratio between three times the standard deviation of the blank signal, and the slope of the calibration line was 0.08 μg L–1. This approach allowed determining the original Hg(II) ion levels in real water samples (verified by atomic absorption spectroscopy) and obtaining good recoveries at the low concentrations added to the studied matrices (1 and 2 μg L–1). Two additional reports fulfilled the requirements of achieving an appropriate LOD (<2 μg L–1), demonstrating a successful application to both spiked and real samples without externally added analyte and comparing the results with those provided by a reference method, but the methods were applied to too few samples (i.e., three or less). Another group includes the development of 15 sensors with adequate LODs but which were only applied to spiked samples, with five of them tested on very few samples. Although real matrices are preferable to synthetic ones for recovery assays, the validation of a new method should be completed by demonstrating its usefulness in real samples with no external addition of the analyte. Depending on the analytes and/or the investigated matrices, it may not be easy to obtain real samples. Despite this drawback, the authors should make an effort to include them in their working protocols and statistically compare the results with those given by a reference method. In any case, if only spiked samples are included in the analysis, at least 10 different water samples with the addition of various analyte levels are required. This would increase the analytical quality of the newly proposed methods and show that the required standards have been fully achieved. More worrying are the 17 papers where the sensor was employed to determine mercury in samples spiked with unreasonably high analyte concentrations, even when they display LODs which apparently allowed to measure mercury levels below the official maximum. In addition, four of these latter reports studied either one or two spiked samples. Finally, Tables 1–4 show six methods with limited applicability for the determination of mercury ion in drinking water, since they present LODs close to or higher than the target concentration of the metal ion in this type of samples. In addition, these sensors were tested on spiked samples with high concentrations of Hg(II). An issue not considered in Tables 1–4 but of analytical relevance is the evaluation of the method selectivity. Although the chemical sensors are described as having high selectivity toward the analyte under investigation, their behavior in the presence of potential interferents must be reliably demonstrated. The most common interferents for metal ion sensors are other cations that may respond similarly to the analyte by interacting with the sensing unit. Nevertheless, it is also important to analyze the effect of certain anions that may react with the cation producing stable species, seriously affecting the analyte detection. In the specific case of the Hg(II) ion, the presence of chloride should be taken into account. In this regard, all reports shown in Tables 1–4 studied the selectivity by analyzing the effect of potentially interfering cations on the sensor signal. However, not all of them analyzed the influence of the anions. In one study, chloride was removed dechlorinating the water sample by previous boiling. In other cases, the fact that this anion does not interfere was indirectly demonstrated either by using chloride salts to prepare the solutions of the investigated cations or by employing HgCl2 as mercuric reagent.
In this section we provide some general recommendations for the development of methods for the determination of inorganic species in drinking water, using Hg(II) as an example. Most scientific reports on the development of sensing platforms for Hg(II) allocate ca. 90% of the space to the detailed description of the synthesis, preparation, structural study, physicochemical characterization, etc., while the analytical sections are usually very brief. We suggest to expand the latter section following some standard rules for analytical calibration and validation. A first phase of any protocol attempting to develop a methodology for detecting Hg(II) in drinking water is to roughly estimate the LOD. If this latter value is above 2 μg L–1, this means that a preconcentration procedure is required. Once the detectability of the method has been found to be feasible, the following steps should be followed: 1. Prepare and measure a set of calibration standards from 0.1 to 0.2 μg L–1 to ca. 50–100 μg L–1. 2. If the calibration graph is linear, use the IUPAC test for setting the maximum concentration at which linearity is complied. Do not rely on the correlation coefficient. 3. Estimate the LOD using the modern IUPAC definition. 4. Analyze at least 10 spiked water samples, with final concentrations of Hg in the range 0.5 to 5 μg L–1. Compare the estimated value of the added concentrations with the nominal ones using a statistical test. 5. Analyze at least 10 real, unspiked drinking water samples and compare the results with a reference method, i.e., atomic spectrometry. Use suitable statistical tests for the comparison.
In view of the above discussion and findings, the conclusions of the present report are rather disappointing. The proposal of new sensors for measuring the concentrations of heavy metal ions in drinking water, particularly in the case of the highly toxic mercury ions, demands authors to comply with a series of requirements for proper method development, validation, and applicability study. These are not met in the vast majority of the reviewed publications, with only a very small number of reports employing the correct analytical standards. The authors of the present review would like to call the attention of reviewers and editors of international journals on the importance of verifying that submitted manuscripts describing spectroscopic sensors for the determination of low concentration species, from heavy metal ions to all potential organic contaminants, meet modern analytical standards. Otherwise, the usefulness of the proposed sensors would be highly limited. |
||
PMC9648130 | Zhen Qin Aw,Chee Keng Mok,Yi Hao Wong,Huixin Chen,Tze Minn Mak,Raymond T. P. Lin,David Chien Lye,Kai Sen Tan,Justin Jang Hann Chu | Early pathogenesis profiles across SARS-CoV-2 variants in K18-hACE2 mice revealed differential triggers of lung damages | 27-10-2022 | SARS-CoV-2,variants of concern,immune response,cytokine storm,K18-hACE2 mice model | The on-going COVID-19 pandemic has given rise to SARS-CoV-2 clades and variants with differing levels of symptoms and severity. To this end, we aim to systematically elucidate the changes in the pathogenesis as SARS-CoV-2 evolved from ancestral to the recent Omicron VOC, on their mechanisms (e.g. cytokine storm) resulting in tissue damage, using the established K18-hACE2 murine model. We reported that among the SARS-CoV-2 viruses tested, infection profiles were initially similar between viruses from early clades but started to differ greatly starting from VOC Delta, where the trend continues in Omicron. VOCs Delta and Omicron both accumulated a significant number of mutations, and when compared to VOCs Alpha, Beta, and earlier predecessors, showed reduced neurotropism and less apparent gene expression in cytokine storm associated pathways. They were shown to leverage on other pathways to cause tissue damage (or lack of in the case of Omicron). Our study highlighted the importance of elucidating the response profiles of individual SARS-CoV-2 iterations, as their propensity of severe infection via pathways like cytokine storm changes as more variant evolves. This will then affect the overall threat assessment of each variant as well as the use of immunomodulatory treatments as management of severe infections of each variant. | Early pathogenesis profiles across SARS-CoV-2 variants in K18-hACE2 mice revealed differential triggers of lung damages
The on-going COVID-19 pandemic has given rise to SARS-CoV-2 clades and variants with differing levels of symptoms and severity. To this end, we aim to systematically elucidate the changes in the pathogenesis as SARS-CoV-2 evolved from ancestral to the recent Omicron VOC, on their mechanisms (e.g. cytokine storm) resulting in tissue damage, using the established K18-hACE2 murine model. We reported that among the SARS-CoV-2 viruses tested, infection profiles were initially similar between viruses from early clades but started to differ greatly starting from VOC Delta, where the trend continues in Omicron. VOCs Delta and Omicron both accumulated a significant number of mutations, and when compared to VOCs Alpha, Beta, and earlier predecessors, showed reduced neurotropism and less apparent gene expression in cytokine storm associated pathways. They were shown to leverage on other pathways to cause tissue damage (or lack of in the case of Omicron). Our study highlighted the importance of elucidating the response profiles of individual SARS-CoV-2 iterations, as their propensity of severe infection via pathways like cytokine storm changes as more variant evolves. This will then affect the overall threat assessment of each variant as well as the use of immunomodulatory treatments as management of severe infections of each variant.
The emergence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has resulted in the coronavirus disease 2019 (COVID-19) pandemic that caused major healthcare and economic burden worldwide (1). The infection SARS-CoV-2 caused is highly heterogeneous ranging from asymptomatic and mild to severe and even death (2). In addition, the heterogeneity is further enhanced by the rapid mutation of the virus giving rise to different clades and variants that caused varying pathologies over the course of the pandemic (3). Hence, it is imperative to characterize the differential pathology of SARS-CoV-2 clades and variants and their mechanisms to improve management of the infection, especially for development of therapies that are based on the modulation of these mechanisms. Coronaviruses have been shown to recombine extensively due to their proofreading mechanism that can facilitate homologous recombination events, giving rise to novel variants of varying pathology and severity (4). To date, SARS-CoV-2 is clustered into 9 major clades based on the Global Initiative on Sharing All Influenza Data (GISAID) database (5, 6). Clade L is the first novel viral genome sequence serving as the reference ancestral strain (7), and other representative clades with major genetic mutations include clades S, V, G, GK, GH, GR, GV, and GRY. Among the clades, clade G and its derivatives have rapidly become the dominant globally circulating SARS-CoV-2 that gave rise to the major variants of concern (VOCs) (8), including Alpha (B.1.1.7), Beta (B.1.351), Gamma (P.1), Delta (B.1.617.2) and Omicron (B.1.1.529). The VOCs have caused concerns as they were shown to have increased virulence, greater transmissibility, resistance to antibody neutralization, and evasion to host- and vaccine-mediated immunity (9–11). It is understood that the SARS-CoV-2, especially the VOCs, resulted in different degrees of severity and symptoms in humans and animal models (12, 13). However, due to the rapid evolution of each VOCs and them rarely overlapping in emergence, there have yet to be a systematic comparison of the early triggers of airway and lung damage among the major SARS-CoV-2 variants, especially between all VOCs. Several studies have shown that the expression of pro-inflammatory cytokines is associated with disease progression in COVID-19 patients (14–17). Hypercytokinemia, also known as cytokine storm, is one of the major triggers for tissue damage, multiple organ dysfunction, and lethal outcomes in SARS-CoV-2 infection (18, 19). Further, it is well documented that the build-up of activated neutrophils and macrophages in the airways and lungs following SARS-CoV-2 infection results in increased pro-inflammatory cytokines that contribute to airway pathology (20). On the other hand, the evolution of SARS-CoV-2 VOCs over the course of the pandemic showed great variation in their severity and pathology, which is especially apparent in Omicron VOC, suggesting differential cytokine activation that contributes to their pathogenesis (21). Henceforth, we employed the established K18-hACE2 mouse model to systematically compare the disease phenotypes across clades and variants and correlate them with their respective early response cytokine profiles, to determine involvement of cytokine storm and selected pathways’ contribution to their pathogenesis. We aim to elucidate the evolutionary progression of SARS-CoV-2 pathogenesis to understand the pathology spectrum of the virus, which will help in current and future management and drug discovery effort against the virus.
African green monkey kidney cells, (Vero E6; ATCC CRL-1586™) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM; Cytiva) supplemented with 10% heat-inactivated fetal calf serum (FCS). SARS-CoV-2 viruses were isolated from nasopharyngeal swabs of qRT-PCR confirmed COVID-19 patients and propagated in Vero E6 for infection experiments, where low passages of not more than 4 passages of the virus were used. Genome sequences from the swab samples uploaded to GISAID by the National Public Health Laboratory, National Centre for Infectious Diseases, Singapore, were used to confirm the variants’ identity. A total of eight viruses were used: 1. Clade L (ancestral), Lineage B (EPI_ISL_574502); 2. Clade V, Lineage B.29 (EPI_ISL_493419); 3. Clade G, Lineage B.1.610 (EPI_ISL_443240); 4. Clade GR, Lineage B.1.1 (EPI_ISL_443199); 5. VOC Alpha variant, Clade GRY (thereafter GRY-α), Lineage B.1.1.7, (EPI_ISL_754083); 6. VOC Beta variant, Clade GH/501Y.V2 (thereafter as GH-β), Lineage B.1.351.3 (EPI_ISL_1173248); 7. VOC Delta variant, Clade GK (thereafter GK-δ), Lineage B.1.617.2 (EPI_ISL_2621925); 8. VOC Omicron variant, Clade GRA (thereafter GRA-o), Lineage B.1.1.529 (EPI_ISL_7195620). All virus experiments were performed in NUS Medicine biosafety level 3 (BSL-3) core facility and all protocols were approved by the BSL-3 biosafety committee (BBC) and the institutional biosafety committee (IBC) of the National University of Singapore (NUS).
Eight to nine weeks-old female K18-hACE2 transgenic mice (InVivos Ptd Ltd, Singapore) were acclimatized in the ABSL-3 facility for 72 hours prior to infection and were infected with approximately 1 × 103 PFU of SARS-CoV-2 virus suspension in PBS, via the intranasal route. Baseline body weight were measured prior to virus infection. Body weight, physiological conditions, and survival were monitored daily by two personnel for the duration of the experiment (14-days post infection, dpi) or until the humane endpoint was reached. Each infection group was performed at n=6, except variant GRY-α at n=5, and mock infection at n=3. Scoring of the infected mice physiological conditions were based on five criteria: appearance of mouse coat, level of consciousness, activity level, eye condition, and respiratory quality. Conditions were scored on a scale from 1 to 5, using an observation system adapted from Shrum, et al. (22), with 1 denoting normal physiological state and 5 being the most severe. Additional groups of mice at n=7, except GK-δ at n=5, and GRA-o at n=6 were sacrificed on 4 dpi to assess the viral load in the brain, lung, liver, and spleen; and histological analyses of the lung. Tissues were halved and homogenized in DMEM supplemented with antibiotic-antimycotic and the supernatant was used for plaque assay, while the remaining tissue pellets were kept in -80°C for RNA isolation. The other half of the tissue was fixed with 3.7% formaldehyde (10% formalin) solution for at least 19 h before removal from BSL-3 containment for histological analysis. All animal experiments were conducted in NUS Medicine Biosafety Level 3 (ABSL-3) facility in accordance with NUS IACUC protocol no. R20-0504, using NUS IBC and BBC approved SOPs.
For virus titre determination, viral supernatants from homogenized tissues were serially diluted in 10-fold increments in DMEM supplemented with antibiotic-antimycotic (Gibco). 250 µl of each serially diluted supernatant were added to confluent Vero E6 cells and incubated for 1 h at 37°C with rocking at 15 min intervals. After 1 h of adsorption, virus inoculum was removed, and cells were washed once with phosphate-buffered saline (PBS). Overlay media containing 1.2% microcrystalline cellulose-DMEM supplemented with antibiotic-antimycotic were added to each well and incubated for 3 days at 37°C at 5% CO2 for plaque formation. Cells were fixed in 10% formalin overnight and stained with crystal violet for plaque visualization. Number of plaques were determined, and virus titre of individual samples were expressed in logarithm of plaque forming units (PFU) per organ.
Formaldehyde fixed tissues were routinely processed, embedded in paraffin blocks (Leica Surgipath Paraplast), sectioned at 5 µm thickness, and stained with hematoxylin and eosin (H&E; Thermo Scientific) following standard histological procedures. A multiparametric, semiquantitative scoring system was further used to assess the magnitude of histo-morphological and -pathological changes in lung tissues based on six criteria: inflammatory cell infiltrates, hemorrhage, edema, degeneration of alveolar epithelial cells, parenchymal wall expansion and bronchiole epithelial cell damage (23, 24). For the histopathological parameters, a score of 0 – 3 was ordinally assigned, where 0 indicated normal; 1 indicated less than 10%; 2 indicated 10 – 50%; and 3 indicated more than 50% of lung regions affected. The average of histopathological scores of the mice from each group was taken as the final evaluation index.
The paraffin embedded lung sections were deparaffinized and rehydrated, followed by heat-mediated antigen retrieval process. The sections were then treated with hydrogen peroxide blocking and protein blocking for 10 min, respectively. After blocking the non-specific binding sites, sections were incubated with anti-SARS-CoV-2 nucleocapsid protein monoclonal antibody (1:1000; Abcam), followed by the respective horseradish peroxidase (HRP)-conjugated secondary antibody. The sections were subsequently visualized using DAB solution and counterstained with hematoxylin.
Homogenized lung tissue pellets were subjected to total RNA extraction using QIAGEN RNeasy® mini kit in accordance with manufacturer’s guidelines. RNA purity and concentration were determined using Tecan Infinite® M200 Pro coupled with NanoQuant Plate™. Real-time reverse transcription polymerase chain reaction (qRT-PCR) was performed using QIAGEN RT2 First Strand Kit, QIAGEN RT2 SYBR® Green Mastermix, and QIAGEN RT2 Profiler PCR Array (PAMM-150ZE-4) in accordance with manufacturer’s recommendations.
Murine cytokines mRNA expression levels from cDNA conversion were analyzed using QIAGEN GeneGlobe Design and Analysis hub, and gene expression were expressed as Log2 fold regulations. Expression levels were calculated relative to Gapdh and normalized to the mock infected mice using the ΔΔCT method with the QIAGEN GeneGlobe software. The genes from the QIAGEN panel were further sub-divided into selected pathways that may potentially contribute to tissue damage using DAVID functional annotation tool (25, 26). Pathways were selected from Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases for visualization of enrichment in each pathway across variants.
Analyses were calculated using two-tailed Mann-Whitney U test to evaluate the data obtained using Graphpad Prism version 9.0 with * denoting that p < 0.05, ** denoting that p < 0.01, and *** denoting p < 0.001. For gene expression analysis, p-values were obtained from the analysis with QIAGEN GeneGlobe Design and analysis hub. Only genes fulfilling both criteria of >2 fold-regulation compared to uninfected controls and having a p-value of <0.05 were considered significant.
To characterize the pathogenesis of various SARS-CoV-2 viruses from different clades and VOCs, and their respective mechanisms leading to airway and lung damage, we tracked a total of eight major SARS-CoV-2 iterations – L (ancestral), V, G, GR, GRY-α, GH-β, GK-δ, and GRA-o ( Figure 1 ). The repertoire of our coverage includes the major SARS-CoV-2 clades and variants of the pandemic to date. The mutations across the iterations based on their clinical isolates’ sequences obtained from GISAID were listed in Table 1 .
We first compared the survival of K18-hACE2 mice following infection with the different SARS-CoV-2 viruses. It was observed that, with the exception of V (34% survival) and GRA-o (83.3% survival), 0% of infected mice survived the infection, where they succumbed between 6 to 8 dpi, with GH-β and GK-δ hitting the end point the earliest at 6 dpi ( Figure 2A ). When comparing the Kaplan-Meier survival curves, it was shown that only GRA-o, the latest evolutionary iteration, was significantly different in mice survival compared with all other infected groups except for V, another iteration of the virus where there were mice that survived ( Figure 2B ). The weight loss trends largely followed the survival curves in which rapid weight loss were observed between 4 and 7 dpi for all groups other than V and GRA-o, for which we observed a rebound in weight from surviving mice ( Figure 2C ). We then compared clinical symptom scores (averaged from appearance, eye condition, consciousness, activity, and respiration) and observed that L and GRY-α showed the most symptom presentation, followed by the others that showed moderate symptoms, while GRA-o was shown to be not presenting any clinical symptom throughout the duration of study ( Figure 2D , Figures S1A–E ). We then examined viral titre retrieved from the tissues (lung, brain, liver, and spleen) of SARS-CoV-2 infected mice. No infectious viruses were recovered from liver and spleen homogenates using plaque assay (data not shown). For viral titre in the lungs, high titres of SARS-CoV-2 was retrieved from the lungs of L, V, G, GR and GRY-α infected mice, at geometric mean titres of 4.27 × 105, 4.55 × 104, 2.75 × 105, 1.09 × 106, and 3.12 × 105 PFU/organ, respectively. Titres from V are slightly lower, while GR slightly higher when compared with L infected group (p < 0.05). On the other hand, significantly lower titre was observed in GH-β, GK-δ and GRA-o when compared with L, at 6.86 × 102, 7.2 × 101 and 1.1 × 101 PFU/organ, respectively (p < 0.05) ( Figure 2E ). As for viral titre in the brain, comparatively high titre was found in the brain tissue of L, GRY-α and GH-β infected groups, at 1.93 × 105, 6.02 × 103 and 3.4 × 103 PFU/organ, respectively; while significantly lower titre in the brain was retrieved in brain tissue of G, GR and GK-δ at 1.03 × 102, 4.4 × 101, and 8.7 × 101 PFU/organ, respectively. No virus was retrieved from the brain of GRA-o infected mice ( Figure 2F ). Notably, unlike in the lung, viral titre in the brain was more variable where about half of the mice did not have viral titre retrieved from the brain in most groups, regardless of the titre observed.
From the lung tissue collected at 4 dpi, it was observed that histopathological damage of SARS-CoV-2 variants was largely similar, with histopathological index score ranging between 0.63 to 1.69. Tissues from GR and GK-δ infected mice were on the higher end of the scores while those from L and GRA-o were with lower scores ( Figure 3A ). The factors that likely contributed to the higher histopathological index were inflammatory cell infiltrates, where the trend was consistent with the overall index compared with other factors ( Figure 3B ; Figures S2A–F ). On the other hand, no histological abnormality was observed in the bronchiole epithelial cells following all variants’ inoculation, and neither bronchitis nor bronchiole epithelial denudation was present in the lung tissues (data not shown). Next, we correlated the extent of infected cells and viral presence in the lungs of the ancestral L infected group, with the VOC infected groups. Using IHC of SARS-CoV-2 nucleocapsid protein, we observed that the viral distribution in the lungs at 4 dpi varied greatly ( Figures 3C, D ). GK-δ, notably, showed major nucleocapsid protein presence in the lung (93.2% coverage), contrary to the low infectious titre detected from the infected murine lungs while GRA-o had almost no nucleocapsid protein detected in the lung tissues (5.83% coverage). Our observation suggested that there was no correlation between nucleocapsid coverage, which are produced in excess during coronavirus infection (27, 28), in the lungs and the lung histopathological damage index between variants.
To further elucidate potential mechanisms in each of the SARS-CoV-2 viruses that led to their respective phenotypes, we performed a qRT-PCR array on 84 genes related to inflammatory responses from lung tissues of SARS-CoV-2 infected K18-hACE2 mice. After taking into account fold-regulation and statistical significance (>2 fold-regulation and p< 0.05), the cytokine profiles between the infected groups differed, some significantly, from each other. We noted the groups with the larger significant changes from the ancestral L virus (18 up; 0 down) were G (23 up; 6 down), GR (21 up; 2 down), GRY-α (26 up; 2 down), GH-β (21 up; 5 down) and GRA-o (26 up; 1 down). On the other hand, V (4 up; 0 down) and GK-δ (6 up; 4 down) had fewer significant changes compared with L ( Figure 4 ). The exact changes in fold-regulation of the genes tested were shown in Table S1 . In general, the host responses of the SARS-CoV-2 infection of all groups tested deviated significantly from the ancestral L virus, with some having unique significant responses as observed in G (Lta, Tnf, Pf4), GH-β (Il24, Il1a, Mstn, Il22), GK-δ (Il7, Tnfsf11b, Il18), and GRA-o (Ccl22, Bmp4, Tnfsf10, Mif, Hc, Bmp7, Gpi1, Cx3cl1). In view of the differences present, particularly in GRA-o which has the most of them, further dissection on how the differential cytokine changes led to the pathogenesis of current and future variants is thus warranted. As cytokine storm was perceived to be a major trigger of lung damage in COVID-19 (18, 19), we grouped the panel of genes we tested into pathways associated with cytokine storm using DAVID. We selected pathways involved directly in cytokine storm (19), and associated pathways like MAPK cascade, TNF responses, neutrophil chemotaxis, IL-17 regulation, negative regulation of inflammatory response, and T-cell mediated cytotoxicity to determine how much they played a role in tissue damage for each virus iteration ( Figure 5 ). Firstly, in the cytokine storm pathway, we observed that for iterations that come after L and V, more genes such as Ifng, Il12b and Il10 were found to be significantly differentially expressed in G, GR, GRY-α and GH-β, before being reduced again in GK-δ and GRA-o ( Figure 5A ). Similar trends were observed in TNF response and neutrophil chemotaxis pathways in genes like Xcl1, Ccl1 and Ccl11 ( Figures 5B, C ). The MAPK cascade pathway was prominently differentially expressed in G, GR, GRY-α and GH-β, but in this case, L and GK-δ as well ( Figure 5D ). On the other hand, negative regulation of inflammatory responses early in infection were linked with cytokine storm (29), and we found that G, GR, GRY-α and GH-β further have differentially up-regulated expression of Il10 ( Figure 5E ). IL-17 pathway, associated with neutrophil infiltration with genes from the pathway like Il15 and Il12b were up-regulated in all groups except V and GK-δ ( Figure 5F ). Finally, T-cell mediated cytotoxicity pathway was found to be prominently changed for G, GR, GRY-α, GH-β, and GRA-o ( Figure 5G ). Interestingly, only GRA-o had a further significant up-regulation of Il12a. Overall, genes from cytokine storm and associated pathways (Ccl2, Ccl7 and Ccl12) were found to be differentially regulated in infections of L, G, GR, GRY-α, and GH-β while not as strongly involved in V and GK-δ. GRA-o, while having similar trend of cytokine storm associated pathway gene changes, had more subdued expression, and a slight difference in T-cell mediated cytotoxicity (Il12a).
The use of K18-hACE2 murine model to understand SARS-CoV-2 infection has been instrumental to the understanding of SARS-CoV-2 pathogenesis mechanisms since the start of the pandemic, including comparisons of early clades’ differential severity (13, 30–34). Moreover, the K18-hACE2 mice were shown to be simulating severe SARS-CoV-2 infection useful in the identification of immune responses leading to severe outcome (35). Hence, here we reported results using the model to correlate early cytokine profiles (4 dpi) of early SARS-CoV-2 evolutionary iterations (L, V, G, GR), and the VOCs Alpha (GRY-α), Beta (GH-β), Delta (GK-δ) and Omicron (GRA-o), with their infection outcome. Interestingly, the majority of the pathogenesis mechanism and infection outcome remained largely similar up until the Beta VOC, before major changes occur in VOCs Delta and Omicron that emerged later. This may be partially due to the larger number of mutations accumulated ( Table 1 ) for Delta and Omicron VOC, which potentially altered their biology significantly. Our study demonstrated that early cytokine profiles, viral titre and histopathological damage worked in concert to influence and determine the infection outcome. Overall, the infection outcome remained largely similar from L to GH-β Beta VOC, and then deviated in VOCs that emerged late in the pandemic i.e. GK-δ Delta and GRA-o Omicron VOCs. Among the early clades and VOCs up to Beta, even the milder clade V virus with ~30% survival rate had similar rates of neurotropism based on viral titre from brain tissue, sufficiently high viral titre in the lungs, and cytokine profiles consistent with previous report (36). While clade V virus had fewer significantly differentially expressed genes, the major genes involved in cytokine storm; TNF pathway and neutrophil chemotaxis, Ccl2 and Ccl7 remained, suggesting similar, but milder presentation of pathology for its infection outcome. The subsequent clade and VOCs in the earlier part of the pandemic, G, GR, GRY-α and GH-β had further inflammatory suppression gene Il10 differentially up-regulated that potentially augmented their pathology slightly, as seen in the histopathological index. Furthermore, the viruses from earlier in the pandemic, up to VOC GH-β, showed higher clinical presentation scores, which may be linked to neurotropism, prominent in these groups of infections. On the other hand, VOCs that emerged later in the pandemic, GK-δ Delta and GRA-o Omicron, followed a different pathogenesis mechanism that resulted in tissue damage. Firstly, neurotropism was observed to be less apparent in Delta, and completely absent in the Omicron variant in the K18-hACE2 mice. It was observed that mechanisms that may contribute to cytokine storm were also less apparent in GK-δ and GRA-o infection. In addition, immune suppression, a mechanism common in SARS-CoV-2 (37), was observed strongly in the GK-δ Delta variant, albeit different from Il10 mediated suppression observed in clades and VOCs that came before. The Delta variant in our study instead showed an almost universal suppression of host responses early post infection at 4dpi (with very little significant differential regulation). This suppression may potentially contribute to the increased nucleocapsid expression and inflammatory infiltrates in the lungs, which may suggest diffusely infected cells in the lungs. On the other hand, the GRA-o variant showed the most varied response among the variants where we observed the least tissue damage, low viral titre and lack of neurotropism. This observation is in line with the finding that Omicron variant has more preference to the upper airway than the lungs (38). In addition, the cytokine profile of our Omicron GRA-o variant infection showed a response largely different than that of its predecessors. It had generally weaker changes in expression of cytokine storm and related pathway genes, coupled with moderately stronger T-cell mediated cytotoxicity gene expression in Il12a and Il12b. The Omicron cytokine profile suggests a different pathogenesis that may contribute to the overall milder tissue damage congruent with other studies to date (21, 39). The effects of the major changes in Omicron cytokine profiles require further studies to validate their correlation with disease outcome, as well as potential interaction with risk factors and chronic conditions (e.g. asthma) due to their major differences and lower lethality. The finding from our study is crucial in the understanding of SARS-CoV-2 pathogenesis mechanisms. It showed the evolutionary potential of SARS-CoV-2 where its ability to accumulate high number of mutations may result in highly variable pathogenesis and infection severity in the variants tested and may further change in future variants (4, 40). This is particularly evident in our elucidation of Delta and Omicron cytokine profiles suggesting a major change in the biology of the VOCs late in the pandemic (41, 42). Indeed, the emergence of Delta and Omicron variants both caused major outbreaks that dwarfed those that of their predecessors (43, 44). Studies have shown that Delta and Omicron VOCs has evolved into their distinct evolutionary group and therefore may explain the significant differences of their infection profiles and pathogenesis mechanisms (41). In addition, other reasons such as gestation and adaptation in immunocompromised patients (45), spill-over and spill-back to and back from animal reservoirs (45–47), and potential vaccine induced selection pressure (48), may all contribute to the accumulation of these mutations and the differential pathogenesis mechanisms. Overall, our study inferred that the adaptation of the virus may likely lead to emergence of future variants having mutations that drives their pathogenesis differently than existing SARS-CoV-2 viruses. Therefore, this implied that management strategies based on immunomodulation and mediation of host responses have to be revisited in different variants, mirroring the management of influenza, where propensity of cytokine storm differs (49). Finally, our finding may also partially explain the continued efficacy of broad spectrum immunomodulators like corticosteroids; but not more targeted therapies targeting specific pathways due to the potential differences in different SARS-CoV-2 iterations’ pathogenesis mechanisms. Our study, however, is not without its limitation. We only elucidated the histopathology and cytokine profile at 4 dpi, that only gave us insights of host response early during infection. This also created potential disconnect such as the case between the high Delta nucleocapsid detection in IHC compared to its low lung infectious viral titre. Nevertheless, high viral protein, especially in the case of nucleocapsid, which were produced in excess during infection (28), may not necessarily translate to infectious viral titre. In addition, we only use female mice for our comparative study and thus may not capture any potential sex specific differences between the cytokine profiles of different virus iterations (50). However, a previous study showed that cytokine and chemokine responses between male and female were largely similar, where the major differences lie at the magnitude and timing of the responses (51). Therefore, our study remained representative of the differences in cytokine profiles between variants despite only testing it in female mice. Nevertheless, future studies with male mice can be performed to investigate the extent of differential cytokine expression levels that the different SARS-CoV-2 virus iteration can induce. Finally, we also did not manage to assess the protein levels of the cytokines in our study. Despite the limitations however, our data showed that, by systematically comparing the profiles across the major evolutionary iterations of SARS-CoV-2, a clear heterogeneity of SARS-CoV-2 evolution and adaptation was observed, especially in VOCs Delta and Omicron that emerged later In conclusion, our study provided insights on the evolution of pathogenesis and its potential mechanisms of SARS-CoV-2 throughout the course of the pandemic. We have observed that adaptation throughout the pandemic can give rise to variants of very different pathogenesis mechanisms that contribute to tissue damage. Our study thus emphasized the importance in differentiating host responses to different SARS-CoV-2 variants. This is especially true as we enter the endemic phase, where the surveillance of differential host responses between variants will become increasingly crucial when assessing future variants, for their threat and impact, as well as when devising immunomodulatory treatments for severe infections.
The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding authors.
The animal study was reviewed and approved by NUS Institutional Animal Care and Use Committee (IACUC) under protocol no. R20-0504.
CKM and JC conceived and designed the experiments. CKM, ZA, TM, DL, and RL contributed to the virus isolation and sequencing. CKM, ZA, HC, and YW performed the experiment. CKM, YW, and KT performed the histological analyses. ZA, CKM, and KT performed the cytokine analyses. CKM, ZA, YW, KT, and JC performed the data analyses. CKM, ZA, YW, KT, and JC wrote the manuscript. All authors have read and approved the final version of the manuscript prior to submission.
This research was supported by the following grants: NUHSRO/2020/066/NUSMedCovid/01/BSL3 Covid Research Work, NUHSRO/2020/050/RO5+5/NUHS-COVID/4, Ministry of Education, Singapore MOE2017-T2-2-014, Singapore NMRC Centre Grant Program – Diabetes, Tuberculosis and Neuroscience CGAug16M009, Ministry of Health MOH-COVID19RF2-0001.
We are grateful to the National University of Singapore, Yong Loo Lin School of Medicine BSL-3 Core Facility for their support with this work.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648141 | Yujia Tang,Jiawei Hu | Enhanced Methane Production from Sludge Anaerobic Digestion with the Addition of Potassium Permanganate | 26-10-2022 | This work aims to reveal the effect of potassium permanganate (KMnO4) on the sludge anaerobic digestion process, as well as the relevant mechanisms. Experimental data showed that the biomethane production was gradually increased from 159.3 ± 3.0 to 211.5 ± 5.1 mL/g VSS (volatile suspended solids) when the KMnO4 content was increased from 0 to 0.08 g/g VSS, with an increasing rate of 32.8%. A further increase in the KMnO4 dosage however resulted in the decline of the methane yield. First-order kinetic model analysis indicated that higher methane production potentials and hydrolysis rates were achieved in KMnO4-added reactors than in the control. Mechanism analysis demonstrated that KMnO4 not only efficiently disintegrated the sludge flocs, which resulted in the increased contents of dissolved organics, but also enhanced the proportion of biodegradable substances in the sludge liquor. Meanwhile, the biodegradabilities of recalcitrant humus and lignocellulose substances were found to be promoted by KMnO4 treatment as higher methane yields were attained from KMnO4-treated model substrates. 16S rRNA analysis illustrated that the functional microbes participated in anaerobic digestion were largely enriched in the KMnO4-pretreated digestor. Furthermore, efficient inactivation of the fecal coliform was achieved by KMnO4 pretreatment. | Enhanced Methane Production from Sludge Anaerobic Digestion with the Addition of Potassium Permanganate
This work aims to reveal the effect of potassium permanganate (KMnO4) on the sludge anaerobic digestion process, as well as the relevant mechanisms. Experimental data showed that the biomethane production was gradually increased from 159.3 ± 3.0 to 211.5 ± 5.1 mL/g VSS (volatile suspended solids) when the KMnO4 content was increased from 0 to 0.08 g/g VSS, with an increasing rate of 32.8%. A further increase in the KMnO4 dosage however resulted in the decline of the methane yield. First-order kinetic model analysis indicated that higher methane production potentials and hydrolysis rates were achieved in KMnO4-added reactors than in the control. Mechanism analysis demonstrated that KMnO4 not only efficiently disintegrated the sludge flocs, which resulted in the increased contents of dissolved organics, but also enhanced the proportion of biodegradable substances in the sludge liquor. Meanwhile, the biodegradabilities of recalcitrant humus and lignocellulose substances were found to be promoted by KMnO4 treatment as higher methane yields were attained from KMnO4-treated model substrates. 16S rRNA analysis illustrated that the functional microbes participated in anaerobic digestion were largely enriched in the KMnO4-pretreated digestor. Furthermore, efficient inactivation of the fecal coliform was achieved by KMnO4 pretreatment.
As the main byproduct of sewage treatment plants (STPs), waste activated sludge (WAS) is hugely generated in daily operations. The total production of WAS in China in 2020 was estimated to be 60 million tons (80% water content), which posed great challenges to the STP operators. Anaerobic digestion is a promising WAS treatment method as it possesses the abilities to simultaneously realize sludge minimization, stabilization, and resource recycle. In WAS anaerobic digestion, the abundantly present organic substances are degraded, accompanied by the generation of methane-contained biogas, which means the concept of “waste to energy” is well-realized. Sludge anaerobic digestion efficiency however is usually very low, which hinders its extensive applications in practical engineering. Disintegration is the first step of sludge anaerobic digestion, during which the organic substances in the solid phase are dissolved into sludge liquor. For the existing extracellular polymeric substances (EPSs) and microbial cytoderm/cytomembrane, sludge disintegration is considered to be the rate-limiting step that restricts the overall reaction rate, and the pretreatment process is needed to facilitate the slow step for ultimately increasing the sludge anaerobic digestion performance. In the last decade, chemical oxidation method has been widely adopted in sludge pretreatment due to the advantages of excellent treatment performance, high efficiency, low investment cost, and easy operation. The strong oxidants ozone (O3), hydrogen peroxide (H2O2), potassium ferrate (K2FeO4), and calcium peroxide (CaO2) are all effective for sludge pretreatment, but the high costs prevent their further applications. Potassium permanganate (KMnO4) is a commonly used oxidant in water and wastewater treatment. For instance, it was found that the organic pollutant triclosan in water was efficiently degraded by 70–80% after 2.0 mg/L KMnO4 treatment for 10 min More recently, the applications of KMnO4 in sludge treatment have raised increasing concerns. Wu et al. adopted KMnO4 to disintegrate sludge flocs and found that the SCOD (soluble chemical oxygen demand) content was promoted by 3473% after 500 mg/L KMnO4 treatment. The research of Zheng et al. also showed that the SCOD value was efficiently promoted by KMnO4 treatment at 0.05 to 0.3 g/g total suspended solids (TSS) dosage range. Li et al. adopted Na2SO3 coupled with KMnO4 for sludge treatment and found that the SCOD of solo KMnO4 and Na2SO3 + KMnO4 treated samples were, respectively, 9.5-fold and 13.2-fold to that of the control, which means that the sludge disintegration extent was largely enhanced by both methods. The promoted disintegration by KMnO4 pretreatment is undoubtedly beneficial to sludge anaerobic fermentation or digestion. Xu et al. found that the sludge volatile suspended solids (VSS) content was largely reduced, while the SCOD content was increased by KMnO4 pretreatment, causing a 7.69 times increase of the volatile fatty acids (VFAs) yield in the subsequent fermentation process. Based on the above discussion, we speculated that the methane generation from sludge anaerobic digestion could be enhanced when pretreated with KMnO4 because a better disintegration effect and more digestion substrates were expected. This issue however has not been investigated to date. The purpose of the work is therefore to find out the impact of KMnO4 on the sludge anaerobic digestion system and explore the relevant mechanisms. The biomethane productions from the sludge in the presence of KMnO4 at the concentration gradients of 0, 0.02, 0.04, 0.06, 0.08, and 0.1 g/g VSS were first investigated. Then, mechanism studies were conducted from the perspectives of EPS destruction, microbial cells leakage, biodegradability of sludge liquor, degradation of model substrates, and abundance of functional microbes. Finally, the inactivation of pathogens after anaerobic digestion by KMnO4 treatment was studied. Compared with the previous literature, this work revealed the correlation between biomethane production and KMnO4 dosages for the first time. For mechanism analysis, the previous studies always only focus on sludge disintegration as well as the generation of biodegradable organic substances. This work not only investigated the variation of disintegration efficiency by KMnO4 treatment but also assessed the impacts of KMnO4 on the biodegradability of recalcitrant organics. Furthermore, the conversion of digestion substrates to methane undergoes several biochemical processes, and how does KMnO4 affect these processes has never been studied till now. In this work, the effects of KMnO4 on these processes were evaluated using model substrates, by which the reaction kinetics of each process were revealed.
The WAS obtained from the Bailonggang STP (Shanghai, China) was first filtered using a 1.5 × 1.5 mm sieve before adopted for anaerobic digestion, whose primary characteristics are summarized in Table S1 (Supporting Information). KMnO4 (99% purity) was provided by Aladdin company in China.
This experiment was conducted in six serum bottles (V = 500 mL). Each one first received 150 mL of raw WAS. Then, these bottles were, respectively, added with 0, 0.0395, 0.079, 0.1186, 0.1581, and 0.1976 g of KMnO4 to attain the KMnO4 dosage gradients of 0, 0.02, 0.04, 0.06, 0.08, and 0.1 g/g VSS. After being mixed for 30 min using a magnetic stirrer (600 rpm), each bottle received 150 mL of inoculum, which was obtained from a pilot-scale sludge anaerobic digestor. Finally, all bottles were aerated with nitrogen for 5 min and immediately sealed for anaerobic digestion under a 35 °C environment in a shaker (150 rpm). The biogas production and methane content were, respectively, measured during anaerobic digestion, till the cumulative methane yield was not significantly enhanced after 30 days.
To reveal the impacts of KMnO4 on sludge disintegration, the VSS, SCOD, soluble protein, and carbohydrate of sludge samples pretreated with different KMnO4 dosages were detected. EPSs grab most organics in sludge, including lightly and tightly bound statuses that are named as LB-EPS and TB-EPS, respectively. The two types of EPSs in the sludge with or without the pretreatment were then extracted using a heat method, and the specific procedures are expounded in Text S1. Beside EPSs, microbial cells also contain abundant organic substrates. The damage of sludge microbes with the KMnO4 treatment was reflected by DNA and lactate dehydrogenase (LDH) releases, which are both intracellular substances and could be liberated into sludge liquor when the microbial cells were broken. Furthermore, the live/dead cells photos of sludge with or without the pretreatment were captured using a fluorescence microscope (Nikon Eclipse 80i, Japan), from which the percentages of live or dead cells can both be calculated according to the procedures demonstrated in Text S2. Besides biodegradable substrates, the sludge also contains abundant nonbiodegradable organics that are difficultly utilized by anaerobes. The excitation emission matrix (EEM) fluorescence technology was adopted to analyze the biodegradability of dissolved organics with or without the KMnO4 treatment. In brief, each EEM spectrum contains five regions (Region I to V), among which Regions I and IV, respectively, refer to tyrosine- and polysaccharide-like substances, which are easily degraded by microbes, while Regions II, III, and V, respectively, represent tryptophan-, fulvic acid-, and humic acid-like substrates, which are difficult to biodegrade. The percent fluorescence response (Pi,n) values were used to assess the relative abundance of each region, as conducted in the literature.
As shown in Table S1, there are many recalcitrant organics detected in raw WAS, including humic acid, fulvic acid, lignin, cellulose, and hemicellulose. These substances are all difficultly utilized by the functional anaerobes. Their biodegradability however could be enhanced when pretreated by KMnO4 because the structure and property could be varied in the KMnO4 oxidization process. This experiment was conducted to evaluate whether KMnO4 had some impacts on the biodegradability of recalcitrant organics using model substrates, including five groups (Groups I to V), with three serum bottles (V = 500 mL) serving as reactors in each group. Group I: First, each bottle received 150 mL of synthetic wastewater with humic acid serving as the only component, and the content of 1348 mg/L was in accordance with that measured in WAS. Then, 0, 0.079 or 0.1581 g of KMnO4 was dosed into the three reactors to obtain the concentration gradients of 0, 0.04, and 0.08 g/g VSS. After being mixed for 30 min at 600 rpm, each reactor received 150 mL of inoculum and was then sealed for anaerobic digestion based on the procedures illustrated in the batch methane producing experiment. During 5 days of anaerobic digestion, the methane productions in different reactors were daily detected. The impacts of KMnO4 on humic acid degradability were revealed through the variations of the methane yield with different KMnO4 dosages. Group II: All the operations were consistent with those performed in Group I, apart from 972 mg/L fulvic acid being substituted for humic acid as the model substrate. Group III: This group was treated in accordance with the treatment in Group I, except that humic acid was substituted by 1035 mg/L lignin alkali. Group IV: The experiment was conducted using the same procedures as those described in Group I, except that 1277 mg/L carboxymethylcellulose sodium was adopted as the solo model substrate. Group V: All operations were the same as those conducted in Group I, except that 1148 mg/L xylan replaced humic acid to be the digestion substrate.
The contents of COD, TSS, and VSS were measured according to the standard methods. Determinations of carbohydrate and protein were conducted using previously established methods. EEM spectra were measured using a fluorescence spectrometer (Hitachi F-7000, Japan). DNA release after the KMnO4 pretreatment was measured according to the literature, and LDH contents were determined using the Cytotoxicity LDH Assay Kit-WST (Dojindo Laboratories, Japan) according to the manufacturer’s instruction. Humus substrates were measured by the same method as that in the literature, while lignocellulosic substrates were detected based on the previously reported method. Methane contents in biogas were detected using a gas chromatograph (Agilent GC7890, USA). The impacts of KMnO4 on the sludge methane production potential and hydrolysis rate were revealed through the first-order kinetic model, with the operations detailed in Text S4. The impacts of KMnO4 on the hydrolysis, acidogenesis, acetogenesis, and methanogenesis processes were revealed through several batch experiments using bovine serum albumin (BSA), glucose, butyrate, hydrogen, and acetate as the model substrate, respectively (Text S5). The structures of the microbial community were detected using the Illumina Miseq sequencing technology, with 515FmodF(5′-GTGYCAGCMGCCGCGGTAA-3′) and 806RmodR(5′-GGACTACNVGGGTWTCTAAT-3′) adopted as the primers, and the raw sequencing data were divided into many operational taxonomic units (OTUs) based on 97% similarity. The fecal coliform in different reactors after anaerobic digestion was enumerated using the Colilert-18 Test kit (IDEXX, USA), with the most probable number (MPN) being adopted as the unit (Text S6).
Figure 1 depicts the cumulative methane productions from digestors pretreated by different doses of KMnO4. It was observed that there was no significant increase in methane production from these reactors after 30 d of digestion (P > 0.05), indicating that the anaerobic reactions were completed and the maximum methane yields were realized. However, the specific productions from different reactors varied a lot. As the KMnO4 content was increased from 0 to 0.08 g/g VSS, the cumulative methane generation was markedly enhanced from 159.3 ± 3.0 to 211.5 ± 5.1 mL/g VSS (P < 0.01), showing an increasing rate of 32.8%. When the KMnO4 content was further increased to 0.1 g/g VSS, the cumulative methane production reduced to 190.8 ± 4.8 mL/g VSS, significantly lower than 211.5 ± 5.1 mL/g VSS, which was obtained from the 0.08 g/g VSS KMnO4 reactor (P < 0.05). This may be attributed to the fact that the residual KMnO4 exerted some negative effects on the inoculated methanogens, which are very sensitive to the change of the environment and could be severely disturbed at a high KMnO4 dosage of 0.1 g/g VSS. Based on the above analysis, the biomethane yield was closely related to KMnO4 dosages, which exhibited an increase followed by a decrease trend with the KMnO4 content increasing from 0 to 0.1 g/g VSS, and the optimal KMnO4 dosage for methane generation was 0.08 g/g VSS. To more deeply understand how KMnO4 affects sludge methane generation, the experimental results were then simulated using the first-order kinetic model (Figure 1). According to the simulative results presented in Table 1, the model captured well the experimental data as the R2 values in all reactors surpassed 99%. Moreover, the parameters B0 and k, which, respectively, represent the methane production potential and hydrolysis rate, were both observed to be enhanced after KMnO4 pretreatment. For instance, B0 was enhanced from 176.3 ± 3.6 to 219.7 ± 3.9 mL/g VSS with the KMnO4 dosage increasing from 0 to 0.08 g/g VSSS, which then decreased to 203.8 ± 4.7 mL/g VSS when the KMnO4 content reached 0.1 g/g VSS. A similar variation trend was perceived in the k value, which was enhanced from 0.092 ± 0.004 to 0.136 ± 0.006 d–1 when the KMnO4 content was increased from 0 to 0.08 g/g VSS and then declined to 0.112 ± 0.006 d–1 when the KMnO4 dosage reached 0.1 g/g VSS. Though B0 and k values both showed decreasing tendencies when the KMnO4 content surpassed 0.08 g/g VSS, the data of 203.8 ± 4.7 mL/g VSS and 0.112 ± 0.006 d–1 were still notably higher than 176.3 ± 3.6 mL/g VSS and 0.092 ± 0.004 d–1 obtained from the control reactor. It is worth noting that the B0 and k values were both positively linearly correlated with the increased KMnO4 dosages at the range of 0 to 0.08 g/g VSS, and the R2 values were 97.88 and 96.24%, respectively (Figure S1). The simulative results demonstrated that the sludge methane production potential and hydrolysis rate were both efficiently enhanced when KMnO4 was added.
The destruction of the sludge EPS by KMnO4 treatment was first studied. According to the results presented in Figure 2a, the COD content of TB-EPS was gradually decreased from 3155 to 1345 mg/L with the KMnO4 content enhancing from 0 to 0.1 g/g VSS, while the COD of LB-EPS was simultaneously promoted from 275 to 506 mg/L, suggesting that the sludge EPS was clearly disturbed when KMnO4 was added. The total contents of TB-EPS and LB-EPS were calculated to be decreased from 3430 mg/L in the control to 3026, 2834, 2352, 2157, and 1851 mg/L by 0.02, 0.04, 0.06, 0.08, and 0.1 g/g VSS KMnO4 pretreatment, respectively, suggesting that a considerable of organics in the EPS were liberated into sludge liquor when treated by KMnO4. The results demonstrated that KMnO4 efficiently destroyed the EPS structure of the sludge, and TB-EPS was partially transformed into LB-EPS and dissolved organics. The damage of microbial cells by KMnO4 pretreatment was then explored. Figure 2b shows the release of DNA and LDH at different KMnO4 dosages. When pretreated by 0.04 g/g VSS KMnO4, the DNA and LDH releases were detected to be 124 and 173%, respectively, which then increased to 167 and 328%, respectively, with the KMnO4 dosage increased to 0.08 g/g VSS, confirming that the sludge cells were largely destructed by KMnO4 pretreatment, and higher dosages caused better effects on microbial cell destruction. To more visually display the destruction of sludge microbes, fluorescence photos of microbial cells with or without the KMnO4 pretreatment were captured, as presented in Figure 3. The green fluorescence intensity, which represents live cells, was distinctly weakened with the increase of KMnO4 dosages. Conversely, the red fluorescence intensity, which refers to dead cells, was increased when KMnO4 was added. The percentages of live cells under different conditions were then calculated based on the specific fluorescence intensities, which decreased from 91.7% in the control to 38.6% when treated by 0.04 g/g VSS KMnO4 and further reduced to 13.3% as KMnO4 content reached 0.08 g/g VSS (Figure S2). The results above indicated that both EPS and sludge cells were effectively destroyed by KMnO4, which means that efficient sludge disintegration was achieved. Table 2 shows the sludge particle size distribution under different conditions, in which the values of Dx(90), Dx(50), and Dx(10) were all found to be observably decreased after KMnO4 pretreatment. For example, the Dx(90) value decreased from 101.36 ± 0.09 μm in the control to 94.53 ± 0.08 and 85.11 ± 0.08 μm, respectively, with 0.04 or 0.08 g/g VSS KMnO4 pretreatment, which further confirmed that the sludge flocs were efficiently disintegrated in the KMnO4 treatment process. As shown in eq 1, plenty of OH– can be generated from the reaction of MnO4– and H2O when KMnO4 appeared in hydrous media. Previous studies reported that OH– possesses the ability to destroy sludge flocs; thus, the abundantly generated OH– served as a crucial reason for KMnO4 accelerating sludge disintegration. Meanwhile, KMnO4 with strong oxidizability can directly disrupt the sludge flocs structure, which was also a crucial factor for the enhanced disintegration efficiency. Figure 2c depicts the changes of SCOD concentrations when pretreated by KMnO4. With KMnO4 content increasing from 0 to 0.1 g/g VSS, the sludge SCOD was clearly increased from 84 to 2110 mg/L, indicating that more dissolved organic substances were generated from the disintegration process by KMnO4 pretreatment. The variations of soluble protein and carbohydrate concentrations by the KMnO4 treatment were also studied, which were, respectively, increased from 56.5 and 11.9 mg/L to 556.2 and 177.7 mg/L with the KMnO4 dosage increasing from 0 to 0.1 g/g VSS (Figure 2d). According to the above results, many biodegradable organics in the sludge were released into the liquid phase upon treatment with KMnO4, providing more digestion substrates for methane production. The variations of sludge liquor biodegradability by KMnO4 pretreatment were then investigated using the EEM fluorescence technology (Figure 4). As seen from the EEM spectra, the percent fluorescence responses of Regions II, III, and V (PII,n, PIII,n, and PV,n) were all distinctly reduced when pretreated by KMnO4, especially under the 0.08 g/g VSSS condition, illustrating that the proportion of non-biodegradable substances in dissolved organics was largely reduced. In contrast, the PIV,n value, which represents the easily biodegraded polysaccharide, was observably enhanced from 54.04% in the control to 62.65% and 71.31% after 0.04 and 0.08 g/g VSS KMnO4 treatment, respectively. The experimental results above demonstrated that KMnO4 efficiently promoted the biodegradability of soluble organics as the proportion of biodegradable organics was largely enhanced after KMnO4 pretreatment. As shown in Table S1, the non-biodegradable humus and lignocellulose were largely detected in the raw sludge. Several batch experiments using model substances were performed to further understand the promoted biodegradability of the sludge when KMnO4 was added. Figure 5 shows the methane generation from different model substances at different KMnO4 dosages, in which the daily methane yields of KMnO4-pretreated samples were generally higher than that from 0 g/g VSS KMnO4 reactors. For instance, the methane yield from the 0 g/g VSS KMnO4-pretreated reactor in the humic acid group was only 7.01 mL on the 4th day of digestion, which then increased to, respectively, 8.75 and 10.17 mL in 0.04 and 0.08 g/g VSS KMnO4-pretreated digestors (Figure 5a), indicating that the biodegradability of humic acid was enhanced by KMnO4. Similar observations were found in other substances (Figures 5b–e), which further confirmed the stimulative efficacy of KMnO4 on sludge organics biodegradability.
After the disintegration, the largely produced dissolved organics are adopted for methane generation through several biochemical processes, including hydrolysis, acidogenesis, acetogenesis, and methanogenesis. The effects of KMnO4 on them were reflected by comparing the degradations of model subtracts with or without the pretreatment. The results in Table 3 showed that except for acidogenesis, the other processes were all suppressed when KMnO4 was added because the degradation rates of relevant model substrates were obviously decreased. Furthermore, Table 4 demonstrates the specific degradation rate (SDR) of each model substrate, which can reflect the impacts of KMnO4 on relevant microbial activities. The SDR of BSA in the control reactor was 26.89 mg/g VSS/h, which was regarded as the original activity of the hydrolytic microbes. When treated with 0.04 g/g VSS KMnO4, the SDR of BSA decreased to 19.52 mg/g VSS/h, suggesting that the activity of hydrolytic microbes was reduced by 27.4%. As KMnO4 content was increased to 0.08 g/g VSS, the activity of hydrolytic microbes was decreased to 13.18 mg/g VSS/h, with a suppression ratio of 51.0%. The activity of acidogenic microbes was not influenced by KMnO4 because the SDR of glucose was not significantly decreased in the presence of KMnO4 at both 0.04 and 0.08 g/g VSS dosages (P > 0.05). From the same calculation with hydrolytic microbes, the microbial activities related to acetogenesis, hydrogentrophic methanogenesis, and acetoclastic methanogenesis processes were, respectively, reduced by 25.7, 33.1, and 28.3% when pretreated by 0.04 g/g VSS KMnO4, which were further decreased by, respectively, 39.8, 65.1, and 68.7% when the KMnO4 dosage reached 0.08 g/g VSS. The results demonstrated that KMnO4 exerted no significant impact on acidogenic microbe activity, while markedly suppressing the activities of all other anaerobes, and a relatively higher KMnO4 dosage led to a more serious impact on each kind of anaerobes. This explained the observation from Figure 1 that the methane production was decreased when the KMnO4 content was ultimately increased to 0.1 g/g VSSS, though higher KMnO4 dosages brought about better disintegration effects (Figure 2).
The microbial community structures with or without the addition of KMnO4 were investigated to reveal the microcosmic mechanisms for KMnO4 affecting sludge methane production. The Chao index and Shannon index, which represent the richness and diversity, respectively, were correspondingly reduced from 2754.4 ± 68.1 and 6.18 ± 0.06 in the control to 2421.7 ± 63.4 and 5.73 ± 0.05 in 0.08 g/g VSS KMnO4-pretreated reactor, indicating that the richness and diversity of the microbial community were both decreased by KMnO4 pretreatment (Table S2). In contrast, the microbial evenness was enhanced in the KMnO4-pretreated reactor because the Simpson index was increased from 0.0055 ± 0.0007 to 0.0121 ± 0.0010 when KMnO4 was added (Table S2). Venn analysis of microbial OTUs between the two reactors with or without the KMnO4 pretreatment was then conducted, as shown in Figure S3. The total OTUs were, respectively, 2149 and 1833 in the control and KMnO4-pretreated reactors, containing 1707 of common OTUs, which further proved the decreased microbial diversity by KMnO4 treatment. The data are in agreement with that observed from Figure 3 as sludge microbes were largely killed when treated with KMnO4. The microbial community structures in the two reactors at the genus level are depicted in Figure 6. Some species participating in anaerobic digestion were detected in both the reactors. However, their abundances varied greatly when pretreated by KMnO4. For example, the total abundance of three typical hydrolytic species norank_f__Caldilineaceae, norank_f__Bacteroidetes_vadinHA17, and Leptolinea was increased from 3.81% in the control to 6.14% in the KMnO4-treated digestor. The species Romboutsia, norank_f__Anaerolineaceae, Exilispira, Longilinea, norank_f__Cloacimonadaceae, and unclassifed_f__Anaerolineaceae are all the functional microbes participating in the acidogenesis or acetogenesis process. Among them, Romboutsia was reported to be a VFA producer by using glucose as the substrate, and the abundance was increased from 0.96% to 1.43% by KMnO4 treatment. The species norank_f__Anaerolineaceae and Longilinea can degrade various carbohydrates for VFA production and were, respectively, enriched from the abundances of 1.28% and 0.87 to 1.64% and 1.55% by the KMnO4 pretreatment. The abundance of Exilispira, a species that was reported to be connected with the medium chain fatty acid (MCFA) generation in anaerobic digestion and to possess the ability to degrade recalcitrant dicamba, was increased from 0.52 to 1.08% when pretreated by KMnO4. The abundance of norank_f__Cloacimonadaceae, an acetogenic bacterium that can convert propionate into acetic acid, was increased from 1.16 to 1.78% after the KMnO4 pretreatment. The species unclassifed_f__Anaerolineaceae can utilize hydrocarbons to produce VFAs, and the abundance was increased from 0.82 to 1.29% by the KMnO4 pretreatment. Further calculation indicated that the total abundances of acidogenic and acetogenic microbes were, respectively, 5.61 and 8.77% in the control and KMnO4-treated reactors. Beyond this, two methanogenic genera Methanosaeta and Methanolinea, which are acetoclastic and hydrogenotrophic methanogens, respectively, were found in the two reactors. As presented in Figure 6, the abundances of Methanosaeta and Methanolinea were, respectively, 0.94 and 0.55% in the control, which were enhanced to, respectively, 1.53 and 1.69% in the KMnO4-treated reactor, suggesting that the two kinds of methanogens were both enriched by the KMnO4 treatment. Beyond this, the abundance of Methanosaeta was higher than that of Methanolinea in the control, which was reversed when treated by KMnO4 (Methanolinea surpassed Methanosaeta), indicating that the major methanogenesis pathway was changed from acetoclastic to hydrogenotrophic after the treatment. In terms of the above analysis, different kinds of functional microbes were all enriched in the KMnO4-pretreated reactor, which was a crucial reason for the promotion of methane generation by the KMnO4 pretreatment (Figure 1).
WAS typically contains many pathogens, such as the fecal coliform, and could result in high health risks to humans without proper disposal. KMnO4 was reported to possess the ability to kill various pathogens; thus, the elimination of the fecal coliform by KMnO4 pretreatment was studied in this work, with the results depicted in Figure S4. After anaerobic digestion, the number of fecal coliform bacteria was 6372 MPN/g TSS in the control reactor, which observably reduced to 1138 MPN/g TSS in the presence of KMnO4 at 0.04 g/g VSS. When the KMnO4 content was increased to 0.08 g/g VSS, the fecal coliform was continuously killed to 272 MPN/g TSS, lower than the Class A level (1000 MPN/g TSS) of the EPA protocol. This result demonstrated that the environmental risks of the digested sludge were largely reduced by the KMnO4 treatment.
This study found that the anaerobic digestion performance of the sludge was effectively enhanced by the KMnO4 pretreatment. The methane generation first increased from 159.3 ± 3.0 to 211.5 ± 5.1 mL/g VSS with the KMnO4 content increas from 0 to 0.08 g/g VSS and then declined to 190.8 ± 4.8 mL/g VSS as the KMnO4 content was further increased to 0.1 g/g VSS. The increasing rate of the methane yield was 32.8% by 0.08 g/g VSS KMnO4 treatment, which was observably higher than the values of 22, 25.1, and 10.3% by H2O2, K2FeO4, and Ca(ClO)2 treatments, respectively, but lower than that by CaO2 treatment (Table 5). The disintegration efficiency of the sludge was sufficiently facilitated when KMnO4 was added, showing a positive correlation with the increased KMnO4 content in the 0 to 0.1 g/g VSS range. Moreover, the biodegradability of sludge organics was remarkably enhanced when pretreated by KMnO4. On one hand, the proportion of easily biodegraded substances in the liquid phase was increased when KMnO4 was added. On the other hand, the methane production potentials of recalcitrant humus and lignocellulose substances were promoted when pretreated by KMnO4. The biochemical reaction kinetics analysis revealed that except for acidogenic microbes, the activities of hydrolytic, acetogenic, and methanogenic microbes were all suppressed by KMnO4, and a higher dosage caused more severe inhibiting effects. The suppression effect of 0.1 g/g VSS KMnO4 on these biochemical processes neutralized the benefits from the enhanced sludge disintegration and organics biodegradability, which then resulted in the decrease of methane production. Microbial community analysis illustrated that all the hydrolytic, acidogenic, acetogenic, and methanogenic microbes were enriched in the KMnO4-pretreated reactor. The results herein reported effectively filled the knowledge gaps of KMnO4 affecting the sludge anaerobic digestion system. However, it should be emphasized that the methane production results in this work were obtained only using batch experiments, and semi-continuous or continuous experiments are needed to further verify the optimum KMnO4 dosage and amend other parameters before scaling up this KMnO4-based technology. In addition, to avoid the adverse impact of KMnO4 on methanogens for further increasing the biomethane generation, the two-phase sludge anaerobic digestion process can be taken into consideration. The literature showed that the inhibition of rhamnolipid on methanogens in the methanogenic phase was relieved compared with that in the acidogenic phase, which means that the suppressive effect of KMnO4 on functional microbes could be alleviated with the two-phase sludge anaerobic digestion process applied.
The impact of KMnO4 on sludge anaerobic digestion, especially for biomethane generation, was studied in this work. The methane yield first increased and then decreased when the KMnO4 content was increased from 0 to 0.1 g/g VSS, with the maximum value of 211.5 ± 5.1 mLg VSS attained at the KMnO4 content of 0.08 g/g VSS, 1.328 times that of the control. Mechanism investigation revealed that both the disintegration extent and biodegradability of the sludge were promoted by KMnO4. Microbial analysis demonstrated that the structure of the microbial community in the KMnO4-treated digestor was more beneficial to methane generation than the control as the functional microbes (e.g., Methanosaeta and Methanolinea) were observably enriched by the KMnO4 treatment. After anaerobic digestion, the number of fecal coliform bacteria was largely reduced in the KMnO4-pretreated reactor. |
||
PMC9648144 | David Wu,Aunoy Poddar,Elpiniki Ninou,Elizabeth Hwang,Mitchel A. Cole,S. John Liu,Max A. Horlbeck,Jin Chen,Joseph M. Replogle,Giovanni A. Carosso,Nicolas W.L. Eng,Jonghoon Chang,Yin Shen,Jonathan S. Weissman,Daniel A. Lim | Dual genome-wide coding and lncRNA screens in neural induction of induced pluripotent stem cells | 14-09-2022 | CRISPR,CRISPRi,human pluripotent stem cells,iPSCs,neurodevelopment,neural induction,long noncoding RNA,genome-wide pooled screens,Perturb-seq,single-cell RNA-seq | Summary Human chromosomes are pervasively transcribed, but systematic understanding of coding and long noncoding RNA (lncRNA) genome function in cell differentiation is lacking. Using CRISPR interference (CRISPRi) in human induced pluripotent stem cells, we performed dual genome-wide screens—assessing 18,905 protein-coding and 10,678 lncRNA loci—and identified 419 coding and 201 lncRNA genes that regulate neural induction. Integrative analyses revealed distinct properties of coding and lncRNA genome function, including a 10-fold enrichment of lncRNA genes for roles in differentiation compared with proliferation. Further, we applied CRISPRi perturbation coupled with single-cell RNA-seq (Perturb-seq) to obtain granular insights into neural induction phenotypes. While most coding hits stalled or aborted differentiation, lncRNA hits were enriched for the genesis of diverse cellular states, including those outside the neural lineage. In addition to providing a rich resource for understanding coding and lncRNA gene function in development, these results indicate that the lncRNA genome regulates lineage commitment in a manner fundamentally distinct from coding genes. | Dual genome-wide coding and lncRNA screens in neural induction of induced pluripotent stem cells
Human chromosomes are pervasively transcribed, but systematic understanding of coding and long noncoding RNA (lncRNA) genome function in cell differentiation is lacking. Using CRISPR interference (CRISPRi) in human induced pluripotent stem cells, we performed dual genome-wide screens—assessing 18,905 protein-coding and 10,678 lncRNA loci—and identified 419 coding and 201 lncRNA genes that regulate neural induction. Integrative analyses revealed distinct properties of coding and lncRNA genome function, including a 10-fold enrichment of lncRNA genes for roles in differentiation compared with proliferation. Further, we applied CRISPRi perturbation coupled with single-cell RNA-seq (Perturb-seq) to obtain granular insights into neural induction phenotypes. While most coding hits stalled or aborted differentiation, lncRNA hits were enriched for the genesis of diverse cellular states, including those outside the neural lineage. In addition to providing a rich resource for understanding coding and lncRNA gene function in development, these results indicate that the lncRNA genome regulates lineage commitment in a manner fundamentally distinct from coding genes.
The human genome expresses thousands of genes—both coding and noncoding,—and many are critical to the complex processes of cell differentiation during development.,, Early in mammalian development, neural stem cells (NSCs) are produced from pluripotent stem cells by the process of neural induction. Long noncoding RNAs (lncRNAs) are transcripts longer than 200 nucleotides that do not encode protein, and many are expressed in neural tissues.,, The recent evolutionary expansion of these loci has led to the hypothesis that lncRNA genes play critical roles in the development of complex organisms.,, However, unlike coding genes, far fewer lncRNA genes have been demonstrated to regulate cell biology. More broadly, systematic understanding of how the coding and lncRNA genomes regulate developmental processes is lacking. Genetic screens are powerful methods for identifying genes underlying phenotypes of interest. The vast majority of CRISPR-based screens have focused on the protein-coding genome, typically excluding lncRNA loci. Nevertheless, these studies provide insight into principles of coding genome function by integrating screen data into a rich foundation of literature, including knowledge of physical and functional interaction networks. Although genetic screens of lncRNAs are now emerging,,, functional knowledge for this class of molecules is still primarily drawn from the study of individual lncRNAs. Genome-wide screens that integrate information from both the coding and lncRNA genomes are rare and have not been performed in complex contexts such as cell differentiation. Such dual genome-wide approaches can provide unique data resources to discover principles of developmental regulation. In this work, we used functional genomics to systematically assess 18,905 coding genes and 10,678 lncRNAs for roles in human neural induction. Using dual genome-wide CRISPR interference (CRISPRi) marker-based screens, we identified 419 protein-coding and 201 lncRNA genes that regulated the production of NSCs from induced pluripotent stem cells (iPSCs). The scale and design of this resource enabled integrated analyses and the discovery of general properties of coding and lncRNA genome function. To obtain deeper insights into the biology of these regulators, we applied this resource to perform a CRISPRi perturbation coupled with single-cell RNA sequencing (RNA-seq), known as Perturb-seq.,,,, Collectively, these systematic studies revealed fundamental insights about the unique developmental roles of the coding and lncRNA genomes at a level that is challenging to ascertain by the study of individual genes.
An early step toward brain development is neural induction from pluripotent stem cells. Using dual SMAD inhibition (dSMADi),, we induced NSCs from iPSCs that express dCas9-KRAB (CRISPRi-iPSCs) under doxycycline-inducible control (Figure 1A). The induction of NSCs was progressive over time, which we characterized by flow cytometry analysis of the canonical marker PAX6 (Figure 1B) and RNA-seq of polyadenylated and total RNA at multiple time-points (0–11 days). Many thousands of coding and noncoding genes were dynamically expressed over the course of neural induction (Figure 1C; Table S1). We applied the transcriptomic data to inform the assembly of a dual genome-wide library (STAR Methods) containing published, validated CRISPRi single-guide RNAs (sgRNAs) targeting human coding (hCRISPRi-v2) and lncRNA (CRiNCL) genes. These sgRNAs were selected based on RNA-seq expression during neural induction and were designed in prior studies using an algorithm that incorporates nucleosome-positioning and FANTOM cap analysis of gene expression data, with off-target activity filtering., We included a total of 212,938 sgRNAs (with 4,523 non-targeting controls) against 29,583 targets, covering 18,905 coding (five sgRNAs/target) and 10,678 lncRNA genes (10 sgRNAs/target). We conducted the dual genome-wide screens using CRISPRi-iPSCs with PAX6 staining as the readout for neural induction (Figure 1D). We selected day 8 of neural induction as the endpoint, when both PAX6+ and PAX6− populations were present (Figure 1B), enabling the identification of hits that either increase or decrease this marker of neural induction. After sequencing the assembled library to ensure uniform distribution, we packaged lentivirus and transduced ∼650 million CRISPRi-iPSCs in two biological replicates. Cells were propagated in self-renewal media under puromycin selection until reaching >80% sgRNA positivity, detected by co-expressed blue fluorescent protein (BFP). We maintained the screens at >1,000× sgRNA coverage per replicate over the course of neural induction. Time zero (T0) aliquots were collected to assess initial sgRNA abundance. After 8 days of neural induction and dCas9-KRAB expression, cells were harvested and quantified. Approximately 2.7 billion cells in total were fixed, permeabilized, and stained with antibodies against PAX6 for fluorescence-activated cell sorting (FACS) into PAX6+ and PAX6− fractions (top and bottom thirds; Figure 1D, right). The abundance of sgRNAs in each fraction were quantified by PCR amplification followed by Illumina sequencing. The differentiation phenotype rho (ρ; log2 enrichment ratio of normalized sgRNA abundance in PAX6+ versus PAX6− fraction) was calculated for all targets and non-targeting controls (Figure 1D, right). This ρ value, used in other marker-based studies, represents the log2 fold change of each sgRNA in the positive fraction relative to the negative fraction.,, Negative values of ρ indicate that the sgRNA decreased neural induction (e.g., knockdown of pro-neural factors), while positive ρ values indicate that the sgRNA promoted the development of PAX6+ cells (e.g., knockdown of pluripotency factors). Independent replicates were correlated (Figure S1A) and non-targeting control sgRNAs produced ρ values centered around zero, as expected (Figure S1B). More than 99% of sgRNAs met a threshold of >100× coverage (with 97% with >500×) providing sufficient data for all 29,583 targets, with 94% of targets having all designed sgRNAs represented. After applying an empirical false discovery rate (FDR) of 0.05, exclusion of sgRNAs targeting multiple loci and gene “neighbor hits” (STAR Methods), we identified 419 protein-coding and 201 lncRNA genes that altered the production of PAX6+ NSCs (Figure 1E). Since each hit was targeted by multiple sgRNAs, we assessed whether these sgRNAs were in agreement by calculating the fraction of sgRNAs in the same direction as the hit. Hits showed a very high median concordance of 1 (indicating that all sgRNAs had the same effect) while those targeting non-hits had a median concordance of 0.5 (indicating random chance) (Figure S1C). Additionally, given the large scale of the screen, we estimated hit identification performance at smaller scales by downsampling the raw data for precision-recall analysis. At 10% downsampling (∼100× coverage), performance was poor (<40% of hits identified). This improved substantially at 200× and 500× coverage, where >70% and >80% of hits were identified, respectively (Figure S1D). Thus, the comprehensive scale of the dual genome-wide screens provides an unparalleled glimpse into this early differentiation process.
Of the 18,905 coding genes screened, PAX6 itself was expectedly the highest scoring negative hit (Figure 1E) with ρ = −3.01, representing an 88% reduction in PAX6+ cells by FACS. We also observed numerous examples of hits with expected positive or negative impact on neural induction. For instance, pro-pluripotency factors (POU5F1/OCT4, GBX2, SMARCC1, PRDM14) were positive hits while genes with known neurodevelopmental roles (SOX2, SOX4, SOX11, HES1, OTX2) were negative hits. Protein-protein network analysis revealed enrichment for known functional interactions among coding gene hits, such as those of the BRG1/BRM-associated factor (BAF) chromatin remodeling complex, the Polycomb repressive complex (PRC), and signaling pathways critical to neurodevelopment such as NOTCH (Figures S2A and S2B). Furthermore, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of coding hits revealed enrichment for processes important in early development (Figure S2C). Thus, our screen recovers a large number of genes known to function in complexes and pathways important for neural induction. To experimentally validate screen results, we selected 16 hits and targeted each with two independent sgRNAs (32 different sgRNAs in total covering each of the hit subcategories, i.e., coding/lncRNA, positive/negative, with two biological replicates per sgRNA). These sgRNAs were individually transduced into iPSCs, and, after 8 days of neural induction and CRISPRi, the cells were analyzed by PAX6 staining via flow cytometry (Figure 2A). Individual sgRNAs targeting both coding and lncRNA genes showed phenotypes matching their screen phenotypes (Figure 2B). Collectively, quantitative results from individual experiments were highly correlated with the screen ρ phenotype (Pearson r = 0.91, Figure 2C), providing experimental validation to the hits identified in the screen.
Similar numbers of coding gene hits exerted positive (52%) and negative (48%) effects on neural induction, and this slight bias was not significant (permutation test p = 0.27; Figure 3A, left). In contrast, the majority (87%) of lncRNA hits identified were negative hits, and this enrichment in the lncRNA hit distribution was highly significant (permutation test p < 1 × 10−6; Figure 3A, right). These results indicate that lncRNA hits were enriched for functions that normally promote neural induction. During development, cell division can have important effects on differentiation. To investigate the effects of proliferation during neural induction, we compared the total sgRNA abundance in the PAX6+ and PAX6− fractions at day 8 (final abundance) with the initial sgRNA abundance in samples collected at the beginning of the screen (Figure S3A). This enabled calculation of the growth enrichment index gamma (γ; negative values indicate a decrease in proliferation; positive values indicate an increase), which we directly validated in a separate screen without FACS (STAR Methods; Figure S3B). We identified 730 coding gene hits and 24 lncRNA gene hits that altered cell proliferation during neural induction (Table S2). As expected, coding gene hits included numerous cell cycle, apoptosis, and other essential genes (e.g., CDC20, CDT1, TP53, MDM2, TOP2A, BAX) (Figure S3C). Proliferation hits were strongly enriched for GO terms relating to essential biological processes, including ribosome biogenesis and DNA helicase activity (Figure S3D). As a group, coding hits were biased toward negative proliferation hits (Figure 3B), consistent with previous studies of essential genes.,, Integrated analyses of both differentiation and proliferation effects (Figure 3C) revealed that the majority (91%) of hits produced a single phenotype (i.e., differentiation or proliferation, but not both). Of the 1,258 hits across both coding and lncRNA genomes, only 9% of hits had dual phenotypes (Figure 3D). For example, knockdown of the dual hit POU5F1/OCT4 increased differentiation (positive p; Figure 1E) and decreased proliferation (negative γ; Figure S3C), consistent with its role in maintaining both pluripotency and self-renewing stem cell divisions., Notably, the coding and lncRNA genomes differed vastly in their propensity for differentiation and proliferation phenotypes. Among coding genes, there were only half as many differentiation hits compared with proliferation hits (permutation test, p < 1 × 10−6). In stark contrast, among lncRNA genes, differentiation hits outnumbered proliferation hits by over 9-fold (permutation test, p < 1 × 10−6; Figure 3E). These differences in differentiation versus proliferation ratios highlight the unique roles of these two aspects of the genome in regulating cell biology. Overall, these integrated analyses of the dual genome-wide screen results indicate that the lncRNA genome is far more specialized for roles in promoting neural induction compared with the coding genome.
We next leveraged the screen data to identify transcriptomic and epigenomic properties that distinguish hits from non-hits. A priori, we hypothesized that differential expression would be predictive of hits. For example, negative hits may have expression patterns similar to PAX6 (high in NSCs, low in stem cells), whereas positive hits may have the expression pattern of POU5F1/OCT4 (high in stem cells, low in NSCs). However, examination of individual genes revealed both negative (e.g., PAF1) and positive (e.g., SMARCE1) hits with stable expression throughout neural induction (Figure S4A). To systematically assess the relationships of transcriptomic and epigenomic data to screen phenotypes, we turned to a machine learning approach. To provide transcriptomic features for this analysis, we used our neural induction RNA-seq time-series data. For each target in the screen, we determined the gene expression (transcripts per million [TPM]), fold change at each time-point relative to day 0, maximum expression, maximum fold change, and scaled expression (Z score representing relative change over time). For epigenomic features, we used data from the Roadmap Epigenomics project that profiled 27 histone marks in human embryonic stem cells (ESCs) undergoing dual SMAD inhibition neural induction similar to that performed in our screen. Specifically, an individual epigenomic feature would be the level of a histone mark in the stem cell and NSC stages. For all coding and lncRNA gene promoters, we quantified the levels of the histone marks at these stages. To compare the overall ability of transcriptomic and epigenomic data to discriminate hits from non-hits, we constructed machine learning classifiers and analyzed the area under the curve (AUC) of the receiver operating characteristic (ROC) (Figure S4B). While transcriptomic data were able to classify coding hits (mean AUC, 0.74), these data performed poorly for lncRNA hits (mean AUC, 0.55) overall and in a bootstrapped analysis of individual features (Figures 4A and 4B). The median expression level of coding hits (50.4 TPM) was more than 4-fold higher than that of non-hits (11.7 TPM), whereas the expression level difference of lncRNA hits (0.7 TPM) and non-hits (0.4 TPM) was smaller and less significant (Figure S4C, left). Differential expression differences (maximum absolute fold change) were also more associated with coding hits than lncRNA hits (Figure S4C, right). Furthermore, in an analysis of temporal expression dynamics, coding hits were associated with certain expression patterns, but lncRNA hits were not enriched for any pattern (Figures S4D and S4E). Thus, common transcriptional heuristics used to predict the biological activity of coding genes—such as expression level or temporal pattern—do not apply to lncRNA genes. The epigenomic data classified both coding and lncRNA gene hits at a similar performance, with mean AUC of 0.75 and 0.74, respectively (Figure 4C). To explore this finding at a more granular level, we turned to the analysis of individual histone marks (Figure 4D), which revealed significant scores for histone-3 lysine-4 trimethylation (H3K4me3), indicating that this mark distinguished hits from non-hits better than random chance for both coding and lncRNA genes. The H3K4me3 modification is associated with active genes, with the top 5% “broadest” domains enriched for genes important for cellular identity and function. Analysis of chromatin immunoprecipitation sequencing (ChIP-seq) profiles revealed elevated H3K4me3 deposition at both coding and lncRNA hit promoter regions (Figure 4E). Additionally, both coding and lncRNA hits were significantly enriched (odds ratio, ∼4–8) in the broadest H3K4me3 domains (Figure 4F), indicating importance in cell identity. Together, these findings illustrate how epigenomic features—as a group as well as at the level of a specific histone mark—distinguish hits from non-hits in a screen for regulators of neural induction.
Some lncRNA loci can function as transcriptional enhancers.,, We therefore investigated what proportion of lncRNA hits have evidence of enhancer-like function. The linear genomic distance of lncRNA gene hits to coding gene hits was somewhat decreased (median, 1.4 Mb) compared with the overall distribution (2 Mb), although these distributions were largely overlapping (Figure S5A). To more comprehensively identify potential enhancer loci among hits, we considered the Functional ANnoTation Of the Mammalian genome (FANTOM5) atlas of 43,011 human enhancers, a massively parallel reporter assay (MPRA) that identified 1,547 candidate regulatory sequences activated during human neural induction, the genomic relationship of each lncRNA gene hit with the nearest coding gene hit, and long-range three-dimensional intrachromosomal interactions between lncRNA and coding genes derived from Proximity ligation-assisted ChIP-seq (PLAC-seq) (Figures 5A–5F). In total, 18% (36 of 201) of lncRNA hits overlapped at least one of these maps (Figure 5G, left). Of note, these broadly inclusive criteria also classified 13% (54 of 419) of coding hits as potential enhancers (Figure 5G, right). At higher stringency—evidence from at least two of the analyses—only 2% (4 of 201) of lncRNA hits were classified as enhancers (Tables S3 and S4). Thus, only a minority of coding and lncRNA gene hits are potential enhancers.
By coupling CRISPRi genetic perturbation with rich single-cell transcriptomic readout, Perturb-seq,, provides deeper insights into gene function and cell biology. While the readout of pooled screens is usually based on simple phenotypes such as cell growth, survival, or marker gene expression, Perturb-seq allows the dissection of different phenotypes and molecular mechanisms that are masked in bulk experiments. We used our functional atlas from the dual genome-wide screens to inform a Perturb-seq experiment that interrogates both coding and lncRNA gene function. We selected targets by prioritizing the highest scoring differentiation hits and excluding any hits with strong proliferative phenotypes; i.e., those with absolute γ greater than 1 (predicted to become substantially overrepresented or underrepresented due to survival differences). For comparative analysis, we also randomly sampled non-hit genes with similar expression levels. The final Perturb-seq library consisted of 480 sgRNAs for 240 unique targets (120 lncRNA and 120 coding genes, with two independent sgRNAs for each target), covering 60 positive differentiation hits, 85 negative differentiation hits, 30 dual hits, and 65 non-hits; additionally, 12 non-targeting control sgRNAs were included for a total of 492 unique sgRNAs. The library was transduced into CRISPRi-iPSCs at a low multiplicity of infection (MOI) of 0.1, corresponding to >95% cells with a single sgRNA integration. After FACS for sgRNA+ cells, we initiated neural induction and activation of CRISPRi (Figure 6A). On day 8, we harvested cells and prepared single-cell RNA-seq (scRNA-seq) libraries using direct sgRNA capture. Following sequencing and data processing, we filtered cells for sgRNA detection, singlet status, and quality metrics (STAR Methods; Table S5). We obtained a total of 78,393 cells that harbored single sgRNA perturbations, with each perturbation represented in a median of 317 cells. Analysis of target gene expression data revealed a median knockdown efficiency of 80% (Figure S6A), comparable with prior studies. The Perturb-seq dataset was visualized in two dimensions using uniform manifold approximation and projection (UMAP). Based on RNA velocity analysis, (Figure 6B) and marker gene expression (Figures 6C and S6B), we identified three major cellular trajectories. The largest trajectory (NSC lineage, representing ∼50% of the cells) corresponded to non-cycling cells undergoing neural induction, with velocities directed toward a final cell state with high expression of neural markers including PAX6, FOXG1, and EMX2. Pluripotency markers such as GBX2 and POU5F1/OCT4 were lowly expressed in this trajectory but present in other cell populations (Figures 6C and S6B). The second largest trajectory (cell cycle, ∼30% of cells) consisted of actively cycling cells (CDC6+ and MKI67+), including both PAX6+ cells as well as PAX6− cells that expressed pluripotency markers. Cells exiting the cell cycle trajectory branched into either the NSC lineage or a third trajectory—non-central nervous system (non-CNS) ectoderm, ∼14% of cells—characterized by markers of ectodermal lineages (e.g., TFAP2A/B) that normally develop outside the CNS and can appear in a subset of cells undergoing neural induction. Approximately 7% of cells did not fall within these three major trajectories. Each Perturb-seq sgRNA was mapped to each cell, and we constructed normalized 2D density heatmaps to visualize the enrichment of hits and non-hits in the UMAP space. As a group, positive-hit sgRNAs were enriched in PAX6+ NSCs, whereas negative-hit sgRNAs were enriched in multiple PAX6− cell states (Figure 6D). The group of non-hit sgRNAs were not statistically distinguishable from non-targeting control sgRNAs (representing non-perturbed cells), indicating that they did not have substantial effects on the neural induction transcriptome. Thus, Perturb-seq validates the differentiation phenotypes of targets from the genome-wide screens. To assess potential sgRNA effects on cell proliferation, we quantified the number of cells expressing each sgRNA, providing a relative measure of this growth phenotype (e.g., a target that reduces proliferation would drop out over time, resulting in fewer sgRNA+ cells). For targets in the Perturb-seq experiment, the sgRNA cell counts at day 8 of neural induction were proportional to the γ proliferation phenotype from the genome-wide screens (Figure S6C), with dual hits showing a strong correlation (Pearson r = 0.92). Thus, Perturb-seq confirms both proliferation and differentiation phenotypes, supporting the findings of the genome-wide screens (Figures 3C and 3D). To study the effects of individual hits, we generated normalized density heatmaps for each target, using density-based spatial clustering and application with noise (DBSCAN) to identify the discrete UMAP regions of high sgRNA density (STAR Methods). Pairwise analysis and clustering of the sgRNA density profiles revealed groups of targets that had similar effects (Figures 6E, S6D, and S6E; Table S6). For instance, BAF1 complex members were positive hits in the genome-wide screen (Figures 1E and S2B), and sgRNAs targeting ARID1A (BAF250A), SMARCA4 (BRG1), SMARCC1 (BAF155), and SMARCE1 (BAF57) were localized in the same patterns in UMAP (Figure 6F, top), suggesting they affected neural induction in a similar manner. Knockdown of the BAF complex led to cells farther along both NSC and non-CNS ectoderm trajectories, consistent with the role of this chromatin regulator complex in maintaining pluripotency and acting as a general barrier to differentiation. Proteins encoded by negative hits PAF1, CTR9, RTF1, and CDC73 physically interact in a complex known as PAF1c that regulates transcription, chromatin structure, and signaling pathways important for embryogenesis. Targeting these PAF1c components produced a transcriptome that is distinct from the major cell trajectories observed in neural induction (Figure 6F, bottom). Similarly, Perturb-seq revealed overlapping phenotypes among physically interacting hits related to the Mediator, DNA synthesis, and Polycomb complexes (Figure S6D). Additionally, genes that function in the same pathway produced similar UMAP density profiles. POU5F1/OCT4 is upregulated by SALL4, and the density heatmaps of these two positive hits were highly overlapping (Figure S6E, left), indicating that repression of POU5F1 and SALL4 led to similar phenotypes. Analysis of the density heatmaps also identified similar patterns among other coding genes that function in the same pathways, such as Wingless (WNT) and mitogen-activated protein kinase (MAPK) signaling (Figure S6E). Collectively, these examples demonstrate that Perturb-seq targeting of genes in the same pathway or molecular complex produces highly similar UMAP profiles that reflect the underlying biological process governed by those genes.
Based on the analysis of density profiles for all Perturb-seq targets, we identified a total of 29 cell states (Figures 7A and S7A–S7C). Each Perturb-seq target was then analyzed for the relative distribution of its sgRNAs mapping to each of the 29 states, and these data were visualized by heatmap (Figure S7D). Hits were color coded according to their positive or negative differentiation phenotype from the dual genome-wide screens, and, although this information was not used to inform clustering, positive and negative hits segregated from each other. For instance, positive hits associated with NSC states (e.g., 16, 12, 9, 23, 24), while negative hits were prominent in less differentiated, intermediate cell states (e.g., 13, 6). For both coding and lncRNA genes, positive hits generally produced similar PAX6+ NSC states. For instance, sgRNAs targeting OGT—which encodes the O-GlcNAc transferase protein that regulates pluripotency and neural differentiation,—were enriched in NSC state 16 (Figure 7B). This state was characterized by the highest expression of neural markers, including genes involved in forebrain development (e.g., PAX6, FOXG1, FEZF1, EMX2) (Figure 7C; Table S6). Targeting the novel lncRNA gene LH09400 (internal identifier) led to enrichment in NSC state 12 (Figure 7B), which expressed a highly similar signature of neural markers but with elevated levels of HES4, HES5, and ID4, genes downstream of NOTCH signaling (Figure 7C). Negative hits showed highly divergent phenotypes between coding genes and lncRNA genes. For coding genes, the most common phenotype (40%) was enrichment in an intermediate cell state. For example, knockdown of the forebrain development factor HESX1, a homeobox, led to enrichment in cell state 6. This intermediate state lies at the junction of the major RNA velocity trajectories (Figure 6B) and is characterized by PAX6−/GBX2+ cells exiting the cell cycle (Figure 6C), suggesting that these cells are most similar to undifferentiated cells, and may have stalled or are slower to progress along their differentiation trajectory to NSCs. The next most common phenotype (16%) was an apoptotic signature (e.g., BAX, CDKN1A), indicating that these cells failed differentiation, most likely due to impaired survival. Together, stalled and apoptotic phenotypes (collectively “non-productive”) represented the majority (56%) of negative coding hits. In contrast, few lncRNA gene perturbations exhibited non-productive states as the main phenotype. Repression of lncRNA genes instead generally led to diverse cell states along multiple trajectories (Figure S7E). For instance, sgRNAs targeting the uncharacterized lncRNA gene SERTAD4-AS1 were enriched in all three trajectories (Figure 7B), even though this hit inhibited neural induction to a similar degree as HESX1. In addition to affecting cells in the NSC lineage (state 12), perturbation of SERTAD4-AS1 also resulted in changes to cell cycle (state 18), and cells at the far end of the non-CNS ectoderm trajectory that appear neural crest derived (state 1; Figure 7C). Thus, despite inhibiting neural induction to a similar degree, the underlying phenotype of SERTAD4-AS1 was vastly different from that of HESX1. Quantitative classification of negative hit phenotypes revealed profound differences between coding and lncRNA genes (Figure 7D). Although coding gene knockdown typically prevented neural induction by generating non-productive (i.e., stalled or apoptotic) phenotypes, lncRNA gene knockdown generally blocked neural induction by dispersing cells along multiple trajectories, including cell identities outside the NSC lineage. Furthermore, the number of trajectories was not explained by neural induction effect size (Figure S7F). These granular Perturb-seq phenotypes therefore support our broader findings of differences in coding and lncRNA differentiation and proliferation phenotypes (Figures 3D and 3E). Collectively, our findings indicate that coding and noncoding genes required for neural induction have markedly different phenotypes, suggesting that lncRNA genes—which have arisen much later in evolution than coding genes—may be employed by the genome for broadly different cellular roles, providing an additional facet of complex gene regulation during development.
To facilitate the widespread use of our resource, we created an interactive data portal (danlimlab.shinyapps.io/dualgenomewide). This website enables intuitive exploration of our collective datasets without any programming experience, from retrieving the differentiation and proliferation effects for genes of interest to visualizing the single-cell gene expression and sgRNA density profiles from the Perturb-seq experiment (Figure S8A). For instance, Perturb-seq revealed that targeting the SERTAD4-AS1 gene inhibited neural induction by causing a multiple trajectory phenotype, producing cells with transcriptomes far outside of the neural stem cell lineage, such as non-CNS neural crest. The SERTAD4-AS1 gene is located on chromosome 1 and produces multiple multi-exon isoforms, in antisense orientation to a transcript isoform of the coding gene SERTAD4. From the transcriptomics data, SERTAD4-AS1 expression is highest in iPSCs and induced cells (TPM > 1), with decreased expression in the transition between these two states (Figure S8B). Importantly, the dual genome-wide nature of the study enabled us to determine that, while the SERTAD4-AS1 lncRNA gene was a differentiation hit, SERTAD4 coding gene was a non-hit (Figure S8C). In our enhancer analysis, SERTAD4-AS1 did not map to any MPRA or FANTOM5 enhancers, and chromatin interaction analysis did not find any statistically enriched long-range 3D interactions with other hits. Using the data portal to explore the Perturb-seq analysis for SERTAD4-AS1 revealed the transgelin (TAGLN) gene—which encodes an actin-binding protein and early marker of smooth muscle cell differentiation,—to be the top marker associated with SERTAD4-AS1 perturbation (Figure S8D). To further explore the underlying biological phenotype of SERTAD4-AS1 gene repression, we designed an internally controlled differentiation assay. We sparsely labeled CRISPRi-iPSCs with viral sgRNA vectors targeting SERTAD4-AS1 (co-expressing GFP) or a non-targeting control sequence (co-expressing RFP) (Figure 7E). After 8 days of neural induction, we performed immunofluorescent staining for TAGLN protein. Significantly more GFP+ cells (sgSERTAD4-AS1) were positive for TAGLN protein compared with RFP+ cells (sgControl), indicating that loss of SERTAD4-AS1 during neural induction leads to the abnormal generation of cells with this early marker of smooth muscle (Figures 7F and S8E). Notably, analysis of cells containing sgSERTAD4-AS1 in the Perturb-seq experiment revealed excellent on-target knockdown of SERTAD4-AS1 without affecting SERTAD4 coding gene expression (Figure 7G). These experimental findings, as predicted by our genome-wide screens and Perturb-seq analyses, further strengthen the neural induction phenotype of this lncRNA gene hit.
In addition to identifying hundreds of coding and lncRNA genes that regulate neural induction, the scale of the dual genome-wide screens provided fundamental insights that would not have been apparent with less comprehensive approaches. Perturb-seq additionally revealed surprising differences in the phenotypes of coding and lncRNA genes when examined at high resolution. Taken together, our systematic studies underscore the unique functional roles of the lncRNA and coding genomes and have important implications for our understanding of gene expression studies, genome evolution, and developmental phenotypes. Gene expression is often used to predict biological function in development., In our systematic analyses, the inference of function by transcriptional information was relatively strong for coding genes, but much weaker for the lncRNA class. In contrast, epigenomic information (e.g., the level of specific histone modifications) distinguished hits from non-hits for both classes. Only a minority of coding and lncRNA hits mapped to potential enhancers, suggesting that most hits do not regulate neural induction through such activity. In addition to providing information that can help prioritize lncRNA genes for functional studies, these insights broadly influence the interpretation of expression data in other biological contexts and certain disease-association studies, highlighting the critical need for functional data rather than reliance on descriptive data (e.g., expression patterns). Coding genes were equally distributed between positive and negative regulators of neural induction, whereas lncRNA genes were strongly enriched for positive regulators. Remarkably, analysis of growth effects uncovered further differences between the two classes: lncRNA genes were ∼10-fold enriched for roles in differentiation, whereas coding genes preferentially regulated proliferation. Given their tissue-specific expression and recent expansion in evolution, lncRNA genes have been suggested to play critical developmental roles, especially in the mammalian nervous system.,,,,,, Our work provides systematic, genome-wide functional evidence that the lncRNA class is enriched for specialized cellular roles (e.g., regulating differentiation) rather than essential housekeeping roles,,, which are dominated by protein-coding genes. The genome-wide screen resource enables highly granular experimental studies, such as our coding-lncRNA gene Perturb-seq experiment. By targeting hundreds of coding and lncRNA genes identified as functional in neural induction and studying transcriptomes at single-cell resolution, we dissected these phenotypes in new detail. Remarkably, most negative coding hits stalled or aborted the NSC trajectory upon knockdown, whereas lncRNA gene knockdown was more permissive of diverse states, including those outside of the NSC lineage. For example, knockdown of HESX1 produced a single, intermediate PAX6− state in the NSC trajectory (Figure 7B), suggesting that these cells become stalled in their differentiation. In contrast, knockdown of lncRNA SERTAD4-AS1—a hit with a similar overall phenotype magnitude as HESX1—was enriched in multiple diverse states, including cell types outside the NSC lineage, such as neural crest cells with early markers of smooth muscle. One interpretation of this comparison is that, in neural induction, HESX1 may function primarily along a specific developmental program, whereas lncRNA SERTAD4-AS1 has function(s) dispersed across multiple cellular programs; for example, immunofluorescence staining revealed that repression of SERTAD4-AS1 produced significantly more cells positive for the TAGLN protein, a canonical marker of smooth muscle cells, suggesting that SERTAD4-AS1 promotes neural induction by suppressing other developmental programs. More generally, the collective results from this Perturb-seq study suggest a conceptual model in which lncRNA hits are enriched for function in “shepherding” cells through the differentiation process, helping prevent cells from “escaping” into non-intended cell trajectories. It is unclear whether this phenotypic heterogeneity associated with lncRNA genes is due to modifying factors, stochasticity, or other unmeasured processes, and detailed analyses of individual genes will be necessary to elucidate the underlying mechanisms. Regardless, these findings emphasize the distinct roles played by the coding and lncRNA genomes in human cell differentiation. Our systematic functional studies showcase the fundamental and surprising differences between the coding and lncRNA genomes, and provide an expansive dual genome-wide resource for investigating their function in human cell differentiation. A variety of studies have implicated both protein-coding and lncRNA genes in a wide range of neurodevelopmental disorders.,,,, Our work reveals that the lncRNA genome enables proper differentiation in critical and unexpectedly unique ways from the coding genome, providing a functional context in which to begin studying potential disease associations. More generally, this vast trove of functional data across the human genetic landscape enables fundamental biological insights that are difficult to obtain by individual gene studies or even screens of the coding or lncRNA genome alone.
The conclusions of our study are limited by the use of iPSCs rather than ESCs. The epigenetic memory of iPSCs may predispose them to specific differentiation pathways related to the parental cell type from which they were derived. Studies in other human pluripotent stem cells and in other developmental contexts (e.g., mesodermal lineages) will be necessary to generalize the findings from this work. Additionally, our characterization of SERTAD4-AS1 in this study has been limited. Full analysis of a lncRNA gene cannot rely on CRISPRi alone and requires many detailed mechanistic studies, such as the case for lincRNA-p21.,, Although we did not find that knockdown of SERTAD4-AS1 affected the expression of the neighbor coding gene, its targets and molecular mechanism are unknown. As CRISPRi can perturb a broad range of function—from cis-regulatory activity to RNA-dependent trans function—genome engineering strategies such as promoter deletion or poly(A) terminator insertion, will be particularly important for understanding the function of this gene.
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Daniel Lim ([email protected]).
This study did not generate new unique reagents.
The male CRISPRi wild-type C human induced pluripotent stem cell line was obtained from and verified by the Gladstone Institutes Stem Cell Core. Cells were maintained in Essential 8 medium (Thermo Fisher Scientific) on Matrigel (Corning).
Neural induction of human iPSCs was performed using the dual SMAD inhibition paradigm., Engineered CRISPRi-iPSCs were grown in Essential 8 (Thermo Fisher Scientific) media on Matrigel (Corning) to 80% confluency. Cells were rinsed with DPBS and dissociated with Accutase (StemPro). After centrifugation at 300 x g for 3 min and resuspension in Essential 8 media with 10 μM Y-27632 ROCK inhibitor (Selleckchem), cells were replated at a density of 250,000 cells/cm2 overnight at 37°C. The next day (D0), cells were rinsed with DPBS and changed to neural induction media, which consisted of Essential 6 media (Thermo Fisher Scientific) with freshly-added SMAD inhibitors 500 nM LDN193189 (Selleckchem) and 10 μM SB431542 (Selleckchem). Media was replaced every 2 days until the endpoint of interest, as described.
Cells were harvested by dissociation with Accutase. After washing twice with DPBS, cells were quantified on the Countess II (Thermo Fisher Scientific) and resuspended in 4% paraformaldehyde at 10 million cells/ml for 20 min at room temperature. Cells were then washed twice in a permeabilization buffer (DPBS 5% goat serum with 0.5% saponin) and blocked in the same buffer for 30 min at room temperature. Primary and secondary antibodies were added for 30 min each with 2 washes in between. Prior to running through the instrument, cells were resuspended in DPBS 5% BSA at 5 million cells/ml. Cells were gated by size (FSC) and granularity (SSC) and then for singlets by FSC-H vs. FSC-W followed by SSC-H vs. SSC-W. Cells were further gated for expression of dCas9-KRAB by coexpression of mCherry and expression of sgRNAs by coexpression of BFP before analyzing stained proteins.
Cells were grown on glass chamber slides until the desired endpoint and fixed in 4% paraformaldehyde for 15 min. After two washes in PBS, cells were permeabilized in 0.1% Triton X-100 for 15 min, followed by blocking with 10% goat serum for 1 h. Primary and secondary antibodies were added for 60 min each with 2 washes in between. DAPI (Thermo Fisher Scientific) was added 1:1,000 with the species-specific secondary antibody (see Key resources table). Slides were mounted overnight with coverslips using Aqua Poly/Mount (Polysciences).
RNA was purified using Direct-zol (Zymo) columns with DNase I treatment. RNA integrity was verified using the TapeStation 4200 (Agilent) prior to library generation. Libraries were generated using the NEBNext Ultra II Directional RNA kit (NEB) according to the manufacturer’s instructions. For polyA RNA, Oligo-dT magnetic bead selection was used prior to library generation. For total RNA, rRNA depletion using hybridization and RNase H-induced cleavage was used prior to library generation. For all samples, 2 μg of RNA was used as input. Samples were sequenced on HiSeq 4000 to >160 M reads per time-point.
Sublibraries of the hCRISPRiv2 and CRiNCL sgRNA libraries were assembled into coding and lncRNA libraries at equimolar ratios and sequenced to ensure uniform distribution. Coding and lncRNA screens were performed separately on a staggered schedule for feasibility. A lncRNA sublibrary (common sublibrary) was also performed in a separate batch to validate proliferation phenotypes (described below in “Differentiation and proliferation screen analysis”) and results were aggregated with the main lncRNA library during analysis. Lentivirus was prepared in the high titer Lenti-X 293T subclonal line (Clontech) using TransIT-LT1 (Mirus) and ViralBoost (Alstem) according to the manufacturer’s instructions. Viral supernatant was collected at 72 h, filtered, and concentrated through ultracentrifugation for 2 h at 4°C. Titer was assessed through serial dilution of single-freeze aliquots to determine the necessary amount of virus to achieve the desired MOI of 0.5. In total, approximately 650 million CRISPRi-iPSCs were transduced for coding and lncRNA libraries (each with 2 independent replicates). Once transduced and seeded, replicates were maintained independently and never re-pooled. Transduced iPSCs were selected under puromycin to >80% positivity and allowed 2 days to recover with >1,000X sgRNA library representation per replicate. At the next seeding, an additional aliquot of each sample (∼100 M cells per replicate) was frozen to measure the initial sgRNA abundance (“T0”). A second, small aliquot of cells was seeded in a sentinel 6W plate for monitoring screen progress. Doxycycline was added at 1 μM at the initiation of neural induction to activate the dCas9-KRAB CRISPRi machinery. At days 6–8, cells from the sentinel plate were fixed, permeabilized, and stained to assess neural induction progress. All samples were harvested, fixed, permeabilized, stained, and sorted into PAX6+ and PAX6- fractions (top and bottom thirds) as described above with final coverage of ∼4000X at day 8. Selection of this time-point was based on the flow cytometry analysis (Figure 1B) showing presence of both PAX6+ and PAX6- populations; earlier time-points may not allow for sufficient differentiation and would enable the discovery of sgRNAs that accelerate neural induction, but not those that prevented it. Later time-points, on the other hand, may preclude the enrichment of sgRNAs that promoted neural induction, and would mainly identify depleted sgRNAs. Genomic DNA was harvested using a chemically-catalyzed FFPE extraction method (CellData) following the manufacturer’s instructions. Sequencing libraries were prepared by targeted amplification of integrated sgRNAs using Q5 High-Fidelity Master Mix (NEB) and sequenced on an Illumina HiSeq 4000 to >50 M reads per replicate. Processing of screen data was performed as previously described using ScreenProcessing.
Individual sgRNAs were cloned using an annealing and ligation procedure, as previously described. Sense and antisense oligos that matched the desired CRISPRi protospacer sequence were annealed and ligated into a U6-driven lentiviral expression vector derived from pSico. All individually-cloned sgRNA vectors underwent Sanger sequencing to verify successful sequence insertion. Lentivirus was prepared and transduced into CRISPRi-iPSCs as described above, except in an arrayed fashion (12-well plates). Transduction efficiency was monitored through detection of a BFP cassette coexpressed by the sgRNA vector. For validation experiments, neural induction was performed as described above and cells were harvested at the endpoint for flow cytometry. Assessment of individual sgRNA phenotypes were performed by comparing the ratio of PAX6+ to PAX6- cells for populations with and without sgRNA (measured by BFP). For immunocytochemistry experiments, GFP and RFP versions of the same vectors were used and cells were plated on glass slides and analyzed as described above (“Immunocytochemistry”).
PLAC-Seq was performed as previously described., Approximately 1–5 million cells were used for library preparation. Digestion was performed using 100 U MboI for 2 h at 37°C, and chromatin immunoprecipitation was performed using Dynabeads M-280 sheep anti-rabbit IgG (Invitrogen 11203D) superparamagnetic beads bound with 5 μg anti-H3K4me3 antibody (Millipore 04-745). Sequencing adapters were added during PCR amplification. Libraries were sequenced at paired-end 150 on an Illumina HiSeq 4000. Quality and adaptor trimming were performed with fastp (0.13).
A direct-capture Perturb-seq library consisting of a mixture of 492 sgRNAs targeting 120 coding genes, 120 lncRNAs, and 12 non-targeting control sequences was cloned as a pool as previously described. Targets were selected from those with the highest genome-wide neural induction screen scores, with strong proliferation hits excluded due to expected growth dropout effects. Dual hits with mild-moderate phenotypes (γ between −1 and 1) were not excluded from the experiment. A set of lncRNA sgRNAs identified as targeting ambiguous loci (described below, under “Differentiation and proliferation screen analysis”) were included in the Perturb-seq library for assessing potential local effects, and these targets were excluded for all reported analysis. Reported analyses included the remaining 195 targets. After library sequencing to ensure sgRNA uniformity, CRISPRi-iPSCs were transduced with the Perturb-seq library at a low MOI of 0.1 (corresponding to >95% of cells with a single sgRNA integration) and >1,000X coverage using the lentiviral protocol described above. After FACS to sort for sgRNA + cells, we recovered the cells in maintenance media for 2 days before initiating dual SMAD inhibition neural induction. We performed single-cell and direct sgRNA capture at our previously determined endpoint of day 8, aiming for >100X singlet coverage per sgRNA on the Chromium V3 single-cell RNA-Seq system with Feature Barcoding (10x Genomics). Gene expression libraries were sequenced to a median depth of >50,000 reads/cell, producing a median library complexity of >3000 unique genes. Directly-captured CRISPRi sgRNAs were sequenced to >5,000 reads/cell. All Perturb-seq sequencing was performed on an Illumina NovaSeq 6000 with paired-end 100 reads.
The lncRNA gene annotation corresponding to the CRiNCL library was merged with the ENSEMBL GRCh37.p13/hg19 annotation that was used to inform the design of the hCRISPRi libraries. Duplicated entries were identified using gffcompare (0.10.6). To maintain consistency across datasets, all analyses were performed using the screen feature identifiers (“feature id”) for targeted genes as the unique identifier. GTF and BED files are deposited on GitHub and a searchable database is available on the interactive web resource with direct links to the UCSC Genome Browser.
RNA-Seq reads were analyzed for quality metrics using FastQC (0.11.8). Quality and adapter trimming were performed using bbduk (38.36) prior to transcript pseudoalignment and quantification using kallisto (0.45) on the unified lncRNA and coding gene annotation described above. Transcripts were aggregated to genes in R (3.6.3) using tximport (1.12.3). Expression values were processed using DESeq2 (1.24) for variance-stabilized transformation. Transcripts per million or Z-scaled variance-stabilized transformation values were used for downstream plotting and analysis. Time-course expression clustering was performed using maSigPro (1.56) at an alpha level of 0.1 using the mclust algorithm. Genes that did not significantly cluster to any of the temporal patterns were aggregated into the “unassigned” cluster.
Based on transcriptomic analysis, we identified expressed and dynamically regulated genes during neural induction. In order to screen equal numbers of coding and lncRNA sgRNAs, we aimed for approximately 100,000 guides for each class. Coding genes were targeted by 5 sgRNAs/locus due to well-annotated transcriptional start sites (TSS) while lncRNAs were targeted with 10 sgRNAs/locus. All sgRNAs were assembled from published CRISPRi libraries based on the human CRISPRi version 2.0 design algorithm,, which uses FANTOM cap analysis of gene expression (CAGE) data to provide highly confident transcription start site coordinates. This enabled coverage of all 18,905 coding loci using the top 5 ranked sgRNAs. For lncRNA loci, we selected the combination of CRiNCL sublibraries that covered the greatest unique number of detected lncRNAs prioritized by differential expression and the temporal clustering. In total, we screened 212,938 sgRNAs targeting 29,583 loci (all 18,905 coding loci, 10,678 lncRNA loci, with 4,523 non-targeting control sgRNAs, Figure 1D).
Analysis and hit scoring of screen data was performed as previously described using ScreenProcessing., Briefly, after sgRNA quantification, all sgRNAs represented with fewer than 50 reads in any sample were excluded. Differentiation phenotypes (ρ) were calculated by taking the log2 enrichment ratio of each sgRNA in the PAX6+ versus PAX6- sorted fractions, providing a symmetric measure of the impact on neural induction (as read out by PAX6 protein) on a log-scale (Figures 1D and S1B). Proliferation phenotypes (γ) were calculated by taking the log2 ratio of the final normalized abundance versus the initial normalized abundance of each sgRNA and normalized by the number of cell divisions. For the genome-wide screen, the final abundance was calculated from the combined sum of sgRNA abundances in sorted fractions. This approach was directly validated through a separate lncRNA sublibrary screen with 37,395 sgRNAs targeting 3,560 lncRNAs (Figure S3B), where we harvested cells without sorting for the final time-point. Upon analysis, the two methods (sorted and unsorted) produced strongly correlated γ values (Pearson r = 0.99 for hits, r = 0.81 for all targets) (Figure S3B). For all analyses, a screen score that incorporated both the effect size and significance was calculated for all targets as previously described,,; briefly, it is the product of the -log10 p-value and the phenotype magnitude of the top 3 sgRNAs. Hits were then identified based on this screen score, at an empirical FDR < 0.05 based on the distribution of non-targeting controls. A subset of CRiNCL sgRNAs that were within the highly active CRISPRi targeting window (1 kb around TSS) of coding genes were identified as ambiguous and excluded from all reported analyses (1468 loci); the coding loci were not excluded as they were more likely the cause of any potential phenotype. However, 142 coding loci were also excluded from analysis as they did not map to ENSEMBL hg19 annotated transcripts. Additional details on the screen scoring procedure and hit identification are previously described.,
For estimation of hit recovery at lower levels of coverage compared to the full dataset, precision-recall analysis was performed by downsampling the raw counts data to 10%, 20%, and 50% with 1% Gaussian noise. The downsampled data then underwent the screen processing and hit identification pipeline described above. Results were compared to the full dataset for determining precision-recall and the proportion of hits recovered at each level of sampling. The median of 3 independent downsampling replicates are reported.
For coding gene hits, gene ontology and KEGG pathway analyses were performed using clusterProfiler (3.14.3). Protein-protein interaction network analysis was performed using STRINGdb (11.0). For all of these analyses, we set the gene universe to contain all screened genes. For the interaction network analysis, statistical background distributions were generated through random sampling an equal number of genes from the gene universe.
We called significant H3K4me3-mediated chromatin interactions using the MAPS pipeline at a resolution of 5 kb. Reads were mapped to hg19/GRCh37 using BWA-MEM (0.7.17). Unmapped reads and reads with low mapping quality were discarded. PLAC-Seq anchor bins were defined by H3K4me3 CUT&Tag using MACS2 with an q-value of 0.0001. To call significant interactions, we used a zero-truncated Poisson regression-based approach to normalize systematic biases from restriction sites, GC content, sequence repetitiveness, and ChIP enrichment. We fitted models separately for AND and XOR interactions and calculated FDRs for interactions based on the expected and observed contact frequencies between interacting 5-kb bins. We grouped interactions whose ends were located within 15 kb of each other into clusters and classified all other interactions as singletons. We defined our significant chromatin interactions as interactions with 12 or more reads, normalized contact frequency (defined as the ratio between the observed and expected contact frequency) ≥ 2, and FDR <0.01 for clusters and FDR <0.0001 for singletons. This was based on the reasoning that biologically meaningful interactions are more likely to appear in clusters, whereas singletons are more likely to represent false positives. Significant interactions were overlapped with all screened genes and annotated by hit category (e.g., lncRNA differentiation hit, coding differentiation hit, and so forth). Interactions between all combinations of categories were tallied and assessed for significance using Fisher’s exact test, with FDR-adjusted p-values for multiple testing correction.
Published neural induction epigenomics datasets were downloaded from the NIH Roadmap Epigenomics Project. Raw reads underwent adaptor and quality trimming using bbduk (38.36) prior to alignment using HISAT2 (2.1). For all overlap analyses, regions were counted if they overlapped by at least 1 bp within a 2 kb window centered around the transcription start site of screened genes. For quantitative analysis of ChIP-Seq signal, reads were mapped to screened genes using featureCounts 1.6 and normalized to input. For peak calling analysis of histone marks, significant peaks were identified using MACS2 (2.2.6) and replicate samples were IDR-filtered before determining overlap with screened genes following parameters of the ENCODE ChIP-Seq Pipeline for broad and narrow peaks. Genomic coordinates of enhancer regions from published datasets, were downloaded as BED files and analyzed for overlap with screened genes by the same criteria. All analysis was performed on the hg19/GRCh37.p13 reference genome build, using the unified coding and lncRNA annotation described above. For visualization, bam files for replicates were merged and converted to bigwig files using deepTools 3.4.0. For analysis of broad H3K4me3 domains, we followed the procedure described by running MACS2 in with the “--broad” flag for broad peak analysis. For all genes, promoter regions (2 kb window centered around the TSS) were analyzed for overlap with H3K4me3 domains, which were categorized by percentile on their peak breadth. The top 5 percentile of peaks were assigned “broad H3K4me3” domains, as described.
Feature data consisted of the transcriptomic and epigenomic datasets described above. Transcriptomic features included the scaled variance-stabilized and TPM values at each time-point (polyA or total RNA), log2 fold-change from day zero for each time-point, as well as variables for the maximum/median expression levels, maximum fold-change, number of exons, gene length, and isoform count. Epigenomic features consisted of the histone mark signal at the promoters (within 2 kb window surrounding the TSS) of screened genes. To prevent confounding epigenomic signal from nearby coding and lncRNA promoters, all coding-lncRNA gene pairs with promoters within the 2 kb window were excluded from classification. All predictor variables were centered around the mean and standardized. To generate machine learning models, the screen hit status was binarized and used as the response variable. For all classification models, coding genes and lncRNAs hits were compared to non-hits of the same class. For example, to analyze features of coding genes, coding hits were binarized as 1 and analyzed against coding non-hits binarized as 0. Several classes of models (elastic net logistic regression, random forest, gradient boosting machines) were generated and tested, producing similar results. Training and validation were performed using randomly-sampled partitions of 70% training data and 30% validation data. Model parameters were estimated using 5-repeats of 5-fold cross-validation. Model performance was evaluated on the validation set using the area under the receiver-operating characteristic (ROC) curve. This resampling, training, validation, and ROC assessment was repeated for 1,000 iterations, and the average AUC is reported. Each feature was additionally analyzed individually using ROC analysis to assess its association with hits. For this individual analysis, statistical significance was determined using 1,000 iterations of bootstrapping at the 99% confidence level. Variables with confidence intervals that crossed AUC 0.5 were considered non-significant.
Differences in phenotype distributions between coding and lncRNA hits were assessed using the Kolmogorov-Smirnov (K-S) test. Skews of positive and negative phenotype distributions were assessed for significance through permutation testing. Using the baseline number of total hits for each library, we permuted the label of “positive” and “negative” hit status and calculated the ratio of positive to negative hits for 1 million trials. Skews between proliferation and differentiation hits were calculated in a similar manner, with permutation of the “proliferation” and “differentiation” hit labels performed for 1 million trials. In each case, the p-value was determined by the fraction of trials producing a more extreme ratio.
Paired-end 100 reads for gene expression and sgRNA libraries were processed using 10x cellranger software (4.0) following developer instructions for CRISPR library analysis. Data was processed on the unified lncRNA and coding gene reference described above as well as the newer GRCh38 genome, which led to similar results. Initial quality filtering was performed with background removal and empty droplet identification using cellbender (2.1). Barcodes identified as those belonging to cells from the cellranger and cellbender pipelines were compiled. Assignment of sgRNAs to cellular barcodes was performed using a two-component mixture model, consisting of Poisson (lower) and Gaussian (upper) distributions, which enabled doublet identification (barcodes with >1 sgRNA). After excluding doublets, we obtained 84,808 single cells harboring the distinct genetic perturbations. Quality scoring was performed based on unique genes detected, mitochondrial RNA percentage, and ribosomal RNA percentage. In order to assess whether apparent low-quality cellular transcriptomes were the result of any perturbations, no cells were filtered on quality metrics until after clustering and analysis (described below). After batches were integrated using multi-canonical correlation analysis in Seurat (3.9) based on the top 5000 variable genes, data was variance-stabilized and transformed using SCTransform (0.3.2). Dimensionality reduction was performed using principal components analysis followed by uniform manifold approximation projection of the top 30 principal components. High-resolution clustering of cells was performed using a shared nearest neighbor network with k = 30 and the Leiden algorithm set at a resolution of 1.2. This resulted in 30 total clusters, and clusters driven by the quality metrics described above were excluded. In total, 78,393 high-quality single-cell transcriptomes containing single sgRNA perturbations were used for all reported analysis.
On-target knockdown of Perturb-seq targets was analyzed in a similar fashion as previously., Within each batch (10x Genomics well), cells harboring sgRNAs against each target were analyzed compared to cells harboring non-targeting control sgRNAs. To increase statistical confidence and minimize bias from gene dropout, cells were merged in a pseudobulk approach, using each batch as an individual replicate.
Output bam files from cellranger were processed for RNA velocity analysis of spliced an unspliced transcripts using velocyto. Velocity vectors were computed and visualized using scVelo, with 30 principal components and 30 neighbors based on the top 3000 highly variable genes for computational feasibility.
To visualize phenotypes of Perturb-seq sgRNAs in the UMAP embedding, normalized density heatmaps of cells were constructed. For each target, cells harboring the relevant sgRNAs were identified and located in UMAP space. Gaussian kernel densities were calculated for these cells in 2 dimensions, with 10,000 total bins (100 bins in each dimension spanning the full coordinate range). To normalize for the background distribution, this density calculation was performed with the same parameters for non-targeting control sgRNAs, and this background was subtracted. The density profile was then visualized by color intensity and overlaid onto the UMAP projection.
After calculating the normalized density for each target in the 2D UMAP embedding, density-based spatial clustering and application with noise (DBSCAN) was applied to identify areas of high density for each target. Regions of zero or near zero density were excluded using a threshold of 1% of the top The DBSCAN epsilon parameter of 1 and a minimum threshold of 25 bins were used for the top 50% of regions with highest densities. These regions were considered the enriched cell states for each target. To identify targets with similar density profiles, the overlap coefficient was determined for all pairwise comparisons and this pairwise table was converted to a distance matrix for unsupervised hierarchical clustering. To merge overlapping cell states from different targets into a universal set of cell states, all states across all targets were compared by overlap coefficient and collapsed to 29 cell states after hierarchical clustering, with the final k determined by the silhouette method for values ranging from 2 to 50. |
|
PMC9648147 | 36386721 | Wei-Gang Xin,Xin-Dong Li,Yi-Cen Lin,Yu-Hang Jiang,Mei-Yu Xu,Qi-Lin Zhang,Feng Wang,Lian-Bing Lin | Whole genome analysis of host-associated lactobacillus salivarius and the effects on hepatic antioxidant enzymes and gut microorganisms of Sinocyclocheilus grahami | 27-10-2022 | Sinocyclocheilus grahami,lactobacillus salivarius,genomic characterization,antioxidant enzymes,gut microbiome | As a fish unique to Yunnan Province in China, Sinocyclocheilus grahami hosts abundant potential probiotic resources in its intestinal tract. However, the genomic characteristics of the probiotic potential bacteria in its intestine and their effects on S. grahami have not yet been established. In this study, we investigated the functional genomics and host response of a strain, Lactobacillus salivarius S01, isolated from the intestine of S. grahami (bred in captivity). The results revealed that the total length of the genome was 1,737,623 bp (GC content, 33.09%), comprised of 1895 genes, including 22 rRNA operons and 78 transfer RNA genes. Three clusters of antibacterial substances related genes were identified using antiSMASH and BAGEL4 database predictions. In addition, manual examination confirmed the presence of functional genes related to stress resistance, adhesion, immunity, and other genes responsible for probiotic potential in the genome of L. salivarius S01. Subsequently, the probiotic effect of L. salivarius S01 was investigated in vivo by feeding S. grahami a diet with bacterial supplementation. The results showed that potential probiotic supplementation increased the activity of antioxidant enzymes (SOD, CAT, and POD) in the hepar and reduced oxidative damage (MDA). Furthermore, the gut microbial community and diversity of S. grahami from different treatment groups were compared using high-throughput sequencing. The diversity index of the gut microbial community in the group supplemented with potential probiotics was higher than that in the control group, indicating that supplementation with potential probiotics increased gut microbial diversity. At the phylum level, the abundance of Proteobacteria decreased with potential probiotic supplementation, while the abundance of Firmicutes, Actinobacteriota, and Bacteroidota increased. At the genus level, there was a decrease in the abundance of the pathogenic bacterium Aeromonas and an increase in the abundance of the potential probiotic bacterium Bifidobacterium. The results of this study suggest that L. salivarius S01 is a promising potential probiotic candidate that provides multiple benefits for the microbiome of S. grahami. | Whole genome analysis of host-associated lactobacillus salivarius and the effects on hepatic antioxidant enzymes and gut microorganisms of Sinocyclocheilus grahami
As a fish unique to Yunnan Province in China, Sinocyclocheilus grahami hosts abundant potential probiotic resources in its intestinal tract. However, the genomic characteristics of the probiotic potential bacteria in its intestine and their effects on S. grahami have not yet been established. In this study, we investigated the functional genomics and host response of a strain, Lactobacillus salivarius S01, isolated from the intestine of S. grahami (bred in captivity). The results revealed that the total length of the genome was 1,737,623 bp (GC content, 33.09%), comprised of 1895 genes, including 22 rRNA operons and 78 transfer RNA genes. Three clusters of antibacterial substances related genes were identified using antiSMASH and BAGEL4 database predictions. In addition, manual examination confirmed the presence of functional genes related to stress resistance, adhesion, immunity, and other genes responsible for probiotic potential in the genome of L. salivarius S01. Subsequently, the probiotic effect of L. salivarius S01 was investigated in vivo by feeding S. grahami a diet with bacterial supplementation. The results showed that potential probiotic supplementation increased the activity of antioxidant enzymes (SOD, CAT, and POD) in the hepar and reduced oxidative damage (MDA). Furthermore, the gut microbial community and diversity of S. grahami from different treatment groups were compared using high-throughput sequencing. The diversity index of the gut microbial community in the group supplemented with potential probiotics was higher than that in the control group, indicating that supplementation with potential probiotics increased gut microbial diversity. At the phylum level, the abundance of Proteobacteria decreased with potential probiotic supplementation, while the abundance of Firmicutes, Actinobacteriota, and Bacteroidota increased. At the genus level, there was a decrease in the abundance of the pathogenic bacterium Aeromonas and an increase in the abundance of the potential probiotic bacterium Bifidobacterium. The results of this study suggest that L. salivarius S01 is a promising potential probiotic candidate that provides multiple benefits for the microbiome of S. grahami.
Sinocyclocheilus grahami, belonging to the family Cyprinidae, subfamily Barbinae, and genus Sinocyclocheilus, is unique to Dianchi Lake in Yunnan. Known as one of the “four famous fish,” S. grahami is fished, consumed, and considered an important economic fish (Yin et al., 2021). However, because of the destruction of its habitat and invasion by exotic species, S. grahami was listed as a Grade II protected animal in 1989, and further as an endangered animal in 1998. In 2007, to save the species from extinction, the artificial reproduction of S. grahami was achieved for the first time. After four generations of manual selection and breeding, a high-quality new national variety (“S. grahami, Bayou No. 1”) was certified in 2018 (Yin et al., 2021). Nevertheless, during the process of artificial breeding, antibiotics were commonly used to treat diseases in specimens, including gill and skin inflammation (Yang et al., 2007). In aquaculture, antibiotics are often used as additives to treat and prevent diseases, because they can inhibit the reproduction of bacteria (Vaseeharan and Thaya, 2014). However, antibiotic overuse has led to the emergence of multidrug-resistant pathogens, damaging the environment, and posing a risk to public safety and health. Previous studies have shown that superbugs resistant to multiple antibiotics can be transmitted among animals and humans through contaminated food and water (Davies and Davies 2010). For instance, contamination with Vibrio parahaemolyticus was observed in 95 (38.0%) of 250 aquatic product samples from Guangdong, China, among which 90.53% of the strains showed streptomycin resistance (Xie et al., 2017). In January 2020, the Ministry of Agriculture and Rural Affairs of China issued a comprehensive ban on the addition of antibiotics to animal feed to address issues related to drug residues and antibiotic resistance (Zhou et al., 2021). Therefore, the excavation of a safe and effective alternative to antibiotics, including probiotics, is urgently needed. Many mechanisms have been proposed to explain the positive effects of probiotics, including stimulating the immune system, helping the host to resist the invasion of external harmful substances and disease-curing organisms, and aiding with digestion (Macfarlane and Macfarlane, 1997; Dong et al., 2018; Sun et al., 2021). For instance, Bacillus cereus NY5 can antagonize Streptococcus lactis by modulating specific and non-specific immunity in tilapia (Ke et al., 2022). Lactobacillus rhamnosus GG normalizes gut dysmotility induced by environmental pollutants (oxytetracycline, arsenic, polychlorinated biphenyls and chlorpyrifos) via affecting serotonin level in zebrafish larvae (Wang et al., 2022). Bifidobacteria are proved to be capable of relieving colitis symptoms in both in vivo and in vitro experiments through following potential mechanisms (e.g., enhancing the hosts’ antioxidant activity, decreasing myeloperoxidase activity, and reactive oxygen species, et. al; Yao et al., 2021). Moreover, in contrast with traditional antibiotics, probiotics fight bacterial diseases and treat inflammation without increasing resistance, through their antimicrobial, antioxidant, anti-inflammatory, and immunomodulatory action (Vanderhoof and Young, 1998; Abd El-Ghany et al., 2022; Yang et al., 2022). Also, probiotics used as a water supplement can improve water quality by affecting the microbial communities of the environment and reducing metabolic waste in the water system (Talpur et al., 2013). These characteristics have led to the widespread use of probiotics in animal farming, particularly in aquaculture. In aquaculture, probiotics can regulate and rebuild the microecological balance of the gut, enhance immunity against diverse pathogens, and improve the conversion rate of feed energy and growth performance (Merrifield et al., 2010; Akhter et al., 2015; Hoseinifar et al., 2018; Chauhan and Singh, 2019; Ringø, 2020). For instance, Lactobacillus salivarius can inhibit the growth of Vibrio spp. within pike-perch larvae, as well as improve ossification and survival rates (Ljubobratovic et al., 2020). However, while different sources of probiotics have shown consistent and favorable results in higher vertebrates, the effects on the gut of fish are variable. Meanwhile, the use of host-associated probiotics as feed additives has a positive effect on fish farming, as reported by Tarkhani et al. (2020). Therefore, probiotic bacteria isolated from the host gut show greater potential to replace antibiotics than non-specific probiotics. In recent years, increasing evidence has shown that the gut microbiota plays a key role in maintaining health and controlling disease, regulating many important physiological functions of the host (Tran et al., 2018; Tang et al., 2021). Lactic acid bacteria (LAB) and their metabolic derivatives can improve the gut microbiota and enhance the host’s immunity against external harmful substances and pathogenic bacteria (Wang et al., 2021). For instance, feeding crucian carp Lactococcus lactis was effective as a treatment for intestinal inflammation and mucosal barrier function damage caused by Aeromonas hydrophila (Dong et al., 2018). Lactobacillus rhamnosus recovered the growth of zebrafish larvae under perfluorobutanesulfonate exposure via its antioxidant properties, reshaping the gut microbiota, and enhancing the production of bile acids (Sun et al., 2021). Previous studies reported that the gut microorganisms and antioxidant system have synergistic effects against harmful external substances in animals (Uchiyama et al., 2022). Interestingly, LAB was reported to be involved in the activation of antioxidizing systems in animals, promoting their adaptation to external changes (Han et al., 2022). However, the effect of LAB on S. grahami is not reported. This study is the first to investigate the intestinal microbial changes in S. grahami after feeding with host-associated LAB. Lactobacillus salivarius, is a host-associated bacterium previously isolated from the gut of S. grahami (Xin et al., 2022). We have isolated and obtained five potential probiotics with high antibacterial activity (i.e., two Bacillus subtilis, two Lactobacillus sake and one L. salivarius) from the gut of S. graham. Particularly, the bacteriocin LSP01 produced by L. salivarius 01 exhibited the best antibacterial activity against A. hydrophila. L. salivarius inhibits the growth of pathogens by disrupting cellular activity and inducing pore size formation in A. hydrophila cells. Therefore, in this study, L. salivarius 01 was used to explore the probiotic potential. The whole genome was sequenced to evaluate the safety and potential properties of the strain at the genetic level. In addition, the effect of L. salivarius on the antioxidant capacity of the hepar and intestinal microbial populations of S. grahami was assessed using feeding experiments. As a result, L. salivarius was confirmed as a safe and effective potential probiotic that can increase the antioxidant capacity and improve the gut microbiota of S. grahami.
Lactobacillus salivarius S01 is a host-associated potential probiotic isolated from the gut of S. grahami (bred in captivity) and preserved in Engineering Research Center for Replacement Technology of Feed Antibiotics of Yunnan College (Xin et al., 2022). For routine use, the strains were grown and subcultured in MRS broth at 37°C for 24 h.
The total genomic DNA of L. salivarius S01 was extracted using the sodium dodecyl sulfate (SDS) method combined with a purification column. Total genomic DNA was sequenced using an ONT PromethION sequencer (Oxford Nanopore Technologies, Oxford, UK). For filtering, low-quality and short-length reads were discarded from the raw reads. The reads were assembled using Unicycler V0.4.9 software (Wick et al., 2017). The annotation of the assembled genome of L. salivarius S01 was performed using Prokka V1.12 software (Seemann, 2014). RepeatMasker V4.1.0 software was used to predict repeat sequences in the genome of L. salivarius S01. The prediction of pseudogenes of L. salivarius S01 was performed using Pseudofinder software. MinCED V0.4.2 software was used to predict the sequence of Clustered Regularly Interspaced Palindromic Repeats (CRISPRs) on the chromosome of L. salivarius S01. Genomic islands in the genome of L. salivarius S01 were predicted by IslandViewer 4 (http://www.pathogenomics.sfu.ca/islandviewer/). The prophage in the genome of L. salivarius S01 was predicted using PhiSpy (https://github.com/linsalrob/PhiSpy). The predicted gene sequences were compared using several functional databases, including Cluster of Orthologous Groups (COG; Tatusov et al., 2000), Kyoto Encyclopedia of Genes and Genomes (KEGG; Kanehisa et al., 2004), Swiss-Prot, and RefSeq, using BLAST+ (2.5.0+). The gene function annotation results were obtained, followed by gene function annotation analysis, such as COG and KEGG metabolic pathway enrichment analysis and gene ontology (GO) function enrichment analysis. Finally, the secondary metabolism gene cluster was analyzed using antiSMASH (v5.2.0), while the bacteriocin synthesis gene cluster of the strain was analyzed using BAGEL4 (Blin et al., 2019). The presence of antimicrobial resistance genes was compared using the comprehensive antibiotic resistance database (CARD) (McArthur et al., 2013). The CARD database is constructed as the Antibiotic Resistance Ontology (ARO) taxonomic unit to correlate information on antibiotic modules and their targets, gene variants, etc.
Standard disc diffusion was performed for antibiotic susceptibility testing according to the Clinical and Laboratory Standards Institute (CLSI; Keter et al., 2022). L. salivarius S01 were cultured on MRS broth for 24 h to determine the antibiotic sensitivity of selected strains to antibiotic (i.e., tetracycline, erythromycin, penicillin, ampicillin, and chloramphenicol). The experiment was repeated three times independently.
An in vitro artificially simulated gastrointestinal juices model was applied, following the previously reported methods, with minor modifications (Li et al., 2020). The simulated gastric juice was prepared by adding 10 g/l pepsin (Solarbio, Beijing, China) to 16.4 ml of sterile 0.1 mol/l HCL, filtered through a membrane with 0.22 μm pore, and adjusted to pH 2.0, 3.0, and 4.0 using sterile 1 mol/l NaOH. The simulated intestinal juice was prepared by adding 10.0 g/l trypsin (Solarbio, Beijing, China) and 6.8 g of KH2PO4 to 500 ml of sterile ddH2O. The pH was adjusted to pH 6.8 using 1 mol/l NaOH and filtered through a membrane with 0.22 μm pore. One milliliter of L. salivarius S01 (approximately 107–108) suspension was inoculated into 5 ml of simulated gastric juice at pH 2.0, 3.0, and 4.0, and incubated for 3 h at 37°C. One milliliter of L. salivarius S01 (approximately 107–108) suspension was also inoculated into 5 ml of simulated intestinal juice for 4 h. Then, the bacterial solutions were cultured on MRS agar for 24 h to determine the tolerance of selected strains to simulated gastrointestinal juice: Survival rate (%) = lg N1 / lg N2 × 100%, where N2 is the total viable counts of the selected strains at 0 h and N1 is the total viable counts after exposure to the simulated gastrointestinal juice for different time periods.
The S. grahami specimens used for the experiments were donated by Yiliang jianzhiyuan Food Co., Ltd. (Kunming, Yunnan). After one week of quarantine, a total of 60 healthy, non-injured, and undeformed S. grahami fingerlings (8.0 ± 1.5-mm long) were randomly assigned to six continuously aerated 300-L aquariums, equipped with temperature and oxygen supply control devices. The experiment was comprised of two groups: the control group (n = 3, 10/cylinder), named as LSC, which was fed basal feed (Satura, Kunming, China), and the treatment group (n = 3, 10/cylinder), named as LSM, which was fed the basal diet supplemented with L. salivarius (approximately 1 × 107 CFU/g). All samples (60) were treated. L. salivarius S01 was inoculated in MRS broth with 1% pitch rate and incubated at 37°C for 24 h. Afterwards, bacterial cultures were centrifuged at 4,000 × g for 10 min, supernatant was discarded, and cell pellet was resuspended in PBS solution. L. salivarius S01 was mixed with the diet and homogenized, and pelleted using oscillating granulator (Daxiang, Guangzhou, China) with 125 μm mesh. At last, mixed diet was dried at room temperature in a ventilated room. Dried pellets for plate-coating inspection. L. salivarius S01 (approximately 1 × 107 CFU/g) was measured by viable bacteria. The specimens were fed a quantity of 3% of their body weight twice a day for 28 consecutive days. The light–dark cycle ratio was 14 h: 10 h, and all water quality standards, including temperature (18 ± 0.5°C), pH (8.0 ± 0.5), and DO (8.5 ± 0.12 ppm), were monitored daily. Half of the water in the tank was replaced each day to ensure the best growth conditions for the fish. The experimental animals were processed in accordance with the recommendations from the Guide for the Care and Use of Laboratory Animals. The experimental protocol was approved by the Ethics Committee of Research of Kunming University of Science and Technology.
At the end of the feeding experiment and after fasting for 24 h, three fish were randomly selected from the control and experimental groups, respectively, and the tissue samples were collected. The hepar of S. grahami were collected by dissection, then triturated in pre-cooled homogenization medium (0.01 M Tris–HCl, 0.001 M EDTA-Na2, and 0.01 M sucrose, pH 7.4). S. grahami hepar were collected by autopsy and ground in pre-chilled homogenizing medium (0.01 M Tris–HCl, 0.001 M EDTA-Na2, and 0.01 M sucrose, pH 7.4), and centrifuged (4,000 × g for 10 min at 4°C). The resulting supernatant was collected and stored at −80°C for subsequent analysis of hepatic antioxidant enzyme activity. Gut samples from the same fish were collected, placed in Eppendorf tubes, and stored at −80°C.
The supernatant was incubated with the enzyme-substrate and read at the indicated wavelength using a UV-8000ST spectrophotometer (Shanghai Yuanxi Instruments Co., Ltd.). The enzyme activity assay was performed in triplicate. The catalase (CAT; E.C.1.11.1.6), superoxide dismutase (SOD; E.C.1.15.1.1), and peroxidase (POD; E.C.1.11.1.7) activities and malondialdehyde (MDA) levels were determined. Commercial assay kits were purchased from Nanjing Jiancheng Institute of Bioengineering (Nanjing, China), and all enzyme activity assays were measured, according to the kit instructions. CAT activity was measured using the CAT Activity Assay Kit (cat. no. A007-1-1). CAT can catalyze the decomposition of hydrogen peroxide, and ammonium molybdate can quickly prevent the decomposition of hydrogen peroxide. The remaining hydrogen peroxide can quickly combine with ammonium molybdate to form a pale-yellow complex that can be measured at 405 nm. SOD activity was measured using a total superoxide dismutase (T-SOD) detection kit at 550 nm (hydroxylamine method; cat. no. A001-1-2). The change in POD activity was measured using a peroxidase assay kit at 420 nm (cat. no. A084-1-1). The MDA levels were measured using the Malondialdehyde Assay Kit at 532 nm (TBA method) (cat. no. A003-1-1).
The fish gut samples were sent to Shanghai Majorbio Bio-pharm Technology Co., Ltd. for sequencing. The sequencing primers were primer 338F (ACTCCTACGGGAGGCAGCAG) and primer 806R (GGACTACHVGGGTWTCTAAT) for amplifying the V3-V4 region of 16S rRNA gene. Sequence analysis was performed using QIIME 1.7 and FLASH 1.2. QIIME 1.7 was used to remove low-quality fragments from the original reads, and FLASH 1.2 was used to complete read merging. Operational taxonomic units (OTUs) were clustered using Uparse based on the threshold of the similarity being above 97%. The RDP classifier algorithm was used to compare the 97% similar OTU representative sequences with the SILVA database for taxonomic analysis.
Gut microbial data validated by multiple comparisons (Knight et al., 2018). Student’s t-test (two-tailed test) was used to identify significant differences between two groups in phylum and genus level. Multiple testing adjustment of the data by Benjamini-Hochberg (BH). All experiments were performed in triplicate, and the results were presented as the mean ± standard deviation (SD). Independent samples t-test (two-tailed test) was used to evaluate between-group variance. One-way ANOVA plus least significant difference (LSD) method was employed to analyze multi-group significance. p-values <0.05 were considered significant.
A genome circle map drawn by integrating the predicted genome information annotation is shown in Figure 1. The innermost circle shows the coding regions (CDS) and non-coding RNA regions (rRNA and tRNA) of the genome. The whole genome sequencing results showed that the genome size of L. salivarius S01 was 1,737,623 bp with a GC content of 33.09%. A total of 1753 coding DNA sequences (CDSs), 22 rRNA operons (including 7 23S rRNA, 7 16S rRNA, 8 5S rRNA), and 78 transfer RNA (tRNA) genes were found in the genome of L. salivarius S01. In addition, one CRISPR sequence and four gene islands were predicted. The CDSs on the chromosome of L. salivarius S01 were annotated using the COG database, and the results are shown in Supplementary Figure S1. One thousand and five genes were annotated into 21 functional categories through the COG database. The major COGs were translation/ribosomal structure and biogenesis (167), amino acid transport and metabolism (106), carbohydrate transport and metabolism (88), replication/recombination/repair (78), and cell wall/membrane/envelope biogenesis (72). Furthermore, GO analysis revealed that 470 genes were classified into biological processes, 991 genes were classified into cellular components, and 1,037 genes were associated with molecular functions (Figure S1). One thousand and hundred ninety genes involved in KEGG metabolic pathway analysis were classified into five major categories: Metabolism class (866), followed by Genetic Information Processing (183), Environmental Information Processing (112), Organismal Systems (15), and Cellular Processes (14; Supplementary Figure S2).
Genes for the following probiotic features were examined: tolerance to stress conditions, aid in adhesion and colonization, antioxidative stress immunity, and protective repair of DNA and proteins. The genomic analysis detected 21 genes encoding proteins that may be related to the tolerance of digestive enzymes, bile salts, and acidic environments. Furthermore, genes related to immune response against oxidative stress, and protein and DNA molecular repair protection were also present in the genome (Table 1).
Three gene clusters related to antibacterial substances synthesis were predicted in the L. salivarius S01 genome (Figure 3). The antiSMASH database predicted the existence of a polyketide synthase (T3PKS) synthesis gene cluster in the genome. Based on BAGEL4 platform, the results showed that the genome contained two bacteriocin synthesis gene clusters, as predicted: Enterolysin A as the core gene, including one immune gene and multiple transporter genes, and sakacin_G_skgA1 (class II bacteriocin) as the core gene, including a bacteriocin immune protein gene and a replication initiation protein gene.
The predicted results of antibiotic resistance genes are shown in Table 2. The identities of tet (L) and ErmC were 98.03 and 93.85%, respectively; The identities of Escherichia coli EF-Tu mutants conferring resistance to kirromycin and Staphylococcus aureus rpoB mutants conferring resistance to rifampicin were 73.03 and 72.14%, respectively; The identities of other antibiotic resistance genes was less than 70%. Antibiotic sensitivity tests showed that Lactobacillus salivarius S01 had good antibiotic sensitivity (Table 3). L. salivarius S01 were sensitive to some antibiotics (penicillin, ampicillin, and chloramphenicol); L. salivarius S01 were intermediate sensitive to tetracycline. L. salivarius S01 were resistant to erythromycin.
The simulated intestinal and gastric juices tolerance of L. salivarius S01 at pH 2.0, 3.0, and 4.0 are shown in Table 4. The results showed that, under simulated gastric juice treatment, the survival rate of the selected strains gradually decreased at lower pH, but maintained a high survival rate (> 79.84%) at all pH conditions. Under the simulated intestinal juice treatment, the survival rate of the strain was 94.04%, indicating that the strain adapted well to intestinal conditions.
As shown in Table 5, the levels of hepatic SOD, CAT, and POD in the experimental group fed with L. salivarius S01 were significantly higher than those in the control group (p < 0.05). Compared with the control, the SOD, CAT, and POD enzyme activity of LSM increased significantly to 25.87 ± 2.21 U/g (1.5-fold), 124.15 ± 3.91 U/g (1.8-fold), and 577.67 ± 40.22 U/g (2.0-fold), respectively (p < 0.05). The MDA levels were significantly decreased by 43.04% in LSM groups compared with control groups (p < 0.05).
After eliminating low-quality reads from the raw sequences of six intestinal bacterial samples of S. grahami, a total of 563,372 high-quality reads were obtained. These were then clustered into 572 OTUs based on the 97% 16S rRNA sequence similarity. Alpha diversity indices were used to evaluate the richness and diversity of the gut microbiota in the experimental and control groups, as shown in Figure 4. Some difference was observed among the indices reflecting community richness (including Chao1 and Ace) and the indices reflecting community diversity (including Shannon and Simpson) between the two groups. The beta diversity of the samples was analyzed using principal component analysis (PCoA) and non-metric multidimensional scaling analysis (NMDS). The PCoA and NMDS results showed that the six samples were clearly divided into two clusters, consistent with the grouping (Figure 5). The reliability of the model was reflected by the stress value, equal to 0.0 in NMDS. These results demonstrated that while the bacterial supplementation did not change the richness of the gut microbiota, the community data was significantly altered compared with the control group. At the phylum level, the two groups of gut microbes were mainly composed of Proteobacteria, Firmicutes, Actinobacteriota, and Bacteroidota, among others (Figure 6). Multiple testing adjustment of the gut microbial data by Benjamini-Hochberg (BH) method. The corrected p-values for all the phylum level are 0.4206. In the control group, Proteobacteria was the most dominant bacterial phylum, accounting for 93.86% of all OTUs, followed by Firmicutes, Actinobacteriota, and Bacteroidota, accounting for 3.91, 1.55, and 0.56%, respectively. Compared with the control, the order of the proportion of dominant bacteria in the experimental group did not change. However, in the experimental group, the proportion of Proteobacteria decreased to 63.78% of the control group, accounting for 59.86%, and the proportion of Firmicutes increased to 8.1-fold that of the control group, accounting for 31.64%. Compared with the control group, the proportions of the Actinobacteriota and Bacteroidota phyla increased to varying degrees, accounting for 3.95 and 3.82% of the bacterial microbiota, respectively. At the genus level, we mapped the top 10 dominant bacterial genera (Figure 7). Multiple testing adjustment of the gut microbial data by Benjamini-Hochberg (BH) method. The corrected p-values for the top 10 dominant bacterial genera are 0.4269. Among them, the proportion of Aeromonas varied between the control and experimental group. The proportion varied from 58.23 to 6.72%. The differences of the other nine genera between the control and experimental groups were not significant. Burkholderia-Caballeronia-Paraburkholderia, Blautia, Ralstonia, Bifidobacterium, and Subdoligranulum increased by 10.75-, 6.30-, 12.32-, 4.72-, and 6.45-fold, respectively, while Candidatus_Bacilloplasma and Acinetobacter increased from a negligible amount to 2.82 and 2.68%, respectively. Among them, the microbiota proportion of unclassified Enterobacteriaceae and Gammaproteobacteria decreased from 25.74 and 2.10% to a negligible amount. These results indicate that after treatment with L. salivarius S01, the proportion of dominant bacteria in the gut microbiota was decreased, while the overall microbiota diversity increased.
To determine the gut bacteria genus involved in the antioxidant capacity observed in the host, we performed correlation analysis between gut bacterial genus and hepatic antioxidant enzyme activity (Figure 8). Shewanella and unclassified_f__Enterobacteriaceae were positively associated with the MDA contents of the hepar. Bifidobacterium, Burkholderia-Caballeronia-Paraburkholderia, Ralstonia, Ruminococcus, Rikenellaceae_RC9_gut_group, Acinetobacter, Faecalibacterium, and Eubacterium_hallii_group were positively associated with the CAT contents of the hepar. Burkholderia-Caballeronia-Paraburkholderia and Ralstonia were negatively associated with the MDA contents of the hepar. Aeromonas and Shewanella were negatively associated with the CAT contents of the hepar. Unclassified_f__Enterobacteriaceae and unclassified_c__Gammaproteobacteria were negatively associated with the POD contents of the hepar.
Microecological preparations can improve the immunity of the host, and other beneficial characteristics, such as a lack of drug resistance, and no toxic side effects (Mingmongkolchai and Panbangred, 2018). At present, probiotic microecological preparations are widely used in aquaculture (Gopi et al., 2022; Zhu et al., 2022). However, there is no standard for the selection of probiotics for use in microecological preparations. Host-associated probiotics isolated from the hosts may be have reportedly the most beneficial effects (Giri et al., 2014; Hao et al., 2017; Sharifuzzaman et al., 2018; Zuo et al., 2019). In this study, L. salivarius S01, originally isolated from the gut of S. grahami, was used as a feed additive in S. grahami. According to our previous study, a bacteriocin produced by L. salivarius S01 exhibited excellent inhibition effects against 12 common pathogens (both fish- and food-derived; Xin et al., 2022). This phenomenon was also confirmed by the prediction results of the antiSMASH secondary metabolite gene cluster and the BAGEL4 bacteriocin synthesis gene cluster prediction. The synthetic gene cluster of T3PKS was identified in the antiSMASH database. T3PKS expressed by the gene cluster can assist in the production of polyketides. Polyketides are a class of substances with broad antibacterial, anticancer, antioxidant, antiparasitic and anti-inflammatory activities (Bandgar et al., 2010; Mao et al., 2016; Patil et al., 2016; Vogel et al., 2008). Two bacteriocin gene clusters, Enterolysin_A and sakacin_G_skgA1, were detected in the BAGEL4 database. Enterolysin_A gene cluster encodes a cell wall degrading bacteriocin, a class III bacteriocin (Dos Santos et al., 2021; Nilsen et al., 2003). The sakacin G bacteriocin encoded by the sakacin_G_skgA1 gene cluster can lyse sensitive cells, leading to the leakage of enzymes and DNA, thereby inducing apoptosis in bacteria (Todorov et al., 2011). In addition, biochemical experiments and genomic analysis showed that Lactobacillus salivarius S01 had good antibiotic sensitivity. Therefore, L. salivarius S01 show great potential for use in the control of pathogens in aquaculture. L. salivarius exhibits good resistance to acid and bile salt, adjusting the gut microecological balance by changing the ratio of symbiotic LAB and other bacteria, as well as reducing the gut pH (Chaves et al., 2017; Messaoudi et al., 2013). Meanwhile, L. salivarius has immunomodulatory, anti-inflammatory, and anti-infectious properties (Langa et al., 2012), it also can stimulate Caco-2 cells to inhibit IL-8 production, as well as promote the recovery of gut epithelial cells (Arribas et al., 2012). The tolerance of L. salivarius to acidic conditions plays a key role in its colonization of the intestine, thereby ensuring its probiotic potential. In this study, the results of the in vitro assay in a simulated gastrointestinal juices environment demonstrated that L. salivarius S01 was able to tolerate the extreme conditions of low pH and proteases. Meanwhile, genes associated with probiotic potential, such as environment resistance, adhesion capacity, protective repair of DNA and proteins, and antioxidant immunity, were identified in the genome. Furthermore, previous studies reported that the antioxidant capacity of animals is important because the body tends to generate reactive oxygen species (ROS) through normal cellular metabolism and in response to factors such as environmental changes and diet (Sagada et al., 2021). Hepatic antioxidant capacity (e.g., SOD, CAT, and POD) is a way of verifying the health condition and nutritional status of fish, which can regulate the balance between oxidants like ROS and antioxidants to avoid oxidative stress (Birben et al., 2012). Also, MDA is an important product of lipid peroxidation, and the level of MDA is a measure of the degree of oxidative damage (Silambarasan et al., 2019). In this study, diet supplemented with L. salivarius S01 significantly increased the antioxidant enzyme activity (SOD, CAT, and POD) of S. graham. Meanwhile, the significantly reduced malondialdehyde levels also confirmed the enhanced repair of oxidative damage by the probiotic-induced antioxidant enzyme activity. Thus, L. salivarius S01 was able to enter the intestine and enhance host immunity, which is necessary for probiotics to exert their beneficial effects. The gut is home to the densest microbial populations of organisms, and plays an important role in many physiological functions, such as host metabolism and nutrition (Devillard et al., 2007). Higher gut microbial diversity provides the host with a higher tolerance to pathogens (Harrison et al., 2019). In this study, the Shannon, Chao1, and Ace indices of the gut microbiome of S. graham were all found to increase after feeding a diet supplemented with L. salivarius, indicating that L. salivarius can promote the gut microbial diversity and richness of the host, consistent with the alterations caused by other LAB as dietary supplements on the host gut microbiota (Ljubobratovic et al., 2017; Lukic et al., 2020). Moreover, previous studies have reported that changes in gut microbiota may be a cause of changes in host immunity (Messaoudi et al., 2013; Chaves et al., 2017; Foysal and Gupta, 2022). In this study, it is important to analyze the changes of specific microbiota in the gut microbiota. At the phylum level, we found that the abundance of the most dominant phylum Proteobacteria decreased after feeding a diet supplemented with L. salivarius. Elevated proportions of proteobacteria in the gut of aquatic animals increase the risk of bacterial infections caused by Eriocheir sinensis and Litopenaeus vannamei (Ding et al., 2017; Wang et al., 2019). The relative abundance of Firmicutes, Actinobacteria, and Bacteroidetes increased to different degrees after feeding L. salivarius compared with the control group. Firmicutes can promote the decomposition of fiber and enable the host to obtain nutrition from fibrous feed (Brulc Jennifer et al., 2009). The main role of Bacteroidetes is to degrade carbohydrates (Spence et al., 2006), and the ratio of Bacteroidetes to Firmicutes in the animal’s gut reflects the organism’s ability to absorb nutrients. In addition, the main bacteria (including Bacillus, Lactobacillus, and Lactococcus) in the Firmicutes phylum can convert carbohydrates into lactic acid, creating acidic environments, thereby inhibiting pathogens, and protecting the gut (Messaoudi et al., 2013; Chaves et al., 2017; Foysal and Gupta, 2022). Actinobacteria can produce a variety of compounds, including antibiotics, enzymes, enzyme inhibitors, signalling molecules, and immunomodulators (Ul-Hassan and Wellington, 2009). As such, a higher abundance of Actinobacteria in the fish gut can improve food digestion and growth performance in S. grahami. Furthermore, the abundance of potential pathogens Aeromonas in the gut tract of the S. grahami was found to be reduced, while the abundance of the potential beneficial microorganisms Bifidobacterium increased, after feeding with L. salivarius S01 supplements. Aeromonas is a common zoonotic pathogen found in fish that can cause systemic sepsis and local infections (Pereira et al., 2022). On the contrary, Bifidobacterium is an important indicator of good health, with nutritional, anti-tumor, and anti-aging potential, and plays an important role in regulating the balance of gut microbiota and promoting normal gut development (Di Pierro et al., 2020). Unfortunately, corrected p-values show a certain probability of false positive gut microbial results. In addition, correlation analysis showed that Aeromonas was negatively correlated with antioxidant capacity in S. grahami, while Bifidobacterium was positively correlated with antioxidant capacity. Higher levels of hepatic antioxidant enzyme activity and changes in certain gut microorganisms suggested that L. salivarius S01 supplements may reduce the probability of disease by enhancing the immunity of S. grahami against the invasion of external harmful substances. Based on genome-wide data of L. salivarius S01 and the bioinformatic analysis of the gut microbiota of S. grahami, the mechanisms by which L. salivarius S01 promotes host health have been elucidated. These results indicate that L. salivarius can be used as a potential probiotic and an antimicrobial food additive to replace chemical drugs and antibiotics in aquaculture, promoting host health and fostering the development of greener aquaculture practices. However, this study still has some drawbacks, but there are also worthwhile points to be considered.
The results presented in this study demonstrated that the antibacterial activity of L. salivarius S01, previously isolated from the gut tract of S. grahami (Xin et al., 2022), may originate from the antibacterial substances produced by the bacteria, such as T3PKS, Enterolysin_A, and sakacin_G. Biological. Also, L. salivarius S01 showed a better antibiotic sensitivity. Function and genetics analysis related to potential probiotics showed that L. salivarius S01 could cope with the pressure of the natural environment, which may contribute to its colonization in the gut tract. Moreover, diet supplementation with L. salivarius S01 significantly altered the gut microbial diversity and hepatic antioxidant enzyme activities of S. grahami. Using L. salivarius S01 as a diet additive markedly reduced the abundance of potential pathogens and increased the abundance of potential beneficial microorganisms in the gut of S. grahami. Furthermore, supplementation was also found to increase the activity of antioxidant enzymes in the hepar and reduce the incidence of oxidative damage in the host. In summary, this study provides both a theoretical and experimental basis for the application of L. salivarius S01 as potential probiotics in aquaculture.
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ncbi.nlm.nih.gov/, CP097639, SRR19351072, SRR19351073, SRR19351074, SRR19351075, SRR19351076, and SRR19351077.
W-GX, X-DL, and Y-CL have carried out the culture and genomic analysis of the bacteria, they conducted the feeding of Sinocyclocheilus grahami and its physiological and biochemical determination, and they also wrote the paper. W-GX, Y-HJ, and M-YX contributed new reagents or analytical tools. Q-LZ, and FW reviewed and revised the manuscript. L-BL has designed the research, provided the reagents and analytical methods and written and revised the paper. All authors contributed to the article and approved the submitted version.
This study was supported by Yunnan Major Scientific and Technological Projects (grant no. 202202AG050008).
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2022.1014970/full#supplementary-material Click here for additional data file. Click here for additional data file. |
PMC9648152 | Tomonori Unno,Masaki Ichitani | Epigallocatechin-3-Gallate Decreases Plasma and Urinary Levels of p-Cresol by Modulating Gut Microbiota in Mice | 28-10-2022 | p-Cresol (PC), a gut bacterial product of tyrosine catabolism, is recognized as a uremic toxin that has negative biological effects. Lowering the plasma PC level by manipulating the gut bacterial composition represents a promising therapeutic strategy in chronic kidney disease. This study was conducted to reveal whether epigallocatechin-3-gallate (EGCG) decreases plasma PC levels by limiting its bacterial production in a mouse model. The PC concentration in the samples was measured by high-performance liquid chromatography (HPLC) after treatments with sulfatase and β-glucuronidase. The results showed that the addition of EGCG to the diet decreased the plasma and urinary concentrations of PC in a dose-dependent manner, with a statistically significant difference between the control group and the 0.2% EGCG group. However, once EGCG was enzymatically hydrolyzed to epigallocatechin (EGC) and gallic acid, such effects were lost almost completely. The addition of 0.2% EGCG in the diet was accompanied by a decreased abundance of Firmicutes at the phylum level and Clostridiales at the order level, which constitute a large part of PC produced from tyrosine. In conclusion, EGCG, not EGC, reduced plasma and urinary concentrations of PC in mice by suppressing its bacterial production with accompanying alteration of the relative abundance of PC producers. | Epigallocatechin-3-Gallate Decreases Plasma and Urinary Levels of p-Cresol by Modulating Gut Microbiota in Mice
p-Cresol (PC), a gut bacterial product of tyrosine catabolism, is recognized as a uremic toxin that has negative biological effects. Lowering the plasma PC level by manipulating the gut bacterial composition represents a promising therapeutic strategy in chronic kidney disease. This study was conducted to reveal whether epigallocatechin-3-gallate (EGCG) decreases plasma PC levels by limiting its bacterial production in a mouse model. The PC concentration in the samples was measured by high-performance liquid chromatography (HPLC) after treatments with sulfatase and β-glucuronidase. The results showed that the addition of EGCG to the diet decreased the plasma and urinary concentrations of PC in a dose-dependent manner, with a statistically significant difference between the control group and the 0.2% EGCG group. However, once EGCG was enzymatically hydrolyzed to epigallocatechin (EGC) and gallic acid, such effects were lost almost completely. The addition of 0.2% EGCG in the diet was accompanied by a decreased abundance of Firmicutes at the phylum level and Clostridiales at the order level, which constitute a large part of PC produced from tyrosine. In conclusion, EGCG, not EGC, reduced plasma and urinary concentrations of PC in mice by suppressing its bacterial production with accompanying alteration of the relative abundance of PC producers.
Uremic toxins are substances that interact negatively with biological functions. The accumulation of uremic toxins is associated with systemic disorders such as chronic kidney disease (CKD), endothelial dysfunction, insulin resistance, and cognitive impairment, leading to higher mortality and lower quality of life. Phenol and p-cresol (PC) are typical uremic toxins generated from tyrosine by intestinal bacteria. They are taken into the blood circulation and then undergo conjugations with sulfation or glucuronidation in the liver. p-Cresyl sulfate (PCS) is considered to be a potential cause of excess cardiovascular disease and mortality in patients with CKD. Mechanistically, it can trigger inflammation and oxidative stress in endothelial cells, contributing to endothelial dysfunction and arterial stiffness. The excretion of PCS mainly depends on the versatile tubular transporter systems in the kidney, and limited renal clearance in patients with CKD leads to a progressive accumulation in the blood. PCS is bound with a high affinity to plasma proteins and therefore is poorly removed by dialysis. Not only preserving renal excretory function but also inhibiting bacterial production of PC would also seem to be a useful way to lessen the disproportionate burden in patients with CKD. Some approaches to controlling PC production have been evaluated; for instance, the use of probiotics, prebiotics, and their combinations, is a topic of keen interest in lowering plasma levels of PCS. Human intervention studies have provided evidence that certain types of dietary fiber bring about beneficial effects in CKD patients. In general, treatments with prebiotics and probiotics aim to modulate gut microbiota based on the concept of increasing bacterial saccharolytic activity and limiting proteolytic activity in the large intestine. Strategies for decreasing the bacterial production of PC show great promise as alternative treatments to mitigate the complications of CKD. Recent studies also provide evidence that a certain type of polyphenols can modulate gut bacterial composition by bilaterally acting as an antimicrobial and stimulating growth. Green tea is a dietary source of polyphenols, mainly epicatechin, epigallocatechin (EGC), epicatechin-3-gallate, and epigallocatechin-3-gallate (EGCG). EGCG is the most abundant polyphenol in green tea. Parts of orally ingested EGCG can reach the large intestine, which is inhabited by a wide variety of bacteria. Green tea, or some of its catechins, stimulates or hinders the growth of specific gut bacterial species, and as a consequence of altered microbial composition, changes in the types and amount of microbially produced metabolites may occur. Therefore, we propose the hypothesis that EGCG influences the bacterial production of PC in the intestine by altering PC-producing bacteria. To test this hypothesis, the present work used animal experiments to evaluate the dose-dependent response (the first experiment, Exp. 1) and structure–activity (the second experiment, Exp. 2) of EGCG by measuring urinary and plasma concentrations of PC.
In both Exp. 1 and Exp. 2, there were no significant differences in food intake or dried feces weight in the 2-week feeding period (Table 1). There were also no significant differences in the final body weight, but the cecal digesta weight of the 0.2% EGCG-treated mice showed statistically significant increases.
The conjugated forms of phenol and PC were converted into free forms by enzymatic hydrolysis with sulfatase and β-glucuronidase, and the urinary and plasma levels of phenol and PC were measured as the total concentrations of their free and sulfate/glucuronide conjugation forms. In the control mice, the mean amount of PC excreted in the urine was 4 times higher than that of phenol (Figure 1A). The dietary addition of EGCG decreased urinary excretion of both compounds in a dose-dependent manner, with the levels in the urine of the mice fed the 0.2% EGCG diet being nearly undetectable. Statistical analysis demonstrated significant decreases in urinary phenol and PC in the 0.2% EGCG group compared to the control group (p < 0.05 for phenol and p < 0.01 for PC). Total plasma concentrations of PC were also higher than those of phenol (Figure 1B). The mean plasma concentration of PC was 9.3 ± 2.2 μM for the control group and 0.2 ± 0.1 μM for the 0.2% EGCG group, a statistically significant difference (p < 0.05). Plasma phenol concentration exhibited a declining trend in line with the addition of EGCG, but it did not reach statistical significance.
Phenol and PC in their free form were determined without an enzymatic hydrolysis reaction. The proportion of the free form to the total concentration of urinary phenol and PC constituted only a small percentage, with values of 1.7% for phenol and 3.7% for PC (Figure 1C). The amount of the glucuronide form was calculated by subtracting the free form from the data obtained after glucuronidase treatment. The amount of the sulfate form was also calculated by subtracting the free and glucuronide forms from the data obtained after glucuronidase plus sulfatase treatments. In the control mice, the sulfated form of phenol in the urine held a dominant share, accounting for 77% of the total concentration. Yet, at the same time, the sulfated form of PC accounted for 45%, showing a different trend a bit lower than the glucuronide form (51%). In mice fed the diets with EGCG at 0.05 and 0.1%, regardless of the additive amount of EGCG, the proportions of the sulfated or glucuronide forms to the total urinary concentration were almost consistent with the mice fed the control diet (Table 2). The proportions of the conjugated forms in the urine of mice fed the 0.2% diet did not follow such a pattern.
In Exp. 2, the dietary addition of EGCG at 0.2% decreased the total urinary concentration of PC (p < 0.001), but this effect was completely lost with the hydrolysis of EGCG (Figure 2A). Between EGCG and its hydrolysate (an equimolar mixture of EGC and gallic acid (GA)), there was a statistically significant difference in the total urinary concentration of PC (p < 0.0001). The case was somewhat different for phenol; urinary phenol concentration did not show significant differences among the groups. Such outcomes appeared consistently in the plasma concentrations. Total PC concentration in the plasma of mice fed the control diet, the EGCG diet, and the EGC + GA diet was 10.8 ± 5.2, 0.2 ± 0.1, and 12.4 ± 8.0 μM, respectively (Figure 2B). The EGCG group had significantly lower levels than other groups (p < 0.05 against the control group and p < 0.01 against the EGC + GA group). There were no significant differences in plasma phenol concentration among the groups.
The concentrations of phenol and PC in their free form per wet weight of cecal digesta are shown in Table 3. In both experiments, the cecal digesta of mice fed the 0.2% EGCG diet contained almost no PC, a statistically significant difference with the control group (p < 0.05 for Exp. 1 and p < 0.01 for Exp. 2). EGCG hydrolysate had little impact on the cecal level of PC. Figure 2C shows that the relationship between the cecal concentration of PC was closely correlated with its plasma concentration. The concentrations of PC both in the plasma and the cecal digesta in the EGCG group were at almost zero levels. In the case of phenol, a dose-dependent decrease was seen in Exp. 1, but dietary addition of EGCG did not lead to a statistically significant difference in Exp. 2.
In Exp. 2, the four main phyla (Firmicutes, Bacteroidetes, Actinobacteria, and Proteobacteria) represented 98.2% of the sequences in the feces of control mice (Figure 3A). In every mouse in the control group, the phylum Firmicutes was the most abundant. The addition of 0.2% EGCG in the diet provided a significant reduction in the relative abundance of Firmicutes compared to the control group (p < 0.05), whereas the EGC + GA diet had no effect (Figure 3B). The phyla Bacteroidetes and Verrucomicrobia were inversely increased (p < 0.05 for both), but in the case of EGCG hydrolysate, there was no practical impact. Principal component analysis (PCA) partially explained much of the variation (PC1, 38.4%, and PC2, 23.4%); the phylum taxonomic profiles of mice fed the control diet and the EGC + GA diet located on the near side, whereas the EGCG diet produced to a different profile (Figure 3C). At the order level, bacteria of 12 orders were detected in the feces of the control group, of which 96.3% were represented by bacteria belonging to Clostridiales, Lactobacillales, Erysipelotrichales, Bacteroidales, Bifidobacteriales, Eggerthellales, and Enterobacterales (Figure 4A). Comparisons among the experimental groups for the top five orders and Verrucomicrobiales in the total sequence are shown in Figure 4B. EGCG treatment caused a downward shift in the relative abundance of the order Clostridiales, a member of phylum Firmicutes, but order Bacteroidales, a member of phylum Bacteroidetes, shifted upward, with significant differences from the control group (p < 0.01 for Clostridiales, p < 0.05 for Bacteroidales). Orders Lactobacillales and Erysipelotrichales, which also belong to the phylum Firmicutes, did not show significant differences. Interestingly, the relative abundance of Verrucomicrobiales rose precipitously in the EGCG group.
In recent studies, polyphenol-rich dietary sources have received much attention for their impact on elevating or depressing the bacterial production of metabolites while also modifying the gut microbial composition. Once orally consumed, parts of polyphenols reach the large intestine, where they have direct contact with a vast variety of bacteria. As a consequence of the modification of the bacterial community by polyphenols, metabolite production could be up- or downregulated. Short-chain fatty acids are an example of bacterial metabolites. Green tea polyphenols suppress their production in the intestine, but black tea polyphenols conversely increase them. Our previous paper also reported that the diet addition of EGCG decreased the cecal PC level in rats, but whether EGCG has a significant impact on the plasma and urinary levels of phenol and PC via direct modulation of their producers in the gut remains unsettled. In addition, the impacts of tea polyphenols in a gallate form or a nongallate form on the relative changes of PC producers have not been systematically compared. Here, we evaluated the effects of EGCG and its hydrolysate (the mixture of EGC and GA) on urinary and plasma PC levels in healthy mice in relation to their modulation effect against PC producers. It is well recognized that phenol and PC in urine and plasma consist largely of conjugated forms. In this study, the conjugated forms of phenol and PC were hydrolyzed by sulfatase and β-glucuronidase to convert them back into their free form, and then, the compounds were purified by the solid-phase extraction (SPE) method. In Exp. 1, the concentrations of PC in urine and plasma were decreased in response to the amount of EGCG added to the diet. This result provides direct evidence that EGCG has a beneficial effect on the suppression of the bacterial production of PC. Especially given that the cecum is one of the major organs where gut bacteria produce this uremic toxin, the finding that the cecal concentration of PC significantly decreased in the 0.2% EGCG group offers conclusive evidence to support the theory. It is reasonable to say that EGCG decreases the plasma and urinary concentrations of PC by reducing their production in the intestine. Next, to evaluate the respective ratio of conjugation with glucuronide and sulfate, we calculated the urinary concentrations of p-cresyl glucuronide (PCG) by subtracting the free form from the data obtained after hydrolysis only with β-glucuronidase. The concentration of PCS was also calculated by subtracting the concentrations of the free form and PCG from the total concentrations. It has been reported that PCG and PCS are found almost equally in rodents. The same was true in this Exp. 1 of this study, which showed that the urine collected from the control mice had almost an equal percentage of glucuronate and sulfated forms. Since the urine of mice fed the 0.1% EGCG diet also maintained a balanced proportion similar to that in the control mice, it is reasonable to suggest that 0.1% EGCG exerted little influence on the conjugating reaction of PC in the liver. Green tea polyphenols are divided into two main classes: one is catechins having a galloyl moiety and the other is catechins not having a galloyl moiety. To reveal the structure–activity relationship between the galloyl type and the nongalloyl type, we next conducted another animal study (Exp. 2). Mice consumed either a diet containing EGCG at the 0.2% concentration or a diet containing the equimolar preparation of EGC and GA (prepared by the enzymatic hydrolysis of EGCG). The results demonstrated that the mice fed the 0.2% EGCG diet had markedly lower concentrations of PC in urine and plasma compared to the control diet, but the mice fed the EGC + GA diet excreted a significant amount of PC in the urine. This observation clearly shows that EGC was ineffective in reducing the bacterial production of PC, implying that the attachment of galloyl moiety to the structure of flavan-3-ol plays an important role. On the one hand, it is known that a part of EGCG entering the large intestine is hydrolyzed to EGC and GA by intestinal bacteria. With the hydrolysis of EGCG, it may become progressively less effective, but the rest of EGCG can play a role in exerting the intended effect. Tyrosine is microbially metabolized to 4-hydroxyphenylacetate and then converted into PC by 4-hydroxyphenylacetate decarboxylase (4-Hpd). In a recent study by Saito et al.,Blautia hydrogenotrophica, Clostridium difficile, Romboutsia lituseburensis, which are members of the order Clostridiales (heterotypic synonym of Eubacteriales, according to the NCBI Taxonomy Database), and Olsenella uli, a member of the order Coriobacteriales, were identified as major PC producers. These four PC-producing bacteria harbor a homolog of 4-Hpd. Amaretti et al. also found that the families Lachnospiraceae and Ruminococcaceae, which are members of the order Clostridiales, had relevance to the production of PC. In light of this knowledge, it seems reasonable to predict that bacteria belonging to the order Clostridiales have a major role in producing PC in the gut. To find out whether the ingestion of EGCG induces an effect on PC-producing bacteria, we also determined the bacterial compositional change based on the taxonomic category. The results showed that the addition of 0.2% EGCG in the diet brought about a significant decrease in the relative abundance of the phylum Firmicutes. Of the major members constituting the phylum Firmicutes, only the order Clostridiales showed a statistically significant decrease as a result of EGCG. The relative abundance of the orders Lactobacillales and Erysipelotrichales could not be influenced by EGCG. A potential explanation for this is that some of the ingested EGCG reached the large intestine and reduced the abundance of the order Clostridiales exclusively, consequently suppressing PC production. However, whether EGCG interferes with enzyme reactions via direct inhibition of bacterial 4-Hpd is not known yet. A vast variety of bacteria with phenol-producing ability seem to reside in the intestine commonly. Saito et al. also identified some types of bacteria belonging to the orders Clostridiales, Fusobacteriales, and Enterobacterales that are capable of effectively producing phenol from tyrosine. These are phylogenetically classified in the phylum Firmicutes, Fusobacteria, and Proteobacteria, respectively. As explained above, EGCG was able to reduce the relative abundance of the phylum Firmicutes, but it showed a reverse trend for Proteobacteria. EGCG might function more to increase the abundance of Proteobacteria than to decrease it. Such a diversified range of phenol producers led us to suppose that EGCG could not by itself meaningfully reduce urinary and plasma levels of phenol. There is a need for more detailed studies to investigate the role of EGCG against phenol-producing bacteria. Dietary supplementations with polyphenol-rich plant extracts may offer an opportunity to control the production of certain uremic toxins. For example, the consumption of a mixture of red wine and grape juice extracts for 4 days brought about a clinical advantage in reducing colonic protein fermentation or changing microbial amino acid metabolism, particularly a reduction of urinary PC. In another study, supplementation with cranberry dry extract (daily dose of 1000 mg) for 2 months did not reduce the plasma levels of PCS in non-dialysis CKD patients. The present study demonstrated a possible beneficial effect of green tea polyphenols on reducing bacterial production of PC in a mouse model. The dietary addition of EGCG had a strong reducing effect on urinary and plasma PC levels with a decreased abundance of PC producers in fecal microbiota. Based on the amount of food intake throughout the experimental period, the daily consumption of EGCG in the 0.2% EGCG group was calculated to be 305 mg/kg body weight of mice. This could be converted to the human equivalent dose at 24.7 mg/kg. If efficient ways were devised to help a greater amount of EGCG reach the large intestine, it should be possible to reduce the dosage of EGCG to some extent. Given that the microbial production of PC has been linked to a significant risk of cardiovascular mortality in CKD patients, EGCG may be a candidate agent for the treatment of the disease. In fact, EGCG has been studied for potential use in the management and prevention of various kidney diseases, with the major mechanisms of action associated with the reduction of oxidative stress and inflammation. The present study looked at the beneficial health effect from a different perspective, focusing on uremic toxin control. Further investigation is warranted to elucidate the clinical benefit to which an EGCG-microbiota interaction is attributed.
This study demonstrated that dietary addition of EGCG reduced the plasma level and urinary excretion of PC in mice. The addition of 0.2% EGCG in the diet was accompanied by decreased abundance of PC-producing bacteria in the feces. However, once EGCG was hydrolyzed to EGC and GA, such effects were lost almost completely. Thus, the intervention of EGCG, not EGC, is a promising strategy for the prevention of disorders derived from this uremic toxin.
Standards of phenol, PC, and p-chlorophenol were purchased from Fujifilm Wako Pure Chemical Co. (Osaka, Japan). Both β-glucuronidase from Helix pomatia and sulfatase from abalone entrails were purchased from Sigma-Aldrich Japan K.K. (Tokyo, Japan). A commercial EGCG product (>94% of purity) was obtained from DSM Nutrition Japan, K.K. (Tokyo, Japan). EGCG hydrolysate (a mixture of EGC and GA) was prepared by enzymatic hydrolysis of EGCG according to a previous procedure with a slight modification. One gram of enzyme preparation (tannase-KTFHR, Kikkoman Co., Chiba, Japan), which consisted of 0.9% (w/w) tannase from Aspergillus oryzae, 99.0% glucose, and 0.1% inositol, was added to an aqueous solution of EGCG (10 g/L) and incubated at 37 °C for 60 min. The reaction mixture was evaporated and freeze-dried. The disappearance of EGCG after tannase treatment was confirmed by high-performance liquid chromatography (HPLC) (Figure S1). Except for EGC and GA, newly generated peaks were undetectable.
In Exp. 1, a total of 12 male ICR mice (4 weeks old) were purchased from Tokyo Laboratory Animals Science Co., Ltd. (Tokyo, Japan) and acclimated for 3 days in stainless steel metabolic cages at 22 °C in a room with an automatically controlled 12 h lighting cycle. During the acclimation period, the mice were fed the AIN93G formulation diet. They were then divided into four groups (n = 3 per group) according to their body weight and fed respective diets: a control diet, a 0.05% (w/w) EGCG diet, a 0.1% EGCG diet, or a 0.2% EGCG diet (Table S1). They were given free access to their experimental diets and tap water for 2 weeks. Feces were collected throughout the experimental period, and urine was collected during the last 2 days of the experiment. The feces were freeze-dried and stored at −40 °C. The mice were humanely killed by inhalation of high levels of carbon dioxide, and blood was immediately collected from the abdominal vein. Plasma was obtained after centrifuging at 2000g for 10 min and stored at −40 °C in a plastic microtube. The cecum was excised, and the cecal digesta was also collected and stored at −40 °C until use. In Exp. 2, we again purchased a total of 12 male ICR mice from the same breeder. After an acclimation period of 3 days, the mice were divided into three groups (n = 4 per group) and fed respective diets, a control diet, an EGCG diet, or an EGC + GA diet (Table S1) for 2 weeks. Feces and urine were collected as in Exp. 1. Blood and cecal digesta were collected on the final day of the experiment. All experiments were approved by the Committee for the Use and Care of Experimental Animals of Tokyo Kasei Gakuin University (approval number 2-11).
Thawed urine was diluted 50-fold with distilled water. For the measurement of the total amounts of phenol and PC (the sum of free, sulfate, and glucuronide forms), 50 μL of the diluted urine sample was first reacted with 20 units of sulfatase solution in 0.1 mL of 0.1 M acetate buffer (pH 5.0) at 37 °C for 2 h and second with 100 units of β-glucuronidase solution in 0.9 mL of 0.1 M phosphate buffer (pH 6.8) containing 5 mM sodium chloride at 37 °C for 15 min. For the glucuronide forms, the diluted urine sample (50 μL) was reacted only with 100 units of β-glucuronidase solution in 0.9 mL of the same phosphate buffer at 37 °C for 15 min. For the free form, the diluted urine sample proceeded without enzyme treatments. After adding 10 μL of 0.5 mM p-chlorophenol solution as an internal standard, the reaction mixture was directly applied to a polymer-based SPE cartridge (Strata-X, particle size of 33 μm, Phenomenex, Inc., CA), which had been preconditioned with 1 mL of water, 1 mL of methanol, and 1 mL of water again. The cartridge was washed with 1 mL of water and 1 mL of 20% (v/v) aqueous methanol and then successively eluted phenol and PC with 1 mL of methanol. After passing through a 0.45 μm filtration membrane, 10 μL of the resulting filtrate was injected into an HPLC system (Shimadzu Co., Kyoto, Japan) that consisted of dual pumps (model LC-20 AD), an autosampler (model SIL-10A), a column oven (model CTO-20A), a fluorescence detector (model RF-20A), and a system controller (model SCL-10A). An analytical column (Unison UK-3C18, 100 mm × 4.6 mm i.d., Imtakt, Co., Kyoto, Japan) was used for separation. The fluorescence detector was set at wavelengths of 270 nm for emission and 305 nm for excitation. Gradient elution was performed by varying the proportion of solvent A (methanol–water, 25:75 v/v) to solvent B (methanol), at a flow rate of 1 mL/min. The mobile phase composition started at 100% solvent A (0% solvent B), after which the ratio of solvent B was linearly increased to 50% over 15 min, followed by a further increase of solvent B to 70% over 2 min. The composition was then brought back to the initial conditions over 2 min for the next run. The measured values were normalized to the urinary creatinine concentration. Urinary creatinine concentrations were measured by the Jaffé assay.
Fifty microliters of thawed plasma was sequentially treated with 20 units of sulfatase in 0.1 mL of 0.1 M acetate buffer (pH 5.0) at 37 °C for 2 h and then with 100 units of β-glucuronidase in 0.1 M phosphate buffer (pH 6.8) containing 5 mM sodium chloride at 37 °C for 15 min. The subsequent procedure for SPE and HPLC measurements was the same as for urine.
To prepare the homogenate, an aliquot of thawed cecal digesta was added to four volumes of distilled water. The homogenate (0.1 mL) was mixed with 0.4 mL of methanol and then centrifuged at 2000g and 4 °C for 10 min. The supernatant (0.4 mL) was diluted with 3.6 mL of water, and the SPE and HPLC analysis proceeded according to the method described above.
DNA was extracted from feces and PCR was performed according to the paper by Takahashi et al. The V3–V4 regions of the 16S rRNA gene were amplified by PCR with universal primers 341F (5′-CCTACGGGAGGCAGCAG-3′) and 805R (5′-GGACTACCAGGGTATCTAAT-3′). Sequencing was conducted using a paired-end modified to a 600 bp cycle run on a MiSeq sequencing system with a MiSeq Reagent Kit version 3 (Illumina, Inc., San Diego, CA). The paired-end reads for each sample were joined using Fastq-join and then processed with quality filtering with the FASTX-Toolkit. The quality of the sequences was checked, and the passed sequences were clustered into operational taxonomic units (OTUs) with 97% pairwise identity. Taxonomic annotation of the representative sequence was performed using the Microbial Identification Database NGS-DB-BA 16.0 (TechnoSuruga Laboratory, Shizuoka, Japan).
p values less than 0.05 were considered significant. In Exp. 1, the statistical analysis was performed by Dunnett’s test to compare the difference with the control group. If the data were not normalized, Dunn’s test was adopted. In Exp. 2, statistics were calculated using one-way ANOVA, followed by the Tukey test. For the comparison of microbiota abundance, a nonparametric approach with Dunn’s test involving pairwise comparisons was employed. All statistical analyses were conducted using GraphPad Prism version 9.03 (GraphPad Software, San Diego, CA). |
||
PMC9648179 | Fabrizio Martora,Teresa Battista,Claudio Marasca,Lucia Genco,Gabriella Fabbrocini,Luca Potestio | Cutaneous Reactions Following COVID-19 Vaccination: A Review of the Current Literature Martora et al | 06-11-2022 | cutaneous reactions,COVID-19 vaccinations,side effects | Abstract The outbreak of coronavirus disease 2019 (COVID-19) represented a new worldwide challenge, strongly impacting on the global economy, overall health and lifestyle. Since then, several strategies have been adopted to contain the widespread of infection. Among these, vaccination is currently the most important measure to fight against the pandemic. However, several concerns such as slower-than-hoped-for rollout, the hurried approval with limited data, the mechanism of action (in particular mRNA-based), and the uncertain duration of protection they afforded were initially raised. Moreover, even if cutaneous reactions have been rarely reported in clinical trials, global mass vaccination showed several dermatologic reactions not initially recognized, leaving dermatologists to decide how to diagnose and treat them. In this scenario, dermatologists should be ready to promptly recognize these clinical manifestations. Thus, the aim of this manuscript is to review current literature on cutaneous reactions following COVID-19 vaccination, particularly inflammatory dermatological diseases, in order to help clinicians to better understand these dermatological conditions and to provide an extensive overview of all the vaccine-related skin manifestations. | Cutaneous Reactions Following COVID-19 Vaccination: A Review of the Current Literature Martora et al
The outbreak of coronavirus disease 2019 (COVID-19) represented a new worldwide challenge, strongly impacting on the global economy, overall health and lifestyle. Since then, several strategies have been adopted to contain the widespread of infection. Among these, vaccination is currently the most important measure to fight against the pandemic. However, several concerns such as slower-than-hoped-for rollout, the hurried approval with limited data, the mechanism of action (in particular mRNA-based), and the uncertain duration of protection they afforded were initially raised. Moreover, even if cutaneous reactions have been rarely reported in clinical trials, global mass vaccination showed several dermatologic reactions not initially recognized, leaving dermatologists to decide how to diagnose and treat them. In this scenario, dermatologists should be ready to promptly recognize these clinical manifestations. Thus, the aim of this manuscript is to review current literature on cutaneous reactions following COVID-19 vaccination, particularly inflammatory dermatological diseases, in order to help clinicians to better understand these dermatological conditions and to provide an extensive overview of all the vaccine-related skin manifestations.
The outbreak of coronavirus disease 2019 (COVID-19) represented a new worldwide challenge, strongly impacting on the global economy, overall health and lifestyle.1 Since then, several strategies have been adopted to contain the widespread of infection.2,3 Dermatologists played a key role during the pandemic, fighting against several challenges such as cutaneous reactions caused by COVID-19 disease, the hesitancy on the efficacy and safety of conventional treatment and biologic drugs in this period, the worsening of several dermatosis due to the wearing of personal protection equipment and the introduction of a new lifestyle.4–6 Indeed, the “stay-at-home” policy and the restrictive measures adopted by the Italian Government during the COVID-19 pandemic period strongly affected the quality of life.7 In addition, COVID-19 restriction measures affect the epidemiology of infectious diseases and skin cancers.5,8–10 Among the developed public health strategies to control the spread of COVID-19, vaccination is currently the most important measure to fight against the pandemic. However, several concerns such as slower-than-hoped-for rollout, the hurried approval with limited data, the mechanism of action (in particular mRNA-based), and the uncertain duration of protection they afforded were initially raised.11,12 Fortunately, worldwide vaccination campaign was a success, showing to be the most effective weapon to prevent and control COVID-19 epidemic, disease progression, hospitalization and mortality.13 According to the WHO COVID-19 dashboard accessed on 11 September 2021, more than 608 million confirmed cased of COVID-19 have been reported, with almost 6.51 million deaths.14 Nowadays, licensed vaccines for COVID-19, use nucleic acid-based vaccination platforms, such as viral vector platforms, messenger ribonucleic acid and inactivated virus.13 Four vaccines have been approved by the European Medicines Agency (EMA): 2 mRNA-based vaccines (Pfizer/BioNTech; BNT162b2 and Moderna; mRNA-1273) and 2 viral-vector-based vaccines (AstraZeneca; AZD1222 and Johnson & Johnson; Ad26.COV2.S).15 However, other vaccines have been approved in other countries such as “CoronaVac” (Sinovac), “Sputnik V” (Gamaleya Research Institute), and “Convidecia” (CanSino Biologics).13 Currently, more than 5.3 billion people have received at least one dose of COVID-19 vaccine.16 Similar to other drugs, some people reported mild-to-moderate adverse events following vaccination, including fatigue, headache, diarrhea, redness or pain at the injection site, fever, muscle aches, chills.17–19 Fortunately, most of the side effects are limited, with a duration of few days.17–19 Even if cutaneous reactions have been rarely reported in clinical trials, global mass vaccination showed several dermatologic reactions not initially recognized, leaving dermatologists to decide how to recognize and treat them. In particular, a wide spectrum of cutaneous reactions has been reported.20 However, the significance of these reactions is still unknown. In this scenario, dermatologists should be ready to promptly recognize these clinical manifestations, which should be considered in personalized medicine.21,22 Thus, the aim of this manuscript is to review current literature on cutaneous reactions following COVID-19 vaccination, particularly inflammatory dermatological diseases, in order to help clinicians to better understand these dermatological conditions and to provide an extensive overview of all the vaccine-related skin manifestations.
For the current review, literature research was carried out on the PubMed, Embase, Cochrane Skin, Google Scholar, EBSCO and MEDLINE databases (until September 11, 2022). Research was performed by using the following keywords: “COVID-19”, “vaccination”, “vaccine”, “cutaneous”, “side effects”, “adverse events”, “skin manifestations”, “mRNA”, “viral-vector”, “Pfizer/BioNTech”, “BNT162b2”, “Moderna”, “mRNA-1273”, “AstraZeneca”, “AZD1222”, “Johnson & Johnson”, “Ad26.COV2.S”, “atopic dermatitis”, “psoriasis”, “lichen planus”, “bullous disease”, “pemphigus”, “pemphigoids”, “hidradenitis suppurativa”, “urticaria”, “rash”, “herpes”, “pityriasis rosea”, “chilblains”, “vitiligo”, “erythematous eruption”, “alopecia”, “local-injection”, “angioedema”, “eczema”. Analyzed articles included meta-analyses, reviews, letter to editor, real-life studies, case series and reports. The most relevant manuscripts were considered. Studies were selected if they provided information on cutaneous reactions following COVID-19 vaccination with BNT162b2, mRNA-1273, AZD1222 and Ad26.COV2.S, both first and second doses, if applicable. Cutaneous reactions following other vaccines, or the booster dose were excluded. Articles regarding skin reactions reported in clinical trials or with a limited number of cases were excluded. Manuscripts reporting local injection site reactions, both immediate and delayed, rash or unspecified cutaneous eruption and delayed inflammatory reactions to dermal hyaluronic acid filler were not considered. Moreover, articles where the vaccine leading to cutaneous reaction was not specified were excluded. Thus, the research was refined by reviewing the texts and the abstracts of collected articles. The bibliography was also reviewed to include articles that could have been missed. Only English language manuscripts were considered. This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors. Details of the included studies are reported in Table 1.
A total of 1922 reports were initially found searching literature. Subsequently, 523 articles and 71 manuscripts were excluded since they were duplicates and in non-English languages, respectively. Then, literature review was refined following inclusion and exclusion criteria. Finally, a total of 183 articles involving 456 patients were selected in the current review. Main findings are summarized in Table 1. Several cases of new onset or exacerbation of inflammatory skin diseases have been reported (Figure 1) as well as the type of vaccine causing these reactions has been investigated (Figure 2). As regards psoriasis, a total of 98 reports on psoriasis following COVID-19 vaccination were reported.23–54 In particular, flare of pre-existing disease and new-onset disease were reported in 81 and 17 cases, respectively. Moreover, several phenotypes of psoriasis were reported, with plaque subtype as the most frequent. Of note, even if biological treatments showed excellent results in terms of effectiveness and safety in psoriasis management,55–58 they seem to reduce the possibility of disease worsening following vaccination, without nullifying the risk. Moreover, the effectiveness of COVID-19 vaccines in patients undergoing treatment with biologics is debated.59,60 Lichen planus is a chronic, inflammatory, autoimmune disease with an unknown pathogenesis.61 To date, 13 cases of new-onset cutaneous lichen planus and 3 cases of cutaneous lichen planus exacerbation have been reported.62–76 Like psoriasis, also cases of new onset and flare of atopic dermatitis or eczema have been reported (7 and 14, respectively).77–83 However, there is not a clear correlation with clinical phenotypes.84 Moreover, undergoing treatment with dupilumab does not seem to prevent the possibility of a flare of the disease, even if its efficacy and safety have been largely demonstrated.85,86 No data of atopic dermatitis worsening in patients undergoing treatment with janus kinase inhibitors are available.87,88 Concerning hidradenitis suppurativa, there are currently few cases of new-onset disease (n = 1)89 or disease exacerbation (n = 5).90 However, patients with hidradenitis suppurativa are not at higher risk for any COVID-19 vaccine‐related adverse outcomes.91,92 Urticarial rashes are the second most common cutaneous reaction following COVID-19 vaccination reported, following local injection site reactions, such as “Covid-arm”.93 Globally, 98 cases of urticarious eruptions following COVID-19 vaccination have been collected in our review,77,78,83,94–107 also during treatment with omalizumab.108 Alopecia areata has been reported following COVID-19 vaccination.109–119 The largest study on 77 patients developing alopecia areata (39) or a worsening of the disease (38) has been reported by Nguyen et al. Unfortunately, it is not possible to correlate alopecia areata development and the type of vaccine.120 Regarding bullous disorders, a total of 26 cased of pemphigus vulgaris have been reported following COVID-19 vaccination,121–135 with several implications in treatment and management.136 Moreover, 40 cases of pemphigoids have been described.137–151 Regarding other cutaneous diseases developed following COVID-19 vaccination, 9, 40, 55, 12 and 11 cases of morphea,152–157 pityriasis rosea,158–177 herpes zoster,178–187 chilblains,188–197 and vitligo198–208 have been reported. Finally, several other dermatoses have been described, even if data are limited. Among these, we want to highlight pityriasis rubra pilaris, leukocytoclastic vasculitis, morbilliform rash, livedo racemosa, fixed drug eruption, erythema annulare centrifugum, granuloma annulare, fascial neutrophilic eruption, annular rash, Henoch-Schonlein purpura, dermatomyositis, regression of viral wart, raynaud phenomenon, eruptive angiomatosis, lichen striatus, pityriasis lichenoides et varioliformis acuta, Rowell’s syndrome, acrocyanosis, … suggesting that a wide type of dermatoses may be triggered by COVID-19 vaccination.209–221 However, most of these are limited to 1 or 2 case reports.
COVID-19 pandemic revolutionized daily clinical practice. Indeed, several strategies were adopted to contain the spreading of the infection.222 Dermatologists had to change their clinical routine in order to avoid the reduction in detection and treatment of several conditions, particularly skin cancer.223–227 Among these, teledermatology allowed physicians to continuously assist patients’ dermatologic conditions with excellent results in terms of treatment adherence and clinical outcomes.228–230 Vaccination campaign is the most important strategy showing excellent results in terms of safety and effectiveness.231,232 Indeed, it allowed to reduce the severity and the impact of COVID-19 pandemic. However, several skin diseases induced or exacerbated by COVID-19 vaccination have been reported. Fortunately, most of them were mild and self-limited, not requiring medical attention. In our review, we highlighted several cutaneous reactions following COVID-19 vaccination such as psoriasis, atopic dermatitis, bullous disease, etc. Even if not specifically investigated, local injection-site reaction was the commonest cutaneous vaccine-related adverse event reported. Of note, cutaneous reactions were reported following vaccination with both mRNA and viral vector-based vaccines, suggesting that the pathogenetic mechanism underlying the cutaneous reaction is not directly related to the vaccine mechanism of action itself.99 Certainly, further studies are needed to understand pathogenetic mechanisms linking cutaneous reaction and COVID-19 vaccination in order to identify “at-risk” subjects and to adopt preventive measures. Of note, among the articles reviewed in our work, the diagnosis of cutaneous reactions was confirmed by histopathological examination in most of the cases. However, a shared immune process was not found assessing the histological reports. Overall, mRNA vaccines, particularly BNT162b2, seem to be most commonly associated with cutaneous reactions. However, mRNA vaccines were previously authorized, produced and administered worldwide. Thus, the different number of administered vaccines may explain the difference between the number of skin reactions following mRNA or viral vector-based vaccines. Further epidemiological studies will clarify if the percentage of cutaneous reactions following vaccination is significantly higher in one of the two types of vaccines, with clinical implications. To sum up, our review analyzed several dermatoses exacerbated or developed following COVID-19 vaccination. However, the temporal association between the administration of the vaccine and the development of skin reaction may be casual. As regards the dose of vaccination, cutaneous reactions were reported following both the first and the second dose of vaccine. Furthermore, skin reactions following both the doses in the same patient have been reported as well. In our opinion, clinicians should be prepared also to cutaneous reaction following the booster dose.233
Main strengths of our review are the systematic method during the literature research and the high number of analyzed article and cutaneous reactions analyzed. Main limitations should be discussed. First, only the four vaccines approved by EMA have been considered. Moreover, several articles reporting registry-based studies did not allow the direct correlation between type of vaccine and cutaneous reaction. Finally, dermatological conditions developed following COVID-19 vaccination are usually mild and patients do not seek for medical attention.
With the worldwide advance of vaccination programs, several cutaneous reactions have been reported. Fortunately, the percentage of these adverse events is extremely low if compared with the number of vaccines administered. In our opinion, other cutaneous reactions following COVID-19 vaccination will be reported. Moreover, the pathogenetic mechanisms linking vaccination and skin reactions should be clarified. Clinicians should keep in mind the possibility of the exacerbation of the new onset of several dermatoses following vaccination in order to promptly recognize and differentiate vaccine-induced cutaneous manifestations from other clinical entities. Certainly, vaccination should not be discouraged. |
|
PMC9648186 | Huan Xiong,Jiaqi Wang,Zewen Chang,Hanqing Hu,Ziming Yuan,Yihao Zhu,Zhiqiao Hu,Chunlin Wang,Yunxiao Liu,Yang Wang,Guiyu Wang,Qingchao Tang | Gut microbiota display alternative profiles in patients with early-onset colorectal cancer | 27-10-2022 | gut microbiota,colorectal cancer,early onset,16S rRNA,functional annotation | Background The incidence of early-onset colorectal cancer (EOCRC) is increasing worldwide. This study aimed to explore whether there is an alternative gut microbiota profile in patients with early-onset colorectal cancer. Methods A total of 24 patients with EOCRC, 43 patients with late-onset colorectal cancer and 31 young volunteers were included in this study. The diversity of their fecal bacteria was explored using 16S ribosomal RNA gene sequencing. Cluster of ortholog genes (COG) functional annotation and Kyoto encyclopedia of genes and genomes (KEGG) were used to detect enrichment pathways among the three groups. Results Community separations were observed among the three groups. The Shannon index of the EOCRC group was significantly lower than the LOCRC group (P=0.007) and the NC group (P=0.008). Both PCoA analysis (Principal co-ordinates analysis, P=0.001) and NMDS (non-metric multidimensional scaling, stress=0.167, P=0.001) analysis indicated significant difference in beta diversity among the three groups. Fusobacteria, Bacteroidetes, and Clostridia were the most abundant bacteria in the EOCRC group, LOCRC group, and NC group, respectively. The results of COG showed that transcription (P=0.01398), defense mechanisms (P=0.04304), inorganic ion transport and metabolism (P=0.00225) and cell wall/membrane/envelope biogenesis (P=0.02534) were differentially expressed among the three groups. The KEGG modules involved in membrane transport (P=0.00856) and porphyrin and chlorophyll metabolism (P=0.04909) were differentially expressed among the three groups. Conclusion Early-onset colorectal cancer patients have a different gastrointestinal microbiota derangement compared to late-onset colorectal cancer patients. This dysbiosis can be reflected in the species diversity of the microbiota, the abundance of bacteria, and the abnormal functional predictions. | Gut microbiota display alternative profiles in patients with early-onset colorectal cancer
The incidence of early-onset colorectal cancer (EOCRC) is increasing worldwide. This study aimed to explore whether there is an alternative gut microbiota profile in patients with early-onset colorectal cancer.
A total of 24 patients with EOCRC, 43 patients with late-onset colorectal cancer and 31 young volunteers were included in this study. The diversity of their fecal bacteria was explored using 16S ribosomal RNA gene sequencing. Cluster of ortholog genes (COG) functional annotation and Kyoto encyclopedia of genes and genomes (KEGG) were used to detect enrichment pathways among the three groups.
Community separations were observed among the three groups. The Shannon index of the EOCRC group was significantly lower than the LOCRC group (P=0.007) and the NC group (P=0.008). Both PCoA analysis (Principal co-ordinates analysis, P=0.001) and NMDS (non-metric multidimensional scaling, stress=0.167, P=0.001) analysis indicated significant difference in beta diversity among the three groups. Fusobacteria, Bacteroidetes, and Clostridia were the most abundant bacteria in the EOCRC group, LOCRC group, and NC group, respectively. The results of COG showed that transcription (P=0.01398), defense mechanisms (P=0.04304), inorganic ion transport and metabolism (P=0.00225) and cell wall/membrane/envelope biogenesis (P=0.02534) were differentially expressed among the three groups. The KEGG modules involved in membrane transport (P=0.00856) and porphyrin and chlorophyll metabolism (P=0.04909) were differentially expressed among the three groups.
Early-onset colorectal cancer patients have a different gastrointestinal microbiota derangement compared to late-onset colorectal cancer patients. This dysbiosis can be reflected in the species diversity of the microbiota, the abundance of bacteria, and the abnormal functional predictions.
Colorectal cancer is the third most common cancer in terms of incidence and second in terms of cancer-related mortality worldwide (Sung et al., 2021). Approximately ten percent of all patients initially diagnosed with colorectal cancer are younger than 50 years of age (Collaborative et al., 2021). Early-onset colorectal cancer (EOCRC) is generally defined as colorectal cancer diagnosed before the age of 50 years (Patel et al., 2022). The incidence of late-onset colorectal cancer has declined due to preventive screening recommendations over the past 10 years (Araghi et al., 2019; Siegel et al., 2020; Sinicrope, 2022). However, the incidence and cancer-related mortality of EOCRC have increased significantly and will continue to show an increasing trend over in next 10 years (Bailey et al., 2015; Araghi et al., 2019; Collaborative et al., 2021). EOCRC always displays adverse clinical and histopathological features, yet the causes are unclear (Chang et al., 2012; Kneuertz et al., 2015; Saraste et al., 2020). In addition to the inherent genetic factors such as family history and germline gene mutations, poor dietary habits, smoking, alcohol, and antibiotics were considered risk factors for EOCRC (Chang et al., 2021; Patel et al., 2022). These risk factors can interact with the gut microbiota (Song and Chan, 2019), and their effects on the host can all be directly reflected by changes in the structure and abundance of the gut microbiota. The gut microbiota, as an ecosystem in direct contact with the gut mucosa, is the potential cause of colorectal cancer (Garrett, 2019). Alterations in the structure of the intestinal microbiota can contribute to the development and progression of intestinal diseases. Increased abundance of certain specific microorganisms (Fusobacterium nucleatum, Prevotella intermedia, Bacteroides fragilis, Porphyromonas asaccharolytica, etc.) can increase the risk of colorectal carcinogenesis through inflammatory responses, evasion of tumor immune responses, and activation of pre-tumor signaling pathways (e.g., β-catenin) (Hernandez-Luna et al., 2019; Wong and Yu, 2019). However, probiotics such as Lactobacillus and Streptococcus thermophilus were significantly less abundant in the gut of colorectal cancer patients (Li et al., 2021). Most of the current data used to explore the microbiota structure of patients with colorectal cancer are derived from late-onset colorectal cancer (Murphy et al., 2019), with few studies characterizing the gut microbiota in early-onset colorectal cancer. In this study, we propose to use high-throughput DNA sequencing technology to analyze the gut microbiota of early onset colorectal cancer patients from our center and to conduct a preliminary.
The fecal specimens of all patients in this study were obtained from the Department of Colorectal Surgery, Second Affiliated Hospital of Harbin Medical University from July 2018 to June 2020. The inclusion criteria for this study were: 1) Patients with colorectal cancer diagnosed with histopathology, and healthy young volunteers without tumors by gastroscopy; 2) Consent for us to collect their feces. The exclusion criteria were: 1) Had taken antibiotics, probiotics, corticosteroids or received fecal microbiota transplantation treatment within 3 months prior to sample collection; 2) Had a familial history of colorectal cancer; 3) Had used evacuant or undergone colonoscopy within 1 week prior to sample collection; 4) Had undergone abdominal surgery or other invasive treatment within 3 months prior to sample collection; 5) Had been diagnosed with multiple primary cancers; 6) Had a history of other cancer or inflammatory bowel disease; 7) Contamination of specimens as a result of failure to collect according to prescribed protocols (Di Segni et al., 2018); 8) Incomplete clinical information. The recruited sporadic CRC patients were divided into two groups based on age: the EOCRC group, aged < 50 years; LOCRC group, aged ≥ 55 years. All recruited young healthy volunteers were less than 50 years of age and they were included in the NC group. Clinical and pathological characteristics of CRC patients including age, gender, body mass index (BMI), history of drinking, tumor location, histological classification of tumors, and TNM stage were collected. The collected information of healthy volunteers included age, gender, BMI, and history of drinking. The stools were rapidly frozen in liquid nitrogen for 30 seconds after acquisition and stored at -80°C until DNA was extracted.
Microbial DNA was extracted from fecal samples using the E.Z.N.A. @ Soil DNA Kit (Omega Bio-tek, Norcross, GA, U.S.) according to the manufacturer’s protocol. The specific steps were performed according to the instructions. Final DNA concentration and purification were determined by NanoDrop 2000 UVVisspetrophotometer (Thermo Scientific, Wilmington, USA), and DNA quality was checked by 1% agarose gel electrophoresis. The extracted DNA was stored in a refrigerator at -80°C. The V3-V4 hypervariable regions (the 338F ~ 806R regions) of the bacterial 16S rRNA gene were amplified by high-throughput sequencing on a thermal cycler PCR system (GeneAmp 9700, ABI, USA) with primer sequences: 338F: 5’-ACTCCTACGGGAGGCAGCAG-3’, 806R: 5 ‘-GGACTACHVGGGTWTCTAAT-3’. The amplified DNA was further purified using the AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, Union City, CA, USA) and quantified using QuantiFluor™-ST (Promega, USA) according to manufacturer’s established guidelines. Then, the normalized equimolar concentrations of each amplicon were pooled and sequenced on the Illumina MiSeq platform (Illumina, San Diego, USA) using 2 × 300 bp chemistry according to the standard protocol from Majorbio bio Pharm Technology Co. (Shanghai, China).
The raw fastq files were filtering and trimming using Trimmomatic and merged by FLASH with the following criteria: (i) The reads were truncated at any site receiving an average quality score <20 over a 50 bp sliding window. (ii) Sequences whose overlap being longer than 10 bp were merged according to their overlap with mismatch no more than 2 bp. (iii)Sequences of each sample were separated according to barcodes (exactly matching) and Primers (allowing 2 nucleotide mismatching), and reads containing ambiguous bases were removed. Operational taxonomic units (OTUs) were calculated via clustering by average neighbor principle at 97% genetic similarity using UPARSE (version 7.1 http://drive5.com/uparse/). The chimeric sequences were identified and deleted after the comparison of the identified taxa. The classification of each 16S rRNA gene sequence was analyzed against the Silva (SSU123) 16S rRNA database using the RDP classifier algorithm (http://rdp.cme.msu.edu/) with a 70% confidence level (threshold).
Alpha diversity between the three groups was compared using Shannon index, Simpson index and the Simpson index. Beta diversity comparison between the three groups was done by PCoA analysis (Principal co-ordinates analysis), NMDS (Non-metric multidimensional scale) analysis and PLS-DA analysis (partial least squares discriminant analysis). PCoA analysis and NMDS analysis were performed using the unweighted UniFrac distance algorithm and weighted UniFrac distance algorithm, and adonis analysis (permutational MANOVA) was used for otherness test. Then, based on the obtained community abundance data, a hypothesis test was performed using rigorous statistical methods to assess the significance level of species abundance differences between the microbial communities of the three groups of samples, and to obtain significantly different species between groups. LEfSe (linear discriminant analysis coupled with effect size analysis) performed linear discriminant analysis (LDA) on samples according to different grouping conditions based on taxonomic composition to find out the significantly different influences on the sample delineation of groups or species that had a significant differential impact on the sample delineation. The OTU abundance table was normalized by PICRUSt1. The effect of the number of copies of the 16S marker gene in the species genome was removed; then the COG corresponding to the OTU was obtained by the greengene corresponding to each OTU family information and KEGG Ortholog (KO) information for each OTU; and calculate the abundance of each COG and KO abundance. According to the information of COG database, the descriptive information of each COG and its functional information can be parsed from the eggNOG database to obtain the potential functional abundance spectrum; according to the information of KEGG database, the KO Pathway can be obtained, and the abundance of each potential functional category can be calculated according to the OTU abundance.
The software mothur (version_1.30.2) was used for Alpha diversity analysis. Principal component analysis and principal co-ordinates analysis were statistically analysed and plotted using R (version 3.3.1). In NMDS analysis, Quantitative Insight Into Microbial Ecology 1 (QIIME, version_1.9.1) was applied to calculate the distance matrix of beta diversity, and then the R packages “vegan” and “mixOmics” were used for analysis and mapping. LEfSe (http://huttenhower.sph.harvard.edu/galaxy/roottool_id=lefse_upload) was used for multilevel species difference discriminant analysis; PICRUSt (version_1.1.0) software was used for functional prediction. All statistical calculations were performed in R 3.3.1. The Kruskal-Wallia H test was used to compare the differences in the measurement data between the three groups, and the Mann Whitney U test was used to compare the differences between two pairs. P-value < 0.05 was considered to be statistically significant, and the correction of the P-value is responsible for the false discovery rate (FDR).
A total of 24 EOCRC patients, 43 LOCRC patients and 31 healthy volunteers were recruited in this study. Their demographic characteristics are shown in Table 1 . We collected 98 samples and obtained a total of 5,362,431 sequence fragments with a total length of 2,261,064,976 bps. The length of all samples ranged from 204 to 528 bp, with an average of 422 bp.
We performed OTU clustering on all valid sequences, and selected OTUs with the number of sequences greater than or equal to 5 in at least three samples and the sum of sequence numbers greater than or equal to 20, and finally obtained 714 OTUs, and the rank abundance curves are shown in Figure S1A . The Shannon curves of all samples can rapidly reach the plateau, indicating that the sequencing depth met the requirements. ( Figure S1B ) We performed alpha diversity analysis on the three groups and found that the Shannon diversity index of the EOCRC group was significantly lower than that of the LOCRC group (P=0.007) as well as that of the NC group (P=0.008). ( Figure 1A ) And the Simpson index of the EOCRC group was significantly lower than that of the LOCRC group (P=0.013) and that of NC group (P=0.011). ( Figure 1B ) The Venn diagram showed that at the genus level, the number of bacterial genera was higher in LOCRC group than EOCRC and NC groups, and the three groups shared 247 bacterial genera, with only 16 unique genera in EOCRC group. ( Figure 1C )
We analyzed the difference of beta diversity among the three groups by PCoA, NMDS and PLS-DA. PCoA based on unweighted unifrac distance showed significant differences on the OTU level among the three groups (R²=0.0695, P=0.001), and adonis analysis showed significant differences between the EOCRC and LOCRC groups (P=0.0003) and between the EOCRC and NC groups (P=0.0002). ( Figure 2A ) PCoA based on weighted unifrac distances also showed significant differences among the three groups on OTU the level (R²= 0.0726, P=0.001). ( Figure S2A ) The results of the NMDS analysis on the OTU level were measured by the NMDS intensity index based on unweighted unifrac distance (stress=0.167, P=0.001, Figure 2B ). The corresponding values based on weighted unifrac distance were as follows: OTU level (stress=0.136, P=0.001), genus level (stress=0.140, P=0.001) and phylum level (stress=0.073, P=0.001), as shown in Figures S2B-D . PLS-DA showed a clear separation of the three groups on the OTU level ( Figure 2C ). These data indicated that EOCRC harbored a peculiar microbiota.
We performed LEfSe to investigate the composition of fecal microbiota in the three groups and identify taxa that were differentially abundant in the EOCRC (linear discriminant analysis (LDA) score > 3.5, P-value < 0.05). There were 48 bacterial taxa whose relative abundances were significantly distinct among the three groups, with 14, 12 and 23 taxa increasing in the EOCRC, LOCRC and NC groups, respectively ( Figure 3A ). As show in Figure 3B , on the phylum (LDA score=4.4283, P<0.001), class (LDA score=4.4283, P<0.001), order (LDA score=4.4283, P<0.001), family (LDA score=4.4247, P<0.001), and genus (LDA score=4.4256, P<0.001) levels, Fusobacteria was mostly abundant, showed a strong relationship with EOCRC. And Porphyromonas was another abundant bacterium in EOCRC group on the family (LDA score=4.0416, P<0.001), and genus (LDA score=4.0714, P<0.001) levels. And Bacteroidetes (LDA score=4.9111, P=0.0011), Bacteroidia (LDA score=4.9111, P=0.0011), and Bacteroidales (LDA score= 4.9110, P=0.0011) were designated as the most powerful markers in LOCRC patients. However, in the NC group, significantly increased Firmicutes (LDA score=4.9069, P=0.0021), Clostridia (LDA score=4.9182, P=0.0022) and Clostridiales (LDA score=4.9182, P=0.0022) were considered as the most significant markers. We performed kruskal-wallis test on the abundance of bacteria in the three groups at different levels to verify the results of LEfSe analysis ( Table 2 ). As shown in Table 2 , in the EOCRC group, Fusobacteria was more abundant on the level of phylum (P<0.001), class (P<0.001), order (P<0.001), family (P<0.001) and genus (P<0.001); and Porphyromonas was more abundant on the genus level (P<0.001), but the proportion of Porphyromonas was low. And in the LOCRC group, the proportion of Bacteroidetes were significantly higher on the level of phylum (P=0.001113), class (P=0.001113), and order (P=0.001113). And Prevotellaceae was more abundant in the LOCRC group on the family level (P<0.001). In the NC group, Clostridia was more abundant on the class level (P=0.002217) and the order level (P=0.002217), and Firmicutes was enriched on the phylum level (P=0.002079). Another abundant bacterium in the NC group is Actinobacteria, which is more abundant at the phylum level and at the phylum level (all P-values=0.002197). These results were consistent with the LEfSe analysis. Therefore, we concluded that the specific bacteria in gut bacterial composition of the EOCRC, LOCRC and NC group were Fusobacteria, Bacteroidetes, and Clostridia, respectively.
To study the functional and metabolic changes of the fecal microbial communities, we compared the measured sequences with the suggested database for the GOG and the KEGG module abundance from bacterial species. The COG potential functional annotation results showed that the EOCRC group as well as the LOCRC group were inferior in the following functions: transcription (P=0.01398) and defense mechanisms (P=0.04304). ( Figure 4A , Figure S3A ) Meanwhile, the three groups showed significant differences in the functions such as inorganic ion transport and metabolism (P=0.00225) and cell wall/membrane/envelope biogenesis (P=0.02534). ( Figure 4A ) Moreover, the KEGG modules involved in membrane transport (ko02010, P=0.00856) and porphyrin and chlorophyll metabolism (ko00860, P=0.04909) were overrepresented in the NC group compared with the EOCRC group and LOCRC group. ( Figure 4B ; Figure S3B )
The structure of the colorectal cancer population is gradually changing, and the rapidly increasing incidence of early-onset colorectal cancer requires vigilance (Collaborative et al., 2021; Sinicrope, 2022). The heterogeneity of clinical and molecular features of early-onset colorectal cancer is quite distinct, which means that it may be independent of traditional colorectal cancer (Silla et al., 2014; Fernandez-Rozadilla et al., 2021). As research progresses, the characteristics of the intestinal flora can be a major consideration in the etiology of many cancers (Murphy et al., 2019). Various studies have shown significant differences in the characteristics of gut microbiome across age, while the gut microbiome was considered to definite risk factor for colorectal cancer (O'Toole and Jeffery, 2015; Garrett, 2019; Wong and Yu, 2019). Therefore, we are more interested in clarifying the characteristics of gut microbiome in EOCRC. We selected patients with sporadic early-onset colorectal cancer from our center and recruited young healthy volunteers and late onset colorectal cancer patients with matched demographic characteristics. We initially delineated the gut flora of patients with sporadic colorectal cancer. Prior studies have shown that imbalanced gut flora in CRC is usually manifested by a decrease in alpha diversity, however studies derived from Chinese populations suggest that the species diversity of gut microbiota of CRC patients is not different from that of healthy populations (Wang et al., 2012; Gagniere et al., 2016; Yachida et al., 2019). A metagenomic sequencing based study suggested that the faecal alpha diversity separation estimates of EOCRC patients were significantly lower than those of the LOCRC patients and healthy young volunteers (Kong et al., 2022). In this study, we found that EOCRC patients had significantly lower alpha diversity than the gut flora of LOCRC patients and healthy young volunteers. The abundance of gut microbiota in the EOCRC group was significantly lower than that in the LOCRC group and NC group, and the number of bacterial genera in the EOCRC group was the lowest of the three groups. The alpha diversity and richness of the gut microbiota are generally considered to be independent of age (Takagi et al., 2019). However, according to our findings, in colorectal cancer patients, the species diversity and abundance were significantly lower in young patients. Meanwhile, significant differences were found in the beta diversity of gut microbiota among the three groups for overall comparison as well as for pairwise comparisons. Combined with alpha diversity analysis and the microbiota variability analysis, it is reasonable to assume that there are some specificities in the gut microbiota of early-onset patients. We compared the differences in abundant gut microbiota among the three groups. The proportion of Bacteroides in CRC patients, including EOCRC patients and in LOCRC patients was higher than that in NC patients (16.89 ± 17.17 vs. 16.18 ± 15.83 vs. 9.409 ± 10.83). But there was no obvious statistical difference among the three groups. Members of the genus Bacteroides account for a major fraction of the gut microbiome and colonize different parts of the colon (Kim et al., 2017). Bacteroides fragilis toxin can induce tumorigenesis through various pathways including IL17, signal transducer and activator of transcription 3 and nuclear factor-κB signaling in colonic epithelial cells (Chung et al., 2018). The Bacteroidetes were significantly enriched in the LOCRC group, and further analysis revealed that this part of the difference might be derived from a higher proportion of Prevotellaceae. Previous study has shown that Prevotellaceae was more abundant in CRC patients (Chen et al., 2012). However, there were only a small number of studies focusing on the association between Prevotellaceae and colorectal cancer. And exploring the role of Prevotellaceae in colorectal carcinogenesis may be a topic for future research. Fusobacterium is one of the definitive causative agents of CRC, and numerous studies have suggested that it can lead to colorectal carcinogenesis and progression (Mima et al., 2016; Yachida et al., 2019; Hong et al., 2021; Kong et al., 2021). In addition, Fusobacterium can promote chemoresistance in colorectal cancer by modulating autophagy, which can lead to poor prognosis in colorectal cancer patients (Yu et al., 2017). A previous study based on 16S rRNA gene sequencing suggested that Fusobacterium could serve as a differentially abundant genus marker for EOCRC, which could validate the results of the present study (Yang et al., 2021). Another study based on integrated metagenomic sequencing suggested that Bacteroides vulgatus and Flavonifractor plautii are unique taxon signatures for EOCRC, while Fusobacterium is a unique taxa signature for the LOCRC group (Kong et al., 2022). We suggest that differences in results are more likely to result from differences in sequencing methods and sample sources. Based on our study, Fusobacterium may play an important role in the gut microbiota of EOCRC patients, although it is present in lower proportions. Another genus enriched in the EOCRC group is Porphyromonas, and different species contained in it could promote colorectal carcinogenesis through butyrate-induced senescence or hematopoietic NLRP3 inflammasome (Okumura et al., 2021; Wang et al., 2021). In addition, we found a decrease in Clostridia in both the EOCRC group and the LOCRC group. Clostridia contains a variety of butyric acid-producing bacteria that can inhibit colorectal cancer development by modulating various signaling pathways and gut microbiota (Montalban-Arques et al., 2021; Stoeva et al., 2021; Zhou et al., 2022). Through functional prediction, we found some changes in certain COG functions and KEGG pathways in each group. Compared with healthy volunteers, the EOCRC and LOCRC groups showed a significant decrease in some functions (such as transcription and defense mechanisms) and some KEGG pathways (such as membrane transport and porphyrin and chlorophyll metabolism). However, we speculated that these distinctions were more derived from the differences between CRC patients and healthy individuals. Although there was no clear mechanism to suggest the difference between gut microbiota and cellular function, we speculated that the gut microbiota can interact and regulate each other through certain specific signaling pathways with the host (Zmora et al., 2019). The functional changes in different groups necessarily produce tumorigenic or protective effects and may serve as targets for the next treatment of colorectal cancer. Although our work has several novel findings, several limitations remain. The sample size of the control group (LOCRC group and NC group) of this study was adequate, but the sample size of the target population of our study needs to be larger. In addition, the male-to-female ratio of CRC patients in this study was slightly skewed, which may cause the findings of this study to be unrepresentative of the entire colorectal cancer population. Furthermore, metagenomic sequencing of the corresponding populations may give more convincing results. In conclusion, our study suggests that patients with early-onset colorectal cancer have a unique gut microbial profile. Gut microbes could be another characteristic of early-onset colorectal cancer. We hope that this study will provide some insight into the use of gut microbes as biomarkers for predicting the risk of early-onset colorectal cancer and contribute to the prevention and treatment of early-onset colorectal cancer.
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: NCBI, PRJNA883949.
The studies involving human participants were reviewed and approved by the Ethics Review committee in the second affiliated hospital of Harbin Medical University. The patients/participants provided their written informed consent to participate in this study.
HX, GW, and QT designed the project. JW, ZC, HH, ZY, YZ, ZH, and CW participated in patient selection and data collection. HX, YL, and YW carried out 16S sequencing, analyzed and interpreted the data. HX and JW preformed statistical analysis. HX, JW, and QT wrote the manuscript. All authors contributed to the article and approved the submitted version.
This work was supported by the Applied Technology Research and Development Project of Heilongjiang Province (number GA19C003) and National Natural Science Foundation Youth Project (number 82103030).
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648192 | 36386682 | Xiaxia Pan,Ziyuan Zhou,Bowen Liu,Zhongwen Wu | A novel therapeutic concern: Antibiotic resistance genes in common chronic diseases | 17-10-2022 | gut microbiota,antibiotic resistance genes,horizontal gene transfer,chronic diseases,monitoring and treatment | Infections caused by multidrug-resistant bacteria carrying antibiotic resistance genes pose a severe threat to global public health and human health. In clinical practice, it has been found that human gut microbiota act as a “reservoir” of antibiotic resistance genes (ARGs) since gut microbiota contain a wide variety of ARGs, and that the structure of the gut microbiome is influenced by the profile of the drug resistance genes present. In addition, ARGs can spread within and between species of the gut microbiome in multiple ways. To better understand gut microbiota ARGs and their effects on patients with chronic diseases, this article reviews the generation of ARGs, common vectors that transmit ARGs, the characteristics of gut microbiota ARGs in common chronic diseases, their impact on prognosis, the current state of treatment for ARGs, and what should be addressed in future research. | A novel therapeutic concern: Antibiotic resistance genes in common chronic diseases
Infections caused by multidrug-resistant bacteria carrying antibiotic resistance genes pose a severe threat to global public health and human health. In clinical practice, it has been found that human gut microbiota act as a “reservoir” of antibiotic resistance genes (ARGs) since gut microbiota contain a wide variety of ARGs, and that the structure of the gut microbiome is influenced by the profile of the drug resistance genes present. In addition, ARGs can spread within and between species of the gut microbiome in multiple ways. To better understand gut microbiota ARGs and their effects on patients with chronic diseases, this article reviews the generation of ARGs, common vectors that transmit ARGs, the characteristics of gut microbiota ARGs in common chronic diseases, their impact on prognosis, the current state of treatment for ARGs, and what should be addressed in future research.
The development and dissemination of antibiotic resistance genes (ARGs), the genetic basis of drug-resistant bacterial strains, are complex. Drug resistance is associated with the existence of genes in the bacterial genome that can produce drug-resistant phenotypes. And different genera, species and strains can exhibit different antibiotic-resistant phenotypes (Davies and Davies, 2010). The prevalence, categories, and resistance mechanisms of ARGs vary geographically and between body sites (Carr et al., 2020). The composition of intestinal ARGs is affected by many factors, and antibiotic use has a significant effect on the composition of gut microbiota and the profile of ARGs therein. Besides, antibiotic drugs can induce the development of ARGs and promote their transmission as well (Nobel et al., 2015; Jutkina et al., 2016; Doan et al., 2020). The gene transmission between individuals except the parent-offspring relationship is a common phenomenon and acts as an important source of genetic diversity which called the horizontal gene transfer (HGT) (Soucy et al., 2015). The human gut microbiome harbors a large number of ARGs, which can spread rapidly among bacterial pathogens through HGT (Penders et al., 2013; McInnes et al., 2020). The spread of antibiotic resistance poses a serious risk to human health. The effects of gut microbiota and ARGs in common chronic diseases, such as liver cirrhosis (Shamsaddini et al., 2021), diabetes (Shuai et al., 2022), and chronic kidney disease (CKD) (Wang X. et al., 2020) have been investigated, characteristic alterations of ARGs in different disease states have been highlighted, and there is an increasing abundance of ARGs with disease progression. In this review, we searched PubMed and Web of Science with keywords such as antibiotic resistance genes, chronic diseases, gut microbiome and horizontal gene transfer. We summarize the latest studies which identify gut microbiota and ARGs in common chronic diseases such as liver cirrhosis, diabetes, and CKD to discuss the role and characteristic of ARGs in the progression of related diseases. This manuscript should provide new concern for the development of individualized treatments targeting the monitoring and management of gut ARGs in common chronic disease.
Gut microbiota have gained increasing attention as a reservoir for ARGs, which can be transferred within the gut microbiota as well as to certain bacteria that simply pass through the gut (Hu et al., 2013; Penders et al., 2013). The factors that contribute to ARG development and prevalence are complex. It is well known that the exogenous substances such as antibiotic drugs have a notable effect on ARGs, and some commonly used non-antibiotic drugs may also alter the constitution of gut microbiota and the transmission of ARGs (Hu et al., 2013; Maurice et al., 2013).
Gut microbiota are frequently exposed to antibiotics, which is the key driver of bacterial antibiotic-resistance and leads to the prolonged presence of ARGs in the gut microbiota (Founou et al., 2016). Antibiotic-induced alteration of the gut microbiota directly causes a shift in the profile of resistance genes including increasing mutations, expression and transmission. Doan et al. investigated the gut resistome in over 500 children who used azithromycin twice a year for 4 years, and found that large-scale use of azithromycin may induce an increase abundance of ARGs (Doan et al., 2020). A study in which cefprozil was administered to healthy volunteers found that standard antibiotic therapy changes the gut microbiota specifically and predictably according to the initial gut microbiota composition which further alters the abundance of ARGs. For example, oral antibiotic treatment led to an increase in point mutations in the β-lactamase resistance gene blaCfxA-6 and an increasing abundance of some conditionally pathogenic bacteria, such as Enterobacter cloacae which may suggest that pre-treatment monitoring of gut microbiota composition can help avoid adverse effects of antibiotic therapy (Raymond et al., 2016). Another study also found that antibiotic-specific resistance gene homologs (AsRGs) had high transcriptional activity during antibiotic therapy and the relative abundance sustained expansion for 3 months which suggested a long-lasting increment of ARGs expression (Kang et al., 2021). Additionally, Nobel et al. (2015) established a mouse model simulating pediatric antibiotic use and found that the two widely used classes of antibiotics including beta-lactam and macrolide, had profound effects on the gut microbiome and metagenome. Early therapeutic-dose pulsed macrolide treatment (PAT) induced alterations in the mouse gut microbiota and increased the expression of four ARGs (acrA, acrB, ant3Ia, and ant2Ia) associated with macrolide-resistance. Except increase the mutations or expression of ARGs, antibiotic drugs can also promote their transmission. Jutkina et al. found that very low concentrations (10 μg/L) of tetracycline drive the transfer of diverse ARGs (Jutkina et al., 2016). Wu et al. discovered that levofloxacin induced the plasmid mediated transformation and increased both the abundance and spread of antibiotic-resistant Escherichia coli (Wu H. Y. et al., 2020). Understanding the role of antibiotics in regulating the expression and propagation of ARGs is critical.
Some commonly used non-antibiotic drugs can influence the gut microbiota composition, which can manifest as a high incidence of antibiotic-like side effects. About 24% of non-antibiotic drugs targeting humans which inhibit the growth of at least one gut bacterial strain could promote antibiotic resistance (Maier et al., 2018). Non-antibiotic drugs use may be a causal factor in the generation and spread of intestinal ARGs. Wang et al. found that five non-antibiotic drugs including anti-inflammatory drugs (ibuprofen, naproxen, and diclofenac), lipid-lowering drugs (gemfibrozil), and β-blockers (propranolol) promoted transmission of ARGs through bacterial transformation (Wang Y. et al., 2020).
Although bacterial communities in human and animal gastrointestinal tracts, rivers, and wastewater are major sources of ARGs, the origin of these ARGs and the process of transmission from the environment to clinical settings are still poorly understood (Larsson and Flach, 2021; Li et al., 2021; Zhuang et al., 2021). The spread of ARGs in the human gut is more complex where ARGs can be transmitted through HGT (including transformation, transduction, and conjugation) (McInnes et al., 2020). The extensive spread of ARGs, due to antibiotic abuse, is associated with different vectors including the plasmids, viruses and bacteria that carry ARGs and act as carriers to transfer genetic information between microbial species (Botelho et al., 2019; McInnes et al., 2020). We describe the emergence of ARGs and the main HGT modes of ARGs in the gut in Figure 1.
Plasmids are autonomously replicating DNA molecules which can carry multiple resistance genes and are the main mode of transmission of ARGs between bacteria. Plasmid-based HGT plays an important role in the spread of ARGs between bacterial conspecifics or different species, and the evolutionary status of plasmid-carrying bacteria is also an important factor in the transmission of antibiotic resistance plasmids between bacterial pathogens in vivo (Navon-Venezia et al., 2017; San Millan, 2018; Stalder et al., 2019). Gumpert et al. provided the first genomic-level characterization of ARG transfer in undisturbed human gut microbiota and demonstrated that plasmids carrying ARGs, in the infant, persist and are transferred even in the absence of antibiotic administration (Gumpert et al., 2017). Besides, it has been proposed that establishing linkage between plasmids and hosts in wastewater environments, through the in vivo proximity-ligation method, can accurately identify carriers of ARGs and help limit the spread of pathogen resistance (Stalder et al., 2019).
Viruses are also thought to be key vectors for ARG transfer, enabling the spread of ARGs more broadly than in bacterial genomes. ARGs have been observed only in a very small number of phages, through virome sequencing and assembly analysis, but phages carrying ARGs are known to exist in the human gut and other environments. Phage-mediated HGT diversifies the microbial gene pool, and since phages can survive in the environment for a long time, they can prolong or delay the transfer of genetic information (Touchon et al., 2017; Debroas and Siguret, 2019). Antibiotic treatment may contribute to the release of bacterial prophages carrying ARGs, leading to an increase in free phages carrying ARGs in the human gut, thereby spreading drug resistance (Fernandez-Orth et al., 2019). AsRGs are frequently present on MGEs, and the relative abundance of mobile AsRGs is often remains consistently high even 3 months after antibiotic treatment. Consequently, MGEs contribute to the expansion and persistence of antibiotic resistance gene pools, and mobile AsRGs also enable transfer of antibiotic resistance through potential HGT (Kang et al., 2021).
Bacteria in the gut not only acquire ARGs, but also contribute to the transfer of ARGs to other bacteria in the gut (Ravi et al., 2014). Jiang et al. demonstrated that ARGs can be transferred from antibiotic-producing actinobacteria to proteobacteria, which may be related to conjugative transfer of a carrier sequence from proteobacteria to actinobacteria (Jiang X. et al., 2017). As early as 1988, quinolone resistant strains of Klebsiella pneumoniae were found that could transmit low level of quinolone resistance to other bacteria (Martínez-Martínez et al., 1998). K. pneumoniae is a major source of global drug resistance with ARGs located on both the chromosome and plasmids of the bacterium. Under antibiotic selection pressure, the bacterium continues to accumulate ARGs through de novo mutations and MGEs, eventually resulting in the appearance of super-resistant strains. ARGs continue to accumulate as the resistome evolves, and K. pneumoniae strains have been shown to harbor 52 antibiotic resistance plasmids capable of promoting the transfer of resistance through HGT (Navon-Venezia et al., 2017). Pseudomonas aeruginosa can also acquire drug resistance through chromosomal mutations and acquisition of ARGs by HGT (Botelho et al., 2019). Salmonella enterica serovar Typhimurium (S.Tm) can survive in host tissues during antibiotic treatment, and reimplantation of these bacteria into the intestinal lumen facilitates the transfer of plasmids carrying ARGs between different Enterobacteriaceae. Small reservoirs of persistent pathogens can also encourage the spread of mixed resistance plasmids among microbiota in the gut even in the absence of selection for resistance genes encoded by plasmids (Bakkeren et al., 2019). Moreover, (Klassert et al., 2021) observed a dynamic process of bacterial colonization and spread of ARGs after initial patient admission and sequencing data revealed that a stable community structure formed at three indoor sites in the hospital environment (floor, doorhandle and sink) in only a few weeks, and that the rapid colonization process of microbiota correlated with a remarkable increase in ARGs. Studying HGT in the gut microbiota could help to identify and develop new interventions aimed at minimizing the transmission of ARGs between commensals and opportunistic pathogens.
Currently, the presence of chronic diseases is now a major challenge to global health,representing a diverse group of diseases that last a year or more and progress slowly. Chronic disease include, but is not limited to, a variety of disorders such as diabetes, cardiovascular disease (CVD), chronic kidney disease (CKD) or chronic liver disease, but could also involve some chronic infections (Bauer et al., 2014; Bergman and Brighenti, 2020). In the long-term clinical progression, those patients with chronic diseases are more vulnerable to various infection and thus, are more likely to be exposed to different antibiotics. However, as discussed above, the emergence of antibiotic resistance affects community structure, and alterations in gut microbiota, caused by antibiotic exposure, are associated with certain metabolic diseases (Blaser and Falkow, 2009; Blaser, 2016; Joossens et al., 2019). A previous study analyzing the ARGs of gut microbiota from 162 people found a total of 1,093 ARGs. Analysis of the microbial origin of the ARGs revealed that they were more likely to be found in Proteobacteria, and regional differences in ARGs revealed different ratios of resistance genes to total intestinal genes in Chinese, Danish and Spanish populations, with ratios of 0.94, 0.44, and 0.89%, respectively. The Chinese population has a higher abundance of resistance genes and a greater variety of resistance gene types, with tetQ and ermB being the two most abundant ARGs (Hu et al., 2013). Yang et al. (Yang et al., 2016) performed a preliminary analysis of 1,267 stool samples from 368 Chinese, 139 American, 401 Danish and 359 Spanish, to investigate country-specific gut resistance profiles, and found a total of 112 ARGs were present in 90% of the samples from different countries, with the highest abundance of ARGs detected in samples from China, and the duration of use and applicability of antibiotics were correlated with the accumulation of ARGs in gut microbiota. The state of diseases can also influence ARGs, and microbial species specifically related to certain diseases may be hosts for ARGs. For example, Escherichia coli carries multiple ARG subtypes, so monitoring ARGs in this bacterium can help predict or treat certain diseases (Qiu et al., 2020). A brief introduction of ARGs in common chronic diseases can be seen in Figure 2.
Patients with cirrhosis are prone to infections and this is may be related to alterations in the composition of gut microbiota in these patients that make them more susceptible to colonization by antibiotic-resistant bacteria and pathogenic species (Bajaj et al., 2019a,b; Tranah et al., 2021). Previous research has demonstrated that progressive changes in gut microbiota of patients with cirrhosis are strongly associated with disease progression and poor prognosis (Bajaj et al., 2014; Sung et al., 2019; Sole et al., 2021; Trebicka et al., 2021). The abundance of pathogens that express ARGs increases as cirrhosis progresses (Bajaj et al., 2014; Qin et al., 2014; Chen et al., 2015). Thus, detecting and targeting ARGs may improve the clinical outcomes of patients with cirrhosis. Shamsaddini et al. (2021) investigated characteristic ARGs that correlated with infection and poor outcomes, in the gut microbiota of patients with cirrhosis. Pathogenic-related resisitomes, such as Enterobacteriaceae, Streptococcus spp., Enterococcus spp. and Fusobacterium spp., were more abundant in cirrhotic patients than in healthy controls, resulting in a higher abundance of ARGs related to β-lactamases, vancomycin and quinolones, in the gut of cirrhotic patients, with disease progression. ARGs associated with hepatic encephalopathy (HE) and ascites have also been identified. Patients with HE had a higher abundance of ARGs associated with β-lactamases, vancomycin resistance and RbpA bacterial RNA polymerase binding protein, than non-HE patients, while quinolone resistance genes were lower than those in non HE patients. This study also found that ascites was connected with an increase in Pseudomonas, Serratia and Clostridium perfringens in the gut, while PPI use was related to higher vancomycin resistance. PPIs can modulate the composition of the gut microbiota in cirrhotic patients, and systematic stopping and judicious use of PPIs is needed in patients with decompensated cirrhosis (Bajaj et al., 2018; Shamsaddini et al., 2021). ARG abundance increases in Gram-positive bacteria in cirrhosis patients, and these ARGs differ from those seen in type 2 diabetes (T2D) and CKD patients. Cirrhosis was significantly separated from both diabetes and CKD by principal coordinate analysis, and six ARGs differed between the cirrhosis and diabetes groups. Patients with cirrhosis had a greater abundance of ARGs, with a broader resistance spectrum. These ARGs belonged to a variety of pathogenic Gram-positive and Gram-negative bacteria (Qin et al., 2012; Shamsaddini et al., 2021). A prospective assessment of two groups of European patients with decompensated cirrhosis, found that infections caused by antibiotic-resistant bacteria were common in cirrhosis patients. Among culture-positive infections, those caused by multidrug-resistant microorganisms accounted for 29% in the year 2011 and 38% in 2017–2018, with the most common multidrug-resistant bacteria being ultra-broad-spectrum β-lactamase-producing Enterobacteriaceae. The study also found that antibiotic resistance patterns were highly heterogeneous, with significant variation in multidrug-resistant bacteria types acrossed countries and centers (Fernandez et al., 2019). In European patients with severe liver disease, multidrug-resistant bacterial infections have become a common and growing medical problem associated with poor prognosis, and preventing the spread of antibiotic resistance in cirrhosis is critical.
Previous studies have shown that gut microbiota composition is strongly associated with T2D and that microbiota in T2D patients change in line with blood glucose levels (Qin et al., 2012; Brunkwall and Orho-Melander, 2017; Wu H. et al., 2020). Antibiotic use disrupts gut microbiota composition and long-term use can increase the risk of T2D in women (Yuan et al., 2020). The study of a large human cohort revealed the detailed profiles of the gut antibiotic resistome in different disease conditions, revealing a new association between the resistome, gut microbiota, and the progression of T2D (Shuai et al., 2022). The most enriched ARGs in this cohort included multidrug, tetracycline and Macrolide-Lincosamide-Streptogramin (MLS), with tetracycline tetQ being the most abundant ARGs subtype, which is more in line with previous study (Hu et al., 2013; Shuai et al., 2022). Due to the overall shift in gut ARGs and microbiota composition in the healthy, prediabetes, and T2D groups, a new Diabetes-ARGs Score (DAS) was proposed. The DAS is strongly associated with the progression of T2D and gut ARGs can be found in a wider range of bacterial species as the disease progresses, which may explain some of the pathophysiological mechanisms associated with T2D. There were 25 ARGs associated with T2D, and genome-wide association analysis of selected ARGs revealed that Vancomycin_vanX, Multidrug_emrE, MLS_ermX, and Quinolone_norB were all positively correlated with a high risk of T2D, and that the abundance of Quinolone_norB and β-lactam subtypes increased with diabetes progression. Genetic tools were constructed for these ARG profiles, and genetic predictions also confirmed that the higher abundance of intestinal ARGs is associated with a high risk of T2D. It was also found that changes in gut antibiotic resistance preceded changes in gut microbiota during T2D progression and the pattern of ARG abundance shifted as disease progressed. Intestinal antibiotic resistance was also associated with fecal metabolites; for example, DAS and Vancomycin_vanX were positively correlated with L-isoleucine and L-leucine in fecal samples (Wu H. et al., 2020; Shuai et al., 2022). The above confirmed the significant association of plasma L-isoleucine and L-leucine with future diabetes risk seen in a previous study (Wang et al., 2011). These results reveal a new horizon between gut microbiota and T2D progression, which was revealed by studying antibiotic resistance, and these characteristic ARGs may become targets for T2D intervention in future studies.
Gut microbiota composition is remarkably altered in patients with CKD, and this is related to the inflammatory state and renal function in CKD patients (Vaziri et al., 2013; Jiang S. et al., 2017; Xu et al., 2017; Meijers et al., 2019). Since patients with CKD, especially dialysis patients, are susceptible to infections they are considered to be reservoirs of antibiotic-resistant pathogens (Su et al., 2018). Antibiotic-resistant bacteria (ARB), including methicillin-resistant Staphylococcus aureus, vancomycin-resistant Enterococcus spp., and several multidrug-resistant gram-negative organisms, colonize and infect patients with CKD who require dialysis or kidney transplantation. Moreover, increased antibiotic resistance is positively associated with the risk of infection and death. Consequently, clinicians must be familiar with the local epidemiological history associated with ARBs and remain alert to the emergence of new resistance patterns (Zacharioudakis et al., 2015; Wang et al., 2019). Imbalance in gut microbiota can promote further deterioration of renal function and lead even to renal failure. Animal models of CKD have demonstrated that two microorganisms, Eggerthella lenta and Fusobacterium nucleatum, can produce serum uremic toxins and exacerbate the development of kidney disease. The prevalence of several other pathogenic bacteria such as Klebsiella pneumoniae, Acinetobacter, Enterobacter, and Legionella, also increases during CKD (Wang X. et al., 2020). When comparing CKD and cirrhosis, only seven ARGs were found to differ between the cirrhosis and CKD groups of patients, and β-lactamase genes were present in both disease groups. The abundance of glycopeptide, vancomycin, cephalosporin, and rifamycin resistance genes was higher in patients with cirrhosis, and the abundance of ARB belonging to a wide spectrum of Gram-negative and Gram-positive pathogenic microorganisms is also increased in patients with CKD (Wang X. et al., 2020; Shamsaddini et al., 2021). The abundance of ARGs is higher in patients with cirrhosis than in those with CKD, and the more highly abundant ARGs are associated with high antibiotic use and increased abundance of antibiotic-resistant pathogens (Su et al., 2018; Shamsaddini et al., 2021). Patients with kidney impairment are at high risk of infections caused by multidrug resistance organisms, and kidney dysfunction at the time of admission should be used as a signal to closely monitor microbiological culture results thereafter (Su et al., 2018). It has been shown that the gut microbiome and the expression of intestinal genes are both altered in patients with CKD. Thus, increases in Proteobacteria at the phylum level, Enterobacteriaceae and Corynebacteriaceae at the family level, Enterococcus and Clostridium at the genus level are all characteristically observed in CKD patients, and the expression of genes related to the trimethylamine (TMA) metabolic pathway are altered in CKD. TMA is absorbed by the liver and oxidized to trimethylamine N-oxide (TMAO), which is ultimately excreted by the kidney. Impaired kidney function and an imbalance of gut microbiota in CKD patients causes elevated plasma TMAO, which increase cardiovascular risk in patients with CKD (Xu et al., 2017).
Alterations in the gut microbiota are also associated with the development of cardiovascular disease. Several members of the Enterobacteriaceae, including Escherichia coli, Klebsiella spp., and Enterobacter aerogenes and oral bacteria such as Streptococcus spp. and Lactobacillus salivarius, are all more abundant in patients with atherosclerotic cardiovascular disease (CVD), whereas the levels of butyric acid-producing bacteria, such as Roseburia intestinalis and Faecalibacterium cf. prausnitzii, are reduced (Jie et al., 2017; Tang et al., 2017; Witkowski et al., 2020). Shuai et al. (2022) also observed that differences in gut antibiotic resistance are associated with cardiometabolic risk; the α-diversity indices of gut ARG were strongly correlated with triglycerides and the total cholesterol/high-density lipoprotein (HDL) cholesterol ratio but negatively associated with HDL cholesterol. Further analysis confirmed that these correlations were unrelated to T2D and that a higher α diversity of gut ARG was linked to higher cardiometabolic risk. Yuan et al. (2020) proposed that prolonged antibiotic use was associated with future cardiovascular disease incidence. Identifying new therapeutic options for altered gut microbiota composition could provide a new direction for future treatments. There are eight species in the gut, including Anaerococcus hydrogenalis, Clostridium asparagiforme, Clostridium hathewayi, that are associated with TMA accumulation (Romano et al., 2015). Elevated plasma TMAO levels can promote the acceleration of chronic diseases including CVD, T2D, and CKD through a variety of mechanisms. TMAO also promotes oxidative stress in blood vessels and dysfunction of the endothelium. Consequently, the inhibition of TMAO production, a gut microbiota metabolite, is a new strategy for the management of several common chronic diseases (Mente et al., 2015; Wang et al., 2015; Chen et al., 2019; Brunt et al., 2020).
ARGs in gut microbiome can affect brain function and significant changes in the gut microbiota of children may cause neurodevelopmental disorders. Kovtun et al. studied the distribution of ARGs in the gut microbiome of 3–5 years old healthy children and children with autism spectrum disorder (ASD) by metagenomic analysis. They found significant differences characterized by increased ARGs in the gut microbiome of children with ASD. Genes aac(6′)-aph(2″) from Enterococcus, cepA-49 from Bacteroides (b-lactams) and tet(40) from Megasphaera, are three specific ARGs in ASD (Kovtun et al., 2020). Another study found significant changes in the gut microbiota in patients with inflammatory bowel disease (IBD). Increased abundance of pathogenic Escherichia coli showed a positive correlation with multiple ARGs, such as mdtO, mdtP, emrK, etc., which may accelerate IBD progression. For ARG types, beta-lactam, fosmidomycin, multidrug and polymyxin resistance genes were significantly enriched in Crohn’s disease (CD) patients (Xia et al., 2021). Duvallet et al. (2017) found that 10–15 genus-level changes among gut flora could lead to the development and progression of multiple human diseases. The close relationship between ARGs and chronic diseases suggests potential value to monitor and evaluate gut microbiota and the ARGs of patients with chronic diseases. The variation of ARGs in common chronic diseases and their impact on disease development are listed in Table 1.
Exposure to ciprofloxacin, amoxicillin, and metronidazole, beyond 5 days, is associated with a marked increase in ARG abundance (Bajaj et al., 2021). Rifaximin treatment, on the other hand, is positively correlated with the abundance of beneficial taxa and negatively correlated with the abundance of Klebsiella spp. resistome and Gram-negative ARGs. Rifaximin is a derivative of rifamycin, which is produced by Streptomyces spp. Interestingly, the Streptomyces spp. resistome increases in cirrhosis patients compared with diabetes or CKD patients (Bass et al., 2010; Shamsaddini et al., 2021). Encouraging the use of non-absorbable enteric-specific antibiotics, such as rifaximin, can improve the prognosis of cirrhosis patients.
Gut microbiota are recognized as a potential reservoir for the transmission of drug-resistant genes from symbionts to pathogens. Microbiota-based therapies can reduce the abundance of ARGs and ARBs in the gut microbiome of patients, reduce their risk of ARB-induced infections, and reduce the risk of transfer of pathogens carrying ARGs to other individuals (Langdon et al., 2021). To restore the biodiversity of microbial communities, probiotics can be used to replace important missing or depleted species or strains (Blaser, 2016). Probiotics impact the ARG reservoir in the human gastrointestinal microbiota in an antibiotic-dependent way. Montassier et al. (2021) showed that disruption of the gut microbiota by antibiotics supports probiotic colonization, but that the probiotics can amplify strains carrying vancomycin resistance genes, aggravating the amplification of the resistome in the gastrointestinal mucosa, and may themselves be reservoirs for the amplification of resistome in the gut. Thus, the potential transfer of ARGs through probiotics in the human intestine needs to be further evaluated.
ARGs associated with poor outcomes can be used as prognostic predictors and new therapeutic strategies could be developed to target these ARGs. Fecal microbiota transplant (FMT) is one such strategy and is associated with a decline in the abundance of ARGs (Saha et al., 2019; Bajaj et al., 2021). Metagenomic analysis revealed that FMT can reduce the burden of ARG in both cirrhosis and non-cirrhosis patients and the expression of gut ARGs was reduced in patients with decompensated cirrhosis after receiving capsule or enema FMT. After capsule FMT treatment patients showed a reduction in abundance of vancomycin (VanH), beta-lactamase (ACT), and rifamycin ARGs. After receiving antibiotics and enema FMT, vancomycin and beta-lactamase ARGs declined by day 15 and the abundance of cephalosporinase cepA, Enterococcus faecalis vanW, lincosamides (including clindamycin), and aminocoumarin resistance genes was also reduced. Irrespective of the mode of FMT dosing, significantly lower abundance of ARGs was observed in the decompensated cirrhosis group than in the pre-FMT and non-FMT groups. Moreover, vanH abundance increased, with time, in the placebo group but this was not found in the FMT group, vanH enhances vancomycin resistance of the cell wall of E. faecium (Bajaj et al., 2021). Thus, the beneficial effect of FMT in reducing the abundance of pathogen-related ARGs, seen for other diseases in previous studies, can now be extended to the field of advanced cirrhosis. Monitoring ARGs should not only improve prognosis, but will also allow the selection of appropriate treatment options for more abundant ARGs. However, extensive trials are still needed to verify that FMT reduces antibiotic resistance in patients with cirrhosis. Targeted intervention of specific microbiomes through FMT is also a potential strategy for the treatment of CKD (Meijers et al., 2019).
The amount, characteristic, and function of ARGs, collectively called the resistome, is the sum of all ARGs and their precursors in the microbiota (D’Costa et al., 2006; Wright, 2007; Casals-Pascual et al., 2018; Schwartz et al., 2020). The next generation of resistome research promises to combat emerging resistance threats (Crofts et al., 2017). We conclude that the composition of the gut microbiota is altered after antibiotic therapy with increased abundance of ARB, which further increases the abundance of ARGs in the gut. While some of the ARB in the gut microbiota were replaced by normal flora after FMT which induces a reduction in the abundance of ARGs. In addition, the individual baseline of gut microbiome can influence resistome alterations. When individual baseline species diversity is high, patients are more likely to experience an increase in ARGs during antibiotic treatment (Willmann et al., 2019). To better predict the impact of individual antibiotic on the gut microbiome and resisitome, further studies focused on the individual baseline of gut microbiome are needed. Antibiotic use has an impact on both gut microbiota and ARGs and identifying disease-host-specific ARGs should provide tailored antibiotic treatment strategies for the clinic. For example, the cfxA gene is significantly enriched in the gut of patients receiving antibiotic therapy in 1 month and is considered a potential biomarker to differentiate between patients and healthy populations (Duan et al., 2020). To date, few studies have evaluated the monitoring and management of ARGs. Thus, more research is needed to confirm the effectiveness of treatment against gut-specific ARGs, in different disease states, and the link between gut ARGs and fecal metabolites. Some studies have already found an association between ARGs and chronic liver disease, diabetes, CKD and CVD, but more research is needed to confirm the effect of ARGs on disease progression in these chronic diseases and to explore whether ARGs are altered in other chronic diseases such as chronic neurological disease and chronic respiratory disease. Monitoring ARGs is particularly important in the chronic disease management (CDM), in order to minimize damage to essential commensal microorganisms, individualized treatment strategies targeting gut-specific ARGs are expected to be a new direction in the future CDM.
The presence of ARGs in the gut microbiota underpins the increasing failure of treatments for fatal bacterial infections. Since the misuse of antibiotics aggravates the production and spread of ARGs, limiting antibiotic overuse is an important tool to alleviate antibiotic resistance. A better awareness of HGT in the gut will help to open up new and effective interventions to reduce the transmission of ARGs. Altered gut microbiota and characteristic ARGs are strongly associated with disease progression in a number of chronic diseases. Future research is needed to confirm the effects of ARGs on the gut microbiota in patients with such diseases. Novel ARGs are still emerging and the detection and targeting of ARGs, as treatment goals, can change the prognosis of chronic diseases. Aiming to reduce the abundance of intestinal ARBs and slow down pathologic progression in chronic diseases, the individualized therapies focus on intestinal ARGs require more attention and further research.
XP carried out the initial literature review and wrote the initial manuscript. ZZ have revised the manuscript. BL have collected the relevant literature data. ZW have provided expertise and insight relating to ARGs and checked the manuscript. All authors have read and approved the final manuscript. |
PMC9648200 | Wenhua Deng,Huan Xu,Yabin Wu,Jie Li | Diagnostic value of bronchoalveolar lavage fluid metagenomic next-generation sequencing in pediatric pneumonia | 27-10-2022 | metagenomic next-generation sequencing (mNGS),conventional microbiological tests (CMTs),bronchoalveolar lavage fluid (BALF),pediatric,pneumonia | Objectives The aim of this study was to evaluate the diagnostic value of bronchoalveolar lavage fluid (BALF) metagenomic next-generation sequencing (mNGS) versus conventional microbiological tests (CMTs) for pediatric pneumonia. Methods This retrospective observational study enrolled 103 children who were diagnosed with pneumonia and hospitalized at Hubei Maternity and Child Health Care Hospital between 15 October 2020 and 15 February 2022. The pneumonia diagnosis was based on clinical manifestations, lung imaging, and microbiological tests. Pathogens in the lower respiratory tract were detected using CMTs and BALF mNGS (of DNA and RNA). The diagnostic performance of BALF mNGS was compared with that of CMTs. Results In 96 patients, pathogens were identified by microbiological tests. The overall pathogen detection rate of mNGS was significantly higher than that of CMTs (91.3% vs. 59.2%, p = 0.000). The diagnostic performance of mNGS varied for different pathogens; however, its sensitivity and accuracy for diagnosing bacterial and viral infections were both higher than those of CMTs (p = 0.000). For the diagnosis of fungi, the sensitivity of mNGS (87.5%) was higher than that of CMTs (25%); however, its specificity and accuracy were lower than those of CMTs (p < 0.01). For the diagnosis of Mycoplasma pneumoniae, the specificity (98.8%) and accuracy (88.3%) of mNGS were high; however, its sensitivity (42.1%) was significantly lower than that of CMTs (100%) (p = 0.001). In 96 patients with definite pathogens, 52 cases (50.5%) were infected with a single pathogen, while 44 cases (42.7%) had polymicrobial infections. Virus–bacteria and virus–virus co-infections were the most common. Staphylococcus aureus, Haemophilus influenzae, rhinovirus, cytomegalovirus, parainfluenza virus, and fungi were more likely to be associated with polymicrobial infections. Conclusions BALF mNGS improved the detection rate of pediatric pneumonia, especially in mixed infections. The diagnostic performance of BALF mNGS varies according to pathogen type. mNGS can be used to supplement CMTs. A combination of mNGS and CMTs may be the best diagnostic strategy. | Diagnostic value of bronchoalveolar lavage fluid metagenomic next-generation sequencing in pediatric pneumonia
The aim of this study was to evaluate the diagnostic value of bronchoalveolar lavage fluid (BALF) metagenomic next-generation sequencing (mNGS) versus conventional microbiological tests (CMTs) for pediatric pneumonia.
This retrospective observational study enrolled 103 children who were diagnosed with pneumonia and hospitalized at Hubei Maternity and Child Health Care Hospital between 15 October 2020 and 15 February 2022. The pneumonia diagnosis was based on clinical manifestations, lung imaging, and microbiological tests. Pathogens in the lower respiratory tract were detected using CMTs and BALF mNGS (of DNA and RNA). The diagnostic performance of BALF mNGS was compared with that of CMTs.
In 96 patients, pathogens were identified by microbiological tests. The overall pathogen detection rate of mNGS was significantly higher than that of CMTs (91.3% vs. 59.2%, p = 0.000). The diagnostic performance of mNGS varied for different pathogens; however, its sensitivity and accuracy for diagnosing bacterial and viral infections were both higher than those of CMTs (p = 0.000). For the diagnosis of fungi, the sensitivity of mNGS (87.5%) was higher than that of CMTs (25%); however, its specificity and accuracy were lower than those of CMTs (p < 0.01). For the diagnosis of Mycoplasma pneumoniae, the specificity (98.8%) and accuracy (88.3%) of mNGS were high; however, its sensitivity (42.1%) was significantly lower than that of CMTs (100%) (p = 0.001). In 96 patients with definite pathogens, 52 cases (50.5%) were infected with a single pathogen, while 44 cases (42.7%) had polymicrobial infections. Virus–bacteria and virus–virus co-infections were the most common. Staphylococcus aureus, Haemophilus influenzae, rhinovirus, cytomegalovirus, parainfluenza virus, and fungi were more likely to be associated with polymicrobial infections.
BALF mNGS improved the detection rate of pediatric pneumonia, especially in mixed infections. The diagnostic performance of BALF mNGS varies according to pathogen type. mNGS can be used to supplement CMTs. A combination of mNGS and CMTs may be the best diagnostic strategy.
The World Health Organization reports that pneumonia is the leading cause worldwide of mortality among children younger than 5 years old (GBD 2015 LRICollaborators, 2017). In clinical practice, identifying pathogens in infectious diseases is a difficult problem. Conventional microbiological tests (CMTs) are limited in their scope for pathogen detection; they are time-consuming, have low detection rates, and usually detect only single pathogens. Although polymerase chain reaction (PCR) tests and serological detection have expanded the detection range of CMTs and increased detection rates, clinicians must first identify the type of pathogen. It is important to diagnose pathogens quickly and accurately in order to shorten the hospital stay and reduce complications and mortality. Metagenomic next-generation sequencing (mNGS) is an unbiased detection technology that can detect multiple pathogens across a wide range. It is relatively time-saving, with a turnaround time of 24–48 h. mNGS has been shown in recent years to be advantageous and viable for the identification of respiratory tract infection pathogens (Leo et al., 2017). However, sequencing DNA and RNA at the same time using mNGS has rarely been reported. In the present study, we compared the diagnostic value of CMTs and mNGS (DNA and RNA) for detecting pneumonia pathogens in children.
This retrospective observational study enrolled children who were diagnosed with pneumonia and hospitalized at the Maternity and Child Health Care Hospital of Hubei Province between 15 October 2020 and 15 February 2022. The inclusion criteria were as follows: (1) the child presented with typical clinical signs of pulmonary infection, such as fever, cough, sputum, and dyspnea; and (2) the diagnosis of pulmonary infection was supported by radiological evidence (e.g., chest computed tomography scan). We excluded patients who were not tested using bronchoalveolar lavage fluid (BALF) mNGS (DNA and RNA). A total of 103 children were enrolled in this study. The recruitment process is illustrated in Figure 1 . Patient age, sex, symptoms, laboratory findings, lung imaging, bronchoscopic findings, and medical history were recorded. All included patients underwent bronchoscopy to obtain BALF samples for use in CMTs and mNGS. Bronchoscopies were performed by experienced bronchoscopy physicians according to standard safety protocols. No serious adverse events were associated with the bronchoscopy procedures. This study was approved by the Institutional Ethics Committee of the Maternal and Child Health Hospital of Hubei Province [2022] IEC (018).
Routine samples were collected, including BALF, sputum, and blood. CMTs were performed within 2 days of admission, including sputum and BALF culture and smear (acid-fast staining for Mycobacterium tuberculosis; India ink staining for Cryptococcus), nasopharyngeal (NP) swab multiplex PCR (13 respiratory pathogens), BALF PCR (for Mycoplasma pneumoniae), serum antibody test (for M. pneumoniae), antigen test (for influenza virus A/B, 1,3-β-D-glucan antigen), and serum and BALF galactomannan test (Aspergillus spp.). The detection methods are specified in Supplementary Tables 1 and 2 .
Based on the clinical diagnosis, two experienced clinicians analyzed all patients’ CMT and mNGS results, along with their medical records. First, each clinician determined whether the patient had pneumonia, based on the Chinese guidelines for the diagnosis of pneumonia in children (National Health Commission of the People’s Republic of China, State Administration of Traditional Chinese Medicine, 2019), according to clinical symptoms, pulmonary imaging, and clinical laboratory examination results. Second, etiology was determined by a comprehensive analysis of the patient’s clinical manifestations, laboratory findings, lung imaging, microbiological examination, and treatment response. If there was disagreement between clinicians, another senior clinician was consulted and a consensus was reached.
Bronchoscopy was performed according to standard procedures using a flexible fiberoptic bronchoscope. A special collector was used to collect 3–5 ml of BALF, which was stored at 4°C. The BALF was sent for mNGS analysis (DNA and RNA). DNA was extracted using a QIAamp® UCP Pathogen DNA Kit (Qiagen), following the manufacturer’s instructions. Human DNA was removed using benzonase (Qiagen) and Tween20 (Sigma). Total RNA was extracted using a QIAamp® Viral RNA Kit (Qiagen). Ribosomal RNA was removed using a Ribo-Zero rRNA Removal Kit (Illumina). Complementary DNA (cDNA) was generated using reverse transcriptase and deoxynucleoside triphosphates (Thermo Fisher Scientific). Libraries were constructed for DNA and cDNA samples using the Nextera XT DNA Library Prep Kit (Illumina). Library quality was assessed using the Qubit dsDNA HS Assay Kit, followed by a high-sensitivity DNA Kit (Agilent) on an Agilent 2100 bioanalyzer. Library pools were then loaded onto an Illumina NextSeq CN500 sequencer for 75 cycles of single-end sequencing, generating approximately 20 million reads per library. For negative controls, we prepared peripheral blood mononuclear cell samples (105 cells/ml) from healthy donors, in parallel with each batch, using the same protocol. Sterile deionized water was extracted alongside the specimens to serve as a non-template control.
Trimmomatic was used to remove low-quality reads, adapter contamination, duplicate reads, and reads shorter than 50 bp. Low-complexity reads were removed using K-complexity, with default parameters. Human sequence data were identified and excluded by mapping to a human reference genome (hg38) using Burrows–Wheeler Aligner software. We designed a set of criteria, similar to the criteria of the National Center for Biotechnology Information (NCBI), for selecting representative assemblies of microorganisms (bacteria, viruses, fungi, protozoa, and other multicellular eukaryotic pathogens) from the NCBI Nucleotide and Genome databases (National Center for Biotechnology Information). These were selected according to three references: (1) Johns Hopkins ABX Guide ; (2) Manual of Clinical Microbiology (Manual of clinical microbiology); and (3) case reports and research articles published in current peer-reviewed journals (Fiorini et al., 2017). The final database consisted of approximately 13,000 genomes. Microbial reads were aligned to the database using SNAP v1.0 beta 18 (Zaharia et al., 2021). Virus-positive detection results (DNA or RNA viruses) were defined by coverage of three or more non-overlapping regions in the genome. A positive detection was reported for a given species or genus when RMP was ≥5 or when RPM-r was ≥5. RPM-r was defined as RPM corresponding to a given species or genus in the clinical sample divided by RPM in the negative control (Miller et al., 2019). To minimize cross-species misalignments among closely related microorganisms, we discounted the RPM of a species or genus that appeared in non-template controls and shared a genus or family designation; a penalty of 5% was used for species (Zaharia et al., 2021).
SPSS 19 (IBM Corporation) was used to perform all analyses. Clinical composite diagnosis and determination of microbiological etiology were regarded as reference standards. At the pathogen level, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were calculated using standard formulas for proportions. Wilson’s method was used to determine 95% confidence intervals for these proportions. McNemar’s test was used to compare diagnostic performance between CMTs and mNGS. All tests were two-tailed. A p-value of <0.05 was considered statistically significant. Note that some children with multiple microbial infections had multiple class labels for this study (bacteria, viruses, fungi, and atypical pathogens). We report sensitivity, specificity, accuracy, and positive predictive value as performance measurements to permit direct comparisons between mNGS and CMTs.
Among 594 eligible patients, 489 were excluded because they did not receive mNGS. Thus, we enrolled 103 (68 male and 35 female) patients. Their mean age was 4.5 years. Their clinical characteristics are shown in Table 1 . The main clinical symptoms were as follows: fever, cough/sputum, wheezing, dyspnea, and hemoptysis. There were 22 cases (21.4%) admitted to an intensive care unit, 39 cases (38%) had recurrent respiratory infections, 54 cases (52.4%) had atelectasis/consolidation, 12 cases (11.7%) had pleural effusion, 4 cases (3.9%) had emphysema/mediastinum, and 6 cases (5.8%) had bronchiectasis. Bronchoscopy revealed 42 cases (40.8%) with poor ventilation, 8 cases (7.8%) with sputum embolus, 5 cases (4.9%) with bronchial mucosal necrosis, and 6 cases (5.8%) with bronchiectasis. All patients received empirical antibiotic therapy prior to admission.
The patients’ microbiological results are provided in Supplementary Table 3 . Among them, 59.2% (61/103) cases tested positive using CMTs, and 91.3% (94/103) tested positive using mNGS (p = 0.000). In clinical comprehensive analysis, 93.2% (96/103) of patients had identified etiology. Figure 2 shows the distribution of pathogens that met the definition of infection. Respiratory syncytial virus (26 cases), cytomegalovirus (25 cases), and parainfluenza virus (20 cases) were the top three viral infections. Haemophilus influenzae (12 cases), Streptococcus pneumoniae (6 cases), and Pseudomonas aeruginosa (6 cases) were the top three bacterial infections. M. pneumoniae (23 cases) was the most frequently detected atypical pathogen. Fungi were also identified, including Aspergillus spp., Pneumocystis jirovecii, and Candida albicans. As shown in Table 2 , there were 52 cases (50.5%) with monomicrobial infection and 44 cases (45.8%) with polymicrobial infection (30 cases were two-microbial infections, 13 cases were three-microbial infections, and 1 case was a four-microbial infection). There were seven cases with unidentified etiology (one patient was positive for Circovirus in blood mNGS but was not clinically considered to be infected, three cases were clinically considered to be viral pneumonia, and three cases were clinically considered to be bacterial pneumonia). Among 52 patients with monomicrobial infection, 27 cases (51.9%, 27/52) were detected using CMTs, while 48 cases (92.3%, 48/52) were detected using mNGS. Among 44 polymicrobial infections, 6 cases (13.6%, 6/44) were detected using CMTS, while 29 cases (65.9%, 29/44) were detected using mNGS. For single and mixed-microbial infections, the detection rate of mNGS was higher than that of CMTs (p = 0.000). The most common mixed infections were bacterial and viral. Staphylococcus aureus, H. influenzae, rhinovirus, cytomegalovirus, parainfluenza virus, and fungi were more likely to be associated with polymicrobial infections.
The diagnostic performance of CMTs and mNGS varied significantly among the different types of pathogens ( Table 3 ). For bacterial detection, the diagnostic sensitivity (88.6% [78.0%–99.1%] vs. 25.7% [11.2%–40.1%], p < 0.001) and accuracy (87.4% [81.0%–93.8%] vs. 70.9% [62.1%–79.6%], p < 0.001) of mNGS were significantly higher than those of CMTs (95% confidence intervals shown). The positive predictive value (PPV) of mNGS was 77.5% (64.6%–90.4%) and the negative predictive value [NPV] was 93.7% (87.6%–99.7%). However, mNGS did not differ significantly from CMTs in the diagnosis of common bacterial infections, such as S. pneumoniae (p = 0.625), S. aureus (p = 0.219), and H. influenzae (p = 0.146). For virus detection, the sensitivity (100% [100%–100%]) and accuracy (86.4% [79.8%–93.2%]) of mNGS were slightly higher than those of CMTs (48.4% [36.2%–60.7%] and 68.0% [58.9%-77.0%], respectively). The PPV of mNGS was lower than that of CMTs (82.1% [73.5%–90.6%] vs. 100% [100%–100%]). The PPV of mNGS varied greatly for the different viruses. Apart from influenza virus (p = 0.125), the diagnostic performance of mNGS was significantly different from that of the CMTs (p < 0.01). For fungi detection, the diagnostic sensitivity of mNGS was higher than that of CMTS (87.5% [64.6%–100%] vs. 25.0% [0%–55%], p < 0.01) but the specificity (83.2% [75.6%–90.7%] vs. 97.9% [95.0%–100%]), accuracy (83.5% [76.3%–90.7%] vs. 92.2% [87.1%–97.4%]), and PPV (30.4% [11.6%–49.2%] vs. 50.0% [10.0%–99.0%]) were lower than those of CMTs. The detection sensitivity of mNGS for P. jirovecii was 100% (100%–100%), which could be verified by CMTs, such as hexamethyltetramine silver staining, implying a sensitivity of 0%. However, the PPV of mNGS for P. jirovecii was relatively low (33.3% [6.7%–60%]). For Aspergillus spp., the diagnostic performances of mNGS and CMTs were not significantly different (p = 0.219). The specificity and accuracy of both methods were higher than 96.1%; however, mNGS had higher sensitivity than CMTs (75% [32.6%–100%] vs. 50.0% [10%–99%]). There were no significant differences in the detection of atypical pathogens between mNGS and CMTs (p = 0.077). However, for M. pneumoniae, the sensitivity of mNGS was lower than that of CMTs (42.1% [20.0%–64.3%] vs. 100% [100%–100%], p = 0.001).
Pneumonia is one of the most common causes of hospitalization for infection in children, and one of the most important causes of their morbidity and mortality (Liu et al., 2015). With extensive use of antibiotics, continuous expansion of the pathogen spectrum, and increasing numbers of hard-to-diagnose infections, it is increasingly difficult to identify the etiology of pneumonia. Relevant literature shows that comprehensive conventional methods do not find pathogens in up to 60% cases (Schlaberg et al., 2017). For patients with severe pneumonia, a long clinical course, the empirical use of antibiotics, and low immunity, CMTs are far from meeting the clinical need for etiology diagnosis; this may lead to the failure of therapy and the overuse of antibiotics. The bronchoalveolar lavage technique can be used to obtain cells and solutions from the lower respiratory tract. It is performed more easily and safely as the technique matures. In clinical practice, for patients with severe illness or suspected mixed infection, clinicians may examine several pathogens at the same time. However, they must verify these pathogens based on their own experience. By contrast, mNGS can detect all possible pathogens for clinicians’ judgment, which can save patients’ time and money. In recent years, BALF mNGS has become a breakthrough application for the diagnosis and treatment of infectious lung diseases. mNGS has potential advantages in terms of speed and sensitivity for detecting lung diseases (Langelier et al., 2018; Li et al., 2018; Miao et al., 2018). Our study showed that, compared with CMTs, mNGS had a significant advantage in its detection rate of pathogens (91% vs. 59%, p = 0.000), even though all patients had used antibiotics. These results are consistent with the conclusions of Miao et al. (2018). mNGS was also superior to CMTs in diagnosing monomicrobial infections (92% vs. 52%, p = 0.000) and polymicrobial infections (66% vs. 14%, p = 0.000). Bacteria and viruses are pathogens commonly found in clinical settings. Our results showed that S. aureus, H. influenzae, rhinovirus, cytomegalovirus, parainfluenza virus, and fungi are more likely to be associated with polymicrobial infections, which suggests the advantages of mNGS in the diagnosis of mixed infections (Fang et al., 2020). Because mNGS can detect almost all microbes in BALF, the technique strongly support improvements in clinical intervention. Many studies (Wang et al., 2019a; Chen et al., 2021) have confirmed the superiority of mNGS in the diagnosis of pulmonary mixed infections and the identification of etiology. A retrospective cohort study (Quah et al., 2018) found that the proportion of respiratory viruses in the pathogen spectrum of severe pneumonia has increased. In our study, 67 cases (70%) had viral infections, of which 28 cases were single infections and 39 were co-infections. Common viruses with high detection rates were respiratory syncytial virus, cytomegalovirus, rhinovirus, influenza virus, parainfluenza virus, and bocavirus. The sensitivity and accuracy of mNGS were higher than those of CMTs for the diagnosis of viral infections ( Table 3 ). Owing to the difficulty of viral culture and the high rate of false positives in nucleic acid detection, it can be difficult to determine the etiology of the viruses identified (Ren et al., 2018). The relative abundance and read ratios of mNGS samples, relative to the negative control, may provide some clues for the determination of viral infections. However, in clinical practice, DNA testing alone may miss some RNA viruses, resulting in a decreased detection rate (Zhang et al., 2020). Messenger RNA of DNA viruses, detected in mNGS RNA testing, may provide clues regarding active transcription (Graf et al., 2016). Thus, performing both mNGS DNA and RNA testing is valuable in diagnosing the etiology of pneumonia. The unbiased nature of mNGS is useful for the detection of new and variant viruses (Chen et al., 2020), evolutionary tracing (Lu et al., 2020), and strain identification (Qian et al., 2021), as well as for guiding epidemiological investigations, public health research, and epidemic prevention and control during infectious disease outbreaks (Deurenberg et al., 2017). mNGS played a key role in the rapid identification of pathogens in the outbreak of the novel coronavirus pneumonia in late 2019 (Chen et al., 2020; Ren et al., 2020). Our study revealed that the diagnostic performance of mNGS varied for different pathogens ( Table 3 ). For detecting bacterial infections, the overall sensitivity (88.6% vs. 25.7%), accuracy (87.4% vs. 70.9%), PPV, and NPV of mNGS were higher than those of CMTs. However, there was no significant difference between mNGS and CMTs for the diagnosis of S. pneumoniae, S. aureus, and H. influenzae (p > 0.05). This differs slightly from previous studies (Xie et al., 2019) and may be related to the fact that these bacteria are clinically common in pediatric pneumonia, where empiric therapy is effective. mNGS has the advantage of being able to detect more pathogens. The sensitivity and accuracy of mNGS were higher than those of CMTs for the diagnosis of viral infection (p < 0.01); however, the PPV varied among different viruses. In particular, our study revealed that the parallel detection of DNA and RNA can determine the activity of DNA viruses and detect RNA viruses. For fungal infections, the overall sensitivity of mNGS was higher than that of CMTs; however, the specificity and accuracy were lower than those of CMTs. The total NPV of mNGS was 98.7% (96.3%–100%). Positive results for P. jirovecii, such as staining microscopic examination and PCR, are important diagnostic criteria; however, the detection rates are low. P. jirovecii was detected in 12 patients using mNGS. In comprehensive clinical analysis, only four cases were considered to be pneumocystis pneumonia. These cases were all infants, in whom the course of disease was >2 weeks and the effect of conventional treatment was not good. This may be related to the fact that fungal infection was secondary to low immunity after infection. The remaining eight patients recovered without antifungal therapy; P. jirovecii was probably colonized in the lower respiratory tract. Unfortunately, our study was not further validated using Gomori methenamine silver staining, which may have influenced the comparison of the two testing methods. Recent studies (Wang et al., 2019b; Lin et al., 2022) have shown that mNGS has good diagnostic performance in the detection of pneumocystis. The identification of Aspergillus spp. by mNGS remains a challenge because of the difficulty of extracting DNA from its thick polysaccharide cell walls (Bittinger et al., 2014). Three cases of severe pneumonia with Aspergillus spp. etiology were reported by He et al. (2019). mNGS results indicated Aspergillus spp., and the patients were adjusted for antifungal treatment; their conditions improved. Thus, the accuracy of mNGS for the detection of Aspergillus spp. is suggested. In contrast with these results, our study did not show an advantage of mNGS for the diagnosis of Aspergillus infections. However, the small number of cases of fungal pneumonia in this study likely introduced biases in the calculation of diagnostic performance. Although the specificity and accuracy of mNGS were high for M. pneumoniae diagnosis, the sensitivity was significantly lower than that of CMTs. In our study, mNGS did not show an advantage for the diagnosis of M. pneumoniae infections. The diagnosis of M. pneumoniae was confirmed by serology in the early stage of the disease (before admission to our hospital); the detection rate may have decreased after treatment. For the detection of M. pneumoniae, it has been reported that combined detection methods can improve the specificity and sensitivity of diagnosis and reduce false-negative and false-positive rates. M. pneumoniae cannot be reliably diagnosed using only a single test (Loens and Ieven, 2016; Tang et al., 2021). Overall, mNGS can improve the detection rate of pathogens and mixed infections in pediatric pneumonia. The diagnostic utility of mNGS differs for different pathogens. For fungi and M. pneumoniae, the CMT approach may need to be combined to improve diagnostic performance. mNGS is valuable as a complement to CMTs, especially when the clinician does not have a presumed pathogen or the local laboratory is without complete CMTs. This study has some limitations. First, the sample size was small, especially for fungal pneumonia. Second, P. jirovecii lacked further validation by Gomori methenamine silver staining; partial PCR failed to evaluate the diagnostic value of CMTs and mNGS. Third, at the time of this study, our hospital had had real-time PCR detection items for some pathogens, including 13 respiratory pathogens; there were no commercial offerings based on real-time multiplex PCR for the detection of community or hospital pathogens. Therefore, we did not compare the performances of multiplex PCR and mNGS for the detection of different pathogens. However, the use of real-time multiplex PCR assays is based on the clinician’s belief that a patient is infected with one or more of these pathogens; it ignores rare and unknown pathogens. Finally, the interpretation of mNGS results, to a certain extent, depended on the subjective judgment of the clinician, which may have led to bias.
The data presented in the study are deposited in the NCBI SRA repository, accession number SRR21425639~SRR21425741.
JL: Designed the study and revised and approved the final version. WD: Drafted the initial manuscript, retrieved pediatric literature, and edited the table and reference list. YW: Participated in formal analysis. HX: Participated in data analysis. All authors contributed to the article and approved the submitted version.
All the authors would like to express their appreciation to all the patients for their cooperation.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648331 | Josef Finsterer | Small fiber neuropathy as a complication of SARS-CoV-2 vaccinations | 22-07-2022 | Adverse reaction,COVID-19,neuropathy,pain,SARS-CoV-2,side effect,small fibers,vaccination | Generally, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccinations are not free of side effects. A rarely reported adverse reaction to SARS-CoV-2 vaccinations is small fiber neuropathy (SFN). Here, we present three patients with SFN after the second dose of messenger ribonucleic acid-based SARS-CoV-2 vaccines. Data for this study were collected via the self-made platform “Pubbly” for reporting side effects of SARS-CoV-2 vaccinations. Three patients with post-SARS-CoV-2 vaccination SFN were identified: a 40 yo Caucasian female (patient 1), a 52 yo Caucasian female (patient 2), and a 32 yo Caucasian female (patient 3). Patient 1 complained about fatigue, dizziness, flushing, palpitations, diarrhea, muscle weakness, and gait disturbance 10 days after the second Pfizer jab. Patient 2 reported dizziness, balance problems, brain fog, palpitations, dysphagia, and sleep problems. Patient 3 complained about profound fatigue, brain fog, vertigo, pre-syncopal sensations, hair loss, chest pain, dyspnea, palpitations, paresthesias, irregular menstrual cycles, muscle weakness, and hives 1 day after the second Moderna dose. All three patients underwent skin biopsy upon which SFN was diagnosed. Patient 1 profited from immunoglobulins, but patient 2 did not require any treatment. Symptoms in patient 3 resolved upon symptomatic treatment. Despite treatment, patient 1 did not completely recover. SFN can be a rare side effect of SARS-CoV-2 vaccinations. Post-SARS-CoV-2 vaccination SFN can be mild or severe and may or may not require treatment. Post-SARS-CoV-2 vaccination SFN is most likely immune-mediated as it responds to intravenous immunoglobulins. | Small fiber neuropathy as a complication of SARS-CoV-2 vaccinations
Generally, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccinations are not free of side effects. A rarely reported adverse reaction to SARS-CoV-2 vaccinations is small fiber neuropathy (SFN). Here, we present three patients with SFN after the second dose of messenger ribonucleic acid-based SARS-CoV-2 vaccines. Data for this study were collected via the self-made platform “Pubbly” for reporting side effects of SARS-CoV-2 vaccinations. Three patients with post-SARS-CoV-2 vaccination SFN were identified: a 40 yo Caucasian female (patient 1), a 52 yo Caucasian female (patient 2), and a 32 yo Caucasian female (patient 3). Patient 1 complained about fatigue, dizziness, flushing, palpitations, diarrhea, muscle weakness, and gait disturbance 10 days after the second Pfizer jab. Patient 2 reported dizziness, balance problems, brain fog, palpitations, dysphagia, and sleep problems. Patient 3 complained about profound fatigue, brain fog, vertigo, pre-syncopal sensations, hair loss, chest pain, dyspnea, palpitations, paresthesias, irregular menstrual cycles, muscle weakness, and hives 1 day after the second Moderna dose. All three patients underwent skin biopsy upon which SFN was diagnosed. Patient 1 profited from immunoglobulins, but patient 2 did not require any treatment. Symptoms in patient 3 resolved upon symptomatic treatment. Despite treatment, patient 1 did not completely recover. SFN can be a rare side effect of SARS-CoV-2 vaccinations. Post-SARS-CoV-2 vaccination SFN can be mild or severe and may or may not require treatment. Post-SARS-CoV-2 vaccination SFN is most likely immune-mediated as it responds to intravenous immunoglobulins.
Small fiber neuropathy (SFN) is a disorder of the peripheral nervous system (PNS), characterized by affection of small nerve fibers (myelinated Aδ fibers, non-myelinated C-fibers) which conduct in an anterograde or retrograde manner either sensory (somatic) or autonomic information.[1] Clinically, SFN usually manifests as chronic pain of uncertain origin or autonomic dysfunction.[12] Causes of SFN are primary (genetic)[3] or secondary (metabolic, infectious, toxic, immune, paraneoplastic).[2] Although SFN has been occasionally reported as a complication of a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection [coronavirus disease 19 (COVID-19)][45] or as a manifestation of the post-(long)-COVID syndrome.[67] SFN has been only rarely reported as an adverse reaction to SARS-CoV-2 vaccinations.[8] Here, we present three patients with SFN following SARS-CoV-2 vaccinations with messenger ribonucleic acid (mRNA)-based vaccines.
Patient 1 is a 40 yo Caucasian female with an uneventful previous history and without a current medication who developed side effects 10 days after the second dose of an mRNA-based SARS-CoV-2 vaccine (Pfizer). Her history was negative for COVID-19 prior to the vaccinations. She particularly complained about severe fatigue, dizziness, flushing, palpitations, diarrhea, muscle weakness, and gait disturbance. On admission, blood pressure was elevated. Immune-mediated dysautonomia triggered by the vaccination was suspected why a skin punch biopsy was carried out, which revealed a reduced intra-epidermal nerve fiber density (IENFD), suggesting SFN. Initially, she was treated with clonazepam, diltiazem, loratadine, steroids, and famotidine. Because the clinical manifestations of SFN hardly resolved upon this treatment, intravenous immunoglobulins (IVIGs) were added with a beneficial effect. Patient 2 is a 52 yo Caucasian female with an uneventful previous history and without taking any current medication who developed dysautonomia 17 days after the second dose of an mRNA-based SARS-CoV-2 vaccine (Moderna). She complained about dizziness, balance problems, brain fog, palpitations, dysphagia, and sleep problems. Her history was negative for COVID-19 prior to the vaccinations. Ambulatory work-up for dysautonomia by a skin punch biopsy revealed SFN. She did not receive any treatment as her symptoms spontaneously resolved. Patient 3 is a 32 yo Caucasian female with an uneventful previous history who developed profound fatigue, brain fog, vertigo, pre-syncopal sensations, hair loss, chest pain, dyspnea, palpitations, paresthesias, irregular menstrual cycles, muscle weakness, and hives 1 day after the second dose of an mRNA-based SARS-CoV-2 vaccine (Pfizer). During hospitalization, SFN was suspected and confirmed upon skin punch biopsy showing reduced IENFD. Upon symptomatic treatment, most of her complaints resolved.
This case series shows that SARS-CoV-2 vaccinations can be complicated by SFN. Clinical presentation of post-SARS-CoV-2 vaccination SFN is not at variance from clinical manifestations of SFN because of other causes. Post-SARS-CoV-2 vaccination SFN is presumably immune-mediated as it responds favorably to IVIG. The study is important for the family physician because he is most frequently the first health care professional who sees the patient and because he needs to take SFN as a complication of SARS-CoV-2 vaccinations into consideration as a differential. Generally, SFN is due to primary (genetic) or secondary causes. Genetic SFN is because of mutations in a number of genes, such as GLA (Fabry disease), TTR (transthyretin-related amyloidosis), or SNCA (alpha-synucleinopathy) and many others. Secondary causes of SFN prevail and include, for example, diabetes, renal failure, thyroid dysfunction, hypovitaminoses, acute infections (SARS-CoV-2,[8] borreliosis[9]), vaccinations (rabies, varicella, human papillomavirus, lyme, SARS-CoV-2),[810] auto-immune disease,[1112] pure autonomic failure because of alpha-synuclein deposition,[13] sarcoidosis,[14] Sjögren syndrome,[15] Parkinson’s disease,[16] and many others.[17] SFN may go along with or without affection of large motor or sensory fibers.[18] Thus, SFN can be associated with polyneuropathy (neuropathy of nerves built up of large fibers) but usually occurs without it. Length-dependent SFN and non-length-dependent SFN are delineated.[19] Generally, SFN manifests clinically as chronic focal or regional pain [complex regional pain syndrome (CRPS)] or with autonomic manifestations, such as fatigue, cognitive impairment, over-sensitivity to light, sicca syndrome, postural tachycardia syndrome, syncope, near-syncope, sudo-motor dysfunction (dyshidrosis), reduced heart rate variability, reduced blood pressure variability, disturbed thermo-regulation, urinary retention, constipation, or impotency. The clinical presentation of the three index cases is in line with these clinical manifestations as they had pain or dysautonomia. Work-up for SFN includes quantitative sensory testing, nerve conduction studies to exclude large fiber neuropathy, micro-neurography, sensory stimulation tests, autonomic testing (deep breathing, Valsalva maneuver, tilt test, cerebral blood flow velocity measurements, quantitative sudo-motor axon reflex test, corneal confocal microscopy, pain-related evoked potentials, and proximal or distal skin biopsy.[720] Skin biopsy of the proximal or distal lower limbs is by far the most widely applied technique and the golden standard to diagnose SFN. Skin biopsies of the three index patients were in line with previously reported findings of skin biopsies including reduced IENFD.[37] Treatment of SFN can be symptomatic, pathogenesis-related, or causal. Causal treatment is available for most of the secondary SFNs. Symptomatic treatment includes systemic pain killers, local analgesics (local anesthetics, capsaicin ointment), transcutaneous electrical nerve stimulation, or sympathectomy. Autonomic disturbance responds favorably to symptomatic treatment. Often, combinations of causal/pathogenesis-related and symptomatic therapies are required. Post-SARS-CoV-2 vaccination SFN has been previously reported in a single patient, a 57 yo female who presented 1 week after receiving the second dose of the Pfizer SARS-CoV-2 vaccine with sub-acute onset of intense burning dysesthesias in the feet, gradually spreading to the calves and minimally into the hands, unaccompanied by other neurological or constitutional symptoms. There was no known prior COVID-19 exposure. She was not on any medication and denied the use of alcohol.[9] In conclusion, SFN can be a rare side effect of SARS-CoV-2 vaccinations. Post-SARS-CoV-2 vaccination SFN can be mild or severe and may or may not require treatment. Post-SARS-CoV-2 vaccination SFN is most likely immune-mediated as it responds to IVIG.
JF: Literature search, discussion, first draft, critical comments, final approval.
Was obtained.
The study was approved by the institutional review board.
Nil.
There are no conflicts of interest. |
|
PMC9648346 | Mingyue Lv,Hongzhe Cao,Xue Wang,Kang Zhang,Helong Si,Jinping Zang,Jihong Xing,Jingao Dong | Identification and expression analysis of maize NF-YA subunit genes | 07-11-2022 | Maize,NF-YA subunit gene,Biological stress,Abiotic stress,Expression analysis | NF-YAs encode subunits of the nuclear factor-Y (NF-Y) gene family. NF-YAs represent a kind of conservative transcription factor in plants and are involved in plant growth and development, as well as resistance to biotic and abiotic stress. In this study, 16 maize (Zea mays) NF-YA subunit genes were identified using bioinformatics methods, and they were divided into three categories by a phylogenetic analysis. A conserved domain analysis showed that most contained a CCAAT-binding transcription factor (CBFB) _NF-YA domain. Maize NF-YA subunit genes showed very obvious tissue expression characteristics. The expression level of the NF-YA subunit genes significantly changed under different abiotic stresses, including Fusarium graminearum infection and salicylic acid (SA) or jasmonic acid (JA) treatments. After inoculation with Setosphaeria turcica and Cochliobolus heterostrophus, the lesion areas of nfya01 and nfya06 were significantly larger than that of B73, indicating that ZmNFYA01 and ZmNFYA06 positively regulated maize disease resistance. ZmNFYA01 and ZmNFYA06 may regulated maize disease resistance by affecting the transcription levels of ZmPRs. Thus, NF-YA subunit genes played important roles in promoting maize growth and development and resistance to stress. The results laid a foundation for clarifying the functions and regulatory mechanisms of NF-YA subunit genes in maize. | Identification and expression analysis of maize NF-YA subunit genes
NF-YAs encode subunits of the nuclear factor-Y (NF-Y) gene family. NF-YAs represent a kind of conservative transcription factor in plants and are involved in plant growth and development, as well as resistance to biotic and abiotic stress. In this study, 16 maize (Zea mays) NF-YA subunit genes were identified using bioinformatics methods, and they were divided into three categories by a phylogenetic analysis. A conserved domain analysis showed that most contained a CCAAT-binding transcription factor (CBFB) _NF-YA domain. Maize NF-YA subunit genes showed very obvious tissue expression characteristics. The expression level of the NF-YA subunit genes significantly changed under different abiotic stresses, including Fusarium graminearum infection and salicylic acid (SA) or jasmonic acid (JA) treatments. After inoculation with Setosphaeria turcica and Cochliobolus heterostrophus, the lesion areas of nfya01 and nfya06 were significantly larger than that of B73, indicating that ZmNFYA01 and ZmNFYA06 positively regulated maize disease resistance. ZmNFYA01 and ZmNFYA06 may regulated maize disease resistance by affecting the transcription levels of ZmPRs. Thus, NF-YA subunit genes played important roles in promoting maize growth and development and resistance to stress. The results laid a foundation for clarifying the functions and regulatory mechanisms of NF-YA subunit genes in maize.
The nuclear factor-Y (NF-Y) transcription factor exists widely in eukaryotes and is also known as hemo-activator protein (HAP) (Thirumurugan et al., 2008; Petroni et al., 2012). NF-Y can bind to CCAAT-boxes in promoter sequences; consequently, it is also called CCAAT-binding factor (CBF) (Laloum et al., 2013). NF-Y is a large gene family composed of NF-YA (CBFB/HAP2), NF-YB (CBFA/HAP3) and NF-YC (CBFC/HAP5) subunits (Nardini et al., 2013). NF-Y is usually located in the nucleus and is evolutionarily conserved (Mantovani, 1999). In animals, NF-YA, NF-YB and NF-YC subunits are encoded by three single genes, and the three subunits function in the form of heterologous trimers (Benatti et al., 2008). In plants, the three subunits are encoded by more than 10 genes, and they can perform their functions independently (Petroni et al., 2012). The CBFB_NF-YA domain is the core conserved domain of the NF-YA family. The N-terminal of this domain can bind to NF-YB and NF-YC subunits, and the C-terminal can bind to DNA CCAAT-boxes (Quach et al., 2015). Compared with the rapid progress in mammalian and yeast (Saccharomyces cerevisiae) NF-Y protein-related research, research progress on the plant NF-Y family has been slow (Liang et al., 2012). Until now, this research has been limited to the preliminary bioinformatics comparisons between A. thaliana and related plant species, as well as the gene expression and function analyses. The single subunit of NF-Y is widely involved in plant growth and development, such as controlling gametogenesis, embryo and plant development (Mu et al., 2013), abscisic acid (ABA) signal transduction (Yu et al., 2021), flowering cycle regulation (Hwang et al., 2019), primary root elongation (Ballif et al., 2011; Zhou et al., 2020), blue light response (Warpeha et al., 2007) and photosynthesis (Tokutsu et al., 2019) as well as stress responses, including to abiotic stresses such as drought, high temperature and salt (Li et al., 2021). The NF-YA subunit genes are involved in multiple processes in the plant lifecycle. In Arabidopsis thaliana, NF-YA3 and NF-YA8 genes mediate cell differentiation and embryo formation through the ABA signaling pathway during early embryonic development (Mu et al., 2013). NF-YA1, NF-YA5, NF-YA6 and NF-YA9 are involved in the development of gametes, embryos and seeds (Mu et al., 2013). AtNF-YA5 is regulated by miR169, thereby improving the resistance of Arabidopsis to drought stress (Li et al., 2008). The overexpression of OsNF-YA7 improves the drought tolerance of rice through an ABA-independent pathway (Lee et al., 2015). In potatos (Solanum tuberosum cv. ‘Desiree’), NF-YAs responds to drought by regulating the number of chlorophylls, stomatal conductance and photosynthesis (Li et al., 2021). PtNF-YA9 plays an important role in the drought resistance of Populus trichocarpa as a positive regulator of stress resistance (Lian et al., 2018). There are five NF-Y genes in tomato (Solanum lycopersicum) that play roles in tomato fruit ripening (Li et al., 2016). In soybean (Glycine max), GmNFYC14 forms heterotrimer with GmNF-YA16 and GmNFYB2, activates GmPYR1 mediated ABA signaling pathway and regulates soybean stress tolerance (Yu et al., 2021). In maize (Zea mays), ZmNFYB16 can form a heterotrimer with ZmNFYC17 and ZmNFYA01, and the heterotrimer binds to CCAAT cis-acting elements in the promoter region of stress response and growth-related genes through the ZmNFYA01 subunit, regulating the expression of multiple genes related to stress resistance and growth, thereby improving the drought resistance of plants (Wang et al., 2018). ZmNFYA03 promotes early flowering by binding to the FT-like12 promoter in maize (Su et al., 2018). Plants have developed complex mechanisms to protect themselves against pathogens. Pathogenesis-related (PR) genes are the key elements of these mechanisms, and activated in response to pathogen attacks. They regulate production of several proteins, peptides or compounds which are toxic to pathogens or prevent pathogen infections where they start (Xie et al., 2010). The PR factors are thermostable, protease-resistant proteins of ~5–43 kDa which are expressed in all plant organs (Zribi, Ghorbel & Brini, 2021). To date, an overall study of the maize NF-YA subunit gene family has not been reported, and the number, physicochemical properties and functions of maize NF-YA subunit genes are not clear. In this study, the maize NF-YA subunit genes were identified using bioinformatics methods, and their phylogenetic relationship, conserved domains, tissue specificity and gene expression patterns under biotic and abiotic stresses, such as salicylic acid (SA) and jasmonic acid (JA) treatments as well as disease resistance, were clarified. This would provide the foundation for elucidating the functions and regulatory mechanisms of maize NF-YA subunit genes.
The information and amino acid sequences of NF-YA subunit genes in maize, rice (Oryza sativa) and Arabidopsis were downloaded from MaizeGDB (http://www.maizegdb.org/), RGAP (http://rice.plantbiology.msu.edu/) and TAIR (https://www.arabidopsis.org/), respectively. The amino acid sequences were aligned using Clustal X software (Larkin et al., 2007). The aligned results were imported into MEGA 7.0 software (Kumar, Stecher & Tamura, 2016), and a phylogenetic tree was constructed using the maximum-likelihood method. The chromosome location and the annotated information regarding the gene structure (including gene length, 5′-UTR, 3′-UTR and the distribution of each intron and exon) of maize NF-YA subunit genes were obtained from MaizeGDB, and their chromosome mapping was performed using RIdeogram software (https://cran.r-project.org/web/packages/RIdeogram/index.html). The gene structure map was drawn using IBS software (Liu et al., 2015). The conserved domains of maize NF-YAs were analyzed using online software SMART (http://smart.embl-heidelberg.de/) and Pfam (http://pfam.xfam.org/), and the domain analysis maps were constructed using IBS software. In accordance with the amino acid sequences of NF-YAs in maize, the subcellular localization was analyzed and predicted using Plant-mPLoc software (Chou & Shen, 2010). For the predicted protein-protein interaction (PPI) analysis, all of the NF-YA amino acid sequences were searched using the STRING database version 11.5 (https://cn.string-db.org/) (Szklarczyk et al., 2021). A “confidence score” of STRING > 0.7 (high confidence) between proteins was used (Mei et al., 2021). The interaction networks of proteins generated using STRING were constructed to determine the relationships of proteins with NF-YAs.
Using the SRA database in NCBI (https://www.ncbi.nlm.nih.gov/), the RNA-seq data of 31 maize tissues (such as Seed_5_days_after_pollination, Endosperm_25_days_after_pollination, Seed_10_days_after_pollination, etc.) and those obtained under both abiotic stresses, such as high temperature (14 days of maize seedlings cultured at 50 °C for 4 h), low temperature (14 days of maize seedlings cultured at 5 °C for 16 h), salt (14 days of maize seedlings irrigated with 300 mM NaCl for 20 h), ultraviolet (14 days of maize seedlings irradiated by ultraviolet lamp for 2 h), drought (14 days of maize seedlings dried filter paper covered for 4 h) and Fusarium graminearum infection (200 μL of F. graminearum conidial suspension was inoculated onto the internodes of maize at ten-leaf stage after punctured with a pipette), were downloaded. Using Hisat2 software (http://daehwankimlab.github.io/hisat2/), RNA-seq data was aligned to the reference genome of maize. Using Cufflinks software (http://cole-trapnell-lab.github.io/cufflinks/), the gene expression value expressed as Fragments Per Kilobase of exon per Million fragments mapped reads (FPKM) was calculated using standardized parameters of gene length and number of reads. The expression heat map of the NF-YA family genes in maize was constructed using HemI software (Deng et al., 2014).
The preserved seeds of maize inbred line B73, Setosphaeria turcica and Cochliobolus heterostrophus were from the Mycotoxin and Molecular Plant Pathology Laboratory, Hebei Agricultural University. F. graminearum strain PH-1 was provided by Prof. Mingguo Zhou at Nanjing Agricultural University. The ZmNFYA01 Mu insertion mutant nfya01 (Chr1, Insertion site 16041887, V4.0) and the ZmNFYA06 Mu insertion mutant nfya06 (Chr1, Insertion site 268308575, V4.0) were obtained from the ChinaMu Project (http://chinamu.jaas.ac.cn/) (Liang et al., 2019) and nfya01 and nfya06 plants used in this study were confirmed by PCR and quantitative real-time PCR (qRT-PCR) (Figs. S1 and S2). The primers for PCR and qRT-PCR identification are listed in Tables S1 and S2. All the maize seeds were soaked in sterile water for approximately 24 h and potted in the mixture of vermiculite and nutrient soil at 1:1. The plants were cultured and grown in an artificial climate chamber with light for 14 h and darkness for 10 h at a temperature of 25–28 °C, and a humidity of 50–60%. The plants were irrigated once every 3 to 5 days, and the roots were irrigated with nutrient solution to ensure sufficient nutrition for the plants after reached the three-leaf stage. S. turcica and C. heterostrophus were inoculated on Potato Dextrose Agar (PDA, 200 g/L potatos, 20 g/L glucose, 12 g/L agar) plates and grown in a 25 °C incubator. F. graminearum was grown on PDA plates at 28 °C for 5 d, transferred to the carboxymethylcellulose sodium medium (CMC, 1.5 g/L CMC-Na, 1 g/L KH2PO4, 1 g/L NH4NO3, 1 g/L yeast powder and 0.5 g/L MgSO4·7H2O) and cultured in a shaker at 25 °C and 200 rpm for 5–7 d. The number of spores in the conidial suspension was counted using a blood cell counting plate, and more than 1 × 106 spores were inoculated onto maize.
In total, 1 L of salicylic acid (SA; 100 μM) and the same volume of jasmonic acid (JA; 100 μM) were sprayed independently and evenly on the aboveground parts of the two groups of maize B73 (five-leaf stage). Leaves were sampled at 0, 3, 9 and 24 h, frozen with liquid nitrogen and stored in a −80 °C refrigerator.
The 4th and 5th leaves of seven-leaf stage maize plants were cut off, and a wound of 1.2 cm in diameter was cut every 8 cm using a disposable syringe needle to facilitate the invasion of pathogenic fungi. Tween 20 was applied to the wound, and cultured S. turcica and C. heterostrophus were punched into the wound to cover the leaf wounds. Two layers of fully wetted filter paper were placed at the bottom of a white culture box. The leaves inoculated with pathogenic fungi were placed into the box, and covered with preservative film. The box had several vents in the box and was maintained in the darkness at 25 °C. The filter paper was kept moist, and the changes in the lesions were observed every day. A total of 4 or 5 days after inoculation, the inoculated leaves were stained with trypan blue. The prepared 0.5% trypan blue staining solution and the leaves of an appropriate size were placed in a 50-mL centrifuge tube and boiled for 15 min to stain the necrotic cells. Then, 100 g chloral hydrate was dissolved into 40 mL water to make the chloral hydrate decolorizing solution. The stained leaves were washed with water to remove the dye solution of trypan blue, and then placed into chloral hydrate solution and shaken for 2 to 3 days to discoloring fully. Image J software (https://imagej.nih.gov/ij/) was used to measure the lesion area caused by fungal infection. The experiment was repeated three times. GraphPad Prism 8 software (https://www.graphpad.com/) was used to calculate the standard deviation (SD), and a Student’s t-test analysis was performed.
A Plant RNA kit (OMEGA, Norcross, GA, USA) was used to extract sample RNAs, and a Reverse Transcription and cDNA Synthesis Kit (Clontech, Mountain View, CA, USA) was used to synthesize cDNA. Specific qRT-PCR primers for the internal reference gene UBQ9 (Jin et al., 2019) and maize NF-YA subunit genes (Table S2) synthesized by Beijing Bomede Biotechnology Co., Ltd. were used to analyze the expression of NF-YA subunit genes with two hormone treated maize inbred line B73 plant samples collected at different times as templates. The qRT-PCR primers of the internal reference gene UBQ1 (da Silva Santos et al., 2021) and ZmPRs (Table S3) were used for the expression analysis of ZmPRs with the leaf cDNAs of maize B73, nfya01 and nfya06 at the five-leaf stage as templates. The reaction system was as follows: 7 μL of 2× M5 HiPer SYBR Premix EsTaq (with Tli RNaseH; TaKaRa, Dalian, China), 1 μL of cDNA template, 0.5 μL of forward primer, 0.5 μL of reverse primer and 5 μL of ddH2O. A fluorescence quantitative PCR instrument (CFX96 Real-time PCR Detection; BioRad, Hercules, CA, USA) was used for a total of 40 cycles, each of which was 95 °C for 30 s, 95 °C for 5 s and 60 °C for 30 s. Each qRT-PCR reaction was repeated three times, and the gene expression level was analyzed using the Ct value method (2−ΔΔCt). The SD of three replicates was calculated using GraphPad Prism 8 software, and the Student’s t-test analysis was performed.
In total, 16 maize NF-YA subunit genes were obtained from MaizeGDB and named ZmNFYA01–16 in accordance with their chromosomal distribution. The 16 corresponding proteins differed in the number of amino acid (aa), relative molecular mass and isoelectric point. The lengths of the 16 NF-YA amino acid sequences were between 90 and 742 aa, with most being approximately 300 aa. The predicted isoelectric points indicated that most of these proteins were alkaline, with only the proteins encoded by ZmNFYA15 and ZmNFYA16 being acidic (Table 1). The 16 maize NF-YA subunit genes, 10 Arabidopsis NF-YA subunit genes and 11 rice NF-YA subunit genes could be divided into three groups: I, II and III. In maize, there were 6, 6 and 4 NF-YA subunit genes in I, II and III, respectively. Among them, ZmNFYA01 was orthologous to OsNFYA2, ZmNFYA06 was orthologous to OsNFYA5, ZmNFYA09 was orthologous to OsNFYA6, ZmNFYA10 was orthologous to OsNFYA4, and ZmNFYA14 was orthologous to OsNFYA1 (Fig. 1). The maize NF-YA subunit genes were unevenly distributed among the chromosomes. Chromosome 1 contained the largest number (seven) of maize NF-YA subunit genes, ZmNFYA01–7. Chromosome 5 contained four maize NF-YA subunit genes: ZmNFYA11–14. Chromosome 2 contained two maize NF-YA subunit genes, ZmNFYA08 and ZmNFYA09. Chromosomes 3, 7 and 10 contained only one maize NF-YA subunit gene each, whereas chromosomes 4, 6, 8 and 9 did not contain any maize NF-YA subunit genes (Fig. S3). The lengths of the NF-YA subunit gene sequences were quite different, with ZmNFYA03, -07, -11 and -13 having no 5′-UTR and 3′-UTR structures. Introns existed in the 5′-UTRs of 10 genes, and most genes had 4–6 exons, suggesting that they may share the same RNA splicing pattern (Fig. S4). Among the 16 members in the NF-YA family, 13 contained a CBFB_NF-YA domain, ZmNFYA03 contained a reverse transcriptase domain, and ZmNFYA13 had a TATA-binding protein domain. ZmNFYA07 lacked any known domains (Fig. 2A). The CBFB_NF-YA domain plays a key role in binding the NF-YB subunit and specifically binds the CCAAT box. ZmNFYA07 was predicted to be located in the chloroplast and cytoplasm, whereas ZmNFYA16 was predicted to be located in the mitochondrion and nucleus. The 14 remaining maize NF-YAs were predicted to be located in the nucleus (Table 2). Thus, most transcription factors were located in the nucleus, but they also played important roles in the mitochondrion and chloroplast. All NF-YA proteins were predicted to interact with proteins encoded by GRMZM2G180947_P01, GRMZM2G473152_P01 and GRMZM2G444073_P01. In addition, GRMZM2G143450_P01 and GRMZM2G099628_P01 may interact with ZmNFYA16. These two proteins may encode a methionine-tRNA ligase, which indicates that ZmNFYA16 may play a role in translation (Fig. 2B).
The expression levels of the 16 maize NF-YA subunit genes in the same tissues at different developmental stages were significantly different (P < 0.05). The expression of the same gene varied in different tissues at different stages. ZmNFYA01 was highly expressed at 25 d of embryonic development, and its expression levels in most tissues were significantly higher than those of the other genes, indicating that it played important roles in maize growth and development. ZmNFYA14 was highly expressed at 16 d and 25 d of embryonic development. ZmNFYA08 was highly expressed during embryonic development, endosperm development and seed germination. Low expression levels of ZmNFYA02, ZmNFYA07 and ZmNFYA13 were observed in all the tissues we studied (Fig. 3A). The expression of ZmNFYA08 was significantly up-regulated under salt and drought stresses. The expression of ZmNFYA11 was similar under salt and drought treatments. ZmNFYA10 was up-regulated under heat and salt stresses. ZmNFYA02 was significantly down-regulated under heat, salt and UV stresses. The expression levels of ZmNFYA05, ZmNFYA12 and ZmNFYA14 decreased significantly under the heat stress. ZmNFYA07 expression was low and maintained without significant change under the various stresses (Fig. 3B). The expression levels of ZmNFYA01 and ZmNFYA15 were the highest at 0 h after the fungal infection, and then decreased gradually. The expression levels of ZmNFYA02, -04–06, -08, -09, -11, -12 and -14 all showed a trend of first decreasing and then increasing. The expression of ZmNFYA03 and ZmNFYA10 increased first, then decreased and finally increased. The expression of ZmNFYA16 did not change significantly at 0–48 h after infection, but it decreased at 72 h after infection (Fig. 3C). Thus, the expression levels of NF-YA subunit genes in maize varied during F. graminearum infection, indicating that NF-YA subunit genes were involved and played important roles in maize disease resistance.
With the SA treatment, maize NF-YA subunit genes were divided into four categories based on their expression. In the first type, the expression increased first and then decreased. The expression levels of ZmNFYA01, -02 and -15 peaked at 3 h after treatment, and then decreased gradually. The expression levels of ZmNFYA03, -11, -13 and -14 peaked at 9 h after treatment and then decreased. In the second type, the expression decreased first and then increased. The expression levels of ZmNFYA04 and ZmNFYA05 were lowest at 9 h after treatment and then increased slightly. The expression of ZmNFYA12 was lowest at 3 h after treatment and then increased to the initial level. In the third type, the expression of the gene such as ZmNFYA08 and ZmNFYA16 continued to increase was maintained at a high level after the SA treatment. In the fourth type, the expression fluctuated after treatment. For example, the expression levels of ZmNFYA06 and ZmNFYA09 decreased first, then increased and finally decreased. The expression of ZmNFYA10 increased first and then decreased to its lowest level before increasing again (Fig. 4A). Under the JA treatment, the maize NF-YA subunit genes were divided into three categories based on their expression patterns. In the first type, after the JA treatment, the gene expression peaked at 3 h after treatment and then gradually decreased, which included ZmNFYA01, -02, -04, -09, -12, -13, -15 and -16. In the second type, the expression increased first and then decreased before increasing again. The expression of ZmNFYA03, -06, -08 and -10 peaked at 3 h after treatment, decreased to their lowest levels at 9 h and then increased. In the third type, which included ZmNFYA05 and ZmNFYA11, expression decreased after the JA treatment. In addition, the expression of ZmNFYA14 decreased first and then increased before decreasing again (Fig. 4B). Thus, the NF-YA subunit genes may be involved and play important roles in SA and JA signaling pathways.
The lesion areas of nfya01 and nfya06 plants inoculated with S. turcica and C. heterostrophus were significantly larger than those of inbred line B73 (Figs. 5A–5D). Compared with B73, the sensitivity of nfya01 and nfya06 to S. turcica and C. heterostrophus was enhanced, indicating that ZmNFYA01 and ZmNFYA06 positively regulate the disease resistance of maize and provide broad-spectrum resistance to pathogenic fungi, thereby playing important roles in maize resistance to pathogen infection.
Compared with in B73, the expression of ZmPR1–3, -5 and -6 in nfya01 and nfya06 plants were significantly down-regulated (P < 0.05), whereas the expression of ZmPR4, -7 and -10 were significantly up-regulated (P < 0.05) (Figs. 6A and 6B), indicating that ZmNFYA01 and ZmNFYA06 affected the expression of ZmPRs and suggesting that ZmNFYA01 and ZmNFYA06 participated in plant disease resistance by regulating the expression of ZmPRs. The expression patterns of different ZmPRs in nfya01 and nfya06 plants were consistent, indicating that ZmNFYA01 and ZmNFYA06 had similar functions.
The roles of NF-YA subunit genes in the molecular mechanisms involved in maize responses to pathogen infection have been rarely reported. In this study, 16 NF-YA subunit genes were obtained by phylogenetic analysis and divided into three categories, with two more NF-YA members identified than what Zhang et al. (2016) have done. The eukaryotic 5′-UTR is critical for ribosome recruitment to the messenger RNA (mRNA) and in start codon choice. Additionally, it plays major roles in the control of translation efficiency and shaping the cellular proteome (Hinnebusch, Ivanov & Sonenberg, 2016). The 3′-UTRs of mRNAs regulate mRNA-based processes, such as mRNA localization, mRNA stability and translation. In addition, 3′-UTRs can establish 3′-UTR-mediated PPIs, and thus transmit genetic information encoded in 3′-UTRs to proteins. This function regulates diverse protein features, including protein complex formation or posttranslational modifications, but it may also alter protein conformations. Therefore, 3′-UTR-mediated information transfer can regulate protein features that are not encoded in the amino acid sequence (Mayr, 2019). However, ZmNFYA03, -07, -11 and -13 do not have 5′-UTR and 3′-UTR structures. Thus, further research is required to determine how they perform these functions. The predicted PPI network indicated that all 16 NF-YA proteins interact with GRMZM2G180947_P01, GRMZM2G473152_P01 and GRMZM2G444073_P01 (Fig. 2B), which were named ZmNF-YB4 (GRMZM2G180947_P01), ZmNF-YB6 (GRMZM2G473152_P01) and ZmNF-YB9 (GRMZM2G444073_P01), respectively (Zhang et al., 2016). The naming is consistent with the binding of CBFB_NF-YA, a conserved domain in the associated NF-YA subunit genes, to the NF-YB subunit. Yang et al. (2022) confirmed that ZmNF-YB16 interacts with ZmNF-YC17 through its histone-folding domain to form a heterodimer in the cytoplasm. Then, the complex enters the nucleus under osmotic-stress conditions to form a heterotrimer with ZmNF-YA1 or ZmNF-YA7 (ZmNFYA08 in this study), forming the first identified ZmNF-Y transcriptional regulatory complex in maize (Yang et al., 2022). These results laid a foundation for elucidating the functions and regulatory mechanisms of maize NF-YA subunit genes. However, the specific functions of NF-YA subunit genes and their relationships with NF-YBs and NF-YCs still require further study. The overexpression of ZmNF-YA1 enhances drought and salt tolerance and promotes root development in maize, whereas the zmnf-ya1 mutant shows drought and salt sensitivity (Yang et al., 2022). In this study, ZmNFYA01 was highly expressed during embryogenesis and positively regulated maize disease resistance, suggesting that it played important roles in maize growth and development. However, its regulatory mechanisms need further study.
In summary, 16 maize NF-YA subfamily genes were identified, and their expression levels in the same tissue at different developmental stages revealed a pattern, and the gene expression levels changed significantly under biotic and abiotic stress, including SA and JA treatments. ZmNFYA01 and ZmNFYA06 positively regulated maize resistance, and provided broad-spectrum resistance to pathogenic fungi. Compared with in B73, the expression levels of the ZmPRs in nfya01 and nfya06 plants were changed significantly, suggesting that this is part of the regulation of maize disease resistance. The important roles of NF-YA subunit genes in maize growth, development and resistance to biotic and abiotic stresses have been preliminarily determined.
10.7717/peerj.14306/supp-1 Click here for additional data file. 10.7717/peerj.14306/supp-2 Click here for additional data file. 10.7717/peerj.14306/supp-3 Click here for additional data file. 10.7717/peerj.14306/supp-4 Click here for additional data file. 10.7717/peerj.14306/supp-5 Click here for additional data file. 10.7717/peerj.14306/supp-6 Click here for additional data file. 10.7717/peerj.14306/supp-7 Click here for additional data file. 10.7717/peerj.14306/supp-8 Click here for additional data file. 10.7717/peerj.14306/supp-9 Click here for additional data file. |
|
PMC9648348 | Gautam Pareek | AAA+ proteases: the first line of defense against mitochondrial damage | 07-11-2022 | AAA+ Protease,Mitochondrial quality control,Mitochondria in neurological disorders,Mitochondrial Unfolded Protein Response,Mitochondrial Translation | Mitochondria play essential cellular roles in Adenosine triphosphate (ATP) synthesis, calcium homeostasis, and metabolism, but these vital processes have potentially deadly side effects. The production of the reactive oxygen species (ROS) and the aggregation of misfolded mitochondrial proteins can lead to severe mitochondrial damage and even cell death. The accumulation of mitochondrial damage is strongly implicated in aging and several incurable diseases, including neurodegenerative disorders and cancer. To oppose this, metazoans utilize a variety of quality control strategies, including the degradation of the damaged mitochondrial proteins by the mitochondrial-resident proteases of the ATPase Associated with the diverse cellular Activities (AAA+) family. This mini-review focuses on the quality control mediated by the mitochondrial-resident proteases of the AAA+ family used to combat the accumulation of damaged mitochondria and on how the failure of this mitochondrial quality control contributes to diseases. | AAA+ proteases: the first line of defense against mitochondrial damage
Mitochondria play essential cellular roles in Adenosine triphosphate (ATP) synthesis, calcium homeostasis, and metabolism, but these vital processes have potentially deadly side effects. The production of the reactive oxygen species (ROS) and the aggregation of misfolded mitochondrial proteins can lead to severe mitochondrial damage and even cell death. The accumulation of mitochondrial damage is strongly implicated in aging and several incurable diseases, including neurodegenerative disorders and cancer. To oppose this, metazoans utilize a variety of quality control strategies, including the degradation of the damaged mitochondrial proteins by the mitochondrial-resident proteases of the ATPase Associated with the diverse cellular Activities (AAA+) family. This mini-review focuses on the quality control mediated by the mitochondrial-resident proteases of the AAA+ family used to combat the accumulation of damaged mitochondria and on how the failure of this mitochondrial quality control contributes to diseases.
Mitochondria are indispensable for cellular life and perform a plethora of functions, including calcium homeostasis, regulation of programmed cell death, regulation of innate immunity, and stem cell fate (Nunnari & Suomalainen, 2012; Bock & Tait, 2020; McBride, Neuspiel & Wasiak, 2006; Kauppila, Kauppila & Larsson, 2017; Chandel, 2015; Vasan, Werner & Chandel, 2020). Most important of all, mitochondria produce almost all the vital cellular energy in the form of Adenosine triphosphate (ATP) by virtue of the different respiratory chain (RC) complexes embedded in the inner mitochondrial membrane (Milenkovic et al., 2017; Gustafsson, Falkenberg & Larsson, 2016; Sousa, D’Imprima & Vonck, 2018; Zhao et al., 2019). These RC complexes are involved in coupling the mitochondrial membrane potential created by the proton gradient to the synthesis of ATP. However, mitochondria face a quality control challenge during their lifetime. In addition to producing ATP, these RC complexes, mainly complexes I and III, produce a large amount of reactive oxygen species (ROS) by electron leakage, which results in the partial reduction of oxygen to form the detrimental superoxide radicals (Quinlan et al., 2013; Murphy, 2009). The excessive ROS produced by this process presents a quality control (QC) challenge for the mitochondria and the cell. The dual genetic origin of the mitochondria also poses a challenge to its homeostasis. Mitochondria harbor their own small circular DNA, which encodes for 37 genes, including 13 proteins of the RC complexes, 22 tRNAs, and two rRNAs of mitochondrial ribosomes in the human (Hällberg & Larsson, 2014). The vast majority of the remaining more than 1,000 mitochondrial proteins are encoded by the nuclear genome and are imported into the mitochondria by the translocase of the outer membrane ‘TOM’ and translocase of the inner membrane ‘TIM’ complex (Pfanner, Warscheid & Wiedemann, 2019; Neupert & Herrmann, 2007). Therefore, mitochondrial biogenesis, specifically the RC complexes’ biogenesis, heavily relies on the coordinated expression of the nuclear and mitochondrially encoded subunits. Any perturbation in this equilibrium can lead to the accumulation of protein aggregates and eventually an adverse effect on mitochondria and the cellular health (Youle, 2019). Thankfully, to combat the damage and the occurrence of ROS and misfolded protein aggregates, mitochondria have evolved a variety of surveillance pathways that are activated upon the trigger of the stress (Rugarli & Langer, 2012; Pickles, Vigié & Youle, 2018; Roca-Portoles & Tait, 2021; Quiles & Gustafsson, 2020). The most studied pathway of all is the ‘mitophagy’ pathway which involves the degradation of the whole mitochondria in the lysosome through a mitochondrial selective form of autophagy in extreme stress conditions (Narendra, Walker & Youle, 2012; Whitworth & Pallanck, 2017). In yeast, the mitophagy pathway is carried out by the ‘Atg32’ protein, which binds to an adaptor ‘Atg11’, and then mitochondria are recruited to and imported into the vacuole for degradation and recycling of its contents (Kanki et al., 2009). However, metazoans possess entirely different machinery for mitophagy, including the kinase ‘Pink1’ and the ubiquitin ligase ‘Parkin’ and several other adaptors, including P62, Optineurin, Nuclear dot protein 52 (NDP52), Human T-Cell Leukemia Virus Type I Binding Protein 1 (TAX1BP1) and Neighbor of BRCA1 gene 1 (NBR1) (Narendra et al., 2008; Narendra et al., 2010; Jin et al., 2010; Lazarou et al., 2015; Martin, Dawson & Dawson, 2011). Interestingly, the mutations in the genes of the mitophagy pathway, including Pink1 and Parkin, have been shown to cause Parkinson’s disease in humans and genetic disruption of these genes causes locomotor and behavioral deficits and accumulation of dysfunctional mitochondria in a variety of model organisms (Greene et al., 2003; Scarffe et al., 2014; Clark et al., 2006; Park et al., 2006; Itier et al., 2003; Lee et al., 2017; Shin et al., 2011; Yang et al., 2006). The mechanistic details of this pathway have been discussed in pioneering reviews in the past and are out of scope for this review article (Pickles, Vigié & Youle, 2018; Whitworth & Pallanck, 2017; Ge, Dawson & Dawson, 2020). The Mitochondrial network is dynamic; mitochondria merge by ‘fusion’ and separate from each other by ‘fission’ (Chen & Chan, 2017; Friedman et al., 2011; Youle & van der Bliek, 2012). The mitochondrial ‘fusion’ allows mitochondria to exchange their content and promotes functional complementation in the event of mitochondrial damage. In contrast, mitochondrial ‘fission’ facilitates the segregation of damaged mitochondria and subsequent removal by mitophagy. These processes are crucial to mitochondrial proteostasis since mutations in the components of the fission/fusion machinery have been shown to cause neurodegenerative diseases in humans, including Charcot-Marie-Tooth Type 2A (CMT2A), peripheral neuropathy, and Dominant Optic Atrophy (DOA) (Chan, 2020). The contribution of ‘mitophagy’ and ‘mitochondrial dynamics’ to mitochondrial quality control is indisputable. However, recent reports have demonstrated that these pathways cannot account for the total mitochondrial protein turnover in the cell (Vincow et al., 2013; Vincow et al., 2019). The quantitative proteomic studies performed in fruit flies and Drosophila S2 cells have demonstrated that the half-lives of different ribosomal proteins lie very close to each other, in agreement with the fact that the whole ribosomes are turned over as a single unit by the ‘ribophagy’ pathway. However, in the same study, mitochondrial proteins exhibited a wide range of half-lives that cannot be explained by a sole mitophagy-mediated mitochondrial protein degradation, suggesting more than one mechanism responsible for the turnover of mitochondria (Vincow et al., 2013). Subsequent studies showed that nearly 35% (one-third) of all mitochondrial protein turnover occurred through autophagy and almost 25% through the parkin-dependent mitophagy (Vincow et al., 2019). Additionally, utilizing mito-keima and mito-QC based reporter assays, it has been shown that the basal mitophagy flux is tissue-specific, and genetic disruption of the Pink1 or Parkin has a very modest effect on this flux (Lee et al., 2018; McWilliams et al., 2016; Sun et al., 2017). That being the case, the quality control mediated by the ATPase Associated with diverse cellular Activities (AAA+) serine proteases represent likely candidates to account for the majority of the mitochondrial protein turnover as they sense and degrade selective damaged mitochondrial components, henceforth, explaining the broad distribution of mitochondrial protein half-lives.
The mitochondrial AAA+ proteases have gained significant attention and broader interest from the scientific community from different research areas due to their role in human health and diseases. To supplement the existing information as relevant literature about AAA+ proteases has increased in the past years, this perspective review discussed the classification, structural insights, functions, and recently reported substrates of the AAA+ proteases. The relevant pieces of literature on the topic were identified using databases like Pubmed, Google Scholar, Web of Science, and ScienceDirect. Briefly, the review identified 154 relevant research articles from research labs working on AAA+ proteases across the globe. To help move the field forward, this study presents a systematic review structured to discuss (1) the need for mitochondrial QC and how AAA+ proteases supplement this need, (2) the neurodegenerative diseases associated with the AAA+ proteases, (3) what have we learned from the disease model systems, and (4) substrates of the AAA+ proteases.
The FtsH (Filamentous temperature sensitive H)-related AAA+ proteases are present in all kingdoms of life, from bacteria to humans, and are classified into different families depending on their localization and domain organization (Sauer & Baker, 2011; Quirós, Langer & López-Otín, 2015). This family of proteases forms oligomeric complexes that use energy from ATP hydrolysis to unfold and transport substrates to their zinc metalloprotease domain for degradation. These proteases are comprised of an axial pore formed by the oligomeric ring of the AAA+ ATPase domain. This pore size is narrow enough to block the passage of any folded protein toward an inner proteolytic chamber where proteolysis occurs (Sauer et al., 2004). Therefore, the energy derived from the cycles of ATP hydrolysis by the ATPase domain is utilized to induce the conformational changes in the substrate protein to unfold and transport substrates to the proteolytic cavity for degradation. Higher eukaryotes, including humans, have five major nuclear DNA encoded AAA+ proteases in the mitochondria, distinguished by their subunit composition and localization (Fig. 1A). The i-AAA (commonly known as Yme1l1) and m-AAA proteases are localized in the inner membrane, and the Lon and Clp family of proteases are matrix-localized (Deshwal, Fiedler & Langer, 2020). Furthermore, the catalytic domains of the m-AAA and i-AAA proteases face the matrix and the intermembrane space, respectively, to protect against the insults on both sides of the inner membrane. The m-AAA protease is classified into two forms: hetero-oligomeric complexes of the paraplegin protein and the ATPase family gene 3-like 2 (Afg3l2) protein, and the homo-oligomeric complexes of the Afg3l2 protein. For all the AAA+ proteases, the ATPase and the proteolytic subunits are contained on a single polypeptide separated by a short linker region, except the Clp family, which is composed of the proteolytic (ClpP) and ATPase (ClpX) subunits present in the two different proteins (Fig. 1A). Recent reports based on cryo-electron microscopy have provided unique structural insights into the Yme1l1, Afg3l2, and Lon protease (Puchades et al., 2017; Puchades et al., 2019; Shin et al., 2021). The quaternary organization of yeast Yme1l1 protease has revealed that the six AAA+ domains form an asymmetric spiral staircase on the top of a planar symmetric hexametric protease ring (Fig. 1B) (Puchades et al., 2017). Notably, four of the six subunits in Yme1l1 are in the ATP-bound form; the lowest subunit has the ADP-like density in the nucleotide-binding pocket, whereas the binding pocket of one of the subunits possesses an “apo-like” nucleotide-free state. The substrate engagement in the asymmetric ring is coordinated by the two conserved tyrosine residues forming a spiral staircase around the translocating substrate by intercalating into its backbone (Fig. 1B). The sequential ATP-hydrolysis around the ring results in subsequent substrate translocation to the proteolytic domain in a stepwise fashion. The cryo-EM structure of the human truncated Afg3l2 protein comprising the ATPase and protease domain has highlighted that the basic structural features and the fundamental mechanisms of ATP hydrolysis and substrate translocation are relatively conserved between the i-AAA and m-AAA family of proteases (Puchades et al., 2019). However, some unique evolutionary protein sequence adaptations related to their extreme C-terminus in i-AAA and m-AAA proteases enable them to process their distinct substrates in a different microenvironment. Surprisingly, the cryo-EM structure of the human Lon protease has revealed that the substrate binding alone to the AAA+ domain is not sufficient to produce allosteric transmission to the proteolytic domain and the activation of the proteolytic chamber was only achieved in the presence of peptidomimetic inhibitor ‘bortezomib’ suggesting an additional level of regulation which requires the substrate-binding within the protease domain to induce the activated conformation of the domain (Shin et al., 2021). The activity of all AAA+ proteases is under strict scrutiny to prevent the uncontrolled degradation of mitochondrial proteins. The precise mechanism by which the AAA+ proteases engage with their specific substrates lies in the binding to the particular sequences that serve as potential degron or degradation signals on the target protein (Glynn, 2017; Steele & Glynn, 2019; Leonhard et al., 2000). These degradation signals are usually highly hydrophobic and are 10–20 amino acids long (Gur & Sauer, 2008; Rampello & Glynn, 2017; Leonhard et al., 1999). The Yme1l1 protease recognizes its substrates through a specific motif of amino acids, including the F-h-h-F (F = phenylalanine and h stands for any hydrophobic residue) accessible for degradation in the unfolded state but hidden in a natively folded form of the protein (Shi, Rampello & Glynn, 2016). The accumulation of protected intermediates from experiments involving dihydrofolate reductase and stabilizing ligand methotrexate has demonstrated that the degradation of the target protein by Yme1l1 is achieved by processive unfolding and translocation to the proteolytic chamber from the degron terminus (Shi, Rampello & Glynn, 2016). Furthermore, a highly similar sequence (WRFAWFP) rich in aromatic amino acids from β-galactosidase has been identified as the primary recognition site for the Lon protease (Gur & Sauer, 2008). The human Lon protease has been shown to preferentially recognize and degrade the unfolded proteins harboring degron signals, including the folding incompetent form of the ornithine transcarbamylase (OTC) but not the aggregated form of the malate dehydrogenase (MDH) (Bezawork-Geleta et al., 2015). Moreover, The peptidase specificity profile of Afg3l2 has demonstrated that this protease discriminates its potential substrates by cleaving peptide bonds adjacent to either the hydrophobic (most commonly phenylalanine) or small polar residues in the P1’ position (the position adjacent to the scissile bond) (Ding et al., 2018). Overall, the recent structural details provide a molecular framework to understand the mechanics of the AAA+ protease machines and to appreciate their role in health and disease.
The m-AAA proteases are embedded in the inner membrane with their active site oriented towards the matrix. There are two different versions of m-AAA; one is composed entirely of the homoligomers of the Afg3l2 protein; the other is a hetero-oligomeric complex of the Afg3l2 and Paraplegin proteins. The heterozygous missense mutations in the AFG3L2 gene cause spinocerebellar ataxia (SCA28) disease characterized by cerebellar dysfunction due to Purkinje cell degeneration (Fig. 1A) (Di Bella et al., 2010). Most of these mutations are clustered in the protease domain’s highly conserved region, and one of the mutations is present at the asparagine 432 positions in the ATPase domain. All these mutations result in the loss of the Afg3l2 activity and impair the degradation/maturation of its substrates. The homozygous missense mutation in AFG3L2 causes spastic ataxia neuropathy syndrome characterized by progressive lower limb spastic paraparesis, cerebellar atrophy, peripheral neuropathy, ptosis, dystonia, and progressive myoclonic epilepsy (Fig. 1A) (Pierson et al., 2011). This homozygous recessive mutation alters the conserved tyrosine at 616 positions to cysteine and impairs the homo- and hetero-oligomerization propensity of the Afg3l2 protein. The homozygous missense mutations in the SPG7 gene (encoding Paraplegin protease) cause the hereditary spastic paraplegia (HSP) disease characterized by the progressive weakness and spasticity of the lower limbs and by the loss of upper motor neurons of the corticospinal tracts (Fig. 1A) (Wilkinson et al., 2004). Additionally, compound heterozygous and heterozygous mutations in the SPG7 gene are also responsible for the progressive external ophthalmoplegia with early or progressive ataxia, dysphagia, and proximal myopathy (Pfeffer et al., 2014). A closer examination of these patients’ skeletal muscle biopsies revealed evidence of the mosaic respiratory chain deficiency and an increase in mitochondrial biogenesis, giving rise to ‘ragged-blue’ fibers. These defects were accompanied by accelerated clonal expansion of mitochondrial DNA mutations and deletions along with increased mitochondrial mass and hyper-fused mitochondria in affected individuals. The deficiency of the yeast homolog of m-AAA protease Yta10/Yta12 is viable but causes a growth defect on a nutrient source containing glycerol that requires a mitochondrial respiration (Arlt et al., 1996). The homozygous knockout mice of AFG3L2 exhibited a defect in axonal development and myelination accompanied by early paraparesis and tetraparesis (Maltecca et al., 2008). These mice do not survive beyond day 16. The heterozygous AFG3L2 mouse recapitulated some pathological features of SCA28, including cerebellar atrophy, dark degeneration of Purkinje cells, and mitochondrial dysfunction (Maltecca et al., 2009). These findings suggest that haploinsufficiency is responsible for the ataxia disease associated with mutations in the AFG3L2 gene. The knockout of AFG3L2 in Drosophila is larval lethal, and RNAi-mediated knockdown causes shortened lifespan, locomotor deficits, and severe neurological and mitochondrial morphological abnormalities (Pareek & Pallanck, 2020). Paraplegin deficiency in various model organisms, including Drosophila and mouse, is viable. Still, adults exhibited severe phenotypes, including shortened lifespan, behavioral deficits, neurodegeneration, and accumulation of dysfunctional mitochondria along with the disorganized cristae network (Ferreirinha et al., 2004; Atorino et al., 2003; Pareek, Thomas & Pallanck, 2018). The paraplegin deficient mice exhibited mitochondrial morphological abnormalities in synaptic terminals and distal regions of axons followed by axonal swelling, degeneration, and behavioral abnormalities. The ultrastructural morphology of the photoreceptor terminals of the Drosophila paraplegin knockout displayed the progressive neurodegenerative phenotype, severely swollen and dysmorphic mitochondria accompanied by altered axonal transport of mitochondria. Despite co-assembling as a heteromeric complex, the spectrum of diseases caused by mutations in the Afg3l2 and paraplegin are distinct. The interplay between these proteins and their substrates in the mitochondria is poorly understood. The m-AAA protease Afg3l2 plays a crucial role in regulating mitochondrial translation by participating in the maturation of one of the large mitochondrial ribosomal subunits, Mrpl32 (Nolden et al., 2005). The Mrpl32 maturation by Afg3l2 protease is conserved in yeast, flies, mice, and human cell lines (Pareek & Pallanck, 2020; Almajan et al., 2012). The defect in the processing led to the defective ribosome assembly and the attenuation of the mitochondrial translation (Fig. 2). This subsequently caused a defect in the RC complexes comprised of the subunits encoded by the mitochondrial DNA, including complex I, III, IV, and V (Pareek & Pallanck, 2020). An imbalance in the stoichiometry of nuclear vs. mitochondrial DNA encoded RC subunits caused the accumulation of protein aggregates and activation of the mitochondrial unfolded protein response (mito-UPR) (Fig. 2). Afg3l2 deficient worms (AFG3L2 has been confusingly referred to as SPG7 in worms) served as one of the first models to understand the mito-UPR (Nargund et al., 2012). The activation of the mito-UPR restores protein homeostasis utilizing a two-pronged approach consisting of increased expression of chaperones and proteases to facilitate the refolding/degradation of the misfolded proteins and phosphorylation-mediated inactivation of the cytosolic translation-initiation factor eIF2α by GCN2 kinase to attenuate cytoplasmic translation to reduce the burden on the mitochondrial protein folding machinery under stress conditions (Baker et al., 2012). However, some of these phenotypes, including the Mrpl32 maturation defect and activation of the mito-UPR, have not been observed in the Paraplegin deficient flies (Pareek & Pallanck, 2020). These recent findings suggest that the Afg3l2 homomers and the Afg3l2/Paraplegin heteromultimers have independent substrates. The result further supports this hypothesis that the SPG7 deficient flies did not exhibit defects in the activity or abundance of several RC complexes that contain subunits encoded by the mitochondrial genome and instead showed a defect in the complex II activity, which is encoded by the nuclear genome (Pareek, Thomas & Pallanck, 2018). The mitochondria also play a crucial role in calcium buffering and have a calcium uniporter complex (MCU) to uptake calcium from the cytoplasm, thereby helping regulate the cytoplasmic calcium levels (Baughman et al., 2011). Recent findings have reported that the m-AAA proteases degrade a regulatory subunit of the MCU complex, known as the Essential MCU REgulator (EMRE) (Sancak et al., 2013; König et al., 2016; Tsai et al., 2017). The formation of the constitutively active MCU-EMRE subcomplex in the absence of another regulatory subunit, Mitochondrial Calcium Uptake Protein 1 (MICU1), caused the calcium overload in the mitochondrial matrix and triggered the opening of the mitochondrial permeability transition pore (mPTP) and necrosis mediated cell death (Bernardi, 2013). However, the role of the m-AAA protease in the regulation of the MCU is disputed since the depletion of the MCU failed to rescue the behavioral defects, Purkinje cell degeneration, and the neuroinflammatory response of the Afg3l2 mutants (Patron, Sprenger & Langer, 2018). While in contrast to this, it has also been shown using cultured Purkinje cells derived from the cerebella of newborn mice that the depolarized mitochondria in Afg3l2-deficient cells are not efficient to buffer induced Ca2+ peaks, increasing the cytoplasmic Ca2+ concentrations, which subsequently triggers the dark cell degeneration (Maltecca et al., 2015). The calcium buffering is particularly critical for the Purkinje cells as they are comprised of the highly branched dendritic arbors receiving inputs from glutamatergic stimulation of various receptors, including ionotropic α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) and metabotropic receptors mGluR1 and, are therefore exposed to massive and sudden Ca2+ influx, making them susceptible to glutamate-mediated excitotoxicity. Notably, in the haploinsufficient Afg3l2 SCA28 mice, the partial genetic depletion of the metabotropic glutamate receptor mGluR1 or administration of the β-lactam antibiotic ceftriaxone, which promotes synaptic glutamate clearance, both decreased Ca2+ influx in Purkinje cells. It also improved the ataxic phenotype (Maltecca et al., 2015). More future work will be required to examine this matter in greater detail to determine how sensitive the mitochondrial calcium uptake is in response to the dosages of Afg3l2 protein. The Afg3l2 protease control several other aspects of mitochondrial biology, including mitochondrial morphology, by regulating the processing of the inner mitochondrial membrane protein Opa1 (Fig. 2) (Ehses et al., 2009; Ishihara et al., 2006). The inner mitochondrial membrane fusion success depends on the balanced stoichiometry of the large and small isoforms of the Opa1 protein (Mishra et al., 2014). The deficiency of the Afg3l2 m-AAA protease caused the activation of another protease Oma1 (Ehses et al., 2009). The enhanced Oma1 activity led to the cleavage of the Opa1 long isoform to the shorter isoform and, subsequently, mitochondrial fragmentation. The Afg3l2 protease also cleaves mitophagy protein Pink1 inside the mitochondrial matrix, and the knockdown of the protease stabilizes the smaller form of Pink1 generated after the matrix processing peptidase (MPP) cleavage (Greene et al., 2012; Thomas et al., 2014). However, the stabilization of this cleavage product did not upregulate the downstream mitophagy pathway and did not affect the parkin recruitment to the mitochondria. Depleting the Afg3l2 protease in mouse primary cortical neurons drastically impaired the anterograde transport of the mitochondria, thereby depleting the synapses out of the mitochondria (Kondadi et al., 2014). The underlying cause of this defect has been attributed to the tau hyperphosphorylation by kinases, including the protein kinase A (PKA) and ERK1/2, which were activated by high ROS caused by the depletion of the Afg3l2 protease.
The Yme1l1 is an ATP-dependent metalloprotease embedded in the inner mitochondrial membrane, with its protease domain facing the intermembrane space (referred to as i-AAA protease). The homozygous missense mutation in the YME1L1 is associated with a multisystemic mitochondriopathy with neurological abnormalities, including intellectual disability, motor development and speech delay, and optic nerve atrophy with visual and hearing impairment (Hartmann et al., 2016). This mutation alters the highly conserved arginine at 149 positions within the predicted mitochondrial targeting sequence to the tryptophan. It abrogates the maturation of the Yme1l1 by MPP after the import into the mitochondria. The homozygous Yme1l1 knockout in Drosophila caused shortened lifespan, locomotor deficit, photoreceptor degeneration, and apoptosis-mediated cell death (Qi et al., 2016). These flies possessed mitochondria with disrupted cristae network, dysfunctional RC activity, and misfolded protein aggregates. The nervous system-specific knockout of YME1L1 in mice caused microphthalmia, cataracts, progressive axonal degeneration of dorso-lateral tracts, and inflammation in the retina and spinal cords accompanied by locomotor impairment of hind limbs (Sprenger et al., 2019). These mice also manifested late onset RC dysfunction, loss of cristae structure, mitochondrial fragmentation, and a defect in the anterograde transport of the mitochondria. The role of the Yme1l1 protease in regulating mitochondrial morphology is well established (Fig. 2). The Yme1l1, along with the Oma1 protease, holds a switch that dictates mitochondrial fusion/fragmentation in response to membrane depolarization and cellular bioenergetic status (Rainbolt et al., 2016). In conditions where membrane depolarization occurs in the presence of ATP, the Oma1 protease is degraded in a Yme1l1-dependent manner. These conditions stabilize the Opa1 short isoform generated by the Yme1l1 cleavage at the S2 site and favor mitochondrial fusion with enhanced RC activity, protection from mitophagy, and protection from apoptosis. However, when the depletion of ATP accompanies mitochondrial membrane depolarization, the Yme1l1 protease is degraded, and Oma1 is stabilized. The stabilization of Oma1 protease caused the cleavage of Opa1 at the S1 site and, subsequently, blockage in fusion, reduced RC activity, susceptibility to mitophagy, and apoptosis. Moreover, a recent study suggested that the Yme1l1 is a part of the enormous protease complex formed by the Yme1l1-PARL proteases along with the scaffold protein Stomatin-like protein 2 (SLP2). This complex SLP2–PARL–YME1L1 not only regulates the Pink1 kinase and Pgam5 phosphatase processing but also inhibits the stress-activated Oma1 protease and promotes the cell survival (Wai et al., 2016). The Yme1l1 protease also participates in the integrated stress signaling (ISR) pathway during stress conditions (Fig. 2). The Yme1l1 degrades the import motor component Tim17A downstream of the stress-regulated translational attenuation induced by the eukaryotic initiation factor 2α (eIF2α) phosphorylation (Rainbolt, Saunders & Wiseman, 2015; Rainbolt et al., 2013). The decrease in Tim17A protein level attenuates the general mitochondrial protein import and promotes the induction of the mito-UPR-associated proteostasis genes and stress-responsive genes. Another intriguing function of Yme1l1 is related to cellular phospholipid metabolism. Yme1l1 drives the degradation of the PRELI-like protein family in the intermembrane space, including Ups1 and Ups2 in the absence of their binding partner Mdm35, a member of the twin Cx9C protein family (Potting et al., 2010). The Ups1 and Ups2 proteins were shown to regulate the accumulation of cardiolipin and phosphatidylethanolamine in the mitochondria (Tamura et al., 2009). Recent studies have also highlighted the role of Yme1l1 in metabolic reprogramming by rewiring the mitochondrial proteome under hypoxic and nutrient-deprived conditions (MacVicar et al., 2019). This metabolic adaptation is operated through the mTORC1-LIPIN-YME1L1 axis. The inhibition of mTORC1 activates LIPIN, a phosphatidic acid phosphatase that causes a decrease in the phosphatidylethanolamine levels in the mitochondrial membrane and subsequently activates Yme1l1. Under these conditions, the enhanced degradation of the protein translocases by activated Yme1l1 inhibits mitochondrial biogenesis. It promotes the metabolic flux of Tricarboxylic acid (TCA) intermediates towards anaplerotic biosynthetic reactions in preexisting mitochondria. These events are critical for the growth of solid tumors, such as pancreatic ductal adenocarcinomas (PDACs), for the anchorage-independent growth of the cells, and for the proliferation of cells in the presence of glutamine as a carbon source. The Yme1l1 is not only required for the development of specific cancer subtypes but also plays a role in the maintenance and self-renewal of neural stem and progenitor cells (NSPCs), and conditional deletion of the protease promotes stem cell exhaustion and pool depletion (Wani et al., 2022). The YME1L1l deletion rewired cellular metabolism in these cells, including a significant reduction in the mitochondrial fatty acid oxidation, and activated a differentiation program in NSPCs.
Lon is the master AAA+ protease that protects against protein aggregation in the mitochondria by degrading the misfolded proteins. The homozygous or compound-heterozygous mutations in the Lon protease are associated with a multisystem developmental disorder known as CODAS syndrome, characterized by cerebral, ocular, dental, auricular, and skeletal anomalies (Fig. 1A) (Strauss et al. 2015). These mutations are clustered in the AAA+ domain, causing a defect in the protease activity. The lymphoblastoid cell lines harboring these mutant substitutions had swollen mitochondria with electron-dense inclusions, abnormal cristae structure, and a defect in the RC activity. The mutation in the proteolytic domain has also been shown to cause the atypical CODAS syndrome with cerebellar atrophy of high intensity, regression, and involuntary movement. However, the molecular basis of the defect for this mutation is not entirely understood (Inui et al., 2017). Additionally, mutations in the Lon protease are associated with a broad spectrum of diseases, including substitution at Pro761Leu in the proteolytic domain causes profound neurodegeneration with progressive cerebellar atrophy, hypotonia, muscle weakness, and intellectual disability (Nimmo et al., 2019). Interestingly, the cultured fibroblast from the affected individuals exhibited electron-dense inclusions in the mitochondria, regular activity of the RC complexes, and glucose-repressed oxygen consumption, while the galactose and palmitic acid utilization were unaffected. Notably, these fibroblasts also had reduced pyruvate dehydrogenase (PDH) activity and elevated intracellular lactate: pyruvate ratios caused by an increase in the phosphorylated E1α subunit of PDH. A recent report has identified a recessive mutation in the N-terminal substrate recognition domain of the Lon protease at aspartate 436 to asparagine, causing a form of mitochondrial cytopathy with early onset ataxia, developmental delay, emotional outbursts, speech and swallowing difficulties, and hypotonia (Hannah-Shmouni et al., 2019). The muscle biopsy from patients revealed increased oxidative stress, a block in autophagy, reduced mitochondrial state 3 respiration, and intra-mitochondrial globular inclusions. The homozygous knockout of Lon is lethal in mice and Drosophila, while the RNAi-mediated knockdown is associated with shortened lifespan and behavioral deficits (Pareek et al., 2018; Quirós et al., 2014). Deleting the Lon protease homolog in worms (lonp-1) is viable but accompanied by disturbed mitochondrial homeostasis, ROS accumulation, impaired growth, behavior, and lifespan (Taouktsi et al., 2022). In addition to the m-AAA proteases, Lon protease has also been shown to regulate the various aspects of the mitochondrial gene expression machinery (Fig. 2). The Lon protease regulates mitochondrial DNA copy number by controlling the mitochondrial transcription factor A (TFAM) levels in Drosophila S2 cells (Matsushima, Goto & Kaguni, 2010). The heterodimer of mitochondrial seryl-tRNA synthetase (SerRS2) and its paralog SLIMP protein bridges the interaction between TFAM and Lon protease, thereby promoting the TFAM degradation in S2 cells (Picchioni et al., 2019). While in mammalian cells, Lon protease promotes the degradation of TFAM phosphorylated by a cAMP-dependent protein kinase, which impairs its binding to DNA (Lu et al., 2013). Additionally, the Lon protease inhibition in fibroblast is associated with a defect in the maturation and solubility of a subset of proteins required for mitochondrial DNA maintenance and translation, including SSBP1, MTERFD3, and FASTKD2 (Zurita Rendón & Shoubridge, 2018). Depleting Lon in these cell lines also resulted in the loss of mitochondrial DNA, suppression of mitochondrial translation associated with impaired ribosome biogenesis, protein aggregates in the matrix, and activation of the ISR pathway. Notably, inhibition of the Lon protease in Hela cell lines caused defects in the turnover of mitochondrial pre-RNA processing nuclease MRPP3, followed by accumulation of many unprocessed mitochondrial transcripts and arrest in the mitochondrial translation (Münch & Harper, 2016). The Lon protease has also been implicated in tumorigenesis, and heterozygous Lon mice are protected against colorectal cancer and skin papillomas induced by the treatment of 7,12-dimethylbenzanthracene and tetradecanoylphorbol acetate (DMBA/TPA) (Quirós et al., 2014). Higher expression of Lon protease is a poor prognosis marker and correlates with poor survival in human colorectal cancer and melanoma. The Lon protease-induced metabolic reprogramming, including reduction in mitochondrial oxidative phosphorylation (OXPHOS) capacity and increase in glycolysis, supports the proliferation of tumor cells and metastasis. In addition to its protease function, Lon protease also possesses chaperone properties, and this function of Lon is critical for its anti-apoptosis activity. The increased Lon expression sequestered the P53 in the mitochondrial matrix and prevented both the transcription-dependent (by reducing the expression of P53 target genes in the nucleus) and independent (by preventing the loss of mitochondrial membrane potential and the release of apoptotic proteins cytochrome C and SMAC/Diablo in the cytosol) functions of P53 and thereby inhibited the apoptosis and promoted tumorigenesis (Sung et al., 2018). The high expression levels of Lon protease correlate positively in oral squamous cell carcinoma (OSCC) patients. One of the well-established functions of the Lon protease is to degrade oxidatively damaged proteins in the mitochondria (Fig. 2). The aconitase’s oxidized form was among the first reported substrates of the Lon protease (Bota & Davies, 2002). Subsequently, the oxidized form of several other proteins, including mitochondrial stress proteins such as Hsp78, Hsp60, Sod2, prohibitins, mitochondrial metabolic enzymes such as pyruvate dehydrogenase complex, alpha keto glutarate dehydrogenase complex, ketol acid reductoisomerase, mitochondrial ribosomal proteins Mrp20 and respiratory chain subunits such as Qcr2 (complex III), Rip1 (complex III), Cox4 (complex IV), and ATP1,2 and 7 (complex V) exhibited altered abundance in Pim1 (Lon protease homolog in yeast) deficient yeast cells (Bayot et al., 2010). In conjunction with the Clp protease, the Lon protease degrades subunits of the matrix exposed ROS generating arm of the complex I, including NDUFV1, NDUFV2, NDUFS1, NDUFS2, NDUFB8, and NDUFA9 in depolarized mitochondria of SH-SY5Y and HeLa cells after CCCP treatment (Pryde, Taanman & Schapira, 2016). The Lon protease also protects against oxidative stress in Drosophila (Pomatto et al., 2017). Notably, the sex-specific isoforms of Lon protease confer protection against hydrogen peroxide-induced stress in female flies and paraquat-induced stress in male flies. The increase in oxidative stress is also associated with protein aggregation. The Lon protease degrades several misfolded proteins and protects against protein aggregation in the human mitochondria (Bezawork-Geleta et al., 2015). Recently, two independent reports have suggested that the chaperone activity of Lon protease is critical to prevent aggregation of several mitochondrial proteins, including DnaJ co-chaperone (TID1/DNAJA3), succinate dehydrogenase complex flavoprotein subunit A (SDHA), ClpX and mitochondrial heat shock protein 75 kDa (TRAP1/mtHSP75) (Matsushima et al., 2021; Pollecker, Sylvester & Voos, 2021). Not only this, but Lon also forms a complex with the components of the ‘TOM’ and ‘TIM’ machinery and maintains the soluble state of newly imported proteins, and degrades the unprocessed form of mitochondrial proteins with signal sequence (Matsushima et al., 2021). Almost any biological process related to mitochondria involves the regulation by the Lon protease. The Lon protease regulates the pyruvate dehydrogenase kinase 4 (PDK4) content in cardiomyocytes mitochondria and thereby controls the activity of pyruvate dehydrogenase complex and metabolic flexibility (Crewe et al., 2017). The ISU (Iron-sulfur cluster assembly protein) protein involved in Fe/S cluster biogenesis is also a substrate of the Lon protease when the sulfur donor Nfs1 and J-protein cochaperone Jac1 (J-type Accessory Chaperone) do not bind it (Nitrogen-Fixing Bacteria S-Like Protein) (Ciesielski et al., 2016; Song, Marszalek & Craig, 2012). The Lon protease also degrades the Steroidogenic acute regulatory protein (StAR) protein which facilitates the rate-limiting step in steroid biosynthesis, which is the transfer of cholesterol from the outer mitochondrial membrane to the inner mitochondrial membrane (Granot et al., 2007). The extensive repertoire of substrates for Lon protease makes it unarguably the master of all trades in the mitochondria.
The Caseinolytic mitochondrial matrix peptidase is a highly conserved member of the serine protease family residing in the mitochondrial matrix. In contrast to other proteases, which contain domains with protease and ATPase activity on the same polypeptide, the mitochondrial ClpP lacks ATP hydrolysis activity and only possesses the domain for the proteolytic activity that can digest small peptides without ATP requirement. Additionally, the ClpP, in conjunction with the chaperone and ATPase ClpX, can digest larger substrates in the matrix compartment. The ClpXP complex comprises the CLPP units arranged as a stacked heptameric ring with a proteolytic cavity flanked at both ends by ClpX (Fei et al., 2020). The recessive mutations in ClpP are associated with Perrault syndrome, a disease characterized by sensorineural hearing loss and ovarian dysgenesis (Jenkinson et al., 2013). The ClpP-deficient mouse model exhibited female and male infertility caused by disrupted spermatogenesis and ovarian follicular differentiation failures (Gispert et al., 2013). These mice also showed reduced lifespan, growth retardation, mild impairment in behavioral and respiratory chain activities, and activation of the Type I IFN responses through the Mitochondrial DNA–cGAS–STING signaling axis (Torres-Odio et al., 2021). Contrary to the mouse model, the deficiency of ClpP in fungal species Podospora anserine led to a healthy and increased lifespan, a phenotype that the expression of human ClpP can revert, demonstrating functional conservation between human and fungal ClpP (Fischer et al., 2013). Over the past years, some of the substrates of ClpP protease have been reported highlighting its role in mitochondrial biology. The ClpP protease has been shown to determine the rate of mitochondrial protein synthesis and mitoribosome assembly by regulating the levels of various components of the transcription and translation machinery (Fig. 2). In the human cell culture system, the deficiency of ClpP causes the accumulation of ERAL1 (Era Like 12S Mitochondrial RNA Chaperone 1), a putative 12S rRNA chaperone, whose timely removal from the small mitochondrial ribosomal subunit is essential for the complete maturation and assembly of the mitochondrial ribosome and the normal rate of mitochondrial translation (Szczepanowska et al., 2016). In Drosophila Schneider S2 cells, the knockdown of ClpP protease caused the accumulation of leucine-rich pentatricopeptide repeat domain-containing protein 1 (LRPPRC1) followed by an increase of mitochondrial mRNAs, accumulation of some unprocessed mitochondrial transcripts, and repression of the mitochondrial translation (Matsushima et al., 2017). Mitochondria are fundamental for cellular metabolism and regulate several metabolic pathways related to glucose homeostasis and fatty acid oxidation. Regarding this, the deficiency of the ClpP protease in mice is metabolically beneficial and has been associated with the lean body phenotype, increased energy expenditure, improved glucose homeostasis, insulin resistance, protection from high fat diet-induced obesity, and protection from hepatic steatosis (Becker et al., 2018; Bhaskaran et al., 2018). However, the depletion of ClpP also caused a decline in brown adipocyte function, impaired cold-induced thermogenesis, and increased fatty acid oxidation in white adipose tissues. It remains to be seen if targeting the ClpP protease using specific inhibitors in fat tissues will help combating obesity-related disorders in humans. In Caenorhabditis elegans, the ClpP homolog (CLPP-1) is required for the mito-UPR signaling (Fig. 2) (Haynes et al., 2007). The CLPP-1-mediated proteolysis of misfolded proteins is an early step necessary for UPR signaling in worms. However, the function of the ClpP in the UPR signaling is debated, and it has been shown in cell culture and mouse models that the ClpP is dispensable for the mito-UPR (Seiferling et al., 2016; Rumyantseva, Popovic & Trifunovic, 2022). Surprisingly, the deficiency of the ClpP protease ameliorated the symptoms of mitochondrial cardiomyopathy and mitochondrial encephalopathy caused by the tissue-specific deficiency of the mitochondrial aspartyl tRNA synthase, DARS2 (Seiferling et al., 2016; Rumyantseva, Popovic & Trifunovic, 2022; Dogan et al., 2014). These findings have demonstrated that the mammalian ClpP is neither required for nor regulates the UPR and introduces ClpP as a possible novel target for therapeutic intervention in mitochondrial diseases characterized by respiratory chain and mitochondrial gene expression dysfunctions. The ClpP is overexpressed in approximately 40% of Acute Myeloid Leukemia (AML) patients (Cole et al., 2015). The inhibition of the ClpP protease by knockdown or by using chemical inhibitors, including A2-32-01, reduced the growth and viability of several AML cell lines with high ClpP expression. Still, it did not affect other cell lines with low levels of ClpP. The effects of the ClpP-mediated growth inhibition on AML cell lines are partly mediated by the increased amount of misfolded complex II subunits, including succinate dehydrogenase A (SDHA), impaired complex II activity, reduced oxygen consumption rates (OCR), and increased ROS (Cole et al., 2015). An independent report has further substantiated the role of ClpP in cancer. It has shown that ClpP is universally overexpressed in primary and metastatic human cancer, correlating with poor patient survival (Seo et al., 2016; Rivadeneira et al., 2015). Using proteomics analysis, this report showed that the ClpP forms a complex with the oncoprotein survivin and the Hsp90-like chaperone TRAP-1. Together, they regulated the solubility and function of the oxidative phosphorylation Complex II subunit succinate dehydrogenase B (SDHB). Furthermore, the inhibition of the ClpP impaired oxidative capacity by causing the accumulation of misfolded SDHB, promoted oxidative stress, and ceased critical downstream signals essential for tumor cell proliferation, invasion, and metastatic dissemination in vivo (Seo et al., 2016). Not only inhibition but the hyperactivation of ClpP by imipridones, including ONC201 and ONC212, has been shown to selectively kill the malignant cells by causing a reduction in the respiratory chain complex subunits, impairing oxidative phosphorylation and mitochondrial morphology. At the same time, these compounds do not exert any effects on non-cancer cells (Graves et al., 2019; Ishizawa et al., 2019). The unfoldase ClpX has functions beyond its partner ClpP in eukaryotes, including its role in heme biosynthesis, which is conserved from yeast to mammalian vertebrates (Fig. 2) (Kardon et al., 2015). The ClpX catalyzes the incorporation of a cofactor pyridoxal phosphate (PLP) into the ALA (5-aminolevulinic acid) synthase apoenzyme by partially unfolding it and thereby generating an active form of ALA synthase and stimulates ALA synthesis, which is the first step of heme biosynthesis (Kardon et al., 2015; Kardon et al., 2020). The morpholino-mediated knockdown of ClpX in D. rerio. Zebrafish caused defects in red blood cell development, indicating that the ClpX is Important for vertebrate heme biosynthesis and erythropoiesis. Additionally, the dominant mutation in ClpX has been reported that results in the pathological accumulation of the heme biosynthesis intermediate protoporphyrin IX (PPIX) in erythroid cells, causing erythropoietic protoporphyria (EPP) in the affected patients (Yien et al., 2017).
The mitochondrial AAA+ family of proteases is believed to provide the first line of defense against any insults. The importance of this protease family is best exemplified by the severe neurodegenerative diseases caused by mutations in their respective genes, including Hereditary Spastic Paraplegia (HSP), SpinoCerebellar Ataxia, CODAS, and Perrault syndrome. While these proteases have been studied for decades, remarkably little is known about their precise biological roles, their substrates in the mitochondria, and the mechanisms by which their mutational inactivation causes disease in humans. It is not entirely known whether manipulating the levels of selective candidate AAA+ substrates influences the behavioral and neurodegenerative phenotypes of AAA+-deficient organisms.
A common characteristic of neurodegenerative diseases is the co-occurrence of dysfunctional mitochondria and cytoplasmic protein aggregates (Ross & Poirier, 2004; Ruan et al., 2020; Vicario et al., 2018; Gao et al., 2019). Recent work in yeast and cell lines suggests that such aggregates can be imported into the mitochondria and degraded by the AAA+ proteases (Ruan et al., 2017; Li et al., 2019; Hu et al., 2019). The various protease-deficient disease models offer an excellent way to test in a system whether the aggregates associated with common neurodegenerative diseases are degraded in the mitochondria. It is tempting to speculate if the overexpression and downregulation of AAA+ proteases in various models of Parkinson’s disease and Alzheimer’s disease (strains known to develop cytoplasmic protein aggregates) will modify the behavioral phenotypes and extend the life expectancy of the organism. The recent work showing that the TDP43 levels are regulated by Lon protease in the Drosophila model offers a glimpse of hope that AAA+ protease might have a role in the degradation of the cytoplasmic pathological aggregates (Wang et al., 2019).
One of the most exciting findings from recently published work on AAA+ proteases is that the respiratory chain defects resulting from the inactivation of the proteases are caused by a global reduction of the mitochondrial translation field (Zurita Rendón & Shoubridge, 2018; Münch & Harper, 2016). While extensive work on the ER and cytoplasmic protein unfolding stress response pathways have documented how unfolded protein stress triggers cytoplasmic translational arrest, the mechanism by which the mito-UPR triggers mitochondrial translational inhibition is not entirely understood. The recently published findings lay the foundation for studies aimed to test: (1) the precise mechanism by which the mito-UPR may inhibit mitochondrial translation. In the nematode Caenorhabditis elegans, the mito-UPR pathway requires the transcription factor ATFS-1; the mammalian ATFS-1 equivalent has only recently been identified (Fiorese et al., 2016). It is unclear if the targets of the ATFS-1 homolog in the mammalian system play any role in mitochondrial translational attenuation and chaperone upregulation under proteotoxic stress; (2) whether manipulations that oppose protein misfolding can restore translation under stress conditions; and (3) whether the mitochondrial translational inhibition during proteotoxic stress is beneficial or harmful. |
|
PMC9648357 | Li Wang,Chunmei Duan,Ruodan Wang,Lifa Chen,Yue Wang | Inflammation-related genes and immune infiltration landscape identified in kainite-induced temporal lobe epilepsy based on integrated bioinformatics analysis | 27-10-2022 | temporal lobe epilepsy,PPI,WGCNA,immunocyte infiltration,neuroinflammatory,kainite | Background Temporal lobe epilepsy (TLE) is a common brain disease. However, the pathogenesis of TLE and its relationship with immune infiltration remains unclear. We attempted to identify inflammation-related genes (IRGs) and the immune cell infiltration pattern involved in the pathological process of TLE via bioinformatics analysis. Materials and methods The GSE88992 dataset was downloaded from the Gene Expression Omnibus (GEO) database to perform differentially expressed genes screening and weighted gene co-expression network analysis (WGCNA). Subsequently, the functional enrichment analysis was performed to explore the biological function of the differentially expressed IRGs (DEIRGs). The hub genes were further identified by the CytoHubba algorithm and validated by an external dataset (GSE60772). Furthermore, the CIBERSORT algorithm was applied to assess the differential immune cell infiltration between control and TLE groups. Finally, we used the DGIbd database to screen the candidate drugs for TLE. Results 34 DEIRGs (33 up-regulated and 1 down-regulated gene) were identified, and they were significantly enriched in inflammation- and immune-related pathways. Subsequently, 4 hub DEIRGs (Ptgs2, Jun, Icam1, Il6) were further identified. Immune cell infiltration analysis revealed that T cells CD4 memory resting, NK cells activated, Monocytes and Dendritic cells activated were involved in the TLE development. Besides, there was a significant correlation between hub DEIRGs and some of the specific immune cells. Conclusion 4 hub DEIRGs (Ptgs2, Jun, Icam1, Il6) were associated with the pathogenesis of TLE via regulation of immune cell functions, which provided a novel perspective for the understanding of TLE. | Inflammation-related genes and immune infiltration landscape identified in kainite-induced temporal lobe epilepsy based on integrated bioinformatics analysis
Temporal lobe epilepsy (TLE) is a common brain disease. However, the pathogenesis of TLE and its relationship with immune infiltration remains unclear. We attempted to identify inflammation-related genes (IRGs) and the immune cell infiltration pattern involved in the pathological process of TLE via bioinformatics analysis.
The GSE88992 dataset was downloaded from the Gene Expression Omnibus (GEO) database to perform differentially expressed genes screening and weighted gene co-expression network analysis (WGCNA). Subsequently, the functional enrichment analysis was performed to explore the biological function of the differentially expressed IRGs (DEIRGs). The hub genes were further identified by the CytoHubba algorithm and validated by an external dataset (GSE60772). Furthermore, the CIBERSORT algorithm was applied to assess the differential immune cell infiltration between control and TLE groups. Finally, we used the DGIbd database to screen the candidate drugs for TLE.
34 DEIRGs (33 up-regulated and 1 down-regulated gene) were identified, and they were significantly enriched in inflammation- and immune-related pathways. Subsequently, 4 hub DEIRGs (Ptgs2, Jun, Icam1, Il6) were further identified. Immune cell infiltration analysis revealed that T cells CD4 memory resting, NK cells activated, Monocytes and Dendritic cells activated were involved in the TLE development. Besides, there was a significant correlation between hub DEIRGs and some of the specific immune cells.
4 hub DEIRGs (Ptgs2, Jun, Icam1, Il6) were associated with the pathogenesis of TLE via regulation of immune cell functions, which provided a novel perspective for the understanding of TLE.
Epilepsy is a chronic neurological disease in which patients suffer from the recurrence of unprovoked seizures and neurocognitive deficits, impacting more than 70 million people’s life worldwide (Englot et al., 2020). Temporal lobe epilepsy (TLE) is a common type of intractable epilepsy. Previous reports have indicated significant changes in multiple cognitive domains, such as judgment, problem-solving, language, executive functions, and memory (Hermann et al., 2006; Keary et al., 2007; Allone et al., 2017). Although lots of genes and pathways involved in the etiology of TLE have been widely reported, the precise pathological mechanisms are still unclear. Besides, there are no effective therapeutic targets or precise biomarkers for the occurrence and development of TLE. Accumulating evidence has revealed that neuroinflammatory, oxidative stress, and excitotoxicity contributed to posttraumatic epileptogenesis (Ambrogini et al., 2019; Eastman et al., 2020). Neuroinflammatory is one of the major biological processes in the development of brain pathology in epilepsy (Paudel et al., 2018; Rana and Musto, 2018). The inflammatory response to the insult induced a cascade of processes resulting in multiple pathological effects, including neurodegeneration, astrocyte dysfunction, damaged blood-brain barrier, long-term plastic alterations, and neuronal excitability. These lesions ultimately resulted in the occurrence and development of TLE (Suleymanova, 2021). Activation of neuroinflammation is commonly presented in epileptogenic brain areas and is implicated in animal models of epilepsy (Okuneva et al., 2016). Besides, IRGs in the molecular imaging and blood of neuroinflammation could provide diagnostic and predictive biomarkers for epilepsy (Vezzani et al., 2019). Furthermore, anti-inflammatory small molecules had the potential therapeutic effects of inhibiting brain inflammation in the treatment of epilepsy (Radu et al., 2017). The inflammatory responses induced the secretion of pro-inflammatory factors, which resulted in blood-brain barrier dysfunction as well as the involvement of resident immune cells in the inflammation pathway (Matin et al., 2015). Experimental models have demonstrated that neural injury and the onset of recurrent epilepsy are regulated by interactions between adaptive and innate immunity (Falsaperla et al., 2014; Vitaliti et al., 2014). Immunocyte infiltration takes part in the development of epilepsy primarily via exacerbating blood-brain barrier damage and activating brain inflammation (Wang et al., 2014). For example, the expression of the chemokines positively correlated with neutrophil infiltration in a rat seizure model (Johnson et al., 2011). Infiltration of T lymphocytes is also implicated in the pathology of epilepsy (Nakahara et al., 2010). Furthermore, PD-1 is an immune-inflammatory diagnostic biomarker for patients with intractable epilepsy (Tang and Wang, 2021). Based on these studies, an inflammatory- and immune-related therapeutic schedule could be effective in the prevention and treatment of TLE. In the present study, we identified inflammation-related genes (IRGs) from the GSE88992 dataset using a combined approach of differentially expressed gene analysis and weighted gene coexpression network analysis. Subsequently, we identified the differential immunocyte types associated with TLE development. And the relationship between immune cell infiltrates and IRGs were evaluated.
We downloaded the gene expression profiles of the GSE88992 dataset (Mus musculus) with 9 control samples and 8 TLE samples from the Gene Expression Omnibus (GEO) database. The mouse model of TLE was established by intrahippocampal microinjection of kainate. We collected the IRGs list from the GeneCards database, the threshold for IRGs screening was set as follows: the relevance score > 5. GSE60772 dataset (Mus musculus) was used as an external dataset. The microarray data were normalized with the normalize between arrays function of the limma package.
We used the limma package of R software to identify the DEGs between the control and TLE groups. The screening criteria were p.adj < 0.05 and |log2 FC| > 1. The volcano and heatmap of DEGs were generated by the ggplot2 package and the Complex Heatmap package in R software, respectively.
We used the WGCNA R package to carry out the WGCNA analysis (Langfelder and Horvath, 2008). A total of 20,812 genes were identified from each sample of GSE88992 dataset. Firstly, the Pearson correlation coefficient between every two genes was calculated to construct the similarity matrix. In the present study, we selected the power value β = 6 (scale free R2 = 0.91) and mean connectivity = 10.54. Then, the similarity matrix was transformed into a topological overlap matrix (TOM). Next, hierarchical clustering was performed to identify modules, the minimum size cutoff value was 30 and the depth segmentation value was 3. The eigengene was calculated, the modules were clustered, and the similar modules were merged (module merge threshold = 0.25). Subsequently, the gene modules that were related to the clinical traits of TLE were identified. Finally, the potential hub genes were identified based on the criteria: gene significance (GS) > 0.8 and module significance (MS) > 0.8.
The cluster Profiler package of R software was used to perform the KEGG and biological process (BP), cellular component (CC), and molecular function (MF) enrichment analyses of DEIRGs. P < 0.05 was considered significant enrichment (Yu et al., 2012).
First, the PPI network of IRGs was established using a STRING and Cytoscape (v 3.7.2) software. Then, we selected the top 8 genes by eight topological analysis algorithms in cytoHubba, namely, radiality, MNC, EPC, MCC, degree, closeness, betweenness, and bottleneck (Chin et al., 2014). The hub genes were identified using an R package “UpSet” (Lex et al., 2014).
GSEA was carried out via GSEA version 4.1.0 software to further explore the potential biological functions of genes in the GSE88992 dataset (Tragante et al., 2017). False discovery rate (FDR) < 0.25 and p.adjust < 0.05 and were considered significant enrichment.
The gene expression profiles of the GSE88992 dataset were uploaded to CIBERSORT to evaluate immune cell infiltration of TLE samples and control samples (Newman et al., 2015). The ggplot2 package and the ComplexHeatmap package in R software were used to visualize the infiltration level of 22 types of immune cells between the control and TLE groups.
The correlation analysis between the infiltrating immune cells and hub genes (Ptgs2, Jun, Icam1, and Il6) was carried out via Spearman, the results were visualized via the ggplot2 package of R software.
The male C57BL/6 mice (8 weeks old and 20–24 g) were obtained from the Experimental Animal Center of Chongqing. All animal experimental protocols were approved by the Animal Care Committee of Army Medical University. All animals were randomly divided into two groups, including control group (Con, n = 6) and kainite model group (TLE, n = 6). Mice were anesthetized with a combination of xylazine (8 mg/kg, i.p.) and ketamine (100 mg/kg, i.p.), and placed symmetrically in the stereotactic apparatus and then drilled into the skull. Then, freshly prepared kainite solution (20 mmol/L in saline) (Araki et al., 2020) was microinjected into the right side of the hippocampus by a microsyringe to induce TLE. The microsyringe was slowly withdrawn and the scalp was stitched. The Con of mice was given the same amount of saline. During the first 4 h after kainite injection, the severity of TLE was evaluated based on the Racine scoring system: 0, no reaction; 1, mild facial clonus, eye blinking, and/or stereotypic mounting; 3, myoclonic jerks in the forelimbs; 4, clonic convulsions in the forelimbs with rearing; and 5, generalized clonic convulsions and loss of balance (Khamse et al., 2015). Total RNA of hippocampus tissues were extracted using TRIzol reagent (Ambion). 2 μg of purified RNA was reverse transcribed into cDNA using cDNA synthesis kit (Bio-Rad, USA). Then, qRT-PCR was carried out using SYBR Green Master Mix reagent (VAZYME) on the PCR System (Applied Biosystems, CA, USA). The mRNA expression levels of genes were normalized using the 2–ΔΔCt method. The sequences of primers used in this study were presented in Table 1.
Drug-Gene Interaction Database has drug-gene interaction data from 30 disparate sources including PharmGKB, NCBI Entrez, clinical trial databases, Ensembl, DrugBank, ChEMBL, and literature in NCBI PubMed. The four core IRGs were selected as the potential targets and imported into the Drug-Gene Interaction Database (DGIdb) to search for existing inhibitors or agonists, then the FDA-approved drugs with agonist or antagonist effects were identified. Finally, the interaction network between the core IRGs and the potential drugs was constructed by Cytoscape software.
First, the data from the GSE88992 dataset were normalized with the normalize between arrays function of the limma package (Figure 1A). And the results of PCA analysis indicated a significant difference in gene expression between control samples and TLE samples (Figure 1B). Therefore, we performed DEGs analysis in the next step. As shown in Figure 1C and Supplementary Table 1, we identified a total of 613 DEGs between TLE and control groups, including 474 up-regulated genes (red dots) and 139 down-regulated genes (green dots). The mRNA expression level of the top 40 DEGs was visualized in Figure 1D.
A total of 20,812 genes obtained from the gene expression profiles of the GSE88992 dataset were used to perform WGCNA. Then, hub modules were identified based on average dynamic tree clipping and hierarchical clustering (Supplementary Figures 1A,B). Two hub modules (blue and turquoise) were visualized in Supplementary Figure 1C and were most correlated with the TLE and selected for further investigation. As shown in Supplementary Figures 1D,E, we screened out the hub genes based on MM > 0.8 and GS > 0.8, and 2,312 and 2,651 key genes were identified in the blue and turquoise modules, respectively (Supplementary Table 2).
A total of 403 IRGs were obtained from GeneCards database (Supplementary Table 3). We intersected the most clinically significant modules (blue and turquoise) with the DEGs and IRGs to further identify 34 core IRGs (Figure 2A). The mRNA expression level of the 34 core IRGs in the GSE88992 dataset was visualized in Figure 2B. Subsequently, we performed a functional enrichment analysis of 34 core IRGs to obtain a better understanding of the underlying BP associated with TLE. As shown in Figures 2C,D, the GO-BP enrichment analysis indicated that the core IRGs involved in leukocyte migration, positive regulation of response to external stimulus, leukocyte chemotaxis, myeloid leukocyte migration, and cell chemotaxis; for GO-CC, the core IRGs mainly enriched in membrane raft, membrane microdomain, membrane region, receptor complex, and extracellular exosome; for GO-MF, the core IRGs significantly enriched in cytokine activity, receptor ligand activity, chemokine activity, G protein-coupled receptor binding, and chemokine receptor binding (Table 2). As shown in Figures 2E,F, the KEGG enrichment analysis indicated that the core IRGs enriched in TNF signaling pathway, IL-17 signaling pathway, Toll-like receptor signaling pathway, AGE-RAGE signaling pathway in diabetic complications, and Rheumatoid arthritis (Table 2).
The top 8 genes were selected by 8 topological analysis algorithms (radiality, MNC, EPC, MCC, degree, closeness, betweenness, and bottleneck) in cytoHubba. Then, we used the R package “UpSet” to identify four hub genes, including Ptgs2, Jun, Icam1, and Il6 (Figure 3A). We also drew the heatmap of four hub genes in the GSE88992 dataset to visualize the mRNA expression levels (Figure 3B). Besides, we used an external dataset (GSE60772) and established a model of TLE to verify the results. As presented in Supplementary Table 4, Con group showed no signs of seizure activity and behavior during the first 4 h after kainite injection. However, all mice (100%) in TLE group showed high scores of seizures (p < 0.001), indicating that the TLE model was established. As shown in Figures 4A,B, the mRNA expression levels of Ptgs2, Jun, Icam1, and Il6 in the control group were significantly lower than those in the TLE group (p < 0.05), the results were consistent with those of bioinformatics screening.
As shown in Figure 5, our results revealed that most of the enriched gene sets were involved in immune- and inflammation-related pathways, including neutrophil degranulation, leukocyte transendothelial migration, disease of the immune system, cytokines and inflammatory response, inflammatory response pathway, IL18 signaling pathway, TNF alpha signaling pathway, and senescence and autophagy in cancer. These findings indicated immune response and inflammation may play a vital role in the pathological process of TLE.
In this study, we performed immune infiltration analysis to identify the immune cell types potentially involved in the pathology of TLE. As shown in Figure 6A, compared to those in the control group, the score of T cells CD4 memory resting and dendritic cells activated were relatively low, while the score of NK cells activated and monocytes were relatively high in the TLE group. The proportions of the 22 types of immunocyte cells were visualized in Figure 6B.
Our correlation analysis indicated that Icam1 was negatively with macrophages M2 (r = −0.578, p = 0.015) and T cells CD4 memory resting (r = −0.575, p = 0.015), and positively correlated with dendritic cells activated (r = 0.715, p = 0.001), mast cells activated (r = 0.634, p = 0.006), NK cells activated (r = 0.588, p = 0.013), and monocytes (r = 0.558, p = 0.019) (Figure 7A). Il6 was negatively with T cells CD4 memory resting (r = −0.654, p = 0.004), and positively correlated with Dendritic cells activated (r = 0.723, p = 0.001), NK cells activated (r = 0.622, p = 0.007), and Mast cells activated (r = 0.580, p = 0.014) (Figure 7B). Jun was negatively with T cells CD4 memory resting (r = −0.786, p = 0.0002), and positively correlated with Dendritic cells activated (r = 0.822, p = 0.0000), Monocytes (r = 0.634, p = 0.006), and NK cells activated (r = 0.514, p = 0.034) (Figure 7C). Ptgs2 was negatively with T cells CD4 memory resting (r = −0.649, p = 0.004), and positively correlated with Dendritic cells activated (r = 0.705, p = 0.001), Monocytes (r = 0.571, p = 0.016), Mast cells activated (r = 0.571, p = 0.016), and NK cells activated (r = 0.531, p = 0.028) (Figure 7D).
As shown in Figure 8, we investigated the hub gene-drug interactions of the 31 potential drugs and 4 hub genes for possibly treating TLE. As shown in Table 3, the potential 31 candidate drugs were all approved by FDA. Most of these potential drugs might interact with the Ptgs2 (13/31). Promising drugs of these hub genes included ETORICOXIB, ALICAFORSEN, SILTUXIMAB, OLOKIZUMAB, etc.
Epilepsy is a brain disorder characterized by recurrent seizures. Imbalanced regulation of inflammation contributes to epilepsy development (Rana and Musto, 2018). Neuroinflammation is associated with an increased risk for epilepsy in related animal models and humans (Terrone et al., 2020). Increasing evidence indicated that systemic autoimmune disorders are the major pathogenesis of epilepsy (Steriade et al., 2021). Investigating the biological functions of immunity and inflammation in epileptogenesis will help in the identification of novel biomarkers for better screening of patients with epilepsy. Therefore, it is of great importance to further understand the role of IRGs and immune cell infiltration in epilepsy, and thus identify novel therapeutic targets. The present study aimed to systematically assess potential IRGs and investigated the correlations of immune cell infiltration in TLE. In the present study, we screened 613 DEGs between TLE and the control group via bioinformatics methods. Then, we constructed a weighted co-expression network using WGCNA and identified hub modules for TLE, and four IRGs (Ptgs2, Jun, Icam1, Il6) were identified as hub genes. Besides, the proportions of T cells CD4 memory resting, dendritic cells activated, NK cells activated and monocytes were significant differences between TLE and control group, implying that immune response is involved in the TLE development. Furthermore, our results showed that these four IRGs were associated with immune cell infiltration in TLE, which could guide the selection of immunotherapy in the treatment of TLE. An important result of this research was that the up-regulation of Ptgs2, Jun, Icam1, and Il6 was closely associated with the development of TLE. Prostaglandin-endoperoxide synthase 2 (Ptgs2) plays an important role in regulating inflammatory responses. It has been reported that Ptgs2 is dynamically regulated during the initiation and resolution of acute inflammation (Hellmann et al., 2015). Besides, down-regulation of peripheral PTGS2 in response to valproate treatment in epileptic (Rawat et al., 2020). Jun is a proto-oncogene, which encoded a variety of key regulatory proteins in the normal tissues. A previous study has indicated that Jun had the potential to serve as a prognostic marker for neuropathic pain (Yang et al., 2018). JUN was identified as a potential biomarker for the prognosis of brain low-grade gliomas (Xu et al., 2020). Besides, the c-JUN-medicated signaling pathway was an important BP involved in the pathogenesis of hippocampal neuronal damage in kainite-induced epilepsy (Lee et al., 2001). Intercellular adhesion molecule 1 (Icam1) is a cell surface glycoprotein that was reported for driving inflammatory responses in the onset of pathologic conditions. For example, increases in soluble Icam1 were associated with inflammatory responses, and some clinical studies used soluble Icam1 as a biomarker to classify patients with inflammatory diseases, as well as non-infectious systemic inflammatory responses syndrome (de Pablo et al., 2013). Icam1 was also reported to play an important role in regulating leukocyte trafficking and transendothelial migration (Kanters et al., 2008). Besides, expression of Icam1 was significantly up-regulated in the hippocampi with hippocampal sclerosis (Nakahara et al., 2010). Interleukin-6 (Il6) is a prototypical proinflammatory cytokine for maintaining homeostasis, and animal experiments showed the involvement of cytokines in epilepsy (Shen et al., 2020). The previous report has revealed that an increased level of IL6 was observed in cerebrospinal fluid and serum from patients with epilepsy (Peltola et al., 1998; Tao et al., 2020). Therefore, we speculated that these hub genes may play an important role in epilepsy. Accumulating evidence has indicated that the activation of immune responses in the brain may be implicated in epilepsy (Granata et al., 2011). Peripheral immune cell invasion into the brain, along with cerebrovascular dysfunction and neuroinflammation, was implicated in the pathogenesis of epilepsy (Yamanaka et al., 2021). A previous study has reported that leukocytes infiltration in brain tissue is involved in the pathogenesis of TLE (Zattoni et al., 2011). As previously indicated, both microglial proliferation and monocytes recruitment resulted in significant microgliosis following epilepsy (Feng et al., 2019). Besides, infiltrating monocytes play an important role in the pathological process of epilepsy (Bosco et al., 2020). It has been reported that immunodeficient (depletion of T cells CD4) mice exhibited significant shortening of the latent stage before epilepsy onset, suggesting the protective effect of T cells in epileptogenesis (Deprez et al., 2011). Furthermore, compared with the healthy individuals, a significant increase of natural killer (NK) cells, while a significant decrease of CD4 T cells was observed in patients with TLE (Bauer et al., 2008). These results were in agreement with our findings. We also used CIBERSORT to comprehensively assess TLE immune cell infiltration to further investigate the role of immunocyte infiltration in TLE development. In this study, the score of T cells CD4 memory resting and dendritic cells activated were relatively low, while the score of NK cells activated and monocytes was relatively high in the TLE group. Moreover, the correlation analysis between Ptgs2, Jun, Icam1, Il6, and immune cells indicated that all these hub gene expressions were positively correlated with NK cells activated and negatively correlated with T cells CD4 memory resting. These findings further indicated the importance of immunocyte infiltration in the pathogenesis of TLE. We speculated that up-regulation of Ptgs2, Jun, Icam1, and Il6 contributes to NK cells activated infiltration and inhibits T cells CD4 memory resting infiltration, and subsequently results in the development of TLE. However, these results require additional in vivo or in vitro experiments to prove the complex interactions between hub genes and immunocyte infiltration.
In conclusion, the present study combined WGCNA and DEG analyses to identify potential genes (Ptgs2, Jun, Icam1, and Il6) associated with specific immune cells involved in TLE development. Up-regulation of Ptgs2, Jun, Icam1, and Il6 contributes to TLE development via regulation of T cells CD4 and NK cells. Our findings provided a novel insight into the immunocyte infiltration pattern of TLE and its potential immune regulatory mechanisms.
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/geo/, GSE88992 and GSE60772.
The animal study was reviewed and approved by the Animal Care Committee of Army Medical University.
LW took part in drafting the manuscript. CD made substantial contributions to conception and design. RW and LC performed data analysis and interpretation of data. YW revised it critically for important intellectual content and gave final approval of the version to be published. All authors have read and agreed to the published version of the manuscript. |
|
PMC9648375 | 36345785 | Barbara Kaproń,Anita Płazińska,Wojciech Płaziński,Tomasz Plech | Identification of the first-in-class dual inhibitors of human DNA topoisomerase IIα and indoleamine-2,3-dioxygenase 1 (IDO 1) with strong anticancer properties | 08-11-2022 | Antiproliferative activity,docking simulations,immunotherapy,thiosemicarbazide derivatives | Abstract Molecular docking of a large set of thiosemicarbazide-based ligands resulted in obtaining compounds that inhibited both human DNA topoisomerase IIα and indoleamine-2,3-dioxygenase-1 (IDO1). To the best of our knowledge, these compounds are the first dual inhibitors targeting these two enzymes. As both of them participate in the anticancer response, the effect of the compounds on a panel of cancer cell lines was examined. Among the cell lines tested, lung cancer (A549) and melanoma (A375) cells were the most sensitive to compounds 1 (IC50=0.23 µg/ml), 2 (IC50=0.83 µg/ml) and 3 (IC50=0.25 µg/ml). The observed activity was even 90-fold higher than that of etoposide, with selectivity index values reaching 125. In-silico simulations showed that contact between 1-3 and human DNA topoisomerase II was maintained through aromatic moieties located at limiting edges of ligand molecules and intensive interactions of the thiosemicarbazide core with the DNA fragments present in the catalytic site of the enzyme. | Identification of the first-in-class dual inhibitors of human DNA topoisomerase IIα and indoleamine-2,3-dioxygenase 1 (IDO 1) with strong anticancer properties
Molecular docking of a large set of thiosemicarbazide-based ligands resulted in obtaining compounds that inhibited both human DNA topoisomerase IIα and indoleamine-2,3-dioxygenase-1 (IDO1). To the best of our knowledge, these compounds are the first dual inhibitors targeting these two enzymes. As both of them participate in the anticancer response, the effect of the compounds on a panel of cancer cell lines was examined. Among the cell lines tested, lung cancer (A549) and melanoma (A375) cells were the most sensitive to compounds 1 (IC50=0.23 µg/ml), 2 (IC50=0.83 µg/ml) and 3 (IC50=0.25 µg/ml). The observed activity was even 90-fold higher than that of etoposide, with selectivity index values reaching 125. In-silico simulations showed that contact between 1-3 and human DNA topoisomerase II was maintained through aromatic moieties located at limiting edges of ligand molecules and intensive interactions of the thiosemicarbazide core with the DNA fragments present in the catalytic site of the enzyme.
According to WHO report (2020), cancer constitutes one of leading cause of death worldwide. The number of cancer deaths is estimated to be approximately 10 million per year. Only in the USA, about 1.9 million new cases of cancer are expected to be diagnosed in 2022. At the same time, the predicted number of cancer deaths in Europe is approximately 1.2 million, According to National Institute of Health (NIH, USA) overall medical costs associated with cancer care in the USA are projected to reach almost 158 billion of dollars. Further estimations showed that global expenditures for cancer grew to 2.5 trillion of dollars. The above-mentioned epidemiological data confirm that cancer is increasingly a global health-care problem that needs urgent action. From the biological point of view, the basic feature of cancer is uncontrolled growth (proliferation) and spread of abnormal cells from the place of origin to another part of the body. Inhibition of the uncontrolled proliferation is one of the main goal of anticancer therapy. Such effect may be obtained, for example, through the use of chemical compounds that affect replication of DNA in cancer cells. For this reason, human DNA topoisomerase inhibitors (e.g. etoposide, topotecan, irinotecan, mitoxantrone, doxorubicin) are effective anticancer agents. Human DNA topoisomerases catalyse topological changes in single- or double-stranded helices of DNA. These enzymes are necessary in the processes of replication, transcription, recombination, and significantly contribute to the genome stability maintenance. Moreover, since human DNA topoisomerases are directly involved in DNA repair, the application of topoisomerase inhibitors as concomitant drugs during radiotherapy or chemotherapy is one of the strategies enhancing the effectiveness of cancer treatment. In the recent years it turned out that 1,4-disubstituted thiosemicarbazide derivatives have the ability to inhibit human DNA topoisomerases and, subsequently, to reduce the viability of cancer cells (Figure 1). Interestingly, anticancer properties of some of the thiosemicarbazide derivatives have been previously known, but their antiproliferative activity had not been linked to DNA topoisomerase inhibition. Therefore, the concept of use of 1,4-disubstituted thiosemicarbazides as anticancer agents acting on the mentioned enzyme is relatively new and not well explored. Much more is known about anticancer properties of thiosemicarbazones, that constitute close analogs of thiosemicarbazides. However, thiosemicarbazones and their metal complexes exhibit antiproliferative activity through the distinct molecular mechanism, that is, through the inhibition of ribonucleotide diphosphate reductase. During our previous studies, we identified a group of thiosemicarbazide-based human DNA topoisomerase II inhibitors (Figure 1, lower row) that decreased viability of cancer cells and inhibited intracellular biosynthesis of their DNA much stronger than etoposide – that is, clinically relevant topoisomerase II inhibitor. Since molecular docking simulations provides insight into the conformation of ligands within the active site of target protein, such approach is an integral part of drug design and discovery. Therefore, in the herein described studies, a large set of 1,4-disubstituted thiosemicarbazides were analysed through the use of molecular docking procedure in order to predict structural features of the most potent inhibitors of human DNA topoisomerase II. Combined application of in-silico and in-vitro techniques enabled us to identify new inhibitors characterised by strong cytotoxic/antiproliferative properties and high selectivity against cancer cells. Moreover, we have discovered and described the first-in-class dual inhibitors of human DNA topoisomerase II/indoleamine-2,3-dioxygenase 1 (IDO 1) that can lead to the future use of thiosemicarbazide derivatives as relevant components of anticancer immunotherapy.
The analysed dataset of ligands included molecules composed of thiosemicarbazide skeleton enriched with structurally diverse R1 and R2 substituents (Table 1). Designed ligands had different aryl, aroyl, alkyl (branched and unbranched), heteroaryl groups at distal positions of thiosemicarbazide core (see Supplementary material). The full set of ligands was docked to a single protein topoisomerase II structure found in the PDB database under entry: 3qx3 (X-ray resolution: 2.16 Å, i.e. the highest available among topoisomerase II structures deposited in the PDB database). The binding energy obtained varied from ca. −6 to ca. −11.8 kcal/mol (vs. −14.4 kcal/mol for native ligand), thus, displaying a broad range of affinities towards the molecular target. The aim of this stage was to identify the potentially most potent compounds, exhibiting the lowest protein-ligand binding free energy. Using this criterion, twelve compounds, characterised by the most favourable binding mode (associated with the lower binding energy), were selected for the synthesis and for further in-vitro experiments (Table 1). Compounds 1-12 were obtained in good yields (70–92%) in a one-step reaction between respective carboxylic acid hydrazides and isothiocyanates using procedure described in the literature and in our previous paper. Due to low solubility in culture media and buffers, compounds 11 and 12 were excluded from further in vitro experiments.
The inhibitory effect of compounds 1-10 and etoposide (positive control) was evaluated using human topoisomerase IIα relaxation assay. As relaxation of supercoiled DNA (scDNA) is one of numerous functions of topoisomerase IIα, so the appearance/disappearance of relaxed DNA bands during agarose gel electrophoresis can be used to identify compounds that inhibit the enzyme. Among the investigated thiosemicarbazide derivatives, compounds 1–3 turned out to inhibit the activity of human DNA topoisomerase Iiα as indicated by the increased intensity of scDNA band and disappearance of relaxed DNA bands (Figure S2, Figure S3). Compound 2 influenced the enzyme activity in a dose-dependent manner with IC50 at ∼43µM (Table 2). The other two compounds (1, 3) exhibited lower inhibitory potency against the enzyme (IC50 ∼71 and 76 µM, respectively), however it was still higher than that of etoposide (IC50 ∼123µM). Importantly, these results are in line with those obtained from docking simulations, since compound 2 exhibited the most favourable free energy of binding (ΔG of −11.8 kcal/mol). During further enzymatic experiments, we also found that compounds 1-3 share the same mechanism of topoisomerase IIα inhibition with etoposide. Both thiosemicarbazide derivatives and etoposide are able to stabilise the covalent DNA-enzyme cleavage complex, which is manifested by the appearance of linear DNA cleavage intermediate (Figure S4). Therefore, compounds 1-3 were found not only to act as inhibitors of human DNA topoisomerase IIα catalytic activity (i.e. relaxation activity), but were also found to be a topoisomerase IIα poisons.
Since the inhibition of human DNA topoisomerase II should result in decreased proliferation of cancer cells, compounds 1-3 were tested using BrdU assay against a panel of cancer cell lines, including MCF-7 (estrogen-dependent human breast cancer cells), MDA-MB-231 (estrogen-independent human breast cancer cells), SCC-25 (squamous cell carcinoma of the tongue), FaDu (squamous cell carcinoma of the pharynx), A549 (human lung carcinoma), AGS (human gastric adenocarcinoma), LS-180, HT-29 (colon cancer cell lines), T98G (glioblastoma cells), and A375 (melanoma cells). Human normal skin fibroblasts (CRL-2072) were also used as a reference cell line. The BrdU assay allows quantification of cell proliferation since it measures incorporation of thymidine analogue (BrdU) during DNA replication. The examined compounds exhibited wide spectrum of anticancer activity and they inhibited the growth of all types of cancer cells much stronger than reference drug – etoposide (Table 3). Among the cell lines tested, lung cancer (A549) cells were the most sensitive to compounds 1 (IC50=0.23 µg/ml), 2 (IC50=0.83 µg/ml) and 3 (IC50=0.25 µg/ml). These three compounds exhibited strong antiproliferative activity against A375 melanoma cells as well (IC50 from 0.56 to 0.96 µg/ml). What is important, while compounds 2 and 3 showed higher selectivity towards A549 and A375 cells, thiosemicarbazide derivative 1 had relatively constant and strong antiproliferative effect over most of the investigated cell lines. Its activity was even 90-fold higher than that of etoposide, with selectivity index values reaching 75 (vs. human normal skin fibroblasts). However, although the antiproliferative effect caused by 3 was mostly weaker (except for A549 and A375) than observed for 1, but the former compound was also characterised by lower toxicity against human normal cells. It results in selectivity index approaching 125 and 57, when it comes to lung cancer (A549) and melanoma cells (A375), respectively. Having in mind that different molecules are able to inhibit the growth of cancer cells through more than one molecular mechanism, it is possible that also other (i.e. other than 1-3) representatives of the set of synthesised thiosemicarbazide derivatives can exhibit anticancer activity that is unrelated to DNA topoisomerase II inhibition. Therefore, the whole set of compounds were tested by using MTT assay in order to check their overall cytotoxic effect. The growth of cells was monitored after 24 and 48 h of incubation with the increased concentrations (1–100 µg/ml) of the thiosemicarbazide derivatives. The results of MTT assay (Table 4) proved that mainly DNA topoisomerase II inhibition contribute to the anticancer effect of the designed compounds, since compounds 1-3 were characterised by the most potent cytotoxic effect in relation to the other thiosemicarbazide derivatives examined in this study. However, even those compounds that lacked inhibitory effect on DNA topoisomerase II exhibited promising cytotoxic activity against some of the cancer cell lines tested, especially when it was monitored after 48 h of incubation. For example, compound 6 was still up to 10-fold more active than etoposide (against T98G cells), with low cytotoxic properties (IC50 > 100 µg/ml) against human normal cells (HEK-293). This confirms that the compounds tested combine different mechanisms of anticancer activity, and DNA topoisomerase II inhibition is only one of them. The results obtained from MTT assay enabled to conlude that R2 substituents play a key role in the anticancer activity of thiosemicarbazide derivatives. It can be observed that bulky substituents (e.g. phenylazophenyl, diisopropylphenyl groups) at N4-position of the thiosemicarbazide core affect topoisomerase II inhibition and cytotoxic properties of the investigated group of compounds. In order to mechanistically explain the primary mode of anticancer action of compounds 1-3, further in-silico simulations were performed.
Three ligands (i.e. compounds 1-3), characterised by exceptionally high binding energies as well as by the favourable experimentally determined properties were docked to 6 further (apart from PDB:3qx3) protein structures of topoisomerase II, found in the following PDB entries: 4g0u, 4g0v, 4j3n, 5gwi, 5gwj and 5gwk of resolutions of 2.70, 2.55, 2.30, 2.74, 2.57 and 3.15 Å, respectively. These structures were also cleaned by removing the co-crystalized ligands and ions (if present) but the DNA fragments were left unchanged. Note that these structures belong either to topoisomerase IIα or IIβ, however, due to high structural similarity within the region of binding cavity, we decided to use both those subtypes. Moreover, the high number of structures used in this stage of the study allowed to account for the inherent flexibility of the DNA strand, exhibiting tight contact with the ligand, according to the available XRD data. Figure 2 shows the typical conformational arrangements found for ligands 1-3 interacting with binding cavity. The most frequent pattern of interactions involves: (1) location of the ligand molecule in the protein cleavage, i.e. the binding site for DNA strand; (2) interactions of the ligand with two protein domains separated by the above-mentioned cleavage; (3) intensive interactions within the DNA fragments, present in the binding cavity. All these three aspects are characteristic for native ligands, co-crystalized with the protein structures considered in the current study. At the same time, depending on the ligand type-protein structure combination, we observed significant scatter of possible conformational arrangements of ligand molecules in the binding cavity. Apart from representative poses shown in Figure 2, alternative poses appeared, differing by the opposite orientation of the ligand in the cavity or interactions with only one site of protein cleavage (and the DNA strand). Although the presence of oppositely oriented poses is expected due to roughly symmetrical structure of ligands (aromatic moieties located at both edges of molecules), the interactions with only one site of catalytic cleavage is contradictory with the available structural data and the knowledge about the mechanism of action of topoisomerases. For this reason (and due to notably higher binding energies), such alternative poses were not analysed in detail. The order of average binding energies (averaged over all seven protein structures of topoisomerase II) obtained for the three lead compounds varies in a rather minor range as follows: −9.83 ± 0.47 (compound 2); −9.77 ± 0.53 (compound 3); −9.23 ± 0.46 kcal/mol (compound 1). This suggests similar potency of all compounds for binding to DNA topoisomerase II. The results of the docking studies have also been analysed with respect to the mechanistic interaction patterns that may be significant in the context of interpretation of the ligand-protein affinities. The summary given below relies on analysing the ligand-protein contacts that occur if the distance between any corresponding atom pair is smaller than the arbitrarily accepted value of 0.4 nm. When considering the most energetically favourable poses, all the studied compounds dock to the protein-DNA complex in a very similar manner (see Figure 3). The contact with protein is maintained through aromatic moieties located at limiting edges of ligand molecules, as well as through smaller substituent of those groups. In the case of compound 2, an additional contact site is created by the azo moiety linking the two phenyl rings. Those contacts involve either sidechains of backbone fragments belonging to the two different protein domains located at the two sides of catalytic cleavage. Namely, sidechains of Met782, Gln778 and Pro819 interact with phenyl rings of ligand molecule through CH–π or π–π interactions. Some hydrogen bonding between Gln sidechain and the azo moiety is also possible. At the other side of the cleavage, the remaining aromatic groups of the ligand interact with sidechains of Arg503, Glu477, Asp479 and backbone fragments of Gly504, Ser480 and Leu502. The character of interactions varies from CH-π interactions to hydrogen bonding (involving mainly Arg503). As indicated by the higher number of potential contacts, interactions at this side of the cleavage are most intensive and stabilise the ligand position to larger extent in comparison to its second limiting edge (Figure 3). Apart from protein-related contacts, all ligands exhibit intensive interactions with the DNA fragments present in all crystal structures of topoisomerase II selected to be used in the docking simulations. Such interactions involve the central part of the ligand molecules, composed of the same chemical motif: -CH2-C(=O)-NH-NH-C(=S)-NH-. Here, the attractive character of such contacts is the consequence of either CH-π interactions (involving aromatic nucleobases) or hydrogen bonding with ligand molecule playing a role of either donor or acceptor.
The concept of use of thiosemicarbazide core as an attractive scaffold in anticancer drug design and development was enriched by the results obtained by Serra et al. Their studies demonstrated that some of the thiosemicarbazide derivatives highlighted inhibitory effect on indoleamine-2,3-dioxygenase (IDO), which is a promising target for anticancer immunotherapy. The use of IDO inhibitors may be a useful strategy to overcome tumour-induced immunosuppression. Moreover, numerous in-vitro and in-vivo clinical trials confirmed that combined application of IDO 1 inhibitors with classical chemotherapy or radiotherapy improved the outcomes of the treatment. Taking into consideration that dual inhibitors, acting both on human topoisomerase II and IDO1, could possess beneficial anticancer properties, the final stage of the study aimed to check if the compounds 1-3 are able to inhibit these two enzymes simultaneously. Preliminary studies demonstrated that compound 1 has almost the same inhibitory activity against IDO 1 as that of epacadostat, which is very promising anticancer drug candidate (Table 5). The other two thiosemicarbazide derivatives (2, 3) were also found as IDO 1 inhibitors, however with relatively lower inhibitory potency. As far as anticancer properties of human DNA topoisomerase inhibitors (e.g. etoposide, topotecan, irinotecan, mitoxantrone, doxorubicin) are well known, contribution of IDO 1 to anticancer effect has been undiscovered for quite a long time. Indoleamine-2,3-dioxygenase (IDO) belongs to the group of cytoplasmic enzymes catalysing the degradation process of tryptophan. As a result of this enzymatic reaction, compounds called kynurenines are formed, which include N-formylkynurenine, L-kynurenine, kynurenine acid, 3-hydroxykynurenine, 3-hydroxyanthranilic acid, quinolinic acid and others. The final effect of these changes is the oxidised form of nicotinamide adenine dinucleotide (NAD+). For many years, the IDO enzyme was only attributed to participation in cellular energy processes. However, IDO, and in particular IDO-1, inhibits the immune system and leads to a state of tolerance to specific antigens. Unfortunately, the same mechanism that allows to protect human organism before the development of autoimmune diseases is used by the cancer to develop and to metastasise in the human body. IDO 1 is overexpressed by tumour cells to escape from a potentially effective immune response. Excessive degradation of tryptophan within the tumour microenvironment leads to immunosuppression and inhibit the function of T lymphocytes and NK cells, mainly due to GCN2 kinase and as a result of inhibition of mTOR signaling,. The result is reduced proliferation of T lymphocytes and their transition into a state of anergy, which causes them to lose their ability to destroy cancer cells. In addition, inhibition of the mTOR pathway leads to the differentiation of T lymphocytes towards regulatory cells that suppress the immune response. Because IDO is involved in the process of tumour growth, high hopes are associated with the use of small molecule inhibitors of IDO. Currently, the effectiveness of IDO inhibitors, for example, epacadostat, indoximod, navoximod, is being tested at various stages of clinical trials. In preclinical models, inhibition of the IDO 1 has been shown to enhance the efficacy of cytotoxic chemotherapy, radiotherapy and immunotherapy without increasing side effects. Importantly, immunotherapy has come to be a dominant treatment option in melanoma and lung cancer,, while the herein described compounds 1-3 possess superior antiproliferative activity against those two types of cancer. Therefore, the above-mentioned dual inhibitors of human DNA topoisomerase II and IDO 1 pave the way to use such thiosemicarbazide derivatives as an effective anticancer strategy to be explored more in further preclinical and clinical studies.
Summarising, three novel and potent dual inhibitors of human DNA topoisomerase II and IDO 1 were discovered using computer-aided drug design techniques. The obtained thiosemicarbazide derivatives exhibited a wide spectrum of antiproliferative and cytotoxic properties since they effectively inhibit the growth of numerous cancer cell lines tested with IC50 values even up to 90-fold lower than etoposide – clinically used chemotherapeutic agent, and selectivity index values reaching 125. Mechanistic studies showed that inhibition of human topoisomerase II by the investigated compounds is maintained through the contact of the protein with aromatic moieties located at limiting edges of ligand molecules and intensive interactions of the thiosemicarbazide core with the DNA fragments present in the catalytic site of the enzyme.
A set of approximately 2000 derivatives of thiosemicarbazide, precisely 1,4-disubstituted thiosemicarbazide derivatives, were docked into the DNA-binding sites of human DNA topoisomerase II. Designed ligands had different aryl, aroyl, alkyl (branched and unbranched), heteroaryl groups at distal positions of thiosemicarbazide scaffold. Docking results were ranked on the basis of binding free energies. During in-silico experiments the AutoDockVina software was applied for docking simulations. All docked molecules were allowed for altering their conformations by introducing rotatable bonds, automatically detected by the prepare_ligand4.py tool of the ATD Tools package. In addition, the flexibility of the following amino-acid residues was allowed: Asp479, Arg503, Gln778 and Met782. The selection of those residues relied on the criterion of contact with the ligand present in the PDB:3qx3 structure. The procedure of docking was carried out within the cuboid region of dimensions of 28 × 28 × 28 Å3 which the originally co-crystallised ligands present in the PDB structures as well as the closest amino-acid residues that exhibit contact with those ligands. All the defaults procedures and algorithms implemented in AutoDockVina were applied during docking procedure. The docking methodology was initially validated by docking simulations of the ligand molecule originally included in the protein structure PDB:3qx3. The validation procedure was analogous to that described above for the studied compounds. The graphical illustration of the validation results is given in Figure 4. The accepted methodology is accurate enough to recover the original position of the bound ligand with a small root-mean square deviation of non-hydrogen atoms, equal to 0.0974 Å.
The selected thiosemicarbazide derivatives, characterised by the most potent affinity towards human DNA topoisomerase II, were synthesised using previously described method. In brief, equimolar amounts of respective acid hydrazides and isothiocyanates were dissolved in anhydrous ethanol and heated under reflux for 10 min. After cooling, the precipitate formed were filtered, dried and crystallised from anhydrous ethanol. Melting temperatures of the synthesised compounds were measured using Fisher-Johns apparatus (Fisher Scientific, Schwerte, Germany). Elemental analyses were determined using AMZ-CHX elemental analyser (PG, Gdansk, Poland) and were within ±0.4% of the theoretical values. NMR spectra were recorded on a Bruker Avance spectrometer (Bruker BioSpin GmbH, Germany) using DMSO-d6 as solvent and TMS as an internal standard. FT-IR spectra were recorded using a Vertex 70 spectrometer (Bruker) equipped with an ATR Platinum Diamond A 225 device. All reagents and solvents were purchased from Sigma-Aldrich and were used without further purification.
Yield: 74%; m.p.: 140–142 °C; 1H-NMR (600 MHz): 1.12 (d, 6H, 2xCH 3, J = 6.9 Hz), 1.17 (d, 6H, 2xCH3, J = 6.9 Hz), 3.05–3.12 (m, 2H, 2xCH), 4.06 (s, 2H, CH 2), 7.12–8.16 (m, 10H, ArH), 9.19 (s, 1H, NH), 9.49 (s, 1H, NH), 10.30 (s, 1H, NH); 13C-NMR (150 MHz): 24.49, 28.11, 37.92, 123.36, 125.26, 125.91, 126.08, 126.31, 127.67, 128.07, 128.38, 128.72, 132.49, 132.80, 133.79, 134.84, 147.15, 170.39, 183.15. IR (ν, cm-1): 3371, 3272, 3016, 2969, 1738, 1518, 1364, 1229; Anal. Calc. for C25H29N3OS (%): C 71.56, H 6.97, N 10.01. Found: C 71.73, H 6.69, N 10.14.
Yield: 92%; m.p.: 178–180 °C; 1H-NMR (600 MHz): 4.07 (s, 2H, CH 2), 7.36–8.21 (m, 16H, ArH), 9.52 (s, 1H, NH), 9.92 (s, 1H, NH), 10.34 (s, 1H, NH)) ; 13C-NMR (150 MHz): 38.25, 122.91, 125.00, 125.96, 126.17, 126.52, 127.78, 128.50, 128.82, 129.94, 131.73, 132.33, 132.49, 133.80, 142.79, 152.52, 170.60, 183.21; IR (ν, cm-1): 3366, 3336, 3109, 2946, 1738, 1501, 1365, 1216; Anal. Calc. for C25H21N5OS (%): C 68.32, H 4.82, N 15.93. Found: C 68.40, H 4.62, N 15.85.
Yield: 87%; m.p.: 168–170 °C; 1H-NMR (600 MHz): 7.24–7.96 (m, 13H, ArH), 9.52 (s, 1H, NH), 9.92 (s, 1H, NH), 10.36 (s, 1H, NH); 13C-NMR (150 MHz): 122.90, 124.99, 125.95, 126.19, 126.56, 127.73, 128.47, 128.81, 129.95, 131.73, 132.41, 133.76, 142.78, 152.54, 170.82, 181.42; IR (ν, cm-1): 3366, 3016, 2969, 2946, 1738, 1501, 1365, 1216; Anal. Calc. for C20H16N6O3S (%): C 57.13, H 3.84, N 19.99. Found: C 57.20, H 3.72, N 19.81.
Yield: 70%; m.p.: 278–280 °C; 1H-NMR (600 MHz): 7.43–7.55 (m, 4H, ArH), 7.81–7.85 (m, 1H, ArH), 7.90–7.94 (m, 1H, ArH), 8.10–8.14 (m, 1H, ArH), 9.31 (s, 1H, NH), 10.28 (s, 1H, NH), 10.51 (s, 1H, NH); 13C-NMR (150 MHz): 124.17, 125.96, 126.13, 126.50, 127.68, 128.29, 128.84, 132.35, 132.50, 133.75, 169.74, 183.15; IR (ν, cm-1): 3348, 3047, 1738, 1536, 1326, 1216. Anal. Calc. for C14H10F2N4O3S (%): C 47.73, H 2.86, N 15.90. Found: C 47.90, H 2.62, N 15.89.
Yield: 80%; m.p.: 194–195 °C; 1H-NMR (600 MHz): 2.40 (s, 3H, CH3), 7.37–8.10 (m, 8H, ArH), 11.62 (s, 1H, NH), 11.82 (s, 1H, NH), 12.67 (s, 1H, NH); 13C-NMR (150 MHz): 22.06, 124.06, 125.32, 126.11, 126.48, 127.08, 127.86, 132.21, 132.48, 133.76, 134.65, 138.13, 146.73, 162.48, 168.91, 180.03; IR (ν, cm-1): 3345, 3046, 2969, 1738, 1516, 1324, 1216; Anal. Calc. for C16H14N4O4S (%): C 53.62, H 3.94, N 15.63. Found: C 53.40, H 3.90, N 15.81.
Yield: 78%; m.p.: 286–288 °C; 1H-NMR (600 MHz): 7.58–7.67 (m, 4H, ArH), 7.93–8.04 (m, 4H, ArH), 11.37 (s, 1H, NH), 11.91 (s, 1H, NH), 12.26 (s, 1H, NH); 13C-NMR (150 MHz): 118.55, 118.89, 125.81, 129.06, 130.76, 131.22, 137.93, 138.58, 139.66, 151.99, 163.68, 167.14, 181.21; IR (ν, cm-1): 3221, 3016, 1738, 1529, 1365, 1228. Anal. Calc. for C15H11N5O6S (%): C 46.27, H 2.85, N 17.99. Found: C 46.43, H 2.72, N 17.88.
Yield: 68%; m.p.: 182–184 °C; 1H-NMR (600 MHz): 7.75–7.93 (m, 3H, ArH), 8.12–8.20 (m, 2H, ArH), 8.32–8.38 (m, 2H, ArH), 11.68 (s, 1H, NH), 12.19 (s, 1H, NH), 12.50 (s, 1H, NH), 13.90 ((s, 1H, OH); 13C-NMR (150 MHz): 118.50, 118.95, 125.91, 127.96, 128.93, 130.87, 133.32, 133.50, 134.40, 138.16, 139.64, 151.90, 163.64, 166.78, 180.97; IR (ν, cm-1): 3750, 3015, 2970, 1738, 1521, 1348, 1216. Anal. Calc. for C15H11ClN4O5S (%): C 45.63, H 2.81, N 14.19. Found: C 45.82, H 2.95, N 14.08.
Yield: 83%; m.p.: 174–176 °C; 1H-NMR (600 MHz): 4.15 (s, 2H, CH2), 7.41–7.59 (m, 8H, ArH), 7.86 (d, 1H, ArH, J = 8.3 Hz), 7.94 (d, 1H, ArH, J = 8.3 Hz), 8.19 (d, 1H, ArH, J = 8.3 Hz), 11.34 (s, 1H, NH), 12.11 (s, 1H, NH), 12.44 (s, 1H, NH); 13C-NMR (150 MHz): 37.53, 124.87, 125.98, 126.21, 126.57, 127.58, 127.84, 128.52, 128.84, 129.70, 130.01, 130.38, 132.10, 132.40, 132.57, 133.81, 134.56, 167.76, 167.91, 176.73; IR (ν, cm-1): 3215, 3016, 2970, 1738, 1495, 1365, 1216. Anal. Calc. for C20H16ClN3O2S (%): C 60.37, H 4.05, N 10.56. Found: C 60.45, H 4.21, N 10.43.
Yield: 86%; m.p.: 164–166 °C; 1H-NMR (600 MHz): 4.14 (s, 2H, CH2), 7.46–7.58 (m, 6H, ArH), 7.62–7.66 (m, 1H, ArH), 7.85 (d, 1H, ArH, J = 8.3 Hz), 7.92–7.96 (m, 3H, ArH), 8.19 (d, 1H, ArH, J = 8.3 Hz), 11.31 (s, 1H, NH), 11.74 (s, 1H, NH), 12.71 (s, 1H, NH); 13C-NMR (150 MHz): 37.53, 124.86, 125.98, 126.21, 126.57, 127.85, 128.53, 128.83, 128.91, 129.16, 132.14, 132.30, 132.39, 133.60, 133.81, 167.87, 168.49, 177.38; IR (ν, cm-1): 3273, 3058, 3005, 2970, 1738, 1525, 1365, 1229. Anal. Calc. for C20H17N3O2S (%): C 66.10, H 4.71, N 11.56. Found: C 66.15, H 4.59, N 11.47.
Yield: 77%; m.p.: 186–188 °C; 1H-NMR (600 MHz): 4.02 (s, 2H, CH2), 7.11–7.56 (m, 7H, ArH), 7.85 (d, 1H, ArH, J = 8.2 Hz), 7.94 (d, 1H, ArH, J = 8.2 Hz), 8.11 (d, 1H, ArH, J = 8.2 Hz), 9.44 (s, 1H, NH), 10.02 (s, 1H, NH), 10.45 (s, 1H, NH); 13C-NMR (150 MHz): 38.18, 112.18, 125.05, 125.94, 126.15, 126.48, 127.76, 128.49, 128.80, 129.38, 132.29, 132.53, 133.80, 158.57, 160.25, 170.35, 183.11; IR (ν, cm-1): 3246, 3056, 2953, 1658, 1525, 1325, 1239. Anal. Calc. for C19H15F2N3OS (%): C 61.44, H 4.07, N 11.31. Found: C 61.33, H 4.21, N 11.30.
Yield: 73%; m.p.: 182–184 °C; 1H-NMR (600 MHz): 7.54–8.13 (m, 7H, ArH), 9.66 (s, 1H, NH), 10.23 (s, 1H, NH), 10.89 (s, 1H, NH); 13C-NMR (150 MHz): 122.92, 123.45, 124.79, 129.96, 130.40, 131.78, 132.19, 134.08, 142.67, 147.72, 152.49, 165.82, 180.11; IR (ν, cm-1): 3328, 3291, 2969, 1738, 1523, 1502, 1378, 1226. Anal. Calc. for C14H10F2N4O3S (%): C 47.73, H 2.86, N 15.90. Found: C 47.78, H 2.59, N 15.96.
Yield: 70%; m.p.: 190–192 °C; 1H-NMR (600 MHz): 2.41 (s, 3H, CH3), 7.22–7.50 (m, 5H, ArH), 8.07–8.10 (m, 1H, ArH), 8.50–8.52 (m, 1H, ArH), 11.26 (s, 1H, NH), 11.85 (s, 1H, NH), 12.27 (s, 1H, NH), 13.95 (s, 1H, OH); 13C-NMR (150 MHz): 20.01, 119.76, 123.22, 125.67, 126.04, 128.84, 131.44, 133.85, 134.25, 134.69, 136.42, 137.28, 155.67, 162.81, 171.25, 180.77; IR (ν, cm-1): 3726, 3245, 3045, 2920, 1738, 1518, 1255; Anal. Calc. for C16H14N4O5S (%): C 51.33, H 3.77, N 14.97. Found: C 51.17, H 3.67, N 15.06.
Cytotoxicity of the synthesised compounds were examined against a panel of the cancer cell lines, MCF-7 (estrogen-dependent human breast cancer cells), MDA-MB-231 (oestrogen-independent human breast cancer cells), SCC- 25 (squamous cell carcinoma of the tongue), FaDu (squamous cell carcinoma of the pharynx), A- 549 (human lung carcinoma), AGS (human gastric adenocarcinoma), LS-180, HT-29 (colon cancer cell lines), T98G (glioblastoma cells), A375 (melanoma cells). Human embryonic kidney (HEK-293) cells and human normal skin fibroblasts (CRL-2072) were used as reference cell lines. The investigated cell lines were obtained from American Type Culture Collection (ATCC; Manassas, VA, USA). A375, MCF-7, MDA-MB-231, T98G cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) (Sigma Aldrich, St. Louis, MO, USA) supplemented with 10% heat inactivated foetal bovine serum (FBS), penicillin (100 U/mL), and streptomycin (100 μg/mL). SCC-25 cells were cultured in 1:1 mixture of Dulbecco’s modified Eagle’s medium and Ham’s F12 Medium (Sigma Aldrich, St. Louis, MO, USA) supplemented with 400 ng/ml hydrocortisone, 10% FBS, penicillin (100 U/mL), and streptomycin (100 μg/mL). A549 and AGS cells were maintained in F-12K Medium (Sigma Aldrich, St. Louis, MO, USA). LS-180 and FaDu cells were cultured in Eagle’s Minimum Essential Medium. HT-29 cells were maintained in McCoy’s 5 A Medium (Sigma Aldrich, St. Louis, MO, USA) supplemented with 10%FBS, penicillin (100 U/mL), and streptomycin (100 μg/mL). All the cells were maintained at 37 °C in a humidified atmosphere of 5% CO2.
The synthesised compounds were dissolved in DMSO in order to obtain stock solutions. At the day of the experiment, the suspension of cells (1 × 105 cells/mL) in the respective medium was applied to a 96-well plate at 100 μL per well. After 24 h of incubation, the medium was removed from the wells and replaced by increasing concentrations of compounds (or etoposide) in medium containing 2% FBS. The control cells were cultured only with a medium containing 2% FBS. Cytotoxicity of DMSO was also checked at concentrations present in respective dilutions of stock solution. After 24 h (or 48 h) of incubation, 15 μL of MTT working solution (5 mg/mL in PBS) was added to each well. The plate was incubated for 3 h. Subsequently, 100 μL of 10% SDS solution was added to each well. Cells were incubated overnight at 37 °C to dissolve the precipitated formazan crystals. The concentration of the dissolved formazan was evaluated by measuring the absorbance at λ = 570 nm, using a microplate reader (Epoch, BioTek Instruments, Inc., Winooski, VT, USA). At least two independent experiments were performed in triplicates. The results of the MTT assay were expressed as mean ± SD. DMSO in the concentrations present in the dilutions of stock solutions did not influence the viability of the tested cells. IC50 values of the investigated derivatives and standards were calculated using the IC50 calculator.
The BrdU assay was performed to determine the effect of synthesised compounds on the proliferation of cancer cells. BrdU assay indicates the rate of incorporation of 5′-bromo-2′-deoxy-uridine (BrdU) during DNA biosynthesis. In brief, cells were plated in 96-well plates (NUNC, Roskilde, Denmark) at the density of 3 × 104 cells/mL. The next day, cells were treated with thiosemicarbazide derivatives or etoposide (in the increasing concentrations, up to 100 µg/ml) or a fresh culture medium. Cell proliferation was examined after 24 h and 48 h of incubation according to the manufacturer’s protocol (Cell Proliferation ELISA BrdU, Roche Diagnostics GmbH, Penzberg,)Germany).
Determination of the inhibitory effect of compounds 1-10 against human topoisomerase Iiα was performed using Human Topoisomerase II Relaxation Assay (Inspiralis Ltd, Norwich, UK) and Topoisomerase II Drug Screening Kit (TopoGEN Inc., Buena Vista, CO, USA) following the manufacturer’s instructions. Briefly, the increased concentrations of the compounds were mixed with the reaction mixture, incubated according to the protocol and, after reaction termination, the samples were analysed by electrophoresis using 1% agarose gel containing 0.5 mg/mL of ethidium bromide in Tris-Acetate-EDTA (TAE) buffer. Visualisation of the bands was performed using an enhanced chemiluminescence system (Syngene G:BOX Chemi XT4, Cambridge, UK). For the quantitative determination of topoisomerase II inhibition, images were densitometrically scanned. The inhibition of topoisomerase II was calculate from the equation:
Inhibitory effects of the investigated compounds were determined using Universal IDO1/IDO2/TDO Inhibitor Screening Assay Kit from BPS Bioscience, Inc. (San Diego, CA, USA). This colorimetric assay is based on the measurement of the ability of IDO1, IDO2 and TDO to convert L-tryptophan into N-formylkynurenine (NFK). The experiments were performed according to the manufacturer’s guidelines. The final concentration of the compounds in the reaction mixture was 50 µg/mL. The amount of NFK was measured spectrophotometrically at 320 nm using Epoch BioTek microplate reader (BioTek Instruments, Inc., Winooski, VT, USA). The samples were run in triplicate and the results were expressed as mean ± SD.
Click here for additional data file. |
PMC9648381 | 36342062 | Amira Awadalla,Eman T. Hamam,Fardous F. El-Senduny,Nisreen Mansour Omar,Mohamed R. Mahdi,Nashwa Barakat,Omar A. Ammar,Abdelaziz M. Hussein,Ahmed A. Shokeir,Salma M. Khirallah | Zinc oxide nanoparticles and spironolactone-enhanced Nrf2/HO-1 pathway and inhibited Wnt/β-catenin pathway in adenine-induced nephrotoxicity in rats A. AWADALLA ET AL. | 07-11-2022 | ZnO-NPs,spironolactone,CKD,Wnt,β-catenin,Nephrotoxicity,Nrf2,HO-1 | ABSTRACT Objective To investigate the renoprotective, the antioxidant, and the anti-inflammatory impact of a combination of SPL and ZnO-NPs to combat against chronic kidney disease (CKD). Methods In total, 50 males of rats were distributed into 5 groups (10 rats each); normal group, adenine sulfate (0.25% in diet for 10 days) (CKD) group. After the last dose of adenine sulfate, rats were divided into three groups: SPL + Adenine sulfate group; rats were treated orally by mixing SPL (20 mg/kg/day) into chow for 8 weeks, ZnO-NPs + Adenine sulfate group; rats were injected intraperitoneally with ZnO-NPs (5 mg/kg) three times weekly for 8 weeks, ZnO-NPs + SPL + Adenine sulfate group; rats were injected with the same previous doses for 8 weeks. Results Each of SPL and ZnO-NPs up-regulated antioxidant genes (Nrf2 and HO-1), down-regulated fibrotic and inflammatory genes (TGF-β1, Wnt7a, β-catenin, fibronectin, collagen IV, α-SMA, TNF-α, and IL-6) compared to CKD. Furthermore, a combination of SPL and ZnO-NPs resulted in a greater improvement in the measured parameters than a single treatment. Conclusion The therapeutic role of SPL was enhanced by the antioxidant and the anti-inflammatory role of ZnO-NPs, which presented a great renoprotective effect against CKD. | Zinc oxide nanoparticles and spironolactone-enhanced Nrf2/HO-1 pathway and inhibited Wnt/β-catenin pathway in adenine-induced nephrotoxicity in rats A. AWADALLA ET AL.
To investigate the renoprotective, the antioxidant, and the anti-inflammatory impact of a combination of SPL and ZnO-NPs to combat against chronic kidney disease (CKD).
In total, 50 males of rats were distributed into 5 groups (10 rats each); normal group, adenine sulfate (0.25% in diet for 10 days) (CKD) group. After the last dose of adenine sulfate, rats were divided into three groups: SPL + Adenine sulfate group; rats were treated orally by mixing SPL (20 mg/kg/day) into chow for 8 weeks, ZnO-NPs + Adenine sulfate group; rats were injected intraperitoneally with ZnO-NPs (5 mg/kg) three times weekly for 8 weeks, ZnO-NPs + SPL + Adenine sulfate group; rats were injected with the same previous doses for 8 weeks.
Each of SPL and ZnO-NPs up-regulated antioxidant genes (Nrf2 and HO-1), down-regulated fibrotic and inflammatory genes (TGF-β1, Wnt7a, β-catenin, fibronectin, collagen IV, α-SMA, TNF-α, and IL-6) compared to CKD. Furthermore, a combination of SPL and ZnO-NPs resulted in a greater improvement in the measured parameters than a single treatment.
The therapeutic role of SPL was enhanced by the antioxidant and the anti-inflammatory role of ZnO-NPs, which presented a great renoprotective effect against CKD.
Chronic kidney disease (CKD) is a severe public health disease that causes irreversible destruction of kidney and eventually leads to tubulointerstitial fibrosis and glomerular damage, which is the main pathway to end-stage renal disease (ESRD) [1]. CKD causes an increase in oxidative stress, which causes inflammation, renal damage, and death of cells through NF-κB activation and Nrf2/HO-1 inhibition in renal tissues [2]. Moreover, CKD activates Wnt/β-catenin pathway, resulting in the elevation of oxidative stress and renal fibrosis [3]. Currently, the use of zinc oxide nanoparticles (ZnO-NPs) has become essential due to their ability to pass through cell membranes. Zinc was reported as a pro-antioxidant agent due to its role in the protection of thiol-containing proteins (antioxidant enzymes) and zinc-finger transcription factors [4]. In rats, ZnO-NPs protect against acute kidney injury (AKI) caused by cisplatin [5]. Furthermore, ZnO-NPs alleviate diabetic nephropathy, the main cause of ESRD [6]. Moreover, ZnO-NPs are dietary supplements with anti-inflammatory properties revealed by down-regulating mRNA expressions of IL-1β, IL-6, and TNF-α. ZnO-NPs are approved as anticancer therapies due to the electrostatic attraction that occurs between negatively charged cancer cells and positively charged ions of ZnO-NPs, inducing apoptosis. Furthermore, ZnO-NPs were conjugated with anti-diabetic drugs to improve their effect due to their antioxidant activity [5]. In addition, Spironolactone (SPL) is a synthetic aldosterone antagonist and was approved as a diuretic medicine for treating hypertension, primary hyperaldosteronism, and edematous states. Therapeutic indications of SPL were subsequently expanded in response to mounting evidence of the systemic pro-inflammatory and pro-fibrotic effects, mainly in the heart, kidneys, and vessels. In CKD, aldosterone levels typically rise as the glomerular filtration rate (GFR) decreases, so both ESRD and CKD reflect relative hyperaldosteronism. Several clinical trials have shown that SPL has a good nephroprotective effect [7]. SPL inhibited profibrotic effects of TGF-β1, preventing fibrosis in kidney tissue, and promoting cellular recovery [8]. Previous research has shown that ZnO-NPs have renoprotective effects against AKI [5] and CKD [9] in rats. Human and rat studies have also demonstrated that SPL can combat against CKD [7,10]. Therefore, this study investigated the impact of this combination in comparison to the effect of each agent alone in the renoprotection, antioxidant, and anti-inflammation against CKD induced by adenine sulfate in rats. Moreover, we aimed to study the mechanisms by which this combination works.
Fifty mature males of Sprague–Dawley rats weighing 180 ± 200 g were divided at a rate of four rats per polycarbonate cage. They were kept at 24°C with 50–70% humidity and 12 h dark light cycle. All procedures used in this study were approved by Institutional Animal Ethics Committee of Faculty of Medicine, Mansoura University, Egypt [R.22.03.1638]. Rats were distributed into five equal groups (10 rats each); (a) normal healthy (control) group, (b) adenine sulfate (CKD positive control) group; for 10 days, the diet included 150 mg/kg/day (0.25% w/w) powdered adenine sulfate [11], (c) SPL + adenine sulfate group; adenine sulfate group was treated orally for eight weeks by mixing SPL (20 mg/kg/day) into chow [10], (d) ZnO-NPs + adenine sulfate group; adenine sulfate group injected intraperitoneal (i.p)with ZnO-NPs (5 mg/kg) three times weekly for eight weeks after the last dose of adenine sulfate [9], (e) ZnO-NPs + SPL + adenine sulfate group; adenine sulfate group treated with the same previous doses of ZnO-NPs and SPL for eight weeks after the last adenine sulfate dose.
Every rat was sited in a metabolic cage for 24 h to collect urine and blood samples from the heart before the sacrifice of the animal under inhalational general anesthesia by sodium thiopental. For biochemical measurements, samples of blood were centrifuged, and serum was collected and stored at −20°C. Then, cervical dislocation was used to sacrifice the animals, and a midline laparotomy and bilateral nephrectomy were performed. The right kidney was fixed in 10% buffered formalin for immunohistochemical and histopathological studies, whereas the left kidney was kept at −80°C till biochemical and molecular parameters were performed.
Serum and urine creatinine (creatinine clearance (CrCl)), serum blood urea nitrogen (BUN), and microalbuminuria were colorimetrically measured according to manufacturer guidelines [12,13]. Kits were provided by Diamond Diagnostics, Egypt.
A colorimetric method was used to measure catalase (CAT), superoxide dismutase (SOD), reduced glutathione (GSH), and malondialdehyde (MDA) in the supernatant of kidney homogenates [12]. Kits were provided by Biodiagnostic, Egypt.
Total renal RNAs were isolated using TRIzol (Invitrogen). mRNAs expression levels were measured using StepOne plus by QuantiFast SYBR Green PCR Kit (Qiagen, Germany). Fibrotic genes (TGF-β1, Wnt7a, β-catenin, fibronectin, collagen IV, and α-SMA), inflammatory genes (IL-6 and TNF-α), and antioxidants genes (Nrf2 andHO-1) were evaluated by real-time PCR. As shown in Table 1, primers were created online at NCBI and manufactured at Vivantis (Malaysia). Gene transcription was normalized to GAPDH. The mRNA expression level was calculated by the 2-ΔΔct Equation [14].
The nuclear and cytoplasmic fractions were prepared by using the sucrose gradient protocol [15]. The concentration of protein was measured by Pierce™ BCA Protein Assay Kit. A 30 µg/well was loaded to SDS-PAGE. Then, the protein was transferred to 0.45 µm nitrocellulose membrane for 90 min at 90 V. The protein transfer was confirmed using Ponceau S stain. The membrane was blocked for 2 h with 3% BSA at room temperature and then incubated with primary antibodies against Nrf2, β-Catenin, PCNA (Proliferating cell nuclear antigen) (Thermo Scientific, MS-106) or actin (Cell signaling technology) at 4°C overnight. After washing the membrane, it was incubated with anti-rabbit HRP-conjugated secondary antibody for one hour at room temperature. After washing the membrane, the signal was detected using WesternBright™ ECL HRP substrate (Advansta, K-12045) and visualized by The ChemiDoc MP Imaging System (Bio-Rad). The fold of change in protein level was calculated using GraphPad Prism Software after normalization to the level of housekeeping protein β-actin in cytoplasmic fraction or PCNA in the nuclear fraction.
The kidney was dehydrated by the serial ascending concentration of alcohol (BDH, UK), and xylene (BDH, UK), then the tissues were embedded in paraffin wax (Sherwood, USA). The block of paraffin embedding tissue was cut at 5 µm by microtome (West Germany). Haematoxylin, and eosin were used to stain the slides to investigate the tissue under the light microscope (400×) [16]. Tubulointerstitial damage, chronicity, and regeneration were evaluated using a semi-quantitative pathological score according to Shi et al. [17].
Immunohistochemical investigation for kidney tissue was done to determine fibrotic markers expression (β-catenin, TGF-β1, and α-SMA) using staining by their antibodies. Slides were examined under light microscope at 200× magnification in order to detect immune reactive cells [16]. The expression of β-catenin, TGF-β1, and α-SMA (the number of positive (brown) cells) were calculated by a semi-quantitative score [13].
The number of renal tubules stained with each marker was counted in each field (HPF) and the mean was calculated to indicate the tubular staining, and number of glomerular cells stained was counted in each glomeruli at HPF and the mean was calculated.
Immunofluorescence staining for kidney tissue was used to show collagen III and IV expression. The samples were treated with the primary antibody at 4°C overnight (1:200). After being rinsed in PBS, the samples were incubated for one hour at 25°C with a secondary antibody, anti-rat IgG (1:400). The expression of collagen III and IV among the study groups was assessed using digital images after tissue sections were examined using fluorescence microscopy [18].
SPSS V 22 was used to conduct the statistical analysis. The data were presented as a mean ± standard deviation. The differences between groups were investigated using a one-way ANOVA analysis of variance with a post-hoc comparison. The Tukey post-test was employed to evaluate whether there were any differences between the groups. P value of <0.05 is considered significant. The Mann–Whitney and Kruskal–Wallis tests were used for statistical analysis of pathological score of histopathological examination.
When compared to the normal group, the adenine (CKD) group revealed that levels of microalbuminuria, serum BUN, as well as creatinine were all significantly elevated, while creatinine clearance(CrCl) was significantly decreased (p < 0.05). Conversely, each of ZnO-NPs and SPL-treated groups indicated a significant reduction in microalbuminuria, serum BUN, and creatinine and a significant increase in CrCl when compared to adenine group (p < 0.05). However, when compared to SPL group, ZnO-NPs group showed a significant decrease in microalbuminuria, serum BUN, and creatinine levels. Furthermore, ZnO-NPs + SPL group exposed more significant attenuation in microalbuminuria, serum BUN, and creatinine levels and a significant rise in CrCl when compared to each of ZnO-NPs and SPL groups (Table 2). In comparison to the normal group, the adenine (CKD) group showed a significant rise in MDA and a decrease in SOD, GSH, and CAT levels (p < 0.05). Interestingly, MDA was significantly reduced while GSH, SOD, and CAT were considerably higher in each of ZnO-NPs and SPL-treated groups when compared to CKD group (p < 0.05). However, as compared to SPL group, there was no significant difference in oxidative stress or antioxidants in ZnO-NPs group. Furthermore, when compared to each of SPL and ZnO-NPs groups, the SPL + ZnO-NPs group had a greater reduction in MDA and a greater improvement in SOD, GSH, and CAT levels (Table 3).
When compared to the normal group, fibrotic genes (TGF-β1, Wnt7a, β-catenin, fibronectin, collagen IV, and α-SMA), and inflammatory genes (TNF-α and IL-6) were significantly up-regulated, while antioxidant genes (Nrf2 and HO-1) were significantly down-regulated in kidney tissues of adenine (CKD) group (p < 0.05). Conversely, when compared to CKD group, each of SPL and ZnO-NPs groups significantly down-regulated the fibrotic genes, and the inflammatory genes. Additionally, each of SPL and ZnO-NPs groups significantly up-regulated the antioxidant genes (p < 0.05). As compared to the SPL group, collagen IV and IL-6 expression were significantly lower in ZnO-NPs group. However, ZnO-NPs and SPL group demonstrated greater attenuation of fibrotic and inflammatory genes and a more significant rise in antioxidant genes when compared to each of ZnO-NPs and SPL groups (Figure 1).
When compared to adenine (CKD) group, western blot analysis after normalization to the cytoplasmic actin or nuclear PCNA protein revealed that in each of ZnO-NPs and SPL + ZnO-NPs groups, β-catenin was decreased in cytoplasm and significantly translocated to the nucleus (p < 0.001). Additionally, when compared to CKD group, Nrf2 nuclear translocation was significantly greater in each of ZnO-NPs and SPL + ZnO-NPs groups (p < 0.0001) (Figure 2).
When compared to the adenine group, the tubulointerstitial damage score of kidney tissues was substantially higher in the adenine (CKD) group (p < 0.05) and dramatically enhanced in all treatment groups, especially the SPL + ZnO-NPs group. (Figure 3(A)). Normal kidney structure was seen in the normal group (Figure 3(B)), while CKD group presented protein leakage in the tubular lumen (Figure 3(C)). On the other hand, SPL group indicated mild interstitial collagen proliferation and prominent nuclei (Figure 3(D)), and ZnO-NPs group revealed mild congestion, mitotic figure and prominent nuclei (Figure 3(E)). ZnO-NPs group represented more regeneration than SPL group. Furthermore, SPL + ZnO-NPs group exposed prominent nuclei and mitotic figure (Figure 3(F)). As shown in Figures 4–6(A), the immunohistochemical positive scoring manifested a significant increase in β-catenin, TGF-β1, and α-SMA expression in the tubular epithelial cells and glomerular cells in CKD group as compared to the normal group. However, their expressions were considerably inhibited in all treated groups, especially SPL + ZnO-NPs group, when compared to CKD group (p < 0.05). Figures 4–6(B–F) depicted the changes in β-catenin, TGF-β1, and α-SMA tubular and glomerular staining among the different treated groups. Marked expression of the studied proteins was observed in CKD, while moderate and mild expressions were detected in the SPL, ZnO-NPs, and combination group, respectively. Additionally, Immunofluorescence examination of renal extracellular matrix proteins collagen III and IV showed marked expression in CKD group, moderate expression in SPL group and mild expression in ZnO-NPs and SPL + ZnO-NP groups (Figures 7 and 8).
Kidney possesses an intrinsic regeneration capacity; this regeneration is limited under chronic condition of kidney disease and cannot prevent the fibrosis process [1]. Chronic inflammation is common in CKD, as are significantly compromised anti-oxidative mechanisms. Inflammation and oxidative stress are important mechanisms of defense, but if they are not adequately managed, they can cause a variety of negative effects [19]. The use of ZnO-NPs is widespread in a variety of applications due to their distinct physical and chemical properties that allow them to interact with cellular macromolecules, resulting in a variety of therapeutic effects. Furthermore, ZnO-NPs have a wide range of biomedical applications, including antioxidant, antibacterial, anticancer, and anti-inflammatory [6]. SPL and its pharmacological properties as a diuretic drug, as well as its antioxidant and renoprotective effects, have been the subject of several studies [7]. The objective of this research was to investigate if a combination of SPL and ZnO-NPs may protect against CKD in comparison to each of the agents alone. The adenine addition to diets of rats has become widely accepted as a model for studying kidney injury. Previous research has supported our findings on the significant declines in kidney functions and significant damage of kidney morphology [11]. Conversely, SPL and ZnO-NPs treatment improved kidney function compared to each of SPL and ZnO-NPs groups, indicating that SPL and ZnO-NPs are renoprotective against adenine-induced CKD. Furthermore, the development in histopathological results of SPL and ZnO-NPs group confirmed this result. Barakat et al. [5] showed that ZnO-NPs improved the decrease in kidney function caused by cisplatin in rats. Also, Elseweidy et al. [8] explained that SPL reversed the adenine effect through kidney function improvement, histopathological findings, and fibrosis recovery enhancement. This could be explained by the antagonistic action of aldosterone by SPL where the renin–angiotensin aldosterone system (RAAS) is considered as the major hormonal circuit that regulates blood pressure by maintaining the level of sodium and potassium [20]. The current study revealed that the oxidative stress increased and antioxidant decreased in adenine group. Previous research found that adenine feeding caused a significant increase in MDA while decreasing GSH, SOD, and CAT levels in renal tissues [21]. Furthermore, ZnO-NPs and SPL group indicated a great rise in GSH, SOD, and CAT, and a significant inhibition in MDA in renal tissues compared to each of SPL and ZnO-NPs groups. Pessôa et al. [22] confirmed that SPL increased antioxidants and reduced oxidative stress due to inflammation reduction. In addition, Abd El-Khalik et al. [6] revealed that ZnO-NPs cause Nrf2 up-regulation that leads to MDA suppression and antioxidant enzymes generation, such as superoxide dismutase and HO-1. Adenine damages the kidney through the elevation of ROS, while zinc stabilizes and protects the antioxidant enzymes to combat the imbalance and protect the kidney from oxidative damage [4]. This study showed the up-regulation of fibrotic genes in the renal tissue of adenine group. Also, western blot analysis proved the up-regulation of β-catenin proteins in adenine-fed rats. In contrast, ZnO-NPs and SPL group caused down-regulation of fibrotic genes, inflammatory genes, and up-regulation of antioxidant genes in renal tissue compared to each of SPL and ZnO-NPs groups. These results are in line with those of Diwan et al. [11], who established that adenine-fed rats have higher TGF-β, collagen, IL-6, and TNF-α expression while having lower Nrf2 and HO-1 expression. Evidence suggests that TGF-β promotes the formation of reactive oxygen species, including fibrogenesis. MMP-9 and TIMP-1 production and activity are increased in CKD by the infiltration of inflammatory cells, the release of TNF-α, TGF-β, and ROS [23]. On the other hand, TGF-β activates both Smad-dependent and independent pathways causing several biological responses. When TGF-β1 binds to its receptor, TGF-β receptor type II (TβRII), the TβRI is activated and forms a heterodimer leading to Smad2 and Smad3 phosphorylation and binding to Smad4. After that, Smad2 and Smad3 are translocated to the nucleus leading to the expression of fibrogenic genes such as fibronectin and collagen [24]. Immunofluorescence examination of collagen III, and collagen IV represented the same findings. Diwan et al. [11] concluded that chronic inflammation was induced by a 0.25% adenine diet, which increased myofibroblast infiltration and macrophage, TNF-α, collagen, TGF-β, and α-SMA expression. Also, Oh et al. [25] revealed that Wnt/β-catenin signaling was up-regulated by an adenine diet by attaching Wnts to receptors on cell membrane, then Wnts dephosphorylate β-catenin. The cytoplasmic β-catenin is translocated to the nucleus, where it regulates Wnt7a target gene transcription. The up-regulation of β-catenin, Wnt1, Wnt2, and Wnt6 mRNA expression activated Wnt/β-catenin signaling. Similar findings were clarified by immunohistochemical investigation of β-catenin, TGF-β1, and α-SMA expressions. Furthermore, western blot revealed a decrease or absence of β-catenin in the cytoplasm, but they were translocated into the nucleus, and Nrf2 nuclear translocation was elevated in ZnO-NPs and SPL group. A similar result was seen in an immunohistochemical examination of β-catenin, as β-catenin accumulates in the cytoplasm before translocation to the nucleus, where it regulates gene transcription [26]. According to Elseweidy et al. [8], SPL has anti-inflammatory properties that are demonstrated by inhibition of nuclear factor kappa B (NF-κB) and TNF-α down-regulation. In addition, the anti-fibrotic action of SPL is thought to be due to insulin growth factor 1 (IGF-1) down-regulation, which inhibits TGF-β1 expression and hence reduces kidney damage. Furthermore, Syngle and Verma [27] demonstrated that SPL decreased the pro-inflammatory cytokines (IL-1, TNF-α, and IL-6). SPL significantly attenuated Wnt/β-catenin signaling activation [28]. Furthermore, Rodri et al. [29] demonstrated a decrease in α-SMA-expressing cells, indicating fibrogenic inhibition in SPL-treated group. Immunohistochemical analysis of TGF-β1 and α-SMA expressions illustrated the same results. In addition, Rombouts et al. [30] showed that procollagen I and IV synthesis was inhibited by SPL, with a tendency to inhibit procollagen III. Similar findings were reported in the immunofluorescence investigation. Binding of the SPL to the mineralocorticoid and glucocorticoid receptors at high doses could be an explanation for decreased collagen production. The inhibition of collagen synthesis by dexamethasone or corticosterone activation of the glucocorticoid receptor in skin fibroblasts is well known. Also, Feria et al. [31] clarified that TGF-β1 deficiency inhibits the formation of extracellular matrix proteins like fibronectin, and collagen I, both of which are related to kidney damage. According to Li et al. [32], after one week of treatment with SPL, the levels of MMP-2, MMP-9, and TIMP-1 can be reduced. Meanwhile, Yuan et al. [33] explained that SPL up-regulated Nrf2 expression as the Nrf2 activation leads to transcriptional regulation of a variety of phase II detoxification and antioxidant enzymes, such as HO-1. These enzymes reduce oxidative stress in tissues and cells. The Nrf2 overexpression decreased renal TGF-β1, fibroblast cells, α-SMA, fibronectin and type 1 collagen. Furthermore, Yim et al. [34] stated that SPL up-regulated HO-1 in rat kidney. According to Gulbahce-Mutlu et al. [35], ZnO-NPs reduced IL-6 in breast cancer in rats, indicating that ZnO-NPs have anti-inflammatory properties. Also, Bashandy et al. [9] established that ZnO-NPs inhibited the IL-6 and TNF-α. Since IL-6 deregulates antioxidant defenses, ZnO-NPs may reduce kidney injury by lowering IL-6 and lipid peroxidation levels. Additionally, ZnO-NPs could inhibit collagen bundles and α-SMA-positive cells in rats. Moreover, Alomari et al. [36] clarified that ZnO-NPs treatment decreased IL-6, TGF-β1, TNF-α, fibronectin, and collagen IV expression as illustrated in immunofluorescence examination. According to Guo et al. [37], ZnO-NPs inhibited MMP-9 and TGF-β1-induced fibroblast activation and epithelial differentiation. Furthermore, Sehsah et al. [38] reported that ZnO-NPs exposure resulted in an increase in CAT, SOD, HO-1, and Nrf2 mRNA levels. In addition, Thomas et al. [39] explained that because ZnO-NPs reduced β-catenin expression, Wnt7a expression was reduced as well.
Rat CKD was reduced by SPL combined with ZnO-NPs. The anti-inflammatory and anti-fibrotic properties of SPL and ZnO-NPs, as well as their antioxidant activities, can be used to treat adenine-induced CKD in rats. The inhibition of oxidative stress, the up-regulation of antioxidant genes (Nrf2 and HO-1), the down-regulation of fibrotic genes (TGF-β1, Wnt7a, β-catenin, fibronectin, collagen IV, and α-SMA), and the down-regulation of inflammatory genes (TNF-α and IL-6) explain their effects. As a result, SPL combined with ZnO-NPs could be a potential CKD therapeutic approach. |
PMC9648387 | 36181398 | Miao Li,Xi Zhang,Mimi Wang,Yaohui Wang,Jiali Qian,Xiaoxia Xing,Zhiming Wang,Yang You,Kun Guo,Jie Chen,Dongmei Gao,Yan Zhao,Lan Zhang,Rongxin Chen,Jiefeng Cui,Zhenggang Ren | Activation of Piezo1 contributes to matrix stiffness‐induced angiogenesis in hepatocellular carcinoma | 01-10-2022 | hepatocellular carcinoma,matrix stiffness,angiogenesis,Piezo1,HIF‐1α ubiquitination | Abstract Background Despite integrin being highlighted as a stiffness‐sensor molecule in matrix stiffness‐driven angiogenesis, other stiffness‐sensor molecules and their mechanosensory pathways related to angiogenesis in hepatocellular carcinoma (HCC) remain obscure. Here, we explored the interplay between Piezo1 and integrin β1 in the mechanosensory pathway and their effects on HCC angiogenesis to better understand matrix stiffness‐induced angiogenesis. Methods The role of Piezo1 in matrix stiffness‐induced angiogenesis was investigated using orthotopic liver cancer SD rat models with high liver stiffness background, and its clinical significance was evaluated in human HCC tissues. Matrix stiffness‐mediated Piezo1 upregulation and activation were assayed using an in vitro fibronectin (FN)‐coated cell culture system with different stiffness, Western blotting and Ca2+ probe. The effects of shPiezo1‐conditioned medium (CM) on angiogenesis were examined by tube formation assay, wound healing assay and angiogenesis array. The underlying mechanism by which Piezo1 participated in matrix stiffness‐induced angiogenesis was analyzed by microRNA quantitative real‐time polymerase chain reaction (qRT‐PCR), matrix stiffness measurement, dual‐luciferase reporter assay, ubiquitination assay and co‐immunoprecipitation. Results Increased matrix stiffness significantly upregulated Piezo1 expression at both cellular and tissue levels, and high expression of Piezo1 indicated an unfavorable prognosis. High matrix stiffness also noticeably enhanced the activation level of Piezo1, similar to its expression level. Piezo1 knockdown significantly suppressed tumor growth, angiogenesis, and lung metastasis of HCC rat models with high liver stiffness background. shPiezo1‐CM from HCC cells attenuated tube formation and migration abilities of vascular endothelial cells remarkably, and analysis of differentially expressed pro‐angiogenic factors revealed that Piezo1 promoted the expression and secretion of vascular endothelial growth factor (VEGF), CXC chemokine ligand 16 (CXCL16) and insulin‐like growth factor binding protein 2 (IGFBP2). Matrix stiffness‐caused Piezo1 upregulation/activation restrained hypoxia inducible factor‐1α (HIF‐1α) ubiquitination, subsequently enhanced the expression of downstream pro‐angiogenic factors to accelerate HCC angiogenesis. Besides, collagen 1 (COL1)‐reinforced tissue stiffening resulted in more expression of Piezo1 via miR‐625‐5p. Conclusions This study unravels a new mechanism by which the integrin β1/Piezo1 activation/Ca2+ influx/HIF‐1α ubiquitination/VEGF, CXCL16 and IGFBP2 pathway participates in matrix stiffness‐driven HCC angiogenesis. Simultaneously, a positive feedback regulation loop as stiff matrix/integrin β1/miR‐625‐5p/Piezo1 and COL1/stiffer matrix mediates matrix stiffness‐caused Piezo1 upregulation. | Activation of Piezo1 contributes to matrix stiffness‐induced angiogenesis in hepatocellular carcinoma
Despite integrin being highlighted as a stiffness‐sensor molecule in matrix stiffness‐driven angiogenesis, other stiffness‐sensor molecules and their mechanosensory pathways related to angiogenesis in hepatocellular carcinoma (HCC) remain obscure. Here, we explored the interplay between Piezo1 and integrin β1 in the mechanosensory pathway and their effects on HCC angiogenesis to better understand matrix stiffness‐induced angiogenesis.
The role of Piezo1 in matrix stiffness‐induced angiogenesis was investigated using orthotopic liver cancer SD rat models with high liver stiffness background, and its clinical significance was evaluated in human HCC tissues. Matrix stiffness‐mediated Piezo1 upregulation and activation were assayed using an in vitro fibronectin (FN)‐coated cell culture system with different stiffness, Western blotting and Ca2+ probe. The effects of shPiezo1‐conditioned medium (CM) on angiogenesis were examined by tube formation assay, wound healing assay and angiogenesis array. The underlying mechanism by which Piezo1 participated in matrix stiffness‐induced angiogenesis was analyzed by microRNA quantitative real‐time polymerase chain reaction (qRT‐PCR), matrix stiffness measurement, dual‐luciferase reporter assay, ubiquitination assay and co‐immunoprecipitation.
Increased matrix stiffness significantly upregulated Piezo1 expression at both cellular and tissue levels, and high expression of Piezo1 indicated an unfavorable prognosis. High matrix stiffness also noticeably enhanced the activation level of Piezo1, similar to its expression level. Piezo1 knockdown significantly suppressed tumor growth, angiogenesis, and lung metastasis of HCC rat models with high liver stiffness background. shPiezo1‐CM from HCC cells attenuated tube formation and migration abilities of vascular endothelial cells remarkably, and analysis of differentially expressed pro‐angiogenic factors revealed that Piezo1 promoted the expression and secretion of vascular endothelial growth factor (VEGF), CXC chemokine ligand 16 (CXCL16) and insulin‐like growth factor binding protein 2 (IGFBP2). Matrix stiffness‐caused Piezo1 upregulation/activation restrained hypoxia inducible factor‐1α (HIF‐1α) ubiquitination, subsequently enhanced the expression of downstream pro‐angiogenic factors to accelerate HCC angiogenesis. Besides, collagen 1 (COL1)‐reinforced tissue stiffening resulted in more expression of Piezo1 via miR‐625‐5p.
This study unravels a new mechanism by which the integrin β1/Piezo1 activation/Ca2+ influx/HIF‐1α ubiquitination/VEGF, CXCL16 and IGFBP2 pathway participates in matrix stiffness‐driven HCC angiogenesis. Simultaneously, a positive feedback regulation loop as stiff matrix/integrin β1/miR‐625‐5p/Piezo1 and COL1/stiffer matrix mediates matrix stiffness‐caused Piezo1 upregulation.
Abbreviations AFP Alpha‐Fetoprotein Ang Angiopoietins CCl4 Carbon tetrachloride CM Conditioned Medium COL Collagen CPA collagen proportional area CXCL16 CXC Chemokine Ligand 16 Co‐IP Co‐Immunoprecipitation EMT Epithelial‐Mesenchymal Transition FGF Fibroblast Growth Factor FN Fibronectin HCC Hepatocellular Carcinoma H&E Hematoxylin‐Eosin HIF‐1α Hypoxia‐Inducible Factor 1α HUVEC Human Umbilical Vein Endothelial Cell IGFBP Insulin‐like Growth Factor Binding Protein IHC Immunohistochemistry LOX Lysyl Oxidase MVD Microvascular Density OS Overall Survival PDGF Platelet‐Derived Growth Factor qRT‐PCR quantitative Real‐Time Polymerase Chain Reaction SDS‐PAGE Sodium Dodecyl Sulfate Polyacrylamide Gel Electrophoresis SEM Scanning Electron Microscopy TB Total Bilirubin TCGA the Cancer Genome Atlas TNM Tumor Node Metastasis UB Ubiquitin UPA Urokinase‐type Plasminogen Activator VEGF Vascular Endothelial Growth Factor VEGFR2 Vascular Endothelial Growth Factor Receptor 2 VHL von Hippel‐Lindau
Metastasis is considered a major obstacle in clinical practice to improve the therapeutic effect and prognosis of tumors. Angiogenesis is the basis of tumor growth, invasion, and metastasis [1, 2, 3, 4]. Currently, anti‐angiogenic monotherapy or combination therapy with immune checkpoint inhibitors has become an effective anti‐tumor strategy [5, 6, 7, 8, 9]. However, only a part of tumor patients are sensitive to anti‐angiogenic drugs alone, such as sorafenib, regorafenib, lenvatinib, and bevacizumab [10, 11], and many initially responsive patients easily develop resistance to drugs after a period of treatment [11, 12], implying that the mechanism of tumor angiogenesis is far from fully understood. Solid tumors generally have a strong ability to induce neovascularization which offers more nutrients and oxygen to feed tumor tissue and helps it to metastasize. Angiogenesis in healthy tissue is tightly controlled by the balance between pro‐angiogenic and anti‐angiogenic factors, but in tumor tissues, this balance is usually destroyed, and the angiogenic switch is almost always activated, thus making tumor tissues possess a hypervascularity phenotype [5, 10]. Besides that, microvasculature within tumor also exhibits the morphological characteristics of hyperpermeable, tortuous, and deformed [13]. These abnormal vasculature structural features remarkably heighten the invasion and metastasis phenotype of solid tumors, also attenuate the ability of chemotherapeutic drugs being delivered into tumors [1, 11], which underlines the significance of angiogenesis in tumor progression and the value of its targeted intervention. The contribution of biochemical cues within the microenvironment to tumor angiogenesis has been well researched during the last several decades [1, 14]. Many pro‐angiogenic and anti‐angiogenic factors, including vascular endothelial growth factor (VEGF), fibroblast growth factor (FGF), angiopoietins (Ang), platelet‐derived growth factor (PDGF), angiostatin, endostatin, thrombospondin‐1, vascular endothelial growth factor receptor (VEGFR), neuropilin, and endoglin (CD105), are validated to participate in the modulation of tumor angiogenesis [2, 14, 15]. Hypoxia and inflammation, the two most common characteristics of the tumor microenvironment, are also found to augment the expression and release of pro‐angiogenic factors and trigger tumor angiogenesis [15, 16]. However, the effects of biomechanical cues within the microenvironment, particularly matrix stiffness, on tumor angiogenesis remain largely uncharacterized. Matrix stiffening and hypervascularity almost co‐occur in all patients with advanced HCC and are associated with poor prognoses [10, 17]. The existing in vivo and in vitro evidence from our and other studies all supports that increased matrix stiffness strengthens the malignant characteristics of HCC cells and promotes their invasion and metastasis in different ways, including triggering epithelial‐mesenchymal transition (EMT) occurrence [18, 19], facilitating pre‐metastatic niche formation [20, 21], enhancing stemness characteristics [22, 23], upregulating invasion/metastasis‐associated gene expression [24, 25], influencing glucose and lipid metabolic reprogramming [26, 27] and attenuating chemotherapeutic effect [18, 28‐30]. Nevertheless, there are few literature about the linkage between matrix stiffness and angiogenesis in HCC. We previously uncovered an important role of stiffness mechanical signaling in driving HCC angiogenesis by increasing both VEGFR2 expression in human umbilical vein endothelial cells (HUVECs) and VEGF expression in HCC cells [31, 32]. Simultaneously, integrin‐based mechanosensory pathways were confirmed to be responsible for HCC angiogenesis and other pathological changes [18, 31, 33]. Yet, little is known about whether there exist other stiffness‐sensor molecules or mechanosensory pathways in matrix stiffness‐induced angiogenesis. The discovery of the Piezo1‐dependent mechanosensitive pathway [34] makes the theory of integrin‐based mechanosensory pathway face new challenges. As a new mechanosensitive ion channel protein, Piezo1 can convert physical stimuli into electrical and chemical signals by accelerating Ca2+ influx and regulate various physiological and pathological processes [34, 35, 36, 37]. It was found that Piezo1 sensed mechanical signals and contributed to the proliferation and invasion of malignant glioma [38]. Shear force‐activated Piezo1 regulates the biological function of endothelial cells and mediates physiological and pathological processes in the vascular tissue [39]. Accordingly, we hypothesized that there might be a certain interplay between Piezo1 and integrin‐based mechanosensitive pathway, and their effects may participate in matrix stiffness‐induced angiogenesis. In this study, we developed a new orthotopic liver cancer SD rat model with high liver stiffness background and an in vitro fibronectin (FN)‐coated cell culture system with different stiffness to explore the interplay between Piezo1 and integrin β1 in the mechanosensory pathway and their effects on HCC angiogenesis for better understanding of matrix stiffness‐induced angiogenesis.
Clinical data of 372 HCC patients and their tumor gene expression profiles were downloaded from TCGA database. Since the expression levels of lysyl oxidase (LOX) and collagen 1 (COL1) can indicate the grade of matrix stiffness [40, 41], the median expression values of LOX and collagen 1A1 (COL1A1) were taken as the threshold to stratify the patients. Of the 372 patients, 119 with COL1A1High ‐ LOXHigh HCC were classified as the high stiffness group, 120 with COL1A1Low ‐ LOXLow HCC were classified as the low stiffness group, and the rest 133 were excluded due to an inconsistent expression trend between COL1A1 and LOX. The mRNA expression levels of target genes, Piezo1, CD31, CD34 and vascular endothelial growth factor A (VEGFA), were compared between the two groups.
Two human HCC cell lines, MHCC97H cells and Hep3B cells, were obtained from the Liver Cancer Institute of Fudan University (Shanghai, China) and Cell Bank of Shanghai Institute of Biochemistry and Cell Biology (Shanghai, China), respectively. MHCC97H cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM, Gibco, New York, NY, USA) supplemented with 10% fetal bovine serum (FBS, Biowest, Riverside, MO, USA) and 1% penicillin/streptomycin (Gibco), and Hep3B cells in Minimum Essential Medium (Gibco) with 12.5% FBS and 1% penicillin/streptomycin. HUVECs, purchased from ScienCell Research Laboratories, Inc, were cultured in endothelial cell medium (ScienCell, San Diego, CA, USA) with 5% FBS (ScienCell), 1% penicillin/streptomycin (ScienCell), and 1% endothelial cell growth supplement (ScienCell). Buffalo rat HCC cells McA‐RH7777, acquired from the American Type Culture Collection (Manassas, VA, USA), were grown in the same culture medium as MHCC97H cells.
For SD rat HCC models with high liver stiffness background, 5‐week‐old male SD rats (Shanghai JieSiJie Laboratory Animal Co., Ltd., Shanghai, China) were subcutaneously injected with 100% carbon tetrachloride (CCl4) (3 mL/kg) in the abdomen, and then injected with 50% CCl4 olive solution (2 mL/kg) twice a week. Twelve weeks later, SD rat models with high liver stiffness were formed, their liver stiffness was evaluated by liver elasticity ultrasound, and the expression of COL1 and LOX in HCC tissues was evaluated by immunohistochemical staining. From two days before transplantation to three days after transplantation, the rats were intramuscularly injected with 2.5 mg dexamethasone per day. Approximately 1.6 × 106 McA‐RH7777 cells mixed in Matrigel (BD Biosciences, Franklin, NJ, USA) were orthotopically injected under the capsule of rat liver. Penicillin (50,000 U/day) was used from the day of surgery to the second postoperative day to prevent infection. Twelve days later, orthotopic liver cancer SD rat models with high liver stiffness background were established, and their blood samples, tumors, and lung tissues were collected for further analysis. The experimental method for the establishment of SD rat HCC models with normal liver stiffness background was the same as the method for orthotopic liver cancer SD rat models with high liver stiffness background, except for CCl4‐free pretreatment before transplantation. All animal care and experiments used in the study followed the guideline for the Care and Use of Laboratory Animals published by the US National Academy of Science (Washington, WA, USA), and the experiment design of the animal model was approved by the Animal Care Ethical Committee of Zhongshan Hospital, Fudan University (Shanghai, China).
The target fragment of genes (Supplementary Table S1) and packaging recombinant plasmid of lentivirus were designed and constructed by GeneChem Co. Ltd. (Shanghai, China). The fragment targeting human gene integrin β1 was inserted into the plasmid pGCSIL, and the fragment targeting human gene Piezo1 was inserted into the GV112 vector. The target sequence to miR‐625‐5p was cloned into the GV280 vector, miR‐625‐5p overexpression sequence was inserted into GV369. The fragments targeting rat genes integrin β1 and Piezo1 were inserted into the GV112 vector, respectively. When HCC cells grew and reached 40% confluence, they were infected with lentivirus and selected by puromycin (2 μg/mL). The efficiency of inhibition or overexpression was evaluated by quantitative real‐time polymerase chain reaction (qRT‐PCR) and Western blotting.
H&E staining: Tissue sections were dewaxed with xylene and rehydrated with ethanol at different concentrations, and then they were stained with hematoxylin for 5 min and rinsed with running water. After that, they were stained with eosin for 2 min. Sirius red staining: Slides were stained with Sirius red reagent (Polysciences Inc, Warrington, PA, USA) for 1 h, and then washed in acetic acid, dehydrated in ethanol, and cleared in xylene. Quantifications of collagen proportional area (CPA) were measured by ImageJ software (National Institutes of Health, Bethesda, MD, USA).
Immunohistochemical staining was the same as the method described previously [31]. The primary antibodies were diluted as follows: LOX (1:100, Abcam, Cambs, Cambridge, UK), COL1 (1:100, Affinity, Cincinnati, OH, USA), Piezo1 (1:50, Abcam), CD31 (1:100, Abcam), hypoxia inducible factor‐1α (HIF‐1α) (1:100, Abcam), VEGFA (1:100, Proteintech, Wuhan, Hubei, China), insulin‐like growth factor binding protein 2 (IGFBP2) (1:200, BOSTER, Wuhan, Hubei, China), CXC chemokine ligand 16 (CXCL16) (1:200, Proteintech). Photos of representative sites were captured with an Olympus microscope (Tokyo, Japan). The density of positive staining was measured by ImageJ software. Microvascular density (MVD) was assessed based on IHC staining of CD31. The slides were examined under 100× magnification to identify the highest vascular density area (hot spot) within the tumor, and five areas of highest MVD under 200× magnification were selected to calculate the average MVD. The average MVD of the five areas was recorded as the MVD level.
FN‐coated substrate gels with the stiffness of 6, 10, and 16 kPa were prepared as described previously [30]. The suspended HCC cells (1 × 106) in 0.6 mL culture medium were spread onto an FN‐coated gel in a dish and cultured for 2‐3 h at 37°C with 5% CO2. Subsequently, 10 mL culture medium was added to the dish, and the attached cells were further cultured for 36‐48 h at 37°C with 5% CO2. Cells were collected from the gel surface with a cell scraper for the following analysis.
Western blot was performed as in our previous work [18] with a little modification. Briefly, 100 μg protein extracted from HCC cells was loaded and resolved on sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS‐PAGE) for Piezo1 detection, and 20 μg protein extract for other target protein detection. The conditions of membrane transfer were optimized as 350 mA, 210 min for Piezo1, 300 mA, 45 min for von Hippel‐Lindau (VHL), 300 mA, and 90 min for other target proteins. The diluted primary antibodies were as follows: integrin β1 (1:1000, Cell Signal Technology, Boston, MA, USA), Piezo1 (1:500, Abcam), COL1 (1:1000, Affinity), α‐tubulin (1:5000, Proteintech), HIF‐1α (1:1000, Abcam), VHL (1:1000, Abcam), IGFBP2 (1:1000, Abcam), CXCL16 (1:1000, Abcam), L‐vascular endothelial growth factor A (L‐VEGFA) (1:1000, Proteintech), ubiquitin (UB) (1:1000, Cell Signal Technology). The secondary horseradish peroxidase (HRP)‐conjugated antibody (Proteintech) was diluted at 1:5000.
An HCC tissue microarray was constructed previously [18] from buffalo rat HCC models with different liver stiffness backgrounds.
Clinical data of 88 HCC patients who underwent curative resection at the Department of Liver Surgery, Zhongshan Hospital of Fudan University between January 2008 and December 2008 were analyzed. Patients were followed up till December 2013. HCC patients were diagnosed according to the diagnostic criteria of the American Association for the Study of Liver Diseases (2018) [42], and their clinical stage was determined according to the Barcelona Clinic Liver Cancer staging system (2004) [43] and the 8th edition of Tumor Node Metastasis (TNM) staging system [44], respectively. Tumor differentiation grade was evaluated by the Edmondson grading system [45]. Vascular invasion, tumor number, tumor size, and other clinical parameters such as age, gender, hepatitis B surface antigen (HBsAg), alpha‐fetoprotein (AFP), and liver function indicators were collected for univariate and multivariate analyses.
Target microRNA (miRNA) to Piezo1 and COL1A1 genes were identified seperately from the miRTarBase database (https://miRTarBase.cuhk.edu.cn/). The common miRNAs targeting both Piezo1 and COL1A1 genes were identified.
Total RNA was extracted from HCC cells using TRIzol reagent (Invitrogen). The cDNA of miRNA was reversely transcribed using All‐in OneTM miRNA qRT‐PCR Detection Kit 2.0 (GeneCopoeia, Rockville, MD, USA) according to the manufacturer's protocol. Specific primers for U6 and miR‐625‐5p were all synthesized by GeneCopoeia. Reaction conditions were performed following the manufacturer's protocol of GeneCopoeia. The target gene was amplified by a QuantStudioⓇ 5 Real‐Time PCR instrument (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer's protocol. Heterogeneous nuclear RNA (hsnRNA) U6 was selected as an endogenous reference. Relative gene expression was normalized to U6 and reported by 2− ΔΔ Ct method.
miR‐625‐5p interference sequence (5'‐GGACUAUAGAACUUUCCCCCU‐3') and its scramble sequence (5'‐ CAGUACUUUUGUGUAGUACAA‐3') were constructed and synthesized by Sangon Biotech (Shanghai, China). When they grew and reached 80% confluence, the cells were transfected with a negative control siRNA or si‐miR‐625‐5p in Opti‐Medium at the concentration of 50 nmol/L via lipofectamine 2000 (Invitrogen).
The wild‐type Piezo1 3’‐untranslated region (UTR) or COL1A1 3’‐UTR containing the binding site of miR‐625‐5p and its mutant type were all designed and synthesized by OBiO Technology Co. Ltd. (Shanghai, China). 293T cells were seeded into a 24‐well culture plate on the day before plasmid transfection. The cells were transfected with the designed plasmids. After 48 h culture, the cells were harvested, and their luciferase activity was analyzed using the Dual‐luciferase Reporter Assay System (Promega, Madison, WI, USA).
Rat rail COL1 (3.58 mg/mL, Corning, New York, NY, USA) and serum‐free DMEM were mixed in volume ratios of 1:20 and 1:4 to prepare two types of substrate gels for SEM analysis. The mixtures containing different amounts of COL1 on a slide were solidified into the gels at 4°C overnight. Then the gels were sampled and fixed in 2% glutaraldehyde at 4°C. Samples were stained with 1% osmium tetroxide and observed under a scanning electron microscope (JEOL, Tokyo, Japan).
Wild‐type, shCtrl‐transfected, and shPiezo1‐transfected MHCC97H cells were cultured on the high‐stiffness substrate for 48 h, and their culture supernatants were collected and concentrated by a centrifugal filter (Millipore, Schwalbach, SL, Germany) at 4000 × g for 30 min at 4°C, and then the concentrated supernatant was filtered through 0.22 μm filters and stored at ‐80°C for subsequent use. The protein concentration of the concentrated supernatant was measured by bicinchoninic acid assay (Beyotime, Shanghai, China).
HUVECs (5 × 105) suspended in extracellular matrix (ECM) containing CM were seeded onto a Matrigel‐coated dish. After 8 h culture, tube‐like structures on Matrigel were photographed by an inverted microscope (Olympus). The state of tube forming in the picture was analyzed using ImageJ software.
Wound healing assay was performed to assess the effect of CM intervention on cell migration. After HUVECs grew to reach a tight cell monolayer in a 6‐well plate, the cell monolayer was scratched with a plastic pipette tip. The remaining cells were washed twice with PBS and then treated with CM for 48 h. The migrated cells at the wound front were photographed using an Olympus microscope and analyzed by ImageJ software.
Angiogenesis‐related factors in CM were analyzed by Human Angiogenesis Array Kit (Proteome ProfilerTM, R&D Systems, Minneapolis, MN, USA). Samples were mixed with a detection antibody cocktail for 1 h. The sample/antibody mixture was then incubated with an angiogenesis array overnight at 4°C. After unbound materials were removed from the array, streptavidin‐HRP and chemiluminescent detection reagents were sequentially added. Array signals were recorded by the Bio‐Rad Chemi Doc XRS imaging system (Hercules, CA, USA) and analyzed using ImageJ software.
HCC cells were cultured on different stiffness substrates for 24 h, and then their culture solution was replaced with a fresh culture medium containing Ca2+ probe (Ca2+‐GPCR analysis‐calcium ion indicator probe Fluo‐4, AM, KeyGEN BioTECH, Nanjing, China, 1 μmol/L). The cells were further cultured for 30 min, and the culture medium was discarded. The cells were washed once with PBS and cultured in a fresh culture medium for 1 h. The fluorescence intensity of intracellular Ca2+ in HCC cells was observed and recorded by a fluorescence microscope (Olympus) with cellSens software (Olympus).
Yoda1 (an agonist for Piezo1, MedChemExpress, Shanghai, China, 25 μmol/L for MHCC97H and 5 μmol/L for Hep3B) and GsMTx4 (an antagonist for Piezo1, Abcam, 2.5 μmol/L for both MHCC97H and Hep3B) were respectively applied to treat HCC cells grown on different stiffness substrates for exploring the effect of Piezo1 activation on pro‐angiogenic factor expression. MG132 (a proteasome inhibitor, Cell Signal Technology, 10 μmol/L) was applied to treat HCC cells grown on different stiffness substrates for exploring the effects of Piezo1 and matrix stiffness on HIF‐1α ubiquitination.
Total protein was extracted from the cells using immunoprecipitation lysis buffer (20 mmol/L Tris‐HCl, pH 7.6; 150 mmol/L NaCl; 1 mmol/L ethylene diamine tetraacetic acid (EDTA); 0.5% NP‐40; 10% glycerol; 1 mmol/L phenylmethanesulfonyl fluoride). The extracted protein was pretreated with protein A/G plus‐agarose beads (Millipore) overnight at 4°C. Subsequently, these mixtures were incubated with IgG or HIF‐1α antibody at 4°C for 24 h, respectively. Afterwards, the collected beads were washed three times using washing buffer (50 mmol/L Tris‐HCl, pH 7.6; 300 mmol/L NaCl; 1 mmol/L EDTA; 0.5% NP‐40; 10% glycerol), and the post‐elution beads were boiled in the loading buffer, resolved on SDS‐PAGE. Immunoblotting was finally performed to detect VHL and HIF‐1α.
GraphPad Prism 8.0 (San Diego, CA, USA) and SPSS 22.0 statistical software (SPSS, Chicago, IL, USA) were used for all statistical analyses. Data are presented as mean ± standard deviation (SD). Quantitative variables were analyzed by the analysis of variance (ANOVA) test among three groups and Student's t‐test between two groups. Qualitative variables were analyzed by the chi‐squared test and rank sum test. Kaplan‐Meier analysis and log‐rank test were performed to evaluate the association between Piezo1 expression and patients’ overall survival (OS). The interval time of OS was calculated from surgery to death of any reason or the most recent follow‐up. COX regression model was performed to clarify the prognostic value of Piezo1 expression. P < 0.05 was considered statistically significant.
On the basis of our previous findings that increased matrix stiffness potentiated HCC angiogenesis [31, 32], we further explored the contribution of Piezo1 to matrix stiffness‐induced angiogenesis in HCC. We first downloaded clinical data of 372 HCC patients and their tumor gene expression profiles from the TCGA database to clarify Piezo1 expression and its association with HCC angiogenesis. Considering that the expression levels of LOX and COL1 can indicate the grade of matrix stiffness [40, 41], we used the median expression values of LOX and COL1A1 as the threshold to stratify HCC patients into COL1A1High‐LOXHigh group (119 patients) and COL1A1Low‐LOXLow group (120 patients). Compared with that in the COL1A1Low‐LOXLow TCGA‐HCC tissues, Piezo1 presented an obvious upregulation in COL1A1High‐LOXHigh TCGA‐HCC tissues (Figure 1A), and its high expression was associated with an unfavorable prognosis (Figure 1B). Additionally, CD31, CD34, and VEGFA also exhibited higher expression in COL1A1High‐LOXHigh TCGA‐HCC tissues than in COL1A1Low‐LOXLow TCGA‐HCC tissues (Figure 1A), in agreement with our previous results in rat HCC tissues [31, 32]. Importantly, Piezo1 was positively associated with CD31 in expression level (Figure 1C). These results described above suggest a potential linkage between Piezo1 and matrix stiffness‐induced HCC angiogenesis. Subsequently, we developed new orthotopic liver cancer SD rat models with high and normal liver stiffness backgrounds (Figure 1D and Supplementary Figure S1A) to validate the role of Piezo1 in matrix stiffness‐induced angiogenesis and metastasis. We first compared liver stiffness levels of SD rat models with high or normal liver stiffness background (healthy liver) using LOX/COL1 expression (Supplementary Figure S1B) and sirius red staining (Supplementary Figure S1C). SD rats with high liver stiffness background had higher expression of LOX and COL1 and more collagen fibers in tumor tissues than those with normal liver stiffness background (Supplementary Figure S1B‐C). Then we detected the stiffness of SD rats with high liver stiffness background using liver elasticity ultrasound and found that the mean stiffness value of SD rats with high liver stiffness background (17.04 ± 1.73 kPa) could represent the tissue stiffness of advanced cirrhosis [46]. These results suggested that liver stiffness levels of the established SD rat models with high liver stiffness background were able to mirror the stiffness ranges of advanced cirrhosis, which were in accordance with reported data [40, 41, 46]. Afterwards, by hepatic subcapsular injection of HCC cells combined with short‐term usage of dexamethasone, we successfully obtained orthotopic liver cancer SD rat models with high liver stiffness background (Figure 1D, Supplementary Figure S1A) and ascertained that their serum phenotype (Supplementary Figure S1D) was almost consistent with that of HCC patients with advanced cirrhosis [47]. So, the established animal models can simulate the pathological properties of HCC with advanced cirrhosis. Using shPiezo1‐transfected McA‐RH7777 cells or shITGB1‐transfected McA‐RH7777 cells (Figure 1E), we also assessed the roles of Piezo1 or integrin β1 in matrix stiffness‐regulated HCC angiogenesis and discovered that tumor volume of the shPiezo1 #1 group was significantly smaller than those of the two control groups (Figure 1F). Simultaneously, the positive expression area of CD31 and the microvascular density (MVD) in HCC tissues of the shPiezo1 #1 group dropped considerably (Figure 1G‐H), and Piezo1 suppression evidently hindered the incidence of lung metastasis (Figure 1I), revealing that Piezo1 mediates matrix stiffness‐induced HCC angiogenesis and influences the incidence of lung metastasis. On the other hand, knockdown of integrin β1, a stiffness‐sensor molecule identified previously [18, 31], also resulted in an apparent decline in tumor growth, angiogenesis, and lung metastasis (Figure 1F‐I), which verified that increased matrix stiffness promoted angiogenesis and metastasis via integrin β1. In orthotopic liver cancer SD rat models with normal liver stiffness background (Supplementary Figure S1A), knockdown of Piezo1 or integrin β1 also significantly inhibited tumor growth and angiogenesis (Supplementary Figure S2A‐F), but they had little effect on the occurrence of lung metastasis (Supplementary Figure S2G), which were significantly different from the results of the SD rat model with high liver stiffness background. These results indirectly support that higher liver stiffness facilitated the occurrence of lung metastasis in HCC, which were consistent with our previous findings in buffalo rat HCC models [18]. Taken together, in addition to integrin β1, Piezo1 may also play a regulatory role in matrix stiffness‐driven angiogenesis and metastasis in HCC.
We applied 6, 10, and 16 kPa stiffness substrates, which represented the stiffness levels of normal, fibrotic, and cirrhotic liver tissues [46, 48, 49], to analyze matrix stiffness‐mediated effects on Piezo1 expression in HCC cells. The expression levels of Piezo1 and COL1 in HCC cells all showed an increasing trend with the increase of matrix stiffness (Figure 2A). Integrin β1 suppression remarkably reduced the expression of Piezo1 and COL1 in HCC cells grown on a high‐stiffness substrate (Figure 2B). Similarly, in a rat HCC tissue microarray analysis, Piezo1 also presented a significant upregulation in HCC tissues with high liver stiffness background compared with that in HCC tissues with normal and medium liver stiffness background (Figure 2C). Because high expression of COL1 and LOX in rat HCC tissues with high liver stiffness background has been validated in our previous study [18], we easily inferred that Piezo1 expression level was associated with matrix stiffness level or COL1/LOX expression level. Additionally, high Piezo1 expression in COL1High‐LOXHigh human HCC tissues (44 cases) (Figure 2D‐E) also supported an association between Piezo1 expression and COL1 expression. Thereby, at cellular and tissue levels, increased matrix stiffness indeed promoted Piezo1 and COL1 expression, and both of them showed the same trend in the expression levels. To evaluate the clinical significance of Piezo1 in HCC patients, we used the median expression value of Piezo1 in HCC tissues as the cut‐off value to divide HCC patients into the Piezo1High group (44 cases) and the Piezo1Low group (44 cases). The results demonstrated that high expression of Piezo1 was associated with high LOX/COL1 expression, cirrhosis‐associated indexes (total bilirubin, albumin), AFP, and tumor size (Table 1), in accordance with the above‐mentioned results in orthotopic liver cancer SD rat models. Additionally, Piezo1High HCC patients had an unfavorable prognosis (Figure 2F), which was identical to the result of the TCGA analysis. Multivariate analysis suggested that apart from TNM stage and vascular invasion, Piezo1 overexpression also served as an independent risk factor for OS of HCC patients (Supplementary Table S2).
Given that Piezo1 and COL1 were all highly expressed in HCC cells under high‐stiffness stimulation (Figure 2A), and the alteration of liver matrix stiffness was mainly attributed to extracellular matrix protein COL1 deposition and crosslinking [40, 41], we speculated that there might be a positive feedback regulation loop in matrix stiffness‐mediated effect on Piezo1 upregulation. Specifically, matrix stiffness promoted Piezo1 expression, simultaneously enhanced the production of extracellular COL1, and COL1‐reinforced tissue stiffening resulted in more expression of Piezo1. Considering that the same miRNA can concurrently regulate different target proteins, we identified 6 common miRNAs from the miRTarBase database that could target both Piezo1 and COL1A1 genes (Figure 3A). Among them, we determined miR‐625‐5p as the target miRNA by literature review and preliminary experiment analysis and then elucidated its role in matrix stiffness‐upregulated Piezo1. Increased matrix stiffness restrained miR‐625‐5p expression in HCC cells (Figure 3B and Supplementary Figure S3A), and suppression of integrin β1 significantly reversed high‐stiffness stimulation‐induced miR‐625‐5p downregulation (Figure 3C and Supplementary Figure S3B). The expression trend of miR‐625‐5p was opposite to that of Piezo1 and COL1 mentioned above (Figure 2A). These data indicate that there may exist a negative regulation of miR‐625‐5p on Piezo1 and COL1 expression under different stiffness stimulation. Based on the predicted binding sites of miR‐625‐5p to the 3’‐UTR of Piezo1 and COL1A1 (Figure 3D‐E), we respectively constructed the wild‐type recombinant plasmids (WT Piezo1 3’‐UTR and WT COL1A1 3’‐UTR) and the mutant‐type recombinant plasmids (MUT Piezo1 3’‐UTR and MUT COL1A1 3’‐UTR) to analyze whether there was a binding site between miR‐625‐5p and the 3’‐UTR regions of these two target genes. Dual‐luciferase reporter assay for Piezo1 and miR‐625‐5p showed that the luciferase fluorescence intensity of the miR‐625‐5p mimics group was decreased by 40.55% compared with that of the control group (P < 0.001), while the luciferase fluorescence intensity of the MUT Piezo1 3’‐UTR group was partially restored compared with that of the WT Piezo1 3’‐UTR group (P < 0.001) (Figure 3D). Similarly, for miR‐625‐5p and COL1A1 3’‐UTR, the luciferase fluorescence intensity of the miR‐625‐5p mimics group was decreased by 38.66% (P < 0.001), and the luciferase fluorescence intensity of the MUT COL1A1 3’‐UTR group was also partially restored (P < 0.001) (Figure 3E). The above results supported that miR‐625‐5p is specifically bound to the 3’‐UTR sites of Piezo1 or COL1A1. However, a partial reversion but not full reversion in the MUT Piezo1 3’‐UTR group or the MUT COL1A1 3’‐UTR group implied that other atypical binding sites might exist except the predicted binding sites. We further analyzed the regulatory role of miR‐625‐5p in matrix stiffness‐upregulated Piezo1 and found that miR‐625‐5p overexpression distinctly decreased the expression of Piezo1 and COL1 in HCC cells grown on high‐stiffness substrate (Figure 3F‐G and Supplementary Figure S3C‐F). Conversely, miR‐625‐5p inhibition improved Piezo1 and COL1 expression in HCC cells grown on low‐stiffness substrate (Figure 3H‐I and Supplementary Figure S3G‐J). All these results demonstrated that miR‐625‐5p participated in matrix stiffness‐upregulated Piezo1 and COL1. To assess the effect of COL1 deposition on matrix stiffness level, we prepared two types of substrate gel by mixing COL1 and serum‐free DMEM in a volume ratio of 1:20 and 1:4 to simulate different deposition amounts of COL1. The results showed that the stiffness level of the substrate gel (1:4, 1230.190 Pa) was significantly higher than that of the substrate gel (1:20, 341.522 Pa), and collagen I fiber bundles of the substrate gel (1:4) were thicker and more tightly packed than that of the substrate gel (1:20) (Figure 3J‐K). It confirmed our speculation that COL1 deposition reinforced tissue stiffening. Together, we proposed that a positive feedback regulation loop as stiff matrix/integrin β1/miR‐625‐5p/Piezo1 and COL1/stiffer matrix was involved in matrix stiffness‐upregulated Piezo1 in HCC (Figure 3L).
Paracrine is the most frequent regulatory way for tumor cells to induce angiogenesis [50, 51, 52]. The above results in animal models and HCC tissues suggested that Piezo1 might participate in matrix stiffness‐induced angiogenesis. By comparing the inhibitory effects of three knockdown clones for Piezo1 in HCC cells, the one with the best inhibitory effect was selected for subsequent function analysis (shPiezo1 #2 for MHCC97H cells and shPiezo1 #3 for Hep3B cells, Figure 4A). Subsequently, the CM from wild‐type, shCtrl‐transfected, and shPiezo1 #2‐transfected MHCC97H cells grown on high‐stiffness substrate (WT‐CM, Scramble‐CM, and shPiezo1‐CM) were collected to appraise their effects on angiogenesis. Compared with the cells treated with shPiezo1‐CM, HUVECs treated with WT‐CM and Scramble‐CM all presented stronger abilities in tube formation (Figure 4B) and migration (Figure 4C), illustrating that WT‐CM and Scramble‐CM contained more pro‐angiogenic factors than shPiezo1‐CM. Thereby, Piezo1 upregulation may influence the expression and secretion of pro‐angiogenic factors and promote HCC angiogenesis. Using a human angiogenesis array comprising 55 cytokines, we successfully found out 9 differentially expressed pro‐angiogenic and anti‐angiogenic factors (fold change >1.2 or <0.5) between shPiezo1‐CM and Scramble‐CM (Figure 4D and Supplementary Table S3), including 7 downregulated factors (Serpin F1, CXCL16, IGFBP‐2, VEGF, PDGF‐AA, IGFBP‐3, interleukin‐8) and 2 upregulated factors [Amphiregulin, urokinase‐type plasminogen activator (UPA)] in shPiezo1‐CM. Among these differentially expressed angiogenesis‐related cytokines, most pro‐angiogenic factors in shPiezo1‐CM exhibited a decreasing trend in content. Based on the above results, we concluded that the overall effect of shPiezo1‐CM was angiogenesis inhibition, and shPiezo1‐CM lacked the pro‐angiogenic factors required for tube formation and migration of HUVECs. Furthermore, we selected three differentially expressed pro‐angiogenic factors, VEGF, CXCL16, and IGFBP2, as the target factors for subsequent validation. The reasons for this selection were as follows: (1) the evidence presented above suggested that Piezo1 was a pro‐angiogenic regulator and mediated matrix stiffness‐induced angiogenesis; (2) VEGF, IGFBP2, and CXCL16 had a common upstream transcription factor hypoxia‐inducible factor (HIF‐1α) [15, 16, 53, 54] and affected tumor proliferation, angiogenesis, invasion and metastasis [55, 56, 57]; (3) HIF‐1α expression was not always dependent on hypoxia [58], and a variety of growth factors and cytokines were able to stabilize HIF‐1α under normoxic conditions [58]; (4) increased matrix stiffness also played a regulatory role in HIF‐1α expression in the polarized M2 macrophages [59]. In validation analysis, knockdown of Piezo1 remarkably suppressed the expression of VEGF, CXCL16, and IGFBP2 in HCC cells cultured on a high‐stiffness substrate (Figure 4E), which was consistent with the results of the human angiogenesis array. Besides that, integrin β1 knockdown also partially attenuated the expression of VEGF, CXCL16, and IGFBP2 in HCC cells grown on a high‐stiffness substrate (Figure 4F), meaning that matrix stiffness was also involved in the regulation of angiogenesis‐related molecules. Consequently, knockdown of Piezo1 reduced the expression of angiogenesis‐related cytokines and restrained matrix stiffness‐induced HCC angiogenesis.
Under mechanical stimulation, Piezo1 allows the influx of positively charged ions (Ca2+, Na+) into the cells, and the content of Ca2+ influx can reflect the activation level of Piezo1 [34]. As shown in Figure 2A, increased matrix stiffness significantly upregulated Piezo1 expression. Accordingly, the relationship between Piezo1 expression and Piezo1 activation becomes an unavoidable problem for understanding the contribution of Piezo1 to matrix stiffness‐induced HCC angiogenesis. We analyzed the levels of Ca2+ influx in HCC cells cultured on different stiffness substrates and discovered that the fluorescence intensity of Ca2+ in HCC cells was increased significantly with the increase of matrix stiffness (Figure 5A). Instead, suppressing Piezo1 significantly diminished the fluorescence intensity of Ca2+ in HCC cells cultured on the high‐stiffness substrate (Figure 5B). It suggests that Ca2+ influx positively correlate with Piezo1 expression and matrix stiffness. On the other hand, Yoda1 intervention noticeably enhanced the fluorescence intensity of Ca2+ in HCC cells grown on low‐stiffness substrate (Figure 5C), whereas GsMTx4 intervention clearly reduced it in HCC cells grown on high‐stiffness substrate (Figure 5D). Overall, high‐stiffness stimulation significantly enhanced Piezo1 expression and activation in HCC cells, and the expression level of Piezo1 was associated with its activation level under high‐stiffness stimulation.
HIF‐1α, an important common transcription factor, governs the expression of many pro‐angiogenic factors such as VEGF, CXCL16, and IGFBP2 [16, 55, 60]. We first elucidated whether matrix stiffness‐caused Piezo1 activation regulated the expression of HIF‐1α and its downstream pro‐angiogenic factors. As shown on Figure 6A and Supplementary Figure S4A‐B, high‐stiffness stimulation significantly upregulated the expression of HIF‐1α and its downstream target genes (VEGF, IGFBP2 and CXCL16) and significantly attenuated the expression of VHL (VH the E3 ligase of HIF‐1α) in HCC cells, suggesting a potential linkage between high matrix stiffness and HIF‐1α ubiquitination. Subsequently, we used proteasome inhibitor MG132 to treat HCC cells cultured on low‐stiffness substrate and found that MG132 intervention obviously increased the expression of HIF‐1α and its downstream target genes VEGF, IGFBP2, and CXCL16 (Figure 6B and Supplementary Figure S4C‐D), indirectly indicating that increased matrix stiffness repressed HIF‐1α ubiquitination and promoted the expression of its downstream target genes. Afterwards, we evaluated the effects of Piezo1 activation on HIF‐1α ubiquitination and the expression of its downstream target genes. The results demonstrated that Yoda1 intervention resulted in a significant upregulation in HIF‐1α, VEGF, CXCL16, and IGFBP2 expression, as well as an obvious downregulation in VHL expression in HCC cells cultured on low‐stiffness substrate (Figure 6C and Supplementary Figure S4E‐F). GsMTx4 intervention achieved an opposite result in HCC cells cultured on high‐stiffness substrate (Figure 6D, Supplementary Figure S4G‐H). These results together with the data in Figure 5 suggested that Piezo1 activation inhibited HIF‐1α ubiquitination and upregulated the expression of its downstream target genes via Ca2+ influx. Besides, Piezo1 or integrin β1 knockdown also led to a significant decline in HIF‐1α, VEGF, IGFBP2, and CXCL16 expression and an increase in VHL expression in HCC cells cultured on high‐stiffness substrate (Figure 6E‐F and Supplementary Figure S4I‐L), but Yoda1 intervention partially reversed the effects of Piezo1 or integrin β1 knockdown on the expression of HIF‐1α, VHL, VEGF, IGFBP2, and CXCL16. Similarly, MG132 intervention also reversed Piezo1 downregulation‐caused changes in the expression of HIF‐1 and its downstream target proteins (Figure 6E‐F and Supplementary Figure S4I‐L). Thereby, matrix stiffness‐caused Piezo1 activation participates in HIF‐1α ubiquitination and the expression of its downstream pro‐angiogenic factors. Taking HIF‐1α as a bait protein to capture VHL protein, we observed an endogenous interaction between VHL protein and HIF‐1α protein in HCC cells (Figure 6G, Supplementary Figure S5A). Additionally, knockdown of Piezo1 or integrin β1 elevated the HIF‐1α ubiquitination level (Figure 6H and Supplementary Figure S5B‐C), showing direct evidence that Piezo1 participates in high matrix stiffness‐downregulated HIF‐1α ubiquitination. In SD rat model with high‐stiffness background, the positive expression areas of HIF‐1α, VEGF, CXCL16 and IGFBP2 were significantly decreased in HCC tissues of the shPiezo1 group and the shITGB1 group (Figure 6I and Supplementary Figure S5D). Moreover, the positive expression areas of HIF‐1α and VEGF were associated with Piezo1‐positive expression areas in human HCC tissues (Supplementary Figure S5E), validating that Piezo1 contributes to matrix stiffness‐mediated effect on HIF‐1α expression and angiogenesis in HCC. Above all, we concluded that the matrix stiffness/integrin β1/Piezo1 activation/Ca2+ influx/HIF‐1α ubiquitination/VEGF, CXCL16 and IGFBP2 pathway participates in matrix stiffness‐regulated HCC angiogenesis (Figure 7).
Our previous works have validated a positive correlation between stiffness mechanical signal and HCC angiogenesis [31, 32]. In the present study, we found that increased matrix stiffness significantly upregulated Piezo1 expression at both cellular and tissue levels. High expression of Piezo1 was associated with HCC angiogenesis and indicated an unfavorable prognosis. Additionally, increased matrix stiffness also noticeably enhanced Piezo1 activation and promoted Ca2+ influx. Under the same stiffness stimulation, the expression level of Piezo1 was consistent with its activation level. Based on the above evidence, we speculated that Piezo1 and Piezo1 activation might be required for matrix stiffness‐induced angiogenesis and metastasis in HCC. We first developed two types of animal models to validate the role of Piezo1 in matrix stiffness‐induced angiogenesis and metastasis in vivo. In orthotopic liver cancer SD rat models with high liver stiffness background, knockdown of Piezo1 or integrin β1 significantly suppressed tumor growth, angiogenesis, and lung metastasis. However, in orthotopic liver cancer SD rat models with normal liver stiffness backgrounds, suppression of Piezo1 or integrin β1 only inhibited tumor growth significantly. By comparing the experimental results of two animal models, it is not difficult to conclude that high matrix stiffness promotes angiogenesis and metastasis of HCC and that Piezo1 is dispensable in matrix stiffness‐mediated effects on angiogenesis and metastasis. On the other hand, shPiezo1‐CM collected from HCC cells under high‐stiffness stimulation also remarkably attenuated tube formation and migration abilities of HUVECs, indicating that Piezo1 upregulation under high‐stiffness stimulation increases the expression and secretion of pro‐angiogenic factors. Similarly, analysis of differentially expressed pro‐angiogenic factors between shPiezo1‐CM and Scramble‐CM also revealed that Piezo1 promoted the expression and secretion of pro‐angiogenic factors. All evidence supports that Piezo1 upregulation participates in matrix stiffness‐induced angiogenesis by regulating the expression of angiogenesis‐related cytokines. Considering that the identified differentially expressed pro‐angiogenic factors VEGF, IGFBP2 and CXCL16 have a common transcription factor, HIF‐1α [16, 55, 60, 61], we determined them as the target proteins for subsequent mechanism analysis. Because Piezo1 expression and its activation exhibited the same increasing trend in HCC cells with an increase in matrix stiffness, we continued to investigate whether Piezo1 activation influenced the expression of HIF‐1α and its downstream angiogenesis‐related factors. Yoda1 intervention significantly reversed the downregulation of HIF‐1α and its downstream angiogenesis‐related factors caused by Piezo1 or integrin β1 knockdown, confirming the regulatory roles of Piezo1 activation and Ca2+ influx in the expression of HIF‐1α and its downstream angiogenesis‐related factors. Changes in intracellular Ca2+ concentration can drive many pathological molecule events such as migration, invasion, proliferation and apoptosis [62, 63, 64]. Calcium channels and pumps were demonstrated to influence HIF‐1 transcription, translation, stabilization and nuclear translocation in diverse types of cancer [65]. In prostate cancer cells, transient receptor potential channel M8 (TRPM8) overexpression reduced the phosphorylation level of the receptor of activated C kinase 1 (RACK1) and hindered its dimerization, and then promoted RACK1 binding to HIF‐1α and calcineurin, ultimately resulted in a decrease in HIF‐1α ubiquitination [66]. In neuroblastoma cells, full‐length transient receptor potential channel M2 decreased the expression of VHL to impede HIF‐1α degradation [67]. Based on an obvious downregulation in VHL in HCC cells under high‐stiffness stimulation and an increase in the expression of HIF‐1α and its downstream target genes under MG132 intervention, we further conjectured that increased matrix stiffness might repress HIF‐1α ubiquitination and promote the expression of its downstream target genes through Piezo1 activation. Yoda1 or GsMTx4 intervention assay under low‐ or high‐stiffness stimulation all supported that Piezo1 activation inhibited HIF‐1α ubiquitination and upregulated the expression of its downstream target genes. Co‐IP assay also revealed an endogenous interaction between VHL protein and HIF‐1α protein and that shPiezo1‐ or shITGB1‐transfected HCC cells had an obvious increase in HIF‐1α ubiquitination level, giving direct evidence that Piezo1 participates in high matrix stiffness‐downregulated HIF‐1α ubiquitination. Taken together, matrix stiffness as an initiator enhanced Piezo1 activation and Ca2+ influx, and then suppressed HIF‐1α ubiquitination, ultimately promoting pro‐angiogenic factor expression to accelerate HCC angiogenesis. In addition to upregulating Piezo1 expression, increased matrix stiffness also significantly improved COL1 expression in HCC cells. In view of the dominant role of matrix protein deposition and crosslinking in matrix stiffening [41], we proposed a hypothesis that COL1‐reinforced tissue stiffening may result in upregulated expression of Piezo1. Our results showed that the increase of COL1 deposition evidently raised the stiffness of the surrounding matrix and caused the appearance of thick bundle collagen, confirming the hypothesis that COL1 deposition reinforced tissue stiffening. Given that a miRNA can simultaneously regulate the expression of multiple target proteins [68], we identified miR‐625‐5p as a common miRNA to analyze its effects on Piezo1 and COL1A1. Under the same stiffness stimulation, the expression of miR‐625‐5p in HCC cells was associated with the expression of Piezo1 and COL1A1. Additionally, increased matrix stiffness significantly downregulated miR‐625‐5p expression, while the suppression of integrin β1 reversed high stiffness‐induced miR‐625‐5p downregulation. Dual‐luciferase reporter assay revealed a specific binding between miR‐625‐5p and the 3’‐UTR sites of Piezo1 or COL1A1. Analysis of miR‐625‐5p overexpression or downregulation also supported that miR‐625‐5p mediated matrix stiffness‐caused Piezo1 and COL1A1 upregulation. In summary, a positive feedback regulation loop as stiff matrix/integrin β1/miR‐625‐5p/Piezo1 and COL1/stiffer matrix was involved in matrix stiffness‐upregulated Piezo1 expression. Because the obtained evidence supports that increased matrix stiffness can enhance the release and expression of multiple pro‐angiogenic factors to accelerate HCC angiogenesis, we think that targeting the anti‐angiogenic effect of initiating factor matrix stiffness may be superior to that of specific angiogenic factors such as VEGF. Additionally, Piezo1 serves as an upstream molecule in mechanosensory pathway to mediate matrix stiffness‐induced angiogenesis and indicates a worse prognosis in HCC, which means that Piezo1 may become a promising novel target in anti‐angiogenic therapy. Nonetheless, it should be mentioned that there were several limitations in the present study. Although the stiffness‐sensor molecule integrin β1 was validated to regulate the expression and activity of Piezo1, the direct effect of mechanical signal on Piezo1 activation could not be excluded. Whether there exists a Piezo1‐based mechanosensory pathway independent of integrin β1 merits to be further investigated in further research. Besides, the detailed mechanism about how elevated Ca2+ influx influences HIF‐1α ubiquitination could be addressed in the follow‐up study.
Our study unravels a new mechanism by which integrin β1/Piezo1 activation/Ca2+ influx/HIF‐1α ubiquitination/VEGF, CXCL16 and IGFBP2 pathway participates in matrix stiffness‐driven HCC angiogenesis. Simultaneously, a positive feedback regulation loop as stiff matrix/integrin β1/miR‐625‐5p /Piezo1 and COL1/stiffer matrix mediates matrix stiffness‐caused Piezo1 upregulation.
All authors contributed to the study conception and design. Jiefeng Cui and Zhenggang Ren proposed conceptualizations and designed the study. Miao Li, Xi Zhang, Mimi Wang, Jiali Qian, Xiaoxia Xing, Yaohui Wang and Yang You developed the experiment methods and completed all the experiments. Jie Chen, Dongmei Gao and Yan Zhao helped finish the animal experiment. Zhiming Wang, Kun Guo, Lan Zhang and Rongxin Chen provided guidance and suggestion for the experiments. Miao Li and Xi Zhang wrote the original draft, and Jiefeng Cui and Zhenggang Ren revised it. All the authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Human HCC tissues and clinical data used in the study were approved by the Zhongshan Hospital Research Ethics Committee (Approval No. B2019‐332R). All animal experiments were approved by the Animal Care Ethical Committee of Zhongshan Hospital (Approval No. 2019‐136).
Not applicable.
Click here for additional data file. |
PMC9648389 | Peipei Shan,Feifei Yang,Jie Yu,Lirong Wang,Yuhua Qu,Huiran Qiu,Hua Zhang,Sujie Zhu | A novel histone deacetylase inhibitor exerts promising anti‐breast cancer activity via triggering AIFM1‐dependent programmed necrosis | 25-09-2022 | A novel histone deacetylase inhibitor exerts promising anti‐breast cancer activity via triggering AIFM1‐dependent programmed necrosis
Abbreviations Ac‐H3 acetylation‐histone 3 Ac‐H4 acetylation‐histone 4 AIFM1 apoptosis inducing factor, mitochondrion‐associated, 1 ALT alanine aminotransferase AST aspartate aminotransferase BUN blood urea nitrogen CypA Cyclophilin A DAPI 4′, 6‐diamidino‐2‐phenylindole DMSO dimethyl sulfoxide EMT epithelial–mesenchymal transition FDA Food and Drug Administration HDAC histone deacetylase HDACI histone deacetylase inhibitor HMGB1 high mobility group protein B1 LDH lactate dehydrogenase PCNA proliferation cell nuclear antigen PI propidine iodide RT‐PCR reverse transcription‐polymerase chain reaction SAHA suberoylanilide hydroxamic acid siRNA short interfering RNA TEM transmission electron microscope ZBG Zn binding group Dear Editor, Breast cancer is one of the most lethal cancers in women, with many patients still succumbing to this disease [1]. Accumulating evidence demonstrates that histone deacetylase inhibitors (HDACIs) are a promising therapeutic intervention for breast cancer [2], and many of them have shown favorable anti‐cancer activities in both preclinical and clinical settings [3]. However, most current HDACIs only exhibit limited efficacy against solid tumors with toxic side effects and readily produce drug resistance [4]. Therefore, it is necessary to develop new HDACIs with improved anti‐tumor activities and decreased toxicities for breast cancer therapeutics and investigate their mechanism of action. To improve the physicochemical properties of new generation HDACIs, the coumarin unit, as a promising pharmacophore in anti‐cancer drug discovery, was incorporated into hydroxamate HDACIs, and a series of new coumarin‐based derivatives were synthesized. After our initial screening, a compound coded YF349 with excellent HDAC inhibitory activity was identified (Supplementary Figure S1A, B). A hallmark index of HDAC inhibition is the increased acetylation of histones H3 and H4 [5]. As shown in Supplementary Figure S1C, D, YF349 significantly increased the acetylation of hH3 and Ac‐H4 compared with Suberoylanilide Hydroxamic Acid (SAHA). SAHA was used as a positive control in our experiments, as it is the first HDACI approved by the US Food and Drug Administration (FDA) for the clinical treatment of breast cancer [6]. Collectively, these results suggested that YF349 was a bona fide HDACI and had potential anti‐breast cancer activity. The chemical structure of YF349 is shown in Figure 1A. We next examined the anti‐breast cancer activity of YF349 in vitro. As shown in Figure 1B, YF349 significantly inhibited the growth of breast cancer cells compared with SAHA. In addition, YF349 significantly inhibited colony formation and the invasion ability of breast cancer cells (Figure 1C, D and Supplementary Figure S2A). Additionally, YF349 induced obvious cell death in breast cancer cells compared with SAHA treatment at the same dosage (Figure 1E). Finally, YF349 significantly altered the expression of proliferation cell nuclear antigen (PCNA), apoptosis‐related proteins (cleaved‐poly ADP‐ribose polymerase [PARP]), and epithelial–mesenchymal transition (EMT)‐related proteins (Supplementary Figure S2B). We then investigated the anti‐breast cancer effect of YF349 in vivo. As shown in Figure 1F, G and Supplementary Figure S2C‐E, YF349 significantly inhibited the tumor growth and metastasis compared with the control group and SAHA treatment at the same concentration. Meanwhile, YF349 significantly increased the levels of Ac‐H3 and Ac‐H3, which confirmed the HDAC inhibitory effect of YF349 on breast cancer in vivo (Supplementary Figure S2F). We then sought to investigate the potential toxicity of YF349. As shown in Supplementary Figure S3A, B, no significant changes in body weight and the major organs of mice treated with YF349 were observed. Histological analyses revealed no obvious damage to major organs (Supplementary Figure S3C). In addition, levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) are representative indicators of liver function, and blood urea nitrogen (BUN) is an indicator of kidney and liver conditions. As shown in Supplementary Figure S3D, YF349 treatment did not significantly affect the ALT, AST and BUN levels. Taken together, these results indicate that YF349 significantly inhibited breast cancer cell growth and metastasis both in vitro and in vivo and showed few adverse effects on the experimental mice at a therapeutic concentration. We then identified the modality of breast cancer cell death caused by YF349. As shown in Supplementary Figure S4A, a significant portion of cells appeared necrotic, and the cell death modality induced by YF349 was similar to that caused by the necrosis‐inducer shikonin. Moreover, pretreatment with the pan‐caspase inhibitor z‐VAD‐fmk did not prevent YF349‐induced cell death, while this treatment inhibited the proteasome inhibitor MG132‐induced cell death (Supplementary Figure S4B). In addition, cells treated with YF349 presented smeared DNA bands on an agarose gel in the DNA large fragment assay (Supplementary Figure S4C). The morphological characteristics of necrosis were also confirmed by transmission electron microscope (TEM) under YF349 treatment (Supplementary Figure S4D). In addition, YF349 promoted the cellular release of high mobility group protein B1 (HMGB1) and lactate dehydrogenase (LDH), which were shown to be necrosis markers [7] (Supplementary Figure S4E, F). Moreover, increased concentration of YF349 significantly upregulated the proportion of necrotic cells (Supplementary Figure S4G). Overall, these results indicated that YF349 induced necrosis in breast cancer cells. To further explore the detailed mechanisms of the anti‐breast cancer effect of YF349. RNA‐sequencing analysis was performed to identify differentially expressed genes between MDA‐MB231 cells treated with or without YF349. Supplementary Table S1 shows the differentially expressed genes after YF349 treatment. We then analyzed the top 10 upregulated genes (cut‐off, fold change > 4.8 and P < 0.05) listed in Supplementary Table S1. As shown in Supplementary Figure S5A, among the top 10 upregulated genes, the mRNA level of AIFM1 (apoptosis‐inducing factor, mitochondrion‐associated, 1) was significantly increased upon YF349 treatment compared with other genes. Supplementary Figure S4 showed that YF349 induced obvious necrosis of breast cancer cells, and AIFM1 has been known to be a key regulator of necrosis [8]. Therefore, we speculated that AIFM1 may play an essential role in the anti‐tumor effect of YF349 on breast cancer. A previous study reported that the increased total AIFM1 expression in cells led to increased sensitivity to cell death [9]. As shown in Figure 1H, YF349 significantly increased the mRNA level of AIFM1. An earlier study has demonstrated that HDAC1 could bind to the promoter region of AIFM1 and thus repress AIFM1 expression [10]. The molecular docking model showed that YF349 could interact with the active site of HDAC1 (Supplementary Figure S5B). These findings provide evidence that the increased mRNA level of AIFM1 induced by YF349 may be through the competitive interference of HDAC1 binding to the promoter of AIFM1. As shown in Figure 1I, J and Supplementary Figure S5C‐E, the promoter activity of AIFM1 was significantly upregulated by YF349, and YF349 significantly disrupted the binding of HDAC1 to the AIFM1 promoter. Taken together, these results indicated that YF349 upregulated the AIFM1 expression by disrupting HDAC1 binding to the AIFM1 promoter. We then clarified the role of AIFM1 in the YF349‐induced necrosis in breast cancer cells. As shown in Figure 1K‐M and Supplementary Figure S6A, B, AIFM1 knockdown cells were less sensitive to YF349‐induced necrosis. These results indicated that YF349 induced‐necrosis of breast cancer cells is in an AIFM1‐dependent manner. Furthermore, we found that YF349 promoted the formation of the AIFM1‐Cyclophilin A (CypA)‐γH2Ax complex in breast cancer cells (Figure 1N‐P and Supplementary Figure S6C, D). Collectively, these results demonstrated that YF349 remarkably induced the nuclear translocation of AIFM1 and significantly promoted the formation of the AIFM1‐CypA‐γH2Ax complex. In conclusion, we identified a novel HDACI, YF349, which displayed promising anti‐breast cancer activity both in vitro and in vivo. Further mechanistic studies revealed that YF349 increased the AIFM1 expression via inducing the disassociation of HDAC1 from the AIFM1 promoter, subsequently accelerating the nuclear translocation of AIFM1, promoting the formation of the AIFM1‐CypA‐γH2Ax complex, and finally inducing AIFM1‐mediated necrosis of breast cancer cells. Collectively, our work highlighted the anti‐breast cancer therapeutic potential of a new HDACI, YF349, via triggering AIFM1‐dependent necrosis. Our results suggest that the HDAC1‐AIFM1‐CypA‐γH2Ax signal axis can be a novel therapeutic target of breast cancer, and YF349 could be a promising preclinical drug candidate for breast cancer treatment.
Author contributions: PPS contributed to experimental design, data acquisition, analysis and interpretation and drafted the manuscript. FFY contributed to data acquisition, analysis and interpretation, and drafted the manuscript. PPS made sequencing and statistical analysis. JY and LRW collected clinical data and collected samples. YHQ and HRQ participated in the synthesis and purification of target compounds. PPS and LRW performed the animal experiments, model improvement and data analysis. HZ and SJZ contributed to the conception, experimental design, data interpretation, and critical revision of the manuscript. All authors reviewed and edited the manuscript. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
This work was supported by the National Natural Science Foundation of China (No. 91849209, 81803016, 81703360, 81903539); the Natural Science Foundation of Shandong Province (No. ZR2019HB012, ZR2021MC189); the China Postdoctoral Science Foundation (No. 2019M650157).
All data needed to evaluate the conclusions are presented in the paper or the Supplementary Materials. The raw data of the RNA‐seq results were submitted to the Gene Expression Omnibus database, with the approval number of GSE203025.
All animal experiments were performed according to the guidelines approved by the Institutional Animal Care and performed following the guidelines for Animal Experimentation of Qingdao University and approved by the Ethics Committee of Medical College of Qingdao University (QDU‐AEC‐2022093).
Click here for additional data file. |
|||
PMC9648390 | 36129048 | Elena Verdugo,Iker Puerto,Miguel Ángel Medina | An update on the molecular biology of glioblastoma, with clinical implications and progress in its treatment | 21-09-2022 | cancer molecular biology,diagnosis,glioblastoma multiforme,ongoing clinical trials,targeted therapy,tumor heterogeneity,tumor metabolism | Abstract Glioblastoma multiforme (GBM) is the most aggressive and common malignant primary brain tumor. Patients with GBM often have poor prognoses, with a median survival of ∼15 months. Enhanced understanding of the molecular biology of central nervous system tumors has led to modifications in their classifications, the most recent of which classified these tumors into new categories and made some changes in their nomenclature and grading system. This review aims to give a panoramic view of the last 3 years’ findings in glioblastoma characterization, its heterogeneity, and current advances in its treatment. Several molecular parameters have been used to achieve an accurate and personalized characterization of glioblastoma in patients, including epigenetic, genetic, transcriptomic and metabolic features, as well as age‐ and sex‐related patterns and the involvement of several noncoding RNAs in glioblastoma progression. Astrocyte‐like neural stem cells and outer radial glial‐like cells from the subventricular zone have been proposed as agents involved in GBM of IDH‐wildtype origin, but this remains controversial. Glioblastoma metabolism is characterized by upregulation of the PI3K/Akt/mTOR signaling pathway, promotion of the glycolytic flux, maintenance of lipid storage, and other features. This metabolism also contributes to glioblastoma's resistance to conventional therapies. Tumor heterogeneity, a hallmark of GBM, has been shown to affect the genetic expression, modulation of metabolic pathways, and immune system evasion. GBM's aggressive invasion potential is modulated by cell‐to‐cell crosstalk within the tumor microenvironment and altered expressions of specific genes, such as ANXA2, GBP2, FN1, PHIP, and GLUT3. Nevertheless, the rising number of active clinical trials illustrates the efforts to identify new targets and drugs to treat this malignancy. Immunotherapy is still relevant for research purposes, given the amount of ongoing clinical trials based on this strategy to treat GBM, and neoantigen and nucleic acid‐based vaccines are gaining importance due to their antitumoral activity by inducing the immune response. Furthermore, there are clinical trials focused on the PI3K/Akt/mTOR axis, angiogenesis, and tumor heterogeneity for developing molecular‐targeted therapies against GBM. Other strategies, such as nanodelivery and computational models, may improve the drug pharmacokinetics and the prognosis of patients with GBM. | An update on the molecular biology of glioblastoma, with clinical implications and progress in its treatment
Glioblastoma multiforme (GBM) is the most aggressive and common malignant primary brain tumor. Patients with GBM often have poor prognoses, with a median survival of ∼15 months. Enhanced understanding of the molecular biology of central nervous system tumors has led to modifications in their classifications, the most recent of which classified these tumors into new categories and made some changes in their nomenclature and grading system. This review aims to give a panoramic view of the last 3 years’ findings in glioblastoma characterization, its heterogeneity, and current advances in its treatment. Several molecular parameters have been used to achieve an accurate and personalized characterization of glioblastoma in patients, including epigenetic, genetic, transcriptomic and metabolic features, as well as age‐ and sex‐related patterns and the involvement of several noncoding RNAs in glioblastoma progression. Astrocyte‐like neural stem cells and outer radial glial‐like cells from the subventricular zone have been proposed as agents involved in GBM of IDH‐wildtype origin, but this remains controversial. Glioblastoma metabolism is characterized by upregulation of the PI3K/Akt/mTOR signaling pathway, promotion of the glycolytic flux, maintenance of lipid storage, and other features. This metabolism also contributes to glioblastoma's resistance to conventional therapies. Tumor heterogeneity, a hallmark of GBM, has been shown to affect the genetic expression, modulation of metabolic pathways, and immune system evasion. GBM's aggressive invasion potential is modulated by cell‐to‐cell crosstalk within the tumor microenvironment and altered expressions of specific genes, such as ANXA2, GBP2, FN1, PHIP, and GLUT3. Nevertheless, the rising number of active clinical trials illustrates the efforts to identify new targets and drugs to treat this malignancy. Immunotherapy is still relevant for research purposes, given the amount of ongoing clinical trials based on this strategy to treat GBM, and neoantigen and nucleic acid‐based vaccines are gaining importance due to their antitumoral activity by inducing the immune response. Furthermore, there are clinical trials focused on the PI3K/Akt/mTOR axis, angiogenesis, and tumor heterogeneity for developing molecular‐targeted therapies against GBM. Other strategies, such as nanodelivery and computational models, may improve the drug pharmacokinetics and the prognosis of patients with GBM.
Abbreviations 2‐HG 2‐Hydroxyglutarate 5hmC 5‐Hydroxymethylcytosine ACC Acetyl‐CoA carboxylase ACSS2 Acyl‐CoA synthetase short chain family member 2 AHNAK2 AHNAK nucleoprotein 2 AKT Alpha serine/threonine‐protein kinase ANKRD10 Ankyrin repeat domain 10 ANXA2 Annexin A2 ANXA7 Annexin A7 APC APC regulator of Wnt signaling pathway ASCL1 Achaete‐scute family BHLH transcription factor 1 ATRX ATP‐dependent helicase ATRX AXIN Axis inhibitor B4GALT3 Beta‐1,4‐galactosyltransferase 3 BBB Blood‐brain barrier BCL6 BCL6 transcription repressor BCORL1 BCL6 corepressor like 1 BIN Bridging integrator BMP2 Bone morphogenetic protein 2 BRAF Serine/threonine‐protein kinase B‐Raf BTK Bruton's tyrosine kinase CAD Carbamoyl‐phosphate synthetase 2 CCL2 C‐C motif chemokine ligand 2 CCNB1 Cyclin B1 CCND1 Cyclin D1 CD24 Small cell lung carcinoma cluster 4 antigen CD27 Tumor necrosis factor receptor superfamily member 7 CD3+ Cluster of differentiation 3 CD31 Platelet endothelial cell adhesion molecule CD41 Integrin subunit alpha 2b CD44 Homing cell adhesion molecule CD8+ Cluster of differentiation 8 CD99 Single‐chain type‐1 glycoprotein CDC6 Cell division cycle 6 CDK4/6 Cyclin‐dependent kinases 4 and 6 CDKN2A/B Cyclin‐dependent kinase inhibitors 2A/2B CELF2 CUGBP Elav‐like family member 2 C‐GBMs Cerebellar glioblastomas CGGA Chinese Glioma Genome Atlas c‐KIT Tyrosine‐protein kinase KIT CMV Cytomegalovirus c‐Myc Master Regulator of Cell Cycle Entry and Proliferative Metabolism C CNS Central nervous system CSF Cerebrospinal fluid CTLA‐3 Cytotoxic T‐lymphocyte‐associated protein 3 CTLA‐4 Cytotoxic T‐lymphocyte‐associated protein 4 Cx43 Connexin 43 CXCL12 C‐X‐C motif chemokine ligand 12 CXCR4 C‐X‐C chemokine receptor type 4 DECR1 2,4‐ dienoyl‐CoA reductase 1 DEGs Differentially expressed genes DHODH Dihydroorotate dehydrogenase DNA Deoxyribonucleic acid DSC‐MRI Dynamic susceptibility contrast magnetic resonance imaging DTI Diffusion tensor imaging E2F7 E2F transcription factor 7 EGFR Epidermal growth factor receptor EGFRvIII Epidermal growth factor receptor variant III ErbB2 ErbB2 Receptor tyrosine kinase 2 EVs Extracellular vesicles FABP3/7 Fatty‐acid binding proteins 3/7 FASN Fatty‐acid synthase FDA Food and Drug Administration FN1 Fibronectin 1 FRP Frizzled‐related protein GABRA1 Gamma‐aminobutyric acid type A receptor subunit alpha1 GBM Glioblastoma multiforme GBP2 Guanylate binding protein 2 GBSCs Glioblastoma stem cells GFAP Glial fibrillary acidic protein GICs Glioma‐initiating cells GLUT1 Glucose transporter 1 GLUT3 Glucose transporter 3 G‐MCI Gene‐mediated cytotoxic immunotherapy GM‐CSF Granulocyte‐macrophage colony‐stimulating factor GNB2 G‐protein subunit beta 2 GNB3 G‐protein subunit beta 3 GNB4 G‐protein subunit beta 4 GNB5 G‐protein subunit beta 5 GO Gene Ontology GPR17 G‐protein‐coupled receptor 17 GSCs Glioma stem cells GSH Glutathione GSK‐3β Glycogen synthase kinase‐3‐beta H3F3A H3.3 histone A HDAC1 Histone deacetylase 1 HDACIs Histone deacetylase inhibitors HER‐2 Human epidermal growth factor receptor 2 HGGs High‐grade gliomas Hh Hedgehog HIF Hypoxia inducible factor HIF‐1α Hypoxia inducible factor 1 subunit alpha HIF‐2α Hypoxia inducible factor 2 subunit alpha HK2 Hexokinase 2 HPSE Heparanase IDH Isocitrate dehydrogenase IDH1 Isocitrate dehydrogenase 1 IDH2 Isocitrate dehydrogenase 2 IDO Indoleamine 2,3‐dioxygenase IGFBP2 Insulin like growth factor binding protein 2 IgG1 Immunoglobulin G1 IL‐10 Interleukin 10 IL‐12 Interleukin 12 JAK Janus activated kinase KEGG Kyoto encyclopedia of genes and genomes KIF20A Kinesin family member 20A KIF23 Kinesin family member 23 KMT2C Lysine methyltransferase 2C KMT2D Lysine methyltransferase 2D LAG‐3 Lymphocyte‐activation gene 3 LDH‐A Lactate dehydrogenase A LGGs Low‐grade gliomas lncRNA Long noncoding RNA LOXL1 Lysyl oxidase like 1 MAPK Mitogen‐activated protein kinase MAPK1 Mitogen‐activated protein kinase 1 MARK4 Microtubule affinity regulating kinase 4 MAX MYC associated factor X MDM2 Murine double minute 2 MEK1/2 Mitogen‐activated protein kinase kinases 1 and 2 MET MET proto‐oncogene receptor tyrosine kinase MGMT O‐6‐Methylguanine‐DNA methyltransferase MHC Major histocompatibility complex MHC‐I Major histocompatibility complex class I MIR4435‐2HG MIR4435 host gene 2 miRNA Micro RNA MKI67 Marker of proliferation Ki‐67 MMP‐2/9 Matrix metalloproteinases 2 and 9 MPC1 Mitochondrial pyruvate carrier 1 MRI Magnetic resonance imaging mRNA Messenger ribonucleic acid mt‐DNA Mitochondrial DNA mTOR Mammalian target of rapamycin NAD+ Nicotinamide adenine dinucleotide NADPH Nicotinamide adenine dinucleotide phosphate hydrogen ncRNA Noncoding RNA NEFL Neurofilament light‐chain gene NES Nestin NF1 Neurofibromin 1 NFATC3 Nuclear factor of activated T cells 3 NF‐κB Nuclear factor kappa‐light‐chain‐enhancer of activated B cells NF‐κB1/2 Nuclear factor kappa B subunits 1 and 2 NG2 Neuron‐glial antigen 2 NOTCH1 Neurogenic locus notch homolog protein 1 NPs Nanoparticles NR4A1 Nuclear receptor subfamily 4 group A member 1 NSCs Neural stem cells OLIG2 Oligodendrocyte lineage transcription factor 2 OPC Oligodendrocyte progenitor cell OS Overall Survival OSMR Oncostatin M receptor beta OX‐40 TNF receptor superfamily member 4 PARP‐1 Poly(ADP‐ribose) polymerase 1 PD‐1 Programmed cell death protein 1 PDGF Platelet‐derived growth factor PDGFR Platelet‐derived growth factor receptor PDGFRA Platelet‐derived growth factor receptor alpha PDIA3 Protein disulfide isomerase family A member 3 PDK1/2 3‐Phosphoinositide‐dependent kinases 1 and 2 PD‐L1 Programmed death ligand‐1 PDPN Podoplanin PFKP Phosphofructokinase platelet PFS Progression‐free survival PHIP Pleckstrin homology domain interacting protein PI3K Phosphatidylinositol 3‐kinase PIK3CA Phosphatidylinositol‐4,5‐bisphosphate 3‐kinase catalytic subunit alpha PLCG1 Phospholipase C gamma 1 POLR2F RNA polymerase II, I and III subunit F poly‐ICLC polyinosinic‐polycytidylic acid PSMA Prostate‐specific membrane antigen PTEN Phosphatase and tensin homolog PTPN11 Protein tyrosine phosphatase non‐receptor type 11 PUFA Polyunsaturated fatty acids RAF Rapidly accelerated fibrosarcoma kinase RAS Rat sarcoma virus GTPase Rb Retinoblastoma protein RNA Ribonucleic acid RNA‐LP RNA‐lipid particle RON Ron receptor tyrosine kinase RPL39L Ribosomal protein L39 like RTK Receptor tyrosine kinase RTK I Receptor tyrosine kinase I RTK II Receptor tyrosine kinase II RT‐qPCR Real‐time polymerase chain reaction RYR2 Ryanodine receptor 2 SDF‐1 Stromal cell‐derived factor 1 siRNA Small interfering RNA SLC12A5 Solute carrier family 12 member 5 SLC7A11 Solute carrier family 7 member 11 SNHG12 Small nucleolar RNA host gene 12 SNRPB Small nuclear ribonucleoprotein polypeptides B and B1 SNVs Single nucleotide variants SOX1 SRY‐box transcription factor 1 SOX10 SRY‐box transcription factor 10 SOX40 SRY‐box transcription factor 40 SREBP‐1 Sterol regulatory element‐binding protein 1 STAT Signal transducer and activator of transcription STAT3 Signal transducer and activator of transcription 3 SUSD2 Sushi domain containing 2 SVZ Subventricular zone SYT1 Synaptotagmin‐1 TAMs Tumor‐associated macrophages TCGA The Cancer Genome Atlas TEM7 Tumor endothelial marker TERT Telomerase reverse transcriptase TGFBR2 Transforming growth factor‐beta receptor II TGF‐β Transforming growth factor beta tGLI1 Glioma‐associated oncogene homolog 1 truncated variant TICs Tumor‐initiating cells TIM‐3 T cell immunoglobulin and mucin‐domain containing‐3 TME Tumor microenvironment TMEM52 Transmembrane protein 52 TMZ Temozolomide TNF‐α Tumor necrosis factor alpha TNTs Tunneling nanotubes TP53 Tumor protein p53 TRADD TNFR1‐associated death domain protein Treg Regulatory T cells USP5 Ubiquitin specific peptidase 5 VCL Vinculin VEGF Vascular endothelial growth factor VEGF‐A Vascular endothelial growth factor A VEGF‐C Vascular endothelial growth factor C VEGFR1 Vascular endothelial growth factor receptor 1 VILL Villin‐like protein VIM Vimentin VPA Valproic acid WGCNA Weighted gene correlation network analysis WHO World Health Organization Wnt Wingless/Integrated WT1 Wilms tumor gene‐1 WWOX WW domain containing oxidoreductase YKL40 Chitinase‐3‐like protein 1 α‐KG Alpha ketoglutarate
Glioblastoma multiforme (GBM) is the most aggressive and common type of malignant primary brain tumor. The incidence of GBM increases with age and is slightly higher in men than in women [1]. GBM's incidence oscillates between 0.59 and 5 cases per 100,000 people and is rising in many countries owing to the aging population and improvements in diagnosis, among other factors [2]. Despite the considerable increase in knowledge about the molecular pathogenesis and biology of this tumor, patients with GBM continue to suffer from poor prognoses. They have a median survival of ∼15 months [3] and a 5‐year relative survival rate of only 6.8%, although this could depend on the patient's sex and age at diagnosis [4]. Since 2005, the treatment regimen for newly diagnosed patients comprises surgery followed by concurrent radiotherapy with temozolomide (TMZ) and further adjuvant TMZ [5]. In recent years, clinical trials testing new drugs and strategies have been rising, particularly those on immunotherapy and targeted therapies [1]. Although our group published a review on the literature related to the molecular biology of glioblastoma in 2019 [6], given the remarkable amount of research related to GBM and its classification, characterization, and treatment conducted within the last 3 years, an update on the topic was advisable. The present review aimed to gather information on the latest advances in understanding the molecular biology of glioblastoma, their clinical implications, and the latest therapeutic advancements.
The origin of IDH‐wildtype GBMs has been described as a neuronal network that starts in the subventricular zone (SVZ) and spreads toward the frontotemporal cortex and lobe, thus creating a “firework” pattern [7]. Tumoral progression is possible because of the presence of astrocyte‐like neural stem cells at the astrocytic ribbon, whose mutations gradually accumulate as they reach the cortex. This origin of primary glioblastomas has been confirmed [8]. On the contrary, a second origin was proposed, highlighting the genesis of outer radial glial‐like cells from astrocytes showing a high expression of ErbB2, a tyrosine kinase receptor implicated in cell proliferation and motility [9]. A series of studies hypothesized the “double origin” of GBM from a mixed population of ventricular and outer radial glial cells in SVZ. The epidermal growth factor receptor variant III (EGFRvIII) is responsible for reprogramming during proliferation, regardless of whether GBM originates from neural stem cells or GFAP‐positive progenitors [8]. Subsequently, glioblastoma consists of a heterogeneous cell population derived from glioma stem cells (GSCs) located within a vascularized tumor niche. GSCs are astrocyte‐like neural stem cells that are prevalent in the SVZ. These cells take advantage of a weakened immune system and proliferate because of the overproduction of growth factors in the perivascular region, stimulated by the release of cytokines. Accompanied by this tumor niche, neural stem cells (NSCs) from the germinal vascular zones establish crosstalk with GSCs and also exhibit cell differentiation capabilities [10]. Lombard et al. [11] summarized the similarities between adult NSCs and GSCs and their effect on the prognosis of a patient with glioblastoma due to recurrence and drug resistance. These two stem cell types are associated with vasculature; niche companions, such as pericytes and endothelial cells; migration and proliferation regulation; and nestin expression. Furthermore, GSCs expose the mutated genes expressed in NSCs: TERT, TP53, PTEN, EGFR, and PDGF [12]. CXCL12 and pleiotrophin might play a role in the migration of GSCs from the tumor niche to the SVZ, heading to the exclusive transformation of NSCs in this brain region. NSCs in other neurogenic niches, such as the hippocampus, are not involved in gliomagenesis [10]. In fact, the SVZ is in contact with the cerebrospinal fluid (CSF), which might interfere with healthy cell growth and is partially responsible for the malignancy of GBM in the proximity of the SVZ. Proliferating GSCs near the SVZ might receive altered genetic information via the CSF. In addition, NSCs in the SVZ can undergo somatic mutations, leading to uncontrolled proliferation and genetic alterations similar to the progenitor cells in IDH‐wt and IDH‐mutant mouse xenografts. Further, Lozano‐Ureña et al. [13] demonstrated that adult NSCs could not be recognized from GSCs based on their genetic expressions. Tumor recurrence might be mediated by glioma‐initiating cells (GICs), reactivated by their presence in the parenchyma, where they stayed during the dormant period [8, 14]. Yoon et al. [7] hypothesized an alternative for tumor recurrence, which comes from the remigration of these dormant cells to the tumor niche from the SVZ. NSCs can be selected as the cell lineage for the origin of gliomas based on their location (SVZ), differentiation properties, and matching variations with glioblastoma. IDH‐wildtype patients show changes in the gene expression of TERT, TP53, PDGFR, and EGFR in these cells. Nevertheless, mature astrocytes can dedifferentiate and reprogram into tumor cells, and oligodendrocyte progenitor cell‐like cells (OPC‐like cells) can redirect their transcriptome to accelerate the uncontrolled proliferation of malignant glioblastomas.
Historically, glioma classification was based on histological and immunohistochemical criteria. The classical diagnostic methods for gliomas are based on imaging or screening tests, such as functional magnetic resonance imaging (MRI), positron emission tomography, computed tomography, and the performance of a liquid biopsy, which is a non‐invasive technique used to confirm the diagnosis and augment treatment prospects [15, 16]. Additional molecular diagnoses can be performed to provide a more personalized prognosis and enhance the chances of therapeutic efficacy. Despite the low impact of molecular classification in medical diagnosis, during the past few years, there have been remarkable advances in this field, especially for the central nervous system (CNS) tumor classification, which was included in the fifth edition of the World Health Organization (WHO) Classification of Tumors of the Central Nervous System, published in 2021 [17]. This new edition integrates molecular changes with clinicopathological utility essential for accurately classifying CNS tumors. This edition also introduces changes to the former taxonomy and nomenclature, including the term “type” instead of “entity” and “subtype” instead of “variant.” Traditional names that refer to histological features, such as anaplastic, malignant, or giant cells, can still be used for medical recognition but are likely to disappear in future classifications [18]. Arabic numerals are now used to grade neoplasms within types, whereas the past edition used Roman numerals to grade neoplasms across different tumor types [19]. Fourteen newly identified neoplasms have been incorporated into the categories of gliomas, glioneuronal tumors, and neuronal tumors. The WHO divided this category of CNS tumors into the following six families: 1) Adult‐type diffuse gliomas, 2) Pediatric‐type diffuse low‐grade gliomas, 3) Pediatric‐type diffuse high‐grade gliomas, 4) Circumscribed astrocytic gliomas, 5) Glioneuronal and neuronal tumors, and 6) Ependymomas (Table 1). As this classification suggests, diffuse gliomas that primarily occur in adults and those that mainly in children have been separated prognostically and biologically into different groups. Moreover, pediatric gliomas are segregated into low‐grade gliomas, those that exhibit diffused growth in the brain but have less‐specific histological features, and high‐grade gliomas. Integrating molecular and histopathological information is essential for precisely diagnosing these tumors. It should also be noted that the term “glioblastoma” has been discarded to identify pediatric‐type gliomas, which are referred to as those that affect the 0‐14 years old age group [17, 20]. Glioblastoma, IDH‐wildtype, was integrated into the gliomas, glioneuronal tumors, and neuronal tumors category within the adult‐type diffuse gliomas. Previously, IDH‐mutant diffuse astrocytic tumors were assigned to three different types: 1) diffuse astrocytoma, 2) anaplastic astrocytoma, or 3) glioblastoma. Singer et al. [21] proposed a new classification for IDH‐mutant astrocytoma because of its lower aggressiveness compared with diffuse midline gliomas and IDH‐wildtype glioblastomas. The current classification identifies all IDH‐mutant diffuse astrocytic tumors as astrocytoma, IDH‐mutant, which can be graded as CNS WHO Grade 2, 3, or 4 [22]. Microvascular proliferation and necrosis rates are proposed as determinants for oligodendrogliomas, now defined by IDH1/2 mutations, 1p/19q codeletion, TERT promoter mutations, and NOTCH1. On the contrary, histone H3.3 G34‐mutant gliomas are characterized by OLIG2 and ATRX mutations, and cerebellar glioblastomas (C‐GBMs) are described as “high‐grade astrocytoma with piloid features” with IDH, ATRX, and CDKN2A/B mutations [21]. Exclusive alterations in ATRX and PDGFRA can define C‐GBMs, with most of these tumors exhibiting IDH1/TP53 mutations and the upregulation of NG2 and NR4A1 [23].
The number of published articles on glioblastoma and its genetics has increased exponentially during the last decade [24]. The 2021 WHO Classification of CNS Tumors defines three genetic parameters for diagnosing glioblastoma, IDH‐wildtype: TERT promoter mutation, epidermal growth factor receptor (EGFR) amplification, and the combined gain of entire chromosome 7 and loss of entire chromosome 10 [17]. However, this neoplasm can be further classified into molecular subtypes, which can impact disease progression and clinical practice. In 2016, Verhaak et al. [25]classified glioblastomas into four subtypes based on their molecular features: neuron, astrocyte, oligodendrocyte, and cultured astrocytic gliomas. Detecting these subtypes relies on the different therapeutic approaches required for each patient and their impacts on tumor progression [26]. Recently, Neftel et al. [27] identified four heterogeneous cellular states using single‐cell RNA‐sequencing and validated the intratumoral heterogeneity present in GBM and the relevance of this subtyping. They classified the development of neural signatures into neural‐progenitor‐like, oligodendrocyte‐progenitor‐like, astrocyte‐like, and mesenchymal‐like states [26]. The neural subtype is derived from astrocytes and oligodendrocytes and expresses neuron‐related genes, whereas the proneural subtype exhibits the characteristics of oligodendroglial cells and develops in young patients [25]. The classical subtype possesses astrocytic features and expresses neuron precursor and stem cell markers, whereas the mesenchymal subtype shows characteristics of cultured astrocytic gliomas [28]. Verhaak's latest update for reclassifying gliomas removed the neural subtype because it is problematic for identifying primary and recurrent gliomas owing to its ongoing genomic signature changes [29]. Further, a study dedicated to the evolution of the tumor determined that the vast majority of The Cancer Genome Atlas (TCGA) subtypes were from a proneural‐like precursor and switched to a mesenchymal‐like state in a differentiation process regulated by TNF‐α/NF‐κB signaling or ASCL1. To sum up, Neftel et al.’s studies [27]and the recent update of the WHO CNS tumor classification reflect the fluidity of GBM's transcriptional states and the influence of the tumor microenvironment (TME) on the development and transition from one subtype to another [22]. Jankowska et al. [30] classified glioblastoma subtypes based on immunochemical expression and concluded that the classical subtype is represented by TP53 mutation, which makes this subtype highly sensitive to classical radiotherapy plus chemotherapy with adjuvant TMZ. The mesenchymal subtype shows NF1, PTEN, AKT, MET, and TRADD mutations. Further, PDGFRA, IDH1, TP53, HIF, and OLIG2 mutations are characteristic of the proneural subtype. Another study revealed the neuronal markers for identifying and profiling neural glioblastomas, such as NEFL, GABRA1, SLC12A5, and SYT1 (Table 2) [31]. In parallel to this classification, Herrera‐Oropeza et al. [32] performed a multi‐omics analysis of driver genes. They concluded that mesenchymal subtype development was related to the upregulation of the MGMT promoter and the downregulation of ATRX, H3F3A, TP53, and EGFR. Complementary information was provided for the proneural subtype, characterized by the overexpression of MKI67 and OLIG2, and the classical subtype by the overexpression of EGFR, NES, VIM, and TP53. The characterization of differential molecular characteristics of histologically similar tumors is relevant to improve the diagnosis of GBM in patients. Besides, the determination of expression profiles is useful for creating progression models and enhancing the prognosis of each tumor subtype. There is a rising tendency to classify gliomas based on their mRNA sequencing and the clustering of samples with computational programs. Using the “Consensus Cluster Plus” package for R v4.0.3, Cai et al. [33] investigated the reclassification of glioma based on the expression levels of Gβ/γ genes from TCGA and the Chinese Glioma Genome Atlas (CGGA) datasets. The result was a differential distribution map correlated with the samples analyzed from these two chosen databases. The Gβ/γ heterodimer can activate the Erk1/2 pathway by inducing the overexpression of guanine nucleotide‐binding protein beta 4 (GNB4), which results in the transformation of epithelial and mesenchymal cells into glioma cells. After clustering, they obtained three subgroups: GNB2, GNB3, and GNB5. GNB2 appeared to be the best indicator of malignant tumors, especially in patients with IDH‐mutated, non‐codeleted 1p/19q low‐grade gliomas (LGGs). This subgroup is characterized by high M0/M2 cell infiltration levels and is highly associated with the immunosuppressive phenotype, thus demonstrating enhanced PI3K‐Akt/JAK‐STAT pathways and high levels of tumor‐associated macrophages (TAMs) and M2 macrophages. Therefore, the GNB2 subgroup would represent the immunosuppressive phenotype in gliomas. Each subgroup has a unique tumor‐related pathway that can answer the selection of a chemotherapeutic drug and enhance the glioma prognosis by choosing the right target [33]. An analysis of the differentially expressed genes (DEGs) revealed 110 upregulated genes and 75 downregulated genes in the GBM samples. This observation identified a four‐protein prognostic signature (SLC12A5, CCL2, IGFBP2, and PDPN) for the segregation of patients into high‐ and low‐risk groups and for the estimation of survival time. These were observed via a weighted gene correlation network analysis (WGCNA) algorithm, a strategy that has also been used to determine disease‐related genes in other oncologic diseases [34]. Another DEG screening found 662 DEGs in patients with GBM and concluded that DECR1, POLR2F, HDAC1, and PDIA3 could be the critical genes related to the overall survival (OS) time of patients with GBM [35]. A WGCNA study comparing the transcriptome and proteome of glioblastoma, IDH‐wildtype tumors found six proteomic modules correlated with survival, but none of the identified RNA modules did [36]. After performing a Kaplan–Meier analysis, 11 proteins were revealed to have a significant association with survival, despite not being significant at the RNA level. Owing to the apparent lack of correlation between RNA and survival, previously established single‐cell‐based signatures were used to define the dominant cell subpopulation of each tumor analyzed. This study revealed that mesenchymal and neural progenitor cell‐like subpopulation signature genes correlated with shorter survival, whereas oligodendrocytic precursor cell‐like and astrocytic subpopulation signature genes correlated with more prolonged survival. Gene Ontology (GO) enrichment analysis from the proteomic and single‐cell‐based signature data revealed that lysosomal activity and amino and nucleotide sugar metabolism were enriched in a cluster of genes and proteins correlated with short survival [36]. Transcriptomic and proteogenomic profiling techniques have been further developed for the classification of gliomas; these could lead to a robust and objective method for the stratification of patients and improve survival prediction, as commented in the previous survival [36]. After a regular medical diagnosis of glioblastoma is achieved via MRI, computed tomography, or biopsy followed by blood analysis, patients await a more personalized and guided treatment [38]. The repertoire of molecular biomarkers characterized by transcriptomics is quite robust. However, only a few are critical for a detailed diagnosis. Further research is needed to classify glioblastomas into subtypes and grades and to estimate survival rates. The most relevant biomarkers described in this section are the mutation of TP53 as the molecular feature of classical glioblastoma, the PDGFRA mutation for the proneural subtype, the presence of GNB2 as an indicator of aggressive tumors, and the cell subpopulation signature as a measure for survival.
Molecular classification can also reveal an age‐related pattern of biomarkers for glioma. This evidence is reflected in a previous computational clustering study that revealed H3F3A, AHNAK2, SOX1, SUSD2, and KMT2C were the most mutated genes in young‐age patients, PIK3CA and TERT were the most mutated genes in middle‐age patients, and RYR2 was the most mutated gene in old‐age patients. Furthermore, two mutations were relevant for young‐ and middle‐age groups: BCORL1 (as an indicator for HGGs) and KMT2D, whereas three mutational events on TERT, PTEN, and NF1 were more frequent in old‐age patients [39]. The characterization of these new biomarkers could provide a more refined molecular classification of HGG/LGGs between age groups when added to the IDH1 mutational status and TERT methylating status. However, these data are based on the IV WHO classification of CNS tumors, and thus they should be updated [18]. RNA‐sequencing and real‐time polymerase chain reaction (RT‐qPCR) quantification followed by Kyoto Encyclopedia of Genes and Genomes (KEGG) and gene ontology (GO) analyses revealed a sex‐related pattern that emerged from the differential expression of hub genes in the cell cycle, DNA replication, and the Fanconi anemia pathway. These DEGs were mainly enriched in women because the involved pathways mediated progesterone release, which leads to oocyte maturation. Four strongly correlated genes (CCNB1, CDC6, KIF23, and KIF20A) were upregulated in the glioma samples and mediated cell cycle, ATP‐binding, and DNA replication. The CCNB1 protein accelerates mitosis and promotes tumoral invasion, thus suggesting a recurrent role in GBM [40]. CDC6 encodes an enzyme that mediates mitosis via E2F regulation [41]. KIF23 is highly expressed in malignant tumors, and KIF20A promotes reverse transport from the Golgi complex to the endoplasmic reticulum and the presentation of the major histocompatibility complex class I (MHC‐I), thereby disguising the tumor from immune response and maintaining its proliferation [42, 43]. These last two hub genes might be potential biomarkers for GBM diagnosis, especially in women [44]. These molecular approaches aim to specify a pattern of DEGs between age or sex groups. On the one hand, distinctive DEG clusters between age groups have been observed that could be useful for classifying the tumors of the CNS following the most recent criteria, especially to distinguish the pediatric‐type diffuse LGGs and HGGs. On the other hand, an RNA‐sequencing study led to the characterization of four hub genes highly related to glioma samples, two of them being possible new biomarkers for GBM diagnosis in women: KIF23 and KIF20A.
Survival prediction can be discussed from an epigenetic perspective. The most frequently observed molecular feature is the status of the MGMT promoter, whose methylation level correlates with the tumor's prognosis and is considered a universal marker to evaluate TMZ sensitivity in glioma chemotherapy. In fact, the MGMT promoter methylation level is more significant than grade or 5‐hydroxymethylcytosine (5hmC) for age‐related prognosis [14, 45, 46]. Conventional chemotherapy with TMZ as adjuvant treatment is an inductor of DNA damage and leads to genetic alterations in the glioma cells, which adapt to the drug dose and develop resistance when the MGMT promoter is hypermethylated [45]. Methylation profiling is another interesting strategy to stratify GBM tumors. DNA methylation‐based GBM subtypes seem related to local T‐cell infiltration [47]. In fact, these immunological characteristics lead to the classification into four methylation subgroups: IDH, RTK I, RTK II, and mesenchymal tumors. Interestingly, IDH methylation groups have the lowest CD3+ T‐cell infiltration and a low PD‐1 expression. Mesenchymal subtype tumors have the highest CD3+/ CD8+ T‐cell infiltration. An increased PD‐1 expression along with higher levels of CD8+ infiltration results from radiochemotherapy, suggesting that CD8+ T‐cells might evolve to an anergic phenotype and activate the immunosuppressive response. Consequently, the mesenchymal subtype might become more aggressive against immune response after conventional therapy. Thus, this information could help identify patients suitable for specific immunotherapy trials [47]. DNA methylation‐based diagnosis could support the histological diagnosis of GBM by combining the transcriptomic and methylation patterns of tumor samples and measuring the methylation degree of the CpG islands [45]. Using the MethylMix algorithm, Wang et al. [48] revealed six highly methylated genes (ANKRD10, BMP2, LOXL1, RPL39L, TMEM52, and VILL) that could be used for the molecular subclassification of GBM. The methylation signature is an independent factor that might predict high‐ and low‐risk glioblastomas and overall survival. Another epigenetic modification contributing to cancer proliferation is the aberrant methylation of histones, a process regulated by histone methyltransferases. An active research field in GBM therapy relies on applying histone deacetylase inhibitors (HDACIs) to improve the patient's OS. An excellent example of the application of HDACIs to treat GBM is a phase II/III trial designed with HDACIs + gene‐mediated cytotoxic immunotherapy (G‐MCI) or gene‐editing treatment mediated by zinc‐finger, CAS enzyme, and new‐generation sequencing findings [45]. A phase II study tested valproic acid (VPA), an HDAC inhibitor, in newly diagnosed patients showing improved overall survival outcomes and lower toxicity. VPA sensitized glioblastoma cells to radiation in 81% of the patients, thus increasing the effectiveness of standard radiotherapy [49]. It has been observed that receptor tyrosine kinases I (RTK‐I) subtype GBM show global hypomethylation. The SOX10 gene, linked to chromatin remodeling and therapy resistance in melanoma, is hypomethylated and overexpressed in RTK‐I subtype GBM. Repression of this gene in an in vivo syngeneic graft GBM mouse model resulted in epigenetic alterations, a phenotypic switch to a mesenchymal subtype, and increased tumor cell invasion. In this case, the RTK‐I subtype is related to better overall survival than the mesenchymal subtype [50]. Some super‐enhancers involved in the regulation of cell identity genes show subtype‐specific enrichment. Consequently, the status of the enhancer landscape plays an essential role in determining tumor subtype identity in GBM, and their enrichment could serve as a biomarker for the molecular diagnosis of specific subtypes [50]. Alternative‐splicing profiling represents a novel technique for glioblastoma classification. ANXA7, MARK4, MAX, USP5, WWOX, BIN, RON, and CCND1 have been suggested as altered biomarkers serving as functional targets for personalized treatment depending on the heterogeneity of the phenotype and genotype of each patient [51, 52, 53]. Moreover, SNRPB, a vital element of the spliceosome complex SmB/B′ implicated in DNA repair and chromatin remodeling, might be a potential target for novel therapies [54]. Additionally, CELF2, a regulator of splicing events, could be a valuable predictor of the prognosis, along with the IDH status and the zinc‐finger motif deletions (3′ ZNF domain alterations) [55]. In summary, the MGMT promoter methylation level is an indicator of age‐related gliomas, prognosis, and immunoresistance. However, this is not the only biomarker identified by epigenetic changes. Hypermethylation, hypomethylation, and alternative splicing play an essential role in tumor heterogeneity. The latest clinical trials focused on the methylation of histones and the evaluation of CELF2 as a potential predictor of prognosis. Epigenetic changes are a hallmark of cancer, and alternative splicing is one of their most frequent manifestations. These subtle changes lead to a wide heterogeneity of phenotypes, making epigenomic profiling and characterization two essentials for personalized prognosis.
Noncoding RNAs (ncRNAs), such as microRNAs (miRNAs) and long ncRNAs (lncRNAs), have interesting regulatory effects on GBM. These nucleic acids can potentially modify the expression levels of proteins involved in the proliferation and migration of tumor cells, like metalloproteinases, cytokines, and growth factors [56]. GBM cells exchange miRNA molecules with oligodendrocytes and endothelial cells within the TME. These molecules can promote angiogenesis and cell differentiation, but some work as tumor suppressors [57]. Identifying miRNA with highly altered expression in glioma provides another method of analyzing patient samples via microarray. The diagnosis can be made by detecting only three miRNAs: miR‐4763‐3p, miR‐1915‐3p, and miR‐3679‐5p. The first and second miRNAs appear oncogenic and are higher in patients with diffuse glioma, while the last might be a tumor suppressor because of its lower levels in patients. Although this could be a promising technique, the current results seem inefficient in discriminating diffuse gliomas from healthy tissue. Nevertheless, these three serum miRNAs represent a powerful tool for GBM diagnosis in combination with histological and molecular characterization [58]. The lncRNA MIR4435‐2 Host Gene (MIR4435‐2HG) is upregulated in GBM tissues. Besides, higher expression of this lncRNA correlated with shorter OS. MIR4435‐2HG targets miR‐1224‐5p, which inhibits TGFBR2. The inhibition of this receptor results in a diminished cell invasive potential compared to MIR4435‐2HG overexpressing U87 cells. This result agrees that MIR4435‐2HG knockdown resulted in the inhibition of cell proliferation and increased cell apoptotic rates in U87 and U251 cell lines. Furthermore, this lncRNA can be found in other tumors (e.g., gastric and colorectal cancer), in which its upregulation is also linked to poor [59]. The small nucleolar RNA host gene 12 (SNHG12) is overexpressed in TMZ‐resistant GBM samples after TMZ treatment. Hypomethylation of the promoter region of this lncRNA induces transcriptional activation of SNHG12 by the SP1 transcription factor. SNHG12 acts as a molecular sponge for miR‐129‐5p, increasing the expression of MAPK1 and E2F7, which regulate TMZ‐induced cell apoptosis and cell proliferation. Even though tumor heterogeneity implies that each patient presents distinct differentially expressed lncRNAs, ncRNAs are promising biomarkers that could have relevant clinical significance [60]. In brief, noncoding RNAs seem to play a role in tumoral proliferation. lncRNAs exhibit a poor prognosis linked to higher apoptosis rates, whereas miRNAs can be targeted for tumor suppression. MIR4435‐2HG and SNHG12 are highly expressed in glioblastomas, increasing the tumoral genotypes and heterogeneity. Targeting these molecules could prevent aggressive or high‐grade glioblastomas.
Deep convolutional radiomics features of diffusion tensor imaging (DTI) [61] and machine‐learning assisted dynamic susceptibility contrast‐magnetic resonance imaging (DSC‐MRI) are novel constructional methods that are worth mentioning. These methodologies provide a better molecular classification of gliomas [62, 63] and subtypes of GBM based on analyzing pathological images and their computational modeling via a deep learning method integrating different biomarkers [37, 64]. Deep learning machine analysis is based on computational artificial intelligence that learns from data samples and builds up neural network models that elucidate the diagnosis, decision‐making, and clinical predictions related to GBM therapy [65]. An example of computational modeling is described in a study by Randles et al. [66], where the authors investigated the dynamics of glioblastoma stem cells within the perivascular niche as they designed a computational model on the Vulcan supercomputer, which let them examine different treatments and their outcomes. Each simulation analyzed the spatial distribution and interactions between cells, giving a fitness value to each cell. Following the motion and the spatial landspace of these cells, the supercomputer could determine tumor growth through time. Thus, they concluded that giving chemotherapy with TMZ right before radiotherapy improved survival because of the timing glioblastoma stem cells spread in space. In this manner, the immunoresistant response to TMZ was more effectively blocked. Computational models can interpret both proteogenomic and metabolomic characterizations of GBM. Wang et al. [67], using computational analysis, revealed how PTPN11 and PLCG1 are signaling hub genes in RTK‐altered tumors, how immune cells characterize GBM subtypes, and how histone H2B acetylation is a biomarker for classical glioblastoma. The processing of data collection and interpretation was facilitated by a non‐negative matrix factorization for multi‐omics subtyping, the use of iPROfun for covariates, and a deep learning histopathology image analysis. An enormous variety of studies can be approached by computational modeling, a potential tool for long‐period analysis and multiple condition evaluations. The computational model is designed in terms of the population census and the experimental conditions. Nevertheless, deep‐learning methods make a difference. Advances in processing and refining data compilations might help researchers head in a direction when giving a diagnosis and prognosis to GBM patients. The list of possible molecular biomarkers for diagnosing and classifying gliomas is endless. The following table (Table 3) collects those molecular targets for diagnosis, prognosis, and the personalized treatment mentioned in the previous sections:
Tumor heterogeneity deviates from the “cell niche” regulation and develops from several signaling and immunosuppressive pathways interceptions. Control of the deactivation of cell proliferation, self‐renewal, and differentiation of glioma‐initiating cells is mediated by the Wnt, Notch, and TGF‐β signaling pathways. The mesenchymal subtype is characterized by an overexpression of TGF‐β and vascular endothelial growth factor (VEGF) pathways and attenuation to both Wnt and Notch signaling as well as the expression of YKL40, a specific biomarker for this subtype. On the contrary, Notch and Wnt signaling pathways were prominently activated in the proneural subtype [68]. There is differential activation of GICs specific to each GBM subtype. The concurrency of TGF‐β signaling and lower activation of both Notch and Wnt signaling pathways suggests that targeting GIC subtypes might improve clinical outcomes. Apart from these pathways, p53 signaling remains essential for immortality by amplifying murine double minute 2 (MDM2), which binds the TP53 gene and inhibits its regulatory role in mutations. The retinoblastoma protein (Rb) pathway is also crucial for regulating the cell cycle and proliferation. Rb protein inhibits the E2F transcription factor, which stimulates the transcription of genes involved in the progress from the G1 to S phase during mitosis [69]. These two latest pathways control the cell cycle and their targeted interception might mitigate the invasiveness and migration of glioblastoma cells. There is an opposing interplay between the IDH1 mutation and the Wnt/β‐catenin pathway [70]. The Wnt signaling pathway is crucial in cell proliferation, migration, and apoptosis. However, this pathway inhibits glycogen synthase kinase‐3β (GSK‐3β), an inflammation and cell membrane signaling regulator. IDH1 mutation is related to a better response to cytotoxic therapy and longer survival in GBM patients. The PI3K/Akt pathway is involved in the phosphorylation of GSK‐3β, which leads to the nuclear transport of β‐catenin. This transport promotes the activation of STAT3, an oncogenic transcriptional factor involved in GBM growth, stimulation of cyclin D1 and c‐Myc (related to angiogenesis and proliferation), and overexpression of MMP‐2/9, which induces cell invasion [71, 72]. A clinical trial based on the combined treatment using sulindac + LY294002 [73] aims to inhibit PI3K for the blockade of GBM invasion. Another biological drug inhibiting invasion is celecoxib, a PI3K inhibitor that can also diminish Akt signaling [71]. The concomitant reduction in tumor proliferation is accompanied by increased cell death [72]. The most remarkable epigenetic silencing of the Wnt pathway occurs because of the hypermethylation of soluble frizzled‐related protein (FRP) genes. FRPs create a receptor complex that binds to Wnt ligands and consequently activates the AXIN/APC/GSK‐3β complex via phosphorylation. This last step promotes the accumulation of β‐catenin in the cytosol and leads to the activation of RTKs, therefore, the stimulation of the HIF‐1α via the PI3K/Akt pathway. HIF‐1α is a hypoxia factor that enhances the Warburg effect by overproducing glycolytic enzymes, such as LDH‐A. The final result of FRP silencing is the inhibition of glucose metabolism in the glioma cells [74, 75]. In recent studies, IDH1‐R132H mutation was found to be correlated with better prognosis owing to the decreased expression of the Wnt/β‐catenin pathway. This result could be explained by the lower intracellular glutathione (GSH) levels due to the reduced availability of NADPH, an essential cofactor in the oxidative carboxylation of α ‐ketoglutarate, and higher levels of reactive oxygen species, which induce apoptosis and reduce cell proliferation. IDH1 is an independent predictor of improvement in the clinical outcomes of TMZ therapy. As mentioned above, IDH1‐mutated tumors correlate with a better prognosis for low‐/high‐grade gliomas. Consequently, this type of mutation in patients with glioma reduces proliferation and induces apoptosis [76]. The Notch signaling pathway regulates cell migration, differentiation, apoptosis, self‐renewal, and homeostasis. This pathway consists of four cytoplasmic receptors (Notch 1‐4) and their ligands, Jagged‐1, Jagged‐2, and DII 1‐4. The expression level of Notch 1, predominantly expressed in neurons, astrocytes, precursor/ependymal, and endothelial cells could be related to the GBM survival period. Notch signaling activity might be useful to predict the overall survival and tumor resistance. Results with a novel therapeutic antibody, functionally validated with a computational‐guided approach, suggest that Notch signaling via Hes1/Hey1 targeted genes could be a druggable and clinically relevant target in GBM. Brontictuzumab (BRON) is the first humanized anti‐Notch 1 blocking antibody directed against cell surfaces to diminish tumor cell invasion [77]. The Hedgehog (Hh) signaling pathway plays a crucial role in embryogenesis and tumorigenesis; furthermore, this pathway plays a pivotal role in tissue repair and regeneration. The terminal effectors of the Hh pathway in glioma are glioma‐associated oncogene homolog 1 (GLI1) zinc‐finger transcription factors. An alternative‐splicing, truncated variant, tGLI1, is expressed in most GBM samples, but it is undetectable in normal brain cells. This tGLI1 is a gain‐of‐function variant able to activate several genes not regulated by GLI1. The targeted genes upregulated by tGLI1 include VEGFR1, VEGF‐A, VEGF‐C, TEM7, HPSE, CD24, and CD44, thus promoting glioblastoma cell proliferation, migration, invasion, and angiogenesis [78]. The aberrant role of the Hh pathway leads to the need to understand the impact of GLI variants, potentiating the development of novel therapies that stop metastasis. In summary, one of the main reasons that glioblastomas are so heterogeneous is that they can modulate the core regulatory signaling pathways involved in immune response, apoptosis, cell growth, proliferation, and migration. The evolution of glioma depends on the upregulation or downregulation of three main pathways: TGF‐β, Wnt, and Notch.
A relevant part of the research devoted to molecular‐targeted therapy of GBM focuses on identifying intrinsic biomarkers in the RTK/PI3K/Akt/mTOR, JAK‐STAT3, and RAS/RAF/MEK pathways as well as the p53 and cell cycle regulation pathways. The RTK/PI3K/Akt/mTOR pathway regulates cell growth, metabolism, and survival in gliomas (Figure 1) [46]. mTOR kinase functions in two complexes: as the nutrient sensor of the cell regulating cell growth (mTOR complex 1 + protein RAPTOR) and coordinates the cytoskeleton's organization and Akt activation via phosphorylation (mTOR complex 2). A remarkable distinction between normal and glioma cells is the loss of function of phosphatase and tensin homolog (PTEN). Consequently, the deactivation of PTEN results in increased Akt activity that triggers mTOR activity that enhances cell proliferation [79]. On the contrary, RTKs activate PI3K and lead to the activation of Akt depending on the phosphorylation of protein kinases 1 and 2 (PDK1/2). Thus, the hyperactivation of Akt is pertinent in understanding why glioma cells are permanently proliferating, changing their metabolism, and promoting a cancer phenotype. Resistance to TMZ treatment stems from the role of autophagy in glioma cells induced by the inhibition of this last pathway [79]. The relevance of the hyperactivation of this axis relies on the control of glioblastoma cell survival. This survival is characterized by changes in metabolic, cell cycle, and cell growth pathways and is translated into radiotherapy + TMZ chemotherapy resistance.
This next section introduces the models available to explain tumor heterogeneity, a hallmark of GBM, which is influenced by epigenetics and metabolism. There are two proposed mechanisms for intratumor heterogeneity in GBMs. First is “the clonal evolution model,” wherein cumulative epigenetic changes in normal cells lead to the genesis of cancer cells, which proliferate and acquire their tumorigenic potential. The second is “the cancer stem cell model,” which suggests that only a portion of cancer cells possess infinite self‐renewal potential and can start and maintain tumor development [80]. Even when there is no clear definition for the origin of the tumor, tumor‐initiating cells (TICs), a subset of highly tumorigenic glioblastoma stem cells (GBSCs), are highly resistant to conventional therapy because TAMs (30%–40%) and tumor‐infiltrating lymphocytes, contributing to the intratumoral vascular density by connecting the neoplastic cells and provide endothelial markers for immunity resistance, such as CD31, CD41, and CD99 [80, 81, 82]. The complex structure of the tumor cell niche can be studied via the connections between tunneling nanotubes (TNTs) established for proliferation and long‐distance communication. TNTs are long cytoplasmic F‐actin extensions of astrocytes and oligodendroglioma cells that may be open‐ended or connected by connexins 43 (Cx43). These extensions invade normal tissue cells and mediate the repopulation of the tumor after radiotherapy through the transfer of cellular material from GBSCs to the targeted cells. The exchange of the altered mitochondrial DNA (mt‐DNA) is particularly relevant since it affects and modifies metabolism and restores tumor adaptation and resistance, providing the tumorigenic phenotype to sensitive‐to‐treatment tumor cells [14]. Interestingly, intratumor spatial heterogeneity can be measured by targeting Bruton's tyrosine kinase (BTK), attributed to GBM core cells. BTK has a pivotal role in the maturing process of B cells. BTK profiling is based on RNA‐sequencing of four transcriptional factors in its pathway: NFATC3, NF‐κB2, BCL6, and NF‐κB1, and distinguishes edge from core‐cell populations. BTK silencing might improve chemotherapy results by promoting core‐cell apoptosis (Figure 2) [83]. One of the challenges of treating glioblastoma is its heterogeneity, affecting genetic expression, modulation of metabolic pathways, and immune system evasion. Cell‐to‐cell communication, extracellular vesicles (EVs), and TNTs mediate the transfer of molecular information between radiotherapy‐resistant and ‐sensitive tumor cells, propagating the tumorigenic phenotype from the tissue's core cells to the marginal zones. Therefore, these changes expand throughout healthy tissues around the tumor.
The conventional classification of LGGs, based on the IDH1/2 mutational status, leads to the metabolic characterization of differentiated astrocytes, one of the cell types that could possibly originate from GICs, as previously described. The traditionally defined IDH‐mutated astrocytoma represents the best example of altered metabolism within the tumoral heterogeneity of gliomas [84]. The brain is highly dependent on glucose intake to function correctly. Glioma cells adapt their metabolism according to glucose availability, which gives them extra resistance to hypoxia or altered redox situations. Selective pressure on tumors makes them overexpress glucose transporters, such as GLUT1/3, on their plasmatic membranes, regulated by the hypoxia factor HIF‐2 α. Even when glucose levels are low, HIF‐1 α guarantees the upregulation of hexokinase 2 (HK2), increasing the glycolytic pathway. Furthermore, many gliomas are characterized by the loss of PTEN function, which causes the constitutive activation of Akt1 and the stabilization of PFKP [85, 86]. MYC is a proto‐oncogene that promotes a bidirectional flow in the mitochondrial transport of lactate‐pyruvate. Deletion of MPC1 and the accumulation of LDH‐A leads to the transformation of pyruvate into lactate, which enhances the Warburg effect [87]. The next modifying step comes with the modulation of the Krebs cycle by extracting C atoms from it (cataplerosis) or introducing C atoms (anaplerosis) to it. These reversible reactions play a crucial role in the de novo biosynthesis of fatty acids, amino acids, and nucleotides. Intermediate metabolites, such as citrate and α‐KG, escape oxidation and serve as precursors for fatty‐acid biosynthesis and aspartate/glutamate synthesis. Aspartate initiates nucleotide biosynthesis, while glutamate provides the C‐skeleton of non‐essential amino acids. Hence, α‐KG DNA repair and demethylating activities become highly inhibited by the overproduction of 2‐HG (sensitizing IDH‐mutant cells to PARP‐1 inhibition and NAD+ deficiency), which also affects the transamination of this compound into keto acids and glutamate [88]. The marginal extensions of astrocyte‐like glioma cells contain high levels of cytosolic citrate, especially in the pseudopodia, where limited access to glucose leads to the uptake of acetate and oxidation, mediated by the ACSS2 enzyme [89]. Lipid metabolism is also altered in GBM. The marked metabolic heterogeneity of GBMs allows the use of this altered lipid metabolism to mark GBM stem and non‐stem cells in separate tumor niches [88]. GBMs can use ketone bodies and fatty acids to maintain growth, thus allowing their progression during ketogenic diet therapy [90]. Two essential enzymes mediate the biosynthesis of fatty acids, acetyl‐CoA carboxylase (ACC) and fatty‐acid synthase (FASN). FASN can be used as a biomarker since it is enriched in GBM‐derived EVs [91]. ACC and FASN are regulated by SREBP‐1, which responds to the EGFR‐PI3K‐Akt1 signaling pathway. By increasing the EGFR signaling, SREBP‐1 favors the tumor evolution of GBSCs into a proliferative status by synthesizing long‐chain ω3/6 polyunsaturated fatty acids. Meanwhile, the marginal and hypoxic regions store fatty acids via FABP3/7, which binds to polyunsaturated fatty acids (PUFAs) in a particular structure, defined as pseudopalisades, a hallmark of GBM [92, 93]. A super‐enhancer in GBM and GSCs promotes PUFA synthesis. These PUFA maintain EGFR signaling and membrane organization in GSCs. This observation suggests that dual targeting of EGFR and PUFA metabolism could be a novel potential therapeutic approach for glioblastoma management [94]. Regarding nitrogen metabolism, glioblastoma cells show both altered expression and activity of the amino acid transporter SLC7A11 [95], key enzymes involved in glutamine metabolism (glutamine synthetase and glutaminase) [96, 97] and cysteine metabolism [98]. The resulting balance of nitrogen metabolism gives rise to cataplerosis (a decreased availability of carbon atoms to enter the Krebs cycle oxidative pathway) [84]. Glutamine dependence exhibited by some tumor cells has motivated the development of therapeutic approaches based on the metabolism of this amino acid, and GBM is no exception. A phase I clinical trial combines a glutaminase inhibitor, Telaglenastat (CB‐839), with radiotherapy and TMZ chemotherapy in IDH1‐mutant astrocytomas. Telaglenastat may stop tumor growth by blocking the enzymes needed for this process (NCT03528642). Altered metabolism is a consequence of tumor heterogeneity and is favored by the complexity of the tumor niche, which is highly vascularized, with infiltrating M2 lymphocytes and TAMs, and a heterogeneous cell population. This characterization of gliomas is exemplified by the study of the mutation in IDH1/2. Knowledge relative to the altered metabolism shown by glioma cells is important to determine how these metabolic changes can affect the development of the tumor and to find new therapeutic targets. Several key intermediates and enzymes from the four main metabolic pathways previously described are discussed and suggested for glioblastoma targeting and treatment. Targeting PTEN could reduce the glucose intake in glioblastoma cells, while targeting PUFA and EGFR might affect the storage of lipids. Finally, the design of inhibitors in the synthesis mechanism of 2‐HG might recover the DNA‐repairing system.
Aggressive invasion potential is a hallmark feature of all subtypes of GBM and entails a struggle for its treatment. GBM invasion mechanisms are well understood in vitro, but this knowledge has yet to be transferred to new treatments in healthcare [99]. Tumor cell‐to‐cell crosstalk within the TME via EVs is involved in migratory phenotypes. EVs generated by mesenchymal subtype cells can affect their environment and contribute to the tumor invasion potential [100]. Annexin A2 (ANXA2) is one of the most abundant proteins in glioma EVs [101]. It is an important mediator in the plasminogen activator system, which mediates the conversion of plasminogen to plasmin and is essential for activating metalloproteinases involved in extracellular matrix degradation. However, the role of the transport of ANXA2 through EVs remains unknown [102]. ANXA2 regulates the molecular phenotype and aggressiveness of GBM via the ANXA2‐STAT3‐OSMR axis, which promotes mesenchymal transition. Consequently, ANXA2 and the ANXA2‐STAT‐OSMR axis could be attractive targets to manage GBM cells’ aggressiveness and migration [103]. Several genes are involved in GBM cell proliferation and invasion. B4GALT3 expression increases in GBM samples, especially in the proneural subtype, and this high expression predicts poor survival for patients with glioma. B4GALT3 depletion reduces cell viability and invasion of U251 glioblastoma cells, presumably due to the reduced expression of β ‐catenin, vimentin, and matrix metalloproteinase‐2, along with an increased expression of E‐cadherin [92]. GBP2 expression is also elevated in GBM samples, particularly in mesenchymal GBM, and this overexpression promotes cell migration and invasion in vitro. Fibronectin (FN1) expression and other genes are induced by GBP2 overexpression in U87 and U251 glioblastoma cell lines. FN1 is an extracellular glycoprotein involved in cell migration, and its depletion avoids GBP2‐induced invasiveness in the studied cell lines. STAT3, which contributes to the maintenance of GBM's mesenchymal subtype, is also involved in GBP2‐promoted FN1 expression [104]. PHIP is another gene involved in GBM motility through its regulatory activity on the focal adhesion complex. Besides, it also promotes cell invasion in melanoma, which shares its neuroectodermal origin with GBM. PHIP physically interacts with VCL, which is located at the force transductor domain of focal adhesions. PHIP downregulation significantly suppresses the migratory potential of U251 cells, an expected effect considering the role of focal adhesions in cell migration. This gene expression has also been suggested to be a biomarker of glioma progression [105]. Another target implicated in regulating GBM invasion and proliferation is ephrinB2, which tends to have a lower methylation status and, consequently, a higher expression in GBM compared with other gliomas. Paradoxically, this gene can act as an oncogene and a tumor‐suppressor gene. EphrinB2 overexpression increases the activation of Eph4 and reduces tumor growth but enhances invasion, while its knockdown has an anti‐invasive but proliferative effect. EphrinB2 knockdown followed by administration of ephrinB2‐Fc fusion protein results in tumor growth suppression along with an anti‐invasive response in U87 ephrinB2 tumor‐bearing mice [106]. Glioblastoma cells are forced to overexpress the neuronal glucose transporters GLUT1/3, as described in section 6.1. Libby et al. [107] observed that the overexpression of GLUT3 promotes GBM invasion in vitro. This invasive phenotype is independent of glycolytic metabolism, as the overexpression of GLUT3 did not have notable effects on glycolytic metabolic flux, which could be associated with the invasive phenotype. Interestingly, the substitution of GLUT3 C‐terminus with GLUT1 eliminated the pro‐invasive phenotype of GBM cells, while on the inverse, the substitution of GLUT1 C‐terminus with that of GLUT3 increased invasive potential. Thus, the GLUT3 C‐terminus could be a valuable target for inhibiting the invasion potential of GBM and other overexpressing GLUT3 cancers, such as breast, lung, liver, colon, head, and neck cancers.
Conventional GBM treatment comprises surgical intervention, which considers the age and medical condition of the patient, followed by radiotherapy and chemotherapy plus adjuvant TMZ. After surgery, the most important postoperative predictor associated with OS is the extent of resection. Recent findings propose hypofractionated radiation for older patients, administered daily in case of focal radiotherapy, and preventing adverse effects as much as possible [108, 109]. Traditional TMZ therapy has two significant issues: high concentrations are toxic to hematopoietic cells, and its administration engenders drug resistance in patients with newly diagnosed or recurrent GBM. Bevacizumab, a humanized anti‐VEGF monoclonal antibody and the first antiangiogenic‐approved therapy for colon cancer, is another drug whose application for GBM treatment remains uncertain based on the OS outcome [108, 109]. However, bevacizumab therapy for recurrent GBM has been approved, and it attempts to improve prognosis and survival [69, 110]. In fact, Meng et al. [111] observed that bevacizumab targets VEGF‐A (a promoter of angiogenesis and vascular permeability) and prevents edemas but does not affect the survival rate of glioblastoma patients [112]. Immunotherapy is gaining relevance, and future clinical trials may orient toward a more personalized diagnosis and treatment. This rising tendency honors the privileged immunoresistance present in CNS cancers. Early studies focused on the design of inhibitors (anti‐PD‐1, anti‐PD‐L1, and anti‐CTLA‐4), among which the combination of anti‐CTLA‐4 and anti‐PD‐1 was the only promising and effective therapy, showing a long‐term cure rate of 75% in GBM. Unfortunately, there are several immune‐resistant mechanisms in GBM, and the effects propagate systematically. The fast expansion and associated infiltrating immune cells quickly eradicate antibody monotherapies by disrupting the antigen flow and immune cell traffic toward the tumor niche. The latest immunotherapy strategies utilize vaccines to enhance T‐cell response [112, 113]. An ongoing clinical trial is evaluating immunotherapy in regard to the OS and progression‐free survival (PFS) of patients with GBM following intervention with durvalumab, an approved IgG1 for treating metastatic urothelial carcinoma (NCT02336165). The following sections describe the last approaches for GBM treatment [114].
Microglia are the principal antigen‐presenting cells in the CNS. GBSCs escape the immune system by increasing STAT3 expression, which translates into the upregulation of Wnt and TGF‐ β pathways (NCT01904123), resulting in the secretion of immunosuppressive factors, such as TGF β‐2 and interleukins IL‐10. This cell‐to‐cell mediated release is regulated by activating the cytotoxic T‐lymphocyte‐associated protein 4 (CTLA‐4) in regulatory T cells (Tregs). Additional help comes from TAMs that show high levels of programmed cell death‐ligand 1 (PD‐L1). PD‐L1 binds to receptors on the surface of T cells, restricting their activity [69, 113]. Based on this previously exposed mechanism, the most cited approach in immunotherapy involves blocking TAMs by creating suitable antibodies against CTLA‐4 and PD‐L1 [35]. The double blockade of CTLA‐4 and PD‐L1 is another suitable checkpoint for immunotherapy trials (NCT03233152, NCT04145115, NCT04003649). PD‐L1 was targeted by a phase I clinical trial that evaluated the effects of nivolumab plus ipilimumab for recurrent GBM. Its purpose was to compare this new treatment's adverse effects and efficacy with bevacizumab monotherapy (NCT02017717) [69]. Another interesting approach is combined therapy comprising PD‐1/PD‐L1 inhibitors plus radiotherapy or antibodies targeting CTLA‐3, TIM‐3, LAG‐3, IDO, or OX‐40 (NCT02658981, NCT04003649) [108, 113]. Antibodies that block the PD‐1/PD‐L1 interaction or the T‐cell response are only effective in some patients. A series of ongoing trials targeting PD‐1 aim to restore immunogenicity, especially CD8 and CD4 T‐cell responses (NCT02287428, NCT04201873, NCT03422094, NCT03899857). The most recent interventions are the development of highly personalized cytomegalovirus (CMV) therapy and neoantigen vaccines. CMV protein pp65 causes a robust CD8+ T‐cell response that benefits survival, especially for CMV transferred into dendritic cells (NCT03688178). This last intervention serves as an adjuvant for vaccination, the same way CpG oligonucleotides and granulocyte‐macrophage colony‐stimulating factor (GM‐CSF) do. Neoantigen vaccines show potent antitumoral activity by inducing both CD8+ and CD4+ T cell responses. However, before these strategies reach clinical practice, further research and optimization are required [112]. To sum up, most immunotherapy therapies in development to treat GBM aim to avoid the inhibition of T cells, mediated by Tregs and TAMs, using antibodies against CTLA‐4 and PD‐L1. However, new strategies that could provide personalized treatment with a robust immune response, such as CMV therapies and neoantigen vaccines, constitute a promising research area.
As previously described, the design of tumor‐specific antigens is a promising application for glioblastoma immunotherapy since they are exclusively expressed on tumor cells. The leading two platforms selected for the design of vaccines are neoantigens or tumor‐specific antigens and plasmid DNA, while the vehicles used are dendritic cells and heat shock proteins. Neoantigens are proteins that originate from mutations within tumor cells and differ from cell to cell. This feature makes them excellent targets to address tumor cells selectively. The development of these vaccines requires sequencing data from the whole exome and the RNA of both healthy and tumor cells from the patient. The sequencing aims to find targetable proteins that produce a T‐cell response in the organism by the specific recognition and binding to the MHC [115]. There is an ongoing clinical trial (NCT02287428) testing the safety of a neoantigen vaccine against glioblastoma in combination with radiation therapy and pembrolizumab or TMZ. This vaccine works well for newly diagnosed patients with unmethylated MGMT promoters. The patients might be receiving immunoadjuvant poly‐ICLC and radiotherapy. These personalized peptide vaccines may be effective for treatment with limited toxicity, modulating the clinical outcomes of glioblastoma patients [116]. The development of nucleic acid‐based vaccines is a promising strategy being tested for treating GBM. This strategy involves developing DNA plasmids encoding tumor‐specific antigens and cytokines, which promote recognition and the CD8+ T‐cell response [115]. DNA‐based vaccines are most beneficial among nucleic acid‐based vaccines since they enter the nucleus and allow the presentation of antigens to the MHC. There is an active phase I/II clinical trial (NCT03491683) evaluating the safety, immunogenicity, and preliminary efficacy of two DNA‐based vaccines: INO‐5401, which is a combination of three DNA plasmids targeting Wilms tumor gene‐1 (WT1) antigen, prostate‐specific membrane antigen (PSMA) and human telomerase reverse transcriptase (hTERT) genes; and INO‐9012, a DNA plasmid for the expression of human interleukin‐12 (IL‐12). Both treatments are administered in combination with cemiplimab, radiation therapy, and TMZ. This is the first study in human GBM to combine DNA plasmids with PD‐1 blockade. The primary outcomes of this trial are related to measuring the percentage of participants with adverse events, which resulted in the usual spectrum for PD‐1 inhibitory agents and the vaccine itself not having significant adverse events either. Secondarily, the preliminary 12‐month overall survival was 84.4%, and the activation of INO‐5401‐specific CD8+ T‐cells was successful. The current outcomes are promising even though there are no conclusions drawn yet [117]. Another phase I clinical trial (NCT04573140) is recruiting patients for the evaluation of the manufacturing feasibility, safety, and maximum‐tolerated dose of an RNA‐lipid particle (RNA‐LP) vaccine in adult glioblastoma patients [115]. The design of personalized peptide or DNA‐based vaccines is currently being tested in clinical trials. Their objective is to improve the outcomes by answering in a more precise mechanism within the patient. This is possible thanks to the immune response behind the treatment. In most cases, vaccines might be a safer and more effective way to increase overall survival with low adverse events or cytotoxicity. Table 4 summarizes the ongoing and finished clinical trials under the label of “immunotherapies.”
An approximation to molecular‐targeted therapy comes from the novel strategies for identifying biomarkers in the RTK/PI3K/Akt/mTOR signaling pathway. One way to forestall the recurrence of GBM is through developing RTK‐targeting drugs, such as imatinib, approved by the Food and Drug Administration (FDA) for chronic myeloid leukemia. Imatinib is directly related to PDGFR inhibitors. The latest phase II studies evaluating the effectiveness of combined imatinib and hydroxyurea showed little improvement in patients with recurrent GBMs. Nevertheless, hydroxyurea is being administered because it sensitizes GBM to TMZ treatment [108]. Recent trials on dasatinib monotherapy and dasatinib plus lomustine therapy showed no considerable effectiveness for recurrent GBM [79]. Besides, enasidenib is used to treat acute myeloid leukemia and is being evaluated in IDH2‐mutated tumors, such as gliomas (NCT02273739). Other metabolic‐related clinical trials are being conducted, designing IDH‐ and PARP‐specific inhibitors (NCT03224104, NCT04740190, NCT03914742). The next possible molecule that intercepts the pathway is EGFR. Gefitinib and erlotinib are efficient first‐generation EGFR reversible inhibitors that impede the binding of ATP to the EGFR tyrosine kinase domain receptors. In fact, a clinical trial has assessed the administration of gefitinib and radiotherapy and its effectiveness in inhibiting cell growth (NCT00052208). Afatinib, a second‐generation EGFR inhibitor, binds irreversibly to Cys 773 EGFR residues and Cys 805 HER2 residues. However, these EGFR inhibitors have only shown some efficacy in in vitro assays since they cannot efficiently cross the blood‐brain barrier (BBB) [79]. A phase II clinical trial demonstrated that direct treatment with rindopepimut was ineffective against EGFRvIII expressing GBM [69]. In addition, sorafenib therapy against tumor growth and proliferation is being tested for GBM. This small molecule can bind to multiple tyrosine kinases (RAF, VEGFR, PDGFR, and c‐KIT) and inhibit them. Unfortunately, a phase III clinical trial on sorafenib also failed [110]. The expectancy of these two trials improved the overall survival and prognosis for patients with severe or metastatic tumors. One ongoing trial is testing TAS2940 safety in candidates who are not approved for currently available therapies, targeting HER‐2 and VEGFR in solid tumor cancers (NCT04982926). This trial aims to predict lower adverse events and improved overall survival. Another important drug is the main mTORC1 inhibitor, rapamycin, which affects kinase conformation. Its monotherapy remains insufficient for recurrent GBM but shows more significant activity in combination with other analogs [79]. Clinical approaches to the PI3K/Akt/mTOR axis illustrate how targeted molecular therapy can improve prognosis in patients with aggressive tumors, such as glioblastoma. Multiple tyrosine kinases can be targeted and inhibited by a biological drug, for example, RTK, EGFR, RAF, or PDGFR. Despite the positive outcomes for imatinib, enasidenib, and sorafenib in other tumors, they remain ineffective for glioblastoma treatment. This recalls for an update on the drugs administered and their respective targets.
There are many possible ways of intercepting or impairing tumor growth in GBM. A promising study assessed the efficacy of G‐protein‐coupled receptor 17 (GPR17) agonist GA‐T0, which crosses the BBB and promotes GBM cell death in murine models through modulation of the MAPK/ERK, STAT, PI3K/Akt, and NF‐κB pathways. This work showed that GPR17 expression in GBM was related to improved overall survival and could be used as a predictive biomarker [118]. Biological drugs or antibody therapies are potent treatments for many types of tumors. However, they have provided little efficacy in GBM so far. Cetuximab (anti‐EGFR), panitumumab, and nimotuzumab are examples of three antibodies designed to target EGFRvIII. The first one, cetuximab, has a five‐month limited effect (intravenous treatment, phase II study) in reducing EGFR mutations, tumor survival, and proliferation. Some small‐molecule inhibitors are non‐specific compounds that target different biomarkers of tumor‐related signaling pathways, for example, lenvatinib, dovitinib, and brivanib [79]. Tyrosine kinases are other antiangiogenic targets in the VEGF signaling pathway for patients with newly diagnosed GBM, characterized by the higher levels of HIF‐1α and SDF‐1, which are responsible for microvascularization. The SDF‐1 pathway induces the recruitment of endothelial progenitor cells from the bone marrow to the tumor niche, where SDF‐1 interacts with CXCR4 receptors on the surface of endothelial matured cells. A way to prevent vasculogenesis would be the design of a new biomolecule that targets CXCR4 or SDF‐1. For example, CXCR4 antagonist AMD3100 treatment could prevent tumor vasculogenesis and growth [110]. Anlotinib is a multi‐target tyrosine kinase inhibitor that blocks the migration and proliferation of endothelial cells and can inhibit tumor angiogenesis and cell growth by targeting specific growth factor receptors, such as VEGFR and PDGFR. Currently, there are clinical trials testing the viability of this treatment against GBM (NCT04725214, NCT05033587, NCT04547855, NCT04959500). Depatuxizumab mafodotin is an antibody‐drug conjugate against activated EGFR, which has not shown positive results in a recent trial (NCT02343406). However, this treatment did show interesting results that suggest that a subgroup of patients could benefit from this therapy. Specific molecular predictors of treatment efficacy, like EGFR SNVs, could help determine which patients would benefit from this conjugate. Pazopanib is an FDA‐approved oral drug for metastatic‐advanced kidney cancer and angiosarcoma. Pazopanib has proven to be an effective therapy when combined with prior exposure to bevacizumab (NCT01931098). Another trial (NCT04704154) is investigating the safety and tumoral response of nivolumab plus regorafenib combo for solid tumors treatment, GBM among them. Both monoclonal antibodies and small‐molecule inhibitors are under research to develop molecular‐targeted therapies focused on angiogenesis to treat glioblastoma. EGFRvIII, CXCR4, SDF‐1 and VEGFR are among the analyzed targets involved in angiogenesis. Nonetheless, even though ongoing clinical trials are testing these strategies, there have not been conclusive results yet.
Alternatively, small interfering RNA (siRNA) nanoparticles designed for gene silencing might be more effective than antibodies. A recent study focused on the comparison of antibody therapy vs. siRNA treatment. The effectiveness of siRNA treatment against cancer invasion and progression was approximately 40%–65%, while cetuximab or trastuzumab (anti‐HER2) therapies showed no reduction in invasiveness. The main reason behind siRNA treatment's efficacy relies on the nature of the materials used for the nanoparticles’ design, such as liposomal carriers, which can transfer siRNA across the BBB [110]. Nanodelivery eliminates the main barrier against pharmacokinetics, the crossing of the BBB, being a suitable mechanism for effective and individualized therapy. This new field is further explored in section 9.4.
GICs appear to be resistant to conventional therapy and responsible for the reappearance of the tumor after surgery resection [109]. Consequently, GBM resistance depends on tumoral heterogeneity and the phenotype of GBSCs. The disruption of key enzymes in the pyrimidine synthesis, such as CAD or DHODH, intercepts GBM resistance. Leflunomide and teriflunomide are two effective inhibitors of the enzyme DHODH, which has a crucial role in stem‐like phenotype maintenance in GBSCs. Despite the significant prognosis of these treatments, GBSCs can still reprogram their pyrimidine metabolism and develop resistance [120].
A phase IV clinical trial (NCT03975829) is assessing the long‐term effects in pediatric patients diagnosed with astrocytoma, oligodendroglioma, neurofibromatosis type 1, and other gliomas. The trial consists of the administration of dabrafenib plus trametinib. The goal of these two inhibitors is to block MEK1/2 (trametinib) and BRAF kinase (dabrafenib), two proteins related to the activation of the RAS/RAF/MEK/ERK signaling pathway. This pathway mediates cell growth and is usually upregulated in tumors, meaning that its inhibition could prevent tumor expansion. This combo has improved the overall survival of metastatic melanoma [121]. The same primary outcomes are expected for glioblastoma treatment. Children diagnosed with high‐grade gliomas may be treated with depatuxizumab mafodotin, designed to be combined with TMZ or lomustine (NCT02343406), or alternatively, repotrectinib (NCT04094610). The primary outcomes of both studies are low toxicity and enhanced overall survival [122]. Moreover, a tentative FDA approval calls for palbociclib isethionate (NCT02255461), which has been tested for young patients with recurrent, progressive, or refractory CNS tumors. This oral drug is a CDK4/6 inhibitor, thereby promoting cell cycle arrest in the G1 phase and decreasing tumor proliferation. Its safety has been tested in pediatric patients with low‐/high‐grade glioma, showing positive results. This trial aims to study the side effects of palbociclib isethionate to determine the maximum tolerated dose and the plasma pharmacokinetics. There is a favorable background for this clinical intervention regarding the safety of palbociclib administered to children and adolescents for other brain tumors [123]. To sum up, these therapeutic trials being tested on children and adolescents aim to design a safer treatment with positive outcomes and less toxicity for the administered drugs. Table 5 depicts information on the ongoing and finished clinical trials under the label of “molecular‐targeted therapies”:
Despite the recent findings on the molecular biology of gliomas and the testing of several biomarkers, their predictive value and efficacy need to be validated and approved. Figure 3 summarizes the main expected trends in glioblastoma therapy. The conventional treatment for glioblastoma remains insufficient to cover the whole spectrum of cases. Due to GBM heterogeneity, patients should be treated on a case‐by‐case basis. This way, patients can obtain a more personalized diagnosis, guiding neurologists into the best ways to diminish tumoral progression and provide the proper treatment. The following section describes the present trends being explored in the latest ongoing trials and could open the door to promising therapies in the future.
Immunotherapy and molecular‐targeted therapies are two major trends in the treatment of glioblastoma. Immunotherapy has not provided positive results thus far due to the immunoresistance of CNS tumors. Even so, the development of new approaches, such as tumor‐specific neoantigen‐based and nucleic acid‐based vaccines, could be a step toward personalized GBM treatment. In fact, an increase in the number of clinical trials based on nucleic acid‐based vaccines may be expected thanks to the remarkable advances in this strategy achieved during the last years. Implementing multiple‐antigen targeting could improve vaccine therapy by increasing OS and PFS, two benchmarks currently failing in recent trials. Most commonly found molecular biomarkers might be targeted simultaneously: EGFR, IL‐10, CD27, SOX40 and WT1, combined with tumor‐specific antigen vaccines, thus providing a more individualized treatment [112, 115]. The challenges of a multifaceted approach remain overwhelming, but the potential benefits that may result from them are substantial.
Circulating tumor cells, EVs, and circulating nucleic acids and proteins are candidate biomarkers for glioma diagnosis, although none of them have been approved for clinical practice thus far [31]. Circulating tumor nucleic acids are a source of comprehensive information relative to glioblastoma cells' genome, while circulating miRNAs can be used to find potential targets related to tumorigenesis and proliferation and to elucidate the grade and response to glioma chemotherapy [124, 125]. As previously mentioned, liquid biopsy can be a helpful tool for diagnosing and developing personalized treatments. Besides being the brain immersed in CSF, this body fluid could be used as a source of tumor metabolites and other biomarkers [126]. This routinely applied clinical intervention could amplify the chances of identifying specific targets with minimal sample invasion. However, further studies with larger cohorts and the standardization of detection approaches are needed to enhance the potential of liquid biopsy applied to GBM [127].
Mathematical models may provide useful information about GBM cell invasion dynamics of prognostic value, which could facilitate the prediction of tumor progression [128]. MRIs contain structural and functional information for diagnosis and can be analyzed by machine‐learning algorithms. Artificial intelligence applied to MRI data could be relevant in the future for diagnosing and prognosticating patients with GBM. However, as gliomas are relatively rare, it is challenging to generate a large amount of clinical data necessary to develop these tools [129]. Nonetheless, certain models and algorithms have already reached diagnostic accuracies higher than 80% [130]. These approaches could also become an excellent non‐invasive method for identifying GBM subtypes [131].
Many pharmacological therapies are inefficient due to the hydrophilic nature of the compound and its low capacity to go through the BBB. Nanodelivery is designed to optimize drug pharmacokinetics across the BBB. The design of nanoparticles (NPs) includes proper factors to assimilate the therapeutic agent and enhance its outcome. Hence, a wide range of nanodelivery systems is being investigated in interventional trials. CMV therapy might be tested for nanodelivery, although there are uncertain values for toxicity and immunogenicity that should be solved [109, 132]. A current trial of siRNA NPs describes an example of liposomal carriers [119]. Another interesting application of nanodelivery comes with autophagy modulators found in natural products, such as resveratrol, curcumin, capsaicin, and others. The limited availability of these phenolic compounds relies on their low solubility. This can be overcome by administering NPs that form complexes and interactions to enhance their solubility and potential efficacy [133]. The promising outcomes from nanodelivery approaches may enhance specific targeting, the reversion of drug resistance, a reduction in the adverse effects, and a more prolonged circulation time [134]. Nevertheless, nanodelivery is still a pilot‐stage field that should be further researched and documented before approval. The lack of treatments based on nanodelivery relies on the extended testing they require, the lack of standardization of the nanotoxicological assays, and the manufacturing costs of the techniques [135].
This review reflects the efforts made to elucidate the molecular biology and genetics underlying the development and complexity of GBM. Although a large body of knowledge related to the molecular basis of this disease has been attained, this knowledge has not resulted in effective remedies for patients who suffer from this unstoppable disease. To achieve effective therapies against GBM, several of its hallmarks must be overcome, such as metabolic heterogeneity, tumor invasion potential, drug resistance, poor pharmacokinetics, and immunoresistance. As long as these issues remain resolved, efficacious treatment for GBM cannot be guaranteed. However, the development of different approaches and strategies to address this disease, such as the future trends discussed herein, guarantees promising advances in the understanding and treatment of this CNS tumor.
Conceptualization, EV, IP and MAM; bibliographic investigation, EV, and IP; writing—original draft preparation, EV, and IP; writing—review and editing, MAM; drawing of figures, EV, and IP; supervision, MAM; funding acquisition, MAM. All authors have read and agreed to the published version of the manuscript.
The authors declare no conflict of interest.
Not applicable.
Not applicable.
The experimental work carried out by our group and this review are supported by grants PID2019‐105010RB‐I00 (Spanish Ministry of Science, Innovation and Universities), and UMA18‐FEDERJA‐220 (Andalusian Government and FEDER) and funds from group BIO 267 (Andalusian Government). The “CIBER de Enfermedades Raras” (Spanish Biomedical Research Network Center for Rare Diseases) is an initiative from the ISCIII (Spain). The funders had no role in the study design, data collection and analysis, publication decision, or manuscript preparation. |
PMC9648392 | 36209503 | Jing Han,Menglin Nie,Cong Chen,Xiaojing Cheng,Ting Guo,Longtao Huangfu,Xiaomei Li,Hong Du,Xiaofang Xing,Jiafu Ji | SDCBP‐AS1 destabilizes β‐catenin by regulating ubiquitination and SUMOylation of hnRNP K to suppress gastric tumorigenicity and metastasis | 09-10-2022 | SDCBP2‐AS1,gastric cancer,hnRNP K,β‐catenin,post‐transcriptional modifications,tumorigenesis | Abstract Background Gastric cancer (GC) is among the most malignant tumors, yet the pathogenesis is not fully understood, especially the lack of detailed information about the mechanisms underlying long non‐coding RNA (lncRNA)‐mediated post‐translational modifications. Here, the molecular mechanisms and clinical significance of the novel lncRNA syndecan‐binding protein 2‐antisense RNA 1 (SDCBP2‐AS1) in the tumorigenesis and progression of GC were investigated. Methods The expression levels of SDCBP2‐AS1 in 132 pairs of GC and adjacent normal tissues were compared, and the biological functions were assessed in vitro and in vivo. RNA pull‐down and immunoprecipitation assays were conducted to clarify the interactions of SDCBP2‐AS1 and heterogeneous nuclear ribonucleoprotein (hnRNP) K. RNA‐sequencing, immunoprecipitation, immunofluorescence, and luciferase analyses were performed to investigate the functions of SDCBP2‐AS1. Results SDCBP2‐AS1 was significantly downregulated in GC tissues and predictive of poor patient prognosis. Silencing of SDCBP2‐AS1 promoted the proliferation and migration of GC cells both in vitro and in vivo. Mechanically, SDCBP2‐AS1 physically bound to hnRNP K to repress SUMOylation of hnRNP K and facilitated ubiquitination of hnRNP K and β‐catenin, thereby promoting the degradation of β‐catenin in the cytoplasm. Silencing of SDCBP2‐AS1 caused SUMOylation of hnRNP K and stabilized β‐catenin activity, which altered transcription of downstream genes, resulting in tumorigenesis and metastasis of GC. Moreover, the knockdown of hnRNP K partially abrogated the effects of SDCBP2‐AS1. Conclusions SDCBP2‐AS1 interacts with hnRNP K to suppress tumorigenesis and metastasis of GC and regulates post‐transcriptional modifications of hnRNP K to destabilize β‐catenin. These findings suggest SDCBP2‐AS1 as a potential target for the treatment of GC. | SDCBP‐AS1 destabilizes β‐catenin by regulating ubiquitination and SUMOylation of hnRNP K to suppress gastric tumorigenicity and metastasis
Gastric cancer (GC) is among the most malignant tumors, yet the pathogenesis is not fully understood, especially the lack of detailed information about the mechanisms underlying long non‐coding RNA (lncRNA)‐mediated post‐translational modifications. Here, the molecular mechanisms and clinical significance of the novel lncRNA syndecan‐binding protein 2‐antisense RNA 1 (SDCBP2‐AS1) in the tumorigenesis and progression of GC were investigated.
The expression levels of SDCBP2‐AS1 in 132 pairs of GC and adjacent normal tissues were compared, and the biological functions were assessed in vitro and in vivo. RNA pull‐down and immunoprecipitation assays were conducted to clarify the interactions of SDCBP2‐AS1 and heterogeneous nuclear ribonucleoprotein (hnRNP) K. RNA‐sequencing, immunoprecipitation, immunofluorescence, and luciferase analyses were performed to investigate the functions of SDCBP2‐AS1.
SDCBP2‐AS1 was significantly downregulated in GC tissues and predictive of poor patient prognosis. Silencing of SDCBP2‐AS1 promoted the proliferation and migration of GC cells both in vitro and in vivo. Mechanically, SDCBP2‐AS1 physically bound to hnRNP K to repress SUMOylation of hnRNP K and facilitated ubiquitination of hnRNP K and β‐catenin, thereby promoting the degradation of β‐catenin in the cytoplasm. Silencing of SDCBP2‐AS1 caused SUMOylation of hnRNP K and stabilized β‐catenin activity, which altered transcription of downstream genes, resulting in tumorigenesis and metastasis of GC. Moreover, the knockdown of hnRNP K partially abrogated the effects of SDCBP2‐AS1.
SDCBP2‐AS1 interacts with hnRNP K to suppress tumorigenesis and metastasis of GC and regulates post‐transcriptional modifications of hnRNP K to destabilize β‐catenin. These findings suggest SDCBP2‐AS1 as a potential target for the treatment of GC.
Abbreviations APC adenomatous polyposis coli β‐TrCP β‐transducin repeat‐containing protein CHX cycloheximide CI confidence interval CTNNB1 catenin beta 1 DFS disease‐free survival EPDR1 ependymin‐related 1 FISH fluorescence in situ hybridization GAPDH glyceraldehyde 3‐phosphate dehydrogenase GC gastric cancer TCGA Cancer Genome Atlas‐Stomach Adenocarcinoma GEPIA gene expression profiling interactive analysis GSEA gene set enrichment analysis GSK3β glycogen synthase kinase 3 beta H3 histone 3 HA hemagglutinin H&E hematoxylin and eosin hFAST human Fas‐activated serine/threonine kinase hnRNP K heterogeneous nuclear ribonucleoprotein K HOXA11‐AS homeobox A11 antisense RNA HR hazard ratio IF immunofluorescence IgG immunoglobulin G IHC immunohistochemical IP immunoprecipitation lncRNA long non‐coding RNA MMP7 matrix metallopeptidase 7 NT non‐target OC ovarian cancer OS overall survival PVT‐1 Pvt1 oncogene RACE rapid amplification of cDNA ends; RIP RNA Immunoprecipitation RT‐qPCR quantitative real‐time polymerase chain reaction SDCBP2 syndecan binding protein 2 SDCBP2‐AS1 SDCBP2 antisense RNA 1 SDS‐PAGE sodium dodecyl sulfate‐polyacrylamide gel electrophoresis SLCO4A1‐AS1 solute carrier organic anion transporter family member 4A1 antisense RNA 1 SNHG5 small nucleolar RNA host gene 5 SUMO‐1 small ubiquitin‐like modifier 1 UCA1 urothelial cancer associated 1
Gastric cancer (GC) is a major cause of cancer‐related death worldwide, although its incidence varies across different geographical regions [1, 2]. Despite standardized treatments have improved patients survival in the last decade, the 5‐year survival rate remains low [3]. Early detection can largely promote the 5‐year survival rate; however, due to the lack of effective diagnostic and therapeutic strategies, more than 80% GC patients in China had advanced diseases at the time of diagnosis, and the 5‐year survival rate is less than 20% [4, 5]. Hence, novel biomarkers and therapeutic targets are urgently needed for the diagnosis and prognosis prediction of GC to reduce mortality. Long non‐coding RNAs (lncRNAs) are important RNA transcripts longer than 200 nucleotides with limited or no protein‐coding capacity [6, 7]. lncRNAs are crucial to various biological processes, especially tumorigenesis and metastasis [8, 9, 10, 11]. However, the specific molecular pathogenic mechanisms of lncRNAs underlying the initiation and progression of GC remain unclear. Wnt/β‐catenin signaling controls several fundamental cell functions, such as proliferation, differentiation, migration, and stemness. It therefore plays a critical role in the gastrointestinal epithelial homeostasis. Even for GC, a highly heterogeneous disease with phenotypic diversity, the dysregulation of Wnt/β‐catenin signaling pathway still has been observed in approximately 50% of GC patients [12, 13]. β‐catenin, the prime effecter of canonical Wnt signaling, contributes to tumor progression by enhancing the proliferation and invasiveness of GC cells [14, 15]. Activation of the canonical Wnt/β‐catenin signaling pathway is dependent on controlling the accumulation of β‐catenin in the cytoplasm and its translocation into the nucleus to regulate the transcription of downstream genes. Regulation of cytoplasmic β‐catenin is controlled by targeting phosphorylated β‐catenin to the proteosomes for degradation via E3 ubiquitin ligase, resulting in low concentrations of cytoplasmic β‐catenin [16]. However, the relative importance of temporal activation of β‐catenin and its crosstalk with lncRNAs is not well understood. Our group has previously identified a collection of candidate lncRNAs, including syndecan‐binding protein 2‐antisense RNA 1 (SDCBP2‐AS1), involved in the GC progression [17]. In continuation, the present study aimed to investigate the molecular mechanism of SDCBP2‐AS1 in mediating the transcription, stability, and transactivation of β‐catenin in the development of GC and to explore potential prognostic implications.
A total of 132 GC tumor tissues and paired adjacent normal tissues were obtained from GC patients via radical resection conducted at the Peking University Cancer Hospital (Beijing, China) between January 2007 and December 2010. None of the patients received chemotherapy or radiotherapy before surgery, and all were followed up until March 2016. All subjects provided informed consent for the use of specimens in this study. The study protocol was approved by the Ethics Committee of Peking University Cancer Hospital (approval number: 2018KT07) and conducted in accordance with the ethical principles for medical research involving human subjects described in the Declaration of Helsinki. After resection, the tissues were snap‐frozen in liquid nitrogen and then stored at −80°C prior to RNA extraction. Detailed information of the patients is provided in Table 1. The tumor‐node‐metastasis (TNM) stage of GC was determined in accordance with the Cancer Staging Manual of the American Joint Committee on Cancer (8th edition). Overall survival (OS) was calculated from the date of surgery to the date of death due to any reason, while disease‐free survival (DFS) was defined as the duration from the date of surgery to the date of death or recurrence of any complications during follow‐up. Patients without an endpoint (progression or death) were censored at the date of the last follow‐up.
The coding potential of SDCBP2‐AS1 was analyzed using the online tools Open Reading Frame Finder (https://www.ncbi.nlm.nih.gov/orffinder/), Coding Potential Calculator 2 (http://cpc2.gao‐lab.org/), Coding‐Potential Assessment Tool (http://lilab.research.bcm.edu/cpat/), and LNCipedia (https://lncipedia.org/) [18, 19, 20, 21]. Gene Expression Profiling Interactive Analysis (GEPIA) web tool (http://gepia.cancer‐pku.cn/) and Asian Cancer Research Group databases (https://consortiapedia.fastercures.org/consortia/acrg/) were used to analyze the correlation between SDCBP2‐AS1 and SDCBP2 expression in GC samples. Kaplan‐Meier survival plot of DFS and OS of GC patients in the dataset GSE22377 (http://www.ncbi.nlm.nih.gov/geo/geo2r/?acc=GSE22377) was analyzed by the web tool Kaplan‐Meier Plotter (http://kmplot.com/analysis/index.php?p=service&cancer=gastric) [22].
Human embryonic kidney 293 (HEK293T) cells and human gastric adenocarcinoma hyperdiploid (AGS) cells were obtained from the American Type Culture Collection (Manassas, VA, USA). GC cells (BGC823, SGC7901, MGC803, HGC27, and N87) and normal gastric epithelial (GES‐1) cells were obtained from the Shanghai Cell Research Institute (Chinese Academy of Sciences, Shanghai, China). Well‐differentiated human gastric adenocarcinoma (MKN28) cells were obtained from the Japanese Collection of Research Bioresources Cell Bank (National Institutes of Biomedical Innovation, Health and Nutrition, Osaka, Japan). All cell lines, which were obtained between 2014 and 2015 and authenticated by short tandem repeat profiling, were cultured in Dulbecco's modified Eagle's medium (DMEM) (Gibco; Invitrogen Corporation, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (Gibco) and maintained at 37°C in 5% CO2 as previously described [23]. Cells were incubated in serum‐free medium and treated with cycloheximide (CHX) (Sigma‐Aldrich Pty. Ltd., Merck KGaA, Darmstadt, Germany) for 4 h or MG132 (Sigma‐Aldrich) for 8 h as indicated. Routine testing for Mycoplasma contamination was performed using polymerase chain reaction (PCR). Cells were grown for no more than 20 passages prior to experimentation.
Full‐length cDNA of human SDCBP2‐AS1 (2479 bp) was synthesized by Invitrogen Corporation and cloned into the expression vector pcDNA3.0. A series of SDCBP2‐AS1 deletion mutants were designed according to the secondary structure of SDCBP2‐AS1 predicted from the RNAfold Web Server (http://rna.tbi.univie.ac.at/). Truncations of SDCBP2‐AS1 were amplified with appropriate primer sets and subcloned into plasmid pcDNA3.0. Human hnRNP K cDNA (1395 bp) was amplified from GC tissues and subcloned into plasmid p3XFLAG‐CMV‐10 (Sigma‐Aldrich). Truncations of hnRNP K were amplified with appropriate primer sets and subcloned into plasmid p3XFLAG‐CMV‐10. The hnRNP K K422R mutant was created with the restriction enzyme DpnI and subcloned into plasmid p3XFLAG‐CMV‐10. Each mutation was verified against the whole hnRNP K cDNA sequence. Human small ubiquitin‐like modifier 1 (SUMO‐1) cDNA was amplified from GC tissues and subcloned into plasmid pCMV‐HA. HA‐catenin beta 1 (CTNNB1) was subcloned into vector GV366 (Shanghai Genechem Co., Ltd., Shanghai, China). The final construct was verified by sequencing. Stable SDCBP2‐AS1‐overexpressing SGC7901 cell line were generated and screened by administration of neomycin (Invitrogen Corporation). Small hairpin RNA (shRNA) against SDCBP2‐AS1 was provided by Invitrogen Corporation, and shRNA against hnRNP K was obtained from Shanghai Genechem Co., Ltd. Stable SDCBP2‐AS1‐knockdown BGC823 and MKN28 cells and hnRNP K‐knockdown SGC7901 cells were established with a lentiviral vector in accordance with the manufacturer's instructions. All primers and shRNA sequences are listed in Supplementary Table S1. GC cells were transfected with plasmid vectors or lentiviruses encoding shRNAs using DNA Miniprep Kits [Tiangen Biotech (Beijing) Co. Ltd., Beijing, China] and Lipofectamine 2000transfection reagent (Invitrogen Corporation) following the manufacturer's instructions and screened by administration of puromycin (Invitrogen Corporation). Plasmid p3XFLAG‐CMV‐10 or the empty vector pCMV‐HA was used as an overexpression control (Con), while non‐target (NT) shRNA was used as a knockdown control.
Total RNA was extracted from the tissue samples and cell lines using TRIzol reagent (Invitrogen Corporation) in accordance with the manufacturer's instructions, and reversely transcribed into complementary DNA using a reverse transcription system kit (Invitrogen Corporation) with the ABI PRISM 7500 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) and the SYBR Green method (amplification condition: 95°C for 10 min, followed by 40 cycles at 95°C for 15 s and 60°C for 1 min). For each sample, gene expression was normalized to glyceraldehyde 3‐phosphate dehydrogenase (GAPDH) or U6. The primer sequences used for RT‐qPCR are listed in Supplementary Table S1. Each RT‐qPCR reaction was performed in triplicate, and relative RNA expression was calculated using the 2−ΔΔCt method.
RACE was performed using the SMARTer 5′/3′ RACE kit (Takara Bio Inc., Shiga, Japan) in accordance with the manufacturer's recommendations. Briefly, first‐strand cDNA was synthesized with SMARTScribe™ Reverse Transcriptase (RT) and a modified oligo (dT) primer that contained an additional sequence as a primer binding site for downstream 3′ PCR reactions. The SMARTer II A Oligonucleotide, which contained several non‐template residues, was annealed to the 5′‐end as an additional template for SMARTScribe RT and as a primer binding site for downstream 5′ PCR reactions. The 5′ and 3′ RACE reactions were performed by PCR with the SDCBP2‐AS1‐specific primer pair SDC5′GSP1/SDC3′GSP1 and the universal primer UPM long (Supplementary Table S1).
GC cells (3000 cells/well) with stable knockdown of endogenous SDCBP2‐AS1 (BGC823 and MKN28) or stable ectopic expression of SDCBP2‐AS1 (SGC7901) were seeded into 96‐well plates, and cell proliferation was monitored with an IncuCyte® Live‐Cell Analysis System (Essen BioScience, Ann Arbor, MI, USA). For the colony formation assay, the established stable cell lines were seeded into 6‐well plates at 600 cells/well and incubated at 37°C in 5% CO2 for 14 days. Afterward, they were carefully washed twice with phosphate‐buffered saline, fixed with 75% ethyl alcohol for 15 min at 21°C, and stained with 0.1% crystal violet. Washing out the dye and drying, the plates were photographed. The ImageJ software (National Institutes of Health, Bethesda, MD, USA) was used to quantify the colonies containing ≥50 cells objectively.
Transwell chambers with and without a Matrigel matrix (Corning Life Science, Woburn, MA, USA) were used to examine cell migration and invasion, respectively. Cells (3 × 104) were suspended in 200 μL of serum‐free DMEM and seeded onto polycarbonate filters, which were either pre‐coated with 100 μL of Matrigel for the invasion assay or left uncoated for the migration assay. The lower chamber was loaded with 600 μL of DMEM containing 10% fetal bovine serum. The wound‐healing assay was performed to assess cell motility. Briefly, wounds were generated by scratching monolayers of cells at 100% confluence using a 200‐μL plastic pipette tip. The growth medium was replaced with serum‐free medium, and the wounds were monitored in real‐time with an IncuCyte® Live‐Cell Analysis System.
Female BALB/c nude mice (age, 4‐6 weeks; body weight, 18‐20 g; Charles River, Beijing, China) were subcutaneously injected via the left and right flanks with 5 × 106 SDCBP2‐AS1‐knockdown BGC823 cells or SDCBP2‐AS1‐overexpressing SGC79901 cells. NT BGC823 cells and Con SGC79901 cells were used as controls. Approximate tumor volume (mm3) was calculated according to the formula [(smallest diameter2 × widest diameter)/2] as previously described [24]. After 3 weeks, mice were euthanized by CO2 inhalation, and the tumors were harvested and immediatelyweighted. A lung metastasis model was created using female BALB/c nude mice that were subcutaneously injected via the tail vein with 2 × 106 SDCBP2‐AS1‐knockdown BGC823 cells or SDCBP2‐AS1‐overexpressing SGC7901 cells. At 4 weeks post‐injection, all mice were euthanized by CO2 inhalation, and the lungs were collected and fixed with Bouin's solution via injection through the main bronchi. The tissue samples were photographed, and metastases were evaluated and finally embedded in paraffin. Generally, body weight loss of more than 20% of the pre‐procedural weight or the size of the subcutaneous tumors reaching 20 mm was considered as a humane endpoint. All animal procedures were conducted following the guidelines of the Laboratory Animal Ethics Committee of Beijing Cancer Hospital (EAEC 2019‐03).
Generally, all dissected tissues were formalin‐fixed and paraffin‐embedded for staining with hematoxylin and eosin (H&E) and IHC analysis. As previously described [19], the slides were dewaxed and gradually hydrated. After retrieving antigens by using citrate buffer (Sigma‐Aldrich) at 120°C for 3 min and incubating within 3% H2O2 to eliminate endogenous enzymes, the slides were incubated overnight at 4°C with antibodies against Ki‐67, hnRNP K, and β‐catenin (Supplementary Table S2). Antigen visualization was performed with ImmPRESS Peroxidase Polymer Detection Reagent (Vector Laboratories, Burlingame, CA, USA) and 3,3′‐diaminobenzidine, followed by counterstaining with Mayer's hematoxylin (Sigma‐Aldrich). The stained specimens were viewed under a light microscope (Nikon Corporation, Tokyo, Japan). For Ki‐67, the proportion of positively stained cells was calculated. For β‐catenin, the scores of 0 (negative), 1 (weak), 2 (medium), and 3 (strong) were used for staining intensity quantification, while the scores of 0 (<5%), 1 (5%–25%), 2 (26%–50%), 3 (51%–75%) and 4 (>75%), which were based on the percentage of the positive staining areas in relation to the whole cancerous area, were used to evaluate the extent of staining. Then, the final IHC score was calculated by multiplying scores of staining intensity and percentage of positivity. Cases with a final staining score of ≤4 were deemed as low expression, and those with a score of >4 were reckoned as high expression. All slides were evaluated independently by two pathologists, who were blinded to the patients’ data and clinical outcomes.
RNA‐FISH with a digoxin‐labeled RNA probe (Supplementary Table S3) was used to detect the subcellular location of SDCBP2‐AS1 as previously described [24]. Briefly, the nuclei were labeled with 4′,6‐diamidino‐2‐phenylindole (blue), and SDCBP2‐AS1 was labeled with a cyanine 3‐conjugated RNA probe (red). Images were captured using an LSM 800 confocal microscope (Carl Zeiss AG, Jena, Germany). To separate the nuclear and cytoplasmic fractions, RNA was isolated from BGC823 and SGC7901 cells using a PARIS™ kit (AM1921, Life Technologies, Carlsbad, CA, USA) in accordance with the manufacturer's instructions. SDCBP2‐AS1 RNA expression levels in the nuclear and cytoplasmic fractions were determined by RT‐qPCR. U6 and GAPDH were used as positive controls for the nuclear and cytoplasmic fractions, respectively. The primer sequences are listed in Supplementary Table S1.
The RNA pull‐down assay was performed using the Pierce Magnetic RNA‐Protein Pull‐Down Kit (20164, Thermo Fisher Scientific, Rockford, IL, USA) in accordance with the manufacturer's instructions. Briefly, the sense or antisense SDCBP2‐AS1 sequence was transcribed in vitro with a biotin RNA‐labeling mix and T7 RNA polymerase (Invitrogen Corporation) in accordance with the manufacturer's instructions. The positive control, negative control, biotinylated SDCBP2‐AS1 or antisense SDCBP2‐AS1 was incubated with streptavidin‐linked magnetic beads and total BGC823 cell lysates at 21°C for 2 h. After washing the bead‐RNA‐protein complexes four times with binding/washing buffer, the proteins were precipitated and diluted in the protein lysis buffer. The collected proteins were separated by electrophoresis for one‐shot mass spectrometry or Western blotting analysis. Pull‐down samples harvested from immunoprecipitation were separated with sodium dodecyl sulfate‐polyacrylamide gel electrophoresis (SDS‐PAGE) and stained with Pierce Silver Stain Kit (24612, Thermo Fisher Scientific) according to the instruction manual. The specific strip was cut for mass spectrometry. The primers used for in vitro transcription are listed in Supplementary Table S1 and the antibodies in Supplementary Table S2. Mass spectrometry was performed at the Institute of Biotechnology of the Peking University Health Science Center (Beijing, China).
RIP assays were performed using the Magna RIP RNA‐Binding Protein Immunoprecipitation Kit (17‐700, EMD Millipore Corporation, Billerica, MA, USA), as previously described [24]. Briefly, 1 × 107 GC cells were harvested and lysed with RIP lysis buffer, and the cell extract was incubated with magnetic beads conjugated with antibodies against hnRNP K or normal rabbit immunoglobulin G (IgG) as a control (Supplementary Table S2). Afterward, the beads were washed and then incubated with proteinase K (EO0491, Thermo Fisher Scientific) to remove proteins. The retrieved RNA was amplified by RT‐qPCR using primers specific for SDCBP2‐AS1 (Supplementary Table S1). Total RNA (as input controls) and normal IgG controls were assayed simultaneously to verify the RT‐qPCR results.
The proteins isolated from the tissue samples and cell lines were separated by electrophoresis and transferred to polyvinylidene fluoride membranes, which were probed with primary antibodies against matrix metallopeptidase 7 (MMP7), cyclin D1, c‐Myc, β‐catenin, phospho‐β‐catenin (Ser33/37/Thr41), glycogen synthase kinase 3 beta (GSK3β), Axin1, adenomatous polyposis coli (APC), histone 3 (H3), Wnt5a, hnRNP K, ubiquitin, Wnt3a, GAPDH, hemagglutinin (HA), and Flag. All antibodies are listed in Supplementary Table S2.
Total RNA was isolated from 1 × 106 BGC823 cells with or without stable SDCBP2‐AS1 using the TRizol reagent. Sequence libraries were generated using the TruSeq® RNA Sample Preparation Kit (Illumina, Inc., San Diego, CA, USA) in accordance with the manufacturer's recommendations. Transcriptome sequencing was performed using an Illumina HiSeq X‐Ten platform by Novogene Biotech Co., Ltd. (Beijing, China). Fragments per kilo‐base of transcript per million fragments mapped of each gene were calculated, and differentially expressed genes (fold change > 2.0, P < 0.01, false discovery rate < 0.05) were selected for gene set enrichment analysis (GSEA), which was performed using GSEA software (v4.0.3; https://www.gsea‐msigdb.org/gsea/index.jsp).
The cells were fixed to the chamber slide with 4% formaldehyde solution, and the cell membranes were penetrated by the addition of 1% Triton X‐100. Normal sheep serum (Dako A/S, Glostrup, Denmark) was used to block the cells at 21°C for 30 min. Afterward, the cells were probed with an antibody against β‐catenin at 4°C for 12‐16 h, followed by a secondary Alexa Fluor™ 594 antibody for IF staining. After Hoechst staining, the cells were observed with an LSM 800 confocal microscope.
As previously described [25], the TOP/FOP‐flash luciferase reporter assay was applied to analyze the activity of the Wnt/β‐catenin signaling pathway. GC cells were seeded into 48‐well plates (6 × 104 cells/well) and transiently co‐transfected with 250 ng of TOP‐FLASH or FOP‐FLASH and 25 ng of the plasmid pRL‐SV40. Luciferase activity was measured with the Dual‐Luciferase Reporter Assay System (E1910, Promega Corporation, Madison, WI, USA) in accordance with the manufacturer's protocol. The TOP/FOP ratio was calculated as an indicator of β‐catenin signaling activity. Each experiment was performed in triplicate and repeated at least three times.
IP was performed as previously described [26] with antibodies specific for hnRNP K, β‐catenin, FLAG, or HA. The cells were transfected as indicated and harvested with IP lysis buffer (20 mmol/L Tris‐HCl at pH 7.5, 150 mmol/L NaCl, 1 mmol/L ethylenediaminetetraacetic acid, 0.1% NP‐40 nonionic detergent, 10% glycerol [vol/vol], 1 mmol/L dithiothreitol, protease inhibitor cocktail). The lysates were rotated at 4°C for 1 h and centrifuged at 12,000 rpm for 15 min. Then, the supernatants were collected and incubated overnight at 4°C with the appropriate antibody or an isotype‐matched antibody against IgG. After washing once with lysis buffer and twice with lysis buffer without 1% NP‐40 to remove all unbound proteins, the precipitants and lysates were subjected to Western blotting analysis with appropriate antibodies.
The cutoff value of RT‐qPCR results was determined by survival significance analysis using the tool Cutoff Finder (http://molpath.charite.de/cutoff/) [27]. Kaplan‐Meier curves were constructed to assess survival, and the log‐rank test was used to compare survival rates between groups. The chi‐square test was used to identify the association between SDCBP2‐AS1 expression levels and clinicopathological features of GC patients. A multivariate Cox proportional hazards model was used to assess the effects of variables, including SDCBP2‐AS1, on survival using IBM SPSS Statistics for Windows, version 22.0. (IBM Corporation, Armonk, NY, USA). The Pearson's correlation test was used to identify correlations between the expression levels of SDCBP2‐AS1 and β‐catenin. For functional analysis, data are presented as the mean ± standard deviation (SD). Comparisons between two and among three or more groups were conducted using the two‐tailed Student's t‐test and analysis of variance, respectively, with GraphPad Prism 7 software (GraphPad Software Inc., La Jolla, CA, USA). All experiments were performed at least three times. A P value of <0.05 was considered statistically significant.
Based our previous study [17], several lncRNAs were found to be significantly related to the progression of GC, one of which was SDCBP2‐AS1. SDCBP2‐AS1 was identified as a novel gene, and therefore, was the focus of the present study. The results of 5′/3′ RACE identified the novel lncRNA transcript as SDCBP2‐AS1 (NCBI reference sequence: NR_040047.1) (Supplementary Figure S1A). SDCBP2‐AS1 was conserved among primates (Supplementary Figure S1B) and was confirmed as a non‐coding gene using the online tools (Supplementary Figure S1C‐F). SDCBP2‐AS1 is located at chromosome 20p13 and overlaps with a short length of the 3’ untranslated regions of the gene encoding SDCBP2 on the opposite DNA strand (Supplementary Figure S1B). Searches of The Cancer Genome Atlas‐Stomach Adenocarcinoma (TCGA) GC database from GEPIA web and Asian Cancer Research Group databases confirmed that there was no correlation between SDCBP2 and SDCBP2‐AS1 at the post‐transcriptional level (Supplementary Figure S1G). Hence, SDCBP2‐AS1 was confirmed as a novel lncRNA. Next, potential associations between SDCBP2‐AS1 expression and different clinicopathological features of our clinical samples were investigated. The results showed that SDCBP2‐AS1 expression was significantly decreased in GC tissues as compared with the matched adjacent non‐tumor tissues (Figure 1A‐B). Based on the cutoff value of SDCBP2‐AS1 (0.0058), obtained from the Cutoff Finder, the patients were divided into SDCBP2‐AS1 high‐expression (n = 59) and low‐expression groups (n = 73). Notably, low expression of SDCBP2‐AS1 was associated with adenocarcinoma (P = 0.033), larger tumor size (P = 0.016), lymph node metastasis (P < 0.001), distant metastasis (P = 0.036), and advanced pathological TNM (pTNM) stage (P < 0.001) in GC patients (Table 1). Kaplan‐Meier survival analysis showed that the DFS and OS rates of the low SDCBP2‐AS1 group were lower than those of the high SDCBP2‐AS1 group (Figure 1C). Univariate and multivariate analyses indicated that SDCBP2‐AS1 expression was an independent prognostic factor for GC (Table 2). To validate these findings, the Kaplan‐Meier plotter was used to explore potential associations between SDCBP2‐AS1 overexpression and survival using the dataset GSE22377. The results showed that lower expression of SDCBP2‐AS1 was related to shorter OS and DFS (Figure 1D). These findings verified that low expression of SDCBP2‐AS1 was predictive of a poor prognosis for GC patients. Further in vitro evaluation revealed that SDCBP2‐AS1 expression was higher in normal GES‐1 cells than in GC cell lines (Supplementary Figure S1H).
SDCBP2‐AS1 was knocked‐down in BGC823 and MKN28 cells using two shRNAs. SDCBP2‐AS1 was overexpressed in SGC7901 cells. RT‐qPCR analysis was used to determine the efficacy of transfection (Supplementary Figure S2A‐B). Notably, SDCBP2 mRNA expression was independent of SDCBP2‐AS1 expression in GC cell lines with SDCBP2 wild‐type, knockdown or overexpression (Supplementary Figure S2C‐E). The cell proliferation and colony formation assays demonstrated that knockdown of SDCBP2‐AS1 significantly promoted the proliferation and colony formation in both cell lines (Figure 2A‐B), whereas overexpression of SCBP2‐AS1 had opposite effects (Figure 2C‐D). Further, the effects of SDCBP2‐AS1 on metastasis of GC cells were explored. The results of the transwell and wound‐healing assays showed that silencing of SDCBP2‐AS1 increased the migration and invasion of BGC823 and MKN28 cells (Figure 2E‐F), while overexpression of SDCBP2‐AS1 enhanced the mobility of SGC7901 cells (Figure 2G‐H). These in vitro data indicate that knockdown of SDCBP2‐AS1 aggressively promoted proliferation, migration, and invasion of GC cells.
The influence of SDCBP2‐AS1 on the tumorigenic and metastatic capacities of GC cells was determined using in vivo xenograft tumor and lung metastasis models in nude mice. The xenograft tumor model was established by subcutaneous injection of BGC823 cells stably transfected with shSDCBP2‐AS1#1 (sh1) or the NT control. The resulting tumors were visibly larger and exhibited faster growth in the sh1 group than in the NT group (Figure 3A‐C). This difference was further confirmed by staining with H&E and IHC analysis of Ki‐67, which showed that Ki‐67 expression in tumor tissues was significantly increased in the sh1 group as compared with the NT group (Figure 3D). Moreover, a lung metastasis model showed that the NT group had fewer metastatic lesions than the sh1 group (Figure 3E‐F). This difference was further confirmed by staining the lung tissue sections with H&E and IHC analysis of Ki‐67 (Figure 3G). Meanwhile, SDCBP2‐AS1 overexpression decreased the tumorigenic and metastatic capacity of SGC7901 cells (Supplementary Figure S3). Taken together, these in vivo results support the tumor‐suppressive role of SDCBP2‐AS1 in GC.
To identify potential mechanisms, the distribution of SDCBP2‐AS1 in GC cells was investigated using RNA‐FISH and RNA subcellular fractionation. The results showed that SDCBP2‐AS1 was more prevalent in the cytoplasm (Figure 4A‐B). Further, the RNA pull‐down assay with biotinylated SDCBP2‐AS1 was employed to identify the protein partner of SDCBP2‐AS1. The retrieved proteins were separated by electrophoresis and subjected to silver staining. Finally, several differential bands were selected for mass spectrometry (Supplementary Table S4), which confirmed that hnRNP K interacted with SDCBP2‐AS1 but not antisense SDCBP2‐AS1 (Figure 4C‐D). Next, in vivo RIP analysis showed that hnRNP K interacted with SDCBP2‐AS1 in BGC823 cells (Figure 4E). Collectively, these findings demonstrated the interplay between SDCBP2‐AS1 and hnRNP K. To determine which regions of SDCBP2‐AS1 bind to hnRNP K, a series of SDCBP2‐AS1 deletion mutants was constructed according to the predicted secondary structure of SDCBP2‐AS1 (Figure 4F). RNA fragments were in vitro transcribed from the deletion constructs for the RNA pull‐down assay (Figure 4G). Immunoblot analysis of hnRNP K in protein samples pulled down by different SDCBP2‐AS1 RNA fragments showed that RNA fragments with deletions of nucleotide 541‐2479 completely lost the ability to bind to hnRNP K (Figure 4H). The sequence of Exon 2 could interact with hnRNP K, its RNA secondary structure was totally different from the full‐length sequence of SDCBP2‐AS1 though (Supplementary Figure S4). Thus, the Exon 3 sequence was determined to be essential for the interaction between SDCBP2‐AS1 and hnRNP K (Figure 4H). Furthermore, hnRNP K harbors several functional domains that participate in RNA‐protein interactions, reportedly involving three K homology (KH) domains (KH1, KH2, and KH3). However, the KI domain, located between the KH2 and KH3 domains, mediates hnRNP K activity, although the underlying mechanism remains unclear [28]. The RIP assay with a series of Flag‐tagged hnRNP K deletion mutants in BGC823 cells (Figure 4I‐J) indicated that KH3 had the greatest ability to bind to SDCBP2‐AS1, while there was no significant difference in the capacities of the other domains to interact with SDCBP2‐AS1 (Figure 4K). These results identified the specific regions of SDCBP2‐AS1 with the ability to bind to the hnRNP K protein domains in GC cells.
To identify the putative targets of SDCBP2‐AS1, RNA‐seq was performed to obtain the transcriptional profiles of BGC823 cells following knockdown of SDCBP2‐AS1. Of 301 differentially expressed genes, 51 were upregulated and 250 were downregulated in the SDCBP2‐AS1‐knockdown group (Figure 5A). Furthermore, GSEA of the RNA‐seq data indicated that knockdown of SDCBP2‐AS1 was positively associated with overexpression of β‐catenin (Figure 5B). The subcellular localization of lncRNAs is related to function [29]. SDCBP2‐AS1 was mostly located in the cytoplasm and associated with β‐catenin expression, suggesting that SDCBP2‐AS1 interfered with the degradation of β‐catenin [30]. For confirmation, Western blotting analysis showed that knockdown of SDCBP2‐AS1 significantly increased the protein levels of β‐catenin and hnRNP K in BGC823 and MKN28 cells, while SGC7901 cells overexpressing SDCBP2‐AS1 had lower expression levels of β‐catenin and hnRNP K (Figure 5C). Consistently, IHC staining showed that the expression of β‐catenin was increased both in SDCBP2‐AS1 low group of our patient cohort and in xenograft tumors formed by SDCBP2‐AS1‐knockdown BGC823 cells (Supplementary Figure S5A‐C). Additionally, Western blotting analysis showed that knockdown of SDCBP2‐AS1 resulted in lower expression of β‐catenin in the cytoplasm and higher expression in the nucleus (Figure 5D). Opposite results were obtained for hnRNP K. The results of the IF assay demonstrated that SDCBP2‐AS1 altered the shuttling of β‐catenin between the nucleus and cytoplasm in GC cells (Figure 5E). The results of the TOP/FOP‐flash luciferase assays showed that β‐catenin signaling was activated in SDCBP2‐AS1‐knockdown cells (Figure 5F). In addition, the results of RT‐qPCR and Western blotting analyses showed that stable overexpression or knockdown of SDCBP2‐AS1 altered the expression patterns of genes downstream of β‐catenin (i.e., CCND1, MYC, and MMP7). Translocation of β‐catenin activated the Wnt signaling pathway, especially via Wnt3a, but not Wnt5a (Figure 5G‐H). Also, the Akt and Erk signaling pathways were activated in SDCBP2‐AS1‐knockdown BGC823 cells, while the Erk pathway was repressed in SGC7901 cells overexpressing SDCBP2‐AS1 (Supplementary Figure S5D‐E). Taken together, these results demonstrate that the knockdown of SDCBP2‐AS1 helped to stabilize β‐catenin and activate the canonical Wnt signaling pathway.
Since SDCBP2‐AS1 was found to regulate the protein levels of β‐catenin and hnRNP K in different directions, SDCBP2‐AS1 might also regulate post‐transcriptional modifications. Hence, the protein synthesis inhibitor CHX was used to evaluate the effect of SDCBP2‐AS1 on the stability of β‐catenin. After treatment with CHX, the protein synthesis of β‐catenin decreased significantly in all groups, implying that knockdown of SDCBP2‐AS1 or overexpression of SDCBP2‐AS1 did not affect the protein synthesis of β‐catenin (Figure 6A). In addition, after incubation with the proteasome inhibitor MG132, knockdown and overexpression of SDCBP2‐AS1 had similar effects on the degradation of β‐catenin in GC cells (Figure 6B). SDCBP2‐AS1 regulated the degradation, but not synthesis, of β‐catenin. IP analysis was performed to determine whether SDCBP2‐AS1 is involved in forming a degradation complex targeting endogenous β‐catenin. Since APC, Axin1, and GSK3β are the major components that induce the degradation of β‐catenin, the interaction between β‐catenin and the destruction complex decreased in the absence of SDCBP2‐AS1 (Figure 6C), suggesting that SDCBP2‐AS1 may interfere with degradation of β‐catenin. Consistently, overexpression of SDCBP2‐AS1 facilitated ubiquitination and degradation of β‐catenin (Figure 6C) by binding to hnRNP K. In the absence of SDCBP2‐AS1, ubiquitination of β‐catenin was incomplete (Figure 6D). To further investigate the capacity of SDCBP2‐AS1 to destabilize β‐catenin and given that the RNA pull‐down assay showed that SDCBP2‐AS1 did not directly interact with β‐catenin or β transducin repeat‐containing protein (β‐TrCP) (Supplementary Figure S6A), the ability of SDCBP2‐AS1 to regulate interactions between β‐catenin and hnRNP K was investigated. Hence, a series of Flag‐tagged truncated hnRNP K protein was constructed as described in the preceding text, and the protein domain mapping study indicated that the KH3 and KI domains of hnRNP K were essential for its interactions with β‐catenin (Figure 6E). Meanwhile, immunnoprecipitation of HA‐tagged β‐catenin truncation mutants showed all domains of β‐catenin were involved in its interactions with hnRNP K, although the full length had the greatest binding ability (Figure 6F). In addition, SDCBP2‐AS1 mainly interacted with the KH3 domain of hnRNP K. Several post‐translational modification sites have been reported in the KH3 region, such as the lysine residue at position 422, which was identified as the major SUMOylation site of hnRNP K [31, 32]. Hence, the ability of SDCBP2‐AS1 to regulate post‐translational modifications of hnRNP K was investigated, especially SUMOylation and ubiquitination of the KH3 domain. Flag‐tagged wild‐type hnRNP K and the K422R mutant showed that the mutant interfered with SUMOylation of hnRNP K (Supplementary Figure S6B). Knockdown of SDCBP2‐AS1 increased SUMOylation of the wild‐type, but not the K422R mutant (Figure 6G), and decreased endogenous poly‐ubiquitination of hnRNP K (Figure 6H). As compared with the K422R mutant, SDCBP2‐AS1 altered SUMOylation by blocking the specific active site of the KH3 domain in the spatial structure. These results indicate that knockdown of SDCBP2‐AS1 led to accumulation of β‐catenin in the cytoplasm and subsequent nuclear translocation via stabilization of β‐catenin in GC cells.
SGC7901 cells with stable knockdown of hnRNP K were established to assess the interactions between SDCBP2‐AS1 and hnRNP K in GC cells (Supplementary Figure S7A). In addition, SDCBP2‐AS1 was overexpressed to determine whether hnRNP K is a key mediator. Silencing of hnRNP K eliminated the tumor suppressor functions of SDCBP2‐AS1, including cell proliferation and migration (Figure 7A‐C). The results of the IF and Western blotting assays showed that without hnRNP K, SDCBP2‐AS1 failed to facilitate nuclear translocation of β‐catenin in GC cells (Figure 7D‐E). In addition, the results of the TOP/FOP‐flash luciferase, qPCR, and Western blotting assays showed that without hnRNP K, SDCBP2‐AS1 could not activate transcription of downstream target genes and the Wnt/β‐catenin pathway (Figure 7F‐H). In xenograft tumor models, stable knockdown of hnRNP K in SGC7901 cells erased the differences in growth rate and tumor weight caused by overexpression of SDCBP2‐AS1 (Figure 7I‐K). In the lung metastasis model, knockdown of hnRNP K also attenuated the impacts of SDCBP2‐AS1 overexpression on the metastatic potential of GC cells (Supplementary Figure S7B‐C). These results demonstrated that SDCBP2‐AS1 inhibited tumorigenicity via hnRNP K as shown in Figure 8.
Although many lncRNAs with altered expression have been reported in gastric tumorigenesis, the functional roles and molecular mechanisms of many GC‐specific lncRNAs have not been determined. In this study, SDCBP2‐AS1 was identified as a novel tumor suppressor in GC tumorigenesis and metastasis. Comparisons of the expression patterns of SDCBP2‐AS1 in GC tissues and validation of the Gene Expression Omnibus data showed that low expression of SDCBP2‐AS1 was associated with poor prognosis of GC patients. A series of in vitro and in vivo experiments showed that knockdown of SDCBP2‐AS1 significantly promoted tumorigenesis and metastasis. An analysis of public datasets showed that SDCBP2‐AS1 was significantly associated with DFS in patients with thyroid cancer [33]. However, the biological functions of SDCBP2‐AS1 were not investigated in that study. SDCBP2‐AS1 was shown to regulate ependymin‐related 1 (EPDR1) by competitively binding to miR‐100‐5p, which targets EPDR1, thereby inhibiting the progression of ovarian cancer [34]. Whether SDCBP2‐AS1 also functions as a competing endogenous RNA in GC is worthy of future research. In the present study, we focused on identifying the protein partner and their post‐transcriptional modifications pattern of SDCBP2‐AS1. The function of lncRNAs is largely determined by subcellular location [35, 36]. According to previous reports, lncRNAs located in the nucleus, such as Pvt1 oncogene (PVT‐1) and homeobox A11 antisense RNA (HOXA11‐AS), regulate transcription [37, 38], while those located in the cytoplasm, such as small nucleolar RNA host gene 5 (SNHG5) and urothelial cancer associated 1 (UCA1), act as sponges for microRNAs to regulate the expression of related mRNAs [39, 40] and others are involved in the regulation of cellular signaling pathways [41, 42, 43]. In the present study, RNA‐FISH was conducted to determine the function of SDCBP2‐AS1 and elucidate the molecular mechanisms in GC cells. Further, the RNA‐seq results showed a significant connection between SDCBP2‐AS1 and β‐catenin, which was confirmed by the Western blotting and TOP/FOP‐flash luciferase assay results. Knockdown of SDCBP2‐AS1 stabilized β‐catenin accumulation in the cytoplasm and its subsequent translocation into the nucleus, resulting in high expression of downstream target genes [15]. Moreover, cytoplasmic localization of SDCBP2‐AS1 promoted the degradation of β‐catenin in GC cells. The lncRNA pancEts‐1 was reported to interact with both hnRNP K and β‐catenin to activate transcription [44], and solute carrier organic anion transporter family member 4A1 antisense RNA 1 (SLCO4A1‐AS1) bound to β‐catenin was found to block the interaction between β‐catenin and GSK3β [45]. However, the results of the RNA pull‐down assay showed that SDCBP2‐AS1 directly bound to hnRNP K, but not β‐catenin. The IP assay showed that when SDCBP2‐AS1 was silenced, hnRNP K was still able to directly interact with endogenous β‐catenin, but not as strongly as in the presence of SDCBP2‐AS1. Therefore, SDCBP2‐AS1 assisted in the interaction between hnRNP K and β‐catenin, thereby effectively regulating the degradation of β‐catenin and achieving the effect of suppressing tumorigenesis. The novel lncRNA human Fas‐activated serine/threonine kinase (hFAST), located in the cytoplasm of human embryonic stem cells, maintained pluripotency of the cells by disrupting the interaction between β‐TrCP and β‐catenin, resulting in reduced degradation of β‐catenin [46]. In the present study, the RNA pull‐down assay did not detect direct interactions between SDCBP2‐AS1 and β‐TrCP, suggesting that SDCBP2‐AS1 promoted degradation of β‐catenin via hnRNP K in GC cells. The results of the RNA pull‐down and RIP assays of different truncated mutants of SDCBP2‐AS1 and hnRNP K showed that the secondary structures of RNAs were important for binding to proteins. As previously reported, the KH domain was the major RNA/DNA‐binding domain of hnRNP K [47, 48]. In the present study, the KH3 domain of hnRNP K was most readily bound to SDCBP2‐AS1, although the other domains also bound to SDCBP2‐AS1, indicating strong binding capacity between SDCBP2‐AS1 and hnRNP K. Previous studies have reported that lncRNA‐OG, lncRNA‐p21, lncRNA THRIL, and LINC01354 physically associate with other members of the hnRNP family to mediate transcriptional regulation [26, 49‐51]. The present study demonstrated that SDCBP2‐AS1 bound to the KH3 domain and blocked the major SUMOylation site of hnRNP K, thereby regulating SUMOylation and ubiquitination of hnRNP K to further facilitate ubiquitination of β‐catenin. Furthermore, SDCBP2‐AS1 was found to promote phosphorylation of cytosolic β‐catenin (Ser33/37/Thr41) as a part of the degradation step. Moreover, ubiquitinated hnRNP K might spatiotemporally recruit other E3 ubiquitin ligases, such as β‐TrCP, a major regulator of β‐catenin degradation. Silencing of hnRNP K almost completely abolished the biological function of SDCBP2‐AS1, suggesting that the function of SDCBP2‐AS1 is mediated by hnRNP K. hnRNP K is shuttled between the cytoplasm and nucleus and interacts with both nucleic acids and proteins, thereby participating in many cellular functions, including transcription, translation, mRNA splicing, and chromatin remodeling [52, 53]. Various post‐translational modifications (i.e., phosphorylation, ubiquitination, SUMOylation, and methylation) determine the biological functions of hnRNP K [54, 55]. In the present study, knockdown of SDCBP2‐AS1 increased the cytoplasmic fraction of hnRNP K and inhibited complex formation with β‐catenin in the nucleus. This phenomenon of SDCBP2‐AS1 in GC contradicts a previous report that hnRNP K and β‐catenin were translocated to the nucleus simultaneously [44]. As a possible explanation for this discrepancy, SUMOylation and ubiquitination serve to rapidly change the activity and abundance of hnRNP K in response to different stress conditions in the cytoplasm [56]. Alternatively, β‐catenin‐induced massive abundance of pre‐RNA requires the accumulation of hnRNP K in the cytoplasm to manage RNA processing. Although the results of this study revealed that SDCBP2‐AS1 could interact with hnRNP K and assist in the degradation of β‐catenin, the precise site of hnRNP K binds to SDCBP2‐AS1 remains unclear. A docking model of SDCBP2‐AS1 and hnRNP K are needed in our future study, which will spatially demonstrate how the ubiquitinated β‐catenin/hnRNP K complex regulates phosphorylation of β‐catenin and other unknown proteins involved in the degradation process.
In summary, we reported SDCBP2‐AS1 as a novel tumor suppressor lncRNA that inhibited proliferation and metastasis of GC cells. The tumor suppressor function of SDCBP2‐AS1 via directly binding to hnRNP K regulated post‐transcriptional modifications of hnRNP K and β‐catenin, thereby promoting degradation of β‐catenin. Consistently, stabilization of β‐catenin via loss of SDCBP2‐AS1 resulted in poor prognosis of GC patients, demonstrating that SDCBP2‐AS1 may be a potential diagnostic and prognostic biomarker for GC patients.
Conceived the hypothesis, Jing Han and Menglin Nie; Performed the experiments, Jing Han, Menglin Nie and Cong Chen; Designed and interpreted the results, Jing Han, Menglin Nie, Cong Chen, Xiaojing Cheng, Ting Guo, Xiaomei Li, Longtao Huangfu, and Hong Du; Wrote the manuscript, Jing Han, Menglin Nie and Cong Chen; Supervised the study, Xiaofang Xing and Jiafu Ji. All authors read and approved the final version of the manuscript.
The authors declare no potential conflicts of interest.
Informed consent was obtained from all patients. The study protocol was approved by the Ethics Committee of Peking University Cancer Hospital (approval no. 2018KT07).
Not applicable.
Click here for additional data file. Click here for additional data file. |
PMC9648394 | 36069342 | Kun Cheng,Ning Cai,Jinghan Zhu,Xing Yang,Huifang Liang,Wanguang Zhang | Tumor‐associated macrophages in liver cancer: From mechanisms to therapy | 07-09-2022 | hepatocellular carcinoma,intrahepatic cholangiocarcinoma,tumor‐associated macrophages,immunotherapy | Abstract Multidimensional analyses have demonstrated the presence of a unique tumor microenvironment (TME) in liver cancer. Tumor‐associated macrophages (TAMs) are among the most abundant immune cells infiltrating the TME and are present at all stages of liver cancer progression, and targeting TAMs has become one of the most favored immunotherapy strategies. In addition, macrophages and liver cancer cells have distinct origins. At the early stage of liver cancer, macrophages can provide a niche for the maintenance of liver cancer stem cells. In contrast, cancer stem cells (CSCs) or poorly differentiated tumor cells are key factors modulating macrophage activation. In the present review, we first propose the origin connection between precursor macrophages and liver cancer cells. Macrophages undergo dynamic phenotypic transition during carcinogenesis. In this course of such transition, it is critical to determine the appropriate timing for therapy and block specific markers to suppress pro‐tumoral TAMs. The present review provides a more detailed discussion of transition trends of such surface markers than previous reviews. Complex crosstalk occurs between TAMs and liver cancer cells. TAMs play indispensable roles in tumor progression, angiogenesis, and autophagy due to their heterogeneity and robust plasticity. In addition, macrophages in the TME interact with other immune cells by directing cell‐to‐cell contact or secreting various effector molecules. Similarly, tumor cells combined with other immune cells can drive macrophage recruitment and polarization. Despite the latest achievements and the advancements in treatment strategies following TAMs studies, comprehensive discussions on the communication between macrophages and cancer cells or immune cells in liver cancer are currently lacking. In this review, we discussed the interactions between TAMs and liver cancer cells (from cell origin to maturation), the latest therapeutic strategies (including chimeric antigen receptor macrophages), and critical clinical trials for hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA) to provide a rationale for further clinical investigation of TAMs as a potential target for treating patients with liver cancer. | Tumor‐associated macrophages in liver cancer: From mechanisms to therapy
Multidimensional analyses have demonstrated the presence of a unique tumor microenvironment (TME) in liver cancer. Tumor‐associated macrophages (TAMs) are among the most abundant immune cells infiltrating the TME and are present at all stages of liver cancer progression, and targeting TAMs has become one of the most favored immunotherapy strategies. In addition, macrophages and liver cancer cells have distinct origins. At the early stage of liver cancer, macrophages can provide a niche for the maintenance of liver cancer stem cells. In contrast, cancer stem cells (CSCs) or poorly differentiated tumor cells are key factors modulating macrophage activation. In the present review, we first propose the origin connection between precursor macrophages and liver cancer cells. Macrophages undergo dynamic phenotypic transition during carcinogenesis. In this course of such transition, it is critical to determine the appropriate timing for therapy and block specific markers to suppress pro‐tumoral TAMs. The present review provides a more detailed discussion of transition trends of such surface markers than previous reviews. Complex crosstalk occurs between TAMs and liver cancer cells. TAMs play indispensable roles in tumor progression, angiogenesis, and autophagy due to their heterogeneity and robust plasticity. In addition, macrophages in the TME interact with other immune cells by directing cell‐to‐cell contact or secreting various effector molecules. Similarly, tumor cells combined with other immune cells can drive macrophage recruitment and polarization. Despite the latest achievements and the advancements in treatment strategies following TAMs studies, comprehensive discussions on the communication between macrophages and cancer cells or immune cells in liver cancer are currently lacking. In this review, we discussed the interactions between TAMs and liver cancer cells (from cell origin to maturation), the latest therapeutic strategies (including chimeric antigen receptor macrophages), and critical clinical trials for hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA) to provide a rationale for further clinical investigation of TAMs as a potential target for treating patients with liver cancer.
Abbreviations ABTAA Tie2‐Activating Antibody ABA Ang2‐Blocking Antibody AD Alzheimer's disease AGER advanced glycosylation end product‐specific receptor Αkg α‐ketoglutarate ApoE apolipoprotein E B7‐H1 B7 homolog 1 BACE1 β‐site amyloid precursor protein‐cleaving enzyme 1 BM bone marrow CA1P combretastatin A‐1 phosphate CAFs cancer‐associated fibroblasts CAR‐M chimeric antigen receptor macrophage CAR‐T chimeric antigen receptor T cell CCL2 C‐C motif chemokine ligand 2 CCL5 C‐C motif chemokine ligand 5 CCL20 C‐C motif chemokine ligand 20 CCL17 C‐C motif chemokine ligand 17 CCL22 C‐C motif chemokine ligand 22 CCR2 C‐C motif chemokine receptor 2 CeO2NPs cerium oxide nanoparticles cGAS cyclic GMP‐AMP synthase CHCCA hepatocellular cholangiocarcinoma Clec4F C‐type lectin domain family 4 member F CSCs Cancer stem cells CSF1 colony‐stimulating factor‐1 CSF1R colony‐stimulating factor 1 receptor CTLA‐4 cytotoxic T‐lymphocyte antigen 4 CX3CR C‐X3‐C motif chemokine receptor CXCL2 C‐X‐C motif chemokine ligand 2 CXCL9 C‐X‐C motif chemokine ligand 9 CXCL10 C‐X‐C motif chemokine ligand 10 DCs dendritic cells ECs endothelial cells ECT2 epithelial cell transforming 2 EMPs erythromyeloid progenitors EMT epithelial mesenchymal transition ERK extracellular regulated protein kinase EZH2 enhancer of zeste 2 polycomb repressive complex 2 subunit FAK focal adhesion kinase FAO Fatty acid oxidation FAS factor related apoptosis FASL factor related apoptosis ligand GATA glutamyl‐tRNA amidotransferase, subunit A GM‐CSF colony‐stimulating factor G‐MDSCs granulocytic myeloid‐derived suppressor cells GPC3 Glypican‐3 HCC hepatocellular carcinoma HDAC6 histone deacetylase 6 HGF hepatocyte growth factor HLA human leukocyte antigen HLA‐DR human leukocyte antigens DR HLA‐E human leukocyte antigen E HLA‐G human leukocyte antigen G HME human macrophage metalloelastase HMGB1 high mobility group box 1 HSCs hepatic stellate cells iCCA intrahepatic cholangiocarcinoma ICIs immune‐checkpoint inhibitors ID3 inhibitor of differention‐3 IFNγ interferon‐gamma IGF insulin growth factor IGF‐R insulin growth factor receptor IL‐1β interleukin 1β IL‐1Rinterleukin 1 receptor;IL‐6 interleukin 1 receptor;IL‐6interleukin 6 IL‐10 interleukin 10 IL‐12 interleukin 12 IL‐13/34 interleukin 13/34 IL21interleukin 21;iNOS interleukin 21;iNOSinducible nitric oxide synthase LOXL4 lysyl oxidase‐like 4 IRF1 interferon regulatory factor 1 KCs Kupffer cells M1 TAMs M1 type tumor‐associated macrophages M2 TAMs M2 type tumor‐associated macrophages MACRO macrophage receptor with collagenous MAPK mitogen‐activated protein kinase MCP‐1 Monocyte chemoattractant protein‐1 MDSCs myeloid‐derived immunosuppressive cells MEK mitogen‐activated protein kinase kinase MHC‐II major histocompatibility complex type II MIF migration inhibitory factor miRNAs microRNAs MMP9 matrix metallopeptidase 9 MoMϕs monocyte‐derived macrophages mTOR mechanistic target of rapamycin NK cells natural killer cells NOS2 nitric oxide synthase NOX2 NADPH oxidase 2 PCR polymerase chain reaction PD‐1 programmed cell death protein 1 PD‐L1 programmed cell death protein 1 ligand 1 PFKFB3 6‐phosphofructo‐2‐kinase/fructose‐2,6‐biphosphatase 3 PGE2 prostaglandin E2 PI3Kγ Phosphoinositide 3‐kinase gamma POSTN Periostin PPAR peroxisome proliferators‐activated receptors PKM2 pyruvate kinase M2 pMacs pre‐macrophages p‐STAT‐3 phospho‐signal transducer and activator of transcription 3 REDD1 Regulated in development and DNA damage response 1 RelB Reticuloendotheliosis viral oncogene homolog B RIG‐I retinotic acid‐inducible gene I; , regulation of Igh‐1b 1 RIPK3 receptor‐interacting protein kinase 3 RNA‐Seq RNA Sequencing ROS reactive oxygen species SALL4 Sal‐like protein‐4 S100A9 S100 calcium‐binding protein A9 SIRT1 sirtuin 1 SIRT4 sirtuin 4 SLC7A11 solute carrier family 7 member 11 SPON2 matricellular protein spondin2 STING stimulator of interferon genes TANs tumor‐associated neutrophils TAMs tumor‐associated macrophages TAZ tafazzin TCA tricarboxylic acid cycle TEMs Tie2‐expressing monocytes TFH cells T follicular helper cells TGF‐β transforming growth factor‐β Th1 type 1 T helper cells Tim‐3 T cell immunoglobulin and mucin‐containing molecule 3 Tim‐4 T‐cell immunoglobulin and mucin domain containing 4 TKIs tyrosine kinase inhibitors TLR Toll‐like receptor TLR2 toll‐like receptor 2 TME tumor microenvironment TNF tumor necrosis factor TNF‐α tumor necrosis factor α TNFR1 tumor necrosis factor receptor 1 TNFSF tumor necrosis factor superfamily TRAF2 tumor necrosis factor receptor‐associated factor 2 Treg regulatory T TREM‐1 triggering receptor expressed on myeloid cells‐1 TSP1 thrombospondin‐1 T‐VEC Talimogene laherparepvec TWEAK Tumor necrosis factor‐like weak inducer of apoptosis YAP yes‐associated protein 1 VEGF vascular endothelial growth factor
Emerging evidence has shown that the TME plays a pivotal role in driving cancer progression and governing the response to standard‐of‐care therapies [1]. Multiple components coexist and interact in the TME, including tumor‐associated macrophages (TAMs), CD4+ and CD8+ T cells, dendritic cells (DCs), natural killer (NK) cells, tumor‐related endothelial cells (ECs), abnormal tumor vasculature, cancer‐associated fibroblasts (CAFs), and myeloid‐derived immunosuppressive cells (MDSCs). As a dynamic system orchestrated by multiple cellular and non‐cellular components, each cellular component in the tumor immune microenvironment represents a potential target for reprogramming the TME (Figure 1). Clinical studies and experimental mouse models have indicated that TAMs are particularly abundant among innate and adaptive immune cells recruited to the tumor milieu [2, 3, 4]. TAMs typically exhibit distinct functional phenotypes. M1‐like TAMs exert pro‐inflammatory and anti‐tumor activities, whereas M2‐like TAMs exert anti‐inflammatory and tumor‐promoting effects. M2 TAMs can promote cancer initiation, suppress antitumor immunity, stimulate angiogenesis, and enhance tumor cell invasion, motility and intravasation [5, 6] (Figure 1). However, the classification of macrophages is far more complex than previously thought, and the markers on the surface of macrophages are constantly in flux as cancer progresses [7, 8]. Rather than simply classifying macrophages, in the present review, we provide a comprehensive discussion of the subpopulation and the dynamic transformation of markers on their surfaces. Specific macrophage subsets assume distinct roles in cancer progression and antitumor immunity [9]. Traditional flow cytometry and histological methods to define TAMs appear to be limited because of the inability to capture the full diversity of the cells or distinguish them from other cell populations. With the development of single‐cell technology and spatial transcriptomics, determining the origin and spatial distribution of TAMs and identifying specific macrophage markers to design targeted and personalized medicine is essential for preventing and treating hepatic malignancies [10, 11]. Very few systematic reviews have discussed the interactions between the origin of macrophages and cancer cells in liver cancer. We first propose a relationship between precursor macrophages and liver cancer stem cells. This potential interaction in cellular evolution contributes to a better understanding of how macrophages can promote early‐stage liver cancer progression and develop appropriate treatment strategies. Therefore, the interactions of cell origins cannot be ignored. Moreover, it is unclear whether macrophage precursors interact with other immune cell precursors, such as lymphoid cell precursors. Additional research is required to test such hypotheses. In the past, associated research on TAMs mainly has focused on how TAMs interact with cancer cells and their phenotype switching [12, 13, 14, 15, 16]. The present review is a more in‐depth discussion than previous reviews. We described the effects of macrophages on cancer cells and summarized the regulation of macrophages by cancer cells. Moreover, we discussed the regulation of macrophage metabolism, including glucose metabolism, lipid metabolism, and amino acid metabolism, which are also essential key research directions of our research group [15, 17, 18]. The liver is vital for metabolic homeostasis; therefore, more metabolomic data on liver TAMs are warranted to confirm that macrophage metabolism regulates liver cancer progression. The role of the formation of an immunosuppressive microenvironment is not limited to TAMs alone. Studies on the relationship between macrophages and CD8+ T cells are currently widespread [19, 20]. However, studies on the relationships between macrophages and other immune cells, such as CD4+ T cells, B cells, Tregs, and MDSCs, are still lacking. The reciprocal regulation between TAMs and these cells is an important and novel research direction. Finally, we divided the macrophage‐targeted therapy for liver cancer into three categories based on published studies and summarized the latest treatment strategies and clinical trials compared to those of the past. Combination therapy is the primary treatment strategy for liver cancer [21]. We explored studies on combined therapy with multi‐target inhibitors such as sorafenib or lenvatinib and evaluated the efficacy of immune checkpoint inhibitors (ICIs) such as anti‐programmed cell death protein 1 (PD‐1) [22, 23, 24]. A critical hypothesis is whether macrophage metabolism disorder is the core resistance mechanism to such drugs. To support such conclusions, more studies on liver cancer are required to explore drug resistance targets around macrophage metabolism. In summary, in the present review, we illustrated the potential origin connection between TAMs and liver cancer cells, the dynamic transitions of TAMs, and the latest insights into therapeutic strategies used with TAMs in the liver field, with particular emphasis on interactions between these and other cells in the liver cancer microenvironment, which could promote the clinical translation of macrophage‐based combination therapy.
TAMs are one of the most abundant infiltrative immune cells in tumor stroma and play a pivotal role in inflammation. They can be divided into two categories according to their source: tissue‐resident macrophages and monocyte‐derived macrophages (MoMϕs) [10]. In the normal liver, macrophages mainly exist in the form of tissue‐resident macrophages, namely Kupffer cells (KCs). Increasing evidence shows that tissue‐resident macrophages may be derived from erythromyeloid progenitors (EMPs) that express the macrophage colony‐stimulating factor 1 receptor (CSF1R) in the yolk sac or fetal liver [25]. During liver cancer progression, tissue‐resident macrophages are stimulated by pro‐tumorigenic factors, which causes them to undergo a phenotypic switch and eventually become TAMs. Recent data have confirmed that EMPs can produce pre‐macrophages (pMacs) and differentiate into KCs in a chemokine‐receptor‐dependent manner [26]. The sources of KCs are also diverse, contributing to the significant heterogeneity in the resident cells of liver tissues. When becoming malignant, macrophages undergo sequential stages as EMPs, monocyte progenitors, mature monocytes, KCs, and TAMs. Therefore, unraveling the mystery of this heterogeneity is essential for targeted macrophage therapy [27, 28] (Figure 2). Another main line of macrophage genesis is very complex and lengthy. Embryonic hematopoietic stem cells that colonize the fetal liver (embryonic period) and migrate to the bone marrow (BM, adult period) can serve as a source of monocytes. MoMϕs penetrate tumor tissues through the blood and differentiate into TAMs during liver cancer [29, 30] (Figure 2). Furthermore, splenic monocytes represent a secondary source of TAMs, which can play a significant role in the inflammatory response following acute injury [31, 32]. Liver stem/progenitor cells are known as hepatic progenitor cells in humans and oval cells in rodents. We previously conducted multiple studies on liver progenitor cells [33, 34]. Clinical and pathological analyses of combined hepatocellular cholangiocarcinoma (CHCCA) have revealed the characteristics of progenitor cells [35]. These findings suggest that CHCCA originates directly from liver progenitor cells. The current consensus is that mature hepatocytes and bile duct cells derived from hepatic progenitor cells or precursor cells transform into hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA), respectively, in the presence of cancer‐stimulating factors [36, 37]. Intriguingly, adult hepatocytes and bile duct cells can dedifferentiate into precursor cells, ultimately transforming into cancer cells with progenitor cell markers [38]. Moreover, hepatocytes can transdifferentiate into bile duct‐like cells and evolve into iCCAs [39] (Figure 2). Recently, a significant discovery revealed that tissue‐resident macrophages could provide a pro‐tumorigenic niche for early cancer [10]. Based on our previous liver cancer research, we proposed that KCs combined with hepatic stellate cells (HSCs) may facilitate the dissolution of the basement membrane. This process could allow residual hepatic cancer stem cells in the canals of Hering to move into the adjacent hepatic lobe and differentiate into cancer cells [34]. Macrophages undergoing malignant transformation can regulate liver precursor cells before maturation and provide a unique niche for maintaining CSCs and controlling their behavior from the time of their origin. CSCs or poorly differentiated tumor cells in the microenvironment act as crucial factors in modulating macrophage activation (Figure 2). Recently, the discovery of oncofetal reprogramming of the tumor ecosystem revealed that reprogramming of embryonic HCC ECs promotes the production of immunosuppressive macrophages [40]. Furthermore, a novel subgroup of cells with a resident CXCR5−PD‐1−BTLA−CD69high phenotype has been identified. These protumorigenic T follicular helper (TFH) cells can create conditions for M2b macrophage polarization through the interleukin 21 (IL21)‐ interferon‐gamma (IFNγ) pathway [41]. Such findings suggest that cancer cells and other immune cells can influence macrophages during the early stages of liver cancer. As more TAM sources are characterized, the diversity and complexity of macrophages in the TME will be better understood. In addition, both macrophage differentiation and cancer cell development are inextricably linked. We discuss these processes in depth in this review, offering the tantalizing possibility of using therapies targeting the recruitment of primitive macrophage subsets and their communication with hepatic progenitors and progeny.
TAMs are recruited and activated by different chemokines in the immuno‐inflammatory microenvironment of liver cancer and differentiate into particular polarized forms associated with specific pathological conditions. Macrophages can be divided into two different polarization states according to the state and function of macrophage activation: classically activated M1 macrophages and alternatively activated M2 macrophages [42]. Both polarization forms are interconvertible under specific circumstances and in the presence of certain stimuli. M1 macrophages usually play pro‐inflammatory roles and secrete large amounts of pro‐inflammatory cytokines. Classical macrophage activation can occur when cells are stimulated with: 1) lipopolysaccharide, a component of the cell wall of gram‐negative bacteria, 2) IFN‐γ, released by NK cells and type 1 T helper (Th1) cells, 3) tumor necrosis factor (TNF), 4) granulocyte‐macrophage colony‐stimulating factor (GM‐CSF), and 5) Toll‐like receptor (TLR) ligands [43]. Activated M1 macrophages secrete certain interleukins, chemokines and TNF‐α to elicit pro‐inflammatory effects and can also exert cytotoxic effects by activating nitric oxide synthase (NOS2) or inducible nitric oxide synthase (iNOS) to produce NO and release reactive oxygen species (ROS). In addition, M1‐type macrophages, which highly express major histocompatibility complex type II (MHC‐II), can regulate and promote Th‐1 type cellular immune responses by presenting antigens to T cells [44]. In contrast, M2 macrophages usually perform functions opposite to those of M1 macrophages. The cytokines IL‐4, IL‐10, IL‐13 and transforming growth factor‐β (TGF‐β) are secreted by Th2 cells and tumor cells, and CSF1 and prostaglandin E2 (PGE2) can induce the alternative activation of macrophages resulting in the M2 polarized phenotype. M2 macrophages can secrete several complex immunosuppressive factors, cytokines, and growth factors; regulate Th‐2 type immune responses; promote tumor cell growth, and participate in tumor angiogenesis [45]. Furthermore, owing to their high heterogeneity, M2‐type macrophages are usually divided into four subtypes: M2a, M2b, M2c, and M2d [46]. The phenotypic heterogeneity and plasticity of the M2 macrophage subtypes have been described in diseases such as atherosclerosis [47]. Table 1 presents a detailed summary of the surface markers of human and mouse macrophages. Research on the specific subtypes of M1/M2 TAMs in liver cancer has been inconclusive. Therefore, it is urgent and necessary to classify the unique phenotypes exhibited by M1/M2 TAMs and propose precise targeted therapies for disparate patients with various stages of cancer. For example, the results of single‐cell RNA sequencing have indicated that microglia‐derived TAMs are predominant in newly diagnosed glioblastoma. In patients with tumor recurrence, monocyte‐derived TAMs outnumber microglia‐derived TAMs [48]. Such findings indicate that tumor intervention strategies may be distinct at multiple stages. Dynamic phenotypic transitions accompany macrophage recruitment and activation. Quiescent KCs originate from yolk sac‐derived CSF1R+ EMPs, which are usually CD11blow, F4/80high, CD68+ and Ly6C− [49, 50]. The predominant hepatic macrophages are located under the sinusoid endothelium and are involved in scavenging dying cells, pathogens, and molecules. In one study, the inhibitor of differention‐3 (ID3)+ progenitors were found to infiltrate the fetal liver during embryogenesis and are essential for giving rise to self‐maintaining KCs [26] (Figure 3). MoMϕs are usually CD11b+, F4/80low, CSF1R+, C‐C motif chemokine ligand 2 (CCL2)+and Ly6C− [51, 52]. These cells may be derived from CX3CR1+CD117+Lin− progenitor cells in the BM [53]. Regarding mouse models of liver disease, hepatic MoMϕs are divided into two main subpopulations according to their Ly6C‐expression levels: Ly6Chigh and Ly6Clow MoMϕs [54]. Similar to pro‐inflammatory M1 macrophages, Ly6Chigh MoMϕs can exacerbate inflammation and fibrogenesis [55]. A previous report indicated that recruited CCR2+Ly6Chigh monocytes replaced embryonic precursor cells and differentiated into mature anti‐inflammatory macrophages. This finding showed that Ly6Chigh MoMϕs had similar characteristics to M1 TAMs, while M2 TAMs usually showed low expression of Ly6C. Therefore, Ly6C is a significant dynamic surface marker during the evolution of TAMs [56] (Figure 3). In addition, macrophages expressing CD11bhigh, F4/80low and Ly6Chigh may constitute an early macrophage phenotype in the dynamic TAM environment in the liver [8, 57]. Moreover, the surface markers of M2‐like TAMs appear to be consistent with those of CD11blow, F4/80high and Ly6Clow MoMϕs. These three markers exhibit dynamic phenotypic transitions in the TAM environment, which suggests that surface markers can be used to track the distribution of TAMs to determine the progression of liver cancer [40, 58]. Although a schematic waterfall model of TAMs development has also been described, we further summarized the published research on liver cancer and generated a more detailed map of dynamic phenotypic transitions in macrophages during carcinogenesis [57] (Figure 3). The transcriptional profiles of the human liver have been reported with the help of single‐cell RNA sequencing. Researchers have segregated intrahepatic CD68+ macrophages into two distinct populations: the CD68+ macrophage receptor with collagenous (MARCO)+ immunoregulatory phenotype and the CD68+ MARCO− proinflammatory phenotype [59, 60]. Human CD68+ MARCO+ cells are transcriptionally similar to the mice population participating in maintaining immune tolerance and suppressing inflammation. This observation indicated that a subtype of suppressive TAMs express MARCO. In contrast, CD68+ MARCO− cells are strongly associated with inflammation. This macrophage subpopulation is characterized by the enriched expression of inflammatory genes [59]. This classification is consistent with the characteristics of our defined macrophage subpopulation, which affects inflammation during development (Figure 3). A recent study has demonstrated that glutamyl‐tRNA amidotransferase, subunit A 6 (GATA6) macrophages could migrate toward sites of injury and repair focal lesions rapidly when the human body suffers peritoneal injury. However, abdominal adhesions can occur as a side effect of this otherwise beneficial repair process [61]. The rapid response of macrophages plays a decisive role in adhesion. However, whether GATA6 macrophages can serve as a source of TAMs that participate in liver cancer progression remains unclear. Tie2‐expressing monocytes (TEMs) play an important role in tumor angiogenesis, and selective elimination of these TEMs through suicide genes has become a novel treatment strategy [62, 63]. Some research groups have developed a mathematical framework for systematically estimating the roles of TEMs in the M1 and M2 macrophage phenotypes during the growth of vascularized tumor lesions [64]. TEMs have been identified as a distinct TAM subpopulation influencing tumor angiogenesis, vascular remodeling and monocyte differentiation [65]. Although previous findings have revealed the recruitment of macrophages and dynamic switching of surface markers, these processes vary across cancer types. In addition, convincing evidence on how human macrophages recruit and find suitable surface markers is required to formulate strategies based on targeting macrophages.
Classically activated (or inflammatory) macrophages exhibit anti‐cancer properties. In HCC, M1 can inhibit tumor progression through various mechanisms [66, 67, 68]. Most current research is focused on regulating genes or proteins in cancer cells that can induce the polarization or infiltration of macrophages. The matricellular protein spondin2 (SPON2) activates RhoA and Rac1 and increases F‐actin reorganization through SPON2‐α4β1 integrin signaling to promote the infiltration of M1‐like macrophages [69]. High sirtuin1 (SIRT1) expression in hepatoma carcinoma cells regulates M1 polarization via the NF‐κB pathway [70]. In addition, an increase in retinoic acid‐inducible gene I (RIG‐I) expression can promote the M1 polarization of mouse peritoneal macrophages via the RIG‐I/MAVS/TRAF2/NF‐κB pathway, thereby inducing apoptosis of HCC cells [71]. Furthermore, monocytes overexpressing IL‐12 can downregulate phospho‐signal transducer and activator of transcription 3 (p‐STAT‐3) and c‐Myc to directionally differentiate into M1 and inhibit HCC growth [72]. However, M1 also shows a positive correlation with cancer, and this abnormal mechanism is relatively rare. For instance, M1 macrophages secrete IL‐1β to activate hepatoma carcinoma cells and induce programmed cell death protein 1 ligand 1(PD‐L1) expression through the transcription factors IRF1 and NF‐κB [73](Figure 4). Therefore, M1 TAMs and M2 TAMs are not always mutually exclusive; on the contrary, the two types of cells often coexist in the TME. Consequently, the two types of macrophages cannot be considered entirely distinct macrophage populations. The function favored by the mixed TAMs phenotype depends on the balance between macrophage activation and suppression and the immune microenvironment. M1 TAMs‐derived exosomes surface‐modified with IL4RPep‐1(named IL4R‐Exo si/mi) reportedly suppress the IL‐4 receptor of M2 TAMs and decrease the levels of M2 cytokines. Such findings indicated that IL4R‐Exo(si/mi) induces the M1 TAMs phenotype and enhances antitumor immunity [74]. Multiple studies have shown that M1 macrophages are positively associated with response to anti‐PD‐1 therapy. Persistent DNA damage contributes to the cyclic GMP‐AMP synthase (cGAS)‐stimulator of interferon genes (STING) signaling pathway in M1 TAMs. Activated M1 TAMs induce T‐lymphocyte infiltration, which could enhance the response to anti‐PD‐1 [75]. Inhibition of p38 kinase phosphorylation and downstream Creb1/Klf4 activity in bone marrow‐derived macrophages by regorafenib reverses M2 polarization and enhances synergistic antitumor effect with anti‐PD‐1 [76]. In a recent study, human leukocyte antigens DR (HLA‐DR)high CD86high glycolytic phenotype macrophages were shown to represent the primary cellular source of PD‐L1 in human HCC tumors. Inhibiting the critical glycolytic enzyme pyruvate kinase M2 (PKM2) or PD‐L1 blockade liberated the inherent antitumor capability of PKM2+ glycolytic macrophages by producing antitumorigenic cytokines such as IL‐12p70. In addition, patients with more PD‐L1+PKM2+ glycolytic TAMs demonstrated a poorer prognosis [77]. In a phase I Hodgkin lymphoma clinical trial, using PI3Kδ/γ inhibitor RP6530 could switch macrophages to the M1 phenotype. PI3Kδ/γ inhibition is an effective therapeutic strategy for Hodgkin lymphoma [78]. The higher the CD137 level in the serum and the higher the M1 density in the matrix in 50 patients with advanced HCC who received sintilimab (anti‐PD‐1) plus IBI305 (a bevacizumab biosimilar), the longer the median progression‐free survival and median overall survival [24]. β‐site amyloid precursor protein‐cleaving enzyme 1 (BACE1) inhibitor MK‐8931 potently reprogramed M2 TAMs into M1 TAMs via inhibiting IL‐6‐STAT3 signaling and activated macrophage phagocytosis in cancer cells [79]. Encouragingly, MK‐8931 has been tested in clinical trials for Alzheimer's disease (AD) treatment and suggested that M1‐type antitumor macrophages were positively associated with a favorable prognosis [80]. Interfering different signaling pathways in macrophages or blocking receptors on the surface of M2 TAMs to convert them into M1 TAMs is an attractive clinical translational strategy.
In hepatic tumors, the essential characteristics of aggressiveness are associated with achieving stemness. CSCs are self‐renewing cells that can facilitate tumor initiation and enhance immune resistance. Cancer cells with these biological properties may positively correlate with tumor development and metastasis. It is particularly important to clarify the crosstalk mechanisms between TAMs and CSCs in HCC. Compared with tumor cells, the interaction between CSCs and TAMs plays a more central role in tumorigenesis, progression, and metastasis formation. Inflammatory microenvironment‐related secreted S100 calcium‐binding protein A9 (S100A9) is highly expressed in TAMs. S100A9 can promote the stemness of hepatoma carcinoma cells by activating the NF‐κB signaling pathway. Intriguingly, HCC cells treated with S100A9 can recruit more macrophages via chemokine ligand 2 [81]. Macrophage‐induced long noncoding RNA H19 upregulation enhances stemness and promotes tumorigenesis, confirming that macrophage‐induced H19 is significantly correlated with HCC progression [82]. Besides, cytokines (TNF‐α, IL‐6, and TGF‐β) and low levels of microRNAs (miRNAs, such asmiR‐125a and miR‐125b) derived from M2 TAMs can also promote cancer stemness and CSC expansion [13, 83‐85](Figure 4). The intervention of CSCs by targeting TAMs is a novel tumor immunotherapy strategy. The interaction between TAMs and CSCs is bidirectional. Periostin (POSTN; a member of the fasciclin family secreted by CSCs) may significantly promote the recruitment of M2 TAMs in iCCA [86]. CSCs in iCCA release multiple molecules, such as IL13, OA and IL34, which can guide macrophage precursors to the M2‐like cancer‐promoting phenotype [87]. In conclusion, further understanding of the biology of CSCs and elucidating the interaction between CSCs and TAMs in various stages of tumor growth are the keys to mitigating liver cancer progression.
Although most of the current related research have focused on tumor‐derived exosomes, the presence of TAM‐derived exosomes is necessary for tumor progression and metastasis [88, 89]. Alternatively, activated (M2) macrophages can secrete the cytokine CCL22 to enhance tumor invasion and induce epithelial‐mesenchymal transition (EMT) via Smad2/3 and Smad1/5/8 activation and Snail upregulation [90]. In addition, CCL17 secreted by M2 macrophages is closely related to tumor stemness and EMT in the TGF‐β1 and Wnt/β‐catenin signaling pathways [91]. In vitro data have indicated that M2‐TAMs orchestrate the immune microenvironment of iCCA by secreting various cytokines, such as TNF‐α, ICAM‐1, IL‐6, and modulating the EMT of cancer cells [92](Figure 4). In an animal model of ICC, tumor cells were shown to accelerate THP‐1 cells (human acute monocytic leukemia cell line) differentiation into M2 macrophages, and the polarized macrophages could secrete IL‐10 to promote the growth, invasion and EMT of liver cancer cells [93]. The accumulation of KCs‐derived ROS and paracrine TNF causes mitochondrial dysfunction and induces oxidative stress, which could lead to the formation of premalignant lesions [94]. TNF has been positively associated with bile duct cell proliferation and carcinogenesis. Blocking the ROS/TNF/JNK axis may be an effective therapeutic strategy for attenuating the growth of iCCA tumors [94]. Intriguingly, enhanced communication between TAMs and tumor‐associated cells also promoted cancer invasion and metastasis. After co‐culturing tumor‐associated neutrophils (TANs) with macrophages, oncostatin M and interleukin‐11 were both expressed at higher levels than in the corresponding individual cultures. Crosstalk between TANs and TAMs was shown to enhance the proliferation and aggression of ICC via STAT3 signaling [95]. The above studies suggest that macrophages play exciting roles in tumorigenesis. Therefore, an attractive therapeutic strategy for liver cancer could be to block communication between M2 TAMs and liver cancer cells, such as using noncoding RNA inhibitors and TAMs receptor inhibitors.
Angiogenesis is essential for hepatocarcinogenesis and metastasis due to the hypervascular nature of most HCC tumors. Angiogenesis in HCC is a multidimensional process orchestrated by hepatoma carcinoma cells and a repertoire of tumor‐associated stromal cells, including TAMs and their bioactive products. The tumor‐microvessel density correlated positively with macrophage counts, which revealed a crucial role for TAMs in early‐stage HCC neovascularization [96]. In recent years, the monocyte/macrophage subpopulation, characterized by the expression of the tyrosine kinase receptor Tie‐2, has attracted attention. TEMs mainly aggregate in the perivascular area of tumor tissues and participate in HCC angiogenesis [97]. Human macrophage metalloelastase (HME) and vascular endothelial growth factor (VEGF) have been implicated in tumor angiogenesis. The balance between HME and VEGF gene expression can significantly affect tumor angiogenesis [98, 99]. For instance, cytokines derived from ICC cells can induce macrophage differentiation into M2‐TAMs, with increased vessels and VEGF expression [100]. Macrophage populations and phenotypes are positively correlated with angiogenesis and clinical prognosis in liver cancer. CCR2+ TAMs are more abundant at the edge of highly vascularized HCC, while the absence of CCR2+ TAM infiltration attenuates pathogenic vascularization [101]. A case‐controlled study showed that CD14+ inflammatory macrophages secreted large amounts of IL‐23 after stimulation by hepatitis virus‐infected hepatocytes, accompanied by the upregulation of IL‐23 receptors and intense macrophage‐associated angiogenesis [102]. In addition, CD14+ CD16+ monocytes from patients with liver cancer express high levels of angiogenic factor‐related genes (epiregulin, VEGF‐A and CXCL3) and predict the tissue invasive character of iCCA [103]. Hypoxic TAMs acquire angiogenic and immunosuppressive properties. Regulated in development and DNA damage response 1 (REDD1), a negative regulator of the mechanistic target of rapamycin (mTOR), was significantly upregulated in hypoxic TAMs. This inhibitor can hinder glycolysis and the vascular‐hyperactivation response in TAMs. Thwarting glycolysis in REDD1‐knockout TAMs may lead to abnormal angiogenesis [104]. The pro‐angiogenic properties of TAMs and vasculogenic mimicry in TME are fundamental reasons for the poor prognosis of tumor patients. The accumulation of macrophages was shown to correlate with the emergence of resistance to anti‐VEGF therapy in a preclinical model [105]. The escape from VEGF‐directed treatment may be due to the downregulation of VEGFR‐1 and VEGFR‐3 and the upregulation of angiogenic‐promoting genes. Such a key finding suggests that using VEGF blockade combined with macrophage blockade (such as CSF1 or CCR2 inhibitors) could enhance the anti‐VEGF therapeutic response.
Substantial evidence has demonstrated that autophagy plays an essential role in cell stress response, thereby maintaining internal environment stability. Listeria‐based HCC vaccines induce autophagy in TAMs via the Toll‐like receptor 2 (TLR2)/Myd88/NF‐κB pathway [106]. The induction of autophagy led to macrophage repolarization from the M2 to the M1 phenotype and recruited mounting anti‐tumor cytokines. Although ICIs have achieved promising therapeutic outcomes in liver cancer, ICIs still have a low response rate. Combined with PD‐1 blockade, this vaccine induced a robust antitumor response and reshaped the tumor immune microenvironment [106]. Additionally, TAMs can induce autophagy in HCC cells and attenuate the toxic effects of oxaliplatin. This autophagy‐mediated, drug‐resistance mechanism provides a new therapeutic strategy [107]. Other findings have unveiled the autophagy‐associated tumor necrosis factor receptor‐associated factor 2 (TRAF2) degradation and RelB/p52 activation can initiate TAM reprogramming to M1 macrophages, as observed in the presence of baicalin [108]. Moreover, autophagy in macrophages is triggered by the high‐mobility group box 1/Toll‐like receptor 2/NADPH oxidase 2 (HMGB1/TLR2/NOX2) autophagy axis, where hepatoma‐derived HMGB1 can skew TAMs to the M2‐like phenotype to support HCC growth [109] (Figure 4). Autophagy is critical for controlling macrophage production and polarization at different stages. Modulating autophagy in TAMs could be a promising strategy for inhibiting liver cancer growth and progression.
Tumor‐derived extracellular vesicles are essential mediators of cell‐to‐cell communication during tumorigenesis. Macrophages can release cytokines and exosomes, which can in turn affect tumor cells. Similarly, liver cancer cells secrete different factors into their local environment, promoting macrophage polarization. Emerging evidence suggests that liver cancer cell‐derived exosomes facilitate cancer progression. Exosomal Sal‐like protein‐4 (SALL4) binds to the promoter region of miR‐146a‐5p and upregulates its expression in HCC‐derived exosomes. In addition, SALL4 plays a key role in T cell exhaustion [110]. Liver tumor‐derived lncRNAs (such as TUC339), circRNAs (such as hsa_circ_0074854), and miRNAs (such as miR150) are implicated as critical signaling mediators that orchestrate macrophage M1/M2 polarization [111, 112, 113](Figure 4). Cholangiocarcinoma cells can also secrete multiple cytokines to guide macrophages into the tumor milieu. Tumor necrosis factor‐like weak inducer of apoptosis (TWEAK), a chemical messenger, and its receptor Fn14 are overexpressed in iCCA cohorts. TWEAK/Fn14 expression correlated positively with CAF proliferation. TWEAK derived from TAMs can bind to the Fn14 receptor on iCCA cells and increase the Monocyte chemoattractant protein‐1 (MCP‐1) secretion. MCP‐1 can recruit more monocytes to enter the tumor and transform into TAMs [114](Figure 4). In this section, we emphasized on how tumor cells regulate TAMs by releasing cytokines and exosomes. However, these studies are mostly limited to the animal level. Consequently, more robust evidence is required to facilitate the translation of basic research into clinical practice. For example, in a phase III clinical trial against melanoma, Talimogene laherparepvec (T‐VEC), a herpes simplex virus type 1‐derived oncolytic immunotherapy, could selectively replicate within tumors and recruit macrophages to enhance systemic antitumor immune responses [115].
Previous studies have highlighted the role of metabolic reprogramming in macrophage activation. This process of switching from a quiescent to an activated state can direct macrophage differentiation and regulate the function of these immune cells. In both mice and humans, glycolytic metabolism is involved in the classical activation of M1 TAMs, whereas mitochondrial oxidative phosphorylation is restricted to alternative activation of M2 TAMs [116]. Thus, glycolysis upregulation in specific macrophages may cause them to acquire an anti‐tumor phenotype. Nevertheless, macrophage metabolism is more complex than previously thought. Previous reports have shown that elevated glycolysis can regulate PD‐L1 levels and lead to M2‐type polarization [117, 118]. Recent data demonstrated a population of macrophages displaying an HLA‐DRhighCD86high PD‐L1+ glycolytic phenotype in HCC tumors. Intrinsic glycolytic metabolism confers PD‐L1+macrophages with anti‐tumorigenic properties [77]. Thus, targeting glycolytic metabolism in macrophages abrogated the PD‐L1‐mediated immune privilege. Tumor‐derived soluble factors, including hyaluronan fragments, can modulate glycolysis in peritumoral monocytes by up‐regulating 6‐phosphofructo‐2‐kinase/fructose‐2,6‐biphosphatase 3 (PFKFB3). This key glycolytic enzyme orchestrates cellular metabolism and induces PD‐L1 expression to attenuate cytotoxic T lymphocyte responses in tumor tissues [119]. Thus, there is an urgent need to understand the glycolytic status of macrophages and their precursor monocytes at different stages and to target critical glycolytic enzymes. In addition, the culture supernatant of HCC cells can activate the Wnt2b/β‐catenin/c‐Myc signaling pathway and augment the glycolysis of M2 TAMs [120]. The process can be inhibited by the TLR9 agonist, suggesting that targeting TLR9 might be a potential therapeutic strategy. Exosomal PKM2 derived from HCC cells induces metabolic reprogramming in monocytes, phosphorylates STAT3 and upregulates differentiation‐associated transcription factors. These data showed that PKM2 could promote monocyte‐to‐macrophage differentiation and remodel the TME [17]. The development of small‐molecule inhibitors similar to PKM2 may prevent monocyte differentiation into tumor‐promoting macrophages. Although essential, identifying tumor metabolites or vital glycolytic enzymes in macrophages would be a complex and challenging task. Lipid metabolic reprogramming in TAMs is indispensable for macrophage polarization and hepatocarcinogenesis. Downregulation of receptor‐interacting protein kinase 3 (RIPK3) in TAMs reduced ROS levels and promoted fatty acid metabolism via PPAR activation, contributing to the accumulation and polarization of M2 TAMs [18]. Fatty acid oxidation (FAO) blockade reverses the immunosuppressive activity of TAMs, which also appears to provide a potential strategy for inhibiting tumor progression in HCC. Several studies have provided mechanistic insights into amino acid metabolism in TAMs, and most of these studies have focused on how amino acid changes in tumor cells affect TAMs. For instance, an amine oxidase involved in extracellular matrix remodeling, lysyl oxidase‐like 4 (LOXL4), is upregulated during liver carcinogenesis in mice concomitantly fed a choline‐deficient, L‐amino acid‐defined diet. LOXL4 from neoplastic cells may facilitate macrophage infiltration into the liver and accelerate tumor growth. Glutamine metabolism is essential for macrophage activation, and α‐ketoglutarate (Αkg) production via glutaminolysis provides synergistic support for the activation [121]. Investigators in our hospital demonstrated that solute carrier family 7 member 11 (SLC7A11) led to intratumoral TAMs and MDSCs accumulation by increasing colony‐stimulating factor‐1 (CSF1) expression through Αkghypoxia‐inducible factor 1 subunit alpha (HIF1α) cascade [15]. Tryptophan‐derived microbial metabolites have been reported to mediate anti‐tumor immunity and activate aryl hydrocarbon receptors in TAMs, suggesting a potential therapeutic strategy [122]. How amino acid metabolism in macrophages affects the direction of polarization and tumor progression remains unclear; therefore, further research is warranted. In conclusion, extensive studies have been conducted on glucose metabolism, lipid metabolism and amino acid metabolism in TAMs. From the perspective of macrophage metabolism, M1 macrophages are characterized by glycolytic metabolism, iNOS expression, and proinflammatory cytokine production. The energy distribution of M2 macrophages is characterized by the increased expression of genes related to improved glucose and fatty acid uptake, transport, and oxidation. Modulating cellular metabolic remodeling to facilitate tumor progression is a novel and promising approach. Therefore, more reliable studies on the metabolic properties, dependencies, and adaptations of TAMs are warranted.
Most hepatocarcinogenesis is based on fibrotic or cirrhotic livers. Patients with HCC and severe cirrhosis usually have a worse prognosis and shorter median survival times. Increased liver stiffness plays a deleterious role in HCC progression. In addition, the degree of matrix stiffness is used to evaluate the histopathological characteristics of HCC [123]. The results of most related studies have shown that increased matrix stiffness can promote the macrophage polarization to the M2 phenotype, and the effects of biomechanical cues on HCC progression remain largely unexplored. Research has shown that a 3D gel‐like microenvironment induces adhesion and differentiation of monocytes via MAPK‐NF‐κβ activation [124]. Activation of the integrin β5‐FAK‐MEK1/2‐ERK1/2 pathway facilitates matrix stiffness‐mediated HIF‐1α and LOXL2 expression in polarized macrophages [125]. The high expression of Nogo‐B in TAMs of patients with HCC is closely correlated with yes‐associated protein 1(YAP)/ tafazzin (TAZ)‐mediated M2 polarization [126]. In summary, high matrix stiffness promotes cancer cell proliferation and resistance to chemotherapeutic agents, regulates angiogenesis, and enhances stemness. The effect of M2 polarization on HCC reflects the above processes. Thus, it is a very innovative point to explore how matrix stiffness affects macrophage polarization and functions, such as cytokine secretion and phagocytosis.
Communication between macrophages and other immune cells involves intricate exposure to different microenvironments. TAMs express an array of effector molecules that inhibit immune response in liver cancer. These immunosuppressing molecules include cytokines, chemokines, enzymes and cell‐surface receptors, which are mainly ligands and receptors expressed by the target immune effector cells (Figure 5).
TAMs can express HLA molecules, such as HLA‐G and HLA‐E, to disable cytotoxic antitumor immune response by interacting with the costimulatory T cells markers Ig‐like transcript 2 (ILT2) and CD94, respectively [127]. In addition to HLA molecules, macrophages can also express ligands for the inhibitory receptors PD‐1 and PD‐2, cytotoxic T‐lymphocyte antigen 4 (CTLA‐4), and T cell immunoglobulin and mucin‐containing molecule 3 (Tim‐3) [128, 129, 130]. These inhibitory ligands induce T cell apoptosis or functional inactivation. Selective CD28 blockade causes 2B4 upregulation in specific CD8+ T cells, which can facilitate the control of antigen‐specific CD8+ T cell responses and functions [131]. Studies on liver metastasis revealed that FasL+CD11b+F4/80+ MoMϕs can mediate apoptosis in activated antigen‐specific Fas+CD8+ T cells [132]. In addition to the molecules mentioned above, many undiscovered mechanisms can mediate direct interactions between macrophages and CD8+ T cells. Therefore, targeting surface markers between TAMs and CD8+ T cells is a promising strategy for reversing the immunosuppressive microenvironment in liver cancer treatment.
Generally, macrophages are influenced by tumor cells after entering the TME and are transformed into a state that promotes tumor growth and inhibits T cell function. Therefore, TAMs are one of the leading factors that induce T‐cell dysfunction. The TME is typically a low‐oxygen, low‐glucose, and high‐lactate milieu due to uncontrolled cancer cell proliferation and immune cell dysregulation. Some research has revealed that HIF1α induces increased expression of triggering receptor expressed on myeloid cells‐1 (TREM‐1) in TAMs in a hypoxic HCC environment. Furthermore, TREM‐1+ TAMs induced CD8+ T cell apoptosis and impaired cytotoxic functions by reducing the release of granzyme B and perforin [133]. Intriguingly, this process is independent of the PD‐L1/PD‐1 axis. In addition, TAMs have been found to express arginase‐1 and suppress the function of activated (but not resting) CD8+ T cells [134]. These results indicated that TAMs could also release small molecule secretions to indirectly repress CD8+ T cells. TAMs are major immunosuppressive components in liver cancer. Nevertheless, reprogramming of tumor‐surveillance phenotypes of macrophages can enhance CD8+ T cell activity. For instance, a low dose of type‐I IFN can effectively reconstitute MoMϕs into CD169high macrophages. These macrophages exhibited significantly enhanced phagocytic and CD8+ T cell activation [135]. In a concurrent study, microRNA‐206 facilitated the infiltration of CD8+ T cells by inducing M1 polarization [20]. In conclusion, future research should explore macrophage remodeling to prevent M2 TAMs from affecting the infiltration of CD8+ T cells. Moreover, attenuating the expression of PD‐L1 on the surface of macrophages is a significant strategy for CD8+ T cell accumulation. [19]. However, the mechanism whereby TAMs exacerbate CD8+ T cell suppression remains unclear. Recently, a study on other cancers demonstrated that the depletion of RNA N6‐adenosine methyltransferase in TAMs induced CD8+ T cell dysfunction [136]. Thus, targeting molecules in TAMs that cause CD8+ T cell suppression represents an attractive immunotherapeutic strategy.
Similar to TAMs, Treg cells are essential for maintaining a suppressive immune microenvironment and contributing to tumor immune escape. Previous findings demonstrated that Tregs could mediate fatty acid synthesis in M2 TAMs [137]. In addition, Tregs indirectly but selectively maintain metabolic fitness, mitochondrial integrity, and cell survival in M2‐like TAMs. Treg cells can mediate the metabolic remodeling of TAMs and influence the direction of macrophage polarization to promote tumor progression [137]. In contrast, CCL5, CCL20, and CCL22 derived from TAMs were responsible for Tregs induction [138, 139]. In epithelial ovarian cancer, exosomes from TAMs, such as miR‐29a‐3p and miR‐21‐5p, can mediate the interaction between TAMs and T ‐cells by suppressing the STAT3 signaling pathway [140]. Regardless of the type of cancer, TAMs and Tregs act as accomplices to maintain an intratumoral immunosuppressive microenvironment. TAMs and Tregs can also cooperate to maintain the immunosuppressive microenvironment [133]. Therefore, while targeting TAMs, Tregs also seem to be affected. However, the current understanding is insufficient, and further work in this area is merited.
Tumor‐infiltrating lymphocytes can secrete TNF‐α and IFNγ, which inhibit TAMs. In turn, IL‐10 and TGF‐β produced by M2 TAMs and Tregs are pivotal chemokines that block the differentiation and maturation of T‐cells, DCs and CD4+ T cells [141, 142]. Cytokines secreted by M1 TAMs contribute to the infiltration and activation of CTLs, NK cells, and CD4+ cells [143, 144]. The CSF1 receptor signaling pathway mediates CD11b+Gr‐1lowLy6Chigh MDSC infiltration while recruiting CD11b+F4/80+ TAMs. Therefore, TAMs and MDSCs usually appear together. Intriguingly, owing to the compensatory appearance of MDSCs, the elimination of TAMs alone cannot inhibit tumor progression in iCCA [145]. As C‐X‐C motif chemokine ligand 2 (CXCL2) is a known chemoattractant for MDSCs, the researchers observed significant upregulation of CAF‐derived CXCL2 in mouse liver cancer tumors when compared with adjacent liver using the chemokine array and quantitative polymerase chain reaction (PCR). Another potential mechanism of TAM blockade–mediated MDSC accumulation is due to their enhanced survival. A distinct apolipoprotein E (ApoE) MDSC subset was uncovered by single‐cell RNA Sequencing (RNA‐Seq) analysis. Both TAMs and MDSCs are immunosuppressive cells in the TME. Eliminating TAMs may lead to the compensatory emergence of MDSCs. ApoE as well as cathepsin D (ctsd) and cathepsin B (ctsb), which mediate MDSC death via interrupted autophagy and endoplasmic reticulum stress, were downregulated in MDSCs. Thus, ApoE agonist GW3965 combined with TAMs blockade have tumor‐suppressive effects in the murine liver cancer model [145]. Moreover, RGX‐104, an LXR/ApoE agonist, was shown to significantly decrease MDSC levels in patients in a phase I human clinical trial [146]. Therefore, the effectiveness of such combinations of MDSCs blockade and TAMs blockade in the treatment of human liver cancers requires further evaluation. In summary, M2 TAMs and other pro‐tumorigenic immune cells such as Tregs and MDSCs constitute the main drivers of the immunosuppressive microenvironment. They can directly or indirectly inhibit the function of immune cells. In contrast, M1 TAMs mediate T cell infiltration and enhance tumor immunity by reshaping the tumor immune microenvironment. Balance the interactions of macrophages with other immune cell types is key for treating malignant tumors.
Over the past decade, experimental and clinical research results have indicated that the destruction or redifferentiation of TAMs may be a viable therapeutic strategy for patients with liver cancer [45, 147, 148]. We have divided TAMs‐related therapeutic strategies into three approaches:1) cutting off the source and eliminating the production of M2 TAMs, 2) remodeling M2 TAMs to M1 TAMs, and 3) blocking communication between M2 TAMs and liver cancer cells (Figure 6). Furthermore, we summarize the most cutting‐edge research on these treatment strategies and provide sufficient and robust evidence to support promising strategies for limiting liver cancer growth and progression (Table 2).
Circulating monocytes are the primary source of infiltrating macrophages in tumors, and the accessibility of mononuclear cells in clinical practice (which enables a treatment strategy based on blocking the source of monocytes) is particularly significant. Blocking the CSF1/CSF1R axis is the most established method of reducing TAM survival [149]. This ligand‐receptor pair is indispensable for TAM differentiation and survival [150]. The CSFR1 inhibitors GW2580 and AZD7507 were shown to prevent the recruitment of peripheral monocytes in iCCA [151]. TAM blockade alone does not inhibit tumor progression, with a compensatory increase of granulocytic myeloid‐derived suppressor cells (G‐MDSCs). Combined treatment with antibodies against CSF1R and Ly6G or ApoE agonist GW3965 could reduce tumor volumes and increase the efficacy of anti‐PD‐1 in the YAP/AKT driven murine iCCA model. Along with the CSF1/CSF1R axis, a growing body of evidence indicates that the CCL2/CCR2 axis‐mediated macrophage infiltration is a potential immunotherapeutic target [152]. CCL2 derived from HSCs can promote macrophage accumulation and modulate TAM polarization. Blocking the CCL2/CCR2 axis with a CCR2 antagonist impaired the accumulation of blood inflammatory monocytes and suppressed murine liver tumor growth [153]. Furthermore, a natural CCR2 antagonist, 747, could recruit more CD8+ cytotoxic T cells and enhance the therapeutic effect of sorafenib by blocking tumor‐infiltrating macrophage‐mediated immunosuppression [154]. In cholangiocarcinoma, monocyte chemoattractant protein‐1(MCP‐1, also known as CCL‐2) directed the trafficking of CCR2+monocytes to the tumor niche, and an anti‐MCP‐1 antibody attenuated the aggregation of circulating macrophages [114]. C/EPBα, a transcription factor, can enhance the function of MDSCs and M2 macrophages. In a phase I/Ib multicenter clinical study, C/EBPα saRNA (MTL‐CEBPA) treatment played an indispensable role in hampering aggregation and reversing the suppressive activities of MDSCs and TAMs, and combination therapy with MTL‐CEBPA and a PD‐1 antibody significantly abrogated tumor progression [155]. Moreover, some researchers have used 10X Genomics single‐cell RNA sequencing technologies to determine that M2 TAMs infiltration was associated with iCCA progression and to measure the expression of macrophage markers and aPKCɩ in 70 human tumor samples at various stages. Furthermore, the study presented a novel combination therapy strategy with cationic liposome‐mediated co‐delivery of gemcitabine and aPKCɩ‐siRNA, which significantly attenuated macrophage recruitment [156]. None of the antibodies and inhibitors above can selectively act on M2 TAMs and affect the antitumor activity of M1 TAMs. Therefore, M2‐selective clearance antibodies and inhibitors need to be developed in the future.
Specific pro‐tumoral tissue‐resident macrophages accumulate close to tumor cells early during tumor formation and provide a pro‐tumorigenic niche and induce a potent regulatory T cell response that protects tumor cells from adaptive immunity [10]. In liver cancer, KCs represent the vast majority of tissue‐resident macrophages. In the presence of carcinogenic factors, these cells will gradually transform into TAMs to promote cancer progression. A previous report showed that combretastatin A‐1 phosphate (CA1P), a microtubule polymerization inhibitor, exerted remarkable antitumor activity against HCC cells and TAMs. CA1P could induce TAM apoptosis by inhibiting the Wnt/β‐catenin pathway and down‐regulating Treg levels. The authors also indicated that combination therapy with CA1P and sorafenib therapy was most likely to achieve treatment effects in drug‐resistant patients [157]. Although emerging clinical evidence and preclinical findings have revealed that sorafenib increases overall survival and improves outcomes in patients with HCC, macrophages may exert a pro‐tumor role under sorafenib treatment [22]. However, zoledronic acid combined with sorafenib could significantly inhibit tumor progression in two xenograft nude mouse models [158]. Other multi‐targeted kinase inhibitors, such as lenvatinib, have been reported to decrease the tumor macrophage population and increase the number of CD8+ cells [159]. That research also indicated that combination treatment with lenvatinib and an anti‐PD‐1 antibody notably enhanced antitumor activity. Recent research showed that treating rats with HCC with cerium oxide nanoparticles (CeO2NPs) reduced the macrophage infiltration into the liver, inhibited the expression of inflammatory response‐related genes, and improved the survival of Wistar rats [160]. Moreover, the macrophage scavenger liposomal clodronate (Lipclod) can selectively inhibit Wnt signaling and the growth of iCCA in bilateral xenografts of human cholangiocarcinoma cells SNU‐1079, CC‐LP‐1 or WITT1 [151]. Tissue‐resident macrophages play a pivotal role in shaping the liver tumor immune microenvironment, making them a crucial target in the prevention and treatment of early liver cancer lesions. Nonetheless, the lack of a comprehensive understanding of macrophage molecular and functional diversity makes modulating them extremely difficult. Therefore, using single‐cell sequencing to trace how such TAM lineages contribute to the TME and liver cancer progression should be explored in future research. [10, 59, 161, 162].
Preclinical experimental data suggest that targeting macrophage repolarization can be beneficial in cancer immune therapy [163]. Sorafenib is a classic multikinase inhibitor approved for treating systemic HCC. Recent data have shown that sorafenib inhibits the macrophage‐induced growth of hepatoma cells [164]. Sorafenib regulates the functions of M2 TAMs and hampers HCC growth mediated by the insulin growth factor (IGF)/insulin growth factor receptor (IGF‐R) signaling axis [165]. Further study on targeting macrophages with sorafenib has shown that the multi‐kinase inhibitors stimulate proinflammatory cytokine production in macrophages and activate hepatic NK cells in cytokine‐ and NF‐κB‐dependent manners [166]; providing robust evidence that tyrosine kinase inhibitors (TKIs) can be used to reverse macrophage polarization and induce anti‐tumor immune responses. RIPK3‐mediated lipid metabolic reprogramming is correlated with tumorigenesis, macrophage accumulation, and polarization in the tumor milieu. The RIPK3 inhibitor, GSK872, or FAO blockade, was shown to remodel the immune activity of TAMs and markedly abrogate tumor progression [18]. A natural compound, baicalin, has been reported to directly induce TAM reprogramming to M1‐like macrophages through autophagy‐associated activation of Reticuloendotheliosis viral oncogene homolog B (RelB)/p52. The authors demonstrated the tumor‐suppressive effect of baicalin in skewing macrophages away from M2‐like macrophages toward the M1‐like phenotype for treating HCC [108]. Recently, it was reported that blocking the CSF1/CSF1R signaling by PLX3397 reduced tumor growth and reduced M2 phenotype polarization [167]. Targeting miRNAs could also be an effective therapeutic strategy for regulating macrophage polarization. MiR‐98 may play a vital role in shifting the polarization of TAMs towards the M1 phenotype [168]. Norcantharidin, a common anticancer drug, modulates macrophage polarization through miR‐214 to exert an anti‐HCC effect [169]. Phosphoinositide 3‐kinase gamma (PI3Kγ) regulates the immunosuppressive properties of TAMs. PI3Kγ inhibitors, such as TG100‐115, can induce the expression of MHC‐II and proinflammatory cytokines, skewing macrophages to the M1 phenotype while reducing immunosuppressive molecules and inhibiting tumor progression [170]. Multiple stages of the immune response are mediated by the tumor necrosis factor superfamily (TNFSF) ligands and their receptors, such as the CD40/CD40L pair. CD40 functions as a crucial communication medium connecting innate and adaptive immunity and is a significant trigger for monocyte to differentiate into M1 macrophages and DC cells. The CD40‐mediated activation of macrophages and DCs in iCCA has been shown to ameliorate responses to checkpoint inhibitors [171]. TLR can trigger the release of pro‐inflammatory cytokines in response to bacteria or viruses infection. Intratumoral injection of TLR agonists can increase monocyte recruitment and infiltration and induce repolarization, preventing TAM polarization to the M2 phenotype [172, 173]. Clinical trials have also been conducted to study the efficacy of TLR agonists, such as RO7119929, against liver cancer. The infiltration of pro‐angiogenic TAMs that express the angiopoietin receptor Tie2 could serve as a diagnostic marker for HCC [97]. Transient expression of a dominant‐negative Tie2 ectodomain (sTie2) can block Ang2‐mediated Tie2 signaling and may be a promising approach for tumor treatment [174]. In addition, data from a study of other cancers showed that Ang2‐Binding and Tie2‐Activating Antibody (ABTAA) induced polarization of TAM toward an anti‐tumor phenotype and promoted tumor vascular normalization more effectively than treatment with Ang2‐Blocking Antibody (ABA) alone. These findings demonstrate that ABTAA is a promising therapeutic for inhibiting tumor growth and reorchestrating the TME [175]. Regorafenib is an oral small‐molecule TKI that potently blocks multiple protein kinases. Interestingly, Tie2 may be a tempting target for regorafenib. Preclinical data and some clinical findings have confirmed that regorafenib could effectively inhibit liver cancer progression [176].
Clinical trials have significantly advanced cancer treatment in adoptive cell therapy, such as chimeric antigen receptor T cell (CAR‐T) therapy, especially for hematologic neoplasms. Recent clinical trials of CAR‐M therapy have opened up a new field for using macrophages to treat solid tumors (Table 2). After modification via genetic engineering technology, macrophages can target specific antigens, such as CD19, CD22, and Her2, to recognize tumor cells [177]. CAR‐M cells can phagocytize tumor cells, secrete pro‐inflammatory cytokines to change the tumor immune microenvironment, present tumor antigens to T cells and activate the immune response [178]. Studies on solid tumors have revealed that highly active CD3‐based CARs on the surface of macrophages can phagocytose and destroy targeted tumor cells in a Syk‐dependent manner instead of soluble opsonizing factors [177, 179]. Ad5f35, a chimeric adenoviral vector, can induce a durable M1 phenotype after macrophage transduction in humanized mouse models. Intriguingly, this group of macrophages did not convert to M2 macrophages upon stimulation with cytokines or a tumor‐conditioned medium, which confirmed that CAR‐Ms could elicit an inflammatory microenvironment, improve antitumor T cell activity, and significantly abrogate solid tumor progression [177]. CAR‐M technologies are constantly evolving. The targets recognized by CAR‐M are primarily expressed in other cancer types. Nevertheless, due to the heterogeneity of liver cancer, exploring liver tumor cell‐specific targets and engineering macrophages in liver TME remains challenging. It is also necessary to pay attention to the therapy's potential off‐target toxicity and immunogenicity [180, 181, 182]. If the results of associated studies are to be translated into clinical practice, safer, more reliable and efficient CAR‐M technologies should be developed. In addition, more research is warranted to confirm whether CAR‐M combined with CAR‐T, multi‐targeted kinase inhibitors, and ICIs can enhance tumor inhibitory effects.
The relationships between tumor cells and macrophages are diverse. Many of the relationships involve the modulation of signaling pathways that contribute to homeostasis in non‐malignant organs. In addition, intervention with M2 TAMs should not affect macrophages not associated with malignancy. Mechanistic data have demonstrated that tumor‐derived adenosine, an ATP‐AMP metabolite, with autocrine GM‐CSF can accelerate suppressive tumor‐infiltrating macrophage proliferation via the PI3K/AKT pathway. Hence, using adenosine receptor antagonists or anti‐GM‐CSF antibodies to inhibit macrophage accumulation may be a novel immunotherapy strategy [14]. CD47 is an immunoglobulin‐like protein known to interact with its receptor on macrophages, SIRPα, and participates in phagocyte‐mediated tumor clearance [183].CD47 expression is modulated by several mechanisms. For instance, IL‐6 secreted by TAMs can upregulate CD47 expression in hepatoma cells through the STAT3 pathway [184]. Moreover, histone deacetylase 6 (HDAC6) mediates thrombospondin‐1 (TSP1) expression, and TSP1 binds the CD47 receptor to block the CD47‐SIRPα‐mediated anti‐phagocytosis of the macrophage in a spontaneous mouse HCC model [185]. Thus, blocking CD47‐SIRPα signaling is a potential strategy for enhancing the ability of macrophages to phagocytose tumor cells and inducing antitumor responses [184, 186‐188]. Combination therapy of CD47 blockade and other target inhibitors presents new insights to improve HCC treatment. Glypican‐3 (GPC3) is a characteristic antigen of hepatocellular carcinoma. Bispecific antibodies to CD47 and Glypican‐3 (GPC3) have more potent antitumor effects and lower toxicity than monotherapy in humanized mice [189]. Furthermore, a preclinical study showed that CD47 blockade with CD47 antibody is associated with a good prognosis. CD47‐blocking increased the time‐to‐progression of metastatic tumors and prolonged survival in a murine splenic injection model of hepatic micrometastatic pancreatic ductal adenocarcinoma [190].
Over the past decades, extensive research has brought forth the revolutionary idea that hepatic immune cell populations based on TAMs play central roles in initiating and perpetuating liver cancer. In the specific immune microenvironment of hepatic neoplasms, TAMs serve functions that differ from those in other organs, including the regulation of signal‐transduction mechanisms, the change of self‐metabolism patterns in response to the context of the internal environment, and the spectrum of secreted factors. Whether starting from the communication between TAMs and cancer cells, the corresponding metabolic changes, or the biological stress of TAMs themselves, there seems to be more room for exploring the field of TAMs in liver cancer. In clinical research, TAMs are closely related to postoperative tumor progression and treatment. Experts have concluded that the TAMs population is a significant clinical prognostic indicator. Many clinical trials are still exploratory and are in an early stage. Some studies, such as using in vitro‐cultured macrophages in patients with advanced cancer (who have failed standard therapy) have been conducted. After intravenous or intraperitoneal injection of these cultured macrophages into 15 patients, the disappearance of ascites and inflammatory reactions occurred in individual patients. It is noteworthy that there were no other adverse reactions except for low‐grade fever and abdominal discomfort. Although lacking significant regression of the primary tumor site and clinical efficacy, these studies provide valuable information for the development of human macrophage‐based therapies. Although preclinical and clinical studies on TAMs have provided encouraging results, the use of macrophages as a targeted therapy for liver cancer still faces some challenges. First, most TAM research has been limited to animal models. There is considerable heterogeneity between mouse models and humans in terms of pathogenesis and responses to drug therapy. To facilitate the introduction of TAMs applications in the clinical arena, it is necessary to explore inhibitors targeting human TAMs and the immune microenvironment in patients with liver cancer. Second, due to the diversity of the origin of macrophages and heterogeneity after differentiation, TAMs show diverse characteristics at different stages. It appears that using specific blocking agents, such as CCL2 antibodies alone, is insufficient to overcome liver malignancies. Additionally, pan‐macrophage therapeutic approaches may also have adverse effects, such as CSF1 blockade targeting all macrophages, leading to systemic toxicity. Therefore, determining an optimal time for treatment and more precise therapeutic targets of TAMs during ongoing treatment are tasks worth exploring. Third, the checkpoint blockade on the surface of macrophages is currently limited to PD‐L1. Several novel immune checkpoints are expressed on the surface of macrophages, such as SIRPα and Tim‐3. Additional preclinical and clinical trials are required to support other ICIs combined with classic therapies. Finally, eliminating TAMs seems to lead to the compensatory emergence of other immunosuppressive cells. Thus, TAM elimination would also need to compensate for other immunosuppressive cells, such as Tregs and MDSCs, which cause tolerance to targeted TAMs alone. In this review, we have elaborated on the origin of TAMs, the communication of TAMs with surrounding cells, and the latest treatment progress, which provides more options and substantial evidence for treating patients with liver cancer by targeting macrophages. In the future, pharmaceuticals targeting macrophages in the specific immune environment of the liver as well as more stable, safe and efficient immune combination therapy could facilitate the further development of immunotherapy for liver cancer.
The authors declare that they have no competing interests.
KC collected the related papers and was a major contributor to the writing of the manuscript. NC prepared the figures and tables. JHZ and XY drafted the paper and revised it critically for important intellectual content. HFL and WGZ provided the final approval for the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Not applicable.
Not applicable.
Not applicable. |
PMC9648395 | 36266736 | Tongxin Ge,Xiang Gu,Renbing Jia,Shengfang Ge,Peiwei Chai,Ai Zhuang,Xianqun Fan | Crosstalk between metabolic reprogramming and epigenetics in cancer: updates on mechanisms and therapeutic opportunities | 20-10-2022 | cancer,epigenetics,metabolic reprogramming,RNA epigenetics,therapy | Abstract Reversible, spatial, and temporal regulation of metabolic reprogramming and epigenetic homeostasis are prominent hallmarks of carcinogenesis. Cancer cells reprogram their metabolism to meet the high bioenergetic and biosynthetic demands for vigorous proliferation. Epigenetic dysregulation is a common feature of human cancers, which contributes to tumorigenesis and maintenance of the malignant phenotypes by regulating gene expression. The epigenome is sensitive to metabolic changes. Metabolism produces various metabolites that are substrates, cofactors, or inhibitors of epigenetic enzymes. Alterations in metabolic pathways and fluctuations in intermediate metabolites convey information regarding the intracellular metabolic status into the nucleus by modulating the activity of epigenetic enzymes and thus remodeling the epigenetic landscape, inducing transcriptional responses to heterogeneous metabolic requirements. Cancer metabolism is regulated by epigenetic machinery at both transcriptional and post‐transcriptional levels. Epigenetic modifiers, chromatin remodelers and non‐coding RNAs are integral contributors to the regulatory networks involved in cancer metabolism, facilitating malignant transformation. However, the significance of the close connection between metabolism and epigenetics in the context of cancer has not been fully deciphered. Thus, it will be constructive to summarize and update the emerging new evidence supporting this bidirectional crosstalk and deeply assess how the crosstalk between metabolic reprogramming and epigenetic abnormalities could be exploited to optimize treatment paradigms and establish new therapeutic options. In this review, we summarize the central mechanisms by which epigenetics and metabolism reciprocally modulate each other in cancer and elaborate upon and update the major contributions of the interplays between epigenetic aberrations and metabolic rewiring to cancer initiation and development. Finally, we highlight the potential therapeutic opportunities for hematological malignancies and solid tumors by targeting this epigenetic‐metabolic circuit. In summary, we endeavored to depict the current understanding of the coordination between these fundamental abnormalities more comprehensively and provide new perspectives for utilizing metabolic and epigenetic targets for cancer treatment. | Crosstalk between metabolic reprogramming and epigenetics in cancer: updates on mechanisms and therapeutic opportunities
Reversible, spatial, and temporal regulation of metabolic reprogramming and epigenetic homeostasis are prominent hallmarks of carcinogenesis. Cancer cells reprogram their metabolism to meet the high bioenergetic and biosynthetic demands for vigorous proliferation. Epigenetic dysregulation is a common feature of human cancers, which contributes to tumorigenesis and maintenance of the malignant phenotypes by regulating gene expression. The epigenome is sensitive to metabolic changes. Metabolism produces various metabolites that are substrates, cofactors, or inhibitors of epigenetic enzymes. Alterations in metabolic pathways and fluctuations in intermediate metabolites convey information regarding the intracellular metabolic status into the nucleus by modulating the activity of epigenetic enzymes and thus remodeling the epigenetic landscape, inducing transcriptional responses to heterogeneous metabolic requirements. Cancer metabolism is regulated by epigenetic machinery at both transcriptional and post‐transcriptional levels. Epigenetic modifiers, chromatin remodelers and non‐coding RNAs are integral contributors to the regulatory networks involved in cancer metabolism, facilitating malignant transformation. However, the significance of the close connection between metabolism and epigenetics in the context of cancer has not been fully deciphered. Thus, it will be constructive to summarize and update the emerging new evidence supporting this bidirectional crosstalk and deeply assess how the crosstalk between metabolic reprogramming and epigenetic abnormalities could be exploited to optimize treatment paradigms and establish new therapeutic options. In this review, we summarize the central mechanisms by which epigenetics and metabolism reciprocally modulate each other in cancer and elaborate upon and update the major contributions of the interplays between epigenetic aberrations and metabolic rewiring to cancer initiation and development. Finally, we highlight the potential therapeutic opportunities for hematological malignancies and solid tumors by targeting this epigenetic‐metabolic circuit. In summary, we endeavored to depict the current understanding of the coordination between these fundamental abnormalities more comprehensively and provide new perspectives for utilizing metabolic and epigenetic targets for cancer treatment.
Abbreviations PPP Pentose phosphate pathway PDA Pancreatic ductal adenocarcinoma ncRNAs non‐coding RNAs CpG cytosine‐guanine DNMT DNA methyltransferase TET Ten‐eleven translocation family proteins HAT Histone acetyltransferase HDAC Histone deacetylase SIRT Sirtuin KMT Histone lysine methyltransferase SAM S‐adenosyl methionine KDM Histone lysine demethylase LSD Lysine‐specific demethylase FAD Flavin adenine dinucleotide JHDM Jumonji C domain‐containing histone demethylase α‐KG α‐ketoglutarate CRC Chromatin remodeling complex ncRNA Non‐coding RNA miRNA MicroRNA lncRNAs Long non‐coding RNA circRNA Circular RNA NADPH Reduced nicotinamide adenine dinucleotide phosphate TCA cycle Tricarboxylic acid cycle SAH S‐adenosyl homocysteine NAD+ Nicotinamide adenine dinucleotide 2‐HG 2‐hydroxyglutarate Acetyl‐CoA Acetyl‐coenzyme A ACL ATP‐citrate lyase ACSS2 Acetyl‐CoA synthetase 2 PDC Pyruvate dehydrogenase complex LKB1 Liver kinase B1 SHMT2 Serine hydroxymethyltransferase 2 PHGDH Phosphoglycerate dehydrogenase NEPC Small cell/neuroendocrine prostate cancer PKCλ/ι Protein kinase C λ/ι MAT2A Methionine adenosyltransferase 2A NNMT Nicotinamide N‐methyltransferase 1MNA 1‐methylnicotinamide YTHDF2 YTH N6‐methyladenosine RNA binding protein 2 m6A N6‐methyladenosine SCC Squamous cell carcinoma OAADPR 2′‐O‐acyl‐ADP ribose FAO Fatty acid oxidation NAM Nicotinamide NAMPT Nicotinamide phosphoribosyltransferase NMNAT‐1 NMN adenylyltransferase 1 IDH Isocitrate dehydrogenase LDHA Lactate dehydrogenase A SDH Succinate dehydrogenase FH Fumarate hydratase GIST Gastrointestinal stromal tumor AML Acute myeloid leukemia EZH2 Enhancer of zeste homolog 2 BCAT1 Branched‐chain amino acid transaminase 1 HK2 Hexokinase 2 G6PD Glucose‐6‐phosphate dehydrogenase ROS Reactive oxygen species SETD2 SET domain‐containing 2 G9A Euchromatic histone‐lysine N‐methyltransferase 2 PSAT1 Phosphoserine aminotransferase 1 PSPH Phosphoserine phosphatase HCC Hepatocellular carcinoma ARID1A AT‐rich interacting domain‐containing protein 1A GLS Glutaminase GSH Reduced glutathione SMARCA4 SWI/SNF‐related, matrix‐associated, actin‐dependent regulator of chromatin, subfamily A, member 4 RISC RNA‐induced silencing complex ENO1 Enolase 1 PKM2 Pyruvate kinase isoform M2 CPT1 Carnitine palmitoyl transferase 1 PFKFB3 6‐phosphofructo‐2‐kinase/fructose‐2,6‐biphosphatase 3 LUAD Lung adenocarcinoma GOT1 Glutamic‐oxaloacetic transaminase GLUT1 Glucose transporter type 1 NPC Nasopharyngeal carcinoma PFK2 6‐phosphofructo‐2‐kinase MTHFD2 Methylenetetrahydrofolate dehydrogenase 2 ccRCC Clear cell renal cell carcinoma FTO Fat mass and obesity‐associated protein LDHB Lactate dehydrogenase B METTL3 Methyltransferase‐like 3 OXPHOS Oxidative phosphorylation ALKBH5 AlkB homolog 5 RNA demethylase RCC Renal cell carcinoma METTL14 Methyltransferase‐like 14 m5C 5‐methylcytosine SAMTOR SAM sensor upstream of mTORC1
Cellular metabolic reprogramming is a core hallmark of cancer [1, 2]. A large body of researches have tried to elucidate the direct effects of metabolism on tumor growth, proliferation, and metastasis. Highly proliferating cancer cells require numerous building blocks for active biosynthesis and an abundant energy supply. To meet the requirements for growth and survival, cancer cells experience significant metabolic alterations, such as upregulated glycolysis and enhanced glutamine catabolism. Oncogenic reprogramming of cellular metabolism is a downstream event of mutant oncogenes or tumor suppressors, dysregulated signal transduction pathways, and perturbed microenvironmental nutrient availability [3, 4, 5, 6]. Emerging researches suggest that metabolism is not merely a passive participant of tumorigenesis; it can serve as signaling molecules and globally control gene expression. Another general mechanism by which metabolism can modulate cellular activities has been proposed. Cellular metabolism provides a pool of intermediate metabolites acting as substrates, cofactors, agonists, or antagonists of chromatin‐modifying enzymes. Significant changes in the metabolic pool accompany the reprogramming of metabolism. Thus, it is reasonable to speculate that fluctuations in these metabolites could regulate the state and function of cells through epigenetic mechanisms. The hyperactive pentose phosphate pathway (PPP) promotes global epigenomic reprogramming and drives the evolution of distant metastasis in pancreatic ductal adenocarcinoma (PDA), providing robust evidence for this hypothesis [7]. The term “epigenetics” was defined as a “stably heritable phenotype resulting from changes in a chromosome without alterations in the DNA sequence” [2, 8]. Beyond oncogenic mutations, four classic epigenetic mechanisms, DNA methylation, histone modifications, chromatin remodeling, and non‐coding RNAs (ncRNAs), dynamically influence various chromatin‐related processes, such as gene transcription, DNA repair, and replication. The basic unit of chromatin is the nucleosome, which is assembled from a histone octamer consisting of H2A, H2B, H3, and H4, with 147 base pairs of DNA wrapped around the octamer [9]. Alterations in chromatin structure caused by epigenetic modifications and chromatin remodelers can change the transcriptional accessibility of regional DNA sequences, thus profoundly influencing gene expression. In human cancers, epigenetic modification profiles and ncRNA expression patterns often change globally [10, 11, 12, 13, 14]. Compelling evidence highlights that epigenetic reprogramming is crucial for the acquisition and maintenance of hallmark capabilities in cancer, including unlocking phenotypic plasticity and deregulating cellular metabolism [2, 15‐19]. Some studies have revealed that the interplay between epigenetics and metabolic reprogramming endows tumor cells with the capability to adapt to ever‐changing conditions during tumorigenesis. Most recently, many discoveries have been made. These findings will be discussed in detail later to provide more supporting evidence for this hypothesis. Additionally, with advances in the fields of cancer metabolism and epigenetics, several intriguing new themes have emerged. One key question is how metabolism tunes transcription through non‐canonical histone modifications like lactylation and succinylation. A second important question is whether a close interaction exists between metabolism and RNA epigenetics. Covering these themes will significantly deepen our understanding of this topic and provide fundamental insights into tumor biology. However, there are still some limitations existing in current studies. First, the causal link between the metabolic‐epigenetic loop and phenotypic outcomes in cancer has not been rigorously proven. That is to say, whether all these outcomes observed are directly caused by metabolically driven changes in epigenetic modifications needs further validation. Newly developed epigenome editing may enable us to confirm which chromatin marks have causal roles in determining tumor behaviors [20]. Second, metabolic and epigenetic heterogeneities within tumors are currently rarely taken into account. High‐throughput techniques, including spatial omics and single‐cell omics, may answer the question of how heterogenous metabolic and epigenetic patterns interweave with each other to amplify intra‐tumoral phenotypic diversity [21]. Cancer metabolism and epigenetics are both attractive therapeutic targets for cancer therapy, which is not surprising, given their important roles in cancer. Unfortunately, successful clinical applications of drugs targeting metabolism are rare. The efficacy of epigenetic drugs has been confined to hematological malignancies, and they are almost ineffective in solid tumors. This indicates the need to identify true metabolic or epigenetic vulnerabilities and develop new drug combinations. The robust association between metabolism and epigenetics has been revealed. It is thus rational to propose some potential treatment strategies targeting these communications (Figure 1).
The most classic example of metabolic reprogramming in cancer is the Warburg effect, also known as aerobic glycolysis. Cancer cells tend to convert pyruvate, the end product of glycolysis, into lactate rather than directing it into mitochondrial metabolism despite the intact function of oxidative phosphorylation (OXPHOS). This may be caused by the increased demand of cancer cells for macromolecule biosynthesis compared with energy production. The intermediate products of glycolysis can be diverted into biosynthetic programs such as serine metabolism, hexosamine pathway, and PPP [22, 23]. These metabolic branches, often deregulated, provide reduced nicotinamide adenine dinucleotide phosphate (NADPH) for reductive biosynthesis and combating oxidative stress and S‐adenosyl methionine (SAM) for methylation reactions and building blocks for proteins and nucleic acids [24, 25]. In addition, cancer cells can utilize intermediates of the tricarboxylic acid (TCA) cycle for de novo fatty acid and non‐essential amino acid synthesis. Researchers reported that cancer cells might be addicted to glutamine and glucose [26, 27]. Glutamine is involved in the synthesis of essential amino acids, purine bases, and pyrimidine bases. Further, glutamine can also be metabolized into α‐ketoglutarate (α‐KG) to replenish the TCA cycle in the mitochondria [28]. Beyond that, lipid metabolism also undergoes reprogramming in cancer. Cancer cells have active fatty acid and cholesterol synthesis that make up the membrane and form signaling molecules. Fatty acid oxidation is an important energy source for rapidly proliferating cancer cells [29, 30]. Altogether, metabolic reprogramming dramatically impacts many biological properties of cancer cells, such as fueling proliferation and growth and promoting invasion and distant metastasis [31, 32].
Epigenetic changes, including DNA methylation, histone modifications, chromatin remodeling, and ncRNAs, are closely related to cancer development and malignant progression. Here, we provide an overview of the basic principles of these epigenetic processes.
DNA methylation refers to the enzymatic addition of a methyl group to a cytosine 5‐carbon, which forms 5‐methylcytosine (5mC). It occurs mainly at scattered cytosine‐guanine (CpG) dinucleotide sites and some CpG islands, which are CpG‐rich sequences [33]. Nevertheless, CpG sites in CpG islands that overlap with the promoter regions of approximately two‐thirds of human genes are commonly unmethylated to maintain a permissive chromatin state for transcription [34]. In cancer, DNA methylation patterns are extensively reshaped with global hypomethylation but with regional hypermethylation of CpG islands in promoters of tumor suppressor genes [33, 35]. DNA methyltransferases (DNMTs) utilize SAM as the methyl group donor and are responsible for the deposition of methyl groups on C5 of cytosines [36]. DNMTs include two major categories: maintenance methyltransferase DNMT1 and de novo methyltransferases DNMT3A and DNMT3B. Ten‐eleven translocation (TETs) family proteins, including TET1, TET2, and TET3, have been demonstrated to be mammalian DNA hydroxylases for active DNA demethylation. TETs require oxygen and α‐KG as substrates and ferrous iron as cofactors to mediate demethylation reactions [37]. Specifically, 5mC is oxidized stepwise into 5‐hydroxymethylcytosine (5hmC), 5‐formylcytosine (5fC), and 5‐carboxylcytosine (5caC) during this process, followed by replication‐dependent dilution or base excision repair [38]. Of note, 5hmC represents both a demethylation intermediate and a stable epigenetic mark. Its abundance could reflect the function and activity of TETs [39, 40]. In cancer cells, global DNA hypomethylation revealed by genome‐wide analyses is the most prominent and earliest identified DNA methylation abnormality [41]. DNA hypomethylation, accompanied by the activation of transcription, repeats, transposable elements, and oncogenes, may contribute to increases in aneuploidies and genomic instability, which are hallmarks of cancer [42]. Furthermore, aberrant hypermethylation of CpG islands in the 5′ promoter regions of tumor suppressor genes in cancer cells can lock them into inactive states, silencing their expression. For example, RB, a well‐known tumor suppressor gene, was discovered to be downregulated by promoter CpG‐islands hypermethylation and promote oncogenesis [43, 44]. Such aberrant DNA methylation patterns were also observed in tumor suppressor genes like CDKN2A, MLH1, and CDH1 [45, 46, 47].
Each histone possesses a highly flexible N‐terminal tail enriched with lysine and arginine residues that can be extensively modified [48]. Covalent histone modifications include acetylation, methylation, acylation (e.g., lactylation, succinylation, and crotonylation), phosphorylation, SUMOylation, and citrullination. Some histone modifications can alter interactions between histones and DNA or can be recognized by specific binding proteins to impact chromatin compaction and regulate transcription processes [49, 50]. Histone acetylation can promote a more open chromatin state and increase chromatin accessibility for gene expression. Histone acetylation is dynamically established by histone acetyltransferases (HATs) and is removed by histone deacetylases (HDACs). There are three major groups of HATs: the GNAT family, the MYST family, and the orphan family. HATs transfer acetyl groups from acetyl‐coenzyme A (acetyl‐CoA) to histone lysine residues [51]. Four classes of HDACs were identified: class I (HDAC1‐3 and HDAC8), class II (HDAC4‐7 and HDAC9‐10), class IV (HDAC11), and class III (Sirtuin/SIRT1‐7). SIRTs require nicotinamide adenine dinucleotide (NAD+) as the cofactor [51, 52]. Some HDACs can also deacetylate nonhistone proteins [52]. Histone methylation occurs in the side chains of lysine, arginine, and histidine residues. Histone lysine methyltransferases (KMTs) can specifically transfer one, two, or three methyl groups from SAM to specific histone lysine residues to generate mono‐, di‐, or tri‐methylated (me1/2/3) histone [53]. There are two kinds of histone demethylases (KDMs). The family of amine oxidases (LSDs) utilizes flavin adenine dinucleotide (FAD) as a cofactor and is limited to demethylating mono‐ and di‐methylated lysine. Jumonji C (JmjC) domain‐containing histone demethylases (JHDMs) utilize ferrous iron and α‐KG and demethylate tri‐methylated lysine [54]. The functions of different histone methylations depend on the location and degree of methylation of lysine residues. Histone methylation plays an essential role in modulating transcription by changing the chromatin structure, recruiting chromatin remodeling factors, or guiding the binding of transcription factors [55]. In cancer cells, a genome‐wide profile revealed the loss of mono‐acetylation and tri‐methylation of histone H4 at a global level [56]. The discovery was subsequently confirmed in skin cancer, and the study suggested that the alteration occurs at the early stage and accumulates during carcinogenesis. With growing evidence supporting this discovery in multiple cancers, it was accepted as a common feature of cancer cells. These losses primarily appeared at the acetylated K16 and tri‐methylated K20 residues of histone H4 and were connected to the well‐described DNA hypomethylation in cancer [56, 57, 58]. In addition, certain combinations of histone modification are associated with extensive CpG island hypermethylation in cancer cells, including H3K9 methylation, H3K27 tri‐methylation, loss of H3K4 tri‐methylation, and deacetylation of histones H3 and H4 [59, 60]. Histone modifications promote tumor pathogenesis and evolution through transcriptional regulation that activates oncogene expression and represses tumor suppressor gene expression. For example, the enhancer of zeste homolog 2 (EZH2) binds to the promoter region of P21, a crucial tumor suppressor gene, and regulates its H3K27me3 modification, which promotes proliferation and tumorigenesis in gastric cancer [61].
Chromatin structure is dynamically regulated by DNA and histone modifiers and ATP‐dependent chromatin remodeling complexes (CRCs). CRCs contain four different families: switch/sucrose non‐fermentable (SWI/SNF), imitation switch (ISWI), chromodomain‐helicase DNA‐binding (CHD), and inositol‐requiring mutant 80 (INO80). CRCs can change the packaging state of chromatin, specialize in chromatin regions, and regulate chromatin accessibility through sliding, ejecting, or reorganizing nucleosomes [62, 63]. Components of the SWI/SNF complex are frequently and extensively mutated in various types of cancer; however, the mechanisms of CRCs mutations in tumorigenesis remain unclear [64]. The SWI/SNF family, composed of 8 to 14 subunits, was initially extracted from Saccharomyces cerevisiae. Eukaryotes usually employ two SWI/SNF family complexes with two relevant catalytic subunits. The family slides and ejects nucleosomes in various processes at many loci but is incapable of chromatin assembly [65]. The ISWI family comprises 2 to 4 subunits. Among the ISWI family, dNURF, dCHRAC and dACF complexes were initially extracted from Drosophila melanogaster and hWICH or hNoRC was subsequently recognized. Eukaryotes develop diverse ISWI family complexes by combining one or two catalytic subunits with specialized proteins [66]. Most ISWI family complexes, including ACF and CHRAC, promote chromatin assembly and transcriptional repression by improving nucleosome spacing [62]. The CHD family, among which Mi‐2 combines 1 to 10 subunits, was initially extracted from Xenopus laevis [67]. Some CHD family complexes promote transcription by sliding or ejecting nucleosomes, whereas others repress transcription, including the vertebrate Mi‐2/NuRD complex. The variability in CHD family complexes may result in chromodomain diversity [68]. The INO80 family, composed of more than 10 subunits, was first extracted from Saccharomyces cerevisiae. INO80 participates in DNA repair and transcriptional activation [69]. Notably, SWR1‐related complexes in the INO80 family reorganize nucleosomes by replacing canonical H2A‐H2B dimers with H2A.Z‐H2B dimers [70]. So far, the studies of chromatin remodeling in cancer have focused on SWI/SNF family. The sequencing of cancer genomes revealed high‐frequency mutations in various SWI/SNF family members in several hematological and solid malignancies, including SNF5, BRG1, MTA1 and ARID1A [71, 72, 73, 74, 75]. These members act as tumor suppressors, the mutations of which contribute to the development and maintenance of cancer. The mutations of these chromatin remodelers provided opportunities to change chromatin accessibility and protein complex topology, yielding oncogenic outcomes. Mutations in the SMARCB1 gene promote tumorigenesis in malignant rhabdoid tumors by preventing SWI/SNF complex interaction with typical enhancers and promoting remaining SWI/SNF subunits to induce gene expression at super‐enhancers [76]. In addition, the SWI/SNF family complexes interact with transcription factors by multiple lineage‐specific subunits to regulate differentiation. They also potentiate malignancy by damaging the balance between differentiation and self‐renewal. Moreover, SWI/SNF family complexes participate in cell motility, cell‐cycle progression, and nuclear hormone signaling [75].
ncRNAs are functional transcripts driven by non‐protein‐coding genomes. Among the ncRNA family, microRNAs (miRNAs), long non‐coding RNAs (lncRNAs), and circular RNAs (circRNAs) are relatively well studied in cancer. They functionally interact with each other and form a sophisticated regulatory network, finely regulating all fundamental biological processes in cells [77, 78]. MiRNAs are small ncRNAs containing about 22 nucleotides, biogenesis taking place through a multi‐step process involving the RNase III enzymes, Drosha and Dicer [79]. They inhibit post‐transcriptional gene expression by regulating mRNA translation into proteins and are estimated to mediate the translation of over 60% of protein‐coding genes. The inhibition is completed through mRNA degradation and the suppression of translation initiation [80]. MiRNAs participate in multiple biological processes, including development, proliferation, differentiation, and apoptosis. Some miRNAs mediate specific individual targets, while others function as major process controllers, simultaneously regulating multiple gene expressions [81]. LncRNAs, comprising the largest portion of the non‐coding transcriptome, are a heterogeneous group encompassing transcripts longer than 200 nucleotides and without protein‐coding capacity [82]. Although lncRNAs were considered to lack open reading frames or conserved codons in transcripts, the recent investigation suggested that some transcripts may produce small peptides [83, 84]. Compared to protein‐coding genes, lncRNAs are commonly expressed at a lower level but display more cell type‐specific expression patterns. Functions of lncRNAs are more complicated and varied than that of miRNAs, including transcriptional regulation, mRNA processing, and post‐transcriptional control [77]. CircRNAs are characterized by the covalent link of the 3′ and 5′ ends in forming single‐stranded continuous loop structures, reshaping RNA structure cognition [85]. They are more stable than liner ncRNAs, owing to the lack of exposed ends that are inclined to nucleolytic degradation and specific RNA folding. In addition, they are evolutionary conserved and abundant in eukaryotes [86]. Splicing and circularization of exons or introns are considered the initial genesis events of circRNAs. CircRNAs exert critical biological functions by serving as sponges to inhibit miRNAs, mediating protein functions or encoding peptides [87]. Growing evidence has revealed that the aberrant expression of ncRNAs is one of the hallmark features of cancers, and distinct ncRNA expression profiles exhibited between tumor cells and normal cells play a vital role in tumor progression and metastasis [88, 89]. Cancer‐associated miRNAs are commonly categorized into tumor suppressor miRNAs and oncogenic miRNAs. Well‐established tumor suppressor miRNAs involve miR‐145, miR‐34a, and the let‐7 family and well‐characterized oncogenic miRNAs include miR‐21 and miR‐155 [90]. Notably, some miRNAs exert dual functions. For example, miR‐200c constrains epithelial‐to‐mesenchymal transition (EMT) to inhibit metastasis initiation in cancer; however, it promotes distant metastasis in late‐stage cancers [91, 92, 93]. Notably, miRNAs can inhibit cell proliferation by targeting cell cycle regulatory genes and mediating the cell cycle. The significantly decreased global expression level of miRNAs was discovered in various tumor cells, leading to the disorder of miRNAs function and deprivation of cell cycle inhibition [94]. LncRNAs display cancer‐related expression profiles based on tumor‐specific features. Specifically, hypoxia is a major cause of cancer progression and chemotherapeutic resistance acquisition, leading to aberrant expression of several lncRNAs. LncRNA p21 is hypoxia‐responsive that develops a positive feedback loop with HIF‐1α to motivate glycolysis in cancer [95]. Upregulation of the hypoxia‐inducing lncRNA EFNA3 accelerates Ephrin‐A3 accumulation at the cell surface to promote tumor invasion and metastasis [96]. Widespread dysregulation of circRNAs has been discovered in multiple cancers, which is frequently accompanied by reduced global circRNA levels in rapidly proliferating cancer cells, indicating that many circRNAs act as tumor suppressors. However, individual circRNAs could be upregulated in cancer cells to promote oncogenesis because their slow generation and high stability guarantee their accumulation in non‐proliferative cells [97, 98, 99, 100].
Many metabolites serve as substrates or cofactors for chromatin‐modifying enzymes, and their cellular concentration ranges overlap with the kinetic parameters of these enzymes [101]. Therefore, the availability of these critical metabolites could influence the activities of chromatin‐modifying enzymes and, thus, the abundance of epigenetic modifications. However, chromatin remodelers are saturated with their substrate, ATP, because of the high intracellular ATP concentration. Their activities are thus generally unaffected by metabolic reprogramming [102]. We think these are general mechanisms explaining how metabolism controls epigenetics in cancer. Researches have revealed that metabolism could regulate tumor initiation, differentiation, proliferation, metastasis, and drug resistance through epigenetics. That is to say, these intricate interactions function in almost all stages of tumorigenesis, even before the malignant transformation. One representative example is that metabolic regulation of the epigenome drives tumorigenesis in posterior fossa A ependymomas. Hypoxia induces metabolic reprogramming, significantly decreasing SAM levels while increasing α‐KG and acetyl‐CoA levels. The perturbations of these three key metabolites attenuate the substrate availability of H3K27 methyltransferases, promoting the activity of H3K27 demethylases, and fueling H3K27 acetyltransferases. Collectively, these changes lead to a unique epigenetic landscape characterized by H3K27 hypomethylation and hyperacetylation [103]. How the aforementioned key metabolites, along with other primary metabolites, build a bridge between aberrant metabolism and the epigenome in cancer will be discussed in detail below. We have also gained some new insights into cancer metabolism beyond conventional wisdom. First, cancer metabolism is subcellularly compartmentalized, which allows metabolites to participate in many distinct biological processes [104]. Several metabolic intermediates, such as acetyl‐CoA and NAD+, can be produced in the nucleus. Recent research has shown that almost all TCA cycle‐associated enzymes exist in the nucleus, forming a local metabolic pool [105]. Thus, the concentration of these metabolites is regulated by, but relatively independent of, mitochondrial and cytoplasmic metabolism. This represents an additional mechanism that tumor cells can exploit to regulate chromatin. Second, newly identified histone post‐translational modifications, such as histone lactylation and succinylation, are also metabolically sensitive [106, 107]. They orchestrate two of the most important metabolic pathways, glycolysis and TCA cycle, and epigenetic transcriptional responses. To delve further into these histone modifications will be very interesting.
Acetyl‐CoA is a crucial metabolite in many cellular compartments. It is mainly produced by pyruvate oxidative decarboxylation, fatty acid β‐oxidation, and branched amino acid catabolism in the mitochondrial matrix [108]. Acetyl‐CoA cannot penetrate the mitochondrial membrane directly. Instead, it forms citrate with oxaloacetate in the TCA cycle, which can be transported into the cytosol and decomposed to acetyl‐CoA by ATP‐citrate lyase (ACL) [109]. Acetate metabolism catalyzed by acetyl‐CoA synthetase 2 (ACSS2) is an alternative source of cytosolic acetyl‐CoA [108]. Histone acetylation relies on the acetyl‐CoA synthesis and can be dynamically regulated by fluctuating concentrations of cellular acetyl‐CoA derived from glucose and lipids under physiological conditions [110, 111, 112, 113]. Metabolic reprogramming could alter the ratio of acetyl‐CoA to coenzyme A and subsequently affect histone acetylation states in cancer cells. AMPK is responsible for promoting glycolysis and the TCA cycle in leukemia. AMPK promotes the production of acetyl‐CoA, which maintains global histone acetylation to facilitate the expression of leukemogenic transcription factors [114]. The PI3K/AKT pathway is activated in human prostate cancer and gliomas. AKT activity correlates with histone acetylation levels in clinical samples. KRAS mutations promote acetyl‐CoA production and histone acetylation by phosphorylating ACL and enhancing glucose uptake in an AKT‐dependent manner [115]. AKT‐induced ACL and histone acetylation are also required for acinar‐ductal metaplasia and pancreatic tumorigenesis. Reduced acetyl‐CoA levels caused by ACL ablation impair early pancreatic tumorigenesis [116]. The ACL is augmented in melanomas. ACL regulates MITF transcription and promotes melanoma growth through P300‐mediated histone acetylation. Targeting ACL increases the sensitivity of MAPK inhibition in BRAF‐mutant melanoma [117]. ACL is essential for maintaining global histone acetylation, whereas ACSS2 can compensate for acetyl‐CoA levels in a dose‐dependent manner when ACL is deficient [118]. Acyl‐CoA thioesterase 12 (ACOT12) could hydrolyze acetyl‐CoA into acetate and coenzyme A. Downregulated ACOT12 increases acetyl‐CoA abundance along with histone H3 acetylation levels in hepatocellular carcinoma (HCC), which epigenetically promote EMT and metastasis [119]. Reprogrammed lipid metabolism is involved in controlling cell state transitions. Enhanced fatty acid oxidation (FAO) contributes to acquiring a mesenchymal phenotype in breast cancer cells by producing acetyl‐CoA to maintain histone acetylation on the promoters of genes associated with EMT [120]. These acetyl‐CoA‐producing enzymes are also located in the nucleus, locally regulating histone acetylation. DNA damage signaling promotes nuclear ACL phosphorylation. Phosphorylated ACL produces acetyl‐CoA locally and promotes histone acetylation at double‐strand break sites, thereby recruiting BRCA1 and favoring homologous recombination repair. These results indicate that acetyl‐CoA production by ACL is spatially and temporally controlled [121]. Growth factors or mitochondrial dysfunction augment pyruvate dehydrogenase complex (PDC) translocation from the mitochondria into the nucleus during the S phase. The nuclear PDC generates acetyl‐CoA and promotes the acetylation of H3K9 and H3K18, which supports S phase progression [122]. In Pten deficient prostate tumors, PDC has a strong nuclear localization. The nuclear PDC regulates H3K9ac and thus affects the expression of lipid synthesis genes [123]. This is an alternative way to generate acetyl‐CoA for histone acetylation in addition to ACL. However, it is astonishing that silencing ACL and PDC affect different sites of acetylation [122, 123]. Under stress conditions, such as nutrient deprivation or hypoxia, acetyl‐CoA generated from glucose is markedly reduced. Specific subsets of cancer cells may be addicted to utilizing acetate as an alternative carbon source for maintaining acetyl‐CoA production, which is mediated by ACSS [124, 125, 126]. Acetate can restore histone acetylation at H3K9, H3K27, and H3K56. Increased histone acetylation at FASN and ACACA promoter regions promotes de novo lipid synthesis [127]. However, the proportion of exogenous acetate‐derived acetyl‐CoA used for histone acetylation is relatively low compared to the amount flowing into mitochondrial metabolism and lipogenesis [128, 129]. Under metabolic stress, ACSS2 translocates to the nucleus and maintains cell survival and growth by promoting H3 acetylation at the promoter regions of lysosomal biogenesis and autophagy‐related genes. The acetate needed for nuclear ACSS2 to produce acetyl‐CoA is generated by histone deacetylation [128]. Nuclear ACSS2 maintains histone acetylation by acetate recapturing, which could explain how cancer cells balance the need for acetyl‐CoA and the lack of nutrition [128, 129]. SAM is synthesized from methionine and ATP during the methionine cycle, which is essential for one‐carbon metabolism [130]. Serine and other amino acids, such as glycine and threonine, are the major one‐carbon unit donors of one‐carbon metabolism [24, 131]. Serine can also contribute to SAM production by supporting de novo ATP synthesis to offer adenosine beyond providing one‐carbon units for remethylating homocysteine [132]. The methylation status is modulated by cellular SAM levels tuned by one‐carbon metabolism [133, 134]. Cancer cells are addicted to serine, which contributes to nucleotide synthesis, methylation, and antioxidant activity. Liver kinase B1 (LKB1) mutation synergizes with KRAS activation to potentiate glycolysis and serine metabolism, which favors SAM production. Elevated SAM generation alters the epigenetic landscape of DNA methylation and dynamically supports retrotransposon methylation and transcriptional silencing. However, it seems to have little effect on histone and RNA methylation levels [135]. SHMT2, the gene encoding serine hydroxymethyltransferase 2 (SHMT2) in serine catabolism, is frequently amplified in B‐cell lymphomas. SHMT2 is responsible for converting serine into glycine and contributes a one‐carbon unit to the folate cycle. Overexpressed SHMT2 changes the DNA methylation state globally and epigenetically silences tumor suppressor genes in lymphoma [136]. Phosphoglycerate dehydrogenase (PHGDH), the critical enzyme in the de novo serine synthesis pathway, directs glycolytic flux into the one‐carbon metabolic network. Upregulated PHGDH increases metabolite levels in the methionine cycle and promotes histone methylation [137]. Small cell/neuroendocrine prostate cancer (NEPC), which is highly aggressive, has a distinct DNA methylation profile from that of adenocarcinoma during differentiation. Protein kinase C λ/ι (PKCλ/ι) deficiency increases the one‐carbon metabolism through the mTORC1/ATF4/PHGDH axis to fuel DNA methylation, which promotes NEPC differentiation [138]. Methionine metabolism can also alter SAM and SAH concentrations, thus quantifying histone methylation. Methionine restriction leads to decreased H3K4me3 at promoters and the expression of colorectal cancer‐associated genes [134]. Cancer stem cells depend highly on methionine because of their high SAM consumption rate. Inhibition of the key enzyme, methionine adenosyltransferase 2A (MAT2A), in the methionine cycle ablates histone methylation in cancer stem cells, which impairs their tumor formation ability and resistance to cisplatin [139]. Deregulation of nicotinamide N‐methyltransferase (NNMT) could alter the epigenetic state by consuming methyl units into 1‐methylnicotinamide (1MNA), which consequently attenuates the SAM/SAH ratio. Deregulated NNMT is found in many different tumors and supports tumorigenesis by selectively reducing the histone methylation of several specific genes and increasing their expression [140]. Other metabolites can also act as substrates for histone modifications [141]. Evidence of the role of these histone modifications in cancer continues to emerge. Lactate is a product of the Warburg effect and is a key metabolite and signaling molecule. It plays essential roles in multiple biological processes during tumor progression, such as angiogenesis, immune escape, and cell proliferation [142]. However, their role in chromatin modification has long been overlooked. Recently, researchers have found that histone lactylation derives from lactate and could contribute to gene expression [143, 144]. Active glycolysis provides sufficient lactate for lactylation in ocular melanomas. H3K18la is enriched in YTH N6‐methyladenosine RNA binding protein 2 (YTHDF2) promoter regions and promotes the transcription of YTHDF2. As an N6‐methyladenosine (m6A) reader, YTHDF2 binds to the m6A sites of PER1 and TP53 mRNAs for degradation [145]. Lactylation provides new insight into the Warburg effect and requires further investigation [146].
α‐KG is an intermediate in the TCA cycle and is produced from isocitrate by isocitrate dehydrogenase (IDH). α‐KG is the co‐substrate for a class of dioxygenase enzymes such as JHDMs, TETs, and prolyl hydroxylase [147]. In human pluripotent stem cells, α‐KG induces histone and DNA demethylation and promotes differentiation [148]. It can be presumed that α‐KG has an important role in regulating epigenomic plasticity. The same mechanism could explain the antitumor effects of α‐KG. In PDA, p53 inactivation leads to reduced α‐KG levels by rewiring the glucose and glutamine metabolism, which impairs TETs activity. This causes tumor cells to gain the characteristics of poor differentiation and high aggressiveness [149]. When exogenous serine is abundant, squamous cell carcinoma (SCC) cells show enhanced mitochondrial pyruvate metabolism and prevent NAD+ regeneration by reducing pyruvate to lactate. Limited NAD+ is not conducive to serine synthesis. Thus, SCC cells inhibit the de novo serine synthesis pathway, resulting in the accumulation of the byproduct, α‐KG. Decreased α‐KG inhibits histone demethylases and H3K27me3 demethylation, which blocks cancer stem cells from differentiating. This feature maintains the stemness of tumor stem cells and promotes tumor initiation [147]. Glutamine replenishes the TCA cycle to produce α‐KG [150]. Increased consumption of glutamine leads to local glutamine deficiency in tumor core regions. Hypermethylation of histones caused by decreased glutamine and α‐KG levels causes cancer cell dedifferentiation and BRAF inhibition resistance [151]. Glutamine supplementation increases the downstream metabolite, α‐KG. An increase in α‐KG concentration could suppress the oncogenic pathway in melanoma by decreasing global H3K4me3 levels and affecting H3K4me3‐dependent transcription [152]. However, the role of glutamine in cancer remains unclear. KRAS‐mutant colorectal cancer cells show increased reliance on glutamine. Mutant KRAS promotes glutaminolysis and succinate, fumarate, and malate accumulation in the TCA cycle, whereas the level of α‐KG decreases. Downregulation of α‐KG to succinate ratio inhibits the activities of demethylases and impacts genome‐wide DNA and histone methylation. Aberrant methylation patterns activate WNT/β‐catenin signaling and increase tumor stemness [153]. NAD+ is a co‐enzyme that mediates oxidation‐reduction (redox) reactions in many metabolic pathways, including glycolysis, TCA cycle, OXPHOS, and FAO. NAD+ regulates cell metabolism, redox homeostasis, genome stability, and histone modifications [154]. SIRTs remove acyl groups from lysine residues and transfer NAD+ into 2′‐O‐acyl‐ADP ribose (OAADPR) and nicotinamide (NAM) [155]. SIRTs can sense NAD+ levels, and their activity may be modulated by cellular concentrations of NAD+ and NAM [156, 157]. The metabolic switch from FAO to glycolysis decreases NAD+ concentration and inhibits SIRT1, thereby blocking H4K16 deacetylation in skeletal muscle stem cells. This directly shows that metabolic reprogramming can rewrite the epigenetic state through NAD+ [156]. For breast cancer cells, nicotinamide phosphoribosyltransferase (NAMPT) and NMN adenylyltransferase 1 (NMNAT1) regulate specific gene expression in a SIRT1‐dependent way. As the key enzymes of the NAD+ salvage pathway, NAMPT and NMNAT1 regulate NAD+ concentration and SIRT1 deacetylation activity, thus affecting H4K16ac levels at gene promoters. SIRT1 can recruit NMNAT1 to target gene promoter regions, creating a locally high NAD+ concentration to control SIRT1 activity [158]. In melanoma, the BRAF/ERK/STAT5 pathway transcriptionally regulates NAMPT expression. Overexpressed NAMPT changes the histone modification landscape and allows melanoma cells to switch to a more invasive phenotype associated with resistance to targeted therapies and immunotherapies [159].
In cancer cells containing mutated metabolic enzymes, 2‐hydroxyglutarate (2‐HG), fumarate, and succinate may be produced and accumulate [160]. It is worth noting that 2‐HG is chiral and exists as the two isoforms, D2‐HG and L2‐HG. These two enantiomers are differentially upregulated in distinct tumor contexts. These abnormal metabolites mix into the metabolic pool and competitively inhibit the activity of α‐KG‐dependent dioxygenases, such as multiple histone demethylases and the TET family of 5‐methylcytosine hydroxylases, because of their similar structure to α‐KG [161, 162]. They are also called oncometabolites because their aberrant accumulation can promote malignant transformation [160]. For example, IDH1/2 encodes isocitrate dehydrogenase 1/2, which usually catalyzes the oxidative decarboxylation of isocitrate to α‐KG. Mutated IDH1/2 gains the function of producing 2‐HG, specifically the D enantiomer, from α‐KG [163, 164]. Emerging evidence indicates that elevated 2‐HG levels could alter global histone and DNA methylation patterns and drive tumorigenesis in leukemia and glioma [165, 166, 167]. Impaired histone and DNA demethylation are associated with blocked cell differentiation [16, 168‐173]. For example, IDH2 mutation impairs the differentiation potential of multipotent cells and endows them with the ability to escape contact inhibition. IDH mutations are sufficient to promote malignant transformation and generate poorly differentiated sarcomas [174]. IDH mutations also cause genome‐wide DNA hypermethylation at the cohesin‐ and CTCF‐binding sites. Decreased CTCF binding widely compromises chromosomal topology and results in oncogenes like PDGFRA aberrant activation through interaction with distant enhancers [175]. IDH mutations alter cell metabolism and DNA repair through epigenetic mechanisms. Mutant IDH silences lactate dehydrogenase A (LDHA) by increasing promoter methylation [176]. D2‐HG increases repressive histone methylation marks at the ATM promoter, resulting in impaired DNA damage repair and self‐renewal of hematopoietic stem cells (HSCs) [177]. There are some similar findings in gliomas and acute myeloid leukemia (AML) that IDH1/2 mutations induce homologous recombination defects and sensitize tumor cells to poly (ADP‐ribose) polymerase inhibition [178]. Besides, mutant IDH produces D2‐HG and epigenetically suppresses the expression of interferon γ response genes, which impedes immune response in cholangiocarcinoma [179]. Under physiological conditions, the L enantiomer of 2‐HG is produced by LDHA and malate dehydrogenase 1 and 2 (MDH1/2) in response to hypoxia stress [180, 181, 182]. It has a far more potent inhibitory effect on α‐KG‐dependent dioxygenases than the D enantiomer [161, 162]. L2‐HG, rather than D2‐HG, mainly elevates in renal cell carcinoma (RCC) due to reduced expression of L2‐HG dehydrogenase (L2HGDH), which can convert L2‐HG back into α‐KG to avoid the accumulation of L2‐HG. Consistently, accumulation of L2‐HG reduces DNA 5hmC and increases repressive trimethylated histone marks like H3K9me3 and H3K27me3 [183]. Restoring L2HGDH can stunt tumor growth [184]. In addition to IDH, mutations in succinate dehydrogenase (SDH) and fumarate hydratase (FH) have been identified. They may share the same oncogenic mechanism. FH and SDH mutants lose their enzymatic activities and lead to fumarate and succinate accumulation, inhibiting α‐KG‐dependent dioxygenases [185]. SDH‐mutant gastrointestinal stromal tumors (GIST), paragangliomas, and FH‐mutant renal cell carcinomas show characteristic hypermethylation patterns [186, 187, 188, 189, 190]. In paraganglioma, hypermethylated and downregulated genes are involved in chromaffin cell differentiation and EMT [187]. Consistent with findings in IDH‐mutant gliomas, abnormal DNA methylation at CTCF sites in SDH‐deficient GIST compromises FGF and KIT insulators, reorganizes chromosome topology, and allows super‐enhancers to interact with and activate oncogenes [191]. Fumarate and succinate accumulation suppresses homologous recombination DNA repair by inhibiting KDM4A and KDM4B and makes tumor cells vulnerable to PARP inhibitors [192, 193]. When it comes to the mutation of enzymes in the TCA cycle, another essential and ubiquitous post‐translational modification, succinylation, is also affected. Succinyl‐CoA, the substrate of succinylation reaction, is mainly generated from the TCA cycle. Histone succinylation can be mediated both enzymatically and non‐enzymatically. KAT1 and KAT2A are responsible for depositing histone succinylation marks, whereas SIRT5 and SIRT7 are histone desuccinylases [194, 195, 196]. Histone succinylation is generally associated with transcriptional activation and broadly regulates the expression of tumor‐related genes [197, 198, 199, 200]. KAT2A interacts with the α‐ketoglutarate dehydrogenase (α‐KGDH) complex in the nucleus. α‐KGDH complex locally catalyzes succinyl‐CoA production and fuels KAT2A‐mediated H3K79 succinylation, which induces gene expression and promotes tumor growth [197]. In IDH1/2‐mutated gliomas, inhibition of SDH and subsequent accumulation of succinyl‐CoA are caused by D2‐HG, which foster widespread histone and nonhistone protein hypersuccinylation in different cellular compartments. Although hypersuccinylation induced by oncometabolites preferentially impacts mitochondrial metabolism, it also profoundly affects chromatin [201]. SDH loss selectively perturbs genome‐wide chromatin succinylation in promoter regions. Genes involved in transcriptional regulation and RNA processing are most affected [202]. However, many tumors, including esophageal squamous cell carcinoma (ESCC), are globally hyposuccinylated. It suggests that the functions of histone succinylation are context‐dependent [203]. Limited researches have provided a glimpse into how succinyl‐CoA is used explicitly by the tumor to alter the epigenetic chromatin state. Further detailed studies are urgently needed to unravel this important link (Figure 2). Aberrant epigenetic modifications have previously been attributed to mutation and abnormal expression of epigenetic enzymes. Cellular metabolism, which provides substrates, cofactors, and oncometabolites for epigenetic enzymes, also dynamically affects the epigenetic landscape. This fundamental process is precisely controlled under normal circumstances. However, these “molecular signals” can be excessive, insufficient, and even erroneous in cancer. Merely inhibiting a specific metabolic pathway or epigenetic enzyme will activate compensating mechanisms. It is conceivable that resistance to monotherapies is almost inevitable. The results presented above provide the molecular bases for the necessity of targeting the intersections between metabolism and epigenetics in cancer. Simultaneously targeting both upstream and downstream epigenetic enzymes of the metabolic‐epigenetic axis may achieve much more significant and durable responses. In addition to being confirmed in preclinical studies, this concept has exhibited promising clinical results in treating leukemia. IDH‐mutant leukemia possesses a hypermethylated phenotype. Although hypomethylating agents and IDH inhibitors have been approved by authorities and improved the clinical outcomes of AML patients, drug resistance invariably occurs. Blocking the source of 2‐HG (IDH inhibitor, ivosidenib) coordinates synergistically with the inhibition of DNA methyltransferase (DNMT inhibitor, azacytidine) in patients unable to receive intensive induction chemotherapy. Combined therapy significantly improved drug responses, event‐free survival, and overall survival compared to azacytidine monotherapy. Toxic effects were durable. These important findings may eventually offer a new treatment option to AML patients with IDH mutations [204, 205].
Genetic and epigenetic alterations actively participate in the metabolic reprogramming of cancer. For example, oncogenic Kras mutations selectively rewire glucose metabolism to promote pancreatic tumor growth [3]. Compared with genetic mutations, epigenetic regulations are reversible and variable. Epigenetic modifiers modulate metabolism by directly changing the transcriptional activities of metabolic enzymes or proteins in metabolism‐related signaling pathways according to the needs of tumor cells. Increased histone and DNA methylation mark transcriptionally repress fructose‐1,6‐biphosphatase (FBP1), which triggers the reprogramming of glucose metabolism to sustain cancer stem cell‐like properties in breast cancer cells [206]. The roles of ncRNAs in regulating metabolic reprogramming are much more complicated, involving both transcriptional and post‐transcriptional regulations. Exploring the epigenetic roles of ncRNAs in regulating metabolism will dramatically expand the list of drug targets. Although studies are emerging, there remain important unanswered questions. One outstanding issue is how these epigenetic processes are coordinated to promote tumor development by regulating metabolism. Here, we introduce the four pivotal epigenetic mechanisms and discuss their contributions. Given that many recurrent mutations in epigenetic regulators have been identified as cancer driver mutations, their roles in promoting cancer metabolism will be highlighted.
Abnormal methylation of promoter DNA occurs in metabolic enzymes. The TET3 protein is often upregulated in AML cells. TET3 induces the expression of glucose metabolism‐related genes by depositing 5hmC marks on their promoters [207]. Hypomethylation of the promoter contributes to the upregulation of hexokinase 2 (HK2) in liver cancer and glioblastoma. Enhanced HK2 levels promote increased glycolytic flux [208, 209]. DNMT1 downregulates FBP1 in basal‐like breast cancer by binding and methylating the FBP1 promoter, inhibiting gluconeogenesis and enhancing cancer cell glycolytic rates [206]. The glucose transporter (GLUT) plays an essential role in glucose metabolism in cancer. Elevated GLUT promotes glucose access to tumor cells and facilitates aerobic glycolysis. Consequently, lactate and pyruvate, metabolites of aerobic glycolysis, acidify the tumor microenvironment and increase tumor proliferation and invasion. Promoter hypermethylation causes the inactivation of DERL3, a crucial regulator of the endoplasmic reticulum‐associated protein degradation pathway, which enhances the expression of GLUT1 and promotes aerobic glycolysis. This is mediated by DNMT1 and DNMT3B [210]. In addition, elevated CAV‐1 expression by hypomethylation of the promoter CpG site upregulates GLUT3 transcription, stimulates glucose uptake, and increases aerobic glycolysis [211].
Loss of histone methyltransferase EZH2 synergizes with oncogenic NRAS mutations to accelerate leukemic transformation in myeloid neoplasms. In terms of mechanism, EZH2 epigenetically silences branched‐chain amino acid transaminase 1 (BCAT1) and disturbs branched‐chain amino acids (BCAAs) metabolism in hematopoietic stem/progenitor cells (HPSCs). Loss of EZH2 abolishes promoter repression and activates enhancers of BCAT1, leading to the accumulation of BCAAs and the subsequent activation of mTOR signaling in leukemia‐initiating cells [212]. The histone methyltransferase KMT2D is frequently mutated in lung cancer. KMT2D deficiency promotes lung tumorigenesis and upregulates glycolysis by impairing super‐enhancers of PER2 [213]. In melanoma, KMT2D loss causes genome‐wide reduction of H3K4me1‐marked active enhancer chromatin states and subsequently activates IGF1R/AKT to increase glycolysis [214]. KMT2D is transcriptionally repressed and mutated in pancreatic cancer. KMT2D repression promotes a metabolic shift to glycolysis and alters the cellular lipid profile of pancreatic cancer cells, which provides energy for cell proliferation [18]. Overexpressed histone methyltransferase NSD2 establishes H3K36me2 marks at the promoters of genes associated with glucose metabolism to upregulate the expression of HK2, glucose‐6‐phosphate dehydrogenase (G6PD), and TIGAR in breast cancer. As a result, glucose flux through PPP and NADPH production is upregulated to alleviate reactive oxygen species (ROS) and promote drug resistance [215]. Mutation and activation of histone methyltransferase SETD2 are frequently observed in renal cancer. SETD2‐deficient cancer cells exhibit enhanced OXPHOS and fatty acid synthesis [216]. The histone H3K9 methyltransferase G9A (KMT1C) is elevated in many types of cancer and promotes tumorigenesis. G9A activates the serine‐glycine biosynthetic pathway by transcriptionally upregulating key enzymes, such as PHGDH, phosphoserine aminotransferase 1 (PSAT1), SHMT2, and phosphoserine phosphatase (PSPH), by increasing H3K9me1 levels around the transcriptional start sites [217]. Consistently, KDM4C, the histone demethylase responsible for removing the repressive mark H3K9me3, could epigenetically coordinate the regulation of amino acid metabolism with G9A. Decreased H3K9me3 level with a concomitant increased ratio of H3K9me1 to H3K9me3 at the promoters of genes associated with the synthesis and transport of seine and glycine promote tumor proliferation [218]. LSD1 (KDM1A) activates glycolysis and represses mitochondrial metabolism and FAO in hepatocellular cancer. H3K4 demethylation in the promoter regions of PGC‐1α and LCAD partially explains the mechanism underlying this metabolic preference [219]. KDM5A specifically removes the active mark H3K4me3 on MPC‐1 genes in PDA. MPC‐1 promotes pyruvate metabolism in mitochondria. Transcriptional inhibition of MPC‐1 endows PDA with reliance on glycolysis [220]. P300/CBP regulates the alteration of cancer metabolism and the transcription of enzymes in glycolysis‐related metabolic pathways, such as amino acid metabolism, fatty acid metabolism, and nucleotide synthesis, by acetylating histone H3K18/K27 directly at the promoters of metabolic genes [221]. SIRT6 is deleted or downregulated in many cancer types, such as pancreatic and colorectal cancer. The deficiency of SIRT6, the co‐repressor of HIF‐1α and MYC, promotes tumorigenesis by supporting glycolytic switch, ribosome biogenesis, and glutamine metabolism without activating other oncogenic signaling pathways. Inhibition of glycolysis in SIRT6‐deficient cells completely inhibits tumor formation [222, 223]. Mechanistically, SIRT6 deletion, transcriptional silencing, and point mutations cannot deacetylate H3K9 and H3K56 and repress glycolytic gene expression [223, 224]. HDAC11 removes H3K9ac on the LKB1 promoter and inhibits its expression. LKB1 inhibition promotes glycolysis and maintains the stemness of HCC cells [225].
Several studies have suggested that the SWI/SNF complex is involved in the rewiring of cancer metabolism. ARID1A, along with other core subunits, can directly bind to the promoter of GLS1. ARID1A inactivation increases the accessibility of the GLS1 promoter and upregulates glutaminase (GLS) expression. ARID1A‐inactivated clear cell ovarian carcinoma cells show dependence on glutamine metabolism for aspartate generation, nucleotide synthesis, and a decrease in glucose consumption [226]. Another study found that ARID1A deficiency in ovarian cancer cells impairs the recruitment of SWI/SNF to the transcription start site of SLC7A11 and subsequently reduces cystine uptake and reduced glutathione (GSH) synthesis. Inhibiting the glutamate‐cysteine ligase synthetase catalytic subunit (GCLC), a rate‐limiting enzyme in the GSH metabolic pathway, induces oxidative stress and the death of cancer cells. Nevertheless, ARID1A‐deficient ovarian cancer cells are insensitive to GLS1 inhibition [227]. SMARCA4 is frequently mutated and inactivated in lung adenocarcinoma. SMARCA4 regulates genes in the hypoxic response pathway and glycolysis to combat energy stress. However, augmented fatty acid and protein synthesis in SMARCA4‐mutant cells results in substantial energy demand. Inconsistent with the Warburg effect, defective glycolytic capacity drives SWI/SNF‐mutant lung adenocarcinoma tumors to shift energy metabolism from glycolysis to OXPHOS [19]. Elevated BRG1 (SMARCA4) increases fatty acid synthesis in breast cancer by transcriptionally activating lipogenic genes, such as ACC, FASN, ACL, and ACSL1. Upregulated de novo lipogenesis can greatly promote tumor proliferation [228]. The above studies summarize the link ATP‐dependent CRCs to cancer metabolism and demonstrate a novel mechanism of how mutant CRCs components contribute to tumorigenesis. Remarkably, these findings provide a new perspective that the vulnerability of SWI/SNF‐mutant tumors to metabolism could be a therapeutic target (Table 1).
MicroRNAs regulate gene expression at the post‐transcriptional level [229]. The role of miRNAs in metabolism has been thoroughly investigated and documented; consequently, it is not discussed in detail in this section [230, 231]. Here, we emphasize that miRNAs are indispensable coordinators of metabolic regulatory networks. Long non‐coding RNAs (lncRNAs) participate in various physiological and pathological processes. LncRNAs are involved in various important cellular processes and play pivotal roles in gene regulation at multiple levels [232]. LncRNAs are involved in cancer metabolism via diverse mechanisms. LncRNAs can recruit chromatin modifiers to target genes and alter their epigenetic status. LINC00184 recruits DNMT1 to the PTEN promoter, increasing the methylation level of the PTEN promoter and inhibiting the expression of PTEN [233]. Fusobacterium nucleatum, an oncobacterium, activates glycolysis in colorectal cancer by increasing lncRNA ENO1‐IT1. LncRNA ENO1‐IT1 interacts with KAT7 specifically and mediates KAT7 binding to the promoter region of ENO1. Increased H3K27Ac levels promote transcription of enolase 1 (ENO1), which increases tumor glucose metabolism and progression [234]. LncRNAs can regulate gene expression by interfering with transcription. In prostate cancer, lncRNA PCGEM1 occupies DNA loci on the promoters of metabolic genes involved in glucose, lipid, and glutamine metabolism that overlap with c‐Myc. LncRNA PCGEM1 promotes the recruitment of c‐Myc to its target genes and induces transactivation activities. These results emphasize that the lncRNA PCGEM1 is a vital transcriptional regulator in restructuring metabolic networks [235]. LncRNAs also bind to other transcription factors, AHR, GLI2, and E2F1, to promote metabolic switching, thereby stimulating tumor progression [236, 237, 238]. LncRNAs mediate the splicing, degradation, and translation of mRNA. The lncRNA CCAT2 alters metabolism by facilitating glycolysis and glutaminolysis. The lncRNA CCAT2 acts as a scaffold binding GLS pre‐mRNA and CFIm complex and regulates alternative splicing of GLS in an allele‐specific manner. Moreover, other metabolic pathways, such as carbohydrate metabolism and fructose and mannose metabolism, may share the same alternative splicing mechanism [239]. LncRNA LNCAROD interacts with SRSF3, a splicer that mediates alternative splicing of PKM. Splicing switching of PKM from PKM1 to PKM2 upregulates glycolysis in HCC [240]. LncRNA GLS‐AS, an intronic antisense lncRNA, is derived from GLS. It can form double‐stranded RNA with GLS pre‐mRNA and recruit the ADAR/Dicer complex, which silences GLS expression. Under nutritional stress conditions, downregulated lncRNA GLS‐AS causes pancreatic cancer to accommodate glutamine and glucose deprivation [241]. Trastuzumab‐resistant breast cancer cells have upregulated lncRNA AGAP2‐AS1. LncRNA AGAP2‐AS1 forms a complex with HuR, which binds to and stabilizes carnitine palmitoyl transferase 1 (CPT1) mRNA to improve its expression, promote FAO, and induce drug resistance [242]. LncRNAs can mediate c‐Myc mRNA decay and glycolysis by virtue of IGF2BPs [243, 244, 245]. LncRNAs can regulate gene expression as sponges of miRNAs. LncRNA PVT1 contains miRNA‐complementary sites and acts as a competing endogenous RNA (ceRNA) of miR‐143, which targets and suppresses HK2 in gallbladder cancer. The sequestration of miR‐143 by lncRNA PVT1 elevates HK2 expression and facilitates the Warburg effect and gallbladder cancer progression [246]. This is the most extensively studied mechanism of the lncRNA‐mediated metabolic switch. The same mechanism fundamentally applies to aberrant regulation of metabolic transporters, key enzymes, and transcription factors associated with glucose, glutamine, and fatty acid metabolism [240, 242, 247‐250]. LncRNAs can bind to metabolic enzymes or transcriptional factors and modulate their activity or block their post‐translational modifications. LncRNA HULC repositions PKM2 and LDHA to the cell membrane and enhances the interaction between these glycolytic enzymes and their phosphorylation regulator, FGFR1. FGFR1 modulates enzymatic activities and promotes glycolysis by elevating their phosphorylation levels [251]. Hypoxia‐induced lincRNA‐p21 competitively binds to VHL and prevents hydroxylated HIF‐1α from interacting with it. Disassociation from VHL prevents HIF‐1α from degradation via the VHL‐dependent ubiquitin‐proteasome pathway [95]. In triple‐negative breast cancer, LINK‐A recruits BRK to phosphorylate HIF‐1α at Tyr565. Phosphorylation of Tyr565 attenuates the Pro564 site hydroxylated by PHD1 [252]. Many other lncRNAs stabilize PKM2, 6‐phosphofructo‐2‐kinase/fructose‐2,6‐biphosphatase 3 (PFKFB3), and c‐Myc by directly binding and blocking these proteins from ubiquitination‐mediated degradation [253, 254, 255, 256]. LncRNAs are found to function as scaffolds for proteins and RNA to form condensates. Under glutamine deprivation, lncRNA GIRGL forms a complex with CAPRIN1 and GLS1 mRNA and promotes the formation of stress granules via liquid‐liquid phase separation. This process contributes to the translational suppression of GLS, which favors tumor growth in a glutamine‐restricted environment [257]. Circular RNAs (circRNAs) have a single‐stranded, covalently closed‐loop structure. Growing evidence indicates that circRNAs play crucial roles in many diseases and have multiple biological functions [258]. Mechanistically, circRNAs can function as ceRNAs to sponge miRNAs and regulate downstream targets. Additionally, circRNAs can regulate transcription, interact with proteins, or even be translated into peptides [87]. Some circRNAs have been identified as key participants in reprogramming cancer metabolism. The overwhelming majority of research has focused on their ability to act as molecular sponges, which could antagonize the regulation of metabolic enzymes, transcription factors, and signaling pathways by miRNA. In HCC, miR‐338‐3p represses glycolysis by targeting and degrading PKM2. CircMAT2B sponges miR‐388‐3p and promotes glucose metabolism reprogramming and tumor cells’ malignancy under hypoxic conditions [259]. CircENO1 upregulates ENO1 and modulates glycolysis by targeting miR‐22‐3p in lung adenocarcinoma (LUAD) [260]. In pancreatic cancer, circMBOAT2 favors glutaminolysis by sponging miR‐433‐3p and upregulating glutamic‐oxaloacetic transaminase 1 (GOT1)[261]. Upstream molecules modulate glycolysis like HIF‐1α [262, 263], PTK [264], and c‐Myc [265], and upstream molecules related to glutamine metabolisms, such as Wnt2 [266], USP5 [267], and IGF [268], are also found to be regulated by the circRNA‐miRNA axis. CircRNAs can directly bind to target mRNA and regulate gene expression at the transcriptional level. CircRNF13 is a tumor suppressor that targets and stabilizes SUMO2 mRNA. SUMO2 accelerates GLUT1 degradation by promoting its SUMOylation and ubiquitination. Downregulated circRNF13 enhances aerobic glycolysis in nasopharyngeal carcinoma (NPC) [269]. Various modes of circRNA‐protein interactions are newly clarified mechanisms responsible for metabolic rewiring, which have not been thoroughly studied [270]. CircACC1 is induced under metabolic stress and plays a critical role in AMPK‐mediated metabolic reprogramming in colorectal cancer. CircACC1 binds to the β1 and γ1 subunits of AMPK and facilitates holoenzyme assembly and stability. AMPK phosphorylates and inactivates ACC1 to increase fatty acid β‐oxidation but has the opposite effect on 6‐phosphofructo‐2‐kinase (PFK2) to promote glycolysis [271]. CircCUX1 binds to EWSR1 and promotes its interaction with MAZ. Activated MAZ promotes the transcription of CUX1, a transcription factor that facilitates glycolysis [272]. CircCDKN2B‐AS1 recruits IMP3 (IGF2BP3) to the HK2 mRNA, making it more stable. Increased expression of HK2 favors glycolysis in cervical cancer [273]. In colorectal cancer, circMYH9 impedes the binding between hnRNPA2B1 and p53 pre‐mRNA. CircMYH9 relieves transcriptional repression of serine and glycine anabolism by impairing the expression of p53 [274]. In LUAD, circDCUN1D4 forms a ternary complex with HUR and TXNIP mRNA and regulates glycolysis in a TXNIP‐dependent manner [275] (Figure 3). Epigenetic modifications, chromatin remodeling, and ncRNAs participate in the precise regulation of metabolism to favor tumor initiation and progression. They control the ability of tumor cells to uptake nutrients, metabolize nutrients, and adapt to nutrition deprivation. Dysregulated epigenetic patterns can cause specific metabolic preferences or dependencies in tumor cells. These weaknesses can be exploited and directly targeted. Furthermore, epigenetic drugs may profoundly remodel cellular metabolic states and thus sensitize tumor cells to other metabolic drugs. One such example is that dual inhibition of DNMT and KMT reverses the Warburg effect and causes OXPHOS dependence in glycolysis‐addicted hematological malignancies [276]. Targeting mitochondrial metabolic stress potentiates the effects of epigenetic drugs. This drug combination shows encouraging results in the clinical trial. In older patients with AML, azacitidine plus venetoclax, a BCL2 inhibitor, significantly improved the median overall survival to 14.7 months, as compared with 9.6 months in the group with azacitidine alone [277]. These basic and clinical studies may open new avenues for developing combination strategies based on epigenetic and metabolic drugs.
Dynamic RNA modification is an emerging research field termed “RNA epigenetics” [278, 279]. Prevalent modifications on mRNA include m6A, N7‐methylguanosine (m7G), 5‐methylcytosine (m5C), N1‐methyladenosine (m1A), pseudouridine (Ψ), inosine (I), and uridine (U). m6A is the most abundant epigenetic mRNA modification, accounting for 60% of RNA methylation. M6A RNA modifications regulate mRNA splicing, nuclear transport, translation, and degradation [280]. As a reversible chemical modification, m6A could also be deposited by writer proteins, removed by eraser proteins, and recognized by reader proteins [281]. M6A is found to regulate gene expression in various biological processes, and disturbed distribution or abundance of m6A could even drive many diseases [282, 283, 284]. Accumulating evidence has demonstrated that m6A RNA modification is affected by cancer metabolism; conversely, it extensively impacts cancer metabolic rewiring by modulating the expression of metabolic genes, which drive tumor development. Although there is a lack of relevant studies in the literature, we could envisage that other novel RNA modifications, such as m5C, m1A, and Ψ, are also closely linked with metabolism in cancer. Elucidating the roles of the crosstalk between RNA epigenetics and cancer metabolism will be an important area for further investigation. In addition to DNA and histone methylation, SAM is also required for RNA methylation. mTORC1 promotes methionine metabolism and increases SAM production via MAT2A, a crucial target for activated mTORC1 signaling. Nevertheless, mTORC1‐dependent regulation of SAM synthesis has little impact on DNA and histone methylation states. Tumors with hyperactivated mTORC1 depend on MAT2A‐mediated m6A RNA for protein synthesis and cell proliferation [285]. Methylenetetrahydrofolate dehydrogenase 2 (MTHFD2), a mitochondrial enzyme involved in one‐carbon metabolism, is elevated in clear cell renal cell carcinoma (ccRCC). MTHFD2 depletion results in decreased global methylation levels of nucleic acids and histones, of which RNA methylation is the most influenced. Increased methylation of HIF‐2α mRNA enhances its translation and subsequently promotes aerobic glycolysis [286]. Similar to DNA and histone disturbances, RNA methylation is significantly elevated in IDH‐mutant tumors because fat mass and obesity‐associated protein (FTO) are α‐KG‐dependent dioxygenases that can also be competitively inhibited by R‐2HG (D2‐HG) [287, 288]. However, R‐2HG‐induced hypermethylation produces contradictory effects on tumorigenesis. In IDH‐mutant leukemia, the decreased m6A demethylase activity of FTO abrogates m6A /YTHDF2‐mediated upregulation of PFKP and LDHB, attenuates aerobic glycolysis, and inhibits leukemogenesis [289] (Figure 1). High methyltransferase‐like 3 (METTL3) expression increased HK2 and GLUT1 expression depending on its m6A methyltransferase activity. M6A modification regulates HK2 and GLUT1 mRNA levels and stability and is closely correlated with the activation of glycolysis in colorectal cancer [290]. In cervical and liver cancer cells, m6A positively regulates glycolysis by stabilizing and promoting the translation of PDK4, which controls glucose flux into glycolysis and OXPHOS [291]. Another potential target is ENO1 in LUAD [292]. FTO has a synthetic lethal interaction with VHL tumor suppressor in ccRCC. VHL‐deficient tumor cells are addicted to glutamine. Increased FTO rewires the metabolic reprogramming and survival of VHL‐deficient ccRCC cells by diminishing m6A methylation and enhancing the expression of the glutamine transporter SLC1A5 [293]. Some key transcription factors or upstream regulators related to metabolic reprogramming are also affected by m6A RNA modifications. METTL3 activates glycolysis by promoting m6A modification of HDGF mRNA in gastric cancer [294], HIF‐1α mRNA in liver cancer [295], APC mRNA in ESCC [296], and USP48 mRNA in liver cancer [297]. METTL3 enhances pre‐mRNA splicing of ERRγ. ERRγ increases FAO via regulating CPT1B [298]. FTO demethylates the transcription factors c‐Jun, JunB, C/EBPβ, and c‐Myc, thus rewiring glycolytic metabolism [299]. In LUAD, decreased FTO upregulates m6A abundance on MYC mRNA and enhances glycolysis [300]. In bladder cancer, decreased AlkB homolog 5 RNA demethylase (ALKBH5) promotes glycolysis by stabilizing CK2α mRNA in an m6A‐dependent manner [301]. In metastatic renal cell carcinoma (RCC), downregulated methyltransferase‐like 14 (METTL14) reduces m6A levels and stabilizes BPTF, which alters the super‐enhancer landscape, affects DNA accessibility, and promotes glycolytic reprogramming [302]. YTHDF2 mediates m6A‐dependent mRNA decay of LXRA, which is involved in cholesterol homeostasis control [303]. M5C RNA modification can bridge transcription and translation. The m5C modification on PKM2 mRNA can be recognized and stabilized by Aly/REF nuclear export factor (ALYREF) to facilitate glycolysis and cell proliferation [304]. Similar to DNA and histone modifications, RNA modifications regulate cancer metabolism, and conversely, cancer‐specific metabolic changes can affect RNA modifications. RNA‐modifying enzymes are potential therapeutic targets for cancer therapy [305]. FB23‐2, a newly developed FTO inhibitor, can inhibit proliferation, promote differentiation, and induce apoptosis in AML cells, showing efficacy in treating AML [306]. However, there are no currently available small‐molecule activators or inhibitors that selectively target RNA methyltransferases. Although the development of targeted drugs is still in a very early stage, their clinical applications might be very promising.
Previous clinical trials have suggested that using a single epigenetic or metabolic agent is insufficient. Based on the topic of this review, it is interesting to test whether metabolic or epigenetic abnormalities sensitize tumor cells to other epigenetic drugs, metabolic agents, or combined therapies. The aforementioned studies have provided a source of inspiration for identifying novel targets.
DNA and histone modifications are both highly dynamic and reversible. Small‐molecule compounds can potentially reverse aberrant epigenetic modification patterns during tumorigenesis, some of which have been approved for clinical use in hematological malignancies [307]. However, the therapeutic effect of monotherapy is not satisfactory for all patients and lacks efficacy for other solid tumors [308]. This raises interest in using combinations of epigenetic therapies with other agents in chemotherapies, immunotherapies, or targeted therapies to achieve synergistic effects. Analogously, despite many drugs targeting cancer metabolism entering clinical trials, few metabolic therapies have been approved [309, 310]. Metabolic heterogeneity and plasticity may account for the failed applications [311]. Therefore, it is necessary to identify bona fide metabolic vulnerability in a certain type of cancer. Metabolic alterations have also been found to be involved in treatment resistance. The combined use of metabolic agents may unlock the potential of epigenetic drugs and provide new clinical opportunities [312]. There are some possible ways to identify novel targets. First, basic researches have employed transcriptomics, epigenomics, and metabolomics to discover many new potential targets. For example, analysis of the metabolome of tumor cells after epigenetic agent GSK126 treatment reveals that lipid synthesis is strengthened to mediate drug resistance. Thus, targeting lipid metabolism can restore sensitivity to epigenetic therapy [313]. Second, combination drug screens with selected drug libraries targeting the metabolic and epigenetic abnormalities exhibited in tumors may provide more direct evidence to develop optimal therapies [314]. Third, current clinically proven treatment strategies may be extended to other cancer types possessing similar metabolic and epigenetic abnormalities. Testing these strategies will offer new therapeutic options for tumors that lack effective treatments.
Recently, the therapeutic potential of epigenetic agents in combination with metabolic inhibitors has attracted considerable attention. For IDH1‐mutant AML, the mIDH1 (mutant IDH1) inhibitor ivosidenib and the hypomethylating agent azacitidine showed promising therapeutic effects in both preclinical stages and clinical trials. Encouraging results from a phase 3 trial showed that patients treated with ivosidenib and azacitidine combined therapy experienced greater clinical benefits than those treated with azacitidine monotherapy [204, 205, 315]. These works remind us of other cancer types with similar mutational and epigenetic patterns, such as glioma, sarcoma, and cholangiocarcinoma [316, 317]. Several small‐molecule inhibitors targeting the glioma epigenome, such as mIDH inhibitors, HDAC inhibitors and DNMT inhibitors, are under clinical evaluation. A new clinical trial is underway to examine the effect of the combination of olutasidenib (mIDH1 inhibitor) with azacitidine in advanced glioma and chondrosarcoma [318]. Another breakthrough was discovering the potent synergistic anticancer effect of hypomethylating agents and BCL2 inhibitor venetoclax in solid tumors and hematological malignancies. Epigenetic drugs that inhibit DNMT, HDAC, and HMT trigger a marked metabolic shift from glycolysis to OXPHOS, which could generate excessive oxidative stress. Venetoclax then boosts the apoptosis of tumor cells by depolarizing the mitochondrial membrane and disrupting mitochondrial metabolism [276, 319, 320]. These drug combinations deliver a powerful one‐two punch to cancer cells and have been successfully translated into clinical trials on leukemia and myelodysplastic syndrome [277, 321]. More importantly, solid tumors, such as liver, lung, colon and breast cancer, synergistically respond to these drug combinations. Further studies are necessary to determine whether their extraordinary results will be recapitulated. Clinical experience suggests that epi‐drugs are often ineffective in solid tumors, restricting their further applications. Thus, unraveling the underlying mechanisms of drug resistance or insensitivity is urgently needed. Epigenetic agents may also induce specific metabolic vulnerabilities in solid tumors, which could be exploited to develop innovative combinatorial treatment regimens. The EZH2 inhibitor GSK126 could change the overall metabolic profiles of melanoma, as evidenced by enhanced lipid synthesis. Drugs targeting fatty acid metabolism can re‐sensitize tumor cells to EZH2 inhibition [313]. In cervical cancer, inhibition of HDAC makes cancer cells rely on glucose and glutamine catabolism for survival. Glycolysis and glutamine metabolism blockers, combined with HDAC inhibitors, further induce oxidative and energetic stress, accelerating cancer cell apoptosis [322, 323]. In glioblastoma, HDAC inhibitors elicit profound metabolic changes characterized by enhanced FAO but a decreased Warburg effect. The interaction and cause‐and‐effect relationship between epigenetic and metabolic processes provide a rationale for the combined use of the pan‐HDAC inhibitor panobinostat and FAO inhibitor etomoxir. Combination treatment has shown better therapeutic effects than any single agent in patient‐derived xenograft models [324] (Table 2).
The concept of synthetic lethality can be summarized as the interaction between two genes. Loss of either gene alone does not affect cell viability, but the loss of both genes simultaneously leads to cell death [325, 326]. In other words, losing one of the two genes renders tumor cells highly dependent on another. Consequently, targeting the synthetic lethal partner is a potent anticancer strategy for oncogenic mutations previously thought to be pharmacologically intractable [327, 328]. One of the most classic examples of synthetic lethal interactions in cancers is the BRCA mutation and PARP inhibition [329, 330]. Since then, many other novel synthetic lethal interactions have been identified [328]. Loss‐of‐function mutations lack targeted therapeutic approaches, and some are vulnerable to metabolic inhibitors or epigenetic agents. Available evidence demonstrates that synthetic lethal screening is a promising therapeutic option for patients with epigenetic or metabolic deficiencies. Cancer cells with epigenetic defects exhibit metabolic vulnerabilities. Recent findings have extended the synthetic lethal partners to proteins closely related to metabolism. BCAT1 inhibitors impair the proliferation of EZH2‐deficient leukemia‐initiating cells both in vitro and in vivo. Inhibition is selective and does not affect normal HPSCs and hematopoiesis. Inhibition of metabolism may also be applied to other types of hematological malignancies with EZH2 mutations or dysregulation [212]. In LUAD, KMT2D loss abolishes the inhibitory effect of PER2 on glycolytic genes. Increased glycolytic activity is an attractive therapeutic vulnerability. 2‐DG preferentially hampers LUAD cell growth and tumor formation in xenotransplantation models [213]. TET3‐depleted AML cells are sensitive to inhibition of glycolysis by 2‐DG [207]. Lung cancer with SMARCA4 or ARID1A loss is characterized by enhanced OXPHOS. Extreme reliance on energy production makes SWI/SNF‐mutant LUAD more susceptible to the OXPHOS inhibitor IACS‐010759 than tumor cells without the aforementioned mutations [19]. ARID1A inactivation was synthetically lethal with GLS and GCLC inhibition. The loss of ARID1A leads to a metabolic phenotype characterized by glutamine dependence. ARID1A‐mutant ovarian clear cell carcinoma cell lines and tumors formed in orthotopic xenograft models are sensitive to GLS inhibitor CB‐839 [226]. Another study on ovarian carcinoma cells reported that ARID1A mutations have synthetic lethal relationships with the glutathione metabolic pathway. Pharmacologically inhibiting the key enzyme GCLC with buthionine‐sulfoximine selectively induces ARID1A‐deficient cancer cell death. Surprisingly, both the genetic and pharmacological inhibition of glutamine transport and catabolism are ineffective in cells with ARID1A deficiency [227]. Metabolic deficiencies create specific vulnerabilities to epigenetic agents. LKB1‐mutant pancreatic tumor cells are susceptible to inhibition of the serine metabolic pathway and DNA methylation, which is the major consumer of SAM. DNMT inhibitor decitabine hinders tumor growth, induces necrosis and apoptosis, and causes significant tumor regression in vitro and in vivo [135]. Similarly, the loss of PKCλ/ι induces NEPC differentiation by controlling global DNA methylation levels. Decitabine blocks NEPC differentiation and inhibits tumor proliferation [138]. A shortage of glutamine leads to histone hypermethylation on H3K27, which helps melanoma cells develop drug resistance to BRAF inhibitors. However, abnormal histone methylation patterns confer crucial vulnerability to histone methyltransferase EZH2 inhibitors. DZNep and EPZ005687 inhibit tumor growth when combined with BRAF inhibitors to overcome tumor drug resistance [151] (Table 3). Suppressing a broad spectrum of metabolic or epigenetic enzymes can cause potential deleterious side effects. New‐generation epigenetic drugs, such as BET inhibitors, HMT inhibitors, and KDM inhibitors, are more specific. Their applications may improve the efficacy and tolerability of synthetic lethal therapies and epigenetic drugs in combination with metabolic therapies.
Cell metabolism and the epigenetic landscape are highly dynamic. Epigenetic abnormalities deregulate metabolic enzymes or signaling pathways to provide energy, nucleotides, amino acids, fatty acids and many other metabolites to cancer cells and support their rapid proliferation. Furthermore, nutritional status and intracellular signals coordinate gene expression at the epigenetic level by churning metabolite pools. These two cooperate to enable cancer cells to quickly adjust to the changing environment. However, it is intriguing that the mutual regulation of metabolism and epigenetics is precise to some extent. Specifically, only limited and certain types of histone methylation are influenced when the intracellular SAM content fluctuates. It can be surmised that this phenomenon is ascribed to the different catalytic properties of the enzymes responsible for those methylation sites. Furthermore, not all metabolic pathways are selectively modulated by epigenetic lesions in cancer cells. KMT2D‐mutated cancer cells consistently showed a dependency on glycolysis. In contrast, different cancers with the same epigenetic lesion as ARID1A inactivation tend to expose distinct metabolic fragilities. The mechanisms underlying these discrepancies warrant further investigation. It is worth noting that concomitant changes in diverse cellular processes occur inextricably when cellular metabolic states shift. For example, the AMPK and mTOR pathways are intrinsic metabolic sensors that monitor intracellular energy production and nutrient supply, controlling cell growth, proliferation, and survival [331]. In addition to being a methyl donor, increased availability of SAM could function as a signal molecule sensed by the SAM sensor upstream of mTORC1 (SAMTOR) and abrogate inhibition of the mTOR pathway [332]. Furthermore, as mentioned above, many enzymes share the same substrates or cofactors. They are also affected by metabolic disturbances and chromatin modifiers. For instance, oncometabolites can drive tumorigenesis by hampering the activity of prolyl‐hydroxylases, which fosters the stabilization of HIF‐1α, in addition to demethylases [333]. In addition, nonhistone proteins are widely modulated by various metabolite‐induced post‐translational modifications, affecting almost all aspects of cell biology, such as gene transcription and signal transduction. One case is p53, whose acetylation and methylation can fine‐tune its transcriptional activity [334]. These extensive and unexpected biological effects on cancer may obfuscate the contribution of epigenetic mechanisms and require careful dissection. Considering the highly intertwined relationship between metabolism and epigenetic regulation, it is not surprising that metabolic drugs can reverse epigenetic alterations, and in turn, epigenetic agents can exert antitumor effects partly by disturbing cancer metabolism [221]. High‐throughput technologies will help characterize the specific epigenetic or metabolic vulnerabilities exposed during this two‐way communication, which could be induced and exploited as potential therapeutic targets. Combined pharmacological intervention and synthetic lethal screening are feasible approaches. In particular, elegant studies combining metabolic therapy and epigenetic therapy in hematological malignancies provide a milestone in targeting the epigenetic‐metabolic circuit, hopefully becoming a novel paradigm for cancer treatment. Although the prospect is exciting, most of our related knowledge is limited to in vitro studies and is usually context‐specific, without considering the effects on immune cells [335, 336, 337]. More confirmatory evidence should be explored before actual clinical practice. Recently, the burgeoning fields of ncRNAs and RNA epigenetics have provided novel insights into the crosstalk between epigenetics and cancer metabolism, the therapeutic values of which have not yet been comprehensively studied. Nonetheless, they can be regarded as candidate targets for developing new therapies. In conclusion, this review highlights the close connections between metabolism and epigenetics in cancer and proposes promising targeting therapeutic strategies. The current preclinical and clinical studies knowledge will potentially open up further research and novel therapeutic opportunities.
Not applicable.
Not applicable.
The authors declare that they have no competing interests.
This work was supported by the National Natural Science Foundation of China (81600766), the Science and Technology Commission of Shanghai (20DZ2270800), and the Innovative research team of high‐level local universities in Shanghai (SHSMU‐ZDCX20210900; SHSMU‐ZDCX20210902).
Xianqun Fan, Shengfang Ge, and Ai Zhuang designed and revised the manuscript. Tongxin Ge, Ai Zhuang, and Peiwei Chai wrote the manuscript and made the figures. Peiwei Chai, Xiang Gu, and Renbing Jia polished the manuscript and gave useful suggestions. All authors read and approved the final manuscript. |
PMC9648396 | 36042007 | Rongfang Wei,Yan Zhu,Yuanjiao Zhang,Wene Zhao,Xichao Yu,Ling Wang,Chunyan Gu,Xiaosong Gu,Ye Yang | AIMP1 promotes multiple myeloma malignancy through interacting with ANP32A to mediate histone H3 acetylation | 30-08-2022 | multiple myeloma,AIMP1,osteoclast differentiation,MAPK signaling,ANP32A,histone H3 acetylation,osteolytic lesions | Abstract Background Multiple myeloma (MM) is the second most common hematological malignancy. An overwhelming majority of patients with MM progress to serious osteolytic bone disease. Aminoacyl‐tRNA synthetase‐interacting multifunctional protein 1 (AIMP1) participates in several steps during cancer development and osteoclast differentiation. This study aimed to explore its role in MM. Methods The gene expression profiling cohorts of MM were applied to determine the expression of AIMP1 and its association with MM patient prognosis. Enzyme‐linked immunosorbent assay, immunohistochemistry, and Western blotting were used to detect AIMP1 expression. Protein chip analysis, RNA‐sequencing, and chromatin immunoprecipitation and next‐generation sequencing were employed to screen the interacting proteins and key downstream targets of AIMP1. The impact of AIMP1 on cellular proliferation was determined using 3‐(4,5‐dimethylthiazol‐2‐yl)‐2,5‐diphenyltetrazolium bromide (MTT) assay in vitro and a xenograft model in vivo. Bone lesions were evaluated using tartrate‐resistant acid phosphatase staining in vitro. A NOD/SCID‐TIBIA mouse model was used to evaluate the effect of siAIMP1‐loaded exosomes on bone lesion formation in vivo. Results AIMP1 expression was increased in MM patients and strongly associated with unfavorable outcomes. Increased AIMP1 expression promoted MM cell proliferation in vitro and in vivo via activation of the mitogen‐activated protein kinase (MAPK) signaling pathway. Protein chip assays and subsequent experiments revealed that AIMP1 interacted with acidic leucine‐rich nuclear phosphoprotein 32 family member A (ANP32A) to regulate histone H3 acetylation. In addition, AIMP1 increased histone H3 acetylation enrichment function of GRB2‐associated and regulator of MAPK protein 2 (GAREM2) to increase the phosphorylation of extracellular‐regulated kinase 1/2 (p‐ERK1/2). Furthermore, AIMP1 promoted osteoclast differentiation by activating nuclear factor of activated T cells c1 (NFATc1) in vitro. In contrast, exosome‐coated small interfering RNA of AIMP1 effectively suppressed MM progression and osteoclast differentiation in vitro and in vivo. Conclusions Our data demonstrate that AIMP1 is a novel regulator of histone H3 acetylation interacting with ANP32A in MM, which accelerates MM malignancy via activation of the MAPK signaling pathway. | AIMP1 promotes multiple myeloma malignancy through interacting with ANP32A to mediate histone H3 acetylation
Multiple myeloma (MM) is the second most common hematological malignancy. An overwhelming majority of patients with MM progress to serious osteolytic bone disease. Aminoacyl‐tRNA synthetase‐interacting multifunctional protein 1 (AIMP1) participates in several steps during cancer development and osteoclast differentiation. This study aimed to explore its role in MM.
The gene expression profiling cohorts of MM were applied to determine the expression of AIMP1 and its association with MM patient prognosis. Enzyme‐linked immunosorbent assay, immunohistochemistry, and Western blotting were used to detect AIMP1 expression. Protein chip analysis, RNA‐sequencing, and chromatin immunoprecipitation and next‐generation sequencing were employed to screen the interacting proteins and key downstream targets of AIMP1. The impact of AIMP1 on cellular proliferation was determined using 3‐(4,5‐dimethylthiazol‐2‐yl)‐2,5‐diphenyltetrazolium bromide (MTT) assay in vitro and a xenograft model in vivo. Bone lesions were evaluated using tartrate‐resistant acid phosphatase staining in vitro. A NOD/SCID‐TIBIA mouse model was used to evaluate the effect of siAIMP1‐loaded exosomes on bone lesion formation in vivo.
AIMP1 expression was increased in MM patients and strongly associated with unfavorable outcomes. Increased AIMP1 expression promoted MM cell proliferation in vitro and in vivo via activation of the mitogen‐activated protein kinase (MAPK) signaling pathway. Protein chip assays and subsequent experiments revealed that AIMP1 interacted with acidic leucine‐rich nuclear phosphoprotein 32 family member A (ANP32A) to regulate histone H3 acetylation. In addition, AIMP1 increased histone H3 acetylation enrichment function of GRB2‐associated and regulator of MAPK protein 2 (GAREM2) to increase the phosphorylation of extracellular‐regulated kinase 1/2 (p‐ERK1/2). Furthermore, AIMP1 promoted osteoclast differentiation by activating nuclear factor of activated T cells c1 (NFATc1) in vitro. In contrast, exosome‐coated small interfering RNA of AIMP1 effectively suppressed MM progression and osteoclast differentiation in vitro and in vivo.
Our data demonstrate that AIMP1 is a novel regulator of histone H3 acetylation interacting with ANP32A in MM, which accelerates MM malignancy via activation of the MAPK signaling pathway.
Abbreviations AIMP1 Aminoacyl‐tRNA synthetase‐interacting multifunctional protein 1 ANP32A Acidic leucine‐rich nuclear phosphoprotein 32 family member A AP‐1 Activator protein 1 APEX The assessment of proteasome inhibition for extending remission BM Bone marrow BMD bone mineral density BV/TV Bone volume / Total volume CHIP‐seq Chromatin Immunoprecipitation and next‐generation sequencing Co‐IP Co‐immunoprecipitation DMEM Dulbecco's modified Eagle medium EDTA Ethylene Diamine Tetraacetie Acid EFS Event‐free survival ELISA Enzyme linked immunosorbent assay ERK Extracellular signal‐regulated kinase FBS Fetal bovine serum GAREM2 GRB2‐associated and regulator of MAPK protein 2 GEP The Gene expression profiling GO Gene Ontology HATs Histone acetyltransferases HDACs Histone deacetylases H&E Hematoxylin‐eosin HMGA1 High Mobility Group AT‐Hook 1 HOVON65 The Dutch‐Belgian Cooperative Trial Group for Hematology Oncology Group‐65 IHC Immunohistochemistry KD Knockdown KEGG Kyoto Encyclopedia of genes and genes MAPK mitogen‐activated protein kinase M‐CSF Macrophage Colony stimulating Factor MGUS Monoclonal gammopathy of undetermined significance MM Multiple myeloma MTT 3‐(4,5‐Dimethylthiazol‐2‐yl)‐2,5‐Diphenyltetrazolium Bromide NC Negative control NFATc1 Nuclear factor of activated T cells c1 OE Overexpression OS Overall survival PBS Phosphate Buffered Saline PDX Patient‐derived tumor xenograft PFS Progress‐free survival p‐ERK1/2 Phosphorylated extracellular‐regulated kinase 1/2 RNA‐seq RNA sequencing RNAKL Receptor activator of NF‐κB ligand ROS Reactive oxygen species RT‐qPCR Real Time Quantitative PCR SD Standard deviation shRNA Short hairpin RNA siRNA Small interfering RNA STAT3 Signal transducer and activator of transcription‐3 TNF‐α Tumor necrosis factor‐α TRAF6 TNF receptor‐associated factor 6 TRAP Tartrate‐Resistant Acid Phosphatase TT2 Total therapy 2 TT3 Total therapy 3 WB Western blotting WT Wild type
Multiple myeloma (MM) is a hematological malignancy with the aggregation of clonal malignant plasma cells in the bone marrow (BM) [1]. MM patients often suffer from many symptoms, such as anemia, renal failure, bone lesions and hypercalcemia [2, 3]. Through the advancement of several therapies such as proteasome inhibitors, immunomodulatory drugs, monoclonal antibodies, and stem cell transplantation, the outcomes of MM patients have improved [4]; however, recurrence and the general incurable nature of the disease remain major challenges in the treatment of MM. One of the characteristics of MM is a bone disease caused by the alteration of a dynamic balance between osteoclast and osteoblast activity [5], which presents as fractures, pain, and a reduction in the quality of life [6, 7]. The BM microenvironment sustains MM cell survival and drug resistance [8]. Despite the improvements in the treatment of MM, it remains incurable and patients with MM are prone to relapse [9]. Common epigenetic mechanisms play a prominent role in the pathogenesis of MM, including DNA methylation [10] and histone acetylation [11, 12, 13]. Acetylation of histones, one of the most common post‐translational modifications, contributes to the chromatin dynamics in the modulation of gene transcription and thus regulates active gene expression [14, 15]. Histone acetylation is reversibly regulated by two families of enzymes, histone acetyltransferases (HATs) and histone deacetylases (HDACs) [13]. As a type of HATs, CREB‐binding protein (CBP)/p300 not only is essential in physiological events but also plays a role during tumor transformation [16]. CBP/p300 catalyzes acetylation of signal transducer and activator of transcription‐3 (STAT3) to activate it, and then activated STAT3 recruits HDAC1 to regulate gene expression and inhibit STAT3 in a negative regulatory loop [17]. It is worth noting that targeting CBP/p300 represents a viable therapeutic strategy for the management of MM [18]. HDAC is a well‐recognized eraser of histone acetylation, and its inhibitors (HDACIs) can induce cell differentiation, cell cycle arrest, apoptosis, reactive oxygen species (ROS) production, mitotic cell death and inhibit cell migration [19, 20, 21, 22, 23]. An additional promising effect of HDACIs for cancer therapy is their selective toxicity against tumor cells compared with normal cells [24, 25]. Preclinical studies have reported that HDACIs trigger cell apoptosis and induce cell cycle arrest in MM cells [26, 27, 28]. These observations reinforce the concept that targeting key factors of histone acetylation is a promising strategy for MM therapy. Aminoacyl‐tRNA synthetase‐interacting multifunctional protein 1 (AIMP1, also known as p43) is initially identified as an auxiliary factor associated with the macromolecular aminoacyl‐tRNA synthetase complex [29], whose secretion can be stimulated by tumor necrosis factor‐α (TNF‐α), heat shock and hypoxia [30, 31]. AIMP1 participates in several cellular processes, including wound healing [30], inflammation [32, 33, 34], angiogenesis [35], and glucose homeostasis [36]. The N‐terminal peptide of AIMP1 induces the phosphorylation of extracellular signal‐regulated kinase (ERK) to promote the proliferation of mesenchymal stem cells [37]. However, AIMP1 possesses both anticancer and carcinogenic effects in different tumors [38, 39, 40]. On the one hand, AIMP1 expression is decreased in gastric and colorectal cancer [38], and it has anti‐tumor activity in a lung cancer xenograft model [39]; on the other hand, AIMP1 promotes cell proliferation, migration, and invasion in laryngeal squamous cell carcinoma cells [40]. However, there is no research on AIMP1 in MM, to the best of our knowledge. It has been reported that AIMP1 induces osteoclastogenesis and acts cooperatively with receptor activator of nuclear factor‐κB (NF‐κB) ligand (RANKL) to promote osteoclastogenesis [32]. In addition, a monoclonal antibody blocking the cytokine activity of AIMP1 suppresses the AIMP1‐mediated osteoclastogenesis in vitro [32]. Therefore, AIMP1 may regulate the BM microenvironment and participate in the progression of MM, and systematic research is required to confirm this hypothesis. In the present study, we evaluated AIMP1 expression in MM patients in Gene expression profiling (GEP) datasets and aimed to identify a novel mechanism of AIMP1 in promoting cell proliferation and altering the BM microenvironment in MM patients to provide novel insight into the mechanisms underlying the development of MM.
The GEP cohorts were obtained from the GEO database as previously described [41]. Total therapy 2 (TT2, GSE2658, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE2658), total therapy 3 (TT3, GSE2658), the assessment of proteasome inhibition for extending remission (APEX, GSE9782, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE9782) and the Dutch‐Belgian Cooperative Trial Group for Hematology Oncology Group‐65 (HOVON65, GSE19784, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE19784) were used in the present study. The TT2 cohort was composed of patients with newly diagnosed MM subsequently treated with high‐dose melphalan and stem cell transplantation. In the TT3 cohort, bortezomib was incorporated up‐front into a tandem transplant regimen for newly diagnosed MM. The APEX cohort was composed of patients with relapsed MM enrolled in phase II/III clinical trials of bortezomib and consented to genomic analyses of pretreatment tumor samples. The HOVON65 cohort was composed of patients with newly diagnosed MM in the Netherlands Belgium / Germany hovon‐65 / gmmg‐hd4 trial. The GEP dataset GSE6401 associated with osteoclast differentiation was used to analyze AIMP1 signaling in MM patients with mild or serious bone lesions (https://www.ncbi.nlm.nih.gov/search/all/?term=GSE6401). The detailed information on the GSE6401 dataset is listed in Supplementary Table S1.
Human MM cell lines H929, OCI‐MY5, ARP1 and CAG were cultured in RPMI‐1640 (05‐065‐1A, Biological Industries, Kibbutz Beit Haemek, Israel). Human embryonic kidney 293T (HEK293T) and mouse leukemic monocyte/macrophage RAW264.7 cells were cultured in Dulbecco's modified Eagle medium (DMEM, 01‐052‐1ACS, Biological Industries). All the above cells were kindly donated by Prof. Siegfried Janz (Division of Hematology and Oncology, Medical College of Wisconsin, Madison, WI, USA). All cell lines used in the present study were verified by short tandem repeat genotyping in October 2021. All culture medium used was supplemented with 10% fetal bovine serum (FBS, 04‐001‐1ACS, Biological Industries) and 1% penicillin/streptomycin (03‐031‐1B, Biological Industries). Cells were cultured at 37°C in a humidified incubator supplied with 5% CO2 air.
ELISA was performed using a human AIMP1 ELISA kit (ZC‐54495, ZCi BIO, Shanghai, China) as per the manufacturer's instructions. The serum samples were collected from 11 healthy subjects and 19 MM patients at the Affiliated Hospital of Nantong University (Nantong, Jiangsu, China; ethics number: 2018‐K007). All participants provided written informed consent.
IHC examination was performed as described previously [42]. The posterior superior iliac spine of MM patients was chosen as the biopsy point. A bone marrow biopsy needle was used to pierce vertically, and the needle was inserted clockwise for 1‐2 mm. Then the needle tube was withdrawn clockwise, the bone marrow tissue was obtained and placed in the fixative solution for IHC. Slides were incubated with the primary antibodies of anti‐AIMP1 (11091‐1‐AP, Proteintech, Wuhan, Hubei, China) or anti‐Ki67 (AF0198, Affinity, Changzhou, Jiangsu, China) at 4˚C overnight. Next day, the slides were treated with secondary antibody at 37˚C for 45 min, followed by applying strept avidin‐biotin complex (SABC) at 37˚C for 30 min, diaminobenzidine (DAB) coloring, and counterstaining with hematoxylin finally. Semiquantitative method for AIMP1 IHC staining was performed by an experimental pathologist using the following staining intensity scores: 0 (negative), 1+ (weak), 2+ (moderate), and 3+ (strong). The total number of cells (0‐100%) at each intensity level were multiplied by the corresponding intensity score. The final staining scores were calculated by summing the four intensity percentages; the minimum possible final staining score was 0 (no staining), and the maximum possible score was 300 (100% of cells with 3+ staining intensity). The samples for IHC were collected from the Affiliated Hospital of Nanjing University of Chinese Medicine (Ethics number: KY2018005). All patients provided written informed consent for their BM tissue samples to be used for research.
The plasmids containing human AIMP1 cDNA or short hairpin RNA (shRNA) cassettes and mouse AIMP1 cDNA were provided by TranSheepBio (Shanghai, China). Mouse AIMP1 plasmid was transfected to RAW264.7 transiently by Lipofectamine Transfection Reagent (40802ES02, Yeasen, Shanghai, China). The AIMP1‐coding sequence was cloned into the lentiviral vector pTSB with Flag tag (TranSheepBio). AIMP1‐targeted shRNA was cloned into vector pLKO.1 (TranSheepBio). Lentiviruses containing cDNA or shRNA were obtained by co‐transfection of the pTSB‐AIMP1 vector or AIMP1 shRNA vector with packaging vectors (Pspax.2 and PMD2.G, TranSheepBio) into HEK293T cells following the protocol of Lipofectamine Transfection Reagent (40802ES02, Yeasen). MM cells were transfected with lentivirus containing AIMP1 cDNA or shRNA to yield AIMP1 overexpression (AIMP1‐OE) or AIMP1 knockdown (AIMP1‐KD) MM cells. Transfected cells were selected by puromycin (60210ES25, Yeasen). Transduction efficiency was determined by Western blotting (WB).
GeminiX2 electrotransmitter (BTX, Holliston, MA, USA) was employed to transfect siRNA into MM cells. MM cells (5 × 105) were resuspended in 500 μL BTXpress Cytoporation Media T4 (47‐0003, BTX) and siRNA was added to a final concentration of 100 nmol/L, then transferred to electric shock cup to mix and transfect. The sequences of siRNAs are listed in Supplementary Table S2.
Cells or tissues were lysed by a Radio immunoprecipitation assay (RIPA) buffer in the presence of protease inhibitor cocktail (20124ES03, Yeasen). Protein concentration was quantified using BCA Protein Assay kit. Total protein (20 μg) was loaded, fractionated by Sodium Dodecyl Sulfate PolyAcrylamide Gel Electrophoresis (SDS‐PAGE) and transferred to polyvinylidene fluoride (PVDF) membrane. After blocked with 5% non‐fat milk at 25°C for 1 h, the membranes were incubated with primary antibody overnight at 4°C, and horseradish peroxidase (HRP) conjugated secondary antibody for 1 h at 25°C. Blots were developed by using the Super ECL Detection Reagent. The following antibodies were used: AIMP1 (11091‐1‐AP, Proteintech), ANP32A (15810‐AP, Proteintech), ERK1/2 (4695S, Cell Signaling Technology, Danvers, MA, USA), p‐ERK1/2 (4370S, Cell Signaling Technology), acetyl‐histone antibody sampler kit (9933, Cell Signaling Technology), nuclear factor of activated T cells 1 (NFATc1, sc‐7294, SANTA, Santa Cruz, CA, USA), acetylated histone H3 (06‐559, Merk Millipore, Boston, MA, USA), Rabbit IgG (7074P2, Cell Signaling Technology), DYKDDDDK (FLAG, 14793S, Cell Signaling Technology), GRB2 associated regulator of MAPK1 subtype 2 (GAREM2, PA5‐20846, Invitrogen, Carlsbad, CA, USA), β‐actin (66009‐1‐Ig, Proteintech), goat anti‐mouse IgG‐HRP (sc‐2005, SANTA), goat anti‐rabbit IgG‐HRP (7074, Cell Signaling Technology). Co‐IP was performed according to the protocol of the Pierce™ Direct Magnetic IP/Co‐IP kit (88828, Thermo Fisher Scientific, Waltham, MA, USA). Bicinchoninic Acid Assay (BCA) Protein Assay kit (20201ES76) and Super Enhanced chemiluminescence (ECL) Detection Reagent (36208ES60) was purchased from Yeasen.
AIMP1 was cloned into a pET28a vector (TranSheepBio) and expressed as a His‐tag fusion protein in E. coli BL21 (DE3) and then purified by nickel affinity chromatography (AKTA Pure, Cytiva, Washington, USA). MM cells were treated with AIMP1 protein (100 nmol/L) for 2 h to measure protein levels of ERK1/2 and p‐ERK1/2.
Cell viability was detected by using 3‐(4,5‐dimethylthiazol‐2‐yl)‐2,5‐diphenyltetrazolium bromide assay (MTT, M8180, Solarbio, Beijing, China). Briefly, AIMP1‐WT, AIMP1‐OE and AIMP1‐KD cells were cultured for a designated length of time in 96‐well plates at a density of 8 × 103 cells/well with 6 duplicated wells in each group. The relative cell viability was calculated as the ratio of absorbance at a certain time relative to the mean value of 24‐hour absorbance. As to detect the effect of GDC‐0994 (S80364, Yuanye, Shanghai, China) on MM cell proliferation, H929 and OCI‐MY5 cells were treated with different concentrations of GDC‐0994 for 48 h. Vehicle‐treated control cells (medium with 1% DMSO) were considered 100% viable, and GDC‐0994‐treated cells were compared these control cells. Absorbance was measured at 570 nm by a microplate reader (Thermo Fisher Scientific).
The quantity and purity of total RNA from AIMP1‐WT, AIMP1‐OE and AIMP1‐KD and AIMP1 protein‐stimulated cells were analyzed by Bioanalyzer 2100 and RNA 6000 Nano LabChip Kit (Agilent, Palo Alto, CA, USA). After purification, the poly(A)‐ or poly(A)+ RNA fractions were fragmented into small pieces using divalent cations under elevated temperature. Then the cleaved RNA fragments were reverse‐transcribed to create the final cDNA library in accordance with the protocol for the mRNA‐Seq sample preparation kit (Illumina, San Diego, CA, USA). Next, we performed the paired‐end sequencing on an Illumina Novaseq™ 6000 (LC Sciences, Houston, TX, USA) according to the vendor's recommended protocol. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was performed by comparing AIMP1‐WT cells with AIMP1‐OE cells, AIMP1‐KD cells and AIMP1 protein‐stimulated cells. More detailed information has been uploaded in the GEO database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162700).
A human proteome chip was established by Guangzhou Bochong Biotechnology Co., Ltd. (Guangzhou, Guangdong, China). The technical route included chip sealing, sample incubation, chip cleaning and chip scanning. Interaction screening of AIMP1 recombinant protein was based on the HuProt™ human whole‐proteome chip (Bochong Biotechnology). After the chip was co‐incubated with Cy3‐labeled AIMP1 protein, it was scanned for bioinformatics analysis. Gene Ontology (GO) categorizations, such as Biological Process, Cellular Component and Molecular Function, were performed to identify proteins interacting with AIMP1 based on the GO database [43, 44].
The Corning® Transwell® 12 mm, 0.4 μm pore polyester membrane insert cell culture system (3460, CORNING, Corning, NY, USA) was used to co‐culture AIMP1‐OE H929 cells with RAW264.7 cells. Briefly, 1 × 105 AIMP1‐OE H929 cells were seeded in the lower chamber, and 5 × 104 RAW264.7 cells were seeded in the upper chamber. After 48 h, RAW264.7 cells were collected for WB analysis.
Osteoclastogenesis assays were performed as described previously [32]. RAW264.7 cells were seeded in 48‐well plates at a density of 5 × 103 cells/well and treated with recombinant murine RANKL (50 ng/mL, PeproTech, Cranbury, NJ, USA) and macrophage‐colony stimulating factor (M‐CSF, 10 ng/mL, PeproTech) on Day 2. Then, the cells were treated with AIMP1 protein (50 and 100 nmol/L) for 6 days. Afterwards, the cells were stained for TRAP activity using the Leukocyte Tartrate‐Resistant Acid Phosphatase kit (Millipore Sigma, Burlington, MA, USA) according to the manufacturer's protocol.
All ChIP‐seq processes were performed by Lc‐Bio Technologies (Hangzhou) Co., Ltd. (Hangzhou, Zhejiang, China). Reads from ChIP‐seq were aligned using Bowtie2 [45], and enriched regions were analyzed using MACS1.4 [46]. The motif detection of the binding peak was performed by MEME. Differential enrichment regions were identified by diffReps. More detailed information has been uploaded in the GEO database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162742).
ChIP was conducted using a Chromatin IP Kit (9003, Cell Signaling Technology). Acetyl H3‐ChIP was followed by real‐time quantitative PCR (RT‐qPCR) using SYBR Green Master Mix (11198ES03, Yeasen). The sequences of GRB2‐associated and regulator of MAPK protein 2 (GAREM2) primers were as follows: forward 5’‐TCACAGAGCATTGGTAGAT‐3’ and reverse 5’‐AGCAGAGGACAAAGGAAA‐3’. The qPCR thermocycling program was as follows: initial denaturation at 95˚C for 3 min, followed by 40 cycles of denaturation at 95˚C for 10 s and annealing at 60˚C for 59 s. The amount of immunoprecipitated DNA in each sample is presented as the signal relative to the total amount of input chromatin.
The supernatant of AIMP1‐WT OCI‐MY5 cells was collected and centrifuged at 300 × g for 10 min, 2000 × g for 10 min, 10000 × g for 30 min as to remove floating cells and debris. The remaining supernatant was centrifuged in an ultracentrifuge at 100,000 × g for 90 min. At last, the collected precipitate was resuspended in 200 μL phosphate buffered saline (PBS) and stored at ‐80˚C.
A Neon electroporation system (Bio‐Rad GenePulser Xcell, Bio‐Rad, Hercules, CA, USA) was utilized in this study. The exosomes were mixed with PBS at a 1:1 ratio. FAM‐labeled siAIMP1 (Sangon Biotech, Shanghai, China) was added to the mixture at final concentrations of 200 nmoL/L for FAM‐siAIMP1 and 20 μg/mL for exosomes. Then electroporation was performed three times at 100 V, 50 μF with the pulse width set at 30 ms with a 2‐second pause in‐between. Following electroporation, one unit of RNase A was added to the mixture to eliminate free siRNA outside the exosomes. Ethylene diamine tetraacetie acid (EDTA) was added to reduce the undesirable electroporation‐induced siRNA precipitation during the loading process [47, 48]. A total of 1 × 106 MM cells were treated with 20 μg siAIMP1‐loaded exosomes, 0.2 nmol/L free siRNA, or nothing (control) for 48 h to measure protein levels of ERK1/2 and p‐ERK1/2.
AIMP1‐WT and AIMP1‐OE OCI‐MY5 cells (2 × 106) were injected subcutaneously into the abdominal left and right flank of 6‐ to 8‐week‐old NOD/SCID mice (n = 6) (GemPharmatech LLC., Nanjing, Jiangsu, China), respectively. Tumor diameter was measured with calipers 2‐3 times weekly. After 28 days, the mice were euthanized by spinal dislocation when the tumor diameter reached 15 mm. The tumors were isolated, weighed, and imaged. The tumor volumes were calculated using the formula: 0.52 × larger diameter × (smaller diameter)2.
AIMP1‐WT and AIMP1‐OE OCI‐MY5 cells (1 × 107/10 μL) were injected subcutaneously into the BM cavity of left and right tibias of 6‐ to 8‐week‐old NOD/SCID mice (n = 10). The mice were randomly divided into two groups (n = 5 per group), and tail intravenous injection of siAIMP1‐loaded exosomes (1.5 mg/kg) or PBS was performed every 3 days. After 1 month, the mice were euthanized by spinal dislocation. The bone mineral density (BMD) and bone volume faction (bone volume / total volume [BV/TV]) of the tibia were analyzed by micro‐computed tomography (micro‐CT) (SkyScan 1176, Bruker microCT, Salbruken, SL, Germany) to assess osteolysis.
To explore the inhibitory role of siAIMP1‐loaded exosomes in tumor growth in vivo, a PDX model was generated using the biopsy sample from a MM patient with a brain manifestation at the Department of Hematology, The First Affiliated Hospital of Nanjing Medical University (Nanjing, Jiangsu, China). The MM patient was a 67‐year old man with IgA‐κ type, Durie‐Salmon stage III stage B group, International Staging System (ISS) stage III MM diagnosed in September 2014. The patient characteristics are listed in Supplementary Table S3. The patient provided written informed consent for the biopsy sample to be used for research. Freshly excised tumor tissue specimens were washed and cut into small pieces in antibiotic‐containing RPMI‐1640. Tumor slices were transplanted into 4‐ to 6‐week‐old male NOD/SCID mice (n = 12). Once the tumor volume reached 100‐150 mm3, siAIMP1‐loaded exosomes (1.5 mg/kg) or PBS were injected via the tail vein every 3 days (n = 6 per group). The mice were euthanized by spinal dislocation on Day 30. This model was used to evaluate the impact of siAIMP1‐loaded exosomes on MM growth in vivo. All animal studies were conducted in accordance with the Jiangsu Province People's Government‐published recommendations for the Care and Use of Laboratory Animals and approved by the Institutional Ethics Review Boards of Nanjing University of Chinese Medicine (ethics number: 201905A003).
Bilateral tibia specimens from NOD/SCID TIBIA mouse model were fixed in 10% neutral formalin for 3 days, and then put in EDTA decalcifying solution (pH = 7.2) for 2 weeks. The decalcifying solution was replaced every 3 days. After decalcification, the tissue was rinsed with running water for 30 min, then embedded in dehydrated paraffin for sectioning. The sections were routinely deparaffinized to water. H&E staining was performed on paraffin tissue sections. The main process was as follows: hematoxylin dye solution, 10 min; water, 2 min; 1% hydrochloric acid and ethanol, 15 s; water, 2 min; dilute ammonia, 1 min; water, 2 min; 80% ethanol, 1 min; eosin, 2 min; water, 1 min; dehydrated by gradient ethanol, transparent xylene, and finally sealed with gum.
All values are presented as the mean ± standard deviation (SD). A two‐tailed Student's t‐test or a one‐way analysis of variance (ANOVA) (≥ 3 groups) were used to compare the differences between groups. Kaplan‐Meier curves were used to evaluate the associations of AIMP1 and acidic leucine‐rich nuclear phosphoprotein 32 family member A (ANP32A) expression with MM patient overall survival (OS), event‐free survival (EFS) or progression‐free survival (PFS) from diagnosis. Spearman's rank correlation analysis was used to evaluate the correlation between AIMP1 and Ki67 protein expression in IHC. P < 0.05 was considered to indicate a statistically significant difference.
We analyzed the GEP dataset to explore AIMP1 expression in plasma cells from healthy individuals, patients with monoclonal gammopathy of undetermined significance (MGUS) and patients with newly diagnosed MM. Intriguingly, mRNA expression of AIMP1 was significantly higher in the plasma cells from patients with newly diagnosed MM (n = 351) than in those from patients with MGUS (n = 44) and healthy individuals (n = 22) (P < 0.001) (Figure 1A). In addition, increased AIMP1 expression was associated with significantly shorter overall survival (OS) in the TT2 cohort (P = 0.008) and APEX cohort (P < 0.001) (Figure 1B). Moreover, increased AIMP1 expression was associated with significantly shorter event‐free survival (EFS) in the TT2 cohort (P < 0.001) (Figure 1C) and progression‐free survival (PFS) in the APEX cohort (P = 0.049) (Figure 1D). We also detected AIMP1 protein levels in the serum by ELISA, and the results showed that the levels of AIMP1 were significantly higher in serum samples from MM patients (n = 19) than that in the serum from healthy subjects (n = 11) (Figure 1E). Subsequently, IHC staining was performed to examine AIMP1 protein expression in BM samples. AIMP1‐staining levels in BM samples from MM patients (n = 12) were significantly higher than those from healthy subjects (n = 9) (P < 0.001) (Figure 1F‐G), while AIMP1 expression was positively associated with Ki67 expression (P < 0.001, R = 0.772) (Figure 1H). Therefore, we infer that AIMP1 may act as an oncogene associated with short OS in MM patients.
To determine whether AIMP1 could promote MM cell proliferation, we functionally overexpressed and knocked down AIMP1 in MM cells, and the transfection efficiency was validated by WB (Figure 2A). MTT assays showed that cell proliferation was increased in AIMP1‐OE cells and decreased in AIMP1‐KD cells relative to AIMP1‐WT cells (Figure 2B‐C). Additionally, upon treatment with exogenous AIMP1 protein (100 nmol/L), the proliferation in four MM cell lines was significantly increased (Figure 2D). In addition to the above in vitro experiments, we further confirmed these observations in vivo. The results indicated that AIMP1‐OE cells generated larger tumors compared with AIMP1‐WT cells (Figure 2E); the mean volume and weight of the AIMP1‐OE tumors were also significantly increased compared with the AIMP1‐WT tumors (P < 0.05) (Figure 2F‐G). WB analysis confirmed that AIMP1 expression was increased in the AIMP1‐OE tumors (Figure 2H). These data suggest that elevated AIMP1 accelerates MM cell proliferation in vitro and in vivo.
To identify the specific pathways regulated by AIMP1, we employed RNA‐seq to assess the potential oncogenic mechanism of AIMP1 in MM cells (GSE162700). The Venn diagrams showed that there were 36 co‐upregulated genes in two AIMP1‐OE cell lines and 116 co‐upregulated genes in two AIMP1‐treated MM cell lines compared with the AIMP1‐WT cells (Figure 3A). KEGG analysis indicated that osteoclast differentiation and the MAPK signaling pathway were significantly dysregulated in AIMP1‐OE and AIMP1‐KD cells compared with AIMP1‐WT cells (Figure 3B). Furthermore, we performed WB to detect the key markers, ERK1/2 and p‐ERK1/2, of the MAPK signaling pathway in AIMP1‐OE and AIMP1‐KD cells, and their expressions were compared to AIMP1‐WT cells. The results revealed that overexpression of AIMP1 increased the expression level of p‐ERK1/2, while knockdown of AIMP1 decreased p‐ERK1/2 expression (Figure 3C‐D). Consistently, treatment with exogenous AIMP1 protein also activated p‐ERK1/2 expression in MM cells (Figure 3E‐F). To further explore the vital role of AIMP1 in the MAPK signaling pathway, we employed GDC‐0994, a highly‐selective inhibitor of ERK1/2 [49], to treat AIMP1‐WT and AIMP1‐OE cells. MTT assays showed that AIMP1‐OE cells were more resistant to cell inhibition induced by GDC‐0994 than AIMP1‐WT cells (Supplementary Figure S1). These findings demonstrate that AIMP1 promotes MM cell proliferation via the MAPK signaling pathway.
MM patients often suffer from bone pain, bone damage, pathological fractures and hypercalcemia; the primary cause is that the infiltration of MM cells stimulates osteoclasts and enhances osteolysis [3]. AIMP1 increases osteoclastogenesis of macrophages via modulation of RANKL [32]. As RNA‐seq analysis confirmed that AIMP1 was involved in osteoclast differentiation (Figure 3B), we next examined the GEP dataset associated with osteoclast differentiation (GSE6401). We found that AIMP1 signaling was increased in patients with serious bone lesions (n = 54) compared with mild bone lesions (n = 42) (P < 0.05) (Figure 4A). The detailed information on the GSE6401 dataset is listed in Supplementary Table S1. To explore whether AIMP1 could be secreted from MM cells, we co‐cultured RAW264.7 macrophages with AIMP1‐OE H929 cells for 48 h, and then detected the expression of AIMP1 expression in co‐cultured RAW264.7 cells compared with un‐co‐cultured RAW264.7 cells. We found a weak expression of flag in co‐cultured RAW26.7 cells (Figure 4B), suggesting that AIMP1 was secreted from MM cells to affect AIMP1 expression in RAW264.7 cells. In addition, we overexpressed AIMP1 in RAW264.7 macrophages (Figure 4C) to verify whether AIMP1 could promote bone lesion formation. TRAP staining and quantitative analysis of multinucleated osteoclasts showed that increased AIMP1 expression enhanced osteoclast differentiation of macrophages when treated with RANKL (50 ng/mL) and M‐CSF (10 ng/mL) (P < 0.001) (Figure 4D‐E). We further detected the expression of NFATc1, an osteoclast marker [50], in RAW264.7 cells. WB results showed that NFATc1 expression was markedly elevated in AIMP1‐OE RAW264.7 cells (Figure 4F). Additionally, exogenous AIMP1 protein also promoted osteoclast differentiation (P < 0.05) (Figure 4G‐H) and stimulated NFATc1 expression (Figure 4I). These data suggest that AIMP1 is a vital activator of osteoclast differentiation.
HuProt™ human proteomics chip V3.1 contains > 20,000 newly sequenced recombinant human proteins, which is currently the highest throughput protein chip and suitable for global protein‐protein interaction analyses [51, 52]. To identify the potential client proteins of AIMP1, we performed protein chip analysis with the purified AIMP1 protein. As shown in Figure 5A and Supplementary Table S4, 127 proteins interacting with AIMP1 were identified by the microarray. Based on the pathway analysis of these proteins in the KEGG database, we found that they were also significantly enriched in the MAPK pathway (Figure 5B), consistent with the RNA‐seq results (Figure 3B). Additionally, these proteins mainly possess protein polyubiquitin and histone‐binding functions (Figure 5C). Among the 127 screened‐out proteins according to the experimental standards, we analyzed the top 15 proteins (Figure 5D) based on the GEP dataset of healthy individuals, MGUS, and MM plasma cells to explore their association with MM. ANP32A, a member of the inhibitor of the histone acetyltransferase complex [53], was significantly increased in patients with newly diagnosed MM (n = 351) relative to patients with MGUS (n = 44) and healthy individuals (n = 22) (P < 0.001) (Figure 5E). MM patients with high ANP32A expression exhibited shorter OS in the TT2 (P = 0.023), HOVON65 (P < 0.001) and APEX cohorts (P < 0.001) (Figure 5F). Co‐IP assays indicated the physical interaction between AIMP1 and ANP32A in MM cells (Figure 5G). It has been reported that ANP32A can regulate histone H3 acetylation [54]. Thus, whether AIMP1 could interact with ANP32A to regulate histone H3 acetylation in MM was next assessed. We tested acetyl‐H2A, acetyl‐H2B, acetyl‐H3, and acetyl‐H4 protein levels in AIMP1‐OE and AIMP1‐KD cells compared with AIMP1‐WT cells. The results showed that the expression levels of acetyl‐histones were notably altered (Supplementary Figure S2), but only acetyl‐H3 was upregulated in AIMP1‐OE cells and downregulated in AIMP1‐KD cells compared to AIMP1‐WT cells, respectively (Figure 5H). In addition, exogenous AIMP1 protein treatment also activated acetyl‐H3 expression in MM cells (Figure 5I). To further confirm the role of ANP32A in MM cells, we interfered ANP32A expression via siRNA (siANP32A) in AIMP1‐OE cells. WB analysis confirmed that siANP32A decreased the expression of acetyl‐H3 (Figure 5J). MTT assays indicated that siANP32A suppressed MM cell proliferation in AIMP1‐OE cells (P < 0.001, Figure 5K). The above results show the vital role of the interaction of AIMP1 with ANP32A in regulating histone H3 acetylation, thus promoting MM progression.
As it had been confirmed that AIMP1 interacted with ANP32A to regulate acetyl‐H3 level, we further performed ChIP‐seq for acetyl‐H3 in AIMP1‐WT and AIMP1‐OE H929 cells to screen potential downstream targets (GSE162742). The schematic of the ChIP‐seq is shown in Figure 6A. KEGG pathway analysis indicated that AIMP1‐regulated acetyl‐H3 was also associated with the MAPK signaling pathway and osteoclast differentiation (Figure 6B). To narrow down the genes regulated by AIMP1/ANP32A‐mediated acetyl‐H3, we focused on the genes with increased acetyl‐H3 enrichment levels in AIMP1‐OE cells compared to AIMP1‐WT cells, which were involved in the MAPK signaling pathway. GAREM2 was considered a unique gene, as it was involved in the MAPK signaling pathway and associated with increased acetyl‐H3 enrichment function (Figure 6C‐D). Furthermore, ChIP‐qPCR confirmed that GAREM2 binding to acetyl‐H3 was significantly increased in AIMP1‐OE cells compared with AIMP1‐WT cells (P < 0.01, P < 0.05) (Figure 6E). To further clarify the function of GAREM2 in MM cells, we interfered GAREM2 expression via siRNA (siGAREM2) in H929 and OCI‐MY5 cells. WB results showed that siGAREM2 evidently decreased p‐ERK1/2 levels (Figure 6F). MTT assay indicated that siGAREM2 suppressed MM cell proliferation significantly (P < 0.01) (Figure 6G). Collectively, these data reveal that GAREM2 is partially involved in the process of the AIMP1‐mediated promotion in MM proliferation.
Exosomes, as natural carriers, are often used as drug delivery vehicles to transport siRNA to target tissues or cells and inhibit tumor development [55, 56, 57]. To further explore the clinical value of targeting AIMP1 in MM treatment, we utilized siAIMP1‐loaded exosomes to verify its functions. First, we designed a siRNA targeting AIMP1 and confirmed that siAIMP1 decreased AIMP1 expression at the protein level (Figure 7A), and the cell proliferation rate was shown to be attenuated by siAIMP1 (P < 0.05) (Figure 7B). Next, we employed exosomes derived from OCI‐MY5 cells to load siAIMP1. MTT assays showed that siAIMP1‐loaded exosomes significantly inhibited MM cell proliferation (P < 0.05) (Figure 7C). Additionally, siAIMP1‐loaded exosomes decreased p‐ERK1/2 expression (Figure 7D), consistent with the results shown in Figure 3. In addition, we employed the PDX model to assess the effects of siAIMP1‐loaded exosomes in vivo. The siAIMP1‐loaded exosomes were injected via the tail vein in the PDX mice, and the growth of tumors was inhibited apparently (Figure 7E). Consistently, the mean volume and weight of tumors in the siAIMP1‐loaded exosomes group were significantly lower than those in the control group (P < 0.05) (Figure 7F‐G). Taken together, the above results suggest that siAIMP1‐loaded exosomes suppress MM cell proliferation in vitro and in vivo, suggesting that targeting AIMP1 may be a promising strategy to prevent MM progression.
As AIMP1 plays a role in promoting osteoclast differentiation, we next assessed whether siAIMP1‐loaded exosomes could inhibit osteoclast differentiation. We found that the differentiation capacity of RAW264.7 macrophages into osteoclasts was decreased upon treatment with siAIMP1‐loaded exosomes (P < 0.05) (Figure 8A‐B). Next, we explored the impact of siAIMP1‐loaded exosomes on the BM microenvironment in the NOD/SCID‐TIBIA mouse model. The symptoms of swelling and foot valgus in the siAIMP1‐loaded exosomes group were milder than those in the control group (Figure 8C). Then, micro‐CT analysis indicated that the bone destruction was more severe in the AIMP1‐OE group than in the AIMP1‐WT group and siAIMP1‐loaded exosomes significantly reduced bone destruction (Figure 8D). BMD and BV/TV were significantly decreased in the AIMP1‐OE group compared with the AIMP1‐WT group, while siAIMP1‐loaded exosomes increased BMD and BV/TV in the experimental mice (P < 0.05) (Figure 8E‐F). In addition, H&E staining indicated that a higher density of tumor cells was presented in the AIMP1‐OE group compared with the AIMP1‐WT group, whereas a lower density of tumor cells was shown in the siAIMP1‐loaded exosome group (Figure 8G). TRAP staining indicated that elevated AIMP1 increased the portion of multinucleated osteoclasts, while siAIMP1‐loaded exosomes reduced the proportion of multinucleated osteoclasts significantly (Figure 8H). These findings demonstrate that siAIMP1‐loaded exosomes can affect the BM microenvironment in vitro and in vivo.
MM is a type of plasma cell dyscrasia that exhibits significant heterogeneity, and novel targets are needed to improve the management of this condition. In this study, we investigated a novel target AIMP1, which could regulate MM cell proliferation and osteoclast differentiation. AIMP1 promoted MM malignancy via promoting histone H3 acetylation and activating p‐ERK1/2 (Figure 9). In addition, AIMP1 acted as a secretory protein to promote osteoclast differentiation. Furthermore, we demonstrated that high expression of AIMP1 was associated with short OS in MM patients. Increasing evidence has shown that secretory AIMP1 activates monocytes/macrophages via signaling cascades including MAPK, ERK, and NF‐κB, leading to the secretion of proinflammatory cytokines, such as TNF, interleukin‐8 (IL‐8), and macrophage chemotactic protein‐1 (MCP‐1) [33, 34]. The MAPK signaling pathway is involved in cell proliferation, differentiation, apoptosis, inflammation, and innate immunity [58, 59]. In this study, the RNA‐seq and WB data confirmed that AIMP1 significantly activated the MAPK signaling pathway to promote MM cell proliferation. To further determine the potential underlying molecular mechanism of AIMP1 in MM, proteomic chip analysis and subsequent experiments were performed that demonstrated AIMP1 interacting with ANP32A. ANP32A is a multifunctional protein involved in regulating histone acetylation and increasing mRNA stability, which promotes cell proliferation, migration and invasion via the high mobility group AT‐hook 1/signal transducer and activator of transcription 3 (HMGA1/STAT3) pathway in hepatocellular carcinoma [60]. ANP32A inhibits p38 and activates the AKT signaling pathway to promote colorectal cancer cell proliferation [61]. In addition, ANP32A acts as a regulator of histone H3 acetylation to promote leukemogenesis and contributes to poor outcomes in acute myeloid leukemia [62, 63, 64]. We found that AIMP1 could regulate histone H3 acetylation, which might be due to the interaction between AIMP1 and ANP32A. The ChIP‐seq and following experiments confirmed that increased enrichment function of acetyl‐H3 upon AIMP1 overexpression significantly increased the expression of GAREM2, which could also mediate ERK activation [65, 66]. Notably, functional analysis revealed that GAREM2 was a key downstream target of AIMP1. Therefore, disruption of AIMP1 to modulate acetyl‐H3 represents a novel therapeutic strategy for the treatment of MM. Osteoclasts are formed by the differentiation and fusion of monocytes/macrophages, which serve as osteoclast precursor cells [67]. We overexpressed AIMP1 in RAW264.7 cells and treated RAW264.7 cells with exogenous AIMP1 protein, both applications promoted the formation of osteoclasts and activated NFATc1 expression in vitro. NFATc1 serves as a foremost transcriptional factor of osteoclast differentiation whose function can be induced by RANKL [68, 69]. As a member of the TNF family, RANKL is a critical osteoclastogenic cytokine [70]. RANKL stimulation induces Ca2+ oscillation to promote the sustained activation of NFATc1 via a calcineurin‐dependent mechanism [68]. Elevated expression of NFATc1 abrogates the effect of IL‐10‐inhibited osteoclastogenesis induced by RANKL [71]. RANKL binding to its receptor can recruit TNF receptor‐associated factor 6 (TRAF6) activating the downstream signaling molecules including MAPKs, NF‐κB, activator protein 1 (AP‐1) and NFATc1, so as to induce osteoclast differentiation and the formation of multinucleated osteoclasts [72, 73, 74]. We confirmed that AIMP1 regulated MM cell proliferation via the MAPK signaling pathway, and AIMP1 cooperatively functioned with RANKL to activate NFATc1 promoting osteoclastogenesis. Exosomes have emerged as a novel delivery system for biotherapeutic and diagnostic molecules, including siRNAs, due to their unique size, structure, and capability in allowing intercellular communication [75]. Exosome‐based immunotherapy becomes a promising strategy for cancer treatment [76]. For example, exosome‐mediated siRNA delivery inhibits postoperative breast cancer metastasis [55], and siPAK4‐loaded exosomes prolong the survival of mice in a model of pancreatic cancer [77]. In addition, RNA nanotechnology has been adopted to remodel the endogenous extracellular vesicles to specifically deliver siRNA to cancer cells to induce cancer regression [78]. In the present study, we used siAIMP1‐loaded exosomes to treat MM cells, and they inhibited MM cell proliferation and osteoclast differentiation in vitro. Additionally, siAIMP1‐loaded exosomes inhibited tumor growth in PDX mice and decreased bone destruction in vivo. Collectively, we clarified that siAIMP1‐loaded exosomes effectively inhibited MM cell proliferation, bone lesion formation, and the pathological changes by remodeling the BM microenvironment. Furthermore, we identified that AIMP1 interacted with ANP32A to regulate histone H3 acetylation and promoted MM malignancy via modulation of the MAPK signaling pathway. AIMP1 also promoted osteoclast differentiation by activating NFATc1. However, the underlying molecular mechanism of AIMP1 in promoting histone H3 acetylation and osteoclast differentiation remains unclear and should be further explored.
AIMP1 is shown to be a novel oncogene in MM, which promotes MM development by interacting with ANP32A to alter acetyl‐H3 enrichment function of GAREM2, thus activating the MAPK signaling pathway. Targeting AIMP1 with siAIMP1‐loaded exosomes may thus be a novel strategy in the treatment of MM.
The samples for IHC were collected from the Affiliated Hospital of Nanjing University of Chinese Medicine (Ethics number: KY2018005). All patients provided written informed consent for their bone marrow tissue samples to be used for research. The serum samples for ELISA were collected in the Affiliated Hospital of Nantong University (Ethics number: 2018‐K007). All participants provided written informed consent. All animal work was performed in accordance with government‐published recommendations for the Care and Use of laboratory animals and guidelines of Institutional Ethics Review Boards of Nanjing University of Chinese Medicine (Ethics number: 201905A003).
Informed consents were received from patients who participated in this study.
YY, XG and CG designed the project, integrated the data and edited the manuscript; RW drafted the manuscript; RW, YZ (Yan Zhu), YZ (Yuanjiao Zhang), WZ, XY and LW performed the experiments and analyzed the data. All authors have read and approved the final version of the manuscript.
No potential conflicts of interest were disclosed.
Click here for additional data file. Click here for additional data file. |
PMC9648397 | 36381195 | Haiyan Wang,Amy Zheng,Edward B. Arias,Gregory D. Cartee | Prior AICAR induces elevated glucose uptake concomitant with greater γ3-AMPK activation and reduced membrane cholesterol in skeletal muscle from 26-month-old rats | 19-05-2022 | insulin sensitivity,glucose transport,AMP-activated protein kinase,aging,cholesterol,skeletal muscle | Attenuated skeletal muscle glucose uptake (GU) has been observed with advancing age. It is important to elucidate the mechanisms linked to interventions that oppose this detrimental outcome. Earlier research using young rodents and (or) cultured myocytes reported that treatment with 5-aminoimidazole-4-carboxamide-1-β-d-ribofuranoside (AICAR; an AMP-activated protein kinase (AMPK) activator) can increase γ3-AMPK activity and reduce membrane cholesterol content, each of which has been proposed to elevate GU. However, the effect of AICAR treatment on γ3-AMPK activity and membrane cholesterol in skeletal muscle of aged animals has not been reported. Our purpose was to evaluate the effects of AICAR treatment on these potential mechanisms for enhanced glucose uptake in the skeletal muscle of aged animals. Epitrochlearis muscles from 26–27-month-old male rats were isolated and incubated ± AICAR, followed by 3 h incubation without AICAR, and then incubation with 3-O-methyl-[3 H] glucose (to assess GU ± insulin). Muscles were also analyzed for γ3-AMPK activity and membrane cholesterol content. Prior AICAR treatment led to increased γ3-AMPK activity, reduced membrane cholesterol content, and enhanced glucose uptake in skeletal muscle from aged rats. These observations revealed that two potential mechanisms for greater GU previously observed in younger animals and (or) cell models are also potentially relevant for enhanced GU by muscles from older animals. | Prior AICAR induces elevated glucose uptake concomitant with greater γ3-AMPK activation and reduced membrane cholesterol in skeletal muscle from 26-month-old rats
Attenuated skeletal muscle glucose uptake (GU) has been observed with advancing age. It is important to elucidate the mechanisms linked to interventions that oppose this detrimental outcome. Earlier research using young rodents and (or) cultured myocytes reported that treatment with 5-aminoimidazole-4-carboxamide-1-β-d-ribofuranoside (AICAR; an AMP-activated protein kinase (AMPK) activator) can increase γ3-AMPK activity and reduce membrane cholesterol content, each of which has been proposed to elevate GU. However, the effect of AICAR treatment on γ3-AMPK activity and membrane cholesterol in skeletal muscle of aged animals has not been reported. Our purpose was to evaluate the effects of AICAR treatment on these potential mechanisms for enhanced glucose uptake in the skeletal muscle of aged animals. Epitrochlearis muscles from 26–27-month-old male rats were isolated and incubated ± AICAR, followed by 3 h incubation without AICAR, and then incubation with 3-O-methyl-[3 H] glucose (to assess GU ± insulin). Muscles were also analyzed for γ3-AMPK activity and membrane cholesterol content. Prior AICAR treatment led to increased γ3-AMPK activity, reduced membrane cholesterol content, and enhanced glucose uptake in skeletal muscle from aged rats. These observations revealed that two potential mechanisms for greater GU previously observed in younger animals and (or) cell models are also potentially relevant for enhanced GU by muscles from older animals.
Epidemiological data indicate that the prevalence of prediabetes and diabetes is progressively increased with advancing age (Cowie et al. 2009). Skeletal muscle is the tissue responsible for the major portion of insulin-mediated glucose disposal, and insulin resistance is an essential defect contributing to the development of type 2 diabetes (DeFronzo et al. 2015). These circumstances motivate efforts to identify and understand interventions that improve glucose uptake by skeletal muscle at advanced ages. Previous exercise research has demonstrated that skeletal muscle retains a considerable reserve capacity to increase muscle glucose uptake even during old age (Cartee et al. 1993; Xiao et al. 2013; Sharma et al. 2015; Cartee et al. 2016; Oki et al. 2020). However, not all older individuals have the ability and (or) willingness to perform sufficient exercise to gain this important health benefit. Accordingly, it is worthwhile to explore other strategies to enhance muscle glucose uptake during advanced age. Strong evidence links the activation of skeletal muscle AMPK (adenosine monophosphate-activated protein kinase) to elevated glucose uptake by skeletal muscle. Fisher et al. (2002) reported that the incubation of isolated rat skeletal muscle with the AMPK activator AICAR (5-aminoimidazole-4-carboxamide-1-β-d-ribofuranoside) results in subsequently increased insulin-stimulated glucose uptake. AMPK is a heterodimeric enzyme that is comprised of alpha (α1 or α2), beta (β1 or β2), and gamma (γ1, γ2, or γ3) subunits (Kjobsted et al. 2016; Kjobsted et al. 2018). The γ3 subunit is notable because it is almost exclusively expressed in skeletal muscle, and the ability of AICAR-treatment to enhance insulin-stimulated glucose uptake by rodent skeletal muscle is absent in γ3-knockout (KO) mice (Mahlapuu et al. 2004; Kjobsted et al. 2015). These results provide compelling evidence that γ3-AMPK is essential for this important outcome. The enhanced insulin-stimulated glucose uptake in skeletal muscle several hours after AICAR treatment is accompanied by greater phosphorylation of a protein known as AS160 (also called Akt substrate of 160 kDa or TBC1D4) (Kjobsted et al. 2015). AS160 is a Rab-GTPase activating protein that plays a key role in regulating insulin-stimulated GLUT4 glucose transporter translocation and glucose uptake (Sano et al. 2003; Cartee 2015). Prior AICAR treatment leads to greater AS160 phosphorylation on key phosphosites that regulate insulin-stimulated glucose uptake, and the improvement in insulin-stimulated glucose uptake after AICAR treatment is not found in the skeletal muscle of AS160-KO mice (Kjobsted et al. 2019). These observations implicate AS160 as a key protein for AICAR-induced improvement in glucose uptake. We recently reported that prior treatment of skeletal muscle from old rats with AICAR resulted in increased insulin-stimulated glucose uptake that was accompanied by greater AS160 phosphorylation (Oki et al. 2018). However, the effect of AICAR on γ3-AMPK activity in the skeletal muscle of aged animals has not been previously reported. Therefore, our first aim was to determine if AICAR leads to greater γ3-AMPK activity in skeletal muscle from old rats. In addition to the evidence that AMPK-induced phosphorylation of AS160 is important for the elevated insulin-stimulated glucose uptake in skeletal muscle after AICAR treatment, there is also support for another possible mechanism to contribute to the AICAR/AMPK-dependent increase in glucose uptake. An inverse relationship has been observed between membrane cholesterol and glucose uptake by skeletal muscle (Grice et al. 2019; Habegger et al. 2012b; Sanchez-Aguilera et al. 2018). Furthermore, research using L6 skeletal muscle cells indicated that AICAR can stimulate AMPK, lower membrane cholesterol, and increase insulin-stimulated glucose uptake (Habegger et al. 2012a). Therefore, the second aim was to determine if AICAR treatment of skeletal muscle from old rats resulted in altered membrane cholesterol content. HMGCR (3-hydroxy-3-methylglutaryl coenzyme A reductase), the rate-limiting enzyme for cholesterol synthesis, is phosphorylated by AMPK on a site that regulates HMGCR activity (Clarke and Hardie 1990), and ABCA1 (ATP-binding cassette transporter A1) is an important protein for cellular cholesterol efflux (Wang and Tall 2003; Larrede et al. 2009). Accordingly, we also evaluated the effect of AICAR treatment on the phosphorylation of HMGCR and ABCA1 in the skeletal muscle of old rats. TBC1D1 is an AS160-paralog and Rab-GTPase-activating protein expressed by skeletal muscle that can be phosphorylated by AMPK in response to AICAR treatment and that regulates glucose uptake (Taylor et al. 2008; Cartee 2015; Chen et al. 2017). However, the effect of prior AICAR treatment on TBC1D1 phosphorylation in skeletal muscle of old rats has not been previously reported. Accordingly, our third aim was to determine prior AICAR treatment’s effect on phosphorylation of TBC1D1 on Ser237, a site implicated in AICAR-stimulated glucose uptake (Taylor et al. 2008; Chen et al. 2017). We also evaluated TBC1D1 Thr590, a site that can be phosphorylated in response to insulin (Pehmoller et al. 2009; Vichaiwong et al. 2010).
Chemicals were purchased from Sigma-Aldrich (St. Louis, MO) or Fisher Scientific (Hanover Park, IL) unless otherwise noted. The reagents and apparatus for SDS-PAGE and nonfat dry milk (no. 170-6404) were obtained from Bio-Rad (Hercules, CA). Pierce MemCode Reversible Protein Stain Kit (#24585), Bicinchoninic acid protein assay (#23225), Tissue Protein Extraction Reagent (TPER; #78510), Protein G magnetic beads (#10004D), and DynaMag™-2 magnet (#12321D) were from Thermo Fisher Scientific (Waltham, MA). Anti-phospho Akt Ser473 (pAktSer473; #9271), anti-phospho Akt Thr308 (pAktThr308; #13038), anti-Akt (#4691), anti-phospho AS160 Thr642 (pAS160Thr642; #8881), anti-phospho AS160 Ser588 (pAS160Ser588; #8730), anti-phospho AMPKα Thr172 (pAMPKaThr172; #2531), anti-AMP-activated protein kinase-α (AMPKα; #5831), anti-acetyl CoA carboxylase (ACC; #3676), anti-phospho ACC Ser79 (pACCSer79; #3661), anti-TBC1D1 (#5929), anti-hexokinase II (HKII; #2867), anti-insulin receptor (IR; #3025), anti-α-Tubulin (#2144), anti-Na+/K+-ATPase (#3010) and anti-rabbit IgG horseradish peroxidase conjugate (#7074) were from Cell Signaling Technology (Danvers, MA). Anti-phospho TBC1D1 Thr590 (pTBC1D1Thr590; #AF2422) was from Sapphire North America (Ann Arbor, MI). Anti-phospho AS160 Ser704 (pAS160Ser704) was provided by Dr. Jonas Thue Treebak (Novo Nordisk Foundation Center for Basic Metabolic Research, University of Copenhagen, Denmark). Anti AMP-activated protein kinase γ3 (γ3-AMPK) was provided by Dr. David Thomson (Brigham Young University, USA) (Hardman et al. 2014). Anti-Akt Substrate of 160 kDa (AS160; #ABS54), Anti-GLUT4 (GLUT4; #CBL243), Anti-phospho-TBC1D1 Ser237 (pTBC1D1Ser237; #07-2268), P81 Phosphocellulose Squares (#20-134) and enhanced chemiluminescence Luminata Forte Western HRP Substrate (#WBLUF0100) were from MilliporeSigma (Billerica, MA). Anti-HMGCR (HMGCR, #BS-5068 R) and anti-phospho HMGCR Ser872 (pHMGCRSer872, # BS-4063 R) were from Bioss Antibodies (Woburn, MA). Anti-ABCA1 (ABCA1, #NB400-105SS) was from Novus Biologicals (Littleton, CO). Anti-phospho ABCA1 Ser2054 (pABCA1Ser2054; #ab12064) was from Abcam (Cambridge, MA). 3-O-methyl-[3 H] glucose ([3 H]3-MG) was from Sigma-Aldrich, and [14C] mannitol was from PerkinElmer (Boston, MA). [γ-33 P]-ATP was from American Radiolabeled Chemicals, Inc. (St. Louis, MO). Liquid scintillation cocktail (#111195-CS) was from Research Products International (Mount Prospect, IL).
Animal care procedures were approved by the University of Michigan Committee on Use and Care of Animals and performed in accordance with the guidelines from the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, USA. Male Fischer-344 X Brown Norway rats were obtained from the National Institute of Aging (NIA) rodent colony at approximately 22–23 months old. Rats were individually housed at the University of Michigan (12:12 h light:dark cycle, lights out at 17:00 h), provided with standard rodent chow (Laboratory Diet no. 5L0D; LabDiet, St. Louis, MO) and water ad libitum. The terminal experiment was performed when the rats were approximately 26–27 months old, and the rats were fasted at approximately 17:00 on the night before the experiment.
Rats were deeply anesthetized by an intraperitoneal injection of ketamine–xylazine cocktail (50 mg/kg ketamine and 5 mg/kg xylazine), and both epitrochlearis muscles from each rat were dissected out and longitudinally split into two muscle strips. The muscle strips were placed in vials that were shaken at 45 oscillations per minute and continuously gassed (95% O2/5% CO2) in a heated (35 °C) water bath. For muscles analyzed to determine 3-MG uptake and signaling proteins, each muscle strip was incubated with a four-step process (Incubation Protocol 1). During step 1 (60 min), muscle strips were incubated with Krebs-Henseleit Buffer (KHB) supplemented with 8 mM glucose ± 2 mM AICAR (2 mM mannitol for without AICAR). During step 2 (180 min), all muscle strips were incubated in KHB supplemented with 8 mM glucose and 2 mM mannitol in the absence of AICAR. During step 3 (30 min), muscle strips were incubated with KHB supplemented with 0.1% bovine serum albumin (BSA), 2 mM sodium pyruvate, and 6 mM mannitol ± insulin (1.2 nM). During step 4 (20 min), muscle strips were incubated with KHB–BSA, the same concentration of insulin as previous step 3, 8 mM 3-MG (final specific activity of 250 μCi/mmol, [3 H]3-MG), and 2 mM mannitol (final specific activity of 50 μCi/mmol [14C]-mannitol). After step 4, the final incubation step, muscles were blotted, freeze-clamped, and stored at −80 °C for later processing and analysis. For muscles analyzed to determine γ3-AMPK activity, two muscle strips from one epitrochlearis muscle were incubated with KHB ± 2 mM AICAR (2 mM mannitol for without AICAR) for 60 min (Incubation Protocol 2), then the muscle strips were blotted, freeze-clamped, and stored at −80 °C for later analysis. Two muscle strips from the contralateral epitrochlearis muscle were incubated with a four-step process (Incubation Protocol 3) in which step 1 and step 2 were the same as Incubation Protocol 1. During steps 3 and 4 (30 min and 20 min), muscle strips were incubated with KHB supplemented with 0.1% BSA, 2 mM sodium pyruvate and 6 mM mannitol with insulin (1.2 nM). For muscles analyzed to determine membrane cholesterol and cholesterol regulatory proteins, each muscle strip from both epitrochlearis muscles was incubated with Incubation Protocol 3. After the final incubation step, muscles were blotted, freeze-clamped, and stored at −80 °C until later processing and analysis.
Frozen muscles were weighed and homogenized with 1 mL ice-cold lysis buffer using a glass pestle attached to motorized homogenizer (Caframo, Georgian Bluffs, ON). For muscle lysates analyzed to determine 3-MG uptake, protein abundance and phosphorylation by immunoblotting, the lysis buffer contained T-PER Tissue Protein Extraction Reagent (#PI-78510; Thermo Scientific, Rockford, IL) supplemented with 1 mM EDTA, 1 mM EGTA, 2.5 mM sodium pyrophosphate (NaPPi), 1 mM sodium orthovanadate (Na3VO4), 1 mM ß-glycerophosphate, 1 μg/mL leupeptin, and 1 mM phenylmethylsulfonyl fluoride (PMSF). For muscle lysates analyzed to determine γ3-AMPK activity, the lysis buffer contained 10% glycerol, 20 mM NaPPi, 1% NP-40, 2 mM PMSF, 150 mM sodium chloride (NaCl), 50 mM HEPES (pH 7.5), 20 mM β-glycerophosphate, 10 mM sodium fluoride (NaF), 1 mM EDTA, 1 mM EGTA, 10 μg/mL aprotinin, 10 μg/mL leupeptin, and 2 m MNa3VO4. Homogenates were rotated for 1 h at 4 °C prior to centrifugation (15,000 g for 15 min at 4 °C). The supernatants were transferred to microfuge tubes and stored at −80 °C until subsequent analyses. Protein concentration was measured using the bicinchoninic acid procedure.
Aliquots of the supernatants (200 μL) from muscle lysates were pipetted into a vial together with scintillation cocktail. A scintillation counter (PerkinElmer) was used to determine the 3H and 14C disintegrations per minute. 3-MG uptake was calculated as described by Cartee and Bohn (1995).
The specificity of the γ3-AMPK antibody used for immunoprecipitation (IP) was previously confirmed in Wang et al. (2018). AMPK activity was determined as described by Kjobsted et al. (2015). Briefly, muscle lysates (300 μg protein) were rotated with antibody γ3-AMPK (1:500) and IP buffer [50 mM NaCl, 1% Triton X-100, 50 mM NaF, 5 mM NaPPi, 20 mM Tris-base (pH 7.5), 500 μM PMSF, 2 mM dithiothreitol (DTT), 5 μg/mL leupeptin, 50 μg/mL soybean trypsin inhibitor, 6 mM benzamidine, and 250 mM sucrose] at 4 °C overnight. 50 μL of protein G-magnetic beads were added to each sample, then the samples were rotated for 2 h at 4 °C. DynaMag™-2 magnet was used to pellet the protein G-immunocomplex. Each immunopellet was washed once in IP buffer, once in 6 × assay buffer (240 mM HEPES, 480 mM NaCl, pH 7.0), twice in 3 × assay buffer (1:1). The reaction was initiated at 30 °C by the addition of 30 μL of kinase mix buffer (40 mM HEPES, pH 7.5, 80 mM NaCl, 800 μM DTT, 200 μM AMP, 100 μM AMARA peptide, 5 mM magnesium chloride, 200 μM ATP, and 2 μCi of [γ-33 P]-ATP). After 30 min, the reaction was stopped by the addition of 10 μL of 1% phosphoric acid. Next, 30 μL of supernatant was spotted on P81 phosphocellulose paper. After 3 × 15 min washing with 1% phosphoric acid, followed by 1 × 5 min washing with acetone, the phosphocellulose paper was dried at room temperature and placed in the vials containing 8 mL scintillation cocktail for scintillation counting. Results were expressed relative to the normalized mean of all the samples from each experiment.
Membrane-enriched and cytosol-depleted fractions were obtained by differential centrifugation as described by Grice et al. (2019). Briefly, epitrochlearis muscles were homogenized in ice-cold HES buffer (20 mM HEPES, pH 7.4, 2 mM EGTA, and 250 mM sucrose, 200 μM PMSF, 10 μg/mL pepstatin, and 1 μg/mL leupeptin) with a Polytron PT-3100 homogenizer. The homogenates were centrifuged at 1,380 g for 30 min at 4 °C, and the resulting supernatant was saved. The pellet was resuspended with HES buffer and centrifuged at 1,380 g for 30 min at 4 °C, the resulting supernatant was saved. Then the two supernatants were combined and centrifuged at 17,000 g for 30 min at 4 °C. This combined supernatant (cytosol fraction) was saved. The pellet was washed with HES buffer and centrifuged at 1,380 g for 30 min at 4 °C, the resulting supernatant was saved. Then this supernatant was centrifuged at 17,000 g for 30 min at 4 °C. The resulting pellet (membrane-enriched fraction) was resuspended in HES buffer and stored at −80 °C until analysis. Membrane enrichment was assessed by immunoblotting with antibodies against membrane marker proteins (insulin receptor, IR; Na+/K+-ATPase) and a cytosolic marker protein (α-tubulin). Cholesterol content in the membrane-enriched fraction was determined using the Amplex Red Cholesterol Assay Kit (Thermo Scientific; #A12216) as described by Grice et al. (2019).
Immunoblotting procedures were described by Wang et al. (2018). An equal amount of protein from each muscle lysate was mixed with 6 × Laemmli buffer, boiled for 5 min, separated using SDS-PAGE, and then transferred to polyvinylidene difluoride membranes. Equal loading was confirmed using the MemCode protein stain (Antharavally et al. 2004). Membranes were blocked with TBST (Tris-buffered saline, pH 7.5 plus 0.1% Tween-20) that was supplemented with either with 5% BSA or 5% nonfat milk for 1 h at room temperature, incubated with appropriate concentrations of primary and secondary antibodies. Then membranes were subjected to enhanced chemiluminescence and quantified by densitometry (AlphaView; ProteinSimple, San Jose, CA). Results for each sample (densitometric units) were expressed relative to the normalized average of all the samples on the blot. These normalized values were divided by the corresponding MemCode loading control value for each sample (using individual sample MemCode values that were normalized by dividing the mean MemCode values for all samples on each blot). Values for phosphorylated proteins were expressed as the ratio of phosphorylated signaling protein or enzyme to total signaling protein or enzyme (determined for each sample using a separate immunoblot with a primary antibody against the appropriate total signaling protein or enzyme).
Student’s t-test was used for comparisons between two groups. Two-way analysis of variance (ANOVA) was used to identify main effects of insulin (0 or 1.2 nM insulin) and AICAR (0 or 2 mM AICAR). Post-hoc analysis was performed using the Tukey test (SigmaPlot version 14.5; Systat Software, San Jose, CA). Data lacking normal distribution and (or) equal variance were mathematically transformed to achieve normality and equal variance prior to statistical analysis.
Glucose uptake with insulin alone exceeded values without either AICAR or insulin (P < 0.05; Fig. 1), and glucose uptake in the AICAR + insulin group exceeded both the insulin alone group (P < 0.01) and the AICAR alone group (P < 0.05).
To gain insight into the time-course for AICAR effects on γ3-AMPK activity, we determined γ3-AMPK activity immediately post-AICAR treatment (immediate post-AICAR) and 3 h post-AICAR treatment (3 h post-AICAR). γ3-AMPK activity was increased immediate post-AICAR (P < 0.001, Fig. 2A). After 3 h of recovery from AICAR stimulation, muscles with AICAR treatment had greater γ3-AMPK activity compared with unstimulated control muscles (P < 0.01, Fig. 2B). This result indicates a persistent effect of prior AICAR stimulation on γ3-AMPK activity.
For all of the phosphorylated proteins, the data were expressed as a ratio of the phosphorylated to total protein values.
There were no significant effects of insulin or AICAR on total Akt, AMPK, and TBC1D1 abundance. Total AS160 abundance in muscles without insulin was less than in muscles with insulin, either without (~35%, P < 0.05) or with prior AICAR treatment (~29%, P < 0.05). ANOVA revealed a significant Insulin × AICAR interaction for total ACC abundance (P < 0.05). Total ACC abundance in the AICAR alone group was less (~22%, P < 0.01) than in the group without either AICAR or insulin (P < 0.01).
AktThr308 phosphorylation in muscles with insulin was greater than in muscles without either AICAR or insulin (P < 0.001; Fig. 3A), as well as insulin + AICAR exceeded AICAR alone (P < 0.001). AktSer473 phosphorylation was increased by insulin, either in the absence (P < 0.001; Fig. 3B) or presence of AICAR (P < 0.001).
ANOVA revealed a significant Insulin × AICAR interaction for AS160Ser704 phosphorylation (P < 0.05; Fig. 4A) and AS160Thr642 phosphorylation (P < 0.05; Fig. 4C). AS160Ser704 phosphorylation with AICAR alone exceeded without AICAR or insulin (P < 0.05; Fig. 4A), and the AICAR + insulin group was greater than both the insulin alone group (P < 0.001) and the AICAR alone group (P < 0.001). AS160Ser588 phosphorylation was increased by insulin, either in the absence (P < 0.01; Fig. 4B) or presence of AICAR (P < 0.01). AS160Thr642 phosphorylation with insulin alone was greater than without either AICAR or insulin, and the AICAR + insulin group exceeded both the insulin alone group (P < 0.01; Fig. 4C) and the AICAR alone group (P < 0.001).
AMPKαThr172 (Fig. 5A) or ACCSer79 phosphorylation (Fig. 5B) was increased by prior AICAR treatment, regardless of insulin concentration (P < 0.05 without insulin; P < 0.01 with insulin for pAMPKαThr172; P < 0.001 without or with insulin for pACCSer79).
TBC1D1Ser237 phosphorylation was increased with prior AICAR treatment, either in the absence (P < 0.05; Fig. 5C) or presence of insulin (P < 0.05). For TBC1D1Thr590 phosphorylation, there were no significant effects of insulin or AICAR (Fig. 5D).
There were no significant effects of insulin or AICAR on either GLUT4 or HKII abundance. (Fig. 6A and 6B).
The membrane fraction was enriched with membrane marker proteins (IR, Na+/K+-ATPase) and depleted of the cytosolic marker protein (α-tubulin; Fig. 7A). Skeletal muscle membrane cholesterol content was decreased by prior AICAR treatment (P < 0.05; Fig. 7B).
Phosphorylated HMGCRSer872 and phosphorylated ABCA1Ser2054 were unaltered by prior AICAR treatment (Fig. 7C and 7D).
The purpose of the current study was to evaluate several potential mechanisms that might contribute to the AICAR-induced enhancement of glucose uptake by skeletal muscle from 26–27-month-old rats. The results revealed that prior AICAR treatment of skeletal muscle resulted in a robust and sustained increase in γ3-AMPK activity. Prior AICAR treatment also resulted in a significant decline in membrane cholesterol content. In addition, AICAR-treated muscle had greater TBC1D1Ser237 phosphorylation. Based on these results, it is possible that one or more of these outcomes contributed to the AICAR-related increase in glucose uptake. Muscles incubated with AICAR had greater glucose uptake determined both in the presence and absence of insulin. We previously used the same incubation protocol with muscles from 24-month-old, male Fischer-344 X Brown Norway (FBN) rats, and found that prior AICAR treatment led to increased glucose uptake in muscles incubated with insulin, but not in muscles incubated without insulin (Oki et al. 2018). Both the current study and Oki et al. (2018) evaluated isolated epitrochlearis muscles from male FBN rats. It is uncertain if the modest difference in age (26–27-month-old in current study versus 24-month-old in Oki et al. (2018)) was a factor in the differing effects of AICAR on insulin-independent glucose uptake. However, the relative AICAR-induced increase in glucose in insulin-stimulated muscles was roughly similar in the current study (49% increase) compared to the earlier study (57% increase). Thus, prior AICAR treatment has consistently resulted in a substantial increase in glucose uptake by insulin-stimulated muscles from old rats. Consistent with our previous study, AICAR caused a substantial increase AMPKThr172 phosphorylation in the skeletal muscle from old rats. AICAR also resulted in elevated phosphorylation of ACCSer79, an ACC phosphosite that is frequently used as a surrogate marker for elevated AMPK activity (Hardie 1992; Scott et al. 2002). AICAR also increased the phosphorylation of other AMPK substrates, AS160Ser704 and TBC1D1Ser237. Muscle γ3-AMPK activity was substantially elevated both immediately after AICAR treatment and more than 3 h after AICAR incubation. However, AICAR did not result in elevated phosphorylation of HMGCRSer872 (an AMPK substrate). The absence of a uniform AICAR effect on the phosphorylation of all AMPK substrates may be related to the fact that, in addition to kinase activation, the phosphorylation status of a specific protein is subject to other regulatory factors, e.g., co-localization of the kinase and substrate, and substrate dephosphorylation by protein phosphatases. Prior AICAR treatment did not amplify proximal insulin signaling at the level of Akt phosphorylation. This observation is consistent with earlier studies in muscles from young mice (Kjobsted et al. 2015; Jørgensen et al. 2018), young rats (Fisher et al. 2002), and old rats (Oki et al. 2018). AICAR treatment also led to enhanced phosphorylation of AS160 on Thr642 (an Akt phosphosite) and Ser704 (an AMPK phosphosite) in insulin-stimulated muscles. Previous research has reported greater phosphorylation on one or both of these sites in muscles from young mice (Kjobsted et al. 2015; Jørgensen et al. 2018). There is evidence that AMPK-mediated phosphorylation of Ser704 may favor greater Thr642 phosphorylation (Kjobsted et al. 2015). Elimination of AS160 expression in young AS160-KO mice prevented the enhanced insulin-stimulated glucose uptake observed after incubation with AICAR (Kjobsted et al. 2019). In this context, it is reasonable to suspect that AS160 contributed to the effect of AICAR on glucose uptake observed in muscles from old rats in the current study. AICAR led to greater TBC1D1 phosphorylation on Ser237, which is an AMPK phosphosite (Espelage et al. 2020). Similar results have been reported for muscles from young mice (Vichaiwong et al. 2010; Kjobsted et al. 2015; Jørgensen et al. 2018), but muscles from old animals had not been previously evaluated. Preventing phosphorylation in young mice with a skeletal muscle-specific knock-in mutation of TBC1D1Ser231Ala (Ser231 in mice corresponds to Ser237) resulted in a partial reduction in AICAR-stimulated glucose uptake at a low AICAR concentration (0.15 mM) (Chen et al. 2017). However, the knock-in mutation did not attenuate glucose uptake in response to 2 mM AICAR (Chen et al. 2017), which is the AICAR concentration used in the current study. As expected, phosphorylation of this AMPK phosphosite was unaffected by insulin in muscles of old rats. Neither AICAR nor insulin resulted in greater TBC1D1 phosphorylation on Thr590, which is an Akt phosphosite (Espelage et al. 2020). Some studies have reported that a pharmacologic insulin dose can enhance Thr590 phosphorylation in muscles from young mice (Pehmoller et al. 2009; Vichaiwong et al. 2010). Conversely, an insulin concentration similar to the dose used in the current study failed to increase Thr590 phosphorylation in muscles from either lean or obese middle-aged humans (Middelbeek et al. 2013). Furthermore, insulin-stimulated GLUT4 translocation is not attenuated in skeletal muscle from TBC1D1-KO rats (Whitfield et al. 2017). Although a role for TBC1D1 remains possible, it seems unlikely that the AICAR effect on glucose uptake by the muscles from old rats was entirely attributable to TBC1D1. Membrane cholesterol was 15% lower for the muscles incubated with AICAR, and this outcome was accompanied by a 49% AICAR-related increase in glucose uptake by insulin-stimulated muscles. These results are similar to the AICAR-related decrease (~10%) in membrane cholesterol and AICAR-related increase (~50%) in plasma membrane GLUT4 content in insulin-stimulated L6 cells (Habegger et al. 2012a). These findings suggest that the magnitude of AICAR-related change in membrane cholesterol in muscles from old rats might have been sufficient to play a role in the increased glucose uptake. The lack of an AICAR-effect on the phosphorylation of either of the enzymes that can modulate membrane cholesterol content (HMGCRSer872, an AMPK substrate, and ABCA1Ser2054, a protein kinase A substrate) indicates that another mechanism was likely responsible for the reduction in membrane cholesterol. However, it is possible that AICAR caused a transient increase phosphorylation of HMGCRSer872 and (or) ABCA1Ser2054 that had reversed when the muscles were sampled, more than 3 h after being incubated with AICAR. Most mammalian cells, including skeletal muscle cells, cannot catabolize cholesterol, and excess cholesterol is either expelled from the cell via transporter proteins or stored as cholesteryl esters in lipid droplets (Luo et al. 2020). While ABCA1 is the cholesterol transporter protein that has been most widely studied in skeletal muscle, mRNA expression of another cholesterol exporter, ABCG1 (ATP binding cassette transporter G1), has also been detected in skeletal muscle (Myers et al. 2006; Cheung et al. 2017; Morgan et al. 2020). AMPK can enhance ABCG1 mRNA and protein expression and cholesterol efflux from macrophages (Li et al. 2010), but evidence is lacking about ABCG1’s protein abundance, regulation, and relative contribution to cholesterol efflux in skeletal muscle. In addition, the possibility that the AICAR-induced decline in membrane cholesterol might be secondary to greater cholesterol esterification and storage in skeletal muscle lipid droplets remains to be evaluated. Several novel results in the current study advanced knowledge related to the mechanisms underlying improved glucose uptake in skeletal muscle of old rats after brief treatment with AICAR. The most striking new observation was that AICAR treatment led to a marked increase in γ3-AMPK activity, and this increase was sustained more than 3 h after the exposure to AICAR. The magnitude and duration of this effect in the muscle of old rats, taken together with the strong evidence from young mice that γ3-AMPK is crucial for long-lasting effects on glucose uptake after transient AICAR incubation (Kjobsted et al. 2015), supports the idea that γ3-AMPK plays a role in the elevated glucose uptake in AICAR-treated muscles from old rats. It has been widely recognized that the highly selective expression of γ3-AMPK in skeletal muscle offers an opportunity to create a compound with selective action on skeletal muscle with reduced chance of unwanted side-effects in other tissues (Mahlapuu et al. 2004; Kjobsted et al. 2018; Steinberg and Carling 2019; Rhein et al. 2021). Another intriguing and novel result of the current study was the AICAR-induced decrement in membrane cholesterol of skeletal muscle from old rats. It will be important for future research to determine if γ3-AMPK has a role in the regulation of membrane cholesterol content and to directly test the extent to which the AICAR-induced decrement in membrane cholesterol contributed to elevated glucose uptake. The role of γ3-AMPK could be evaluated using genetically modified mouse models that lack γ3-AMPK expression (Barnes et al. 2004; Rhein et al. 2021). The extent to which a decline in membrane cholesterol plays a role in enhanced glucose uptake could be assessed using the cholesterol-depleting chemical methyl-β-cyclodextrin (MβCD). Previous research demonstrated that MβCD can lower membrane cholesterol in cultured mouse myofibers and palmitate-treated L6 muscle cells (Habegger et al. 2012b; Llanos et al. 2015). Incubating AICAR-treated L6 muscle cells with MβCD that was complexed with cholesterol (MβCD-cholesterol), which can be used to replenish membrane cholesterol, eliminated the AICAR-mediated decrease in membrane cholesterol (Habegger et al. 2012a). MβCD-cholesterol could be used to eliminate the AICAR-induced decrement in skeletal muscle membrane cholesterol content to test if reduced membrane cholesterol is necessary for AICAR’s effect on glucose uptake. |
PMC9648401 | 36350790 | Yongjian Guo,Liang Wang,Zhen Qi,Yu Liu,Kun Tian,Huanran Qiang,Pei Wang,Guohua Zhou,Xiaobo Zhang,Shu Xu | A novel strategy for orthogonal genetic regulation on different RNA targeted loci simultaneously RNA BIOLOGY | 09-11-2022 | Orthogonal genetic regulating,mRNA,miRNA,FEN1,mis-hpDNA | ABSTRACT No current RNA-targeted interference tools have been reported to simultaneously up and down-regulate different gene expressions. Here we characterized an RNA-targeted genetic regulatory strategy composed of a flap endonuclease 1 (FEN1) and specific mis-hairpin DNA probes (mis-hpDNA), to realize the orthogonal genetic regulation. By targeting mRNA, the strategy hindered the translation and silenced genes in human cells with efficiencies of ~60%. By targeting miRNA, the strategy prevented the combination of miRNA to its specific mRNA and increased this mRNA expression by about 3-folds. In combination, we simultaneously performed CXCR4 gene knock-down (~50%) and EGFR gene activation (1.5-folds) in human cells. Although the functional property can be further improved, this RNA-targeted orthogonal genetic regulating strategy is complementary to classical tools. | A novel strategy for orthogonal genetic regulation on different RNA targeted loci simultaneously RNA BIOLOGY
No current RNA-targeted interference tools have been reported to simultaneously up and down-regulate different gene expressions. Here we characterized an RNA-targeted genetic regulatory strategy composed of a flap endonuclease 1 (FEN1) and specific mis-hairpin DNA probes (mis-hpDNA), to realize the orthogonal genetic regulation. By targeting mRNA, the strategy hindered the translation and silenced genes in human cells with efficiencies of ~60%. By targeting miRNA, the strategy prevented the combination of miRNA to its specific mRNA and increased this mRNA expression by about 3-folds. In combination, we simultaneously performed CXCR4 gene knock-down (~50%) and EGFR gene activation (1.5-folds) in human cells. Although the functional property can be further improved, this RNA-targeted orthogonal genetic regulating strategy is complementary to classical tools.
Present RNA-level interference tools are siRNAs [1], antisense oligonucleotides (ASOs) [2], RNA-targeted CRISPR/Cas systems [3–6], miRNA mimetics [7], antimiRs [8] and so on. Up to now, no individual tool has been reported to be capable of simultaneously up and down-regulating gene expression when targeting RNA loci. In fact, in some cases, researchers need to simultaneously work on different splice isoforms and regulatory RNA elements, or even up-regulate and down-regulate the expression of different genes simultaneously to characterize gene functions comprehensively. Traditionally, it requires various tools working in the same cell, leading to inconvenience in operation and maybe mutual interference in effect. In this study, we designed an RNA-targeted strategy composed of a Flap Endonuclease-1 (FEN1, ~35 kDa) [9] and specific mis-hpDNA probes (including stem-loop and guide sequence), to orthogonal regulate gene expression (Figure 1A). FEN1 captures the mis-hpDNA probe due to its stem-loop structure [10,11]. Then this FEN1-probe complex locates on the target loci, which is guided by the guide sequence in the mis-hpDNA probe, and makes a steric hindrance effect on the targeted mRNA or miRNA. The FEN1-probe complex further blocks the moving forwards of the ribosome to inhibit the mRNA translation to down-regulate gene expression (Figure 1B), or blocks miRNAs from forming miRNA:mRNA [12,13] complexes and makes the free mRNA more to up-regulate gene expression (Figure 1C). We named this strategy as hpDNA-assistedstructure-guidednuclease mediating interference (designated the HpSGNi system hereafter) and tested the loss/gain function separately and together here.
Firstly, we tested whether we could target mRNA and mediate gene silencing in human cells. HEK293A cells were co-transfected with plasmids encoding FEN1-NES (an NES was into the C terminal of FEN1, which made the protein expressed only in the cytoplasm, not in the nucleus, Figure S1), plasmids encoding EGFP, and mis-hpDNAs targeting the EGFP mRNA gene (Figure 2A). Then the fluorescence in cells was detected by flow cytometric analysis, and the efficiency of down-regulation of the EGFP gene was calculated by comparing with control groups. As shown in Figure 2A and Figure S2, the normalized fluorescence of the group transfected with mis-hpDNA plus FEN1-NES was 62% lower on average than that in the group transfected with unrelated probes, with a significant difference (p-value was 0.0002). To know the effect of the guide sequence length on down-regulation efficiency, we shorten the length from 26-nt to 14-nt with a 4-nt interval (Figure 2B). For mis-hpDNA with 26- and 22-nt guide sequences, the EGFP expression was reduced obviously (p-value were 0.02 and 0.03, respectively). But for 18- and 14-nt guide sequence, the reduction was no longer obvious (p-value were 0.08 and 0.20, respectively). It indicated that too short guide sequences with relatively low Tm values are not beneficial for binding with targeted RNA loci. So, we suggested the guide sequence should not be too short, more than 22-nt. Secondly, we co-transfected A549 cells with plasmids encoding FEN1-NES and mis-hpDNAs targeting the CXCR4 mRNA. Compared with the control groups that were transfected with FEN1-NES alone, certain (~45%) down-regulation of CXCR4 expression in the group transfected with both FEN1-NES and specific mis-hpDNA was observed (Figure 2C). The location of the mis-hpDNA for CXCR4 mRNA crossed two exons in genomic DNA, and the knockdown cannot be attributed to DNA-level transcription blocking. Then, we quantified the CXCR4 mRNA to see whether it was cleaved or not. The CXCR4 mRNA level in the group transfected with FEN1-NES plus specific mis-hpDNA shows an indistinctive difference from that in the group transfected with FEN1-NES alone (Figure 2D). Therefore, the knockdown of CXCR4 expression cannot be attributed to mRNA cleavage. Thus, it was concluded that the FEN1-probe complex can be reprogrammed to preferably mediate gene silencing by blocking but not cleaving mRNA in human cells.
After verifying the ability of down-regulating gene expression, we then test whether it can also up-regulate in human cells. The miRNA [12] was reported with the ability to capture the specific mRNA and guide degradation/blocking. In this study, as shown in Figure 3A, we tried to use FEN1 plus mis-hpDNA to capture the miRNA, then inhibit the combination between miRNA and its targeted mRNA, followed by holding back the mRNA degradation/blocking. In studies reporting the binding ability of miRNA and targeted RNA [14], miRNA-21-5p has been reported in relationship with regulating EGFR [15]. FEN1-NES plasmid was co-transfected with mis-hpDNA targeting miRNA-21-5p. As Figure 3B showed, the group transfected with the FEN1-NES plus mis-hpDNA increased EGFR protein expression (best ~ 3-folds) than the group transfected with FEN1-NES alone. To confirm that HpSGNi did not cleave targeted miRNA, we used quantitative PCR to quantify the amount of miRNA-21-5p. Compared with control groups, no depletion in experimental groups was observed (Figure 3C). Then we further test whether multiple miRNAs could be blocked. The miRNA-145-5p [16] was reported to be capable of targeting OCT4 (also named POU5F1P1) mRNA. We transfected mis-hpDNAs targeting miRNA-145-5p and miRNA-21-5p into A549 cells expressing FEN1-NES. Results in Figure 3D showed that both EGFR and OCT4 expression were up-regulated (~1.5 and ~2.5-folds on average, respectively). Thus, it was concluded that the FEN1-probe complex could be reprogrammed to preferably mediate gene up-regulation by targeting miRNA in human cells. However, we worry that the efficiency will decrease when the number of targeted miRNA increases. It was confirmed when we transfected three kinds of mis-hpDNAs to up-regulate both OCT4, EGFR and YWHAZ expression. One of the reasons is that the FEN1-NES will be shared with different targeted loci. Another reason is the amount of each mis-hpDNA will be decreased when the number of kinds of mis-hpDNA increases. The ability to up-regulate many genes simultaneously still needs to be improved in future studies.
To explore whether one gene expression could be suppressed and another gene expression could be active simultaneously, we mediate orthogonal gene control (down- and up-regulation) using the FEN1-NES, a mis-hpDNA targeting CXCR4 mRNA and a mis-hpDNA targeting miRNA-21-5p (Figure 4A). A549 cells were transfected with plasmids encoding FEN1-NES, then were transfected with mis-hpDNA targeting miRNA-21-5p for EGFR gene activation and mis-hpDNA targeting CXCR4 mRNA for CXCR4 gene silencing. As a result, CXCR4 protein expression decreased by~50% (Figure 4B, and C), and EGFR protein expression increased by about 2-folds (Figure 4B, and D) in this orthogonal condition.
FEN1 is a nuclease being able to recognize and cleave nucleic acid. During DNA replication and repair processes, the newly synthesized DNA and the displaced region compete for base pairing with the template strand, forming a double-flap structure [17]. The FEN1 recognizes the double-flap structure by the 3’ flap and catalyzes phosphodiester cleavage of the 5’ flap [11]. Then FEN1 participates in removing RNA primers. Then in 2020, we used the characteristic of FEN1 to build a hairpin DNA probe structure-guidednuclease system (designated the HpSGN system hereafter) [18]. The HpSGN system comprises FEN1 nuclease and hairpin DNA probe (designated hpDNA below). Significantly, a notable feature is that the cleavage is invalid when the base at the position of 1 on the target mismatches the hpDNA [18]. We inferred that this feature could be taken to build an interference platform. And to confirm this point, we mutated the bases in the hpDNA at positions of 1–3, followed by reacting with single-strand RNA targets (ssRNA) and FEN1. As the electrophoretic mobility shift assay (EMSA) in Figure S3 shows, the mis-hpDNA frustrated FEN1 cleavage but formed a FEN1-hpDNA-target ternary complex. These observations suggested that we can build the interference platform HpSGNi, which can capture ssRNA targets without cleavage. To our knowledge, HpSGNi was the first reported RNA-targeted loss and gain-function methodology. But at present, the number of genes that can be regulated simultaneously is limited by transfection and cell toxicity (Figure S4). And when compared separately, this orthogonal genetic regulation system has lower efficiency of up-regulation than ORFs and does not perform better than RNAi and CRISPRi to down-regulate. However, the HpSGNi system still has its own advantages. It has no limitation on targets’ sequences (like PFS for RNA-targeted CRISPR/Cas system) and is theoretically suitable for RNA targets with any sequence. In contrast with the Cas proteins, which were 1367 to 422 amino acids, the FEN1 protein is 337 amino acids. The smaller size of the FEN1 is of great benefit to delivery. It has no need to be chemically modified, which makes the strategy cheaper than synthetic oligonucleotides. Theoretically, the stability of the DNA probe is better than the RNA probe, which may be one of the advantages of HpSGNi system. So, we modified a Cy5 label on the mis-hpDNA and transfected it into cells to observe after 12, 24 and 48 hours. As shown in Figure S5, the mis-hpDNA remained stable after being transfected into cells 48 hours later. Compared with other endogenous miRNAa strategies [19], the mechanism of HpSGNi is relatively simple and relies on a fixed FEN1:mis-hpDNA:miRNA complex. But this kind of miRNAa methods may be partly affected by the cell type and targeted miRNA type, for example, with no strong miRNA /mRNA interaction, there will be no mRNA up-regulation In summary, we developed a simple and modular RNA-targeted platform HpSGNi system for targeted gene regulation. Although the efficiency should be improved in further study and the compatibility of application on the miRNA was limited, the strategy reported here is still complementary to classical methods.
For RNA loci regulating in cells, NES was added to the C-terminus of synthesized A. fulgidus FEN1 ORF and subcloned to pcDNA3.1(+) to generate pcDNA3.1(+)-HA-A. fulgidus FEN1-NES (Figure S1).
The mis-hpDNA consisted of a stem-loop and a guide sequence. The stem-loop was fixed as 5’-aga gtc ggc ctt ttg gcc gac tct ctt atc aac ttg aaa aag ttg gca ccg agt cgg tgt-3’. The guide sequence is shown in Table 1. The unpaired bases are marked in red.
HEK293A and A549 cells were maintained in Dulbecco’s modified Eagle’s Medium (DMEM) or RPMI-1640 supplemented with 10% foetal bovine serum (HyClone), 100 U/mL penicillin, and 100 µg/mL streptomycin at 37°C with 5% CO2 incubation. Cells were seeded into 6-well plates/24-well plates (Corning) 24 h before transfection at a density of 300,000 /70,000 cells per well. Cells were transfected using GenJet™ Reagent (SignaGen) following the manufacturer’s recommended protocol. For each well of a 6-well/24-well plate, 1 μg/0.5 μg plasmids expressing FEN1 were transfected and 500 pmoL/100 pmoL of DNA probes were transfected 24 h later.
The targets of miR-145 and miR-21 were predicted by Miranda (http://www.miRBase.org).
Protein samples were isolated with lysis buffer, eluted with SDS buffer, separated by SDS-polyacrylamide gels, and electroblotted onto NC membranes. The specific protein bands were stained with High-sig ECL Western Blotting Substrate (Tanon) and imaged using the Amersham Imager 600 (GE Healthcare). Primary antibodies included antibodies against OCT4 (Abcam, ab200834, 1:10,000), EGFR (Abclonal Technology, A11577, 1:1000) and β-actin (Abclonal Technology, AC026, 1:50,000).
The cells were resuspended in PBS, and the fluorescence intensity (EGFP 488 nm excitation and 525 nm emission) was measured immediately using FACSCalibur (Becton Dickinson). For CXCR4, cells were dissociated and then stained in PBS for 1 h at room temperature, followed by incubating with antibody (Miltenyi Biotec, 130–120-778, 1:50) and measured using FACSCalibur.
Eukaryotic cells were trypsinized and washed once with PBS, and total RNA was isolated with RNA-easy Isolation Reagent (Vazyme, R701-01/02) following the manufacturer’s instructions. For mRNA and miRNA, cDNA synthesis was performed using the HiScript III 1st Strand cDNA Synthesis Kit (+gDNA wiper) (Vazyme, R312-01) and 1st Strand cDNA Synthesis Kit (by stem-loop) (vazyme-MR101-01), respectively. The cDNA was used in quantitative PCR analyses with AceQ qPCR SYBR Green Master Mix (Low ROX Premixed) (Vazyme, Q131-02) and Universal SYBR qPCR Master Mix (Vazyme-MR101-01). Relative gene expression was calculated using the ΔΔCt method. Results were normalized to Gapdh or U6 for mRNA and miRNA experiments. The primers used in this study are shown in Table 2.
All data were expressed as mean ± SD or mean ± SEM from two to eight independent experiments performed in a parallel manner. Comparisons between two groups were analysed using two-tailed Student’s t-tests. P values < 0.05 were considered statistically significant.
Click here for additional data file. |
PMC9648404 | Andrei V. Chernov,Veronica I. Shubayev | Sexual dimorphism of early transcriptional reprogramming in degenerating peripheral nerves | 27-10-2022 | Mus musculus,sexual dimorphism,peripheral nerve injury,axotomy,RNA-seq,Wallerian degeneration | Sexual dimorphism is a powerful yet understudied factor that influences the timing and efficiency of gene regulation in axonal injury and repair processes in the peripheral nervous system. Here, we identified common and distinct biological processes in female and male degenerating (distal) nerve stumps based on a snapshot of transcriptional reprogramming 24 h after axotomy reflecting the onset of early phase Wallerian degeneration (WD). Females exhibited transcriptional downregulation of a larger number of genes than males. RhoGDI, ERBB, and ERK5 signaling pathways increased activity in both sexes. Males upregulated genes and canonical pathways that exhibited robust baseline expression in females in both axotomized and sham nerves, including signaling pathways controlled by neuregulin and nerve growth factors. Cholesterol biosynthesis, reelin signaling, and synaptogenesis signaling pathways were downregulated in females. Signaling by Rho Family GTPases, cAMP-mediated signaling, and sulfated glycosaminoglycan biosynthesis were downregulated in both sexes. Estrogens potentially influenced sex-dependent injury response due to distinct regulation of estrogen receptor expression. A crosstalk of cytokines and growth hormones could promote sexually dimorphic transcriptional responses. We highlighted prospective regulatory activities due to protein phosphorylation, extracellular proteolysis, sex chromosome-specific expression, major urinary proteins (MUPs), and genes involved in thyroid hormone metabolism. Combined with our earlier findings in the corresponding dorsal root ganglia (DRG) and regenerating (proximal) nerve stumps, sex-specific and universal early phase molecular triggers of WD enrich our knowledge of transcriptional regulation in peripheral nerve injury and repair. | Sexual dimorphism of early transcriptional reprogramming in degenerating peripheral nerves
Sexual dimorphism is a powerful yet understudied factor that influences the timing and efficiency of gene regulation in axonal injury and repair processes in the peripheral nervous system. Here, we identified common and distinct biological processes in female and male degenerating (distal) nerve stumps based on a snapshot of transcriptional reprogramming 24 h after axotomy reflecting the onset of early phase Wallerian degeneration (WD). Females exhibited transcriptional downregulation of a larger number of genes than males. RhoGDI, ERBB, and ERK5 signaling pathways increased activity in both sexes. Males upregulated genes and canonical pathways that exhibited robust baseline expression in females in both axotomized and sham nerves, including signaling pathways controlled by neuregulin and nerve growth factors. Cholesterol biosynthesis, reelin signaling, and synaptogenesis signaling pathways were downregulated in females. Signaling by Rho Family GTPases, cAMP-mediated signaling, and sulfated glycosaminoglycan biosynthesis were downregulated in both sexes. Estrogens potentially influenced sex-dependent injury response due to distinct regulation of estrogen receptor expression. A crosstalk of cytokines and growth hormones could promote sexually dimorphic transcriptional responses. We highlighted prospective regulatory activities due to protein phosphorylation, extracellular proteolysis, sex chromosome-specific expression, major urinary proteins (MUPs), and genes involved in thyroid hormone metabolism. Combined with our earlier findings in the corresponding dorsal root ganglia (DRG) and regenerating (proximal) nerve stumps, sex-specific and universal early phase molecular triggers of WD enrich our knowledge of transcriptional regulation in peripheral nerve injury and repair.
Successful regeneration of a severed peripheral nerve relies on a well-coordinated multicellular distal degeneration process first described by Waller (1850). Significant insight into the cellular and molecular processes of Wallerian degeneration (WD) was obtained using rodent models of the peripheral nerve transection (Fawcett and Keynes, 1990; Stoll et al., 2002; Vargas and Barres, 2007). A high-efficiency acute-phase (within 24 h) peripheral WD process in the distal to transection segment, disconnected from neuronal soma, determines functional repair of the nerve (Gordon et al., 2003; McDonald et al., 2006; Wood et al., 2011; Zochodne, 2012). These prior acute-phase peripheral WD studies were done using rodents of one sex. Sex-dependent response to peripheral nerve injury in rodents (Sorge et al., 2015; Chernov et al., 2020; Yu et al., 2020; Chernov and Shubayev, 2021, 2022) is thought to relate to sex differences in prevalence, incidence, mechanisms, or clinical presentation of peripheral neuropathies (Unruh, 1996; Greenspan et al., 2007; Fillingim et al., 2009; Mogil, 2012; Sorge and Totsch, 2017; Boerner et al., 2018). The transcriptional landscape of peripheral nerve in control human subjects and patients with radiculopathy assessed by RNA-sequencing (RNA-seq) analyses reveal sexual dimorphism in metabolic, neuroendocrine, immune regulatory, and sex chromosome-related programs both at baseline and in response to peripheral nerve damage (North et al., 2019; Ray et al., 2019; Stephens et al., 2019; Chernov et al., 2020; Mecklenburg et al., 2020; Paige et al., 2020; Tavares-Ferreira et al., 2020; Ahlström et al., 2021). Whether sexual dimorphism in acute-phase WD exists remains not understood. During an acute-phase (within 24 h) peripheral WD, an immediate influx of extracellular calcium activates calpain- and ubiquitin-proteasome-dependent disintegration of the axonal cytoskeleton. A dominant cell type in peripheral nerve, denervated Schwann cells initiate axonal and myelin breakdown and phagocytosis. In concert with resident macrophages, endothelial cells, and fibroblasts, Schwann cells control a time-dependent and continuous stream of hematogenous immune cells, starting with neutrophils, mast cells, monocytes, and lymphocytes (Hirata and Kawabuchi, 2002; Bauer et al., 2007; Scholz and Woolf, 2007; Barrette et al., 2008; Jessen and Arthur-Farraj, 2019). In addition, Schwann cells engage vast molecular machinery required to support their robust phenotypic changes induced by injury, including de-differentiation, accompanied by myelin protein gene silencing and glial fibrillary acidic protein (GFAP) activation, followed by neuregulin-epidermal growth factor receptor (EGFR, also named ERBB) controlled mitosis, and c-Jun-regulated redifferentiation, axonal partnership and remyelination (Clemence et al., 1989; Triolo et al., 2006). Extensive molecular signatures reveal a dynamic interplay between immune, metabolic, kinase, neurotrophic, and other gene families jointly contributing to the functional activity of many cell types in a time-dependent manner. Thus, within 24 h post-axotomy, calcium- and zinc-dependent activation of extracellular matrix (ECM) remodeling by matrix metalloproteinase (MMP) family of 24 members, which control permeability of blood-nerve and perineurial barriers, Schwann cell signaling, as well as myelin and ion channel proteolysis (Shubayev and Myers, 2002; Myers et al., 2006; Shubayev et al., 2006; Chattopadhyay et al., 2007; Kobayashi et al., 2008; Chattopadhyay and Shubayev, 2009; Liu et al., 2010, 2012; Liu and Shubayev, 2011; Kim et al., 2012; Nishihara et al., 2015; Remacle et al., 2015). Our comparative high-depth (over 5.0E + 07 paired-end reads per sample) RNA-seq and predictive bioinformatics analyses of the early response genome-wide transcriptional changes at 24 h after sciatic nerve axotomy in male and female mice indicated immediate and sex-dependent control of the transcriptional landscape in dorsal root ganglia (DRG) (Chernov and Shubayev, 2021), regenerating (proximal) nerve segment (Chernov and Shubayev, 2022). The present study of degenerating (distal) identified new sexually dimorphic protein-coding and non-coding (nc)RNAs and respective signaling pathways specific to peripheral nerve remodeling.
To assess sex differences in the early phase peripheral WD transcriptional (protein-coding mRNAs and ncRNA) response, at 24 h after complete sciatic nerve axotomy or sham operation, distal nerve stumps (Figure 1A) were collected in female and male mice (n = 6 mice/group). To obtain the optimal amount of total RNAs for whole-genome transcriptomics analysis by high-depth RNA-seq nerve stumps from two mice were pooled in all RNA samples (n = 3/groups, 2 mice/RNA sample). RNA-seq of the respective lumbar (L) 4/5 DRG and proximal sciatic nerve stumps from the same animal cohorts were reported earlier (Chernov and Shubayev, 2021, 2022).
Genes with greater than 10 transcript counts across all samples were used for normalization and identification of differentially expressed genes (DEGs) in axotomy samples relative to respective shams by DESeq2 using Wald’s test (Love et al., 2014). The standard significance criteria Padj < 0.1 and log2FC > 1 were used to identify significant DEGs (Figure 1B) in this report and our prior comparative analyses of regenerating peripheral nerves (Chernov and Shubayev, 2022) and DRG (Chernov and Shubayev, 2021). Baseline gene expression [log2(counts)], log2FC, and significance scores of 1,017 DEGs identified by DESeq2 were used for predictive system analysis (Supplementary Table 1). Comparative analysis of RNA-seq and protein immunoblotting data of select genes was conducted in DRG of the same animal cohort, paired-end library preparation, and RNA-seq. A high level of correlation between mRNA and protein levels was demonstrated in post-axotomy and sham samples (Chernov and Shubayev, 2021). The principal component (PC) analysis (Figure 1C) identified sex-specific variance-driving DEGs related to ECM homeostasis, nerve regrowth, axonal guidance, and cytokine signaling among the most significant contributors to injury-related transcriptional changes in distal nerve stumps. We predicted that 80.6% (PC1) and 8.9% (PC2) of DEGs could contribute to variance based on axotomy and sex differences. Specifically, sonic hedgehog (Shh), cardiotrophin-like cytokine factor 1 (Clcf1), ECM-localized corticotropin-releasing factor urocortin-2 (Ucn2), and glia-derived neurotrophic factor (Gdnf), leucine-rich repeat-containing protein 15 (Lrrc15), and major urinary proteins Mup18/Mup22 genes were the most significant drivers of sexually dimorphic transcriptional variance. Heatmap histograms ranked by absolute expression in three biological replicates (Figure 2A) and volcano plots (Figure 2B) display DEG’s ranked by statistical significance (Padj) and log2FC ranking.
In both sexes, we identified high expression levels of genes related to signaling pathways specific to nerve injury response. Some of these pathways exhibited additional activation or inhibition at 24 h post-axotomy as outlined below. RhoGDI, ERBB, ERK5 Signaling pathways, and Epithelial/Mesenchymal Transition by Growth Factors Signaling pathway showed significant activation in both sexes (Figure 3). Compared to males, females demonstrated distinct, albeit mild, upregulation of the Osteoarthritis Signaling, Protein Kinase A Signaling, and Wnt/β-catenin Signaling pathways. Males activated signaling pathways related to Neuregulin Signaling consistent with upregulation of Ereg, Areg, and Hbegf genes (Figure 4). In addition, the activity of HMGB1 Signaling and IL-17 Signaling pathways additionally increased in males due to stronger upregulation of IL-6, LIF, Fos, Fosl1, and related genes. GDNF Family Ligand Receptor Interactions, CSDE1 Signaling, Mouse Embryonic Stem Cell Pluripotency, and Aryl Hydrocarbon Receptor Signaling demonstrated a male-specific up-regulation. Cardiac Hypertrophy Signaling and Wound Healing pathways were activated in males but inhibited in females. Baseline expression levels of DEGs related to Axonal Guidance Signaling in both sexes due to a robust expression of endothelin-converting enzyme Ecel1, Shh, Ngfr, Glis1, and other related genes (Figure 4). Cholesterol Biosynthesis, Reelin Signaling in Neurons, Synaptogenesis Signaling Pathway, and GP6 Signaling pathways were downregulated in females but were not significantly altered in males. Signaling by Rho Family GTPases, cAMP-mediated signaling, and sulfated glycosaminoglycan biosynthesis were downregulated in both sexes. The iodothyronine deiodinase genes Dio2 and Dio2, which regulate thyroid hormone conversion, increased in both sexes, yet the increase was higher in males. It is important to note that many DEGs related to pathways activated in males exhibited robust expression in females in both axotomized and sham nerves (Figure 4). Thus, male-specific post-injury DEG regulation balanced the activity of respective pathways to levels observed in females. The most significant sex-specific DEGs were further highlighted on heatmap diagrams grouped by gene relevance to peripheral nerve injury processes according to IPA predictions and GO terms.
The cascade of predicted upstream transcriptional regulators (small endogenous molecules, proteins, including transcription factors, and RNAs) could explain the observed gene expression changes. According to IPA predictions, growth factors (including Egf, Fgf2, Vegf, Hgf, and others) could contribute to transcriptional regulation in both sexes (Figure 5A). In addition, IPA predicted that steroid hormone β-estradiol could regulate downstream transcription distinctly in females in males. In females, among 144 DEGs targeted by β-estradiol, 44 DEGs were upregulated, and 68 DEGs were downregulated. In males, among 77 genes 64 were upregulated more than twofold but only 12 DEGs were downregulated (Padj < 0.1) (Figure 5B and Supplementary Table 2). The Esr1 gene, encoding one of two main types of nuclear estrogen receptors ERα that transmit β-estradiol signaling in the nucleus (Rettberg et al., 2014), maintained moderate levels of expression in both sexes. Remarkably, the Esr2 gene encoding ERβ demonstrated a 10-fold decrease in females (Figure 6). In males, ERβ expression decreased fivefold, yet a moderate expression level was maintained. We proposed that estrogen receptor signaling could distinctly regulate early phase WD responses in nerves of both sexes, which is modulated, in part, via estrogen receptors’ sexually-dimorphic expression.
Gene ontology (GO) analysis (Ashburner et al., 2000) identified a relatively small number of regulated DEGs generally attributed to the innate immunity system (GO term 0045087) (Figure 6). Likewise, IPA did not detect significant changes of immune-related canonical pathways (Figure 3). Consistent with time-lapse microarrays in axotomized male rats (Yu et al., 2016), robust baseline expression of pro-inflammatory interleukins IL-1b and IL-6 was observed in both sexes in axotomized and sham mice, albeit post-axotomy activation of IL-6 was stronger in males. Upstream regulator analysis in IPA predicted that interleukins IL-6 and IL-1B could promote upregulation of a larger number of DEGs in males, including important growth and transcription factors (Figure 5B and Supplementary Table 2).
Both females and males attenuated expression levels of kinases, including protein kinases (Figure 6). Males predominantly increased Sphk1 encoding sphingosine kinase 1 that catalyzes the phosphorylation of sphingosine, a lipid mediator with intra- and extracellular functions, including neuroinflammation. The following kinases showed a moderate mRNA increase. The protein kinase adapter gene involved in ubiquitin-dependent protein degradation Trib1 (Satoh et al., 2013), Plk2 (a serine/threonine-protein kinase involved in synaptic plasticity), cyclin-dependent kinase 18 (Cdk18), cGMP-dependent protein kinase 2 (Prkg2), paralemmin A kinase anchor protein (Pakap), and serine/threonine-protein kinase Nek6 important for mitotic cell cycle progression. Females, but not males, exhibited significant downregulation of multiple kinases and kinase-associated protein genes, including Cdkn1c, Plk3, Mapk3k13, Camk2n1, and Camk2b.
In both sexes, MMPs gelatinase A (Mmp2), stromelysin-1 (Mmp3), neutrophil collagenase (Mmp8), gelatinase A (Mmp9), Mmp19, Mmp23, membrane-type MMPs (MT-MMPs) MT1-MMP (Mmp14), MT2-MMP (Mmp15), MT3-MMP (Mmp16), and MT4-MMP (Mmp17) genes exhibited high expression levels in both distal nerve stumps and sham at 24 h (Figure 6). Stromelysin-2 (Mmp10) exhibited a low expression level in both sexes. Several MMPs colocalized in the genomic locus on chromosome 9, Mmp12, Mmp13, and Mmp27, exhibited mild activation in females and a decrease in males. Many ADAM/ADAM-TS genes were expressed at high levels irrespective of sex (Supplementary Table 1). Timp1, Timp2, and Timp3 but not Timp4 expressed at high levels. Timp1 and Timp3 showed a mild increase in both sexes in axotomized nerves. MMP, ADAM/ADAM-TS, and TIMPs gene expression patterns were remarkably consistent in axotomized distal rat nerve stumps (Chernov et al., 2015), suggesting universal proteolytic activities.
Both sexes reduced levels of many myelin-associated genes, including myelin basic protein (MBP), periaxin (Prx), myelin protein zero (Mpz), and myelin-associated glycoprotein (Mag). GFAP maintained high expression in females and was significantly increased in males to a matching mRNA level (Figure 6).
Males, but not females, significantly upregulated Neuregulin Signaling implicated in growth factor-dependent repair and maintenance of the nervous system (Fricker and Bennett, 2011) (Figure 4). Rho GDP-dissociation inhibitor (RhoGDI) encoded by the ARHGDIA gene, p75 neurotrophin receptor (p75NTR, also named NGFR) demonstrated a twofold increase in both sexes. In males, we observed a 4–8-fold increase of the epiregulin (Ereg) and amphiregulin (Areg) autocrine growth factors, ligands of EGF receptor, known to stimulate Schwann cells in axotomized PNS (Rosenbaum et al., 1997; Meier et al., 1999; Jessen and Mirsky, 2021). The proheparin-binding EGF-like growth factor Hbegf associated with macrophage-mediated cellular proliferation exhibited a fourfold increase in males. In addition, probetacellulin (Btc) EGFR ligand showed eightfold upregulation in both sexes. The GDNF Family Ligand Receptor Interactions signaling is regulated by a spectrum of polypeptide growth factors, including the GDNF (Gordon, 2009). Regulation of Mouse Embryonic Stem Cell Pluripotency was male-dominant potentially due to the increase of LIF extrinsic pluripotency factor accompanied by the induction of the intrinsic transcription factors Sox2 and c-Myc.
As reported in the DRGs (Chernov and Shubayev, 2021) and proximal nerve stumps (Chernov and Shubayev, 2022) of the same animal cohort genomic localization of DEGs on a sex-chromosome could determine sexually dimorphic expression. Because females can up- or down-regulate the dosage of X-linked genes by female-exclusive X-chromosome inactivation epigenetic mechanism, the expression of X-linked genes could exhibit significant sexual dimorphism. In the distal stump, females upregulated two X-linked genes for the gap-junction β-1 protein (Gjb1) and the fibronectin type III domain-containing protein 3C1 gene (Fndc3c1). Gjb1 was also activated in males (Figure 7A). In addition, females reduced the expression of over a dozen genes, including Gm7598, Cltrn, Prrg3, Tmem28, Zcchc18, Col4a5, Fgf16, Tnmd, Cnga2, Itm2a, and Xlr3a that did not significantly change in males. Several Y-linked genes in males were strongly expressed, including lysine (K)-specific demethylase 5D (Kdm5d), eukaryotic translation initiation factor Eif2s3y, DEAD box helicase Ddx3y, and ncRNAs (Gm21860 and Gm47283) in both shams and axotomized nerves.
In the distal nerve stump, 172 ncRNA DEGs were detected, including 153 in females, and 51 in males (Figures 7B,C). Reduced expression of opposite strand ncRNA genes occurred in females and included Hoxb3os, H3f3aos, Prdm16os, Rapgef3os2, Tspan32os, Tbx3os1, Pard3bos3, Slc36a1os, Igf1os, and Sap30bpos. Runx2os1, and Junos genes exhibited upregulation in females. Slc36a1os and Igf1os genes were reduced in males.
Interactive network predictions in IPA suggested that networks of cytokines and growth factors, including neuron-specific growth factors, act via their specific receptors to positively regulate neuron growth and proliferation of cells of supporting tissues. The upregulation of these networks was higher in males as compared to females (Figure 8A). Cytokines localized in the extracellular milieu could interact with plasma membrane-associated EGFR, ERBB, and HRAS GTPase and mediation of cytoplasmic and nuclear factors. This regulatory network leads to the activation of axonal guidance, motor neuron outgrowth, mitogenic activity, lipids synthesis, and other pathways promoting axon regrowth. In addition, female-specific networks predicted signaling processes and metabolic changes that could cause a deficit in the myelin maintenance system, leading to pathologic states, such as dysmyelination (Figure 8B).
Sciatic nerve axotomy triggers a strong sex-specific early phase transcriptional response in regenerating (proximal) nerve stump (Chernov and Shubayev, 2022) and corresponding DRGs (Chernov and Shubayev, 2021). The present study documents sexual dimorphism in early phase WD response in the distal sciatic nerve segment of the same animal cohort. WD is a prerequisite to a successful peripheral nerve repair post-injury (Gerdts et al., 2016) orchestrated by denervated Schwann cells. Due to their own remarkable plasticity, Schwann cells de-differentiate immediately after injury to acquire an injury-specific phenotype that drives a sequence of reparative processes in both neurons and immune cells (Arthur-Farraj et al., 2012; Jessen et al., 2015). Through a coordinated transcriptional reprogramming, Schwann cells expedite axonal and myelin proteolytic fragmentation, phagocytosis, and secretion of cytokines and chemokines to foster the immune cell recruitment (Tofaris et al., 2002; Chernov et al., 2015; Jessen et al., 2015). Interleukins IL6 and IL1b could play a significant role to promote transcriptional reprogramming of a large number of genes in WD nerves, especially in males. Except for the moderately upregulated IL-17b, the panel-wide activation of the IL-17 family cytokines was not observed in either sex, consistent with the well-established notion that recruitment of IL-17-expressing immune [T helper (Th)17] cells starts days post-injury (Kim and Moalem-Taylor, 2011). It is important to note, that the upregulation of many cytokine genes in males led to the equalized absolute expression of respective mRNAs to matching levels in both sexes. Estrogens, including β-estradiol, and targeted genes showed sex-dependent WD response. Both Esr1 and Esr2 are expressed in nerves of both sexes and the corresponding DRGs (Chernov and Shubayev, 2022). Injury induced a 10- and 5-fold decrease in Esr2 in female and male nerves, respectively. Under acute pathological conditions, estrogens regulate the neuronal (Hucho et al., 2006) and immune functions (Straub, 2007). In the corresponding DRGs, Esr1 signaling mediates the female-specific pain hypersensitivity (Khomula et al., 2017; Chernov et al., 2020). In the nervous system, rapid attenuation of estrogens in DRG can be complemented by a neuron-specific β-estradiol biosynthesis (Evrard and Balthazart, 2004). In addition to estrogens of gonadal origin (Rettberg et al., 2014) the downstream signaling is likely influenced by the naturally higher β-estradiol levels in females. We concluded that estrogen signaling is a potent regulator of early phase WD response in nerves of both sexes, which is modulated, in part, via estrogen receptors’ sexually-dimorphic expression. Sex specificity in Schwann cell signaling, the dominant cell type in the damaged nerve within 24 h post-injury, manifested in myelin protein and growth factor expression patterns. Thus, the GDNF family of neurotrophic factors known as survival factors for neurons, bind to respective receptors, including upregulated NGFR, which could trigger the phosphorylation of RET tyrosine kinase receptor (Durbec et al., 1996) that is implicated in the regulation of PI3k/Akt and Plcγ/Ip3r dependent Ca2+ signaling during neurogenesis (Fukuda et al., 2002; Lundgren et al., 2012) and, sometimes, hyperalgesia (Bogen et al., 2008). In addition, GDNF/Gfra1 could interact with neural cell adhesion molecules such as NCAM to induce an axonal expansion (Nielsen et al., 2009). GDNF family of ligands can act in a synergetic manner with other growth factors, including transforming growth factor-β (Tgf-β) and sonic hedgehog (Shh) (reviewed in Jessen and Mirsky, 2021). GDNF signaling can stimulate the migration of neuronal precursors and Schwann cells (Iwase et al., 2005) to the site of the injury in both sexes. Action by the neurotrophin growth factors, their receptors, and RhoGDI could inhibit the Rho signaling pathway in both sexes and stimulate the neurite outgrowth (Yamashita and Tohyama, 2003; Dovas and Couchman, 2005). Proteinases and their intrinsic inhibitors are indispensable for remodeling ECM after nerve damage, establishing a permissive environment for axonal regrowth, promoting regeneration and remyelination (Shubayev and Strongin, 2018; Pellegatta and Taveggia, 2019). MMP and disintegrin ADAM/ADAM-TS protease families control cytokine and growth factor receptors and their ligands in peripheral nerves, including IGF and neuregulin/ERBB after the sciatic nerve axotomy (English et al., 2000; Chattopadhyay and Shubayev, 2009; Liu et al., 2010). MMPs in the orthologous human chromosome 11q22.3 region could carry distinct histone modification marks (Chernov et al., 2010), epigenetic control of this set of metzincins could be involved. Consistent with our previous report (Chernov et al., 2015), of the ECM-remodeling metzincins, demonstrate regulated expression in injured nerves. Protease expression levels were high in intact and axotomized nerves. Processes of post-translational protease activation and binding to intrinsic proteinase inhibitors control potentially cytotoxic proteolytic activities. Aberrant cleavage of mediators of nociceptive signaling during neurogenesis could cause neuropathic pain (Remacle et al., 2015). TIMP family members are important regulators of extracellular proteolysis in normal and damaged nerves (Kim et al., 2012; Liu et al., 2015; Nishihara et al., 2015; Chernov and Shubayev, 2021). Timp1 and Timp3 demonstrated high levels of expression and further increased post-axotomy in both sexes. Females reduced the expression of over a dozen of sex chromosome-related genes, including Gm7598, Cltrn, Prrg3, Tmem28, Zcchc18, Col4a5, Fgf16, Tnmd, Cnga2, Itm2a, and Xlr3a that did not significantly change in males. As reported in the DRGs (Chernov and Shubayev, 2021) and proximal nerve stumps (Chernov and Shubayev, 2022), at least in part, the sexual dimorphism may be attributed to differences in baseline expression of sex chromosome-linked genes rather than due to post-axotomy regulation. Sexually dimorphic regulation of lipocalins of the major urinary protein (MUP) family was observed in our study. The male-prevalent activity of MUPs has been attributed to regulating metabolic rate, toxin removal, and survival (Penn et al., 2022). In addition, MUPs carry on chemical communication and pheromone signaling functions (Beynon and Hurst, 2003) in response to a physical injury. The thyroid hormones (T4 and T3) play an essential role in peripheral nerve regeneration producing a lasting and stimulatory action on axotomized neurons and Schwann cells (Barakat-Walter, 1999). Limited data exist on their role in the development of demyelinating diseases (reviewed in Zorrilla Veloz et al., 2022). Disorders of thyroid metabolism are possible causes of inflammation and autoimmune reactions. Selenocysteine-containing Dio2 and Dio3 deiodinases are primarily responsible for T3/T4 enzymatic activation and inactivation, respectively. At 24 h post-axotomy, the upregulation of both enzymes was stronger in males. Thyroid hormone metabolic enzymes exhibited sexually dimorphic expression at the site of nerve axotomy. Their role in nerve injury response requires a focused investigation. We predicted regulatory changes in female and male mice to constitute an immediate pro-regenerative response to nerve injury relative to shams. It is conceivable that, at least partially, sex differences in gene transcription could be matched over a more extended time course of nerve injury response. It is important to note that protein-level expression profiles cannot be directly interpreted using RNA-seq data. Single-cell RNA-seq, spatial transcriptomics analysis could provide additional functional information and a precise understanding of sexually dimorphic mechanisms.
Reagents and resources are listed in Supplementary Table 3.
Female and male C57BL6/J mice (6–8 weeks old, Jackson Labs) were randomly assigned to axotomy (n = 6/group) and sham (n = 6/group). The mice were housed in a temperature-controlled room (∼22°C), on a 12 h light/dark cycle, and had free access to food and water. All procedures were conducted between 8.00 and 12.00 daytime. Under isoflurane anesthesia, the left sciatic nerve was exposed at the mid-thigh level. The entire width of the nerve was axotomized using sterile microsurgery scissors. In a sham surgery animal cohort, nerves were exposed using surgical scissors without transection. The muscle was then sutured, and the skin stapled. At 24 h after surgery, distal stumps of the axotomized sciatic nerve were collected for RNA isolation.
All surgical and tissue harvesting instruments were sterilized and repeatedly treated with RNase Away reagent followed by a rinse in RNase-free water. Tissues were submerged in 500 μl RNAlater stabilization solution, placed at 4°C overnight, then transferred for storage at −20°C. All sample groups were processed in parallel to minimize batch effects.
DRG tissues from 2 animals were pooled to obtain at least 500 ng of total RNAs. Tissues were transferred in Trizol solution and disrupted by mechanical homogenization. Total RNAs were purified using RNeasy RNA purification reagents. RNA concentrations and quality were determined using NanoDrop absorbance ratios at 260/280 nm and 260/230 nm. RNA integrity was determined using the Agilent Bioanalyzer Nano RNA chip. Five hundred nanogram of total RNA samples with RIN ≥ 7.0 were used for RNA-seq.
mRNA libraries were generated following the TruSeq Stranded mRNA library preparation protocol (Illumina). In brief, the Poly-A enriched mRNAs were purified using poly-T oligo coupled magnetic beads, followed by mRNA fragmentation, first and second strands synthesis, cleaning on AMPure XP beads, and 3′-adenylation. Ligation of TruSeq dual-index adapters was used for barcoding. The quality of RNA-seq libraries was validated using qPCR. Libraries were sized on Agilent Bioanalyzer DNA high sensitivity chip and normalized. RNA-seq was performed using the paired-end 100 cycle program on the NovaSeq 6000 system at the Genomics High Throughput Facility (University of California Irvine). Base calls were recorded and converted to FASTQ files containing sequencing reads and the corresponding quality scores using Illumina software. Sequencing was conducted until we acquired at least 50 million paired-end reads per sample.
Data analysis steps are summarized in Supplementary Figure 1. FASTQ files were processed using Ubuntu server 20.04 LTS (64-bit ARM). FASTQ files were filtered to remove low-quality bases, TruSeq dual-index adapter sequences, and unpaired reads using Trimmomatic (Bolger et al., 2014). Transcript-level quantification was performed using Salmon (Patro et al., 2017) in quasi-mapping mode using version M27 of the Gencode mouse genome. To correct systematic biases commonly present in RNA-seq data, both -seqBias and -gcBias features were used. Transcript- to gene-level conversion was done using Tximeta (Love et al., 2020). RNA-seq quantification data quality was assessed using MultiQC (Ewels et al., 2016).
Gene count matrices were imported into the DESeq2 package (Love et al., 2014). Outliers were identified by Cook’s distance method and excluded. Dataset’s normalization was conducted using trimmed M-values (TMM). Log2FC we calculated using the Wald test. The adjusted (shrunken) log2FC values were calculated using the adaptive t-prior apeglm method (Love et al., 2014). Significant DEGs defined by Padj < 0.1 and used in downstream analyses. Batch effects were controlled using removeBatchEffect (Ritchie et al., 2015) and RUVseq (Risso et al., 2014) functions. DEGs were visualized using PCAtools, ComplexHeatmap, and EnhancedVolcano R packages.
Biological interpretation of the regulated signaling pathways in female and male animals was conducted by the Ingenuity Pathway Analysis (IPA) as described previously (Chernov and Shubayev, 2021, 2022) based on causal network approaches (Krämer et al., 2014). Cut-off filtering was applied (Padj < 0.1, |Log2FC| > 1) for IPA. Signaling pathway regulation directionality and upstream regulator analysis were based on statistical z-scores calculated in IPA using default parameters. Ontology terms were retrieved from the GO public database (Ashburner et al., 2000).
The original contributions presented in this study are publicly available. This data can be found here: www.ncbi.nlm.nih.gov/geo/ (GSE182751 and GSE182709).
The animal study was reviewed and approved by the Institutional Animal Care and Use Committee at the VA San Diego Healthcare System.
AC: conceptualization, methodology, software, formal analysis, investigation, data curation, writing—original draft, review and editing, and visualization. VS: conceptualization, resources, writing—review and editing, project administration, and funding acquisition. Both authors contributed to the article and approved the submitted version. |
|
PMC9648410 | 36381660 | Nobuyuki Takahashi,Sehyun Kim,Christopher W. Schultz,Vinodh N. Rajapakse,Yang Zhang,Christophe E. Redon,Haiqing Fu,Lorinc Pongor,Suresh Kumar,Yves Pommier,Mirit I. Aladjem,Anish Thomas | Replication Stress Defines Distinct Molecular Subtypes Across Cancers | 24-06-2022 | Endogenous replication stress is a major driver of genomic instability. Current assessments of replication stress are low throughput precluding its comprehensive assessment across tumors. Here we develop and validate a transcriptional profile of replication stress by leveraging established cellular characteristics that portend replication stress. The repstress gene signature defines a subset of tumors across lineages characterized by activated oncogenes, aneuploidy, extrachromosomal DNA amplification, immune evasion, high genomic instability, and poor survival, and importantly predicts response to agents targeting replication stress more robustly than previously reported transcriptomic measures of replication stress. Repstress score profiles the dual roles of replication stress during tumorigenesis and in established cancers and defines distinct molecular subtypes within cancers that may be more vulnerable to drugs targeting this dependency. Altogether, our study provides a molecular profile of replication stress, providing novel biological insights of the replication stress phenotype, with clinical implications. Significance: We develop a transcriptional profile of replication stress which characterizes replication stress and its cellular response, revealing phenotypes of replication stress across cancer types. We envision the repstress score to serve as an effective discovery platform to predict efficacy of agents targeting replication stress and clinical outcomes. | Replication Stress Defines Distinct Molecular Subtypes Across Cancers
Endogenous replication stress is a major driver of genomic instability. Current assessments of replication stress are low throughput precluding its comprehensive assessment across tumors. Here we develop and validate a transcriptional profile of replication stress by leveraging established cellular characteristics that portend replication stress. The repstress gene signature defines a subset of tumors across lineages characterized by activated oncogenes, aneuploidy, extrachromosomal DNA amplification, immune evasion, high genomic instability, and poor survival, and importantly predicts response to agents targeting replication stress more robustly than previously reported transcriptomic measures of replication stress. Repstress score profiles the dual roles of replication stress during tumorigenesis and in established cancers and defines distinct molecular subtypes within cancers that may be more vulnerable to drugs targeting this dependency. Altogether, our study provides a molecular profile of replication stress, providing novel biological insights of the replication stress phenotype, with clinical implications.
We develop a transcriptional profile of replication stress which characterizes replication stress and its cellular response, revealing phenotypes of replication stress across cancer types. We envision the repstress score to serve as an effective discovery platform to predict efficacy of agents targeting replication stress and clinical outcomes.
Genomic instability is an enabling characteristic of cancer, which by generating genetic diversity expedites the acquisition of multiple hallmark capabilities (1). DNA damage resulting from unabated replication—referred to as replication stress—is a major driver of genomic instability (2). Cells have evolved multiple mechanisms to sense and respond to replication stress, together referred to as the replication stress response (3). When replication fork stalls, the exposed single-stranded DNA (ssDNA) is rapidly coated by ssDNA-binding proteins such as replication protein-A (RPA), leading to activation of ataxia telangiectasia and Rad3-related kinase (ATR), which subsequently phosphorylates downstream kinases including CHK1 (4). ATR and CHK1 negatively regulate cyclin-dependent kinase (CDK) activity through phosphorylation of WEE1 and other substrates. ATR also delays exhaustion of RPA and global breakage of active forks by limiting origin firing (5). Together, the replication stress response cascade prevents stalling of replication forks, controls the initiation of DNA replication, ensures sufficient supply of nucleotides, and limits mitotic entry of cells that have not yet completed DNA replication. Failure to resolve replication stress can lead to collapse of replication forks, DNA double-strand breaks, and acquisition of mutations that are deleterious to genome integrity (2). Replication stress is a feature of precancerous (6) and cancerous cells (7). Cancer cells exhibit heightened replication stress response, for example through CHEK1 amplification, to support rapid proliferation and tolerate the higher levels of replication stress (8). Replication stress itself and the mechanisms that mitigate replication stress are increasingly recognized as cancer cell–specific vulnerabilities that could be exploited therapeutically (9–12). However, rational targeting of these dependencies requires reliable approaches to assess replication stress and its cellular response in patient tumors. Measures of replication stress—including ssDNA or ssDNA-bound RPA levels, phosphorylated form of histone H2AX (γH2AX)—are widely used in experimental settings (13, 14), but are not optimized for use in large cohorts of clinical tumor samples. Here we develop and validate a transcriptional profiling–based approach—the repstress gene signature—that characterizes the cellular response to replication stress at a functional network level (Supplementary Fig. S1).
RNA sequencing (RNA-seq), mutations, copy-number states, drug activity, and doubling time in NCI Development Therapeutics Program small cell lung cancer (NCI-DTP SCLC), Cancer Cell Line Encyclopedia (CCLE), Genomics of Drug Sensitivity in Cancer, Cancer Therapeutics Response Portal (CTRP), and NCI60 were downloaded from CellMiner CDB (15, 16). Clinical, pathologic, and molecular characteristics, survival, RNA-seq, expression of reverse phase protein array (RPPA), genomic alteration, and copy-number alteration for The Cancer Cell Genome Atlas (TCGA) samples were retrieved from data hub of Pan-Cancer TCGA dataset in University of California Santa Cruz Xena platform (17). For other dataset used in this study, please refer Supplementary Text in Supplementary Materials and Methods.
To develop repstress gene signature, we focused on four biological characteristics associating with replication stress in SCLC cell lines: MYC-paralog genes amplification, sensitivity to cell-cycle checkpoint inhibitors, high expression of phosphorylated Chk1 (p-Chk1), and neuroendocrine (NE) differentiation. We defined MYC-amplified SCLC cell lines using the cutoff of 0.7 or more of copy-number score (the average log2-transformed probe intensity ratio of gene specific chromosomal segment DNA relative to normal DNA) in either of MYC family genes (MYC, MYCL, MYCN). Cell-cycle checkpoint inhibitor–sensitive SCLC cell lines were defined as those with drug activity score [standardized, z-score normalized measurements provided from the mean and SD of −log10 (molar concentration causing 50% cell growth inhibition, GI50) values over NCI-DTP SCLC cell lines] of more than 6 with CHK1 inhibitor AZD-7762 (drug ID: 754352) or WEE1 inhibitor MK-1775 (drug ID: 757148). For details of these scores, please refer a previous report describing methods used in CellMiner CDB (16). High expression of p-Chk1 was defined as Chk1_pS345 RPPA expression of more than 0.15. We subsequently applied gene set enrichment analysis (GSEA) using Hallmark gene sets (18) comparing differentially regulated pathways between SCLC cell lines with one of these characteristics and those without. By using adjusted P value of <0.05, we identified two shared hallmark gene sets (HALLMARK_E2F_TARGET and HALLMARK_G2M_CHECKPOINT) as commonly upregulated pathways in SCLC cell lines with one of the repstress characteristics across all of the hallmark genesets. During the GSEA, 11 genes (AURKB, CCNA2, GINS1, KPNA2, LIG3, MTF2, ORC6, PRPS1, SRSF1, SUV39H1, TNPO2) were found as shared leading-edge genes of the two gene sets. Neuroendocrine status of SCLC cell lines (19) and clinical tumors in an independent cohort (20) were assessed using single-sample GSEA (21) of previously described 50 NE gene set, containing 25 genes associated with high neuroendocrine and 25 genes associated with low NE (22). High-neuroendocrine score and low-neuroendocrine score were calculated by single-sample GSEA separately using each of the 25 high of low NE genes and compared the two scores with define high versus low neuroendocrine differentiated SCLC cell lines (15) and clinical tumors (20). Subsequently, differentially expressing genes were analyzed between high versus low neuroendocrine differentiated SCLC cell lines or tumors in each cohort. Among identified highly expressing genes in neuroendocrine differentiated SCLC, by FDR of <10% by Mann–Whitney U test followed by adjusting multiple testing with Benjamini–Hochberg test, those identified in both two cohort and involved in DNA damage repair pathways (23) were defined as additional repstress signature genes (GADD45G, POLA1, POLD4, POLE4, RFC5, RMI1, and RRM1). We finally excluded the gene KPNA2 from the repstress gene signature because it did not frequently express in cell lines other than SCLC (Supplementary Table S1). Repstress score was calculated by applying principal component analysis–based weighting score. In detail, SCLC cell lines were projected onto principal component analysis plot using the scores for biological characteristics associated with replication stress described above and the 17 repstress gene expression were also projected onto the plot, which achieved variable loadings of first principal component dimension for each gene as gene weight (Supplementary Fig. S2A; Supplementary Table S1). We summed up the measurements of repstress signature gene expressions (Z score–normalized in each cell line across all of sequenced gene expressions) multiplied by each gene weight and defined as repstress score. Repstress scores were Z score–normalized among samples used in each analysis and shown in figures.
Nine SCLC cell lines (NCI-H1048; RRID: CVCL_1453, NCI-H1341; RRID: CVCL_1463, NCI-H841; RRID: CVCL_1595, DMS114; RRID: CVCL_1562, NCI-H211; RRID: CVCL_1529, NCI-H446 RRID: CVCL_1562, NCI-H889: RRID: CVCL_1598, NCI-H146; RRID: CVCL_1473, NCI-H524; RRID: CVCL_1568) were purchased from ATCC and maintained in cell culture. H211, H889, H1048, and H1341 cell lines are female and the rest are male. Cell lines were authenticated using short tandem repeat analysis, and were monthly tested for Mycoplasma contamination. Cell media was RPMI1640 supplemented with 10% FBS for all lines to maintain consistency. Cells were grown at 37°C and 5% CO2 were used in subsequent experiments.
Cells were lysed with RIPA buffer containing protease inhibitor cocktail (Thermo Fisher Scientific) and micrococcal nuclease (Thermo Fisher Scientific). The resulting mixtures were incubated on ice for 30 minutes, then centrifuged 20 minutes to get the supernatants. After adding Tris-Glycine SDS sample buffer including 5% of 2-mercaptoethanol, the lysates were boiled for 10 minutes, analyzed by SDS-PAGE, and immunoblotted with various antibodies as follows: RPA phosphorylation (pS4/8, from Bethyl Laboratories; RRID: AB_2891810); total RPA (from Bethyl Laboratories; RRID: AB_185548); pATR (T1989, from Cell Signaling Technology; RRID:AB_2722679); and pCHK1(S345, from Cell Signaling Technology; RRID:AB_330023). To start Western blot analysis, nitrocellulose membrane was blocked with 5% nonfat milk, then incubated with primary antibodies at 1:1,000 dilution in PBST buffer (PBS containing 0.1% Tween 20) containing 1% nonfat milk, at 4°C overnight. After washing with PBST three times, the membrane was incubated with second antibody at 1:2,000 dilution in PBST buffer containing 1% nonfat milk, at room temperature for 1 hour. The Western blot analysis results were developed by Bio-Rad ChemiDoc MP Imaging System.
Cells were fixed with 2% paraformaldehyde, followed by the incubation with 70% cold ethanol. After blocking with 5% BSA, primary antibody staining was performed as follows: anti-γH2AX (1:500, Millipore, 05-636), anti-phosphorylated replication protein A (pRPA; 1:500, Bethyl Laboratories, A300-245A; RRID: AB_210547). Secondary antibody staining was performed as follows: Alexa 488–conjugated anti-mouse lgG and Alexa 594–conjugated anti-rabbit lgG (1:500, Cell Signaling Technology, 4408 and 8889). 4′,6-diamidino-2-phenylindole (DAPI) staining was performed with VECTASHIELD mounting medium with DAPI (H-1200, Vector Laboratories). A Zeiss LSM780 confocal microscope was used to capture the fluorescence. The Colocalization Plugin of the FIJI-ImageJ software was used to calculate the fluorescence density.
Cell lines were plated at 1 million cells per 10-cm plate. After 24 hours, cells were treated for 2 hours with either DMSO control or 10 μmol/L topotecan, and for 1 hour (the second hour of topotecan treatment) with 1 μmol/L 5-ethynyl-2′-deoxyuridine (EdU). Cells were fixed in and stained for γH2AX as described previously (24), followed by Click-iT Chemistry as per manufacturer's instructions utilizing the Click-iT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit C10634 (Thermo Fisher Scientific). Flow cytometry data were collected using a BD LSRFortesa and analyzed utilizing FlowJo V10.7.1.
As described previously (25), asynchronous DMS114 and H524 cells were sequentially labeled with 20 μmol/L IdU for 20 minutes and 50 μmol/L CldU for 20 minutes. To preserve long genomic DNA fibers, cells were embedded in low melting point agarose plugs and incubated in cell lysis buffer with proteinase K at 50°C overnight. Washed plugs with TE buffer, and then melted plugs in 0.1 mol/L MES (pH 6.5) at 70°C for 20 minutes. Agarose was subsequently degraded by adding 2 μL of β-agarase (New England Biolabs). DNA fibers were then stretched onto salinized coverslips (Genomic Vision, cov-002-RUO) using an in-house combing machine. Combed DNA on coverslips was then baked at 60°C for 2 hours and denatured in 0.5 N NaOH for 20 minutes. IdU, CldU, and ssDNA were detected using a mouse antibody directed against BrdU (IgG1, Becton Dickinson, 347580, 1:25 dilution), a rat antibody directed against BrdU (Accurate Chemical, OBT0030, 1:200 dilution), and a mouse antibody directed against ssDNA (IgG 2a, Millipore, MAB3034, 1:100), respectively. The secondary antibodies used were goat anti-mouse Cy3 (Abcam ab6946), goat anti-rat Cy5 (Abcam, ab6565), and goat anti-mouse BV480 (Jackson ImmunoResearch, 115-685-166) for ssDNA. Slides were scanned with a Fiber-Vision Automated Scanner (Genomic Vision). Replication signals on single DNA fibers were analyzed using FiberStudio (Genomic Vision).
All figures were generated using CellMiner CDB (16), GraphPad PRISM software version 8.1.2 (GraphPad Software), R version 1.2.135 (R Foundation for Statistical Computing), and STATA software version 16.0 (StataCorp). Box plots in this article were shown by Tukey box and whisker plots, unless specifically indicated in figure legends. Methods for statistical analyses were indicated in the article and figure legends and were performed using softwares described above. Overall survival (OS) curves were created by the Kaplan–Meier method and compared by log-rank test. All statistical tests were two sided.
The data analyzed in this study were obtained from public database. The experimental data generated in this study are available upon request from the corresponding author.
While replication stress is widely prevalent across cancers, it is more central to the tumorigenesis of some cancers than others (7). We chose to develop a replication stress response signature in SCLC, a fast-growing and deadly cancer with molecular and clinical features distinct from other lung cancers. We reasoned that signatures that report replication stress response in SCLC could then be extended to other tumors that also exhibit this phenotype. SCLCs are characterized by high degree of genomic instability, an important consequence of replication stress (26). Nearly all SCLCs have loss-of-function alterations in tumor suppressors RB1 and TP53, and frequently exhibit amplification and overexpression of oncogenes such as MYC (20). SCLCs also exhibit sustained high expression of lineage transcription factors, which contribute to replication stress (27), and are highly vulnerable to perturbation of the transcriptional state (28, 29). Not surprisingly, the standard treatment of SCLC consists mostly of DNA-damaging agents such as platinum compounds, topoisomerase I and II inhibitors, and an alkylating agent temozolomide. To obtain a comprehensive molecular understanding of the replication stress response, we examined a panel of 67 SCLC cell lines characterized by microarray-based gene expression, representing the molecular diversity of the disease (15, 19). We reasoned that SCLC cells under high replication stress might be characterized by amplification of MYC and its paralogs MYCN and MYCL (30, 31); expression of p-Chk1 (32); sensitivity to inhibitors of cell-cycle checkpoints CHK1 and WEE1 (33); and NE differentiation (12, 29, 34). GSEA was performed to define differentially regulated biological processes between SCLCs with and without these features, revealing cell cycle–related targets of E2F transcription factors and genes involved in the G2–M checkpoint (AURKB, CCNA2, GINS1, LIG3, MTF2, ORC6, PRPS1, SRSF1, SUV39H1, TNPO2) and DNA replication and repair genes associated with NE differentiation (GADD45G, POLA1, POLD4, POLE4, RFC5, RMI1, and RRM1), together designated as the repstress gene signature (Fig. 1A; Supplementary Table S1). Repstress signature score was calculated using weighted principal component analysis (Supplementary Fig. S2A; Supplementary Table S1), with most genes providing positive signature weightings except POLD4 and POLE4. Repstress signature included genes involved in mitosis (AURKB), cell-cycle progression (CCNA2), initiation of replication and replisome progression (GINS1, ORC6, RFC5), nuclear transport (TNPO2), DNA and RNA metabolism (LIG3, PRPS1, RMI1, RRM1), transcriptional regulation (MTF2, SUV39H1), RNA splicing (SRSF1), and DNA polymerases (POLA1, POLD4, POLE4). High repstress cells had elevated expression of MDC1, CLSPN, and TIMELESS, genes involved in replication stress tolerance by protecting the replication fork, downstream effectors CHEK2 and CDC25A, and genes associated with proliferation PCNA and MKI67 (ranges of Spearman correlation coefficient and multiple testing Padj value: 0.22 to 0.61 and 5.5 × 10−7 to 7.7 × 10−4, respectively). In contrast, DNA damage sensors RAD9A and RAD17 and sensor kinases ATM and ATR were less correlated with repstress score (Fig. 1B and C; Supplementary Fig. S2B–D). Repstress score correlated positively with the expression of genes involved in solving topological problems during replication (TOP2A), facilitating the repair and restart of stalled replication forks (FANCD2), resolving barriers to replication fork progression (RNASEH2A), and DNA repair (POLQ and PARP1; Supplementary Fig. S2E–S2I). Stalled replication forks require the surrounding chromatin to be compacted for their stabilization (35); the expansion of heterochromatic regions is mediated by histone modifications and attenuates replication stress signaling. We reasoned that if repstress score captures replication stress response at a functional network level, it may be able to predict the heterochromatin response as well. To test this possibility, we examined pairwise correlations between the repstress score and expression of chromatin remodelers and histone modifiers. Repstress score correlated positively with the expressions of heterochromatin proteins HP1α, HP1β, and HP1γ that associate with methylated histone H3 on nucleosomes and mediate heterochromatin formation (ranges of Spearman correlation coefficients and multiple testing Padj values: 0.44 to 0.56 and 1.4 × 10−5 to 2.8 × 10−3, respectively). In contrast, genes involved in INO80 chromatin remodeling complex (INO80 and ARP8) were less correlated with repstress signature and clustered separately (0.11–0.25 and 0.6–1.0, respectively; Supplementary Fig. S2J). Stressed DNA replication results in DNA double-strand breaks, which induce rapid phosphorylation of H2AX on Ser139, termed as γH2AX. γH2AX is a sensitive albeit indirect indicator of replication stress (36). We detected higher basal endogenous expression of γH2AX by Western blot analysis in SCLC cells with high repstress score compared with cells with low repstress score (Spearman correlation coefficient and P value: 0.80 and 0.0096, respectively; Fig. 1D and E). Other replication stress–associated proteins such as phosphorylated RPA, Chk1, and ATR also had positive correlations with repstress score (Fig. 1F and G; Supplementary Fig. S3). Higher basal levels of γH2AX and phosphorylated RPA were also detected by fluorescence microscopy in repstress-high H524 cell line compared with repstress-low DMS114 (Supplementary Fig. S4). We then assessed whether cells with variable repstress scores responded differentially to exogenous replication stress, using topotecan which produces replication blocks by generating topoisomerase I−DNA cleavage complexes, in two representative cell lines H524 and DMS114 with high and low repstress scores, respectively. At basal levels without drug treatment, H524 cells exhibited lower DNA synthesis and more DNA damage during S-phase, as indicated by the proportion of cells labeled with EdU and γH2AX, respectively, compared with DMS114 cells. Upon treatment with topotecan, DNA synthesis and cell proliferation were inhibited to a much lesser extent in H524 cells compared with DMS114 (Fig. 1H; Supplementary Fig. S5), resulting in higher induction of γH2AX in H524 (Fig. 1H and I). The γH2AX induction by topotecan treatment correlated with the repstress score in a larger panel of SCLC cell lines (Supplementary Fig. S6). To further elucidate the dynamics of DNA replication, we performed DNA combing assay. H524 cells had markedly lower fork velocities and inter-origin distances compared with DMS114 (Fig. 1J–L). Shorter inter-origin distances can result from activation of dormant origins due to oncogene-induced replication stress which slows or stalls replication forks (37). Furthermore, the patterns of bidirectional fork movement were more asymmetric in H524 cells compared with DMS114 (Fig. 1M and N), indicating that higher repstress gene expression associates with replication fork stalling. Together, we find that the molecular components involved in replication stress response are interconnected. Repstress score captures the coordinate expression of key components of this cascade downstream of checkpoint sensors and kinases with the associated chromatin changes. Even in an unchallenged S-phase, high repstress score cells exhibit more endogenous replication stress and robust activation of DNA damage response (DDR) than low repstress cells. However, they are hypersensitive to exogenous replicative stress likely because further recruitment of replication stress response is less effective. Thus, the repstress gene signature could allow for interrogation of endogenous replication stress and efficiency of the replication stress response in SCLC cell lines.
To determine whether the repstress gene signature was generalizable and able to predict replication stress response signaling in cancers beyond SCLC, we queried RNA-seq and RPPA data from the CCLE of 937 cell lines across 20 cancer types (Fig. 2A; ref. 38). Highest repstress scores were found in SCLC (the number and proportion of SCLC cells with repstress score ≥95% confidence interval of repstress score across all CCLE cell lines: 48/50, 96.0%), hematopoietic malignancies [non–Hodgkin lymphoma (43/49, 87.8%) and leukemia (57/78, 73.1%)], and sarcoma (55/87, 63.2%), consistent with previous reports of these malignancies exhibiting high replication stress phenotype (39, 40). Low repstress scores were observed in renal cell carcinoma (the number and proportion of cells with repstress score <95% confidence interval of repstress score across all CCLE cell lines: 22/31, 71.0%), pancreatic cancer (15/23, 65.2%), ovarian cancer (30/46, 65.2%), melanoma (35/56, 62.5%), and thyroid cancer (6/11, 54.5%). The distribution of repstress score across cancer types was overall similar when DNA repair genes associated with NE were excluded from the signature, with SCLC and hematopoietic malignancies exhibiting the highest scores (Supplementary Fig. S7), suggesting that the high repstress score in SCLC is not confounded by NE, a pathophysiologic characteristic of this cancer. Similar to SCLC cell lines, the repstress score was positively correlated with expression of key genes involved in increasing replication stress tolerance across cancer types (Fig. 2B). Pairwise correlations recapitulated the correlation of repstress score with expression of DDR mediators, effectors, and heterochromatin, in contrast to sensors and sensor kinases at the mRNA and protein levels (Supplementary Fig. S8). Genotoxic agents currently used for cancer therapy include many potent inducers of replication stress, such as platinum derivatives, topoisomerase inhibitors, and nucleotide analogs (41). We hypothesized that repstress gene signature may profile drug induced modulation of replication stress in diverse cancers types. To investigate this possibility, we examined repstress score dynamics pretreatment and posttreatment with 15 anticancer agents across a panel of 60 human cancer cell lines of different lineages (42). Cells were exposed to these agents at concentrations below the human peak plasma concentration and the average concentration resulting in 50% cell growth inhibition. In a group of cell lines, we identified similar transcriptional responses to gemcitabine, cisplatin, and topotecan, which resulted in notable induction of repstress gene expression after treatment (Fig. 2C; Supplementary Fig. S9A–C). Topotecan and cisplatin induce replication blocks respectively by generating topoisomerase I−DNA cleavage complexes and platinum–DNA adducts, whereas gemcitabine stalls replication through its integration into DNA and depletion of the deoxyribonucleotide pool. In contrast, treatment with tyrosine kinase inhibitors sorafenib and dasatinib, and the histone deacetylase inhibitor vorinostat resulted in uniformly decreased repstress gene expression (Fig. 2D; Supplementary Fig. S9A, S9D, and S9E). Together, repstress gene signature stratifies cancer cell lines across tumor types based on their adaptability to replication stress and profiles transcriptional responses to drug-induced modulation of replication stress. Molecular features that contribute to the replication stress phenotype including drug responses across cancer cell line databases may be explored at this web-based resource: https://discover.nci.nih.gov/cellminercdb/ (15, 16).
Cancers with heightened replication stress response may be particularly vulnerable to drugs that target this dependency. We investigated whether the repstress score predicts drug sensitivity using 481 anticancer drugs across 823 cell lines of the CTRP (43). Drug sensitivities were compared between cell lines defined by the lowest (<25th) and highest (≥75th) repstress score percentiles. With FDR of 5%, 280 compounds were identified as significantly more or less active in repstress-high compared with repstress-low cell lines (Supplementary Fig. S10A). High repstress score cells were more sensitive to inhibitors of polo-like kinase-1 (BI-2536: Padj = 2.4 × 10−28), topoisomerase I (topotecan: Padj = 1.1 × 10−21), aurora kinase A and B (alisertib: Padj = 2.0 × 10−20), and regulators of cell-cycle progression and DNA replication (gemcitabine: Padj = 9.4 × 10−17; Fig. 2E; Supplementary Fig. S10). In contrast, low repstress score cells were more sensitive to compounds targeting pathways such as mitogen-activated protein kinase (MAPK) and EGFR (Fig. 2E; Supplementary Fig. S10A). This observation is consistent with a recent study in isogenic cell lines which reported MAPK signaling dependence in replication stress response defective cells (44). Repstress score exhibited a higher positive correlation with response to agents that induce replication stress, including alisertib, BI-2536, topotecan, and gemcitabine, than the currently available cell-cycle proliferation genes (refs. 39, 45–47; Fig. 2F, Supplementary Fig. S11). Because of the critical functions of ATR in protecting cells under replication stress, small-molecule ATR inhibitors are being explored as cancer therapeutic agents to selectively kill cancer cells under replication stress (9). A reliable method to measure replication stress levels could in principle enable patient stratification for ATR inhibitor therapies. We examined whether the repstress signature predicted sensitivity to ATR inhibitors (48). Across 16 cancer cell lines from different histologies, cells with high repstress score showed higher sensitivity to ATR inhibitor M4344 than cells with low repstress score (Spearman r = 0.88, P < 2.0 × 10−16; Fig. 2G). Repstress score better predicted ATR inhibitor response than the previously described signatures of replication stress and proliferative gene expression signatures (Fig. 2H; Supplementary Fig. S12; refs. 39, 45–47).
Replication stress is a driver for cancer progression and is linked to genomic instability in precancerous lesions and cancers (7). In precancerous lesions, the replication stress response provides a barrier to delay or prevent tumorigenesis (6, 8, 49). Using repstress score, we assessed replication stress along the continuum of cancer development (50). Repstress scores were higher in bronchial precancerous lesions which eventually regressed and those that progressed to become cancers, compared with lesions that maintained stable precancerous characteristics (Fig. 3A), supporting the dual roles of replication stress in promoting genomic instability, and in slowing down cell proliferation and activating anticancer barriers (8). To explore the replication stress response profiles of cancers, we analyzed over 10,000 tumors of 33 cancer types from TCGA. As with cell lines, expression of genes required for survival of replication stress and DNA damage repair (TIMELESS, CLSPN, TOP2A, FANCD2, RNASEH2A, POLQ, and PARP1) positively correlated with repstress scores (Supplementary Fig. S13A–G). These associations were also maintained at the protein level across tumor types; expression of proteins that most highly correlated with repstress score included CYCLINB1, CYCLINE1, CHK2, 4EBP1, phosphorylated CDK1 and PCNA (Supplementary Fig. S13H). We next assessed repstress scores across normal tissue, localized, and metastatic cancers. Normal tissue had the lowest repstress score compared with cancers, and hematologic malignancies had higher repstress score than epithelial cancers (Fig. 3B). We observed large variance in repstress scores across cancer types, implying significant differences in replication stress response proficiency among different cancers (Fig. 3C). High repstress gene expression was observed in testicular germ cell tumors (TCGT, the number and proportion of TCGT with repstress scores ≥95% confidence interval of repstress score across TCGA: 148/156, 94.9%), cervical squamous cell carcinoma (302/307, 98.4%), and hematologic malignancies (diffuse large B-cell lymphoma: 46/48, 95.8%; and acute myeloid leukemia: 161/173, 93.1%). In general, tumors with high repstress scores were highly proliferative tumors typically treated with DNA-damaging therapies such as platinum and topoisomerase inhibitors. A notable exception was thymoma which had high repstress scores (THYM: 96/120: 80.0%) despite a relatively indolent growth pattern. This may be explained by the prominent role of E2F2 in promoting unscheduled cell division and oncogenic transformation of thymic epithelial cells (51). Cancer types with lower repstress scores included thyroid cancers (THCA: the number and proportion of THCA with repstress scores <95% confidence interval of repstress score across TCGA: 513/513, 100%), kidney cancers [renal papillary cell carcinoma (KIRP): 284/291, 97.6%; renal clear cell carcinoma (KIRC): 521/534, 97.6%; kidney chromophobe (KICH): 63/66, 95.5%], and pancreatic adenocarcinoma (PAAD: 172/179, 96.1%). The distribution of repstress score across cancers was overall similar even after excluding the seven genes associated with NE differentiation (Supplementary Fig. S14). Because replication stress is driven by activation of oncogenes and absence of tumor suppressor genes (52), we examined the association between repstress score and mutations or copy-number states in these genes. Tumors with mutated oncogenes (Fig. 3D) and tumor suppressor genes (Fig. 3E) had higher repstress scores compared with tumors with no mutations affecting these genes. In most cancer types, repstress score was significantly higher in tumors harboring mutations in DNA repair and cell cycle–related genes (Supplementary Fig. S15A), suggesting deregulation of these pathways underlying increased replication stress. Tumors with TP53 or RB1 mutations had significantly higher repstress score compared with those without (Supplementary Fig. S15B and S15C) and a loss of Rb1 function score (53) positively correlated with repstress score (Supplementary Fig. S15D). Notably, there was no association between repstress score and the number of point mutations (Supplementary Fig. S15E). In contrast, somatic copy-number alterations (54) at chromosome, arm, and focal levels (Fig. 3F; Supplementary Fig. S15F) and whole-genome doubling (Supplementary Fig. S15G) were positively correlated with repstress score. Extrachromosomal DNA (ecDNA) amplification has recently been reported to promote aneuploidy and genomic instability (55). Tumors with ecDNA amplification had higher repstress scores compared with those without (Fig. 3G), with increasing number of ecDNA amplicons associated with higher repstress scores (Supplementary Fig. S16). Consistent with cancer stem cells displaying robust replication stress response to prevent the accumulation of genetic lesions (56), a cancer stemness gene signature score (57) positively correlated with repstress score (Fig. 3H). Next, we examined repstress score among previously defined cancer immune subtypes (58). The wound healing and IFNγ dominant subtypes had higher repstress scores compared with the other immune subtypes, including notably the inflammatory subtype which had lower repstress scores (Fig. 3I). The association of wound healing and repstress score (Pearson r = 0.81, P < 0.0001; Supplementary Fig. S17A; ref. 58), consistently observed across nearly all cancer types (Supplementary Fig. S17B), is supported by previous work showing the similarities in cellular responses to cancer progression and wound healing (59). Th cells play a key role in the adaptive immune system by coordinating effector functions leading to destructive responses, including pathogen clearance and autoimmunity. A proinflammatory Th1 subtype response score was negatively correlated with repstress score (Pearson r = −0.34, P < 0.0001), whereas immunosuppressive Th2 subtype response score correlated positively (Pearson r = 0.76, P < 0.0001; Fig. 3J and K). Accordingly, high repstress score was associated with poor survival in an independent cohort of melanoma patients treated with immune checkpoint inhibitor nivolumab (ref. 60; Supplementary Fig. S18). Finally, we analyzed the impact of repstress score on patient outcomes. Patients with high repstress tumors had poorer OS compared with patients with low repstress tumors [HR (95% confidence interval): 2.0 (1.8–2.3), P < 0.0001 by log-rank test; Fig. 3L]. Multivariate Cox regression analysis revealed that the repstress score independently contributed to poor survival after adjusting known variables associated with survival including age at diagnosis, sex, pathologic/clinical stage, and cancer type (Supplementary Table S2; Supplementary Fig. S19). Together, these analyses functionally link replication stress and its cellular response as measured by the repstress score with oncogene alterations, tumor aneuploidy, ecDNA amplification, cancer stemness, immunosuppressive T-cell responses, and inferior survival across cancers.
Given the wide range of repstress scores in individual cancers (Fig. 3C), we hypothesized that the repstress score can identify distinct molecular subtypes within cancer types. Among breast cancers, the basal subtype, characterized by expression of markers such as cytokeratins 5 and 6 (61), had significantly higher repstress score compared with the luminal A, luminal B, and HER2-enriched subtypes (Fig. 4A). Triple-negative breast cancers, which share similarities to the basal subtype, were also characterized by higher repstress score gene expression than tumors that expressed estrogen, progesterone, or HER2 receptors (Supplementary Fig. S20A). Pancreatic cancers with transcriptionally defined basal characteristics and squamous features on histology harbored higher repstress score than those without these features in TCGA and an independent cohort (Fig. 4B; Supplementary Fig. S20B–S20F; ref. 62). Malignant mesothelioma with sarcomatoid histology, defined by infiltrative spindle or mesenchymal appearing cells and poor prognosis, were characterized by higher repstress score than epithelioid mesothelioma (Fig. 4C). Among prostate cancers, repstress score showed a positive correlation with Gleason score (Fig. 4D), an indicator of prostate cancer differentiation, with the highest Gleason score associated with the most poorly differentiated and aggressive subtype (63). In addition, prostate cancers with higher copy-number alterations (64) had higher repstress scores compared with those with less frequent copy-number alterations (Fig. 4E). Similarly, uterine corpus endometrial carcinoma with genomic instability defined by high copy-number alterations, POLE mutations, and microsatellite instability (65) had higher repstress score compared with low copy-number altered tumors (Fig. 4F). Repstress score also identified a proliferative subtype of ovarian cancer (ref. 66; Fig. 4G), and aggressive subtypes of hepatocellular carcinoma (iCluster 3; ref. 67) with higher degree of chromosomal instability and TP53 mutations (Fig. 4H). Given recent studies linking oncoviruses with genomic instability and replication stress (68), we examined repstress score in oncovirus-derived cancers. Human papilloma virus (HPV)-associated head and neck cancers had significantly higher repstress scores compared with non–HPV-associated cancers (Fig. 4I). A similar trend was also observed in cervical cancer, another HPV-related cancer (Supplementary Fig. S20G). Replication stress exposes tracts of ssDNA that form substrates for APOBEC3-deaminase–mediated mutagenesis (69). Accordingly, repstress score positively correlated with APOBEC3B expression in breast cancer, lung adenocarcinoma, and acute myeloid leukemia, malignancies wherein APOBEC3B is upregulated and plays a key role in mutagenesis (ref. 70; Fig. 4J–L). STK11 and KEAP1 co-mutated lung adenocarcinoma, which are associated with aggressive tumor growth and immunotherapy resistance (71), had higher repstress scores compared with lung adenocarcinoma without concomitant loss of these genes (Fig. 4M). Among KRAS-mutant lung adenocarcinoma, a particularly aggressive subset with STK11 comutations (72) had higher repstress scores compared with tumors without comutations (Fig. 4N). Non–small cell lung cancer cell lines with KRAS/STK11 comutations were more sensitive to a CHK1/2 inhibitor than cell lines without STK11 comutations (Supplementary Fig. S21). Together, our analysis brings to light the dependence of certain tumor types and subtypes of tumors on replication stress response, potentially representing important therapeutic opportunities.
DNA replication is a tightly regulated process. Replication stress and DNA damage ensue when these regulatory mechanisms fail. Causes of replication stress are diverse. Even single oncogenes can induce replication stress by different mechanisms depending on the context (73). In fact, the causes of replicative stress might be quite dynamic during tumorigenesis. Independent of the causes of replication stress, cells have evolved a complex mechanism which ensures that the genome is accurately duplicated in each cell cycle. Despite its critical role in tumorigenesis and emerging importance as a potential therapeutic target, replication stress and its phenotypic characteristics have not been explored in high-throughput sequencing studies of human cancers. Many available studies examining replication stress to date have focused on individual tumor types, for example in ovarian cancer (74), pancreatic cancer (75, 76), or selected features that drive replication stress, for example overexpression of oncogenes (via overexpression of CDC25A, CCNE1 or MYC; ref. 77) or replication stress response defects (via depletion of ATR, ATM, CHEK1, or CHEK2; ref. 44). Here we describe a gene expression signature, capturing broad measures of replication stress–related gene expression using an approach compatible with formalin-fixed paraffin-embedded clinical samples, allowing interrogation of replication stress at a functional network level across cancers, independent of the underlying mechanisms. The global view of replication stress provided by the repstress signature reveals heightened genomic instability, immune evasion, and poor survival in subsets of tumors across lineages, and enabled identification of cancer subtypes that may be more vulnerable to replication and replication stress response inhibitors including the novel ATR inhibitors (Fig. 4O; Supplementary Fig. S1). Repstress score provides a framework to investigate the link between replication stress and its functional consequences. Our analyses implicate copy-number alterations rather than base-pair mutations as a key consequence of genomic instability linked to DNA replication stress. These results support the oncogene-induced DNA replication stress model for cancer development wherein chromosomal instability in sporadic cancers results from oncogene-induced collapse of DNA replication forks, which in turn leads to DNA double-strand breaks and genomic instability (78). Another consequence of replication stress is abnormal chromosome segregation which may result in formation of micronuclei (79) and nonchromosomal DNA elements (55). Indeed, we find a positive correlation between repstress gene expression and ecDNA amplification, suggesting that oncogene-induced replication, abnormal chromosome segregation, and chromosome instability may be driving ecDNA formation. Repstress gene signature reveals the dynamic nature of the replication stress response during tumorigenesis and following drug treatment. Bronchial precancerous lesions that eventually regress and those that progress to become cancers are characterized by high repstress score compared with lesions that maintain stable precancerous characteristics. These results are consistent with the fundamental role of replication stress response in early stages of cancer development maintaining genomic integrity and preventing tumorigenesis (6, 8) while generating DNA damage and contributing to rapid evolution and genetic heterogeneity in established cancers (52). Whether these insights could enable the currently sparse toolset to identify and treat premalignant lesions at risk for progression to cancer needs further study (80). Modulation of repstress score following treatment suggests the utility of the signature to profile to study agents in terms of their impact on replication stress. Repstress score provides insights into tumor phenotypes associated with high replication stress. Across multiple datasets, repstress score was an independent predictor of poor survival after adjusting known variables associated with survival. Notably, we find substantial enrichment of TCGA wound healing and IFNγ dominant phenotypes among high repstress tumors. The dominant anti-inflammatory Th2 response and rapid tumor growth that preclude immune control may explain the notably less favorable outcomes in high repstress score tumors despite a substantial immune component. It is also likely that these tumors have already been remodeled by the existing robust Th1 infiltrate and have escaped immune recognition. Furthermore, the repstress score enabled delineation of several prognostically relevant subtypes within diverse cancer types, including high Gleason score prostate cancer, basal subtype of breast cancer, sarcomatoid mesothelioma, proliferative subtypes of ovarian cancer and hepatocellular carcinoma, and pancreatic cancer with squamous differentiation. Additional studies are warranted to define clinically relevant and tumor-type specific repstress score thresholds, but it is notable, and probably the singular strength of the study, that repstress gene signature stratifies tumors across and within cancer types beyond SCLC based on the likelihood of drug response and prognosis. The generalizability of repstress score beyond SCLC suggests that while the causes of replication stress are varied, the replication stress response pathways are conserved across cancers, and thus may represent a shared therapeutic vulnerability. Upregulation of cell-cycle genes is a common denominator between highly proliferative cells and cells under high replication stress, and further studies are needed to understand the contribution of individual repstress genes to these characteristics. It is notable that repstress signature better predicted response to ATR inhibitors than previously described gene signatures of proliferation (39, 45–47), suggesting that repstress signature captures transcriptional changes of replication stress in addition to proliferation. In conclusion, gene expression profiling–based assessment of replication stress using the repstress signature represents a powerful approach to dissect the replication stress response. We anticipate the repstress score to have therapeutic implications, enabling stratification of patients for therapies that modulate replication stress.
Click here for additional data file. |
|
PMC9648412 | 36381237 | Kateryna Krytska,Colleen E. Casey,Jennifer Pogoriler,Daniel Martinez,Komal S. Rathi,Alvin Farrel,Esther R. Berko,Matthew Tsang,Renata R. Sano,Nathan Kendsersky,Stephen W. Erickson,Beverly A. Teicher,Kumiko Isse,Laura Saunders,Malcolm A. Smith,John M. Maris,Yael P. Mossé | Evaluation of the DLL3-targeting Antibody–Drug Conjugate Rovalpituzumab Tesirine in Preclinical Models of Neuroblastoma | 11-07-2022 | Neuroblastomas have neuroendocrine features and often show similar gene expression patterns to small cell lung cancer including high expression of delta-like ligand 3 (DLL3). Here we determine the efficacy of rovalpituzumab tesirine (Rova-T), an antibody–drug conjugated (ADC) with a pyrrolobenzodiazepine dimer toxin targeting DLL3, in preclinical models of human neuroblastoma. We evaluated DLL3 expression in RNA-sequencing datasets and performed IHC on neuroblastoma patient-derived xenograft (PDX), human neuroblastoma primary tumor and normal childhood tissue microarrays. We then evaluated the activity of Rova-T against 11 neuroblastoma PDX models using varying doses and schedules and compared antitumor activity with expression levels. DLL3 mRNA was differentially overexpressed in neuroblastoma at comparable levels to small cell lung cancer, as well as Wilms and rhabdoid tumors. DLL3 protein was robustly expressed across the neuroblastoma PDX array, but membranous staining was variable. The human neuroblastoma array, however, showed staining in only 44% of cases, whereas no significant staining was observed in the normal childhood tissue array. Rova-T showed a clear dose–response effect across the 11 models tested, with a single dose inducing a complete or partial response in 3 of 11 and stable disease in another 3 of 11 models. No overt signs of toxicity were observed, and there was no treatment-related mortality. Strong membranous staining was necessary, but not sufficient, for antitumor activity. Rova-T has activity in a subset of neuroblastoma preclinical models, but heterogeneous expression in these models and the near absence of expression seen in human tumors suggests that any DLL3-targeting clinical trial should be only performed with a robust companion diagnostic to evaluate DLL3 expression for patient selection. Significance: GD2-directed antibody therapy is standard of care for high-risk neuroblastoma; therapy is toxic, and relapses often occur. DLL3, an inhibitory Notch ligand, is overexpressed in several neuronal cancers. A DLL3-targeting ADC showed objective activity only in neuroblastoma models with high DLL3 expression. These data provide vigilance about clinical development of DLL3 immunotherapies for neuroblastoma. | Evaluation of the DLL3-targeting Antibody–Drug Conjugate Rovalpituzumab Tesirine in Preclinical Models of Neuroblastoma
Neuroblastomas have neuroendocrine features and often show similar gene expression patterns to small cell lung cancer including high expression of delta-like ligand 3 (DLL3). Here we determine the efficacy of rovalpituzumab tesirine (Rova-T), an antibody–drug conjugated (ADC) with a pyrrolobenzodiazepine dimer toxin targeting DLL3, in preclinical models of human neuroblastoma. We evaluated DLL3 expression in RNA-sequencing datasets and performed IHC on neuroblastoma patient-derived xenograft (PDX), human neuroblastoma primary tumor and normal childhood tissue microarrays. We then evaluated the activity of Rova-T against 11 neuroblastoma PDX models using varying doses and schedules and compared antitumor activity with expression levels. DLL3 mRNA was differentially overexpressed in neuroblastoma at comparable levels to small cell lung cancer, as well as Wilms and rhabdoid tumors. DLL3 protein was robustly expressed across the neuroblastoma PDX array, but membranous staining was variable. The human neuroblastoma array, however, showed staining in only 44% of cases, whereas no significant staining was observed in the normal childhood tissue array. Rova-T showed a clear dose–response effect across the 11 models tested, with a single dose inducing a complete or partial response in 3 of 11 and stable disease in another 3 of 11 models. No overt signs of toxicity were observed, and there was no treatment-related mortality. Strong membranous staining was necessary, but not sufficient, for antitumor activity. Rova-T has activity in a subset of neuroblastoma preclinical models, but heterogeneous expression in these models and the near absence of expression seen in human tumors suggests that any DLL3-targeting clinical trial should be only performed with a robust companion diagnostic to evaluate DLL3 expression for patient selection.
GD2-directed antibody therapy is standard of care for high-risk neuroblastoma; therapy is toxic, and relapses often occur. DLL3, an inhibitory Notch ligand, is overexpressed in several neuronal cancers. A DLL3-targeting ADC showed objective activity only in neuroblastoma models with high DLL3 expression. These data provide vigilance about clinical development of DLL3 immunotherapies for neuroblastoma.
Neuroblastoma arises from neural crest cells of the developing sympathetic nervous system and accounts for 12% of all childhood cancer mortality (1). This disease remains a significant challenge largely due to its underlying biological heterogeneity, and outcomes for patients with the high-risk form of the disease remain poor despite intensive upfront chemoradioimmunotherapy, with over 50% of patients ultimately dying and survivors burdened with significant treatment-related morbidities (2). Neuroblastoma is the only pediatric solid tumor with an FDA-approved immunotherapy, and three separate mAbs that target GD2 are commercially available. While a randomized phase III study showed a 10% improvement in relapse-free survival, the therapy is toxic, and relapses occur on or after therapy (3, 4). New immunotherapeutic strategies are clearly needed. Neuroblastomas have neuroendocrine features and often show similar expression patterns to small cell lung cancer (SCLC) as well as other neuroendocrine cancers. Rovalpituzumab tesirine (Rova-T) is an antibody–drug conjugate (ADC)-targeting delta-like protein 3 (DLL3), a Notch ligand that inhibits Notch signaling (5). DLL3 is highly expressed in SCLC and large-cell neuroendocrine carcinoma (LCNEC) models and minimally expressed in normal tissues (6). Rova-T is an ADC comprised of a humanized anti-DLL3 mAb conjugated to a DNA-damaging pyrrolobenzodiazepine (PBD) dimer toxin that has been shown to induce sustained tumor regression across neural crest–derived malignancies including SCLC and LCNEC patient-derived xenograft (PDX) models (5). Noting in our RNA-sequencing data that the median DLL3 expression level in neuroblastomas was higher than in SCLC, here we sought to develop DLL3 as an immunotherapeutic target for high-risk neuroblastoma.
RNA-sequencing data from patients with high-risk neuroblastoma (n = 126), osteosarcoma (n = 88), rhabdoid tumor (n = 68), and Wilms tumor (n = 136) were retrieved from the Therapeutically Applicable Research to Generate Effective Treatments project. RNA-sequencing data for 79 patients with small cell lung cancer were retrieved from the GSE60052 dataset in the NCBI Gene Expression Omnibus database. Lung adenocarcinoma (n = 230), lung squamous cell carcinoma (n = 501), and mesothelioma (n = 87) RNA-sequencing datasets were retrieved from The Cancer Genome Atlas. Normal tissue RNA-sequencing data were obtained by the Genotype-Tissue Expression Project (GTEx). All RNA-sequencing datasets were aligned using STAR and gene-level expression was quantified with RSEM normalization using hg37 as reference genome and Gencode v23 gene annotation. RNA sequencing for the PDX models was performed and analyzed as described previously (7). Fragments Per Kilobase of transcript per Million mapped reads (FPKM) data were available from 21 of 35 neuroblastoma PDX models and were correlated with IHC quantification as described below.
All tissue microarrays (TMA) were constructed using standard methods (8, 9). Each tumor sample was punched in duplicate using 0.6-mm cores with a Galileo CK3500 Tissue Microarrayer (Integrated System Engineering SRL). For the neuroblastoma PDX array, PDX bearing mice were euthanized when the tumor reached approximately 1 cm3 in volume, the tumor was harvested and immediately placed into a 50 mL conical tube with enough 10% neutral buffered formalin (NBF, Thermo Fisher Scientific) to cover the entire tumor (minimum 10:1 formalin to tissue ratio). The tissue was fixed in 10% NBF for 24–48 hours depending on the size of tumor. Fixed tumor tissue was processed in the Excelsior ES tissue processor from Thermo Fisher Scientific using the standard overnight protocol. Formalin-fixed, paraffin-embedded tissue blocks were sectioned, and hematoxylin and eosin (H&E) staining was performed to obtain a template guide slide to match the face of each tissue block. H&E slides were reviewed to validate the sample quality and preservation. PDX samples were excluded if they showed greater than 20% necrosis or inadequate tissue preservation. Slides were annotated to create a template for punching, then used as guides for core selection as described previously (10). The neuroblastoma PDX TMA was constructed of 35 distinct PDX tumor models that were collected in duplicate for a total of 70 tumor samples. Two 0.6-mm cores were punched per each tumor, resulting in a total of 140 tissue sample cores along with nine control tissues (placenta, human brain cortex, human brain cerebellum, human tonsil, human adrenal cortex, human adrenal medulla, murine brain cortex, murine brain cerebellum, and murine adrenals). Human placenta was used as an orientation marker on the array to mark the starting point (upper left-hand corner), and a row down the middle to aid in manual reading of the array. For the human neuroblastoma tumor and normal pediatric tissue arrays, all samples were collected and deidentified at Children's Hospital of Philadelphia from 2005 to 2016 under Institutional Review Board approval. The normal childhood TMA represents tissues from patients ranging in age between 0 and 253 months (Supplementary Fig. S1). The human neuroblastoma array included 64 tumors from patients between the ages of 0 months to 12 years of age. Eight of these samples were from cases with matched primary and metastatic tumors.
IHC with rabbit anti-DLL3 SP347 (Ventana, 790-7016) and anti-CD56 (Cell Marque, 156-R-95) antibodies was performed on formalin-fixed paraffin-embedded TMA slides. Staining was performed on a Bond Max automated staining system (Leica Biosystems). The Bond Refine polymer staining kit (Leica Biosystems) was used. The standard IHC protocol F was used apart from the mouse polymer step, which was excluded. For CD56 staining, the primary antibody incubation time was extended to 1 hour. DLL3 antibody was prediluted and antigen retrieval was performed with E1 (Leica Biosystems) retrieval solution for 20 minutes. CD56 was used at a 1:200 dilution with E2 (Leica Biosystems, AR9640) retrieval solution for 20 minutes. Slides were rinsed, dehydrated through a series of ascending concentrations of ethanol and xylene, and then covered with coverslips. Stained slides were then digitally scanned at 20× magnification on an Aperio CS-O slide scanner (Leica Biosystems). TMAs were scored for the most prominent intensity (0–3 with 1 representing equivocal, 2 weak, and 3 strong positive staining) as well as for percentage of staining. Both the overall staining (any pattern) and membranous specific staining patterns were recorded. A modified H score was calculated as intensity multiplied by percentage of positively stained cells. All scores for each of the two tumors per PDX model were averaged for the final score.
All xenograft studies were conducted in compliance with protocols approved by the Institutional Animal Care and Use Committee approved by The Children's Hospital of Philadelphia. Felix-PDX, COG-N-452x, COG-N-519x, COG-N-415x, NB-FLY-623m, KWK-6062x, COG-N-421x, and COG-N-424x PDXs, and SH-SY5Y and SK-N-AS cell line xenografts (CDX) were implanted subcutaneously into the right flanks of female CB17 SCID mice (CB17/Icr-Prkdcscid/IcrIcoCrl, Charles River Laboratories, strain code 236). When tumors reached 200–300 mm3, the animals were randomized into groups of 2–10 mice per arm. Felix-PDX, COG-N-452x, COG-N-519x, or COG-N-415x tumor-bearing mice n = 8–10 were enrolled in randomized controlled preclinical trials and dosed via intraperitoneal injection with one single dose according to the following treatments (vehicle, Rova-T at 0.1, 0.3, and 0.6 mg/kg, or IgG1-ADC at 0.6 mg/kg). In a separate study, the COG-N-415x model was treated with 1 or 3 weekly injections of 1 mg/kg of Rova-T, IgG1-ADC, or vehicle. NB-FLY-623m, KWK-6062x, COG-N-421x, COG-N-424x, SH-SY5Y, and SK-N-AS tumor-bearing mice were part of n = 2 animal trials and dosed intraperitoneally with one single dose of vehicle or Rova-T at 0.1 and/or 0.6 mg/kg. Tumor volume was estimated using the spheroid formula = (π/6) × (a + b/2)3, where “a” and “b” are two diameters measured with an electronic caliper. Total body weight and tumor volume were recorded two to three times weekly. Events were defined as quadrupling of a mouse's tumor volume from day zero.
The exact time to event is estimated by interpolating between the measurements directly preceding and following the event, assuming log-linear growth. Differences in event-free survival (EFS) between experimental groups (e.g., treated vs. controls) are tested using the Peto and Peto modification of the Gehan–Wilcoxon test (α = 0.05, two-sided alternative) and plotted as Kaplan–Meier EFS curves. The objective response categories are progressive disease (PD, which is subdivided among treated mice into PD without and with growth delay, PD1 and PD2, respectively), stable disease (SD), partial response (PR), complete response (CR; no measurable tumor mass for one recording), and maintained complete response (no measurable tumor mass for at least three consecutive weekly recordings). Response rate is defined as the percentage of mice with PR or better. Mice experiencing a possible treatment-related death (i.e., drug toxicity), mice with failed engraftment, and mice which unexpectedly die for reasons unrelated to treatment are excluded from statistical analyses of time-to-event, minimum tumor volume, and objective response. For experiments of groups of n = 2 mice per group, we did not perform any statistical tests of group differences but have reported all other summary statistics. All studies performed had written informed consent from the patients when relevant, studies were conducted in accordance with recognized ethical guidelines (including the Declaration of Helsinki, CIOMS, Belmont Report, U.S. Common Rule), and the studies were approved by an Institutional Review Board.
Data were generated by the authors and included in the article.
We used harmonized RNA-sequencing data to compare DLL3 gene expression of high-risk neuroblastoma primary tumors (n = 126) to other pediatric and adult tumors (n = 1,189) as well as 7859 GTEx samples across 31 unique normal tissues. In comparison with normal tissues, except for brain, pituitary gland, and testes, DLL3 showed significantly higher expression in neuroblastoma and some other pediatric solid tumors, such as malignant rhabdoid tumor and Wilms tumor (Fig. 1). Notably, the median DLL3 expression level in neuroblastomas was higher than in SCLC (n = 79), albeit with a broader range of expression.
We next optimized an IHC assay using a commercially available anti-DLL3 mAb (Ventana, SP347 clone) to evaluate DLL3 expression in a panel of neuroblastoma PDX models (Fig. 2A–C). This array contains 140 cores from 35 models with two tumors from each model and two cores from each tumor. The majority of the cores showed completely undifferentiated neuroblastic histology, with 49 of 140 cores (35%) showing some neuropil (generally minimal), and two of these cores (from the same tumor) showed rare ganglion cell differentiation. There was a wide range of DLL3 staining across the 35 models ranging from intense membranous staining to no staining at all (Fig. 2C). In some tumors, there was clear membranous staining, often with cytoplasmic staining as well, others showed cytoplasmic staining only, and in others there was paranuclear accumulation of the signal (Supplementary Fig. S1). Mixtures of patterns were also identified. In most cases with membranous staining or paranuclear staining, a subset of cells (sometimes only a small minority) showed clear membranous (Supplementary Fig. S1A), or paranuclear accentuation (Supplementary Fig. S1C), while a larger number of cells had diffuse (usually weaker) cytoplasmic staining (Supplementary Fig. S1B), resulting in a lower membranous specific score. Some degree of 2+ or 3+ (i.e., greater than equivocal) staining was seen in 116 of 140 cores (83%) or 30 of 35 (86%) of PDX models. Staining was overall concordant between the two cores, with occasionally one degree of difference in intensity between cores. RNA-sequencing data from 21 of the 35 PDX models included on neuroblastoma PDX TMA showed weak correlation with the overall staining IHC H scores (Fig. 2D and Fig. 3). A human neuroblastoma tumor array with 33 specimens from initial diagnosis, 25 from postchemotherapy local resections, and five from relapsed and/or progressive disease (and one from unknown timing) was also stained. A total of 125 cores were evaluable for DLL3 staining from 64 unique cases (Supplementary Fig. S2). In general, there was much weaker DLL3 staining on this array compared with the PDX array and the robustly expressed cell surface protein NCAM1 (CD56; Supplementary Fig. S3). Fifty cores (40%) from 28 cases (44%) contained at least some cells with staining, but with a very wide range from 1% to 100% of cells (Supplementary Fig. S2 and S3). Only two cases showed an H-score > 100. The normal human childhood TMA (Supplementary Fig. S4) showed no clear membranous DLL3 staining in any normal tissue (Supplementary Fig. S2). There was weak cytoplasmic staining of placental syncytiotrophoblasts as well as the epithelium in the appendix, ileocecum, fallopian tube, bladder, and gallbladder with supranuclear granular accentuation in the gallbladder. Weak cytoplasmic staining was also seen in a subset of bone marrow mononuclear cells, and in a subset of adrenal cortical cells with a somewhat granular pattern. Moderate granular cytoplasmic staining was present in a subset of neurons and in a subset of gastric glandular epithelial cells. Strong paranuclear cytoplasmic staining was present in a subset of thyroid follicular cells.
We first tested the efficacy of Rova-T in four neuroblastoma PDX models using a conventional design of 8–10 mice per arm. In the PDX array, all four of these models had robust DLL3 expression in the majority of tumor cells including some degree of clearly membranous staining (Supplementary Fig. S5). There was a clear dose–response effect, and all models showed some or significant evidence of antitumor activity at a single 0.6 mg/kg dose (i.e., one injection) of Rova-T (Table 1; Fig. 4; Supplementary Fig. S6). Doses of 0.3 and 0.6 mg/kg induced statistically significant increases in EFS, except for the COG-N-415x dosed at 0.3 mg/kg (Table 1; Fig. 4). We next compared a single 1 mg/kg dose to 1 mg/kg weekly ×3 in mice bearing the COG-N-415x PDX, showing a maintained complete response for 7 weeks with the weekly ×3 treatment schedule, with slow resumption of tumor growth thereafter (Fig. 5; Table 1). Next, we sought to further understand biomarkers of antitumor activity by studying six additional PDX and CDX models with varying levels of DLL3 expression using an n = 2 design (10), with one single dose of vehicle or Rova-T at 0.1 or 0.6 mg/kg. No objective responses were observed and there was SD in three models at 0.6 mg/kg (Table 1). There were no overt signs of toxicity and no treatment-related mortality (Supplementary Fig. S7).
Improved clinical outcomes for patients with high-risk pediatric cancers, especially those with solid tumors, have been hampered by aggressive and nonspecific multimodal therapy such as empiric high-dose chemotherapy. There is an urgent need to identify new drug targets that can be harnessed for the development of more rationale therapies. The Notch pathway is a highly conserved cell-cell signaling pathway involved in a variety of cellular processes and controls fate decisions in developing organs (11–13). Here, we show that the Notch ligand DLL3 is robustly expressed in most high-risk neuroblastoma preclinical models, but much less so in the patient tumor samples. We hypothesize that the predominant adrenergic cell type of neuroblastomas must maintain suppression of Notch signaling, and that PDXs are enriched for adrenergic cell type and thus DLL3 overexpression (14–16). The low protein expression noted in our human tumor array must be viewed in the context that many of the samples were derived from definitive surgery after chemotherapy, and thus enriched for a more mesenchymal cell type (17). Thus, any DLL3-targeting clinical trial should strongly consider real-time assessment of DLL3 protein expression at study entry and explore a minimum expression level required for antitumor efficacy. Preclinical studies support the development of novel therapies that target DLL3 in SCLC and possibly other neuroendocrine tumors, and have demonstrated that targeting of DLL3 using an ADC approach inhibits tumor growth (5, 18–20). The efficacy seen with Rova-T against a panel of neuroblastoma PDXs suggests that DLL3 is a potential candidate therapeutic target in this disease. Rova-T was evaluated in a pivotal phase II study in patients with DLL3-expressing SCLC (defined by IHC). There was no difference in overall survival between patients high or low DLL3 expression, and results demonstrated modest clinical activity with associated toxicities (21). All patients received 0.3 mg/kg Rova-T intravenously infused over 30 minutes once every 6 weeks for two cycles as the initial treatment. Ultimately, a phase III placebo-controlled trial evaluating Rova-T as a first-line maintenance treatment therapy for advanced SCLC demonstrated no survival benefit and Research and Development of this drug was hence discontinued. A recent trial combining Rova-T with nivolumab plus or minus ipilimumab in patients with relapsed/refractory SCLC showed an objective response rate of 30% but unfortunately the combination was not well tolerated (22). Despite the cessation of trials with Rova-T as an ADC approach using a PBD dimer to target DLL3, this remains a conceivable target in neuroblastoma and other neuroendocrine tumors due to its relatively restricted normal expression and moderate expression in preclinical PDX models. However, in light of the marginal expression on the cell surface of human neuroblastoma tumors, this target should be contemplated with caution for immunotherapeutic approaches that exploit not only antigen specificity but also and receptor density. DLL3-targeted Bispecific T-cell Engager (BiTE) molecules and chimeric antigen receptors (CAR) T cells are in development to promote the tumor-suppressive functions of DLL3 and to avoid oncogenicity. These include Amgen's 757 Anti-DLL3 x CD3 BiTE antibody (23) and AMG 119 CAR, as well as Boehringer Ingelheim's bispecific DLL3/CD3 IgG-like T-cell engaging antibody (24). In addition, we should not rule out the possibility of other DLL3-ADC approaches using different payloads. Clinical trials suggest that a better understanding of the mechanistic basis of ADC activity as well as exploration and modification of dose and schedule of DLL3-targeted ADC are necessary to reduce toxicity and improve efficacy. Along these lines, recent studies indicate that in addition to intracellular release of the cytotoxic payload, ADCs can concomitantly suppress infiltrating lymphocytes that overall limits their efficacy (25). Collectively, these data provide a foundation for vigilance about future consideration of clinical development of DLL3-targeted immunotherapeutic agents for neuroblastoma and other potentially other DLL3-expressing neuroendocrine tumors. Any neuroblastoma clinical trial must consider the heterogeneity and plasticity of DLL3 expression on Notch pathway activity in this disease.
Click here for additional data file. |
|
PMC9648417 | 36382088 | Mariafausta Fischietti,Frank Eckerdt,Ricardo E. Perez,Jamie N. Guillen Magaña,Candice Mazewski,Sang Ho,Christopher Gonzalez,Lukas D. Streich,Elspeth M. Beauchamp,Amy B. Heimberger,Aneta H. Baran,Feng Yue,C. David James,Leonidas C. Platanias | SLFN11 Negatively Regulates Noncanonical NFκB Signaling to Promote Glioblastoma Progression | 13-09-2022 | Glioblastoma (GBM) is an aggressive and incurable brain tumor in nearly all instances, whose disease progression is driven in part by the glioma stem cell (GSC) subpopulation. Here, we explored the effects of Schlafen family member 11 (SLFN11) in the molecular, cellular, and tumor biology of GBM. CRISPR/Cas9-mediated knockout of SLFN11 inhibited GBM cell proliferation and neurosphere growth and was associated with reduced expression of progenitor/stem cell marker genes, such as NES, SOX2, and CD44. Loss of SLFN11 stimulated expression of NFκB target genes, consistent with a negative regulatory role for SLFN11 on the NFκB pathway. Furthermore, our studies identify p21 as a direct transcriptional target of NFκB2 in GBM whose expression was stimulated by loss of SLFN11. Genetic disruption of SLFN11 blocked GBM growth and significantly extended survival in an orthotopic patient-derived xenograft model. Together, our results identify SLFN11 as a novel component of signaling pathways that contribute to GBM and GSC with implications for future diagnostic and therapeutic strategies. Significance: We identify a negative regulatory role for SLFN11 in noncanonical NFκB signaling that results in suppression of the cell-cycle inhibitor p21. We provide evidence that SLFN11 contributes to regulation of stem cell markers in GBM, promoting the malignant phenotype. In addition, SLFN11 targeting triggers p21 expression and antitumor responses. Our studies define a highly novel function for SLFN11 and identify it as a potential therapeutic target for GBM. | SLFN11 Negatively Regulates Noncanonical NFκB Signaling to Promote Glioblastoma Progression
Glioblastoma (GBM) is an aggressive and incurable brain tumor in nearly all instances, whose disease progression is driven in part by the glioma stem cell (GSC) subpopulation. Here, we explored the effects of Schlafen family member 11 (SLFN11) in the molecular, cellular, and tumor biology of GBM. CRISPR/Cas9-mediated knockout of SLFN11 inhibited GBM cell proliferation and neurosphere growth and was associated with reduced expression of progenitor/stem cell marker genes, such as NES, SOX2, and CD44. Loss of SLFN11 stimulated expression of NFκB target genes, consistent with a negative regulatory role for SLFN11 on the NFκB pathway. Furthermore, our studies identify p21 as a direct transcriptional target of NFκB2 in GBM whose expression was stimulated by loss of SLFN11. Genetic disruption of SLFN11 blocked GBM growth and significantly extended survival in an orthotopic patient-derived xenograft model. Together, our results identify SLFN11 as a novel component of signaling pathways that contribute to GBM and GSC with implications for future diagnostic and therapeutic strategies.
We identify a negative regulatory role for SLFN11 in noncanonical NFκB signaling that results in suppression of the cell-cycle inhibitor p21. We provide evidence that SLFN11 contributes to regulation of stem cell markers in GBM, promoting the malignant phenotype. In addition, SLFN11 targeting triggers p21 expression and antitumor responses. Our studies define a highly novel function for SLFN11 and identify it as a potential therapeutic target for GBM.
Glioblastoma (GBM), classified as World Health Organization grade 4 glioma (1), is the most frequent primary tumor in the brain, with a 5-year survival estimate of only 7% (2). Maximal safe surgical resection, followed by chemoradiation and adjuvant temozolomide (TMZ) treatment has been the standard of care in treating GBM for the past 17 years (3). However, this treatment approach results in modest survival benefit, with 14.6 months being the median value (3). Tumor cell dissemination in normal brain tissue and extensive subpopulation heterogeneity are hallmarks of GBM (4, 5), and are key factors in the relative lack of success in treating this cancer. Isocitrate dehydrogenase (IDH) mutational status is an important prognostic factor for patients with GBM, with IDH wild-type (WT) patients exhibiting a lesser median survival compared with patients with IDH mutant tumor (6). As such, GBM is now genetically defined as IDH WT versus mutant (1). Transcriptomic analysis has also been used in the subclassification of GBM, with mesenchymal, proneural, and classical tumors being widely accepted as transcriptionally defined subtypes (7, 8). Thus far, there has been no success in identifying therapeutics that are effective in controlling genetically nor transcriptionally defined GBM subtypes. As such, there is a compelling need for increased understanding of GBM molecular biology which may reveal therapeutically actionable targets for treating GBM. Previously, we described Schlafen 5 (SLFN5) as a potential biomarker and therapeutic target in GBM by demonstrating that elevated SLFN5 expression promotes GBM malignant phenotypes (9). This implicated, for the first time, the family of Schlafen (SLFN) genes in the pathogenesis of GBM. SLFN genes were initially identified for their ability to induce a reversible G0–G1 cell-cycle arrest in thymocytes (10). SLFNs are found in vertebrates from frogs to mammals with variable homology across species (11). Previous studies have established SLFNs as IFN-responsive genes, with implications in cell differentiation, proliferation, immune cell regulation (10), and as inhibitors of viral replication (12). Recent studies have explored roles for human SLFN family members in cancer biology (reviewed in ref. 13) and found that the contributions of SLFNs in the regulation of oncogenic processes are complex. On one hand, SLFN5 overexpression suppresses breast tumor growth in mice and elevated SLFN5 expression correlates with better survival in breast cancer (14) and renal cell carcinoma (15). In addition, SLFN5 knockdown increases transformation and invasion in malignant melanoma (16). On the other hand, SLFN5 is highly expressed and contributes to tumor progression in pancreatic ductal adenocarcinoma (17), castration-resistant prostate cancer (18), and gastric carcinoma (19). This indicates that SLFNs can have diverse and, sometimes, opposing functions in cancer, possibly in a tissue-specific manner. A similarly complicated role can be assumed for SLFN11, which was found to exhibit a broad range of expression in a The Cancer Genome Atlas (TCGA) pan-cancer dataset (20), and the NCI-60 cancer cell line panel (21, 22). Also, SLFN11 is highly expressed in some cancers, such as Ewing sarcoma, pediatric sarcomas, mesothelioma, and renal cell carcinoma, while its expression is low in other types of cancer such as tumors of the ovary and pancreas (20, 22–24). In this work, we provide evidence for SLFN11 contributing to GBM cell proliferation, neurosphere growth, and expression of progenitor/stem cell markers. We demonstrate that SLFN11 associates with components of the NFκB family of inducible transcription factors that are involved with the regulation of numerous cellular processes (25). Activation of the noncanonical NFκB pathway is stimulated by regulated processing of p100 into the DNA-binding transcription factor p52, which either homodimerizes (p52:p52) or heterodimerizes with RelB (p52:RelB) to modulate transcription of a plethora of target genes (26). Using immunoprecipitation (IP) mass spectrometry analysis, we found association of SLFN11 with NFκB2 in GBM. Genetic disruption of SLFN11 stimulated expression of NFκB target genes, including CDKN1A (p21) and significantly delayed tumor growth and improved survival in a GBM orthotopic patient-derived xenograft (PDX) mouse model.
All cell lines were incubated in a 37°C humidified incubator with 5% CO2 and were grown in DMEM supplemented with 10% FBS. U87 cells were kind gift from Dr. Alexander Stegh, LN229 cells from Dr. Chi-Yuan Cheng, and GBM6 PDX from Dr. C. David James (all Northwestern University, Chicago, IL). All cell lines were tested for Mycoplasma every 4 months and every 6 months cell lines were tested by short tandem repeat analysis and authenticated using published reference databases.
GBM cell lines and GBM6 PDX cells were grown under cancer stem cell (CSC) culture conditions in three-dimensional (3D) and plated for neurosphere assays as described previously (27). GBM6 PDX line, stably expressing Luciferase was described previously (28). Neurosphere assay was performed as in ref. 29, and neurosphere cross-sectional area was determined as described before (30).
All animal studies were carried out under an approved protocol of the Institutional Animal Care and Use Committee (IACUC) of Northwestern University (Chicago, IL). Luciferase-expressing GBM6 cells were suspended in RPMI at a concentration of 1.5 × 105 cells per μL. Anesthetized female homozygous NCr nude mice (5–6 weeks; Taconic) were placed on a heating pad, the surgical area was cleaned with 70% ethanol and betadine solution. A para-sagittal skin incision was made (∼10 mm) over the middle frontal to parietal bone. The exposed skull surface was treated with 3% hydrogen peroxide solution and a 25-gauge needle was used to create a burr hole 2 mm lateral right of the bregma and 1 mm posterior to the coronal suture. GBM6 WT and SLFN11 knockout (KO) cells (2 μL cell suspension) were implanted through a Hamilton syringe over a period of 1 minute. After 1 additional minute, the syringe was slowly withdrawn, and the wound was closed with staples. Mice received postsurgical care according to IACUC guidelines and were imaged weekly by bioluminescence imaging using a Lago/Lago X—Spectral Instruments Imaging system as described previously (17).
LN229-TetON-SLFN11-Myc-Flag stable cell lines were generated as described previously (17).
U87 were transfected using Turbofect (Thermo Fisher Scientific) with Cas9, single-guide RNA (sgRNA) targeting the SLFN11 gene, and homology direct repair plasmids. Cells were kept under Puromycin selection for 2 weeks and then sorted for high expression of red fluorescent protein using FACS. LN229 and GBM6 cells were transduced with pLVX-hEF1α-Cas9-Blast and pLVX-CMV-SLFN11 sgRNAs-PURO using TransDux MAX Lentivirus Transduction enhancer (System Biosciences) according to the manufacturer's instructions. Cells were kept under Puromycin selection. LN229-Cas9 cells were used as control. LN229 SLFN11 KO cells were seeded as single cells in 96-well plates by FACS for generation of single clones. The clone showing the most efficient KO was used for experiments in this study.
pLenti-CMV-Hygro-SLFN11-Flag plasmid was generated as described previously (17). U87 and LN229 cells reexpressing SLFN11 (SLFN11 KO+SLFN11) were generated using lentiviral transduction as described above.
Mouse brains with GBM6 SLFN11 WT and KO tumors were collected and processed for hematoxylin and eosin (H&E) staining and IHC staining by the Human Pathology Core of Northwestern University as described previously (30). Slides were analyzed and scored for p21 expression by a board-certified pathologist Dr. Lukas D. Streich. Slides were scanned using Hamamatsu NanoZoomer Digital slide scanner and images were exported using NDP.view2 Viewing software.
SLFN11 WT and KO U87 and LN229, and KO+SLFN11 U87 and LN229 cells, and SLFN11 KO siCTRL/siCDKN1A U87 and LN229 cells were plated in 6-well plates in duplicate (50,000 cells/well). Cells were dissociated by Trypsin and counted after 2, 4, and 7 days using TC20 Automated Cell Counter (Bio-Rad).
Cell lysis and immunoblotting were performed as described previously (31). Chemiluminescence was detected using a ChemiDoc MP imager or autoradiography film. Films were digitally scanned with Adobe Photoshop using a Canon CanoScan 8800F scanner.
SLFN11 WT and KO U87, LN229, and SLFN11 KO+SLFN11 U87 and LN229, and GBM6 cells were lysed in RLT buffer (Qiagen). Tumors from mice bearing SLFN11 WT and KO GBM6 were homogenized in RLT buffer using TissueRuptor (Qiagen). Total RNA was isolated using RNeasy minikit (Qiagen) and retrotranscribed using High Capacity cDNA reverse transcription kit (Applied Biosystems). TaqMan qRT-PCR was performed using SsoAdvanced Universal Probes Supermix (Bio-Rad) according to the manufacturer's instructions, using Taqman probes (see Supplementary Table S1).
Cells were lysed with NP-40 lysis buffer (40 mmol/L HEPES pH 7, 120 mmol/L NaCl, 1 mmol/L EDTA, 10 mmol/L Na Pyrophosphate, 50 mmol/L NaF, 10 mmol/L β-glycerophosphate and 0.1% NP-40). FLAG-M2–conjugated sepharose beads (Sigma-Aldrich) and Myc-Tag–conjugated sepharose beads (Cell Signaling Technology) were used for IP as described previously (17).
LN229-TetON-SLFN11-Myc-Flag cells were plated in 150 mm dishes. The following day, cells were either left untreated or treated with doxycycline (DOX) for 48 hours. Subsequently, cells were either left untreated or irradiated with 8 Gy for 30 minutes. Cell lysates were prepared in NP-40 lysis buffer and then IP was performed using Myc-Tag–conjugated sepharose beads (Cell Signaling TEchnology). No DOX-treated samples were used as negative controls. Samples were then processed and analyzed as described previously (17).
Protein lists identified in LC/MS-MS were converted to gene lists and were submitted to the Metascape database, a gene annotation and analysis resource (http://metascape.org/), for pathway and process enrichment analysis as described previously (17).
TCGA_GBM gene expression data using the RNA sequencing (RNA-seq) platform or the Agilent (Agilent-4502A) array were analyzed using the GlioVis portal (http://gliovis.bioinfo.cnio.es/; ref. 32).
SLFN11 WT and KO U87 and LN229 cells were plated on coverslips in 12-well plates (25,000 cells/well). After 5 days, cells were washed with PBS and fixed with 4% paraformaldehyde (PFA) for 30 minutes. For permeabilization, cells were incubated with PBS+0.1% Triton for 20 minutes at room temperature. Subsequently, cells were washed and incubated with blocking buffer (2% BSA+0.1% Triton in PBS) for 50 minutes at room temperature. Cells were then incubated with anti-p21 primary antibody (Cell Signaling Technology) overnight at 4°C. The next day, cells were washed and sequentially stained with AlexaFluor546-phalloidin and 4ʹ,6-Diamidine-2ʹ-phenylindole dihydrochloride (DAPI). After five washes, coverslips were mounted on microscope slides using ProLong Gold Antifade Mountant (Thermo Fisher Scientific). Images were acquired using a Leica DMi8 inverted microscope with objective lens 20× air Plan Fluotar, NA 0.40. Cells positive for p21 were counted manually using Fiji-ImageJ software.
Microscopy was performed using a Nikon A1plus inverted microscope. Objective lens was from Nikon: 20× air Plan Apo objective, NA 0.75. Fluorochromes were from Invitrogen and included AF488 (green) and AF546 (red). DAPI (blue) was from Roche. For microscopic analysis, the acquisition software NIS Elements (Nikon) was used.
Control and NFKB2-targeting siRNAs were from Dharmacon, Control and CDKN1A were from Santa Cruz Biotechnology and used with Lipofectamine RNAiMAX reagent and Opti-MEM medium (Thermo Fisher Scientific), as described previously (30).
Cells were grown as 3D Neurospheres. Chromatin immunoprecipitation (ChIP) was performed using the SimpleChIP Enzymatic Chromatin IP Kit with Magnetic Beads from Cell Signaling Technology, as per the manufacturer's instructions. Antibodies for NFκB2 p100/p52 and RelB were purchased from Cell Signaling Technology. Normal rabbit IgG was used as a negative control. qRT-PCR was performed on purified immunoprecipitated DNA for the CDKN1A promoter (the RPL30 promoter served as a negative control) using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad) according to the manufacturer's instructions. All qRT-PCR signals were normalized to the input DNA.
To determine potential NFκB binding sites on the CDKN1A promoter, 3,000 bp upstream and 100 bp downstream from the transcription start site were analyzed using the JASPAR database (33) at a relative profile score threshold of 80% for known human NFκB binding motifs.
3D tumor cell invasion was determined using the Cultrex 3D Spheroid Cell Invasion Assay (Trevigen) as described previously (9).
All statistical analyses were performed using GraphPad Prism 8.0 and P values <0.05 were considered statistically significant.
The data generated in this study are available upon request from the corresponding author. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD033913.
Using the Sun and Rembrandt datasets that were available through the Oncomine database, we previously found elevated SLFN11 expression in GBM and this was associated with worse prognosis (9). To corroborate and extend these findings, we now interrogated TCGA dataset (TCGA_GBM) using the GlioVis portal (http://gliovis.bioinfo.cnio.es/; ref. 32). Results from microarray (Agilent-4502A) as well as RNA-seq revealed that SLFN11 expression is significantly elevated in GBM, as compared with normal brain tissue (Fig. 1A). Among transcriptionally defined subtypes of GBM, SLFN11 expression is higher in the mesenchymal subtype relative to classical and proneural subtypes (Fig. 1B). With regard to GBM genetic subclassification, SLFN11 is higher in IDH WT than IDH mutant tumors (Fig. 1C). There is no significant gender-associated difference in SLFN11 expression (Fig. 1D), but there is clear indication of increasing SLFN11 expression being associated with decreasing GBM patient survival (Fig. 1E).
To investigate the biological effects of SLFN11 loss in GBM cells, CRISPR/Cas9-mediated gene KO was used to eliminate endogenous SLFN11 protein expression in U87 (Fig. 2A, left) and LN229 (Fig. 2A, right) GBM cell lines. The human SLFN11 gene clusters together with SLFN5, SLFN12, SLFN12L, SLFN13, and SLFN14 on chromosome 17 (11). Thus, modulating expression of one SLFN family member may alter expression of other SLFN family members in certain cellular contexts (17). While deletion of SLFN11 resulted in some alterations in expression of SLFN13 and SLFN14, these alterations were not consistent throughout cell lines, indicating the effects seen after SLFN11 KO are specific (Supplementary Fig. S1). Loss of SLFN11 significantly inhibited proliferation of KO cells in vitro (Fig. 2B) and reduced neurosphere formation as well as invasiveness in 3D cultures (Fig. 2C; Supplementary Fig. S2). We expanded our analysis to PDX cells which are known for improved preservation of patient tumor characteristics, relative to that of highly passaged GBM cell lines (34). CRISPR/Cas9-mediated KO of SLFN11 in GBM6 PDX cells (Fig. 2D, left), significantly inhibited the ability of these cells to form neurospheres (Fig. 2D, right). To validate whether these growth inhibitory effects are indeed due to loss of SLFN11, we reexpressed SLFN11 in U87 and LN229 SLFN11 KO cells (Fig. 2E) and found that this reverted the antiproliferative effects to a level mirroring WT cells (Fig. 2F). Similar results were obtained when SLFN11 expression was rescued under 3D spheroid conditions (Fig. 2G). These results support the notion that elevated SLFN11 expression promotes GBM cell proliferation and invasion and confirm these effects are specifically mediated by SLFN11.
Glioma stem cells (GSC) reside at the apex of GBM cellular hierarchy and contribute to long-term GBM progression, malignancy, and therapy resistance (35, 36). We investigated whether SLFN11 may modulate expression of genes associated with stem/progenitor markers. Expression data from cells grown in 3D as neurospheres under stem cell–permissive conditions revealed that KO of SLFN11 in LN229 and U87 significantly reduced neural stem/progenitor cell marker expression including VIM (vimentin), SOX2, NES (nestin), CDH2 (N-cadherin), CD44, and CTNNB1 (β-catenin; Fig. 3A and B). In addition, we employed the PDX line GBM6 that has been shown to reflect GBM cellular heterogeneity including GSCs (8). Similar results were obtained for GBM6 PDX cells (Fig. 3C). Next, we employed our cell lines reexpressing SLFN11 (SLFN11 KO+SLFN11) and observed increased expression of stem/progenitor markers (Supplementary Fig. S3), suggesting a partial rescue, and indicating the effects seen after SLFN11 KO are specific.
To gain mechanistic insights into the pathways regulated by SLFN11 in GBM, lysates from LN229 cells expressing DOX-inducible Myc-tagged SLFN11 were incubated with anti-Myc antibody, and immunoprecipitates were analyzed using nano-LC/MS-MS. As SLFN11 is known to mediate responses to DNA damage (37), we also analyzed immunoprecipitates from cells treated with radiation. Myc-tagged SLFN11 was efficiently immunoprecipitated in lysates from DOX-induced LN229 cells treated with and without irradiation (Fig. 4A). Proteomic analyses identified 75 putative SLFN11 interacting proteins, 20 of which were discovered exclusively in irradiated cells and 8 exclusively in untreated cells (Fig. 4B, left; Supplementary Table S2). Ontology analysis of the 47 candidates found associated with SLFN11 regardless of treatment revealed that 15 of these genes are involved in the positive regulation of cytokine-mediated signaling pathways (Fig. 4B, top right panel in green; Supplementary Table S3). Of the 20 candidates associated with SLFN11 after cell irradiation, eight are known to be involved in IL1 family signaling (Fig. 4B, bottom panel in blue; Supplementary Table S4). Among these was the transcriptional regulator NFκB2 (Supplementary Table S4, highlighted in yellow). Changes in SLFN11 expression result in alterations of gene transcription (38). As NFκB2 represents a key transcription factor involved in various cellular responses, we sought to investigate the biological effects of this potential association in more detail. To corroborate this interaction, we used FLAG IP and found that under these conditions, SLFN11 associated with NFκB2 (NFκB2/p100) from LN229 cell lysates regardless of irradiation treatment (Fig. 4C). To investigate whether SLFN11 regulates NFκB transcriptional activity, we monitored mRNA levels of established NFκB target genes, such as CD82, CDKN1C, TRAF2, and TRAF3 (39). As NFκB is known to be part of a positive regulatory feedback loop, we also investigated expression of NFKB2 and RELB (39). We found that the transcript levels of all these NFκB target genes were significantly elevated in LN229 spheroids lacking SLFN11 (Fig. 4D). Our results suggest that SLFN11 physically associates with NFκB2/p100 in LN229 cells. Furthermore, loss of SLFN11 triggers expression of numerous NFκB target genes in untreated cells, consistent with stimulation of NFκB transcriptional activity, independently of irradiation.
Evidence indicates that NFκB can inhibit cell proliferation through induction of the cell-cycle inhibitor p21CIP1 (herein p21, encoded by CDKN1A) in certain cell types (40). qPCR and immunoblot results show that CDKN1A transcript and encoded protein are significantly elevated in LN229 and U87 SLFN11 KO spheres (Fig. 5A and B). In addition, immunofluorescence analysis of cells lacking SLFN11 revealed a significantly higher proportion of p21-positive cells (Fig. 5C). This indicates that KO of SLFN11 stimulates induction of p21 protein expression in GBM cells. Next, we sought to determine whether the induction of p21 expression after loss of SLFN11 is specifically dependent on NFκB2. Efficient knockdown of NFKB2 by siRNA in SLFN11-deficient cells (Fig. 5D, left) blocked the increase in CDKN1A expression seen after loss of SLFN11 (Fig. 5D, right). In addition, NFκB2 activity was significantly increased in SLFN11 KO LN229, U87, and GBM6 neurospheres as indicated by ELISA assay results (Fig. 5E). To investigate the mechanism responsible for p21 induction by NFκB2 in SLFN11-deficient cells, we next performed ChIP experiments. We designed primers able to hybridize in the CDKN1A promoter region that contains the NFκB consensus DNA-binding sequence. In SLFN11 KO neurospheres, we found a significant enrichment of p52 (the mature, activated NFκB2 form) occupancy on the CDKN1A promoter in LN229, U87, and GBM6 3D neurospheres (Fig. 5F). As activation of the noncanonical NFκB pathway triggers p52:RelB dimerization, we also assessed RelB in ChIP. Similar to p52, we detected significantly higher RelB occupancy at the CDKN1A promoter (Fig. 5G). We used the promoter for RPL30 (encoding 60S ribosomal protein L30) as a control and found no significant enrichment of p52 or RelB occupancy on the RPL30 promoter (Fig. 5F and G), indicating the enrichment on the CDKN1A promoter is specific. Together, these results indicate that CDKN1A is under the control of SLFN11-NFκB2 signaling in GBM and raise the possibility that p21 mediates, at least in part, the antineoplastic effects observed after SLFN11 loss (see Fig. 2). In line with this, we found that efficient CDKN1A knockdown (Fig. 5H) restored neurosphere growth (Fig. 5I) and proliferation (Fig. 5J) in SLFN11 KO LN229 and U87 cells.
On the basis of the potent biological effects of SLFN11 loss observed in vitro, we proceeded to examine KO effects on tumor cell growth in vivo. In athymic mice intracranially implanted with GBM6-SLFN11 KO cells, tumor growth was greatly inhibited as compared with mice that had received GBM6 WT cells (Fig. 6A and B). As a result, the lack of SLFN11 prolonged survival (Fig. 6C). IHC analysis revealed that GBM6 WT tumors strongly expressed SLFN11, whereas GBM6-SLFN11 KO tumors depicted greatly reduced SLFN11 staining (Fig. 6D and E, left). In agreement with our in vitro results (see Fig. 5), GBM6-SLFN11 KO tumors exhibited significantly increased proportions of p21-positive cells (Fig. 6D, bottom) as well as increased expression of CDKN1A (Fig. 6E, right). In summary, loss of SLFN11 increases p21 expression, blocks tumor growth and prolongs survival in an intracranial PDX model.
The results of this study show that expression of SLFN11 contributes to GBM growth and malignancy. Our investigation began with an analysis of TCGA data, which revealed that SLFN11 expression is inversely correlated with malignant glioma patient survival. This observation prompted our detailed examination of the molecular, cellular, and tumor biologic effects of SLFN11 in GBM. Genetic disruption of SLFN11 in three distinct GBM cell sources inhibited cell proliferation and neurosphere growth, and reduced the expression of genes associated with progenitor/stem cell characteristics in neurosphere models, suggesting a GSC-supportive function of SLFN11. The potential role of SLFN11 in the regulation of stem cell properties is of particular interest given the importance of GSCs in contributing to GBM heterogeneity, response to treatment, and evolution (i.e., transcriptional subtype transitions; refs. 41–45). Our results show that disruption of SLFN11 expression greatly impairs tumor growth and significantly improved survival in an orthotopic PDX model. This finding is of utmost importance because SLFN11 mediates cell death in response to DNA-damaging agents (DDA) such as topoisomerase inhibitors and alkylating agents like cisplatin and TMZ (37, 46, 47). Hence, stimulation of SLFN11 expression via promoter demethylation by histone deacetylase inhibitors has been suggested as a strategy to sensitize cancer cells to DDA (48). However, our findings necessitate careful assessment of such strategies in GBM because stimulation of SLFN11 expression might trigger some undesired glioma-promoting effects. Supporting this notion are data from pediatric sarcomas including Ewing sarcoma, where SLFN11 is highly expressed. In these sarcomas, elevated SLFN11 protein expression was associated with worse outcome in terms of recurrence-free survival, and recurrent and resistant sarcomas still exhibited high SLFN11 expression (24). Hence, in some cancers SLFN11 may execute additional roles, besides its DDA sensitizing ability, that may contribute to tumor progression. SLFN11 is a well-established predictor of response to a variety of DDAs and PARP inhibitors (49). Still, whether SLFN11 expression may serve as a treatment biomarker in GBM remains to be elucidated. Our findings are consistent with SLFN11 as a potential prognostic biomarker for GBM. SLFN11 might also represent a potential target for therapeutic anti-GBM strategies with the caveat that SLFN11-depleted cells may exhibit reduced response to DNA damage induced by chemoradiation. Importantly, inhibition of ataxia telangiectasia and rad3-related (ATR) kinase was shown to reverse resistance in SLFN11-deficient cancer cells (21). Furthermore, a recent genome-wide RNAi chemosensitization screen identified several components of the ATR/CHK1 signaling pathway as potent hits in SLFN11 KO cells and clinical inhibitors of these targets reversed the resistance to a broad range of DDAs seen in SLFN11-deficient cells (50). Thus, inhibitors of ATR pathway components might represent promising combinatorial candidates in SLFN11-depleted cells. Further studies are required to determine whether combined targeting of SLFN11 and components of the ATR/CHK1 pathway might enhance antitumor effects in patients with GBM treated with chemoradiation, the current standard of care. Our data provide, for the first time, definitive evidence that SLFN11 associates with NFκB2 in GBM cells. The association of SLFN11 with NFκB2, appears to repress NFκB transcriptional activity because loss of SLFN11 stimulated expression of NFκB target genes. Mechanistically, loss of SLFN11 triggered enrichment of both p52 and RelB on the CDKN1A promoter and induced expression of p21 in an NFκB2-dependent way. Thus, based on these findings, it appears that SLFN11 blocks the p52:RelB heterodimer from occupying target gene promoters. On the basis of these findings, we propose a model in which SLFN11 associates with and inhibits NFκB2 to repress p21 in GBM. Consistent with this interpretation, p21 protein was enriched in orthotopic PDX tumors established from SLFN11 KO cells. Importantly, knockdown of CDKN1A restored cell proliferation and neurosphere growth in SLFN11 KO cells in vitro indicating the antineoplastic effects after SLFN11 loss can be rescued by concomitant suppression of p21 expression. Together, these results provide compelling evidence for a SLFN11-NFκB2-p21 axis, in which SLFN11 suppresses NFκB2–mediated p21 expression, and by extension promotes GBM progression. Irradiation and TMZ are essential components of the current treatment regimen for GBM, and both are potent DNA damage inducers. Besides p53, NFκB signaling is a major element for transcriptional reprogramming in response to DNA damage (51). DNA damage results in nuclear RelB enrichment and processing of p100 into p52, indicating a role also for the noncanonical NFκB pathway (51, 52). While SLFN11 is an established marker for sensitivity to DDA-mediated cancer cell killing, its role as a repressor of NFκB2 mediated transcription may complicate targeted approaches aiming to activate SLFN11 in GBM. Further studies are required to carefully dissect the effects of irradiation and TMZ on the transcriptional activity of SLFN11/NFκB2 associated signaling.
Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. |
|
PMC9648418 | 36382086 | Natalia E. Cortez,Cecilia Rodriguez Lanzi,Brian V. Hong,Jihao Xu,Fangyi Wang,Shuai Chen,Jon J. Ramsey,Matthew G. Pontifex,Michael Müller,David Vauzour,Payam Vahmani,Chang-il Hwang,Karen Matsukuma,Gerardo G. Mackenzie | A Ketogenic Diet in Combination with Gemcitabine Increases Survival in Pancreatic Cancer KPC Mice | 08-09-2022 | Pancreatic ductal adenocarcinoma (PDAC) continues to be a major health problem. A ketogenic diet (KD), characterized by a very low carbohydrate and high fat composition, has gained attention for its antitumor potential. We evaluated the effect and mechanisms of feeding a strict KD alone or in combination with gemcitabine in the autochthonous LSL-KrasG12D/+; LSL-Trp53 R172H/+; Pdx1-Cre (KPC) mouse model. For this purpose, both male and female pancreatic tumor-bearing KPC mice were allocated to a control diet (CD; %kcal: 65% carb, 15% protein, 20% fat), a KD (%kcal: 1% carb, 15% protein, 84% fat), a CD + gemcitabine (CG), or a KD + gemcitabine (KG) group. Mice fed a KD alone or in combination with gemcitabine showed significantly increased blood β-hydroxybutyrate levels compared with mice fed a CD or CG. KPC mice fed a KG had a significant increase in overall median survival compared with KPC mice fed a CD (increased overall median survival by 42%). Interestingly, when the data were disaggregated by sex, the effect of a KG was significant in female KPC mice (60% increase in median overall survival), but not in male KPC mice (28% increase in median overall survival). Mechanistically, the enhanced survival response to a KD combined with gemcitabine was multifactorial, including inhibition of ERK and AKT pathways, regulation of fatty acid metabolism and the modulation of the gut microbiota. In summary, a KD in combination with gemcitabine appears beneficial as a treatment strategy in PDAC in KPC mice, deserving further clinical evaluation. Significance: This article is the first preclinical study to comprehensively evaluate the effect of a KD alongside chemotherapy using a standard autochthonous genetically modified mouse model (in both male and female KPC mice). | A Ketogenic Diet in Combination with Gemcitabine Increases Survival in Pancreatic Cancer KPC Mice
Pancreatic ductal adenocarcinoma (PDAC) continues to be a major health problem. A ketogenic diet (KD), characterized by a very low carbohydrate and high fat composition, has gained attention for its antitumor potential. We evaluated the effect and mechanisms of feeding a strict KD alone or in combination with gemcitabine in the autochthonous LSL-KrasG12D/+; LSL-Trp53 R172H/+; Pdx1-Cre (KPC) mouse model. For this purpose, both male and female pancreatic tumor-bearing KPC mice were allocated to a control diet (CD; %kcal: 65% carb, 15% protein, 20% fat), a KD (%kcal: 1% carb, 15% protein, 84% fat), a CD + gemcitabine (CG), or a KD + gemcitabine (KG) group. Mice fed a KD alone or in combination with gemcitabine showed significantly increased blood β-hydroxybutyrate levels compared with mice fed a CD or CG. KPC mice fed a KG had a significant increase in overall median survival compared with KPC mice fed a CD (increased overall median survival by 42%). Interestingly, when the data were disaggregated by sex, the effect of a KG was significant in female KPC mice (60% increase in median overall survival), but not in male KPC mice (28% increase in median overall survival). Mechanistically, the enhanced survival response to a KD combined with gemcitabine was multifactorial, including inhibition of ERK and AKT pathways, regulation of fatty acid metabolism and the modulation of the gut microbiota. In summary, a KD in combination with gemcitabine appears beneficial as a treatment strategy in PDAC in KPC mice, deserving further clinical evaluation.
This article is the first preclinical study to comprehensively evaluate the effect of a KD alongside chemotherapy using a standard autochthonous genetically modified mouse model (in both male and female KPC mice).
Despite extensive efforts to develop new treatment strategies, pancreatic ductal adenocarcinoma (PDAC) continues to be a major health problem, with a 5-year survival of approximately 11% (1). While surgery is a viable option in a limited number of patients, the majority of patients with PDAC (>80%) are diagnosed with advanced, unresectable, or metastatic disease (2). For these patients, the standard treatments include combination of gemcitabine plus nanoparticle albumin bound (nab)-paclitaxel (Abraxane), or the combined therapy of leucovorin-modulated 5-fluorouracil, irinotecan, and oxaliplatin (FOLFIRINOX; refs. 2–5). Unfortunately, these chemotherapeutic strategies still provide limited clinical benefit. Hence, there is an urgent need to develop therapies that can improve outcomes in patients with PDAC, and the exploration of dietary interventions is a critical component. During the last years, there has been considerable interest in the antitumor evaluation of ketogenic diets (KD; 6). KDs are characterized by a high fat, moderate protein, and very low carbohydrate content. These diets mimic changes in metabolism that are similar to fasting by elevating circulating levels of ketone bodies (i.e., acetoacetate, β-hydroxybutyrate, and acetone), which serve as an alternative energy source (7) and as signaling molecules (8). Numerous studies have indicated that a KD inhibits tumor growth and increases survival (9–11), including in PDAC (12–14). Multiple cellular mechanisms might explain the beneficial effects of a KD in tumor growth. These include anti-inflammatory, antiangiogenesis, cell metabolism, and epigenetic effects, as well as modulation of the microbiome (15). Unfortunately, many of these studies were performed using xenograft models of pancreatic cancer, which do not closely recapitulate human PDAC, so their clinical significance is limited (16). Recently, Yang and colleagues, reported that a KD was effective as an chemotherapy adjuvant reducing tumor growth in syngeneic subcutaneous pancreatic tumors and prolonged survival in the clinically relevant LSL-KrasG12D/+, LSL-Trp53R172H/+, Pdx1-Cre (KPC) genetically engineered mouse model of pancreatic cancer (17). However, this study was performed using a small cohort of only male mice. In this study, we evaluated the impact of feeding a strict KD alone or in combination with gemcitabine in the autochthonous and clinically relevant KPC mouse model of pancreatic cancer (18, 19). Furthermore, we examined whether there might be sex-related differences in the response to a KD in PDAC. We observed that a KD in combination with gemcitabine extends survival in KPC mice and that female mice appear to be slightly more responsive to the KD. The mechanisms by which a KD plus gemcitabine increases survival response appear to be multifactorial, including inhibition of ERK and AKT pathways, regulation of fatty acid metabolism, and the modulation of the gut microbiota.
All animal used procedures were approved by the University of California, Davis Animal Care and Use Committee.
The genetically engineered LSL-KrasG12D/+; LSL-Trp53R172H/+; Pdx-1-Cre (KPC) mice were bred at the UC Davis Animal Facility in Meyer Hall. KPC mice were generated from three mouse parental strains (LSL-KrasG12D/+; LSL-Trp53R172H/+; and Pdx-1-Cre), obtained from NCI mouse repository, following established procedures described by Hingorani and colleagues (18). After weaning, mice were individually housed in polycarbonate cages in a room with controlled temperature (22°C–24°C) and humidity (40%–60%), maintained on a 12-hour light-dark cycle, and fed chow diet ad libitum LabDiet 5001 (LabDiet) until enrolled in the studies.
Enrollment of KPC mice was based on tumor size, measured using a high-resolution ultrasound imaging of the pancreas with the Vevo 2100 System with a 35 MHz RMV scan-head (Visual Sonics, Inc.), when KPC mice were around 3–4 months old (Supplementary Fig. S1A). Imaging was obtained and tumor volumes measured following previously published guides (20, 21). Once tumor size was assessed (Fig. 1A), male and female KPC mice were assigned randomly to one of four groups: a control diet (CD), KD, a control diet plus gemcitabine (CG) or a ketogenic diet plus gemcitabine (KG).
Following tumor size determination, male and female KPC mice (7–12 mice per sex per group; 16–23 mice/group) were allocated to either a CD (%kcal: 65% carb, 15% protein, 20% fat), a KD (%kcal: 1% carb, 15% protein, 84% fat), a CD + gemcitabine, or a KD + gemcitabine group. Mice were fed ad libitum, and food was changed and food intake was recorded three times per week. The composition of diets was adapted from the study by Roberts and colleagues (22), and is shown in Supplementary Table S1. The Envigo mineral mix TD94046 was used for the CDs and the TD79055 was used for the KDs due to their lower carbohydrate contents. For both diets, TD40060 (vitamin mix) was used.
Gemcitabine (>99% 2′-Deoxy-2′,2′-difluorocytidine; dFdC; Gemzar; LY-188011) from Thermo Fisher Scientific was administered to the CG and KG groups at 100 mg/kg by intraperitoneal injections twice per week for 3.5 weeks (seven total injections). Throughout the survival study, mice were observed daily for signs of significant weight loss, hemorrhagic ascites, and for other signs of clinical failure including loss of thermoregulation, inactivity, and presence of malignant ascites. Endpoint criteria included the development of abdominal ascites, weight loss exceeding 20% of the initial weight, or extreme weakness or inactivity. When an animal reached the endpoint criteria, it was euthanized by carbon dioxide asphyxiation, blood was collected and tissues, including pancreatic tumors were dissected, weighted, and then stored in liquid nitrogen, RNA later, and 10% buffered formalin.
Nonfasting glucose levels were measured using a glucometer (Easy Plus II, Home Aid Diagnostics Inc), and β-hydroxybutyrate levels were measured using the Precision Xtra glucose and ketone monitoring system (Abbott) according to the manufacturer's instructions.
A cohort of mice was allocated to either the CG or the KG groups after tumor detection and euthanized at 2 months after interventions. We chose CD + gemcitabine as our control group to specifically depict the contribution of a KD to the effect. At the end of the 2 months, pancreas and pancreatic tumors were dissected, weighed, sectioned, and then stored in liquid nitrogen, RNAlater, and 10% buffered formalin.
Blood samples were collected via cardiac puncture and serum was isolated after centrifugation at 3,000 × g for 10 minutes at room temperature. Insulin was assayed using the V-PLEX mouse metabolic kit and mouse leptin kit. Inflammation-related biomarkers were assayed using the V-PLEX Proinflammatory panel I kit (Meso Scale Discovery).
After necropsy, pancreas specimens were fixed in 10% buffered formalin overnight at 4°C. Tissues were processed and embedded by routine methods. Tissue sections (5 μmol/L) were stained with hematoxylin and eosin or Masson's Trichrome (Chromaview, Thermo Fisher Scientific). Tumors were classified by morphologic pattern (glandular, spindled, solid), and each morphologic pattern was scored as a percentage of total tumor surface area. Presence and extent of tumor necrosis, and presence and type of background pancreatic fibrosis (e.g., interlobular, intralobular) were also scored. All histologic sections were evaluated in a blinded fashion.
Pancreas were fixed in 10% buffered formalin overnight at 4°C, processed and embedded using routine methods. Paraffin sections were deparaffinized, rehydrated, and heated for 12 minutes at 95°C in 10 mmol/L (pH 6) citrate buffer (M-15704, Thermo Fisher Scientific). Afterward, sections were incubated with 3% hydrogen peroxide (59105926, Millipore corporation) for 10 minutes and blocked-in animal-free Blocker (SP-5030, Vector laboratories) for 1 hour at room temperature and then incubated overnight at 4°C with primary antibody against p-ERK1/2 (1:200 dilution, Cell Signaling Technology, catalog no. 4376, RRID:AB_331772). The following day, paraffin sections were incubated with biotin-conjugated secondary antibody for 1 hour at room temperature (856743, Life Technologies), horseradish peroxidase (HRP) streptavidin for 1 hour at room temperature (856743, Life Technologies), and developed by DAB (SK-4100, Vector Laboratories), followed by hematoxylin (MHS16, Sigma) staining. Sections were then dehydrated, mounted in Cytoseal 60 mounting medium (8310-16, Thermo Fisher Scientific), and analyzed using an Olympus BX51 microscope. Scoring: Five or more fields per sample (at magnification × 200) were scored and the percent of positive cells was calculated as described previously (23).
Pancreas tissue homogenates were prepared and Western blots were performed as described previously (23). Aliquots of total homogenates containing 25–40 μg protein were separated by reducing 8%–12.5% (w/v) PAGE and electroblotted onto nitrocellulose membranes. Membranes were blocked in 5% (w/v) nonfat milk for 1 hour and subsequently incubated with the following antibodies from Cell Signaling Technology: p-ERK (catalog no. 4376, RRID:AB_331772), ERK (catalog no. 9102, RRID:AB_330744), p-Akt (Ser473; catalog no. 4060, RRID:AB_2315049), AKT (catalog no. 9272, RRID:AB_329827), p-AMPKα (Thr172: catalog no. 2535, RRID:AB_331250), AMPKα (catalog no. 2532,RRID:AB_330331), p-4E-BP1 (Thr37/46; catalog no. 2855, RRID:AB_560835), 4E-BP1 (catalog no. 9452, RRID:AB_331692), HKII (catalog no. 2867, RRID:AB_2232946), PDH (catalog no. 3205, RRID:AB_2162926), LDH (catalog no. 2012, RRID:AB_2137173), p-IGFR-R (catalog no. 3024, RRID:AB_331253), IGFR-R (catalog no. 9750, RRID:AB_10950969), and PKM2 (catalog no. 4053, RRID:AB_1904096), using a 1:1,000 dilution, overnight at 4°C. After incubation for 1 hour at room temperature in the presence of secondary antibodies (either HRP or biotinylated antibodies, followed by 1 hour incubation with streptavidin when biotinylated antibody was used in a 1:5,000 dilution), the conjugates were visualized and quantified by chemiluminescence detection in a Chemidoc Imaging-System, Bio-Rad Laboratories (RRID:SCR_008426), Inc. β-Actin (catalog no. A1978) from Millipore-Sigma, was used as a loading control. The densitometry analysis was performed using the Image J Program (RRID:SCR_003070).
Mice for the RNA sequencing (RNA-Seq) data were treated for 2 months with diet and chemotherapy. Tissues were stabilized in RNAlater (Thermo Fisher Scientific). Total RNA was extracted following the manufacturer's instructions using a RNeasy mini kit (74104, QIAGEN) from frozen pancreas and/or tumors. RNA quality was confirmed using Nano drop one (Thermo Fisher Scientific). Library preparation and RNA-Seq were performed by Novogene Co., LTD. In brief, mRNA was enriched using oligo(dT) beads, and rRNA was removed using the Ribo-Zero kit. The mRNA was fragmented, and cDNA was synthesized by using mRNA template and random hexamers primer, after which a second-strand synthesis buffer (Illumina), dNTPs, RNase H, and DNA polymerase I were added for the second-strand synthesis, followed by adaptor ligation and size selection. The library was sequenced by the Illumina Novaseq platform. Raw data were aligned to mm10 genome using HISAT2, read counts and normalized read count were generated using the feature counts, and the differentially expressed genes were identified using DESeq2 (RRID:SCR_000154).
Fatty acid content in KPC pancreatic tumors of CG- and KG-treated mice was measured using gas chromatography (GC). Briefly, pancreatic samples were freeze dried and direct methylated with sodium methoxide as described previously (24). Cis-10–17:1 methyl ester (Nu-Check Prep Inc.) was added as an internal standard prior to methylating reagent. Fatty acid methyl esters (FAME) were analyzed by GC using a CP-Sil88 column (100 m, 25 μm ID, 0.2 μm film thickness) in a TRACE 1310 gas chromatograph (Thermo Fisher Scientific) equipped with a flame-ionization detector (GC-FID, Thermo Fisher Scientific). Each sample was analyzed twice by GC using a 175°C plateau temperature program (24). The FAME were quantified using chromatographic peak area and internal standard-based calculations.
Fecal samples were collected from KPC mice at baseline and 1 month after dietary intervention ± gemcitabine treatment (from the KPC survival study groups). All fecal samples were collected directly from the animals on Eppendorf tubes and immediately frozen in liquid nitrogen. Genomic DNA was extracted from all samples using a commercially available kit (Qiagen QIAamp PowerFecal Pro DNA Kit, catalog no. 51804) and following manufacturer's instructions. DNA concentrations of each sample were evaluated using Qubit dsDNA High Sensitivity Assay Kit (catalog no. Q32851) with Qubit 4.0 Fluorometer, following manufacturer's instructions. Quality assessment was performed by agarose gel electrophoresis to detect DNA integrity, purity, fragment size, and concentration. The 16S rRNA amplicon sequencing of the V3-V4 hypervariable region was performed with an Illumina NovaSeq 6000 PE250. Sequences analysis were performed by Uparse software (Uparse v7.0.1001; 25), using all the effective tags. Sequences with ≥97% similarity were assigned to the same operational taxonomic units (OTU). Representative sequence for each OTU was screened for further annotation. For each representative sequence, Mothur software was performed against the SSUrRNA database of SILVA Database (26). OTUs abundance information was normalized using a standard of sequence number corresponding to the sample with the least sequences.
The accession number for the RNA-Seq data reported in this study is NCBI Gene Expression Omnibus: GSE208398. The accession number for the microbiome data is under Bioproject number: PRJNA858994.
The data, obtained from at least three independent experiments, were expressed as the mean ± SEM. Statistical evaluation was performed by t-test or one-factor ANOVA followed by the Tukey test adjusted for multiple comparisons. Analyses were performed by GraphPad (Prism version 9.2, RRID:SCR_002798) and R version 4.0.4. Two-sided P < 0.05 was regarded as statistically significant. Kaplan–Meier methods and the log-rank tests were used to compare unadjusted survival outcome (time from the start of treatment to death) between treatments in overall and key subgroups. There is no censoring in the survival outcome. To adjust for possibly unbalanced age and sex between treatment groups and explore potential interactions, linear regression models for survival days since treatment to death were performed, which include diet (CD, KD), gemcitabine (no, yes), sex, age at the start of treatment (centered at 90 days), interaction between diet and gemcitabine, two- and three-way interactions between sex and treatments (diet and gemcitabine). All interactions were removed from the final model due to nonsignificance. Model diagnosis was performed to ensure that the assumptions of linear regressions hold. For the microbiota analysis, alpha- and beta-diversity were assessed by using standard metrics (e.g., Simpson and Shannon H diversity index) and Bray–Curtis principal coordinates of analysis (PCoA), respectively. Statistical significance was determined by Kruskal–Wallis or permutational multivariate ANOVA (PERMANOVA). Comparisons at the Phylum and Genus level were made using classical univariate analysis using Kruskal–Wallis combined with a FDR approach used to correct for multiple testing. Finally, LEfSe (linear discriminant analysis effect size) was also employed to determine the features most likely to explain differences between classes.
We first conducted a survival study in the clinically relevant KPC mouse model to evaluate the effect of feeding a strict KD alone or in combination with gemcitabine as a treatment protocol in male and female KPC mice bearing pancreatic tumors. For this purpose, we enrolled KPC mice with similar tumor sizes, measured by high-resolution ultrasound imaging of the mouse pancreas (Fig. 1A; Supplementary Fig. S1). Male and female KPC mice were divided into CD, KD, CD + gemcitabine (CG), or KD + gemcitabine (KG) groups (16–23 mice/group; Fig. 1B). While a KD alone had no significant effect on KPC survival, the combination of a KD with gemcitabine synergistically prolonged survival. The overall median survival times among the four groups were 80, 94, 88, and 119 days for CD, KD, CG, and KG groups, respectively. While KD alone or CG treatments were unable to extend KPC mouse survival, KPC mice fed a KG had a significant increase in overall median survival compared with KPC mice fed a CD (increased overall median survival by 42%; Fig. 1C). A 26% increase survival was observed when comparing KG group with CG. Interestingly, when we separated by sex, the effect of a KG was significant in female KPC mice (60% increase in median overall survival (P = 0.028), but not in male KPC mice [28% increase in median overall survival compared with CD mice (P = 0.089),] (Fig. 1D and E; Supplementary Fig. S1B). The median survival times for CD and KG groups were 77 and 123 days in females; and 80 and 103 in males, respectively. Interestingly, the weights of the pancreas/tumors were comparable among the groups, with only a significant decrease of the tumor weight of KG-treated female mice was observed compared with KD alone (Fig. 1F). Of note, treatment with KD, CG, or KG was well tolerated with no body weight loss throughout the treatment, as compared with the baseline body weight levels (Fig. 1G). Histopathologic evaluation of the tumors showed the classic glandular morphology of PDAC, as originally described by Hingorani and colleagues (18). Spindled and solid patterns were also observed as secondary or primary patterns [also described by Hingorani and colleagues (18); Fig. 1H]. Interestingly, an increased proportion of CG and KG tumors demonstrated classic glandular morphology compared with CD and KD tumors (Fig. 1H), whereas the latter showed an increased proportion of spindled and solid patterns, generally considered indicative of more aggressive behavior (27, 28). In addition, the increase in glandular morphology in the KG-treated group was found to be the result of a marked predominance of glandular morphology in the tumors of the female mice. Some tumors showed some degree of necrosis. Overall, we noted a slight increase in tumor necrosis in females fed the KD compared with CD, with and without gemcitabine (Supplementary Fig. S2). The residual background pancreas (when present) showed a combination of intralobular and interlobular fibrosis. No significant difference in the pattern of fibrosis (intralobular vs. interlobular) was noted among the four groups (Supplementary Fig. S3). To adjust for possibly unbalanced age and sex between intervention groups and further explore whether a KD ± gemcitabine's survival effect is sex-dependent, we conducted linear regression models for survival days since treatment to death, which was adjusted by sex and age at the start of treatment. As shown in Table 1, gemcitabine significantly extended mean survival by 25.8 days (P = 0.002), KD extended mean survival by 13.8 days with a trend toward significance (P = 0.052). Compared with CD, KG significantly extended mean survival by 39.6 days (P < 0.001). Although Kaplan–Meier curves in sex subgroups suggested that the effect of KG is likely more effective in females than in males, the interactions between sex and treatments in linear regressions were not significant and hence removed from the final model. Thus, the treatment effect was comparable across both sexes, benefiting both females and males. Mice fed a KD, alone or in combination with gemcitabine, showed significantly increased blood ketones compared to mice fed a CD or CG (Fig. 2A). The β-hydroxybutyrate levels in the KD and KG groups remained elevated throughout the study. The increase in β-hydroxybutyrate levels were observed in both female and male mice fed a KD or KG (Fig. 2A). In contrast, glucose levels were significantly higher in the CD and CG groups when compared with KD only at 1 month (Fig. 2B). When disaggregated by sex, such effect was only observed in males (Fig. 2B). Furthermore, because KDs have been shown to exert anti-inflammatory effects (29), we assessed the levels of several proinflammatory cytokines in the serum of male and female KPC mice at endpoint. In males, there was a decrease of TNFα in KD compared with CD and a decrease in IL6 in the KG group when compared with KD. In addition, higher levels of IL1β were observed in KG males compared with CD. In males, higher levels of KC/GRO were observed in the KG group compared with CD and CG groups. On the other hand, in females, no significant changes in serum cytokines were observed among the CD, KD, CG, and KG groups (Fig. 2C). Moreover, no significant differences, in both males and females, were observed in IFNγ, IL10, or MCP-1 levels among the groups (Supplementary Fig. S4).
To elucidate the cellular mechanisms underlying the beneficial effects of a KD plus gemcitabine on pancreatic tumors, we conducted a study in which male and female KPC mice bearing pancreatic tumors (3 months old) were treated with either CG or KG for 2 months (Fig. 3A). We chose CD plus gemcitabine as our control group to specifically depict the contribution of a KD to the effect. After 2 months of treatment, no differences in the weights of the pancreas and/or tumors were observed between CG- and KG-treated mice (Fig. 3B). To investigate whether KD plus gemcitabine induces changes in PDAC tumors in females that would suggest general antitumor activity, we initially performed an RNA-Seq analysis followed by HALLMARK gene set enrichment analysis (GSEA) on female pancreatic tumors obtained from KG or CG mice after 2 months of treatment. KG treatment was broadly associated with increased changes in the expression of genes involved in early and late estrogen response, xenobiotic metabolism, glycolysis, and fatty acid metabolism. In contrast, KG treatment was associated with decreased changes in the expression of genes involved in allograft rejection, IFN alpha and gamma response, PI3K-AKT-mTOR as well as unfolded protein response (Fig. 3C).
Two pathways commonly activated in PDAC are PI3K-AKT-MTOR and Kras/MAPKs (30, 31). GSEA data indicated that PI3K-AKT-MTOR was one of the pathways downregulated in the KG group, compared with CG (Fig. 4A). Thus, we validated these data by assessing the activation status of key proteins in the PI3K/Akt/mTOR, as well as the Raf/MEK/ERK pathways by immunoblot. Although there was no significant difference in AKT, ERK, or AMPK phosphorylation in pancreatic tumors between KG and the other groups in the survival study (Supplementary Fig. S5), KG treatment significantly reduced AKT, ERK, IGFR, and AMPK phosphorylation in pancreatic tumors of female, but not male, KPC mice, compared with CG-treated mice, at 2 months of treatment (Fig. 4B and C). In contrast, no significant changes were observed in the expression levels of phosphorylated 4EBP-1 between the two groups (Fig. 4B and C). To confirm these results, we assessed ERK activation by IHC of tumor sections prepared from CG- and KG-treated female and male KPC mice. KG reduced p-ERK levels by 79% in females, compared with CG-treated mice (Fig. 4D; P < 0.08 for females). Furthermore, because AKT activation can be regulated by insulin, we assessed serum insulin levels. Compared with CG-treated mice, after 2 months KG treatment reduced insulin levels by 85.5% in female and 78.2% in male KPC mice (Fig. 4E).
Among many signatures, GSEA of differentially expressed genes in tumors from KG- and CG-treated female mice identified glycolysis signatures as highly affected (Fig. 5A). Thus, we assessed the expression levels of several enzymes linked to glucose metabolism in the pancreas/tumors of KPC mice. In both female and male KPC mice in the survival study, KG reduced hexokinase 2 (HK2) levels when compared with CD (Fig. 5B), but no changes were observed in animals euthanized at 2 months between KG and CG groups (Fig. 5C). Moreover, no significance differences were observed in the expression levels of LDH, PKM2, and PDH (Supplementary Fig. S6).
GSEA of differentially expressed genes between KG and CG female KPC tumors identified fatty acid metabolism signatures as highly enriched in KG-treated mice compared with CG-treated mice (Fig. 6A). To understand more comprehensively, which fatty acids are affected in pancreatic tumors, we analyzed the fatty acid composition of tumors isolated from female and male KPC mice treated with a KG or a CG for 2 months. As shown in Fig. 6B, there were no significant differences in concentrations of total saturated fatty acids (SFA), total cis-monounsaturated fatty acids (c-MUFA), total n6-polyunsaturated fatty acids (n6-PUFA), and n3-PUFA in the pancreas of KG mice compared with CG mice. This holds true when separated by sex. Interestingly, when examining changes of individual fatty acids between KG and CG, we observed that KG significantly reduced concentrations of asclepic acid (cis11–18:1), palmitoleic acid (cis9–16:1), and eicosatrienoic acid (20:3n-3), while increased margaric acid (17:0) content, compared with CG (Fig. 6C). Distinctively in female KPC mice, KG significantly reduced the concentrations of palmitic acid (16:0), myristoleic acid (cis9–14:1), palmitoleic acid (cis9–16:1), and linoleic acid (18:2n-6), and significantly increased the concentrations of margaric acid (17:0) and stearic acid (18:0) when compared with CG-treated females (Fig. 6D). No significant changes in any fatty acid concentrations were observed between KG- and CG-treated KPC male mice.
Given that diet influences the composition of the gut microbiota, and the gut microbiota can affect PDAC growth and response to treatment (32, 33), we next performed 16S rRNA sequencing to evaluate the impact of a KD alone or in combination with GEM (KG) on the gut microbiota. For this purpose, we collected fecal samples at baseline (KPC mice fed chow diet, prior to dietary and/or chemotherapeutic treatments) and after 1 month of treatment with CD, KD, CG, or KG and assessed the α-diversity among groups. As expected, at baseline, there were no significant differences on the microbiota composition and/or diversity among the four groups. As shown by the Shannon and Simpson diversity indices, there were no significant differences on the microbiota diversity in the CD-fed group pre to post dietary intervention, but a significant difference was observed in both gemcitabine-treated groups (P < 0.001). When comparing mice fed a CD with those in the KG group, a significant difference was observed (P = 0.0003). Interestingly, a significant difference was observed when comparing animals fed a KD with those in the KG group, as depicted by the Shannon index (P = 0.0163; Fig. 7A). We next analyzed the taxonomic components for all groups to confirm the specific changes of the microbial community. At the phylum level, Firmicutes and Bacteroidetes dominated the gut microbiota, and lower levels of Proteobacteria were detected. Compared with CD-fed mice, there was an increase in the relative abundance of Firmicutes in both KD-fed groups at 1 month of treatment. At 1 month of treatment, the ratio Firmicutes/Bacteroidetes was significantly higher for the KG (ratio = 9) group when compared with all others [CD (ratio = 2.9), KD (ratio = 5.3), CG (ratio = 2.5)] (Fig. 7B). At the Genus level, all post-treatment groups increased the levels of Faecalibaculum, Romboutsia, and Erysipelatoclostridium, while reduced Lactobacillus levels, compared with baseline levels. Romboutsia levels were higher in the CD groups, while the increase in Erysipelatoclostridium was more apparent on the KD-fed animals. Interestingly, Dubosiella increased only in both GEM-treated groups. Of note, the levels of Faecalibaculum were significantly increased in the KG-treated mice when compared with all three other groups (Fig. 7C). On the basis of the differences in microbial community composition among groups, we next performed a Bray–Curtis PCoA to define the similarity of species diversity among groups on OTU level (Supplementary Fig. S7A). Although there was a significant impact of treatment on microbial beta-diversity (PERMANOVA: F-value: 4.6619, P < 0.001), no significant changes due to sex were observed (PERMANOVA: F-value: 1.1273; P < 0.321; Supplementary Fig. S7B). Finally, given that only KD plus gemcitabine increased overall survival, we aimed to identify some key species of bacteria that were differentially present in the KG group compared with KD or CG groups alone by performing a LEfSe analysis. The linear discriminant analysis (LDA) histogram was used to calculate the significant changes in the gut microbiota and interpret the degree of consistent difference of relative abundance between treatment groups. LDA results showed several discriminative features in the KG group (LDA>3.6, P < 0.05), compared with either KD or CG groups (Supplementary Fig. S8). The major species that were significantly increased in KG versus KD and KG versus CG include: genus_Faecalibaculum, class_Erysipelotrichia and order_Erypiselotrichales and family_Eryspelotrichales. Moreover, major species that were significantly decreased in KG versus KD and KG versus CG were order_Lactobacillales, family_Lactobacillae, genus_Lactobacillus phyllum_Bacteroidetes, order_Bacteroidales, class_Bacilli (Supplementary Fig. S8). Of note, increases in order_Erypiselotrichales and a decrease in order_Lactobacillales, family_Lactobacillae, genus_Lactobacillus was observed when comparing as the KD versus CD (Supplementary Fig. S8).
Dietary interventions hold promise in cancer treatment, including PDAC. Previous studies in animal models suggested that a KD is an effective adjuvant therapy for pancreatic cancer, yet the significance of the clinical benefit of KDs was limited because of the use of xenograft models, only one sex, or the use of small cohorts (12–14). We observed that in the clinically relevant KPC mouse model, mice fed a strict KD in combination with gemcitabine exhibited a significant increase in overall median survival, compared with KPC mice fed a CD, and this beneficial effect was superior in female mice compared with male mice. Although our linear regression model indicates that the effect of a KD plus gemcitabine is likely not sex dependent, benefiting both males and females, the survival curves suggest that the effect of a KD plus gemcitabine is somewhat more effective in females. Indeed, when disaggregating the data between females and males, the effect of a KD plus gemcitabine was significant in female KPC mice (60% increase in median overall survival), but not in male mice (28% increase in median overall survival). It is important to note that treatment with a KD alone had no effect on KPC survival, indicating that the dietary changes themselves were insufficient to cause the tumor responses. Consistent with our findings, other investigators have recently evaluated the use of a KD in preclinical KPC allograft tumor models. For instance, Hopkins and colleagues observed that a KD rendered PI3K inhibitors, which are normally inactive against PDAC, effective in a KPC cell line–based orthotopic allograft tumors (14). In addition, Yang and colleagues recently showed that a KD synergized with a clinically relevant chemotherapeutic regimen of gemcitabine, nab-paclitaxel, and cisplatin, significantly increasing survival in subcutaneous KPC allograft tumors (17). Overall, these findings, together with our data, strongly indicate that a KD is an effective adjuvant dietary strategy for PDAC, and supports the initiated clinical trials (i.e.,NCT04631445), currently underway, to investigate its benefit in humans. Mechanistically, the survival response to a KD plus gemcitabine appears to be multifactorial, including the inhibition of ERK and AKT pathways, regulation of fatty acid metabolism and the modulation of the microbiota. Interestingly, we noted some discrepancies between ours and Yang and colleagues's RNA-Seq data (17). For example, while allograft rejection, IFN alpha and gamma response gene sets were down regulated in our data, they noted the opposite. These discrepancies might be the result of the differences in tumor types used in the analysis (KPC tumors vs. allografts), differences of the tumor microenvironment, or the variances in the duration of KD intervention and other interventions (i.e., gemcitabine). Therefore, and as suggested by our RNA-Seq analysis, at this time, we cannot rule out that other mechanisms, including modulation of xenobiotic metabolizing enzymes or estrogen responses, could also contribute to the effect of a KD in PDAC. Many features contribute to the reduced effectiveness of gemcitabine, including the dysregulation of signaling pathways related to cell metabolism (34), such as the insulin/IGF-1R, ERK, and PI3K/AKT pathways. For example, the PI3K/AKT pathway is aberrantly activated in multiple tumor types, regulating tumorigenesis, cancer metabolism, and drug resistance (35, 36). On the other hand, the deregulation of the ERK pathway is a signature of many epithelial cancers, including PDAC (37), whereas the upregulation of the insulin/IGF-1R pathway in PDAC occurs in over 70% of patients (38). Interestingly, compensatory upregulation of IGF-1R and ERK signaling limits the efficacy of select inhibitors, such as autophagy inhibitors, and their concurrent inhibition synergistically increases autophagy dependence and chloroquine sensitivity in PDAC (39). Therefore, the fact that a KD inhibits ERK, AKT, and IGFR activation might explain, at least in females, the survival benefit of its combination with gemcitabine. Lipid metabolism is essential for cancer progression (40), with increased levels of specific fatty acids known to regulate pancreatic cancer progression (41). For example, Lien and colleagues recently showed that the upregulation of stearoyl-CoA desaturase, which synthesizes MUFAs from SFAs, is essential for cancer cells to grow (42). Interesting, they suggest that modifying the composition of the dietary fat could lead to higher tumor inhibitory effect. For instance, altering the KD fat composition, by using palm oil instead of lard as the source of fat, slowed tumor growth, by increasing tumor saturated fatty acid levels, lowering MUFAs and decreasing tumor stearoyl-CoA desaturase activity. Although we did not observe significant differences in overall SFAs or MUFAs between KG and CG groups, we observed a reduction in select MUFAs in the KG group compared with the CG group. Because the KD used in our study was mainly prepared with lard, it would be important to evaluate whether a KD from other fat sources that increase SFAs might provide an additional beneficial effect. Several studies have also shown a positive association between higher consumption of certain fatty acids and pancreatic cancer risk. For example, high linoleic acid intake was shown to increase the risk of pancreatic cancer when compared with the individuals with the low linoleic acid intake (43). In a prospective nested case–control study, Yang and colleagues identified a fatty acid pattern using principal component analysis, associated with an increased risk of prostate cancer, which was characterized by higher levels of 14 and 16 carbon SFA and MUFA including myristic acid, palmitic acid, myristoleic acid and palmitoleic acid, along with low levels of α-linolenic acid (44). Interestingly, many of the fatty acids were reduced in pancreatic tumors following KG treatment, such as palmitic acid, myristoleic acid, palmitoleic acid, asclepic acid, and linoleic acid. Additional studies are warranted to validate whether one or more of these fatty acids could explain, in part, the beneficial effect of a KD in PDAC, and whether the modifying the type of fat used in the KD could lead a higher tumor inhibitor effect. The gut microbiota is an emerging mediator of PDAC progression (45), with many strategies to modulate the gut microbiome in PDAC being actively explored (46). For example, the transplantation of human fecal microbes can affect PDAC tumor response by modulating the gut microbiota and the immune system (47). In addition, two bacterial communities (Faecalibaculum and Lactobacillus) have been recently documented to play a critical role regulating tumor growth. Zagato and colleagues identified that Faecalibaculum rodentium, belonging to the Erysipelotrichaceae family, was strongly underrepresented during the early phases of tumorigenesis in the ApcMin/+ mice compared with wild-type mice, and that it was responsible for inhibiting intestinal tumor cell proliferation (48). Moreover, Faecalibaculum can inhibit tumor growth in breast cancer models (49). On the other hand, bacteria belonging to the genus Lactobacillus, which are gut commensals with an ability to produce indoles from tryptophan (50), can drive suppression in the pancreatic tumor microenvironment promoting tumor growth (51). Our findings, showing that KG treatment leads to increased relative abundance of Faecalibaculum and the reduction of Lactobacillus, might provide a partial explanation of the beneficial effects of KD in combination with gemcitabine observed in KPC mice. Future investigations will determine whether the selective modulation of these bacteria can be used to improve the therapeutic response in PDAC. In summary, a KD in combination with gemcitabine is beneficial as a treatment strategy for PDAC in KPC mice. The mechanisms by which KD plus gemcitabine increase survival response are multifactorial, including inhibition of ERK and AKT pathways, regulation of fatty acid metabolism and the modulation of the microbiota. These data in an autochthonous and clinically relevant mouse model strongly suggest that a KD should be evaluated concomitant to chemotherapeutic treatment in the clinical setting.
Click here for additional data file. |
|
PMC9648433 | Biçem Demir,Elmas Beyazyüz,Murat Beyazyüz,Aliye Çelikkol,Yakup Albayrak | Long-lasting cognitive effects of COVID-19: is there a role of BDNF? | 10-11-2022 | BDNF,Cognitive,COVID-19,Impairment | Coronavirus disease 2019 (COVID-19) affects numerous systems of the body during the illness, and there have been long-lasting effects. BDNF plays an important role in synaptic plasticity and synaptic communication. According to the inclusion and exclusion criteria, 54 patients who had COVID-19 infection participated in this study. Thirty-six age-, sex-, body mass index (BMI)-, education level- and smoking status-matched healthy controls were included in the present study. All participants were individually administered the Stroop test and Visual Aural Digit Span Test Form B (VADS-B). Serum BDNF levels were measured by ELISA. Stroop test word reading spontaneous correction number and reading time, word color saying wrong number, spontaneous correction number and reading time, box color speaking spontaneous correction number and reading time, Stroop interference and speed factor duration were significantly higher in the COVID-19 group than in the control group. All scores of the VADS-B test were found to be significantly lower in the COVID-19 group. The mean serum BDNF levels were found to be 10.9 ± 6.9 ng/ml in the COVID-19 group and 12.8 ± 6.4 ng/ml in the healthy control group. Two-way ANOVA showed that the serum mean BDNF level was significantly lower in the COVID-19 group than in the control group. Gender had a significant effect on BDNF levels (F = 12.21; p = 0.008). The present study is the first to demonstrate the association between the role of serum BDNF and cognitive decline in patients with COVID-19 infection. Additionally, there is a significant role of male gender in terms of lower BDNF level and cognitive decline. | Long-lasting cognitive effects of COVID-19: is there a role of BDNF?
Coronavirus disease 2019 (COVID-19) affects numerous systems of the body during the illness, and there have been long-lasting effects. BDNF plays an important role in synaptic plasticity and synaptic communication. According to the inclusion and exclusion criteria, 54 patients who had COVID-19 infection participated in this study. Thirty-six age-, sex-, body mass index (BMI)-, education level- and smoking status-matched healthy controls were included in the present study. All participants were individually administered the Stroop test and Visual Aural Digit Span Test Form B (VADS-B). Serum BDNF levels were measured by ELISA. Stroop test word reading spontaneous correction number and reading time, word color saying wrong number, spontaneous correction number and reading time, box color speaking spontaneous correction number and reading time, Stroop interference and speed factor duration were significantly higher in the COVID-19 group than in the control group. All scores of the VADS-B test were found to be significantly lower in the COVID-19 group. The mean serum BDNF levels were found to be 10.9 ± 6.9 ng/ml in the COVID-19 group and 12.8 ± 6.4 ng/ml in the healthy control group. Two-way ANOVA showed that the serum mean BDNF level was significantly lower in the COVID-19 group than in the control group. Gender had a significant effect on BDNF levels (F = 12.21; p = 0.008). The present study is the first to demonstrate the association between the role of serum BDNF and cognitive decline in patients with COVID-19 infection. Additionally, there is a significant role of male gender in terms of lower BDNF level and cognitive decline.
Severe acute respiratory syndrome (SARS-CoV-2) first appeared in Wuhan, China, in December 2019 and has been defined as coronavirus disease 2019 (COVID-19). COVID-19 was declared a pandemic by the World Health Organization (WHO) on March 11, 2020 [1]. COVID-19 affects numerous systems of the body during the course of the illness, and there have been long-lasting effects of the disease after recovery. It has been well established that cognitive deficits can be seen during the illness, and moreover, cognitive problems have also been reported to be seen after recovery [2]. The most common neurological symptoms in COVID-19 are headache, dizziness, anosmia, fatigue, myalgia, anorexia and ageusia. Severe neurological manifestations include confusion, seizures, cerebrovascular diseases, meningoencephalitis, acute necrotizing encephalopathy, posterior hemorrhagic encephalopathy syndrome, myopathy, radiculopathy, cerebellar ataxia, myoclonus and Guillain–Barre syndrome [3, 4]. The most common psychiatric symptoms of COVID-19 are as follows: depression, anxiety, sleep disorders, chronic fatigue syndrome and posttraumatic stress disorder symptoms [5]. Additionally, cognitive symptoms have recently been addressed [6]. SARS COV-2 binds to the ACE-2 receptor and enters epithelial cells in the lung. The S protein is cleaved by proteases such as TMPRSS2, cathepsin G, trypsin or disintegrin, and metalloprotease 17 (ADAM17) to facilitate viral entry. As a result, ACE-2 receptors are blocked. When ACE-2 activity is lost, the levels of angiotensin 1–7 and angiotensin 1–9 decrease. Based on these decreases, MAS/G protein-dependent receptors cannot be activated, vasodilation cannot occur, and cell protective mechanisms cannot be activated. All these mechanisms result in vasoconstriction, fibrosis, proliferation and atherogenesis, which are significantly associated with thrombophilia, microthrombosis, alveolar epithelial damage and respiratory failure [7]. BDNF is a protein member of the neurotrophin family, which includes neurotrophin 3 and neurotrophin 4. BDNF plays an important role in synaptic plasticity and synaptic communication [8]. The neurotrophic functions of BDNF are associated with memory, learning, sleep, appetite and neuronal survival. It is also well established that BDNF plays a critical role in hippocampal long-term potentiation (LTP), which is a long-term result of synaptic activity [9]. BDNF participates in many neurophysiological processes [10]. Angiotensin 1–7, which is produced by ACE-2, increases BDNF levels through the MAS receptor/PI3K/Akt/BDNF pathway. Given the decrease in the activity of ACE-2 receptors in the brain in COVID-19 patients, the level of BDNF may decrease, which causes neurodegeneration [11]. In the present study, we aimed to investigate whether there might be an association between cognitive impairment, which has been observed after mild COVID-19 infection, and serum BDNF levels.
The present study was conducted at Tekirdağ Namık Kemal University Hospital, Department of Psychiatry, between July 1, 2021, and January 1, 2022. The inclusion criteria were as follows: (1) a positive COVID-19 PCR test during the disease period and two negative tests postdisease, (2) having had mild disease according to the WHO's COVID-19 disease severity classification, (3) being between the ages of 18 and 50, (4) having a minimum education of 12 years, (5) having a BMI ≥ 18 and < 30 and (6) volunteering to participate in the study. The exclusion criteria included: (1) having a score above 7 on the Hamilton Depression Scale (HAM-D), (2) having a score of 6 or above on the Hamilton Anxiety Rating Scale (HAM-A), (3) having a psychiatric illness or a previous psychiatric illness and treatment, (4) having an alcohol or substance use disorder or a history of alcohol or substance use, (5) having current neurological disease or a history of neurological disease, (6) being treated with antidepressant, antipsychotic, mood stabilizer, antiepileptic, benzodiazepine and other drugs that may affect neurocognitive test evaluation, (7) presence of a known chronic inflammatory disease, cancer or autoimmune disease, (8) having acute or chronic infectious disease, (9) having a history of head trauma, (10) having a disease that increases intracranial pressure, (11) having a physical disease that affects the main organs of the body or that prevented neurocognitive testing, (12) presence of a defect in visual function that could not be corrected with lenses, (13) diagnosis of color blindness, (14) presence of a known allergy. According to the inclusion and exclusion criteria, 54 patients who had COVID-19 infection participated in the study. Thirty-six age-, sex-, BMI-, education level- and smoking status-matched healthy controls were included in the study. The inclusion criteria for healthy controls were as follows: (1) having no history of COVID-19 infection, (2) being between the ages of 18–50, (4) having a minimum education of 12 years, (5) having a BMI ≥ 18 and < 30 and (6) volunteering to participate in the study. The exclusion criteria for healthy controls were the same as those for the COVID-19 group. All participants were vaccinated with the BNT162b2 mRNA COVID-19 vaccine.
This form was designed based on the literature. The form consisting of a total of 19 questions prepared in order to collect demographic information about the participants in the COVID-19 and healthy control groups and was completed by the researcher for all participants.
The Hamilton Depression Rating Scale (HDRS) was established in 1960. It uses the 5-level rating method of 0 to 4 points. The total score is 0–78, and the depression level can be divided as follows: < 8 means no depression, 8–17 means possible depression, 18–24 means mild to moderate depression and > 24 means severe depression [12].
The HAMA-14 is one of the most commonly used clinician-rated measurements of anxiety in studies of depression. The HAMA-14 is rated from 0 to 4 with general guidelines provided for distinguishing stagewise anxiety severity. It is a reliable and valid measure of the severity of anxiety in depressed patients and has become the standard in this field. A score higher than 7 indicates the presence of anxiety symptoms [13].
All participants in our study were individually administered the Stroop test and Visual Aural Digit Span Test Form B (VADS-B) by a supervised test practitioner to evaluate cognitive function.
The Stroop test was first developed by Stroop in 1935 as a neuropsychological test that measures focused attention, selective attention, response inhibition, resistance to interference and information processing speed in order to assess frontal lobe functions [14]. The reliability and validity study of the Turkish version of the Stroop test was performed by Karakaş et al. in 1999 [15]. The Stroop test consists of 5 cards, which are used as follows: In the 1st part, the subjects are asked to read the names of colors printed in black ink on the 1st card; in the 2nd part, they are asked to read the names of colors printed in colors different from the cards themselves as presented on the 2nd card; in the 3rd part, they are asked to say which color the colored circles are as presented on the 3rd card; in the 4th part, they are asked to say some neutral words printed in different colors; and finally, in the 5th part, they are asked to name the colors of the mismatching words printed in colors different from themselves. In each part, the total time for a subject to read words or say the colors, the number of correct answers, the number of errors and the number of spontaneous corrections are calculated. The Stroop interference score is calculated as the difference of 3 points, which is obtained by subtracting the duration of the 3rd part from that of the 5th part. The reading time of the 1st card with the color names printed in black, that is, the duration of the first part, shows the basic level of reading speed and is calculated as the speed factor [14].
The Visual Aural Digit Span Test Form B (VADS-B) is a neuropsychological test developed by Karakaş et al. based on the Visual Aural Digit Span test developed by Koppitz for use in children in 1977 to measure the attention and short-term memory function of the hippocampus and prefrontal cortex regions of the brain. One of the validity and reliability studies of the VADS-B was conducted by Karakaş et al. in 1995 [16]. The VADS-B is a test in which visual and aural stimuli are given and responses are received both orally and in writing. The VADS-B consists of consecutive number sequences, with the shortest sequence consisting of 2 numbers and the longest sequence consisting of 9 numbers. When the number sequences are repeated incorrectly, the subject is given a second try. This test consists of four subtests: aural oral (AO), visual oral (VO), aural written (AW) and visual written (VW). The VADS-B has a total of 11 points. Four of these scores consist of the basic scores obtained from each subtest, namely, AO, VO, AW and VW, and 6 of them are related to the combined scores of the aural input score (AO + AW), visual input score (VO + VW), oral expression score (AO + VO), written expression score (AW + VW), intrasensory integration score (AO + VW) and intersensory integration score (VO + AW). The total score is calculated as follows: AO + VO + AW + VW. A maximum of 9 points can be obtained for each subtest, a maximum of 18 points for each combined test and a maximum of 36 points in total [14].
Peripheral blood samples (5–8 ml) were collected in a red-capped gel tube between 08:00 and 10:00 in the morning after 8 h of fasting. All peripheral blood samples were centrifuged at 1000 rpm for 15 min to obtain serum, and the obtained serum samples were stored in a deep freezer (− 80 °C). Serum BDNF levels were measured by ELISA. A commercial ELISA kit (Catalog No: E1302Hu) from Bioassay Technology Laboratory (Shanghai Korain Biotech Co., Ltd. Shanghai, China) was used for this measurement. The mass was measured using the sandwich ELISA principle.
Power analysis was used to determine the sufficiency of the sample size for the study. For the comparison of patient and control groups, the Mann‒Whitney U test was performed for two independent samples. Additionally, the normal distribution assumptions were checked by using the Shapiro‒Wilks normality test. In correlation analysis, Spearman's coefficient of correlation was used for non-normally distributed data or ranked data. Otherwise, Pearson's coefficient of correlation can be used for normally distributed data. Statistical analyses were performed using SPSS version 23.0 (SPSS Inc., Chicago, IL, USA). Two-way ANOVA was used to compare serum BDNF levels between groups. Specifically, sex and group were selected as fixed factors, and the serum BDNF value was selected as the dependent variable. A post hoc Tukey test was used for comparisons.
To calculate the power of the study, the Mann‒Whitney U test results were used. The effect size was derived by G*Power statistical software. The sample size of 72 achieved 91.6% power to detect an effect size of 0.83 using a Mann‒Whitney U test with a significance level (alpha) of 0.05. A sample size of 90 was considered, and the power was approximately 96% at the alpha level.
There was no significant difference between the two groups in terms of sociodemographic characteristics and HAM-D and HAM-A scores. The data are shown in Table 1. The COVID-19 clinical characteristics are presented in Table 2. The duration of recovery was found to be as follows: 35 (64.8%) patients had COVID-19 6–12 months before participating in the study (Table 2).
Stroop test word reading spontaneous correction number and reading time, word color saying wrong number, spontaneous correction number and reading time, box color speaking spontaneous correction number and reading time, Stroop interference and speed factor duration were significantly higher in the COVID-19 group than in the control group (p < 0.05). All scores of the VADS-B test were found to be significantly lower in the COVID-19 group than in the control group (p < 0.05). Stroop test and VADS-B test data are shown in Table 3.
The mean serum BDNF level was selected as an independent factor, and sex and group were administered as fixed factors. BDNF levels were found to be 10.92 ± 6.91 ng/ml in the COVID-19 group and 12.83 ± 6.41 ng/ml in the healthy control group. Gender had a significant effect on BDNF levels (F = 12.21; p = 0.008). Two-way ANOVA showed that the serum mean BDNF level was significantly higher in the COVID-19 group than in the control group (F = 12.22; p = 0.044). A comparison of the serum BDNF levels of the two groups is shown in Table 4 (Table 4).
There were no significant correlations between neurocognitive tests and serum BDNF levels in female participants in case group. In male participants, there were significant negative correlations between Stroop Word Reading (number of correct word and reading time), Stroop Saying The Word’s Color (number of correct word), Stroop Saying The Box’s Color (number of correct word and reading time) and speed factor duration and serum BDNF level (Table 5). There were not any correlations between both female and male groups’ neurocognitive tests and serum BDNF level in control group (Table 6). There were not any significant correlations between the scores of HDRS, HAMA-14, times passed after COVID-19 and serum BDNF levels (respectively, r = 0.076, p = 0.637; r = 0.126, p = 0.744; r = 0.214, p = 0.432).
In the present study, the main findings were as followings: Stroop test word reading spontaneous correction number and reading time, word color saying wrong number, spontaneous correction number and reading time, box color speaking spontaneous correction number and reading time, Stroop interference and speed factor duration were significantly higher in the COVID-19 group than in the control group. All scores of the VADS-B test were found to be significantly lower in the COVID-19 group. The mean serum BDNF levels were found to be 10.9 ± 6.9 ng/ml in the COVID-19 group and 12.8 ± 6.4 ng/ml in the healthy control group. Two-way ANOVA showed that the serum mean BDNF level was significantly lower in the COVID-19 group than in the control group. Gender had a significant effect on BDNF levels (F = 12.21; p = 0.008). Several studies have investigated the effects of COVID-19 infection on cognitive function after recovery. In a previous study, 18 men and 11 women who had experienced COVID-19 were assessed, and it was found that cognitive functions were impaired in the field of selective attention three weeks after the disease [17]. In another study in which 97 patients were included, cognitive functions were screened 8 months after COVID-19. It was found that 33% of the patients reported impaired attention, and 27% of them reported a decrease in memory [18]. In a study that evaluated the cognitive function of patients who did not need to be hospitalized due to COVID-19, it was shown that there were decreases in attention and short-term memory function compared to healthy controls [19]. The results of the present study are in line with the literature and indicate a decline in cognitive function, especially in attention and short-term memory. The patients who had mild COVID-19 were evaluated 6 months later in terms of cognitive function, and it was found that cognitive function decreased in these individuals compared to the preepidemic situation [20]. Another study showed that memory, attention, executive functions and language were lower in people who had COVID-19 than in those who had not, and it has been shown that decreased cognitive function was not associated with the severity of the disease [21]. In another study, the cognitive function of people who had COVID-19 was examined 3 months after the infection, and it was reported that one-third of these people had deterioration in cognitive function. However, it was also found that the severity of the disease did not correlate with the deterioration in cognitive function [22]. In a meta-analysis that included 43 studies, it was reported that approximately 20% of people showed cognitive dysfunction for 3 or more months after COVID-19; however, there was no significant association between deterioration of cognitive function and severity of illness [23]. A recent study demonstrated that cognitive dysfunction was more common in people who had severe illness and who needed to stay in the intensive care unit for a longer period of time [24]. Although there are conflicting results about the relationship between the degree of cognitive function and the severity of the disease, our study indicated that cognitive decline can be observed even in young people with mild illness and in people who have had the disease for more than 6 months. There have been few studies that have investigated the role of serum BDNF and cognitive decline in patients with COVID-19 infection. Azoulay et al., showed that lower serum BDNF levels were found in patients with severe disease, and serum BDNF levels returned to normal over time. They also reported that the serum BDNF levels in males were lower than those of females, and thus, it was interpreted that the serum BDNF level could be a prognostic indicator, especially in male patients [25]. Studies have reported that the more severe course of COVID-19 in men may be related to the higher expression of ACE-2 in men [26–28]. In our study, we found that serum BDNF levels were significantly lower in the COVID-19 group, when a two-way ANCOVA model was applied. Sex was shown to have a significant effect on serum BDNF levels. Higher expression levels of ACE-2 in males may be associated with lower levels of serum BDNF in male patients with COVID-19 infection. Although we calculated the sample size for the present study, the small sample size can be considered a limitation. The inclusion of only patients who recovered from mild COVID-19 and the exclusion of patients who recovered from severe to moderate COVID-19 might have resulted in false-negative findings, and this issue is another limitation of the present study. Pro-BDNF is the precursor of mature BDNF and has been reported to have different effects on the etiology of major depressive disorder [29]. We could not measure serum pro-BDNF levels, which is another limitation of the present research. The present study is the first to demonstrate the association between the role of serum BDNF and cognitive decline in patients with COVID-19 infection. Additionally, there is a significant role of male gender in terms of lower BDNF level and cognitive decline. Our results indicated that cognitive decline occurred after recovery and that this decline persisted. Further studies are needed to demonstrate the effects of COVID-19 infection on long-lasting cognitive dysfunction. |
|
PMC9648440 | Shufeng Liu,Charles B. Stauft,Prabhuanand Selvaraj,Prabha Chandrasekaran,Felice D’Agnillo,Chao-Kai Chou,Wells W. Wu,Christopher Z. Lien,Clement A. Meseda,Cyntia L. Pedro,Matthew F. Starost,Jerry P. Weir,Tony T. Wang | Intranasal delivery of a rationally attenuated SARS-CoV-2 is immunogenic and protective in Syrian hamsters | 10-11-2022 | Live attenuated vaccines,SARS-CoV-2 | Few live attenuated severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccines are in pre-clinical or clinical development. We seek to attenuate SARS-CoV-2 (isolate WA1/2020) by removing the polybasic insert within the spike protein and the open reading frames (ORFs) 6–8, and by introducing mutations that abolish non-structural protein 1 (Nsp1)-mediated toxicity. The derived virus (WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A) replicates to 100- to 1000-fold-lower titers than the ancestral virus and induces little lung pathology in both K18-human ACE2 (hACE2) transgenic mice and Syrian hamsters. Immunofluorescence and transcriptomic analyses of infected hamsters confirm that three-pronged genetic modifications attenuate the proinflammatory pathways more than the removal of the polybasic cleavage site alone. Finally, intranasal administration of just 100 PFU of the WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A elicits robust antibody responses in Syrian hamsters and protects against SARS-CoV-2-induced weight loss and pneumonia. As a proof-of-concept study, we demonstrate that live but sufficiently attenuated SARS-CoV-2 vaccines may be attainable by rational design. | Intranasal delivery of a rationally attenuated SARS-CoV-2 is immunogenic and protective in Syrian hamsters
Few live attenuated severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccines are in pre-clinical or clinical development. We seek to attenuate SARS-CoV-2 (isolate WA1/2020) by removing the polybasic insert within the spike protein and the open reading frames (ORFs) 6–8, and by introducing mutations that abolish non-structural protein 1 (Nsp1)-mediated toxicity. The derived virus (WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A) replicates to 100- to 1000-fold-lower titers than the ancestral virus and induces little lung pathology in both K18-human ACE2 (hACE2) transgenic mice and Syrian hamsters. Immunofluorescence and transcriptomic analyses of infected hamsters confirm that three-pronged genetic modifications attenuate the proinflammatory pathways more than the removal of the polybasic cleavage site alone. Finally, intranasal administration of just 100 PFU of the WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A elicits robust antibody responses in Syrian hamsters and protects against SARS-CoV-2-induced weight loss and pneumonia. As a proof-of-concept study, we demonstrate that live but sufficiently attenuated SARS-CoV-2 vaccines may be attainable by rational design.
The rapid development of multiple vaccines has afforded mankind powerful tools to curb the severe coronavirus disease 2019 (COVID-19) pandemic. Among the eleven vaccines granted emergency use by the World Health Organization (WHO), seven of them, including the Pfizer and Moderna mRNA vaccines, three adenovirus-vector-based vaccines (AstraZeneca/Oxford, AstraZeneca/Serum Institute of India, J&J), and two protein-based vaccines (Novavax and Covovax with Novavax formulation), all express SARS-CoV-2 spike protein as the immunogen. The other three inactivated, whole virus vaccines (SinoPharm, Sinovac, and Bharat Biotech) appear to be less immunogenic in inducing neutralizing antibodies. The efficacy of existing vaccines in preventing symptomatic infections, especially against new variants of concern, declines considerably over a period of six months. The stringent storage conditions and requirement of medical supplies to administer the vaccines put additional constraints on the worldwide distribution of some vaccines. For these reasons, new vaccines that are potent, broadly protective, and elicit durable immunity, as well as being easy to administer, store, and transport must be continuously pursued. Live attenuated viral vaccines (LAV) utilize a living but weakened virus as immunogen, and there are many examples of effective LAVs including the measles, mumps, and rubella vaccine, oral polio vaccine, yellow fever virus vaccine, chickenpox vaccine, and one influenza vaccine. LAV causes a real, but often asymptomatic infection in vaccinees, and hence usually elicits both humoral and cellular immune responses. In addition, an intranasally administered LAV will not only avoid needle sticks but may be more effective in eliciting immunity at the mucosal membrane. The latter is especially desirable for prevention of COVID-19 because the human upper respiratory airway tends to be less protected by existing vaccines that are administered intramuscularly. An obvious obstacle to the development of LAVs against SARS-CoV-2 is the safety of the vaccine virus. Multi-layer attenuation of pathogenesis is expected to ensure a LAV does not revert back to virulence. To date, the only LAV that is in clinical development against COVID-19 is a genetically recoded virus with a segment of the viral spike protein being codon-pair deoptimized (clinicaltrials.gov Identifier: NCT04619628). It is worth noting that SARS-CoV-2 attenuation does occur naturally and in cell culture. The most common mechanism of attenuation is the loss of the polybasic insert (PRRA) or the furin cleavage site, which impairs the ability for the virus to infect the lung. SARS-CoV-2 is also known to encode several powerful viral proteins that subvert the host innate immunity, especially interferons and interferon-stimulated genes (ISGs), and hence promote infection. Numerous studies have reported that non-structural protein 1 (Nsp1), and accessory proteins ORF6, 7, and 8 are potent interferon (IFN) antagonists. Nsp1 is also cytopathic in cells of human lung origin. Clinical isolates containing a 382-nucleotide deletion in ORF8 seem to be associated with a milder infection. Taken together, there is a strong rationale to explore further development of attenuated SARS-CoV-2 vaccines and sufficient information currently available to provide a basis for the rational attenuation. Here, we demonstrate a generic strategy to attenuate SARS-CoV-2 through reverse genetics. The derived virus (WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A) replicates to 100- to 1000-fold-lower titers than the ancestral virus, induces robust antibody responses in Syrian hamsters and protects against SARS-CoV-2 challenge.
Although a single mutation may attenuate a virus, a LAV that differs from the wild-type virus by only one or two mutations poses an inherent safety concern due to the possibility of reversion. For that reason, we made the following modifications to the ancestral WA1/2020 viral genome: First, the polybasic insert (PRRA) immediately upstream of the furin cleavage site was removed from the virus. Such a modification abolishes the S1/S2 cleavage of the spike protein and significantly reduces infection of the lung. Second, known IFN antagonists, ORFs6-8, were deleted from the viral genome. Thirdly, a pair of mutations (K164A/H165A) was introduced into the C-terminus of Nsp1 as we previously showed that these mutations significantly reduced cytotoxicity of SARS-CoV-2. The three-pronged genetic modifications are to decrease infection of the lung, reduce inflammation and interferon antagonism, alleviate Nsp1-mediated toxicity. Ultimately, a recombinant virus termed “WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A” was obtained. For comparison purpose, we also generated two other recombinant viruses, WA1-ΔPRRA and WA1-ΔPRRA-ΔORF6-8-Nsp1N128S/K129E, respectively (Fig. 1a). WA1-ΔPRRA only has the polybasic insert removed, whereas WA1-ΔPRRA-ΔORF6-8-Nsp1N128S/K129E has both the polybasic insert and ORFs6-8 deleted and then contains a pair of mutations (N128S/K129E) that did not efficiently abolish Nsp1-mediated cytotoxicity as did K164A/H165A. All recombinant viruses formed plaques in Vero E6 cells and reached titers to 107 pfu/ml (Fig. 1b). For brevity, the three recombinant viruses are also referenced as “ΔPRRA”, “Nsp1-K164A/H165A”, and “Nsp1-N128S/K129E” throughout the text. To monitor the genome stability, recombinant Nsp1-K164A/H165A virus was passaged five times in Vero E6 cells and human lung cell line H1299/hACE2 before viral genome sequencing. Passaging in H1299/hACE2 did not result in any new mutations, whereas passaging in Vero E6 cells led to a few single nucleotide variants (SNVs) outside the designed region of mutations (Supplementary Fig. 1). Sanger sequencing was also performed to detect the presence of K164A/H165A in Nsp1 after passage 5 (Fig. 1c). Overall, no reversion was found, and the genome of recombinant viruses appeared to be stable after cell passages in H1299/hACE2 cells. The three recombinant viruses grew with similar kinetics in A549-hACE2 cells to comparable titers (Fig. 1d). Compared to the ancestral virus (WA1/2020), all three recombinant viruses poorly infected primary human airway cells that were cultured at the air-liquid interface except that ΔPRRA mutant yielded infectious virus after 4 days. Nsp1-K164A/H165A and Nsp1-N128S/K129E displayed nearly no infectivity on primary cells over a period of 5 days (Fig. 1e, f).
To test attenuation in vivo, adult K18-hACE2 transgenic mice were divided into five groups (n = 10/group) and intranasally inoculated with 105 plaque forming unit (PFU) of WA1/2020, ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E or left uninoculated. Weight, survival, and clinical signs of illness were monitored for eight days. Notably, all infected mice succumbed to the infection by day 8 (Fig. 2a–d). Because encephalitis following SARS-CoV-2 infection is known to cause lethality in this model, possible attenuation of the virus in the respiratory tract could have been masked by the lethality caused by encephalitis. To compensate for the limitations of the K18-hACE2 mouse model, we quantified the viral loads in nasal turbinates, lungs, and brains at 2, 4, 6 days post infection (dpi). In nasal turbinates, the median log10-transformed infectious titers peaked by 2 dpi at 5.3 [interquartile range (IQR), 5.0 to 5.7], 3.8 (IQR, 2.6 to 5.7), 3.2 (IQR, 2.2 to 3.7), and 4.3 (IQR, 2.8 to 4.7) for WA1/2020, ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected mice, respectively. Infectious titers subsided to approximately the lower limit of quantification by 4 dpi (Fig. 2e, f). In the lungs of Nsp1-K164A/H165A infected mice, infectious viral titers were ∼2log10 lower compared to WA1/2020-infected animals at 2 and 6 dpi (Fig. 2g–j). The lung viral loads in Nsp1-K164A/H165A infected mice also trended lower than those from ΔPRRA and Nsp1-N128S/K129E groups, reaching statistical significance at 4 and 6 dpi. The viral loads in the brain, however, were largely comparable among the four infected groups except that at 6 dpi the Nsp1-K164A/H165A group had the lowest viral titers (Fig. 2k–m). Interestingly, at 2 dpi, there was very little detectable infectious virus in the brain, but the viral load in the brains rose in delayed kinetics as opposed to that in the respiratory tract. High viral loads in the brains of the Nsp1-K164A/H165A group at 6 dpi were coupled with a lack of pathology in lung tissues. Haematoxylin and eosin (H&E) staining (Fig. 2n–w) found lung lesions in WA1/2020, ΔPRRA, and Nsp1-N128S/K129E infected mice. Generally, there were peribronchiolar and perivascular immune infiltration. By contrast, Nsp1-K164A/H165A infected lungs had less than 1% impacted area with little pathology, which are nearly indistinguishable from the uninfected mice. Altogether, these results demonstrated that the Nsp1-K164A/H165A virus was primarily attenuated in the lower respiratory tract of K18-hACE2 mice but was still neuroinvasive in this highly sensitive mouse model.
Syrian hamsters are highly susceptible to SARS-CoV-2 and have been widely used in COVID-19 research. Given the intrinsic shortcomings of the K18-hACE2 mouse model due to fatal neuroinvasion, we further evaluated the possible attenuation of Nsp1-K164A/H165A in hamsters. To this end, five groups of 6-month-old Syrian hamsters were intranasally inoculated with 104 PFU of each virus or left uninoculated. This inoculum has consistently yielded weight loss, clinical signs, and lung pathology in Syrian hamsters. Shown in Fig. 3a, WA1/2020-infected hamsters showed 18% weight loss on 7 dpi, whereas ΔPRRA, and Nsp1-N128S/K129E groups displayed no more than 5% weight loss over a period of 14 days. Nsp1-K164A/H165A infected animals (n = 8), like the uninfected group, had no weight loss at all through the study. During the first four days following infection, the log10-transformed infectious viral titers in nasal wash samples were measured by a TCID50 assay. Shown in Fig. 3b, the infectious titers from Nsp1-K164A/H165A infected animals were about two log10 lower than that of WA1/2020-infected hamsters at 1 and 2 dpi and trended lower than those from ΔPRRA and Nsp1-N128S/K129E-infected animals. Infectious nasal viral titers from all groups subsided to just above the limit of quantification at 4 dpi. Similarly, subgenomic RNA (sgRNA) titers in nasal turbinates from all groups were just above limit of quantification at 4 dpi (Fig. 3c). Log10-transformed sgRNA copies per ml in lung homogenates were 5.5 (IQR, 5.1 to 6.1), 4.0 (IQR 3.5 to 4.4), 3.5 (IQR 3.3 to 4.3), 4.5 (IQR 4.0 to 4.9), for WA1/2020, ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected hamsters, respectively, at 4 dpi (Fig. 3c); median log10-transformed infectious viral titers in lung homogenates were 7.1 (IQR 7.0 to 7.2), 6.0 (5.8 to 6.1), 4.3 (IQR 3.7 to 5.3), 6.1 (IQR 5.4 to 6.4) for WA1/2020, ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected hamsters, respectively (Fig. 3d). RNAscope revealed that viral RNA was present only along the bronchial epithelium in Nsp1-K164A/H165A-infected hamsters and the amount of staining was much less than the other three infected groups (Supplementary Fig. 2). Overall, viral loads of Nsp1-K164A/H165A-infected hamsters were 100–1000-fold lower than WA1/2020-infected animals and were half-log lower than ΔPRRA and Nsp1-N128S/K129E-infected animals. The lung pathology was subsequently scored. Once again, Nsp1-K164A/H165A-infected hamsters had minimal histopathological changes in the lung at 4 dpi (Fig. 3e). By contrast, WA1/2020 infection induced massive peribronchiolar edema and perivascular immune cell infiltrates, which led to significant consolidation (Fig. 3f–t). ΔPRRA and Nsp1-N128S/K129E groups occasionally had areas where type II hyperplasia and immune infiltration were found. Nsp1-K164A/H165A infected hamsters showed minimal pathological changes in the lung, sometimes indistinguishable from uninfected animals. The pathology of trachea followed the same trend observed in the lung with Nsp1-K164A/H165A infected animals showing only minimal submucosal lymphoplasmacytic infiltrates (Supplementary Fig. 3). None of the infections led to noticeable changes in heart and other critical organs as we have previously reported. Immunofluorescence analyses showed that regions of consolidation in WA1/2020-infected lungs were dominated by Iba1-expressing macrophages (Fig. 4). Consolidated Iba1 staining was not detected or minimal in uninfected, ΔPRRA and Nsp1-K164A/H165A groups while consolidated Iba1 regions were visible in Nsp1-N128S/K129E-infected animals. Prominent staining for viral nucleocapsid was present in alveolar epithelium surrounding consolidated regions and within affected bronchioles in WA1/2020 (Fig. 4a, b). Nucleocapsid staining was limited to bronchiolar epithelium in Nsp1-K164A/H165A group (Fig. 4a), but reached alveolar epithelium in ΔPRRA and Nsp1-N128S/K129E infected animals. Reduced staining for RAGE and ProSPC, markers of type 1 and type 2 epithelial cells, respectively, highlighted the excessive epithelial damage in regions of consolidation (Fig. 4f, g). To further assess changes at the molecular level, we isolated RNA from both nasal turbinates and lung homogenates at 4 dpi and performed RNAseq analyses. In nasal turbinates, comparison between WA1/2020 and ΔPRRA, Nsp1-N128S/K129E and Nsp1-K164A/H165A infected animals showed that WA1/2020 upregulated 34 and downregulated 17 genes in pathways of inflammation, upregulated 33 and downregulated 25 genes of pathways of type I IFN responses, and then upregulated 39 and downregulated 8 genes in pathways of type II IFN responses (Fig. 5). Most noticeably, Nsp1-K164A/H165A infection had least effects on the expressions of proinflammatory markers, such as Mx2, Ifit3, Tlr6, Cxcl10, and Nfkb1 (Fig. 5a). Nsp1-K164A/H165A infection upregulated least numbers of genes of the interferon-alpha and gamma responses (Fig. 5b, c), presumably due to the lowest viral load among all tested groups. In some cases, gene expression profiles of Nsp1-K164A/H165A-infected hamsters were indistinguishable from those of uninfected hamsters. Strikingly, Nsp1-K164A/H165A specifically upregulated genes like Irf8, Tap1, and Stat1, all of which are important for antiviral defense (Fig. 5b). In the lungs, WA1-2020 induced 50 proinflammatory genes, 14 TLR signaling genes, 25 genes of the type I IFN and 37 genes of type II IFN pathways to higher levels than in ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected hamsters. For those genes that were significantly downregulated in WA1/2020-infected animals, their expressions were restored in ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected hamsters (Supplementary Fig. 4). Notably, we did not observe significant differences in terms of gene expression profiles in the lungs between ΔPRRA, Nsp1-K164A/H165A, and Nsp1-N128S/K129E infected animals, which may be a result of limited numbers of samples or an overall very low viral loads in the lungs.
Based on the data obtained from above studies, Nsp1-K164A/H165A appears to be the most attenuated recombinant virus and hence was chosen for subsequent evaluation of immunogenicity and efficacy as a LAV candidate. To this end, adult Syrian hamsters were intranasally immunized with 102, 103, and 104 PFU Nsp1-K164A/H165A. For comparison, we also inoculated hamsters with 104 PFU WA1/2020, ΔPRRA, and Nsp1-N128S/K129E and housed animals until convalescence (Fig. 6a). A single dose of Nsp1-K164A/H165A induced binding and neutralizing antibodies (Fig. 6b, c) to levels that are comparable to those from WA1/2020-infected hamsters at 14 or 28 days after immunization. Interestingly, a dose of 100 PFU was just as potent as a dose of 104 PFU regarding induction of antibodies. Immunized and convalescent hamsters were subsequently challenged with 104 PFU WA1/2020 virus and monitored for 7 days before necropsy. Mock vaccinated hamsters lost more than 15% body weight by day 7 post-challenge (dpc), whereas immunized hamsters from all three dosage groups did not lose any weight (Fig. 6d). At 1- and 2-days post-challenge, infectious viral titers in nasal wash samples collected from immunized animals were 3 to 4 log10 lower than those from unvaccinated but challenged animals (Fig. 6e, f) and largely resolved by 4 dpc (Fig. 6g, h). Infectious viral titers and sgRNA titers in trachea and lungs were frequently below limit of quantification in many of the immunized and convalescent hamsters at 4 and 7 dpc (Fig. 6i–l). Viral loads in the nasal turbinates of the immunized and convalescent hamsters were at least 4 logs lower at 4 dpc and then went undetectable at 7 dpc. Lastly, single-dose immunization with Nsp1-K164A/H165A completely protected hamsters from developing pneumonia upon challenge, with nearly 0% consolidation and no histopathological changes at 4 and 7 dpc (Fig. 7a–c and Supplementary Fig. 5). Interestingly, we did not observe anamnestic antibody response in the Nsp1-K164A/H165A-vaccinated hamsters (Supplementary Fig. 6), likely reflecting the robust protection and minimal viral replication in these animals, as we have found in convalescent animals.
For a pandemic respiratory pathogen like SARS-CoV-2, an important factor to consider when evaluating a LAV is safety, i.e., how pathogenic the vaccine virus is. Naturally, a poorly replicating virus is likely to be less immunogenic. We presented a rational approach to attenuate SARS-CoV-2 and abolish its pathogenicity while preserving immunogenicity. In K18-hACE2 transgenic mice and Syrian hamsters, viral loads of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A infected animals were 100−1000-fold lower in the nose and the lungs than those of WA1/2020-infected animals and were also noticeably lower than two comparators, WA1-ΔPRRA and WA1-ΔPRRA-ΔORF6-8-Nsp1N128S/K129E-infected animals. The attenuation of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A pathogenesis was further confirmed by the absence of lung pathology in infected animals. The mechanism of attenuation is likely threefold: (1) removal of the polybasic insert (PRRA) rendered the virus less infectious in the lung; (2) removal of ORF6-8 and the introduction of Nsp1K164A/H165A further weakened the ability of the virus to antagonize IFNs; (3) K164A/H165A mutations also alleviated Nsp1-mediated cytotoxicity as we have recently reported. In consistence, WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A-infected hamsters showed more pronounced attenuation of proinflammatory genes in the nasal turbinates compared to WA1-ΔPRRA and WA1-ΔPRRA-ΔORF6-8-Nsp1N128S/K129E-infected animals. Infection with WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A was limited to bronchiolar epithelium and hence recruited nearly no macrophages into the lung. As we did not create a recombinant virus that only lacks ORFs6-8, we are unable to quantify the contribution of each modification to the observed attenuation in vivo. Nonetheless, WA1-ΔPRRA-ORF6-8-Nsp1K164A/H165A appears to be the least pathogenic among the three recombinant viruses in animal models in terms of overall tissue viral load and pathology. Splitting attenuating modifications in three regions within the viral genome, rather than concentrating them, is likely to reduce the chances of reversion to virulence through a single recombinational event. Therefore, WA1-ΔPRRA-ORF6-8-Nsp1K164A/H165A may be a safer candidate for a LAV or a challenge virus in human challenge studies than those attenuated SARS-CoV-2 with only one or two-mutation difference from the wild type. Unlike vaccines that express only the spike protein as immunogen, LAVs confer broader and/or more durable protection because the whole organism is recognized by the host immune system. As to the humoral immune response, profiling antibody binding in 40 COVID-19 convalescent patients identified B cell epitopes derived from many viral proteins, including S, M, N, ORF1ab, ORF3a, ORF6, and ORF8. In this study, as low as an inoculum of 100 PFU WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A elicited potent humoral immune response and prevented hamsters from developing lung pathology. The observed protection was accompanied by more than 5-log10 reductions in viral loads in the lung and trachea following challenge. Impressively, immunized animals also displayed over 4-log10 reductions in nasal viral load. It is known that natural infection by SARS-CoV-2 induces both mucosal antibody responses and systemic antibody responses. Secretory immunoglobulin A (IgA) is thought to play a major role in protecting the upper and lower respiratory tract from acute infection. Unfortunately, vaccines that are administered intramuscularly or intradermally potently induce IgG but not much secretory IgA. By contrast, intranasal administration of a vaccine likely induces more potent mucosal immunity. Ongoing efforts in our group are the development of appropriate reagents and assays to reliably determine mucosal antibody titers in hamsters and to evaluate cellular immunity induced by WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A. Future studies are also warranted to directly assess the effectiveness of intranasal delivery of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A against SARS-CoV-2 transmission in comparison to mRNA or protein-based vaccines. We expect that administration of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A will activate broader cellular immunity that is cross-protective against variants of concern than spike-based vaccines because replication of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A offers many more targets to derive T cell epitopes. While this area of research is currently under investigation, if needed, variant-specific LAVs can be rapidly generated using our attenuation strategy. The furin cleavage site, ORFs 6–8, and the Nsp1-K164/H165 residues are conserved among all major SARS-CoV-2 variants of concern. Although a single intranasal administration of WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A protected against SARS-CoV-2-induced pneumonia in Syrian hamsters, we noted some limitations in the study. First, WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A remains neurovirulent in K18-hACE2 mice at an inoculum of 105 PFU. Whether this reflects an intrinsic neurovirulence of the virus or the limitation of the mouse model will need to be investigated. Second, we and others have shown that immunity acquired through natural infection does not lead to sterilizing immunity in nasal cavity. Hence, even if immunization with WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A significantly reduces SARS-CoV-2 infection and pathology, to achieve sterilizing immunity in the upper respiratory tract may not be a realistic goal. Finally, studies are needed to monitor immune responses over time to establish the durability of the protective response after intranasal vaccination with WA1-ΔPRRA-ΔORF6-8-Nsp1K164A/H165A. In summary, our studies demonstrate the value of rationally attenuated SARS-CoV-2 in facilitating the development of new vaccines. Intranasal delivery of the attenuated SARS-CoV-2 induces potent humoral immunity, provides excellent protection, and possibly promotes sterilizing immunity in the lung. Our results support intranasal delivery of rationally attenuated SARS-CoV-2 as a promising platform for preventing COVID-19 and thus warrants further exploration.
Research described here complies with all relevant ethical regulations and has been approved by the US Food and Drug Aministration Institutional Biosafety Committee. All critical reagents are listed in Supplementary Table 1.
Vero E6 cell line (Cat # CRL-1586) was purchased from American Type Cell Collection (ATCC) and cultured in Eagle’s minimal essential medium (MEM) supplemented with 10% fetal bovine serum (Invitrogen) and 1% penicillin/streptomycin and L-glutamine. A549-hACE2 (NR-53821) cells were obtained from BEI Resources and maintained in DMEM supplemented with 5% penicillin and streptomycin, and 10% fetal bovine serum (FBS) at 37 °C with 5% CO2. Lenti-X cell line was purchased from Takarabio (Cat No. 632180) and maintained in DMEM supplemented with 5% penicillin and streptomycin, and 10% fetal bovine serum (FBS) at 37 °C with 5% CO2. EpiAirway cells (AIR-100-HCF) and culturing media were purchased from MatTek. EpiAirway is a ready-to-use, 3D mucociliary tissue model consisting of normal, human-derived tracheal/bronchial epithelial cells cultured at the air-liquid interface (ALI). Cells were cultured in MatTek proprietary media for 2 days prior to usage. Mucus was washed off at the time of infection. The SARS-CoV-2 isolate WA1/2020 (NR-52281, lot 70033175) was obtained from BEI Resources, NIAID, NIH, and had been passed three times on Vero cells and 1 time on Vero E6 cells prior to acquisition. It was further passed once on Vero E6 cells in our lab. The virus has been sequenced and verified to contain no mutation to its original seed virus.
SARS-CoV-2 recombinant virus was generated using a 7-plasmid reverse genetic system which was based on the virus strain (2019-nCoV/USA_WA1/2020) isolated from the first reported SARS-CoV-2 case in the U.S.. The initial 7 plasmids were generous gifts from Dr. P-Y Shi (UTMB). Upon receival, fragment 4 was subsequently subcloned into a low-copy plasmid pSMART LCAmp (Lucigen) to increase stability. To introduce Nsp1N128S/K129E and K164A/H165A mutations, pUC57-CoV2-F1 plasmids containing mutated Nsp1 were first created by using overlap PCR method with the following primers: M13F: gtaaaacgacggccagt N128S/K129Ef: taagaacggtAGTGAGggagctggtggccatagtta N128S/K129E r: caccagctccCTCACTaccgttcttacgaagaagaa K164A/H165Af: aaaactggaacactGCcGCcagcagtggtgttacccgtga K164A/H165Ar: gggtaacaccactgctgGCgGCagtgttccagttttcttgaa NheIr: cacgagcagcctctgatgca PCR fragments were digested by Bgl II/Nhe I and ligated into Bgl II/Nhe I digested F1 plasmid. The spike ΔPRRA mutation was introduced in to pUC57-CoV2-F6 by using overlap PCR with primers: M13F: gtaaaacgacggccagt ΔPRRA-f: actcagactaattctcgtagtgtagctagtcaatc ΔPRRA-r: actagctacactacgagaattagtctgagtctgat BglIIr: cagcatctgcaagtgtcact PCR fragments were digested by Kpn I/Bgl II and ligated into Kpn I/Bgl II digested F6 plasmid. To delete the ORF6-ORF8 region, an overlap PCR was performed using the following primers: Mf: ttaattttagccatggcaga ORF68f: tttgcttgtacagtaaacgaacaaactaaaatgtc ORF68r: ttttagtttgttcgtttactgtacaagcaaagcaa AvrIIr: gaagtccagcttctggccca PCR fragments were digested by Mlu I/Avr II and ligated into Mlu I/Avr II digested pCC1-CoV-2-F7 plasmid. The resulted plasmids were validated by restriction enzyme digestion and Sanger sequencing. In vitro transcription and electroporation were carried following procedures that were detailed elsewhere. To recover the virus, the RNA transcript was electroporated into Vero E6 cells. Virus after passage 1 was titrated by plaque forming assay in Vero E6 cells and verified by deep sequencing.
Adult male outbred Syrian hamsters were previously purchased from Envigo and held at FDA vivarium. All experiments were performed within the biosafety level 3 (BSL-3) suite on the White Oak campus of the U.S. Food and Drug Administration. The animals were implanted subcutaneously with IPTT-300 transponders (BMDS), randomized, and housed 2 per cage in sealed, individually ventilated rat cages (Allentown). Hamsters were fed irradiated 5P76 (Lab Diet) ad lib, housed on autoclaved aspen chip bedding with reverse osmosis-treated water provided in bottles, and all animals were acclimatized at the BSL3 facility for 4–6 days or more prior to the experiments. The study protocol details were approved by the White Oak Consolidated Animal Care and Use Committee and carried out in accordance with the PHS Policy on Humane Care & Use of Laboratory Animals. For challenge studies, adult (6–12 months old) Syrian hamsters were anesthetized with 3–5% isoflurane following procedures as described previously. Intranasal inoculation was done by pipetting 104 PFU or desirable doses of SARS-CoV-2 in 50 µl volume dropwise into the nostrils of the hamster under anesthesia. Following infection, hamsters were monitored daily for clinical signs and weight loss. Nasal wash samples taken on days 1-, 2-, 3-, and 4-day post infection to test for sgRNA by RT-qPCR and infectious virus by TCID50 in Vero E6 cells. Nasal washes were collected by pipetting ~160 µl sterile phosphate-buffered saline into one nostril when hamsters were anesthetized by 3–5% isoflurane. For tissue collection, a subset of hamsters was humanely euthanized by intraperitoneal injection of pentobarbital at 200 mg/kg and lungs for histopathology. Blood collection was performed under anesthesia (3–5% isoflurane) through gingival vein puncture or cardiac puncture when animals were euthanized.
Female adult K18-hACE2 mice (12 weeks ago) were previously purchased from the Jackson laboratory and held at FDA vivarium. All experiments were performed within the biosafety level 3 (BSL-3) suite on the White Oak campus of the U.S. Food and Drug Administration. The study protocol details were approved by the White Oak Consolidated Animal Care and Use Committee and carried out in accordance with the PHS Policy on Humane Care & Use of Laboratory Animals. For infection studies, mice were first anesthetized by 3–5% isoflurane. Intranasal inoculation was done by pipetting 105 PFU SARS-CoV-2 in 50 µl volume dropwise into the nostrils of the mouse. Mice were weighed and observed daily. For tissue collections, mice were euthanized by CO2 overdose on days 2, 4, 6 as necessary.
Procedures as described previously. In brief, 50 μL of SARS-CoV-2 S pseudovirions were pre-incubated with an equal volume of medium containing serum at varying dilutions at room temperature for 1 h, then virus-antibody mixtures were added to Vero E6 cells in a 96-well plate. After a 3 h incubation, the inoculum was replaced with fresh medium. Cells were lysed 48 h later, and luciferase activity was measured using luciferin-containing substrate. Controls included cell only control, virus without any antibody control, and positive control sera. The end-point titers were calculated as the last serum dilution resulting in at least 50% SARS-CoV-2 neutralization. A NIBSC anti-SARS-CoV-2 antibody (20/130) was included as a positive control.
Procedures as described previously. In brief, RNA was extracted from 50 μl NW or 0.1 g tissue homogenates using QIAamp vRNA mini kit or the RNeasy 96 kit (QIAGEN) and eluted with 60 μl of water. 5 μL RNA was used for each reaction in real-time RT-PCR. When graphing the results in Prism 8, values below the limit of quantification (LoQ) were arbitrarily set to half of the LoQ values. Unless otherwise specified, the unit for RNA copies are as presented as Log10 RNA copies/5 μl nasal wash or Log10 RNA copes/0.1 g tissue homogenates,
To prepare sequencing libraries, RNA was first extracted using the Trizol-chloroform method from the lung homogenates and nasal turbinates. The aqueous portion was further purified using RNeasy mini kit (Qiagen, Gaithersburg, MD). RNA quality was assessed using Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA), and the RNA integration numbers (RIN) were all greater than 9. An aliquot (1 μg) of each sample of total RNA was used to prepare sequencing libraries using Illumina Stranded Messenger RNA Prep (ligation based). The cDNA libraries were normalized and loaded onto a NovaSeq 6000 sequencer (Illumina, San Diego, CA) for deep sequencing of paired-end reads of 2 × 100 cycles. The sequencing reads for each sample were mapped to the respective reference genomes of Mesocricetus auratus (BCM_Maur_2.0) by Tophat (v2.1.2). Cufflinks (v2.2.1) was then used to assemble transcripts, estimate abundances and test for differential expression. The sequencing and initial data analysis using Qiagen CLC Genomics Workbench (version 21) was performed by FDA Next Generation Sequencing Core Facility. The sequencing and initial data analysis using Qiagen CLC Genomics Workbench (version 21) was performed by FDA Next Generation Sequencing Core Facility. Raw and processed data have been deposited to NCBI (GEO accession number GSE199922s). Further data analysis was done using R Studio 1.4.1106 (http://www.R-project.org). Heatmaps were constructed using heatmap library. The gene list for signaling pathways was obtained from hallmark gene sets in Molecular Signatures Database (MSigDB). The figures were assembled in Adobe Photoshop. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).
Procedures as described previously. Tissues (hearts, brains, lungs, trachea, and nasal turbinates) were fixed in 10% neural buffered formalin overnight and then processed for paraffin embedding. The 5-μm sections were stained with hematoxylin and eosin for histopathological examinations. Images were scanned using an Aperio ImageScope. For hamster tissues, blinded samples were graded by a licensed pathologist for the following twelve categories: consolidation, alveolar wall thickening, alveolar airway infiltrates, perivascular infiltrates, perivascular edema, type II pneumocyte hyperplasia, atypical pneumocyte hyperplasia, bronchiole mucosal hyperplasia, bronchiole airway infiltrates, proteinaceous fluid, hemorrhage, vasculitis. Grading: 0 = none, 1 = mild, 2 = moderate, 3 = severe. A graph was prepared by summing up the score in each category. Mouse tissues were scored based on the following categories: consolidation, alveolar wall thickening, alveolar airway infiltrates, perivascular Infiltrates, perivascular edema, peribronchiolar infiltrates, type II pneumocyte hyperplasia, necrosis (alveoli and bronchiole), bronchiole mucosal hyperplasia, bronchiole airway infiltrates, proteinaceous fluid, hemorrhage, vasculitis. Grading scale: 0 = none, 1 = mild, 2 = moderate, 3 = severe.
To detect SARS-CoV-2 genomic RNA in FFPE tissues, ISH was performed using the RNAscope 2.5 HD RED kit, a single plex assay method (Advanced Cell Diagnostics; Catalog 322373) according to the manufacturer’s instructions. Briefly, Mm PPIB probe detecting peptidylprolyl isomerase B gene (housekeeping gene) (catalog 313911, positive-control RNA probe), dapB probe detecting dihydrodipicolinate reductase gene from Bacillus subtilis strain SMY (a soil bacterium) (catalog 310043, negative-control RNA probe) and V-nCoV2019-orf1ab (catalog 895661) targeting SARS-CoV-2 positive-sense (genomic) RNA. Tissue sections were deparaffinized with xylene, underwent a series of ethanol washes and peroxidase blocking, and were then heated in kit-provided antigen retrieval buffer and digested by kit-provided proteinase. Sections were exposed to ISH target probes and incubated at 40 °C in a hybridization oven for 2 h. After rinsing, ISH signal was amplified using kit-provided pre-amplifier and amplifier conjugated to alkaline phosphatase and incubated with a fast-red substrate solution for 10 min at room temperature. Sections were then stained with 50% hematoxylin solution followed by 0.02% ammonium water treatment, dried in a 60 °C dry oven, mounted, and stored at 4 °C until image analysis.
Formalin-fixed paraffin-embedded (FFPE) lung sections 4 µm thick were dewaxed, rehydrated, and heat-treated in a microwave oven for 15 min in 10 mM Tris/1 mM EDTA buffer (pH 9.0). After cooling for 30 min at room temperature, heat-retrieved sections were blocked in PBST with 2.5% bovine serum albumin (BSA) for 30 min at RT followed by overnight incubation at 4 °C with primary antibodies in 1% BSA. Primary antibodies used included SARS nucleocapsid protein (1:800, Sino Biologicals, 40143-MM05), prosurfactant protein C (1:200, EMD Millipore, AB3786), Iba1 (1:100, Abcam, ab5076), and RAGE (1:400, Abcam, ab216329). Sections were rinsed and incubated with 1:500 secondary antibodies congujated with Alexa Fluor 488 (A-21206) and Alexa Fluor 647 (A-31571, A-21447) for 1 h at RT (ThermoFisher, Waltham, MA). Nuclei were counterstained with Hoechst 33342. For double labeling experiments, primary antibodies were mixed and incubated overnight at 4 °C. For negative controls, sections were incubated without the primary antibody or mouse and rabbit isotype antibody controls. Sections stained with conjugated secondary antibodies alone showed no specific staining. Whole-slide fluorescence imaging was performed using a Hamamatsu NanoZoomer 2.0-RS whole-slide digital scanner equipped with a ×20 objective and a fluorescence module #L11600. Analysis software NDP.view2 was used for image processing (Hamamatsu Photonics, Japan).
Procedures as described previously. In brief, Vero E6 cells were plated the day before infection into 96-well plates at 1.5 × 104 cells/well. On the day of the experiment, serial dilutions of 20 μl nasal wash samples were made in media and a total of six to eight wells were infected with each serial dilution of the virus. After 48 h incubation, cells were fixed in 4% PFA followed by staining with 0.1% crystal violet. The TCID50 was then calculated using the formula: log(TCID50) = log(do) + log(R) (f + 1). Where do represents the dilution giving a positive well, f is a number derived from the number of positive wells calculated by a moving average, and R is the dilution factor.
Nasal wash samples were 10-fold serially diluted and added to a 24-well plate containing freshly confluent with Vero E6 cells. For tissue samples, entire trachea or nasal turbinates or the left lobe of the lung (~0.2 g) were resuspended in 1 milliliter MEM and homogenized on a Precellys Evolution tissue homogenizer with a Cooling Unit (Bertin). Tissue homogenates were then 10-fold serially diluted and added to Vero E6. After 1 h the mixture was removed and replenished with Tragacanth gum overlay (final concentration 0.3%). Cells were incubated at 37 °C and 5% CO2 for 2 days, then fixed with 4% paraformaldehyde, followed by staining of cells with 0.1% crystal violet in 20% methanol for 5–10 min. The infectious titers were then calculated and plotted as plaque forming units per milliliter (PFU/ml).
ALI cell culture supernatants were 10-fold serially diluted in 96-well plates and dilutions added to 96-well black-well plates for fluorescent focus-forming assays in H1299-hACE2 cells. After 1 h the Tragacanth gum overlay (final concentration 0.3%) was added. Cells were incubated at 37 °C and 5% CO2 for 1 day, then fixed with 4% paraformaldehyde, followed by staining of cells with primary rabbit anti-nucleocapsid Wuhan-1 antibody (custom made by Genscript) overnight followed by secondary anti-rabbit Alexa-488 conjugated antibody and DAPI staining. The infectious titers were then counted using Gen5 software on a Cytation7 machine and calculated and plotted as focus-forming units per milliliter (FFU/ml).
SARS-CoV-2 S and RBD antigens for ELISA were prepared in a baculovirus expression system using procedures as published elsewhere. Briefly, Immunlon 2HB plates were coated with recombinant S or RBD protein at 1 µg/mL overnight at 4 °C. Test serum samples were pre-diluted in assay diluent (PBS containing 0.05% Tween-20 [PBST] and 10% fetal bovine serum), followed by serial two-fold dilutions of each sample in duplicates across the plate. Plates were incubated with the test serum samples for 2 h at 37 °C. After rigorous plate washes in a microplate washer, a secondary antibody (anti-hamster IgG) conjugated to HRP (6060-05, SouthernBiotech, Birmingham, AL) was added to wells at 1:4000 dilution (hamster IgG ELISA). Plates were incubated with secondary antibody for 1 h, washed, and ABTS/H2O2 peroxidase substrate (SeraCare, Gaithersburg, MD) was added to assay wells. After 20 to 30 min at ambient temperature, reactions were stopped with 1% SDS, and OD405 values were captured on the Versamax microplate reader with the Softmax Pro 7 software installed (Molecular Devices, San Jose, CA). The assay endpoint was a mean OD405 of 0.05 for duplicate wells for the full-length S ELISA and a mean OD405 of 0.01 for the RBD ELISA. The reciprocal of the highest serum dilution at which the mean OD405 value averaged ≥0.05 (full-length S ELISA) or ≥0.01 (RBD ELISA) was the IgG titer. To determine the serum IgA antibody titers, each test serum sample was diluted from 1:160. The sandwich Rabbit anti-hamster IgA antibody (sab 3001a, Brookwood Biomedical, Jemison, Alabama) was added at 1:4000 dilution followed by 1:4000 Goat anti-rabbit-HRP (4030-05, Southern Biotech, Birmingham, AL).
One-way ANOVA was used to calculate statistical significance through GraphPad Prism (9.1.2) software for Windows, GraphPad Software, San Diego, California, USA.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Supplementary Information Supplementary Data File Reporting Summary |
|
PMC9648446 | 36357401 | Jinchao Jia,Mengyan Wang,Jianfen Meng,Yuning Ma,Yang Wang,Naijun Miao,Jialin Teng,Dehao Zhu,Hui Shi,Yue Sun,Honglei Liu,Xiaobing Cheng,Yutong Su,Junna Ye,Huihui Chi,Tingting Liu,Zhuochao Zhou,Liyan Wan,Xia Chen,Fan Wang,Hao Zhang,Jingjing Ben,Jing Wang,Chengde Yang,Qiongyi Hu | Ferritin triggers neutrophil extracellular trap-mediated cytokine storm through Msr1 contributing to adult-onset Still’s disease pathogenesis | 10-11-2022 | Neutrophils,Rheumatic diseases,Mechanisms of disease,Inflammatory diseases | Hyperferritinemic syndrome, an overwhelming inflammatory condition, is characterized by high ferritin levels, systemic inflammation and multi-organ dysfunction, but the pathogenic role of ferritin remains largely unknown. Here we show in an animal model that ferritin administration leads to systemic and hepatic inflammation characterized by excessive neutrophil leukocyte infiltration and neutrophil extracellular trap (NET) formation in the liver tissue. Ferritin-induced NET formation depends on the expression of peptidylarginine deiminase 4 and neutrophil elastase and on reactive oxygen species production. Mechanistically, ferritin exposure increases both overall and cell surface expression of Msr1 on neutrophil leukocytes, and also acts as ligand to Msr1 to trigger the NET formation pathway. Depletion of neutrophil leukocytes or ablation of Msr1 protect mice from tissue damage and the hyperinflammatory response, which further confirms the role of Msr1 as ferritin receptor. The relevance of the animal model is underscored by the observation that enhanced NET formation, increased Msr1 expression and signalling on neutrophil leukocytes are also characteristic to adult-onset Still’s disease (AOSD), a typical hyperferritinemic syndrome. Collectively, our findings demonstrate an essential role of ferritin in NET-mediated cytokine storm, and suggest that targeting NETs or Msr1 may benefit AOSD patients. | Ferritin triggers neutrophil extracellular trap-mediated cytokine storm through Msr1 contributing to adult-onset Still’s disease pathogenesis
Hyperferritinemic syndrome, an overwhelming inflammatory condition, is characterized by high ferritin levels, systemic inflammation and multi-organ dysfunction, but the pathogenic role of ferritin remains largely unknown. Here we show in an animal model that ferritin administration leads to systemic and hepatic inflammation characterized by excessive neutrophil leukocyte infiltration and neutrophil extracellular trap (NET) formation in the liver tissue. Ferritin-induced NET formation depends on the expression of peptidylarginine deiminase 4 and neutrophil elastase and on reactive oxygen species production. Mechanistically, ferritin exposure increases both overall and cell surface expression of Msr1 on neutrophil leukocytes, and also acts as ligand to Msr1 to trigger the NET formation pathway. Depletion of neutrophil leukocytes or ablation of Msr1 protect mice from tissue damage and the hyperinflammatory response, which further confirms the role of Msr1 as ferritin receptor. The relevance of the animal model is underscored by the observation that enhanced NET formation, increased Msr1 expression and signalling on neutrophil leukocytes are also characteristic to adult-onset Still’s disease (AOSD), a typical hyperferritinemic syndrome. Collectively, our findings demonstrate an essential role of ferritin in NET-mediated cytokine storm, and suggest that targeting NETs or Msr1 may benefit AOSD patients.
Hyperferritinemic syndrome, an overwhelming inflammatory condition, is characterized by high serum ferritin level and sustained by excessive release of pro-inflammatory cytokines, leading to a cytokine storm. It encompasses four clinical conditions, including adult-onset Still’s disease (AOSD), macrophage activation syndrome (MAS), catastrophic anti-phospholipid syndrome (CAPS), and septic shock, all of which are burdened by multi-organ failure and high mortality rate. Recently, severe coronavirus disease-19 (COVID-19) has been recognized as a fifth member of this spectrum due to similar clinical and laboratory features. Understanding the potential mechanisms triggering the development of a pernicious inflammatory loop in cytokine storm is critical to make targeted therapeutics. Ferritin is an iron storage protein and preserved throughout species. It was previously thought to be an acute-phase protein only as a biomarker of systemic inflammation, but there is accumulating evidence that hyperferritinemia is not only a consequence of the inflammatory process but also a critical part of the pathogenic mechanism. Regulated by pro-inflammatory cytokines, ferritin can further promote expression of pro-inflammatory mediators, providing interesting insights on its association with cytokine storm. To date, the mechanism of how ferritin plays a pathogenic role in facilitating the inflammatory burden in hyperferritinemic syndrome remains to be determined. In fact, ferritin is largely over-expressed in AOSD patients. AOSD, a systemic inflammatory disorder triggered by virus infections, is characterized by spiking fever, evanescent rash, arthralgia or arthritis and hepatosplenomegaly. Liver involvement is common in AOSD, ranging from mild hepatitis to life-threatening hepatic failure. Hallmarks of AOSD are high levels of ferritin and innate immune cell hyperactivation, with neutrophils being the most abundant. The involvement of neutrophils in the pathogenesis of AOSD has been widely acknowledged and investigated. Our previous study has demonstrated the enhanced ability of neutrophil extracellular trap (NET) formation in AOSD patients, which is associated with the leukocyte immunoglobulin-like receptor A3 gene, a genetic risk factor for AOSD. We also reported that circulating NETs can serve as potential biomarkers for identifying liver injury and response to glucocorticoid in AOSD. NETs are extracellular web-like structures that are released by neutrophils and decorated with histones, proteinase, and granular proteins. As potential drivers of amplified inflammatory storm in AOSD, neutrophils crosstalk with macrophages to activate NLRP3 inflammasome by NETs, leading to a vicious loop accelerating inflammatory response. However, the precise mechanisms regulating NET formation in AOSD have not been completely elucidated. Since its discovery, dysregulated NET formation has extensively been investigated in the initiation and progression of autoimmune and autoinflammatory diseases. The molecular pathways to release NETs depend on the diverse inflammatory context of each disease, leading us to hypothesize the potential role of ferritin as a driver of NET formation in hyperferritinemic syndrome. With emerging recognition of the pathogenic role of ferritin as a pro-inflammatory mediator in cytokine storm, there is now a compelling reason to explore ferritin-NETs interplay. A recent study has also demonstrated that macrophage scavenger receptor 1 (Msr1), a pattern recognition receptor to recognize a broad spectrum of ligands, binds ferritin and may facilitate the internalization of ferritin. It has been reported that Msr1 promotes fulminant hepatitis by enhancing NETs. We thus hypothesized that Msr1 might be involved in the ferritin-neutrophil interaction. Although high ferritin levels have always been considered as a sign of hyperinflammation, their pathogenic roles in triggering cytokine storm remains elusive. In the present study, we show that ferritin, as a ligand to scavenger receptor Msr1, promotes inflammatory response by inducing NET formation in a peptidylarginine deiminase 4 (PAD4), neutrophil elastase (NE), and reactive oxygen species (ROS)-dependent way. Overall, these data reveal a mechanism through which high ferritin levels contribute to neutrophil-mediated inflammation and highlight the importance of NETs and Msr1 as therapeutic targets in hyperferritinemic syndrome.
To investigate the pathogenetic role of ferritin in inflammation, we administrated ferritin intraperitoneally to mice. Ferritin-treated mice developed a marked hepatosplenomegaly at 6 h post injection (Fig. 1a). Administration of ferritin also resulted in neutrophilia and subsequently a mild increase in monocyte numbers in peripheral blood (Fig. 1b; Supplementary Fig. 1). Consistent with hyperferritinemic syndrome, ferritin-treated mice developed elevated levels of serum pro-inflammatory cytokines after 3 h, including interleukin (IL)−6 and tumor necrosis factor (TNF)-α. There was also an elevation of IL-10 at 3 h, monocyte chemoattractant protein (MCP)−1 at 3–12 h, and interferon (IFN)-γ at 12 h (Fig. 1c). Abnormal liver function is very common in the spectrum of the hyperferritinemic syndrome. Ferritin-treated mice developed hepatic inflammation with abundant inflammatory cell infiltration (Fig. 1d). In addition, ferritin-treated mice developed mildly increased serum ALT levels, markedly elevated by 6 h and persisted beyond 12 h post injection (Fig. 1e). We also observed an elevation of inflammatory genes (Fig. 1f), suggesting that ferritin contributes to liver injury by enhancing hepatic inflammation. We next explored infiltrated immune cell in the liver of ferritin-treated mice. Neutrophil infiltration in the liver from ferritin-treated mice started at 3 h, reaching the peak at 6 h post injection (Fig. 1g; Supplementary Fig. 2). Particularly, early neutrophil recruitment was accompanied by monocyte infiltration (Fig. 1g). Macrophages began to rise at 12 h and rose significantly at 24 h. Percentage of T cells and B cells was not statistically elevated (Fig. 1g). As confirmed by immunochemistry, livers from mice treated with ferritin for 6 h showed an enhanced neutrophilic infiltration (Fig. 1h; Supplementary Fig. 3). Taken together, these data indicate that ferritin drives systemic and hepatic inflammation, mainly manifesting as excessive release of cytokines, neutrophilia, hepatosplenomegaly, and neutrophil infiltration in the liver.
Given the critical role of NETs in neutrophil-mediated inflammation, we examined neutrophil NET formation in ferritin-treated mice. After ferritin injection, bone marrow-derived neutrophils (BMDN) showed augmented NET formation at 3 h and most robust at 6 h (Supplementary Fig. 4a). We further isolated BMDNs from ferritin-treated mice for 6 h, and stimulated them with LPS or PMA ex vivo. BMDNs from ferritin-treated mice for 6 h exhibited an increased capacity of releasing extracellular DNA (Fig. 2a, b; Supplementary Fig. 4b). Moreover, the liver tissue from ferritin-treated mice for 6 h manifested a significantly increased expression of NET markers, NE and citrullinated histone H3 (citH3) measured by western blotting (WB) (Fig. 2c). In line with this observation, enhanced NET formation was displayed by immunofluorescence staining of NE, citH3, and myeloperoxidase (MPO) (Fig. 2d). To track and identify NETs after ferritin injection, we used intravital imaging experiments with a combination staining approach: a specific neutrophil marker (Ly6G), extracellular DNA dye (Sytox Green) and a NET marker (NE). In ferritin-treated mice, a robust increased recruitment of neutrophils (Ly6G+) into the liver and elevated extracellular DNA and NE were observed (Fig. 2e; Supplementary Movies 1–2). Taken together, these data suggest a contribution of ferritin on NETosis on top of its effects on neutrophil recruitment into the liver.
After confirming that neutrophils were increased in blood and liver of ferritin-treated mice, we investigated their roles in ferritin-mediated hyperinflammatory response. Since mice treated with ferritin for 6 h showed the most severe inflammatory phenotype and highest degree of liver damage, we selected these mice as models for subsequent experiments. We applied anti-Ly6G antibody to deplete neutrophils. The anti-Ly6G antibody efficiently depleted circulating and hepatic neutrophils in ferritin-treated mice, but hepatic monocytes were enhanced (Fig. 3a, b). Next, we assessed the effects of neutrophil depletion on the systemic inflammation and hepatic injury, and found that anti-Ly6G antibody ameliorated the degrees of hepatomegaly in ferritin-treated mice (Fig. 3c). Meanwhile, mice with anti-Ly6G antibody had lower systemic inflammation as demonstrated by diminished serum levels of IL-6, TNF-α, and MCP-1 (Fig. 3d). Serum levels of IL-10 and IFN-γ showed no significant difference (Fig. 3d). Further, mice with neutrophil depletion had reduced NE and citH3 expression and less NET structures in the livers (Fig. 3e, f). haematoxylin and eosin (H&E) staining confirmed that neutrophil depletion reduced ferritin-induced liver inflammatory infiltration (Fig. 3g). Reduction in Il1b, Il6, and Tnfa mRNA levels in the liver tissue homogenates and serum ALT levels were also seen in mice with neutrophil depletion (Fig. 3h, i). Together, these observations suggest that neutrophils are required for ferritin-triggered systemic inflammation and liver injury.
Next, the ability of ferritin to promote NET formation was confirmed in neutrophil from healthy donors. Ferritin with concentration from 10 nM to 1000 nM significantly facilitated NET formation in control neutrophils, with 100 nM ferritin having the best stimulation effect (Fig. 4a). The protein level of citH3 and the activity of NE were gradually increased with ferritin stimulation (Fig. 4b, c). Further, intracellular ROS levels were increased both in ferritin-treated human neutrophils and BMDNs from ferritin-treated mice at 6 h (Fig. 4d). Next, we pre-treated neutrophils with Cl-amidine (a PAD4 inhibitor), Sivelestat (a NE inhibitor) or DPI (a NADPH oxidase inhibitor), prior to stimulation with 100 nM ferritin. All inhibitors abrogated ferritin-induced NET formation as confirmed by cell-free DNA, MPO-DNA and immunofluorescence staining of NET structures (Fig. 4e, f). After demonstrating the indispensable roles of PAD4, NE, and ROS in ferritin-induced NET formation, we assessed the effects of these inhibitors in mice. Cl-amidine, sivelestat, and DPI did not affect the number of circulating neutrophils, but ameliorated hepatic neutrophil infiltration and the degree of hepatomegaly in ferritin-treated mice (Fig. 4g, h; Supplementary Fig. 5a). Decreased serum levels of IL-6, TNF-α, and MCP-1 were also observed (Fig. 4i; Supplementary Fig. 5b). Importantly, Cl-amidine, sivelestat, and DPI significantly blocked NET formation in the livers determined by WB analysis and intravital imaging (Fig. 4j, k). Meanwhile, liver inflammation cells infiltration, mRNA levels of Il1b, Il6, and Tnfa, and serum ALT levels were significantly attenuated (Fig. 4l–n). To confirm the hypothesis that PAD4, NE and ROS-mediated NET formation contributed to ferritin-induced inflammation, we transplanted BM from wild type (WT), Padi4−/−, Elane−/− or Cybb−/− mice into WT mice and gave them ferritin injection (Fig. 5a). As expected, BMDNs from mice transplanted with Padi4−/−, Elane−/− or Cybb−/− BM displayed lower levels of NET release after ferritin injection (Fig. 5b). We also observed less NET deposition in the livers of mice transplanted with Padi4−/−, Elane−/− or Cybb−/− BM after ferritin injection (Fig. 5c, d). Deletion of these genes resulted in decreased peripheral neutrophil frequency and hepatic neutrophil infiltration (Fig. 5e; Supplementary Fig. 6a), reduced hepatomegaly (Fig. 5f), and suppression of serum cytokine levels and liver inflammation (Fig. 5g–j; Supplementary Fig. 6b). In summary, these results indicate that ferritin promotes NET formation and inflammatory response in a PAD4, NE, and ROS-dependent way.
To further investigate the molecular changes of neutrophil in ferritin-treated mice, we performed RNA sequencing (RNA-seq) of BMDNs. By gene ontology analysis, we found genes involved in inflammatory response, such as innate immune response and cytokine production, were enriched in the BMDNs from ferritin-treated mice (Fig. 6a). We then focused on the expression levels of potential ferritin receptors and found Msr1 was significantly increased (Fig. 6b). This change was confirmed in the FACS-sorted BMDNs by quantitative real-time PCR (qRT-PCR) (Fig. 6c; Supplementary Fig. 7a). WB confirmed the augmented protein level of Msr1 in BMDNs (Fig. 6e). The mRNA and protein levels of Msr1 in the livers were also significantly increased (Fig. 6d, e). Flow cytometry revealed that CD204+ neutrophils were increased both in the peripheral blood and liver in mice with ferritin treatment (Fig. 6f; Supplementary Fig. 7b, c). Similarly, ferritin stimulation led to enhanced Msr1 expression on cell surface in human neutrophils (Fig. 6g; Supplementary Fig. 7d). Fucoidan, a Msr1 antagonistic ligand, reduced the ability of human neutrophils to release NETs in response to ferritin (Fig. 6h, i). In line with this result, ferritin promoted neutrophil to generate NETs from WT mice but not Msr1−/− mice (Fig. 6j). Msr1−/− mice were then exposed to systemic ferritin, and NET formation of BDMNs and NET infiltration within the liver were measured. A significant reduction in the capacity to form NETs spontaneously was observed, and the ability to generate NET in response to LPS and PMA was also attenuated (Fig. 6k, l). In addition, a dramatic reduction in the NET infiltration within the liver was validated by WB and confocal intravital imaging (Fig. 6m, n). Collectively, these results reveal that Msr1, as a receptor of ferritin on neutrophils, is critical for ferritin-induced NET formation.
We then investigated the role of Msr1 in ferritin-induced inflammation in vivo. Though Msr1 ablation did not affect neutrophil number in blood (Fig. 7a), Msr1−/− mice with ferritin treatment displayed less neutrophil infiltration in the liver and lower degrees of hepatomegaly compared with WT mice (Fig. 7b, c). Besides, serum levels of IL-6, TNF-α, and MCP-1 were significantly attenuated in Msr1−/− mice than WT mice, while serum levels of IL-10 and IFN-γ were not affected (Fig. 7d). Consistently, a dramatic reduction in the inflammatory cell infiltration within the liver of ferritin-treated Msr1−/− mice was observed by H&E staining, a result supported by qRT-PCR, where a significant diminished expression levels of Il1b, Il6, and Tnfa were detected (Fig. 7e, f). In line with this observation, Msr1 deficiency led to attenuated liver damage with lower ALT activity (Fig. 7g). Because of the important role of mitogen-activated protein kinases (MAPK) and Akt in the initiation of NETosis, we examined the expression of these kinases in human neutrophils upon ferritin stimulation. Significant enhanced levels of ERK, JNK, and p38 phosphorylation were observed in ferritin-treated neutrophils, but not Akt phosphorylation (Supplementary Fig. 8), suggesting activating MAPK pathways in human neutrophils in response to ferritin. Notably, BMDNs from ferritin-treated Msr1−/− mice displayed lower levels of MAPKs phosphorylation compared with ferritin-treated WT mice (Fig. 7h). Collectively, these results indicate that Msr1 on neutrophils is critical for ferritin-triggered hyperinflammatory conditions and liver injury.
AOSD, an overwhelming systemic inflammatory disorder, is characterized by high serum ferritin and neutrophil hyperactivation with enhanced NET formation, we then investigated whether ferritin could play a pathogenic role in AOSD through NET formation. A positive correlation was observed between ferritin and circulating NETs, including serum cell-free DNA, citH3-DNA, MPO-DNA, and NE-DNA levels (Fig. 8a). To further determine whether serum ferritin could mediate NET release, neutrophils from healthy controls were treated with sera from 3 patients with active AOSD with ferritin either absorbed away (after absorption) or not absorbed (before absorption). The capacity of sera to promote NET formation was significantly attenuated after absorption of ferritin (Fig. 8b). To obtain evidence of NET formation in the liver of AOSD patients, we analyzed the liver biopsy from an AOSD patient with ongoing liver injury. H&E staining revealed mild hepatitis with neutrophilic infiltration and immunofluorescence analysis demonstrated the deposition of NETs in areas of neutrophilic infiltration (Fig. 8c). Next, we explored the potential of ferritin-Msr1 pathway to promote NET formation in AOSD neutrophils. We observed that the mRNA and protein levels of Msr1 as well as Msr1 expression on neutrophils’ membrane were significantly increased (Fig. 8d–f; Supplementary Fig. 9). In AOSD patients, neutrophils showed increased basal level of NET formation compared with that in health controls (Fig. 8g). And neutrophils in response to ferritin showed a slight increase in NET formation, probably due to overactivated neutrophils in AOSD (Fig. 8h). Furthermore, using Msr1 antagonistic ligand, NET release was significantly reduced (Fig. 8h, i), suggesting the cardinal role of Msr1 in ferritin-NETosis in patients with AOSD.
Due to the catastrophic outbreak of COVID-19, the fifth type of hyperferritinemic syndrome, much attention has been paid on the pathogenetic role of ferritin in cytokine storm sustained by a vicious pro-inflammatory loop. Using murine models and clinical samples, we comprehensively assess the pro-inflammatory effect of ferritin. We show here that ferritin activates neutrophils to form NETs and contribute to cytokine storm and liver inflammation. Ferritin activates Msr1 of neutrophils to facilitate ERK, JNK, and p38 activation, thus enhancing neutrophil infiltration in the liver with abundant NET formation. Importantly, we also reveal that treatment with Msr1 inhibitor or genetic loss of Mrs1 result in a significant reduction in neutrophil infiltration in the liver with ameliorated systemic inflammatory response. Cytokine storm was first used to describe the pathogenesis of graft-versus-host disease (GVHD) and subsequently demonstrated to be associated with various infectious, autoimmune, and inflammatory diseases. Cytokine storm is characterized by increased production of IL-1β, IL-6, IL-10, IFN-γ, TNF-α, and other cytokines. These inflammatory mediators activate immune system and lead to a life-threatening uncontrollable inflammation. However, the understanding of cytokine storm is still in the early stage. Recently, COVID-19, a virus-induced respiratory disease has brought attention to the cytokine storm. In severe COVID-19 patients, a hyperinflammatory status with a massive release of pro-inflammatory cytokines is proved, which shows similarity to hyperferritinemic syndrome. Interestingly, neutrophilia and hyperferritinemia predict poor outcomes in patients with severe COVID-19. Thus, some studies have considered severe COVID-19 as a fifth member of the spectrum of hyperferritinemic syndrome. AOSD, triggered by viral infections, also belongs to hyperferritinemic syndrome. In our previous study, a significant higher serum level of ferritin was observed in active AOSD patients in comparison with severe COVID-19. Ferritin was traditionally regarded as a conserved protein with the mere function of iron storage. However, increasing evidence supports the idea that high circulating ferritin may not only reflect an acute-phase response, but also play a critical role in inflammatory response. It has been revealed that ferritin regulates NF-κB activation and subsequent expression of pro-inflammatory molecules in an iron-independent way. In addition, the deletion of ferritin heavy chain ameliorates the inflammatory burden in the model of sepsis, including reduction of IL-1β, IL-6, IL-12, and IFN-γ and improves survival. Moreover, ferritin can induce Toll-like receptor (TLR) 9 expression and other TLRs in a leukemic cell line. Herein, we confirmed the pathogenetic role of ferritin in vivo. We found systemic inflammation in ferritin-treated mice, characterized by cytokine storm and liver damage, which is similar to clinical manifestations of the spectrum of hyperferritinemic syndrome. AOSD, a rare systemic autoinflammatory disorder, is typically characterized by hyperferritinemia and cytokine storm. After the first trigger, there is an amplification of inflammation activated by innate immune system, leading to uncontrollable inflammatory cascade reaction. Our previous study has revealed an enhanced NET formation to link neutrophils and macrophages. It is noteworthy that NETs can facilitate the production of inflammatory mediators and further be enhanced by these mediators, leading to a vicious uncontrollable, inflammatory loop. Although it has yet to be determined whether NETs contribute to the amplified inflammatory process in AOSD patients, there is accumulating evidence to indicate inflammatory cytokines such as IL-1β, IL-18, and IL-6 in the AOSD milieu, which can interact with NETs. Indeed, a NETs-cytokine loop exists in various disease, including COVID-19, atherosclerosis and systemic lupus erythematosus (SLE). Despite recent advances in exploring the role of neutrophil and NET in the pathophysiology of AOSD, little is known about underlying mechanism. Herein, our observation that high ferritin levels correlates with increased circulating NETs levels in patients with AOSD and ferritin promotes neutrophils to release NETs in an animal model and human primary cells implies that ferritin may play an important role in the initiation of NET formation. Although we have not tested the importance of ferritin for NET formation in COVID-19, the ferritin-NETs-cytokine storm loop may be validated in the future since enhanced NET formation and hyperferritinemic state are features of patients with severe COVID-19. Considering that many endogenous pathways are involved in NETosis, we investigated the underlying molecular mechanisms of ferritin-induced NET formation. Our previous studies have already shown that the generation of ROS is needed for NET formation in AOSD. Besides, several studies have suggested that PAD4 inhibitor and NE inhibitor prevent NETosis in human neutrophils and in mice. In contrast, it has been noted that NETosis induced by different physiological stimuli is very diverse in the engaged pathways. Indeed, granulocyte-macrophage colony-stimulating factor (GM-CSF) and TNF could induce NETosis in a ROS-independent but PAD4-dependent way. In this regard, our study analyzed these biological processes in ferritin-stimulated neutrophils and revealed that ROS, NE, and PAD4 are essential molecules for ferritin-induced NET formation. Furthermore, MAPKs and Akt are two important signalling pathways that have been closely linked to NETosis. This study shows that ERK, JNK, and p38 are activated, but Akt is not affected in neutrophils after ferritin stimulation. The receptor of ferritin on neutrophils has been marginally addressed in the past decades. Recently, Msr1 was shown to interact directly with extracellular ferritin. Msr1, a membrane-bound scavenger receptor on macrophages and monocytes that recognize various targets, exerts a regulatory function in inflammatory response and cytokine production. In a murine model of fulminant hepatitis, Msr1 acts as an inflammatory accelerator through activating neutrophils and promoting the release of NETs, supporting the important pathogenetic role of Msr1 in neutrophilic inflammation. However, the physiological effect of their interaction remains unknown. In our current work, we reveal that Msr1 is expressed on neutrophils from AOSD patients and ferritin-treated mice, and accompanied by p38, ERK, and JNK pathway activation. Using pharmacological and genetic approaches, Msr1 is found to be indispensable for ferritin-induced NET release and subsequent inflammatory response. The mechanism of how ferritin is internalized by Msr1 will be investigated in our further study. Liver involvement is very common in AOSD. The most frequent manifestation of AOSD is mild hepatitis, but liver failure with hepatic necrosis can be life-threatening in several cases. A major histologic finding of the liver from AOSD is the infiltration of neutrophils. Our previous study has shown that neutrophil-derived lipocalin-2 could serve as a potential biomarker to identify liver injury of AOSD and evaluate the severity of liver dysfunction in AOSD patients. Excessive activation of neutrophils and NETs are believed to promote liver injury in many other liver diseases. In virus-induced fulminant hepatitis, enhanced NETosis resulted in complement activation and subsequent cytokine production. In addition, it has been described that NETs exacerbate inflammatory cascades and sterile inflammation in liver ischemia and reperfusion injury. Decreased levels of NET clearance also contributed to liver injury in alcoholic liver disease. In our present study, we observed that ferritin treatment induced hepatic neutrophilic inflammation, which was ameliorated by neutrophil depletion and suppression of NET formation. This indicates a dominant role of NETs as inflammatory mediators in hyperferritinemia-related hepatic inflammation. There are several limitations in our study. First, the ferritin used in human and mouse experiments is from equine spleen, which is composed of 24 subunits in variable ratios of heavy and light chain encoded by the ferritin heavy chain (FTH) and ferritin light chain (FTL) genes, respectively. We did not test other sources of ferritin, such as human/mouse ferritin or heavy/light chain ferritin, for functional validation. Second, we demonstrated the pathogenic role of ferritin in an animal model and to some extent in patients of AOSD. Additional research is needed to verify the link between ferritin and neutrophils in other hyperferritinemic conditions. In conclusion, our findings demonstrate the underlying relationship between hyperferritinemia and NET formation. Ferritin induces the release of NETs in a Msr1-dependent way, which then contributes to systemic and hepatic inflammation. Accordingly, abolishing the NETs or Msr1 could abrogate the ferritin-induced hyperinflammatory process. Our study highlights the important role of ferritin-Msr1-NETs pathway in the overwhelming inflammatory response, serving as a therapeutic target against the spectrum of hyperferritinemic syndrome.
Human biological samples were obtained under a protocol approved by the Institutional Research Ethics Committee of Ruijin Hospital (ID: 2016–62), Shanghai, China. All biological samples from patients and healthy donors were collected after obtaining informed consent from all participants. All experimental animal protocols described in this study were approved by the Animal Care Committee of Shanghai Jiao Tong University School of Medicine.
A total of 64 AOSD patients (45 active and 19 in active AOSD patients) admitted to the Department of Rheumatology and Immunology, Ruijin Hospital from May 2017 to December 2018 were consecutively included in the present study, and serum samples were collected from all participants. All patients fulfilled Yamaguchi’s criteria after exclusion of those with infectious, neoplastic and autoimmune disorders. All serum samples were stored at −80 °C immediately after collection. The AOSD disease activity of each patient was assessed using a modified Pouchot score. The biological samples of AOSD patients and healthy donors were obtained under a protocol approved by the Institutional Research Ethics Committee of Ruijin Hospital (ID: 2016–62), Shanghai, China., and all the participants provided informed consent. Supplementary Table 1 shows the main characteristics of AOSD patients at the time of the blood sampling.
Female, 8–12 weeks WT FVB/n (#215), and C57BL/6 (#219) mice were purchased from Vital River Laboratories (Beijing, China). Msr1-deficient (Msr1−/−) mice on a C57BL/6 background were provided from Prof. Jingjing Ben in Nanjing Medical University (Nanjing, China), which were purchased from Jackson Laboratory (#006096, RRID: IMSR_JAX: 006096). Padi4−/−, Elane−/− and Cybb−/− mice were obtained from Shanghai Model organisms (Padi4−/−: #NM-KO-190334, Elane−/−: #NM-KO-201544, Cybb−/−: #NM-KO-18031). Animals were maintained under pathogen-free conditions and housed with no more than five animals per cage under a 12 h light/dark cycle with free access to mouse chow and water, ambient temperature 22–24 °C and humidity 50–70%. All experiments were performed on sex-and 8 to 12-week-old age-matched animals. Ferritin (F4503, sterile-filtered, from equine spleen, composed of heavy chains and light chains, Sigma-Aldrich, USA) was injected intraperitoneally (i.p.) to WT FVB/n mice at a dose of 60 μg/g of body weight. After treatment for 3, 6, 12 and 24 h, mice were anesthetized with isoflurane inhalation for body weight measurement and blood collection, and then euthanized by 2.5% chloral hydrate (0.1 mL/10 g) and rapid cervical dislocation for liver and spleen collection. The size of liver and spleen were measured. The blood and serum were harvested for flow cytometry and cytokine analysis. Liver samples were collected, fixed in 4% paraformaldehyde and embedded in paraffin. Additional liver samples were stored at −80 °C. In some experiments, to deplete neutrophil, rat anti-Ly6G antibody (clone 1A8, BE0075, BioXcell, 200 μg/mouse, i.p.) was given 24 h and 2 h prior to ferritin injection. Rat IgG2a isotype control (BE0089, BioXcell) was administered in the same way. To inhibit NET formation, Cl-amidine (20 mg/kg, HY-100574A, MCE, China, i.p.) was given 24 h and 1 h before ferritin injection, DPI (1 mg/kg, D2926, Sigma-Aldrich, USA, i.p.) was given 0.5 h before ferritin injection and sivelestat (50 mg/kg, HY-17443, MCE, China, i.p.) was given 1 h after ferritin injection. For exploring the effect of Msr1 in ferritin-NETs pathway, female, 8–12 weeks Msr1-deficient (Msr1−/−) C57BL/6 mice were injected with ferritin (60 μg/g) intraperitoneally. Female, 8–12 weeks WT C57BL/6 mice were used as controls. All experimental protocols described in this study were approved by the Animal Care Committee of Shanghai Jiao Tong University School of Medicine.
For performing bone marrow transplants (BMTs), the recipient mice (C57BL/6 wild-type mice, female, 8 weeks old) were lethally irradiated with 10 Gy from a caesium gamma source. Donor bone marrow cells (5 × 106 cells) obtained from WT, Padi4−/−, Elane−/− and Cybb−/− donor mice (female, 8–12 weeks old) were intravenous injected into the recipient mice (after irradiation). Four weeks after BMT, the recipient mice were injected with ferritin (60 μg/g) intraperitoneally for further experiments.
Briefly, heparinized blood from AOSD patients and healthy controls was isolated by density gradient centrifugation on Polymorphprep (AS1114683, Axis-Shield, Dundee, UK) for 40 min at 400 × g without braking. The median layer containing neutrophils/red blood cells (RBCs) was transferred to a fresh tube. The neutrophil/RBC pellet was suspended in RBC lysis buffer (Servicebio) and neutrophils were washed in sterile PBS and suspended. Neutrophils (1 × 106 cells/mL) were cultured in RPMI 1640 (Hyclone) supplemented with 10% fetal bovine serum (FBS) and were stimulated with PMA (20 nM, P1585, Sigma-Aldrich) or ferritin (10–1000 nM, F4503, Sigma-Aldrich) for 3.5 h at 37 °C. In some assays, to inhibit NET formation, neutrophils were pre-treated with PAD4 inhibitor Cl-amidine (10 μM, HY-100574A, MCE, China), DPI (25 μM, D2926, Sigma-Aldrich, USA), sivelestat (20 μM, HY-17443, MCE, China) or fucoidan (100 μg/mL, 20357, Cayman chemical company, MI). In some experiments, to determine whether ferritin in plasma could mediate NETosis or not, we absorbed ferritin protein away from sera samples by incubating with ELISA plates from ferritin ELISA kit (CSB-E05187h, CUSABIO) twice for 2 h at 37 °C for no ferritin blockers or neutralizing antibodies available in the current market. The sera before and after absorption were collected from 3 AOSD patients. 10% sera were added to stimulate the neutrophils from healthy control (HC) for 3.5 h at 37 °C.
BMDNs were isolated by density gradient centrifugation using Histopaque 1119 (11191, Sigma-Aldrich) and Histopaque 1077 (10771, Sigma-Aldrich). Total bone marrow cells were collected from tibias and femurs and the RBCs were lysed. Mature neutrophils were purified by centrifugation for 30 min at 845 × g without braking on a Histopaque 1119 and Histopaque 1077. The neutrophils were collected at the interface of the Histopaque 1119 and Histopaque 1077. Mice neutrophils (1 × 106 cells/mL) cultured in RPMI 1640 supplemented with 10% FBS and were stimulated with ferritin (100 nM, F4503, Sigma-Aldrich), LPS (100 ng/mL, L4391, Sigma-Aldrich) or PMA (20 nM, P1585, Sigma-Aldrich) for 3.5 h at 37 °C.
Cell-free DNA was quantified using the Quant-iT PicoGreen double-stranded DNA (dsDNA) assay kit (P11496, Invitrogen) according to the manufacturer’s instructions. 10% serum or cell culture supernatants was added per well, followed by incubation for 10 min. NE-DNA, MPO-DNA, and citH3-DNA complexes were quantified using the Quant-iT PicoGreen. As the capturing antibodies, anti-citH3 (1:1000, ab5103, Abcam), anti-NE (1:2000, ab68672, Abcam) or anti-MPO monoclonal antibodies (1:1000, ab25989, Abcam) were coated onto 96-well microtiter plates overnight at 4 °C. After blocking in 1% BSA for 90 min at room temperature, 10% serum or cell culture supernatants was added per well, followed by incubation overnight at 4 °C. PicoGreen was added to detect cell-free DNA and NET-DNA complexes.
After stimulation, neutrophils were fixed with 4% paraformaldehyde. Protein staining was performed using a rabbit polyclonal anti-citH3 antibody (1:200, ab5103, Abcam), a mouse monoclonal anti-NE antibody (1:50, sc-55549, Santa Cruz), and a goat monoclonal anti-MPO antibody (1:200, AF3667, R&D) overnight at 4 °C. After three washes, appropriate fluorochrome-conjugated secondary antibodies (1:200, Alexa Fluor 594-conjugated rabbit anti-mouse IgG, 33912ES60; 1:200, Alexa Fluor 488-conjugated goat Anti-rabbit IgG, 33106ES60; 1:200, Alexa Fluor 647-conjugated rabbit anti-goat IgG, 33713ES60; all from YEASEN, Shanghai, China) were applied for 1 h incubation at room temperature. DNA was stained with Hoechst 33342 (1:2000, H3570, Invitrogen) for 5 min. After three washes, images were obtained using FV3000 confocal system (Olympus). For quantification, total extracellular DNA generated by cultured neutrophils was digested with 30 mU/mL micrococcal nuclease (MNase, Thermo Fisher Scientific) for 10 min at 37 °C, and then stopped with 5 mM EDTA. Cell-free DNA and MPO-DNA complex in the supernatants was quantified by PicoGreen. For BMDNs, NET formation was confirmed by visualization using 0.3 μM SYTOX green (S7020, Invitrogen, USA) and images were captured using an Olympus microscope (IX73).
Mice were bled from the retro-orbital plexus after isoflurane anaesthesia. 50 μl blood was collected into an EDTA-containing tube. The samples were analyzed using a SYSMEX Hematology Analyzer (pocH-100iVD, Sysmex, Japan).
Serum was collected from whole blood by centrifugation at 845 × g for 10 min. Serum ALT measurement was performed using a commercially available kit (700260, Cayman chemical company, MI). 150 μl of substrate, 20 μl of cofactor and 20 μl of positive control or sample were added to the wells. After incubation for 15 min at 37 °C, 20 μl of ALT initiator was added and the absorbance at 340 nm was measured for 10 min at 37 °C. The ALT value was determined by calculating the slope of the absorbance curve. Serum IL-6, IL-10, TNF-α, IFN-γ, MCP-1 and IL-12p70 were measured with Cytokine Bead Array (552364, BD Biosciences, USA). 50 μl of the standard dilutions or samples were incubated with capture beads and detection reagent for 2 h at room temperature. Assays were performed by a FACS Canto II cytometer (BD).
Liver tissue sections were stained with H&E staining. To detect neutrophils, macrophages and lymphocytes in the liver tissue, paraffin-embedded mouse liver sections were stained by antibodies against Ly6G (1:1000, GB11229, Servicebio, Wuhan, China), F4/80 (1:1000, GB11027, Servicebio, Wuhan, China), CD3 (1:1000, GB11014, Servicebio, Wuhan, China) and B220 (1:4000, GB11066, Servicebio, Wuhan, China). To detect NET formation in the liver tissue, the sections were incubated with anti-citH3 (1:200, ab5103, Abcam), anti-NE (1:50, sc-55549, Santa Cruz) and anti-MPO (1:200, AF3667, R&D). 4′,6-Diamidino-2-phenylindole (2 μg/ml, DAPI, Servicebio, Wuhan, China) was used to detect DNA. Finally, slides were visualized using an Olympus microscope (IX73, Tokyo, Japan).
Gradient-centrifuged BMDNs from PBS- and ferritin-treated mice were used for total RNA isolation. Oligo(dT)-attached magnetic beads-purified mRNA was fragmented into pieces at appropriate temperature. cDNA was generated by random hexamer-primed reverse transcription. Afterwards, RNA Index Adapters and A-Tailing Mix were added to end repair. The cDNA fragments were amplified by PCR, and products were purified using Ampure XP Beads. The double-strand PCR products were denatured and circularized by the splint oligo sequence to construct the final library. Then the single-strand circle DNA was formatted as the final library. The final library was amplified by phi29 to produce DNA nanoball (DNB) with more than 300 copies of one molecule. DNBs were loaded into patterned nanoarray and single end 50 bases reads were generated on BGIseq500 platform (BGI-Shenzhen, China). HISAT2 (v2.0.4) was used to map the clean reads to the genome. Bowtie2 (v2.2.5) was applied to align the clean reads to the reference coding gene set. RSEM (v1.2.12) was used to calculate the expression level of gene. Differentially expressed genes (DEGs) with a fold change >2 and p value < 0.05 were determined using DESeq2(v1.4.5).
Total RNA was extracted using Trizol reagent following manufacturer’s instructions (9109, Takara, Japan), and reverse-transcribed into cDNA using PrimeScript™ RT Reagent Kit (RR036, Takara). qRT-PCR was performed with SYBR Green (B21703, Bimake, China). The relative expression levels of mRNA were normalized against β-actin (mouse) or GAPDH (human). Specific primers of mouse Ly6G, IL-1β, IL-6, IL-10, TNF-α, EMR1, PPARγ, IFN-γ, ARG1, TGF-β, CD163, CD206, iNOS, Msr1, Scara5, Timd2, Tfrc and human Msr1 were used. Primer sequences were listed in Supplementary Table 2.
Intrahepatic and blood leukocytes were stained for 15 min at 4 °C using fluorescently labeled antibodies: PerCP-conjugated anti-mouse CD45 (30-F11 clone, 1:100, 557235, BD), FITC-conjugated anti-mouse CD11b (M1/70 clone, 1:100, 101206, Biolegend), PE-Cy7-conjugated anti-mouse Ly6G (1A8 clone, 1:100, 560601, BD), APC-conjugated anti-mouse Ly6C (HK1.4 clone, 1:100, 128016, BD), FITC-conjugated anti-mouse CD3 (17A2 clone, 1:100, 561798, BD), PE-Cy7-conjugated anti-mouse CD4 (RM4-5 clone, 1:100, 552775, BD), APC-H7-conjugated anti-mouse CD8a (53-6.7 clone, 1:100, 560182, BD), APC-conjugated anti-mouse CD19 (1D3 clone, 1:100, 550992, BD), PE-conjugated anti-mouse CD49b (DX5 clone, 1:100, 553858, BD), PE-conjugated anti-mouse F4/80 (BM8 clone, 1:100, 123110, Biolegend) and Alexa Fluor 647-conjugated anti-mouse Msr1 (2F8 clone, 1:100, MCA1322A647, AbD Serotec, NC, USA). Human blood was stained with PE-conjugated anti-human CD11b (ICRF44 clone, 1:100, 555388, BD), PE-Cy7-conjugated anti-human CD66b (G10F5 clone, 1:100, 305116, Biolegend) and APC-conjugated anti-human Msr1 (7C9C20 clone, 1:100, 371905, Biolegend). All assays were performed by a FACS Canto II cytometer (BD). For cell sorting, FACSAria was used. Data were analyzed using FlowJo software (Tree Star, Inc., Ashland, OR).
DCFH-DA (1:1000, Beyotime Institute of Biotechnology, Shanghai, China) was used to detect intercellular ROS according to the manufacturer’s instructions. The 5 × 105 cells in a final volume of 500 μl were incubated for 20 min with 10 μM DCFH-DA. The cells were washed with serum-free RPMI 1640 for 3 times. Flow cytometry was used for quantitative analysis. Data were analyzed using FlowJo software (Tree Star, Inc., Ashland, OR).
NE activity was quantified using neutrophil elastase activity assay kit (600610, Cayman chemical). Standards or cell culture supernatants from human neutrophils treated with ferritin were plated in a 96-well plate. 10 μl of neutrophil elastase substrate was added to each well, and the plate was incubated at 37 °C for 1.5 h. Fluorescence was quantified at excitation and emission wavelengths of 485 nm and 525 nm, respectively.
Fluorescence imaging of NETs and neutrophils was performed with intravital imaging analysis. Mice were anaesthetized by an initial intraperitoneal injection of avertin (100 mg/kg). The tail vein was catheterized to allow delivery of fluorescent probes and to maintain anaesthesia as required. A heating pad was used to maintain the temperature of the mice at 37 °C. The exposed liver was bathed in normal saline and cover slipped. Neutrophils were visualized by injection of PE anti-mouse Ly6G (2.5 μg, 1A8 clone, 127608, Biolegend) 10 min before intravital imaging. Then NETs were visualized by co-staining of extracellular DNA (Sytox green, 5 μM, S7020, Invitrogen) and neutrophil elastase (Alexa Fluor 647 anti-NE antibodies, 1.6 μg, sc-55549 AF647, Santa Cruz). Images and videos were acquired using Olympus FV3000 inverted microscope with a Galvano scanner and 20×/NA 0.75 UPLANSAPO objective lenses. Sytox green, PE, and Alexa Fluor 647 were irradiated using 488, 561 and 640 nm laser lines, respectively. 20 min videos were recorded every 20 s after ferritin injection for 6 h. Fiji software on Image J (v1.53c, Bio-Rad, USA) was used to create movies in 6 h control group and 6 h ferritin group.
Liver biopsy from patients with AOSD (n = 1) was analyzed for NET formation by immunofluorescence using anti-citH3 (1:200, ab5103, Abcam) and anti-MPO (1:1000, ab25989, Abcam). The liver biopsy of healthy control (n = 1) is obtained from a patient with haemangioma during surgery with normal liver histology, which could be used as the normal liver tissue according to previous studies.
Human neutrophils, mouse livers, and BMDNs were lysed in RIPA lysis buffer (Beyotime Institute of Biotechnology, Shanghai, China) containing protease inhibitor cocktails (Roche Diagnostics, Mannheim, Germany) and phosSTOP (Roche Diagnostics, Mannheim, Germany). The protein lysates (20–40 μg protein) were separated on 10% SDS/PAGE gel and transferred onto polyvinylidene fluoride membrane. Membranes were incubated with primary antibodies against rabbit anti-ERK (1:1000, 137F5 clone, 4695, CST), rabbit anti-phospho-ERK (1:1000, D13.14.4E clone, 4370, CST), rabbit anti-JNK (1:1000, 56G8 clone, 9258, CST), rabbit anti-phospho-JNK (1:1000, 81E11 clone, 4668, CST), rabbit anti-p38 (1:1000, D13E1 clone, 8690, CST), rabbit anti-phospho-p38 (1:1000, D3F9 clone, 4511, CST), rabbit anti-AKT (1:1000, C67E7 clone, 4691, CST), rabbit anti-phospho-AKT (1:1000, C31E5E clone, 2965, CST), rabbit anti-Msr1 (1:1000, ab151707, Abcam), rabbit anti-NE (1:1000, ab68672, Abcam) and rabbit anti-citH3 (1:1000, ab5103, Abcam) overnight at 4 °C followed by HRP-conjugated anti-rabbit IgG (1:5000, 7074 S, CST) or HRP-conjugated anti-mouse IgG (1:5000, L3032, Signalway Antibody, Maryland, USA), and the signals were detected by ECL assays (WBKLS0500, Millipore, USA). Anti-GAPDH (1:1000, AF1186, Beyotime Institute of Biotechnology, Shanghai, China) or β-actin antibody (1:1000, 3700 S, CST) was used as an internal control. Bands were quantitated using Image J (v1.48 & v1.53c, Bio-Rad, USA), and results are expressed as fold change relative to the internal control.
All data were statistically analyzed using the SPSS version 20.0 (SPSS Inc., Chicago, IL, USA) and Graphpad Prism v8.0 software. Quantitative data are expressed as the means ± SEM (standard error of the mean) or means ± SD (standard deviation) as indicated. Data with a Gaussian distribution was analyzed using an unpaired two-sided t-test, one-way or two-way analysis of variance (ANOVA), while nonparametric data were assessed using the Mann–Whitney U test or Wilcoxon rank-sum test. Bonferroni post hoc tests were used to compare all pairs of treatment groups when the overall p value was <0.05. All tests were two-sided, and p values < 0.05 were considered statistically significant.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Movie 1 Supplementary Movie 2 Reporting Summary |
PMC9648449 | 36357786 | Wei Pang,Ying Lu,Yan-Bo Zhao,Fan Shen,Chang-Fa Fan,Qian Wang,Wen-Qiang He,Xiao-Yan He,Ze-Kai Li,Tao-Tao Chen,Cui-Xian Yang,You-Zhi Li,Si-Xuan Xiao,Zu-Jiang Zhao,Xu-Sheng Huang,Rong-Hua Luo,Liu-Meng Yang,Mi Zhang,Xing-Qi Dong,Ming-Hua Li,Xiao-Li Feng,Qing-Cui Zhou,Wang Qu,Shibo Jiang,Songying Ouyang,Yong-Tang Zheng | A variant-proof SARS-CoV-2 vaccine targeting HR1 domain in S2 subunit of spike protein | 10-11-2022 | Mechanisms of disease,Molecular modelling | The emerging SARS-CoV-2 variants, commonly with many mutations in S1 subunit of spike (S) protein are weakening the efficacy of the current vaccines and antibody therapeutics. This calls for the variant-proof SARS-CoV-2 vaccines targeting the more conserved regions in S protein. Here, we designed a recombinant subunit vaccine, HR121, targeting the conserved HR1 domain in S2 subunit of S protein. HR121 consisting of HR1–linker1–HR2–linker2–HR1, is conformationally and functionally analogous to the HR1 domain present in the fusion intermediate conformation of S2 subunit. Immunization with HR121 in rabbits and rhesus macaques elicited highly potent cross-neutralizing antibodies against SARS-CoV-2 and its variants, particularly Omicron sublineages. Vaccination with HR121 achieved near-full protections against prototype SARS-CoV-2 infection in hACE2 transgenic mice, Syrian golden hamsters and rhesus macaques, and effective protection against Omicron BA.2 infection in Syrian golden hamsters. This study demonstrates that HR121 is a promising candidate of variant-proof SARS-CoV-2 vaccine with a novel conserved target in the S2 subunit for application against current and future SARS-CoV-2 variants. | A variant-proof SARS-CoV-2 vaccine targeting HR1 domain in S2 subunit of spike protein
The emerging SARS-CoV-2 variants, commonly with many mutations in S1 subunit of spike (S) protein are weakening the efficacy of the current vaccines and antibody therapeutics. This calls for the variant-proof SARS-CoV-2 vaccines targeting the more conserved regions in S protein. Here, we designed a recombinant subunit vaccine, HR121, targeting the conserved HR1 domain in S2 subunit of S protein. HR121 consisting of HR1–linker1–HR2–linker2–HR1, is conformationally and functionally analogous to the HR1 domain present in the fusion intermediate conformation of S2 subunit. Immunization with HR121 in rabbits and rhesus macaques elicited highly potent cross-neutralizing antibodies against SARS-CoV-2 and its variants, particularly Omicron sublineages. Vaccination with HR121 achieved near-full protections against prototype SARS-CoV-2 infection in hACE2 transgenic mice, Syrian golden hamsters and rhesus macaques, and effective protection against Omicron BA.2 infection in Syrian golden hamsters. This study demonstrates that HR121 is a promising candidate of variant-proof SARS-CoV-2 vaccine with a novel conserved target in the S2 subunit for application against current and future SARS-CoV-2 variants.
SARS-CoV-2 and its emerging variants of concern (VOCs), particularly Omicron sublineages with strong immune evasion and transmission, pose substantial challenges to the control of COVID-19 pandemic, thus calling for the development of variant-proof SARS-CoV-2 or pan-sarbecovirus vaccines. The step of SARS-CoV-2 entry into host cells is a main target for the development of vaccines and therapeutic approaches. The spike (S) protein mediates SARS-CoV-2 entry by binding its S1 subunit to the host receptor angiotensin-converting enzyme 2 (ACE2), and subsequently promoting viral and cellular membrane fusion by its S2 subunit, leading to the release of viral genome into the cytoplasm. The S1 subunit, particularly its receptor-binding domain (RBD) and N-terminal domain (NTD), induces dominant neutralizing antibody (nAb) production in the host and serves as a prime antigen for vaccine design. However, selective pressure from the host acts on the S1 subunit in a manner that has increased the number of mutations in an equally growing number of new variants. This “domino” effect is steadily weakening the efficiency of some current antibodies and vaccines, and leading to constantly and possibly breakthrough infections in vaccinated individuals from some variants, such as Delta, Omicron BA.1 and its sublineages, including BA.2, BA.3, BA.4, and BA.5, as well as other emerging recombinant or hybrid variants. The S2 subunit, unlike S1 subunit, is buried inside the S protein in a prefusion conformation, and induces few nAbs against SARS-CoV-2 after viral infections or post vaccinations with its full amino acid sequence. The S2 subunit contains two important domains, heptad repeat 1 (HR1) and heptad repeat 2 (HR2), which are highly conserved among coronaviruses. According to the currently accepted membrane fusion model of class I enveloped viruses, such as HIV-1, influenza virus, and coronaviruses, fusion occurs when the hydrophobic fusion peptides (FPs) in the S2 homotrimer are inserted into the host cell membrane. Consequently, HR1 and HR2 adjacent to FP are instantly exposed to form a “fusion intermediate” or “prehairpin intermediate” conformation. Subsequently, three HR2 peptides move backward in an antiparallel manner and fold into the three surface grooves of the HR1 trimeric α-helical inner core, thus forming an irreversible six-α-helical bundle (6-HB) structure that maintains close contact between viral and cellular membranes and promotes membrane fusion. In this process, some HR2-derived peptides, binding to the HR1 trimer, and rare monoclonal antibodies (mAbs) isolated from COVID-19 convalescent individuals, with high affinities to HR2, can interfere with the conformational transition of the HR domains from fusion intermediate to post-fusion structure of 6-HB, thus demonstrating broad-spectrum antiviral activities against multiple SARS-CoV-2 variants, coronaviruses, and even HIV-1. Therefore, the conserved HR1 and HR2 domains, present in the fusion intermediate conformation of the S2 subunit, may serve as potential targets for vaccine development. However, previous reports on HIV-1 and influenza viruses suggested that the fusion intermediate was transient and unstable, and its structure in enveloped viruses has not been resolved so far, making it difficult to design an immunogen capable of mimicking its conformation and evoking highly active nAbs in vivo. For instance, using core regions of HR1 and HR2, a lead study in HIV-1 had designed a 5-helix protein by linking three truncated HR1s and two truncated HR2s together (consisting of HR1–HR2–HR1–HR2–HR1). This protein lacked an HR2 helix; therefore, one surface groove in the HR1 trimeric inner core was unoccupied and could serve as HR1-based antigenic epitopes. However, the mAb targeting this vacancy only demonstrated weak anti-HIV-1 activities at subnanomolar ranges. We also showed that the 5-helix protein from human respiratory syncytial virus (hRSV) only elicited weak nAb titers to hRSV. Besides, some other studies on HIV-1 have developed some HR1-based trimers consisting of three different truncated HR1s, such as N35CCG-N13, (CCIZN36)3, (CCIPN36)3, and N46FdFc, but the nAbs induced by these proteins were proved similarly weak and unsatisfactory. In this study, we provided a new strategy to develop SARS-CoV-2 vaccines. We designed a recombinant protein HR121 from SARS-CoV-2, which could highly mimic the conformation of the HR1 trimeric inner core in the fusion intermediate of the S2 subunit. Immunization with HR121 induced potent broad-spectrum nAbs against SARS-CoV-2 and its main variants, including the current pandemic Omicron sublineages BA.1, BA.2, BA.3, BA.4/5. Furthermore, vaccination with this protein provided almost complete protection against the prototype SARS-CoV-2 (hereafter as SARS-CoV-2) infection in hACE2 transgenic mice, Syrian golden hamsters and rhesus macaques, and effective protection against Omicron BA.2 variant challenge in golden hamsters. Thus, this study demonstrates that the conserved HR1 domain in the S2 subunit, mimicking the fusion intermediate conformation, can serve as a new target for the development of variant-proof SARS-CoV-2 or pan-sarbecovirus vaccines.
In our previous studies, using the truncated HR1 and HR2 sequences from HIV-1, hRSV, and SARS-CoV, we have separately designed three small recombinant HR121 proteins as fusion inhibitors against these viruses. These three proteins, with molecular weights of ~13 kDa, exhibited antiviral activities at nM–μM ranges, and showed interactions with HR2, implying that the motif of HR121 could mimic the function of HR1-based fusion intermediate. In this study, to improve the immunogenicity of HR121 and develop it as a novel antibody-based vaccine, we used the full amino acid sequences of HR domains derived from S2 subunit of SARS-CoV-2 Wuhan-Hu-1 strain, and linked HR1–HR2–HR1 together to produce a much larger recombinant protein, SARS-CoV-2-HR121 (abbreviated as HR121) (Fig. 1a). Initially, we attempted to design an immunogen of trimeric HR121, in which three HR1s and three HR2s form an HR1–HR2 trimer (6-HB scaffold), while three free HR1s would aggregate together, thereby mimicking the conformation of the HR1 trimeric inner core in the fusion intermediate, as predicted for the other three viruses using computer models. HR121 was abundantly expressed in Escherichia coli strain BL21 (DE3) in a soluble state, and was easily purified from lytic cellular supernatant by one step of elution on a Superdex 200 increase column (Fig. 1b). Every liter of the cultured cells could yield 10–20 mg of high-purity (> 95%) protein. The molecular weight of the HR121 monomer was determined to be 22 kDa by its amino acid sequence, which was confirmed by SDS-PAGE. However, it was unexpectedly observed to be a dimer in PBS by Native-PAGE (Fig. 1b). X-ray crystal diffraction in 2.41-Å resolution demonstrated that HR121 dimerized into a stable asymmetric 6-HB conformation with a rod-like shape (115 Å length and 25 Å diameter) (Fig. 1c, d). The HR121 dimer (or HR121) consists of a parallel HR1 tetrameric coiled-coil helix (the four HR1s were termed α1–α4), packed with two antiparallel HR2 helices (Supplementary information, Video S1, Fig. S1a, b). Each α-helix in HR121 shares a typical seven-amino acid wheel of 3–4 heptad repeats (the positions are denoted by letters “a” to “g”). The three internal HR1s (α1, α3, and α4) interact with each other via hydrophobic residues located at the “a” and “d” positions. Furthermore, they bind two exterior HR2s and one HR1 (α2) packed at their three hydrophobic surface grooves using residues located at the “e” and “g” positions, commonly found in the 6-helix coiled-coil proteins. Compared with the post-fusion structure of SARS-CoV-2 6-HB (PDB code: 6LXT), HR121 displays a very similar conformation and amino acid residue interaction, except that a parallel HR1 (α2) in HR121 replaces an antiparallel HR2 in 6-HB, via its residues located at the “a” and “e” positions interacting with residues located at the “e” and “g” positions in the other two HR1s (α4 and α3), respectively (Supplementary information, Figs. S1b, c, S2a–d). This results in a 0.5-coil dislocation in this HR1 (α2) helix, compared to the other three HR1s (α1, α3, and α4) (Supplementary information, Fig. S1d). As such, the four HR1 helices in HR121 combine to form two sets of trimeric HR1s. The inner HR1 trimer (α1, α3 and α4) are filled with two HR2s and one HR1 (α2) in its surface hydrophobic grooves (Fig. 1c). The outer HR1 trimer (α2, α3 and α4) also bind two HR2s and one HR1 (α1), but its two surface grooves formed by two adjacent HR1s (between α2 and α3, α2 and α4) remain unoccupied (Fig. 1d). These two exposed surface grooves show higher charged surfaces than the counterparts in the HR1 trimer of SARS-CoV-2 6-HB post-fusion structure, explaining the high solubility of HR121 (Supplementary information, Fig. S2e, f). Notably, these two vacancies may provide two possible binding sites for exogenous HR2s, and are, therefore, expected to create efficient antigenic epitopes to elicit nAbs targeting the HR1 trimeric core in the S2 fusion intermediate. To verify the above conjecture, we first explored the affinity between HR121 and HR2 or HR1. Both HR1 and HR2 were expressed in E. coli BL21 containing pET-30a vector, thus each of them containing a redundant 52 amino acid sequence from the vector. Using a GST pull-down assay, we observed that HR121 selectively binds to HR2, but not HR1 (Fig. 1e), implying that HR121 is functionally analogous to the HR1 trimer inner core in the fusion intermediate conformation, and could interact with its counterpart HR2. Using surface plasmon resonance (Biacore), we confirmed HR2 binding to HR121. The potent interaction between these two proteins had a dissociation constant (KD) of 0.686 nM, an association rate constant (Ka) of 5.48 × 105 M−1 s−1, and a dissociation rate constant (Kd) of 3.76 × 10−4 s−1 (Fig. 1f). These values are comparable to the affinity between RBD and human ACE2 (hACE2). However, HR1 could not bind to HR121 at the concentration of 200 nM (data not shown). Circular dichroism (CD) spectrum also showed that HR121 exhibited a salient α-helix character, while HR2 alone showed a random structure, and HR1 alone showed a partly α-helix structure. When HR121 was mixed with either HR1 or HR2, it interacted with HR2 to form a more stable α-helix structure with typical double spectral minima at 208 nm and 222 nm. However, HR121 did not interact with HR1, and their mixture exhibited a less α-helical structure than that of HR121 (Fig. 1g). Together, our purified HR121 showed a high affinity to HR2, mostly mimicked the conformation, and resembled the function of the HR1 trimeric inner core in the fusion intermediate.
To assess the immunogenicity of HR121, we formulated HR121 with Freund’s adjuvant, and subcutaneously injected the combination into four rabbits three times at three-week intervals. Complete Freund’s adjuvant (CFA) was used in the prime immunization, and incomplete Freund’s adjuvant (IFA) was used in the two boosts (the same procedure was used in the mouse and hamster immunizations). Sera were collected two weeks after the last immunization. Two adjuvant-immunized rabbits were used as mock controls (Fig. 2a). We observed that HR121 evoked homogeneously high endpoint titers of autologous IgG antibodies (geometric mean: 1.3 × 107), slightly lower geometric mean titers (GMTs) to 6-HB scaffold (5.4 × 106) and HR1 (2.4 × 106), and weak antibodies to HR2 (3.0 × 105) in the rabbit sera (Fig. 2b). By serially diluting sera on cell culture plates, we observed that the nAbs evoked by HR121 were highly inhibitory, preventing VSV pseudotyped SARS-CoV-2 env (Wuhan-Hu-1 strain) entry into 293T-hACE2 cells and authentic SARS-CoV-2 prototype strain replication in human pulmonary alveolar epithelial cells (HPAEpiCs), with a geometric mean of 50% neutralization titer (NT50) = 1.7 × 104 and 1.0 × 104, respectively (Fig. 2c; Supplementary information, Fig. S3a, c). The corresponding IgGs purified from the sera also demonstrated high neutralizing activities against the pseudovirus and authentic SARS-CoV-2, with geometric mean half-maximal inhibitory concentration (IC50) of 346 ng/mL and 1312 ng/mL, respectively (Fig. 2d; Supplementary information, Fig. S3b, d). To determine whether HR12 protein can also be used as a vaccine antigen to elicit nAbs against SARS-CoV-2, we immunized three rabbits with HR12 protein (the 6HB scaffold) using the same procedure as that for immunization with HR121 (Fig. 2e). Interestingly, we found that HR12 could induce higher endpoint titers of HR12-binding IgGs (geometric mean: 4.6 × 107), ~7.5-fold higher than that of sera from HR121-immunized rabbits (geometric mean: 5.4 × 106) (Fig. 2f). The antibodies evoked by HR12 showed low neutralizing activity against the prototype SARS-CoV-2 pseudovirus with a GMT value of 4.8 × 102, which was 34.4-fold lower than that of HR121 (GMT: 1.7 × 104) (Fig. 2g; Supplementary information, Fig. S3a). These results suggest that HR12 can elicit high titer HR12 binding antibodies (bAbs) with weak SARS-CoV-2 neutralizing activity and that HR12 protein is not a good vaccine antigen. Among the antibodies to HR121, those targeting the two unoccupied surface grooves formed in the outer HR1 trimer of HR121 (Fig. 1d), thus inhibiting HR121 binding to HR2, would play the main antiviral roles. Using a competitive enzyme-linked immunosorbent assay (ELSIA) previously reported, we found that both rabbit anti-HR121 sera and IgG antibodies could potently inhibit HR2 binding to HR121, with geometric mean NT50 and IC50 at 2.3 × 103 and 663 ng/mL, respectively (Fig. 2i, j; Supplementary information, Fig. S3e, f), which were consistent with their antiviral activities (Fig. 2c, d; Supplementary information, Fig. S3a–d). These results provide a proof of principle that the nAbs evoked by HR121 mainly target HR2 binding to the HR1 trimer, thus blocking membrane fusion and viral entry into host cells. In human COVID-19 convalescent and vaccinated populations, rare nAbs targeting HR1 and HR2 domains in the S2 subunit were isolated. Here, we detected moderate levels of anti-HR1, -HR2, and -HR12 antibodies in the sera of these individuals (Fig. 2h), which is consistent with the results of a previous report. However, none of their sera could inhibit the binding between HR2 and HR121 in the competitive ELISA (Fig. 2i; Supplementary information, Fig. S3g, h), implying that few nAbs targeting membrane fusion could be generated from these populations. These results suggest that HR121 is a unique vaccine candidate that evokes nAbs targeting the HR1 domain in the S2 fusion intermediate, whereas none of the currently reported COVID-19 vaccines target the viral and cellular membrane fusion steps. Because anti-HR121 sera and IgGs from the four rabbits showed similar activity against pseudotyped or authentic SARS-CoV-2, we pooled the sera and assessed their antiviral spectrum using a VSV pseudotyped SARS-CoV-2 spike system. The pooled sera exhibited high dose-dependent neutralizing activities against a series of SARS-CoV-2 variants, including 7 important point mutants (S477N, E484K, N439K, A222V, K417N, D839Y, and D614G) that decrease the efficacy of nAbs, and 12 historically pandemic variants generated from SARS-CoV-2-infected populations, as well as the current globally emerging Omicron BA.1 and its sublineages BA.2, BA.3, and BA.4/5 (BA.4 and BA.5 share the same spike). The NT50s of the rabbit sera ranged from 6.0 × 103 to 3.6 × 104 against the 7 point mutants and 1.8 × 103 to 4.3 × 104 against the 12 previous pandemic variants. Notably, the anti-serum potently neutralized the 6 main VOCs and variants of interest (VOIs), with NT50s of 2.9 × 104 to B.1.1.7 (Alpha), 3.5 × 103 to B.1.351 (Beta), 1.9 × 103 to B.1.1.28 (Gamma), 1.5 × 104 to B1.617.2 (Delta), 6.3 × 103 to C.37 (Lambda), and 4.3 × 103 to B.1.621 (Mu). The serum also neutralized the circulating variant B.1.1.529 (Omicron BA.1) at a titer of 6.6 × 103, and its current sublineages BA.2, BA.3, and BA.4/5 at titers of 2.6 × 103, 2.0 × 103, and 5.0 × 103, respectively (Fig. 2k). These data suggest that our HR121 is a promising immunogen that induces high nAb titers and exhibits broad activity against the main ancestral and current SARS-CoV-2 variants entering host cells. Previous studies have demonstrated that the nAbs in the convalescent sera from SARS-CoV-2-infected hamsters or monkeys exhibited protective roles in these animals against subsequent viral rechallenge, and passive administration of the convalescent sera to naïve hamsters or monkeys effectively suppressed the viral replication in their lung tissues. To further evaluate the in vivo antiviral role of nAbs in the rabbit sera, we injected the purified IgG from the rabbit sera intraperitoneally into hACE2 transgenic mice and Syrian golden hamsters at a dose of 5 mg IgGs/20 g body weight, and estimated the viral genomic RNAs (gRNAs) and subgenomic RNAs (sgRNAs) in their lung tissues 3 days post SARS-CoV-2 challenge using real-time qRT-PCR assays. The gRNA is a marker of viral particles, and sgRNA is an indicator of viral replication. After injecting anti-HR121 IgGs, no gRNAs were detected in 6/6 mice and 7/9 hamsters. In the remaining 2 hamsters, the sgRNAs were not detectable (Supplementary information, Fig. S4), suggesting that the residual gRNAs in the lung tissues of these two hamsters were inactive. This result indicates that the nAbs evoked by HR121 can completely block SARS-CoV-2 replication in vivo.
To evaluate antibody production, we subcutaneously injected three groups of BALB/c mice with 2 μg, 10 μg, and 50 μg of HR121 in Freund’s adjuvant (CFA/IFA), respectively. Immunizations were performed three times at two-week intervals, and sera were collected 7 days after each dose for antibody examination (Fig. 3a). Mice immunized with the adjuvant and PBS were used as controls. There were no remarkable differences in anti-HR121 IgG production among the administered HR121 doses. The first dose induced low HR121 IgG titers (GMTs ranging from 2.2 × 102 to 6.8 × 102); antibodies increased sharply after the second dose (GMTs ranging from 4.3 × 105 to 9.0 × 105), and peaked after the third dose (GMTs ranging from 1.0 × 106 to 1.6 × 106). Notably, even 2 months after the third immunization, HR121 IgG titers remained stable, only decreasing ~4-fold in the 2 μg and 10 μg dose groups, and ~8-fold in the 50 μg dose group (Fig. 3b). We also evaluated nAb production by immunization with 10 μg dose of HR121 at a two-week interval. After the third immunization, the geometric mean NT50 against authentic SARS-CoV-2 reached 4.0 × 103 (Fig. 3c). Three months after the third immunization, the mice in this group were euthanized and their splenocytes were isolated and stimulated with a pool of 15-amino-acid overlapped peptides covering the full HR1 sequence. In the HR121-immunized mice, an increased level of cytokine interferon gamma (IFNγ) was observed using enzyme-linked immune absorbent spot (ELISpot) assays (Fig. 3d). These data suggested that even after three months of the last immunization, these mice still maintained a strong cellular immune response. To evaluate the number of antibody-producing cells in the HR121-immunized mice, we cultured their splenocytes without stimulation and found an increased secretion of anti-HR121 IgGs even 90 days after the last immunization by ELISpot (Fig. 3e), which was consistent with the robust and sustained antibody responses detected in the sera, suggesting that the sustained plasma cell and memory B cell responses were elicited in the HR121-immunized mice. We also immunized BALB/c mice with 10 μg of HR121 at three-week intervals (Fig. 3f), and observed antibodies 2.3-fold lower (GMT: 6.2 × 105) than those generated at two-week intervals (Fig. 3g). Together, these data suggest that HR121 is a strong immunogen, consistent with the results of previously reported RBD-HR vaccines, in which the HR1 and HR2 were linked together as a 6-HB scaffold to enhance both humoral and cellular immune responses in the vaccine-immunized mice.
Thereafter, we selected a 10 μg dose of HR121 to vaccinate hACE2 transgenic mice. Using the same vaccination procedure as that in the BALB/c mouse experiment (Fig. 4a), we detected similar titers of HR121-binding IgGs (GMT: 1.3 × 106) (Fig. 4b) and nAbs to authentic SARS-CoV-2 (GMT: 3.3 × 103) (Fig. 4c). After being challenged with a high titer of SARS-CoV-2 (TCID50 = 107), 7/8 mice were completely protected from the viral infection. One mouse had a few remnant virions in its lung tissue (Fig. 4d), but they were inactive as no viral sgRNAs were detected (Fig. 4e). Additionally, mice vaccinated with HR121 showed decreased levels of some inflammation-related cytokines (IL-4, IL-6 and IP-10) and antiviral cytokines (IFNγ and MX2) and their genes in the lung tissues, compared to those of the adjuvant controls (Fig. 4f; Supplementary information, Fig. S5c), implying that the inflammatory immune responses induced by SARS-CoV-2 infection were reduced in these mice. Hematoxylin and eosin (H&E) staining showed that mice vaccinated with HR121 did not have obvious histopathological changes in their lung tissues (Fig. 4g; Supplementary information, Fig. S5a). In contrast, in the two control groups (PBS and adjuvant alone), the lung tissues showed typical features of viral interstitial pneumonia, including infiltration of lymphocytes and macrophages, detached bronchial mucosa, and thickened alveolar walls (Fig. 4g). Immunohistochemistry staining also did not detect the nucleocapsid protein of SARS-CoV-2 in the lung tissues of HR121-vaccinated mice. By contrast, in PBS and adjuvant controls, the nucleocapsid protein was aggregated around the bronchial epithelial cells and alveolar epithelia (Supplementary information, Fig. S5b). Thus, these data suggest that HR121 elicits potent nAbs in hACE2 transgenic mice against SARS-CoV-2 infection.
We further assessed the protective efficiency of HR121 in Syrian golden hamsters. Two groups of hamsters were subcutaneously vaccinated with 3 doses of HR121 (15 μg each) formulated with Freund’s adjuvant (CFA/IFA) at two-week intervals. Then, they were separated into short-term protective (STP) and long-term protective (LTP) groups. The SARS-CoV-2 (TCID50 = 104) challenge was carried out two weeks or three months after the last vaccination in the STP and LTP groups, respectively (Fig. 5a). We observed that HR121 induced a relatively moderate humoral immune response in hamsters compared with that in mice. The anti-HR121 IgG titers reached a geometric mean of 8.1 × 104 after the third HR121 dose, with no obvious attenuation three months after the last immunization (Fig. 5b). Correspondingly, the GMT of nAbs to authentic SARS-CoV-2 in the STP group was 3.5 × 102 and was similar in the LTP group (Fig. 5c). Although the nAbs were relatively lower than those generated in mice, the nAbs showed dose-dependent inhibition against SARS-CoV-2 replication in HPAEpiC cells (Fig. 5d). The lung tissues of hamsters were collected on day 3 after the viral challenge. In the STP group, only 3/10 HR121-vaccinated hamsters had detectable viral gRNAs (Fig. 5e), most likely from the remnant inactive viral genomes, since no viral sgRNAs were determined (Fig. 5f). In the LTP group, even three months after the vaccination, HR121 showed efficient protection in hamsters. Viral gRNAs were detected in 6/8 hamsters with ~5 Log reduction (median), compared to those in adjuvant controls (Fig. 5g). Again, these gRNAs should be inactive since no viral sgRNAs were detected (Fig. 5h). The lung tissues of hamsters in both STP and LTP groups showed no obvious pathological changes by H&E staining, similar to the healthy hamsters (Fig. 5i, j; Supplementary information, Fig. S6a). In the controls, pathological changes of the lung were observed, including severe infiltration of lymphocytes and macrophages, apparent thickened alveolar walls, diffuse alveolar damage, detached bronchial mucosa, and disappearance of recognizable architecture (Fig. 5i, j). The nucleocapsid protein staining also confirmed the uninfected states of HR121-vaccinated hamsters (Supplementary information, Fig. S6b). These results indicate that a relatively low level of nAb vaccination can provide long-term protection in hamsters against SARS-CoV-2 infection.
To assess the efficacy of HR121 vaccine in nonhuman primates, we subcutaneously injected four rhesus macaques with 50 μg HR121 plus IFA, and another four rhesus macaques with IFA only as mock controls. Vaccinations were given three times at one-month intervals, and sera were collected seven days after each dose for antibody examination. The SARS-CoV-2 challenge was performed seven days after the last vaccination (Fig. 6a). We found that HR121 elicited a strong antibody response in rhesus macaques. The first dose induced low HR121-binding IgG titers with GMT of 1.0 × 102; GMTs of the IgG titers increased to 9.5 × 104 after the second dose, and peaked at 6.9 × 105 after the third dose (Fig. 6b). Correspondingly, the nAbs to SARS-CoV-2 were undetectable after the first dose, increased to 2.7 × 102 and 1.1 × 103 after the second and third doses, respectively (Fig. 6c; Supplementary information, Fig. S7a). The cross-nAbs in the sera of the HR121-vaccinated macaques were evaluated after the third dose using the VSV pseudotyped viruses. Notably, HR121 induced broad-spectrum and homogeneous nAbs to the main circulating variants in these macaques. Compared with the naïve (Wuhan-Hu-1) strain, the GMTs of nAbs to the 6 previously circulating variants (Alpha, Beta, Gamma, Delta, Lambda, and Mu) only dropped by 1.5–2.9 folds, and no significant differences were observed among them (P > 0.05, ordinary one-way analysis of variance (ANOVA)). Encouragingly, the nAbs had a considerable effect against the current circulating variants of Omicron, with GMTs of 3.9 × 102 to BA.1, 2.5 × 102 to BA.2, 2.3 × 102 to BA.3, and 4.0 × 102 to BA.4/5. The titers of nAbs to BA.2 and BA.3 had a slight decrease by 3.4 and 3.8 folds, respectively, compared with that of the naïve strain. The nAbs demonstrated similar activities against Omicron and its sublineages, implying that the mutations in the sublineages of Omicron, which are mainly located in the S1 subunit of the S protein, are possibly unrelated to the efficiency of the membrane fusion and thus cannot dampen the efficacy of HR121-elicited nAbs (Fig. 6d; Supplementary information, Fig. S7b). Interestingly, the homogeneity of the nAbs evoked in macaques was slightly different from those in rabbits, which varied among different strains (Fig. 2k). This discrepancy may be due to the diverse Ig germline generations in different species and individuals. Generally, in both rabbits and macaques, HR121 induced potent cross-nAbs against the main previous and current variants of SARS-CoV-2.
Taking into consideration the high levels of nAbs elicited by HR121 in macaques, which are similar to those elicited in BALB/c (Fig. 3c) and hACE2 transgenic mice (Fig. 4c), and the possibility of the undetectable viral RNAs in some pulmonary tissues of SARS-CoV-2-challenged macaques as previously reported, we challenged these eight macaques with a high titer of SARS-CoV-2 (TCID50 = 3 × 107) equally through intranasal and intratracheal routes. Subsequently, we collected nasal and throat swabs on days 1, 3, 5, and 7 after the challenge. In adjuvant-injected macaques, viral gRNAs were consistently detectable, and remained at high levels in nasal swabs (medians ranged from 1.1 × 106 to 2.8 × 107 copies per test) and throat swabs (medians ranged from 1.3 × 104 to 1.6 × 105 copies per test) at each sampling time point. In HR121-immunized macaques, viral gRNAs were only detected at a relatively high level in nasal swabs on day 1 (median: 1.2 × 105 copies per test), with a decline on day 3 (median: 2.8 × 104 copies per test) after the challenge. The gRNAs were occasionally detectable in throat swabs. Accordingly, it can be inferred that these viral gRNAs were from remnant inactive virions or from the remaining inoculations, as no viral sgRNAs in HR121-immunized macaques were detected in both the nasal and throat swabs at all sampling time points (Fig. 6e, f). On day 7 after the challenge, the macaques were euthanized and their lung tissues were collected for viral load evaluation and pathological analysis. In adjuvant-immunized macaques, the viral gRNAs (medians ranged from 9.3 × 103 to 1.8 × 106 copies/μg RNA) remained at high levels in all tested lung lobes, showing a similar pattern to those in nasal and throat swabs. Likewise, few viral gRNA remnants were detected in only one HR121-immunized macaque, and they were undetectable in later viral sgRNA tests (Fig. 6g). After H&E staining, severe pathological lesions were observed in adjuvant-injected macaques; the typical pathological damages included thickened alveolar walls, lymphoid proliferation, intra-alveolar hemorrhage, alveolar macrophage infiltrates, and perivascular macrophage infiltrates. However, these pathological lesions were minimal and occasionally revealed in HR121-immunized macaques (Fig. 6h, i; Supplementary information, Fig. S8), which coincided with the reported results of some SARS-CoV-2 vaccines. The observed pathological changes in the HR121-immunized macaques may be highly attributed to the acute immune responses following the high inoculation of viral particles into the trachea of these macaques. In HR121-immunized macaques, there were no considerable changes in body weight and body temperature after the viral challenge, whereas in the adjutant-injected macaques, the average body weight decreased by ~5% seven days (at the endpoint) after the challenge (Fig. 6j), and the average body temperature was a little higher than that of HR121-immunized macaques at all monitoring time points (Fig. 6k). These clinical signs observed suggest that HR121 has a protective role in rhesus macaques.
Finally, we assessed the in vivo efficacy of HR121 vaccine against the currently emerging SARS-CoV-2 Omicron variant BA.2. One group of hamsters were subcutaneously vaccinated with HR121 formulated with Freund’s adjuvant (CFA/IFA) (Freund’s adjuvant only as control), while another group of hamsters were intramuscularly vaccinated with HR121 formulated with a clinical-grade aluminum adjuvant (aluminum adjuvant only as control). Both vaccinations were administered with the same dose of HR121 (15 μg each) three times, at two-week intervals. Omicron BA.2 (TCID50 = 103) challenge was carried out two weeks after the last vaccination (Fig. 7a). After the vaccination, we found that HR121/aluminum adjuvant could elicit HR121 bAbs (GMT: 1.5 × 105) at the similar level to that induced by HR121/Freund’s adjuvant (GMT: 3.2 × 105) (Fig. 7b). The titers of nAbs against authentic Omicron BA.2 in the sera from these two groups of hamsters were also at a similar level, with GMT values of 2.8 × 102 in the HR121/Freund’s adjuvant group and 1.8 × 102 in the HR121/aluminum adjuvant group, respectively (Fig. 7c). At 3 days post the Omicron BA.2 challenge, the viral loads and pathological changes in lung tissues of these hamsters were examined. The pulmonary gRNAs in both Freund’s adjuvant and aluminum adjuvant controls were detected with median values of 2.1 × 105 and 1.0 × 105, respectively. There were no significant differences between them. gRNAs were detected in 5/12 hamsters in the HR121/Freund’s adjuvant group and 4/12 hamsters in the HR121/aluminum adjuvant group, with 5.3 and 5.0 Log reductions of the median values of the total numbers (12/12), respectively (Fig. 7d). Correspondingly, the pulmonary sgRNAs in both Freund’s and aluminum adjuvant controls were detected with median values of 5.8 × 104 and 5.9 × 104, respectively (no significant differences). The sgRNAs were detected in 3/12 and 1/12 hamsters in the HR121/Freund’s adjuvant and the HR121/aluminum adjuvant groups, with 4.4 and 4.0 Log reductions in the median values of the total numbers, respectively (Fig. 7e). After the H&E staining of the lung tissues, no apparent pathological changes were observed in both the HR121/Freund’s adjuvant and HR121/aluminum adjuvant groups, whereas in the Freund’s or aluminum adjuvant controls, slight to moderate degrees of infiltration of lymphocytes and macrophages, along with alveolar damage were presented (Fig. 7f). Together, these results suggest that HR121 formulated with Freund’s or aluminum adjuvants could induce similarly effective protection against Omicron BA.2 infection in the vaccinated Syrian golden hamsters.
The mechanisms of viral and host cellular membrane fusion are intriguing. Efforts have been made to design mimetics of the HR1 domain present in the fusion intermediate conformation, such as various HR1 trimers or 5-Helix, as fusion inhibitors against HIV-1, and SARS-CoV-2. Other research groups have applied the mimetics of the HR1 domain in the HIV-1 gp41 to design antibody-based HIV-1 vaccines, but these studies were terminated due to the poorly elicited nAbs. To date, there has been no report showing that the mimetics of HR1 domain present in the fusion intermediate conformation from a coronavirus is applied to vaccine design, although some groups have used the fragment consisting of HR1 and HR2 in SARS-CoV-2 S2 subunit for its conjugation with RBD to design RBD-HR-based vaccines. In this study, we designed a recombinant protein HR121, which showed a stable structure with a partially exposed HR1 trimeric structure, in which two surface grooves remained unoccupied, analogous to the theoretical HR1 trimeric inner core in the fusion intermediate conformation of the S2 subunit. As a proof of principle, HR121 is able to induce high titers of nAbs against SARS-CoV-2 in the immunized animals, such as rabbits, BALB/c mice, hACE2 transgenic mice, Syrian golden hamsters, and rhesus macaques. The high nAb titers evoked by HR121 from SARS-CoV-2 were different from those induced by other fusion intermediate mimetics from HIV-1. The reason for this discrepancy is unknown but may be related to: (1) differences of the efficient antigenic epitopes in different immunogens; the full HR1 and HR2 sequence in HR121 may improve its immunogenicity, and the two exposed surface grooves in one of HR1 trimeric structure of HR121 may provide more efficient antigenic epitopes for induction of nAbs; (2) differences of steric hindrances in the fusion intermediates; the fusion intermediate structure in HIV-1 was estimated with a length of 100 Å, which has provided enough space for IgGs to enter their binding sites, whereas in SARS-CoV-2, with an unusually prolonged HR1 domain, the length of the fusion intermediate complex was calculated to be at least 180 Å, providing a larger space for IgG accession. In other studies, the serum nAb titers have been proved a main immune correlate of protection in evaluating the efficacies of COVID-19 vaccines. Here, we also demonstrated that the potent nAb responses induced by HR121 could provide nearly full protection against SARS-CoV-2 infection in several SARS-CoV-2-susceptible animal models, including hACE2 transgenic mice, Syrian golden hamsters and rhesus macaques. Aside from the robust antibody responses, we did not observe the HR121-elicited antibody-dependent enhancement of viral infection in either in vitro or in vivo experiments, in line with the result of the RBD-HR-based nanoparticle vaccine in which HR1 and HR2 domains were linked together as a scaffold and immunoenhancer. Thus, these results suggest that HR121 is a promising COVID-19 vaccine candidate. Currently, the continuous emergence and rapid evolution of Omicron sublineages including BA.1, BA.2 and BA.3 sublineages had showed widespread escapes from the neutralization by most antibodies and vaccines. Besides this, the recently evolved Omicron sublineages BA.4 and BA.5 demonstrated stronger immune evasion against the plasma of 3-dose vaccinees, and BA.1 or BA.2 convalescents. These two sublineages BA.4 and BA.5 have emerged and dominated in South Africa, and are rising worldwide. Apart from them, the ongoing COVID-19 pandemic and concomitantly continuous evolution of SARS-CoV-2 will constantly evoke more newly mutated or recombinant variants. Therefore, those vaccine strategies, such as updating the spike antigens, or vaccine boosters, may be cumbersome and inaccurate to match the speed of antigenic drift, which is similar to the current status of seasonal influenza vaccines. In this study, we demonstrated another strategy to design a novel spike antigen HR121, targeting the conserved HR1 domain of S2 subunit. The nAbs elicited in both HR121-vaccinated rabbits and rhesus macaques exhibited broad-spectrum neutralizing activity against the main previous and current SARS-CoV-2 variants, particularly the current globally circulating variant Omicron and its sublineages. This result coincides with the observation that most of the mutations among the SARS-CoV-2 variants were located in the S1 subunit, and provides an alternative target for the development of the next-generation COVID-19 vaccine. The design strategy of the recombinant protein vaccine HR121 may also be applied to other COVID-19 vaccine platforms, such as viral vector-based vaccines, virus-like particles and DNA/mRNA vaccines, and combined with other targeted COVID-19 vaccines, such as RBD vaccines and inactive viral vaccines. In addition, several broadly neutralizing antibodies (bnAbs) targeting the conserved regions in S2 subunit, including the FP and stem helix (SH) in the region between HR1 and HR2, have been isolated from SARS-CoV-2-infected or vaccinated individuals. They all act at the membrane fusion steps and show cross-binding and neutralizing activities against sarbecoviruses, implying that fusion mechanism of S2 subunit is a promising target for broad-spectrum vaccine designs. However, there has been no report so far about the clinical or per-clinical studies on SARS-CoV-2 vaccines targeting the S2 subunit, although Ng et al. has reported that a DNA vaccine containing genes encoding membrane-bound SARS-CoV-2 S2 subunit can induce bnAbs against sarbecoviruses, while the bacterially produced recombinant S2 protein cannot elicit bnAbs. These results, together with those from our study, suggest that these S2-specific bnAbs isolated from the SARS-CoV-2-infected or vaccinated individuals may be elicited by the neutralizing epitopes in the conserved regions of S2 subunit at a special fusion conformation (e.g., pre-fusion, fusion intermediate or post-fusion conformation), while our HR121 vaccine can induce potent bnAb responses in the vaccinated animals, possibly because HR121 vaccine contains the transiently exposed neutralizing epitopes in the HR1 region of S2 subunit present in the fusion intermediate conformation. Therefore, it is both a great opportunity and a challenge to design an effective S2 subunit vaccine containing the exposed neutralizing epitopes in the conserved regions of S2 subunit at a proper conformation that can induce potent and broad-spectrum cross-nAb responses. Meanwhile, the approach taken herein provides a specific immunogen to evoke HR domain-targeted nAbs. This means that further isolation of the mAbs from mice immunized with HR121 is feasible, thus making the resultant humanized antibody constructed from mouse antibodies for SARS-CoV-2/COVID-19 therapy promising. Finally, HR domains in S2 subunit are extraordinarily conserved among coronaviruses. Further, other class I enveloped viruses, including some of the most studied viral families, such as retroviruses, orthomyxoviruses, paramyxoviruses, filoviruses, and arenaviruses, share a similar membrane fusion mechanism, in which HR1 and HR2 participate. Therefore, our strategy for vaccine design may extend or optimize HR121 for better antiviral activity against some types of these viruses, with careful consideration for the accessibility of IgGs to the binding sites formed by fusion intermediates and binding valences of nAbs to viruses. In summary, this study demonstrates that HR121 can mimic the HR1 domain present in the fusion intermediate conformation of the S2 subunit, and induce highly potent broad-spectrum antibodies against SARS-CoV-2 and its main variants. Thus, it provides another important target for the development of novel COVID-19 vaccines.
There are several limitations in this study. First, we are unable to solve the structure of HR121/HR2 complex, or HR121/antibody complex. Particularly, the conformation of the target sites in the HR1 part of HR121 may change significantly after HR121 binding to HR2 or antibodies, it is difficult for us to show the critical neutralizing epitopes and HR2-binding sites in HR121. Second, the homogeneous antibody and nAb responses elicited in HR121-immunized animals, and the small size of animal numbers, especially rabbits and rhesus macaques, limited the performances of the correlation between the inhibition of HR121 binding with HR2 and nAbs, or the relationship of antibodies and nAbs. Third, with the outcomes of the undetectable viral sgRNAs in different SARS-CoV-2-challanged animals, we did not measure the anamnestic nAb titers after infection, as they had showed little to no changes compared to the nAb titers evoked by pre-challenge of some SARS-CoV-2 vaccines with nearly full protection. Finally, we mainly focused on the evaluation of the vaccine-elicited nAb responses and the protective outcomes in different HR121-immunized animals, while the cellular immune correlates of protection, particularly the roles of CD4+ T and CD8+ T cell responses, which are likely to participate in the control of viral infection and disease severity, were not well investigated.
The Ethics Review Board of the Kunming Institute of Zoology, Chinese Academy of Sciences (CAS) approved this study. All animal experiments were approved by the Ethics Committee of Kunming Institute of Zoology, CAS (assurance No.: SMKX-2021-01-006), and were carried out in strict compliance with the guidelines and regulations of the Animal Care and Use Committee, Kunming Institute of Zoology, CAS. All SARS-CoV-2 infection experiments were approved (assurance No.: KIZP3-XMZR-2021-05) and carried out in the biosafety level-3 (BSL-3) laboratory of Kunming Institute of Zoology, CAS. The human serum samples in this study were collected in accordance with the Declaration of Helsinki. Informed consent was obtained from each participant.
The prototype SARS-CoV-2 strain (Accession No.: NMDCN0000HUI in the China National Microbiology Data Center (NMDC)) was provided by the Guangdong Provincial Center for Disease Control and Prevention (Guangzhou, China). The SARS-CoV-2 Omicron BA.2 variant was isolated from a patient in Yunnan Provincial Infectious Disease Hospital, China. Its whole genome was sequenced by researchers at BSL-3 laboratory of Kunming Institute of Zoology, CAS, Kunming, China, and submitted to NMDC (No. SUB1663744906574). The viruses were propagated in African green monkey kidney epithelial cells (Vero-E6) (ATCC, #1586) and titrated. Virions from the 4th passage were used in the current experiments. VSV-G pseudotyped virus (G*ΔG-VSV-Rluc) was kindly provided by Prof. Geng-Fu Xiao (Wuhan Institute of Virology, CAS). HEK293T cells were obtained from ATCC. 293T-ACE2 cells were provided by Prof. Yuxian He (Institute of Pathogen Biology, Chinese Academy of Medical Sciences). These cells were cultured in basal DMEM (Gibco, Beijing, China) supplemented with 10% FBS (Gibco, New Zealand). Human pulmonary alveolar epithelial cells (HPAEpiCs) were purchased from the ScienCell Research Laboratory (San Diego, CA, USA) and were cultured in basal DMEM supplemented with 10% FBS, and passages 6–8 were used in this study.
Nine serum samples from SARS-CoV-2-vaccinated participants and one healthy unvaccinated serum sample were collected from our laboratory. The participants were injected with two doses of inactivated SARS-CoV-2 vaccine (Sinovac Biotech, Beijing, China) within 1–2 months. Eleven serum samples from COVID-19 convalescent individuals were collected at Yunnan Provincial Infectious Disease Hospital. All convalescent patients had recovered from COVID-19 for 1–2 months.
BALB/c mice and New Zealand white rabbits were purchased from the Experimental Animal Center of Kunming Medical University, China. The hACE2 transgenic mice were established as we previously reported. Syrian golden hamsters were purchased from Charles River Company (Beijing, China). Rhesus macaques were purchased from Kunming Primate Research Center, Kunming Institute of Zoology.
Recombinant protein HR121 from the S protein of SARS-CoV-2 was designed as HR1–linker1–HR2–Linker2–HR1. Genes encoding the full amino acid sequence of HR1 (residues 912–988) and HR2 (residues 1163–1206) were derived from the S protein of SARS-CoV-2 isolate Wuhan-Hu-1 (accession No.: NC_045512). The nucleotide sequence encoding linker 1 (GSSGG) was GGAGGAAGCGGAGGA and the nucleotide sequence encoding linker 2 (SGGRGG) was AGCGGAGGAAGAGGAGGA. The gene encoding HR121 was synthesized and cloned into the E. coli expression vector pMCSG7 with an N-terminal SUMO tag using the ligation-independent cloning method. To express the GST-HR121 fusion protein, the full HR121 sequence was inserted into the E. coli expression vector, pGEX-6P-1, at the restriction enzyme sites of EcoRI and XhoI. To express HR1, HR2, and HR12 proteins, the genes encoding these proteins were separately cloned into the E. coli expression vector, pET-30a, at EcoRI and XhoI restriction enzyme sites. Therefore, each of these expressed proteins contains the redundant 52 amino acid residues from the vector.
SUMO-tagged HR121 in the pMCSG7 vector was expressed in E. coli BL21 (DE3), and bacteria harboring the expression vector were grown at 37 °C in LB media supplemented with 100 μg/mL ampicillin. Protein expression was induced with 0.5 mM IPTG when the cells reached an optical density of 0.6 at 600 nm. The cells were then cultured at 16 °C for another 16 h. Then, the cells were harvested by centrifugation at 5000× g for 10 min at 4 °C. The cells were resuspended in lysis buffer (25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM β-ME), and lysed using ultrasonication. Then, the supernatant containing the recombinant proteins was separated by centrifugation at 12,000× g for 30 min at 4 °C. The fusion proteins were isolated by Ni Sepharose 6 FF (GE Healthcare, Beijing, China), and the SUMO tag was removed by TEV enzyme (1:100 w/w) cleavage. The target protein was loaded to a Superdex 200 increase column (GE Healthcare) with buffer containing 25 mM Tris-HCl, pH 8.0, 150 mM NaCl and 2 mM DTT. Peak fractions containing HR121 dimer were pooled, concentrated to 25 mg/mL and stored in a –80 °C freezer. The recombinant proteins HR1, HR2, and HR12 were expressed in E. coli BL21 (DE3) cells. Bacteria were induced with 1 mM IPTG for 12 h at 20 °C before harvesting by centrifugation. The collections were lysed in PBS buffer by ultrasonication. The supernatants were separated by centrifugation at 12,000× g for 30 min at 4 °C, and the target proteins were purified using Ni Sepharose 6 FF affinity column. Briefly, the column was washed in gradients with 20 mM and 50 mM imidazole. Proteins were then eluted by 100 mM imidazole and concentrated to 10 mg/mL in PBS through a concentrating column with a 3 kDa cutoff (Millipore, Bedford, MA, USA). The fusion protein GST-HR121 was expressed and induced using the same method as that for the recombinant proteins HR1, HR2, and HR12. Purification was carried out by glutathione–Sepharose 4B affinity column (GE Healthcare).
Crystals were obtained using the sitting-drop vapor diffusion method by commercial crystallization kits and incubation at 16 °C for 10 days. The crystals appeared in a solution containing 0.2 M sodium fluoride and 20% (w/v) polyethylene glycol 3350. Then, the crystals were harvested using 20% ethylene glycol (v/v) as a cryoprotectant before flash freezing in liquid nitrogen. The crystals were analyzed with beamline BL02U1 at the Shanghai Synchrotron Radiation Facility. The structure was determined by the molecular replacement method using PHASER and refined using PHENIX. The structure of 6-HB of SARS-CoV-2 (Protein Data Bank: https://www.rcsb.org/, with the accession number 6LXT) was used as the initial search model. The model of HR121 was manually adjusted in COOT and refined to a resolution of 2.41 Å with Rwork and Rfree values of 24.9% and 29.1%, respectively. Details of data collection and refinement statistics are provided in Supplementary information, Table S1. All figures representing structures were prepared using PyMOL (http://pymol.org).
Excess HR1 or HR2 in the bacterial supernatants was mixed with glutathione–Sepharose 4B affinity gel containing GST-HR121. Blank glutathione–Sepharose 4B affinity gel mixed with HR1 or HR2 and glutathione–Sepharose 4B affinity gel containing GST-HR121 were used as negative controls. The mixture was incubated for 30 min at room temperature with gentle agitation. Then, the supernatants were removed by centrifugation at 500× g for 5 min. The gels were washed three times with PBS by the same method of centrifugation, and collected for 10% SDS-PAGE analysis.
HR121 binding to HR2 or HR1 was determined by surface plasmon resonance using a BIAcore 3000 instrument (GE Healthcare). Briefly, HR121 was immobilized on the flow cell of a CM5 sensor. HR2 or HR1 protein (12.5 nM, 25 nM, 50 nM, 100 nM, and 200 nM) was injected to run across the chip. A separate channel was set as a control. The binding assays were performed at 25 °C. HR2 or HR1 was dissolved in BIAcore running buffer and injected at a constant flow rate of 35 μL/min for 3 min. Dissociation data were collected for 10 min. The kinetic parameters were obtained using an automated program.
CD spectra were recorded using a Jasco spectropolarimeter (model J-815). 1 μM each of HR121 and HR2 or HR1 was dissolved in PBS. Using a 0.1 cm pathlength cuvette, wavelength spectra were recorded with a 1-nM step size and 1-nM bandwidth from 195 nm to 260 nm at 20 °C. The spectra were corrected by subtracting the solvent blank, PBS.
Rabbits, mice, and hamsters were subcutaneously immunized with HR121 or HR12 formulated with CFA (Sigma-Aldrich, Saint Louis, MI, USA) and IFA (Sigma-Aldrich). Macaques were subcutaneously immunized with HR121 formulated with IFA. Another group of hamsters were intramuscularly immunized with HR121 formulated with aluminum adjuvant (Alhydrogel® adjuvant 2%, Invivogen, San Diego, CA, USA). All immunizations were performed in a prime boost-reboost manner. For rabbit immunization, 11 adult female New Zealand White Rabbits (average weight: 2.8 kg) were used. Four and three rabbits were injected with HR121 plus CFA/IFA and HR12 plus CFA/IFA in the same procedure, respectively. Each rabbit was immunized with 100 μg protein on day 0, and then 150 μg protein on days 21 and 42. The other 4 rabbits were injected with equal volumes of Freund’s adjuvant as mock controls. Serum samples were collected 14 days after the third immunization. IgGs were purified from the serum samples using Protein A (BBI, Shanghai, China). For BALB/c mouse (male, 8 weeks old) immunization, three groups of mice (n = 6 per group) were injected with 2 μg, 10 μg, or 50 μg HR121 plus CFA/IFA at 14-day intervals. Two groups of mice (n = 6 per group) were injected with equal volumes of CFA/IFA or PBS (mock controls). Serum samples were collected at 7 days post each immunization, and 60 days post the third immunization. At 90 days post the third immunization, the mice were euthanized, and their splenocytes were isolated as previously reported. To optimize the HR121 immunization interval in mice, another group of mice (n = 8) was injected with 10 μg HR121 at 21-day intervals. Serum samples were collected 7 days after the third immunization. For hACE2 mouse (male, 8 weeks old) vaccination, 8 hACE2 mice were vaccinated with 10 μg HR121 plus CFA/IFA at 14-day intervals. Equal numbers of mice injected with Freund’s adjuvant or PBS were used as controls. Serum samples were collected 7 days after the third immunization. SARS-CoV-2 challenge was carried out 14 days after the third immunization. Syrian golden hamsters (male, 8 weeks old) were injected with 15 μg HR121 formulated with adjuvant, or adjuvant, or PBS only at 14-day intervals. For SARS-CoV-2 challenge studies, 39 hamsters were divided into two groups for short- and long-term protection studies. In STP group (n = 25), 10 and 9 hamsters were immunized with HR121 plus CFA/IFA and CFA/IFA only, respectively, while 6 hamsters were injected with PBS as mock control. Serum samples were collected 7 days after the third immunization. SARS-CoV-2 challenge was carried out 14 days after the third immunization. In LTP group (n = 14), 8 and 6 hamsters were injected with HR121 plus CFA/IFA and CFA/IFA only, respectively. Serum samples were collected 83 days after the third immunization. SARS-CoV-2 challenge was carried out 90 days after the third immunization. For Omicron BA.2 challenge studies, 48 hamsters were divided into 4 groups (12/group). Hamsters in Groups 1 and 2 were injected with HR121 plus CFA/IFA and CFA/IFA only, respectively, while those in Groups 3 and 4 were immunized with HR121/aluminum adjuvant and aluminum adjuvant only, respectively. Serum samples were collected 7 days after the third immunization. Omicron BA.2 challenge was carried out 14 days after the third immunization. For rhesus macaque (3 males and 5 females, 9–13 years old) vaccination, 4 macaques were vaccinated with 50 μg HR121/IFA at 30-day intervals. Equal numbers of macaques injected with adjuvant were used as controls. Serum samples were collected 7 days after each immunization. SARS-CoV-2 challenge was carried out 7 days after the third immunization.
To evaluate the in vivo neutralizing activity of nAbs from HR121-immunized rabbits, sera from the 4 HR121-immunized rabbits were pooled. The IgGs purified from them (~5 mg IgGs obtained from 1 mL sera) were transferred intraperitoneally to 6 hACE2 mice (8 weeks old) or 9 hamsters (8 weeks old) at a dose of 5 mg IgGs/20 g body weight, while isotypical IgGs purified from adjuvant-immunized rabbits were transferred to 4 mice or 8 hamsters as mock controls. 24 h after transfer, SARS-CoV-2 challenge was carried out.
To challenge hACE2 mice vaccinated with HR121 (adjuvant and PBS as controls), the mice were anaesthetized with isoflurane (RWD, Shenzhen, China) and inoculated intranasally with 107 TCID50 of SARS-CoV-2 in 30 μL. Lungs were collected 5 days post infection (dpi). To challenge Syrian golden hamsters vaccinated with HR121, the hamsters were anesthetized and inoculated intranasally with 104 TCID50 of SARS-CoV-2 or 103 TCID50 of Omicron BA.2 in 100 μL. Lungs were collected at 3 dpi. To challenge rhesus macaques, the macaques were anesthetized with Zoletil-50 (FeiBo, Beijing, China) and inoculated equally by intranasal and intratracheal routes with 3 × 107 TCID50 of SARS-CoV-2 in 2 mL. Lungs were collected at 7 dpi. To challenge the animals passively administered with rabbit sera, hACE2 mice were inoculated with 106 TCID50 of SARS-CoV-2 and Syria golden hamsters were inoculated with 104 TCID50 of SARS-CoV-2, respectively. Lungs were collected at 3 dpi.
A sandwich ELISA was used to detect the bAbs of HR1, HR2, HR12, and HR121 in serum samples. Briefly, HR1, HR2, HR12, or HR121 protein (1 μg/mL) in coating buffer (15 mM Na2CO3, 35 mM NaHCO3, pH 9.6) was coated on 96-well polystyrene plate at 4 °C overnight. After removing the coating buffer, the plate was washed 3 times with PBS containing 0.05% Tween-20 (PBST) and blocked with blocking buffer (PBST containing 5% BSA) at 37 °C for 2 h. Then the plate was washed 3 times, and serially-diluted serum samples were added to the plate and incubated for 2 h at 37 °C. After washing again, OPD substrate was added to each well. The reaction was stopped with 2 M H2SO4 and the optical density (OD) values of the wells were read on an ELISA reader at 490 nm/630 nm. The endpoint titers of the serum samples were determined as the reciprocal of the last dilution exhibiting an OD value ≥ 2.1-fold that of the average background values.
Competitive ELISA was used to detect antibodies blocking the binding of HR2 and HR121. Briefly, HR121 protein (1 μg/mL) was coated onto the ELISA plate at 4 °C overnight. After the coating buffer was removed, the plate was washed with PBST and incubated with blocking buffer at 37 °C for 2 h. Then, the plate was washed 3 times with PBST. Meanwhile, serum or IgG samples from rabbits or humans were serially diluted in PBST and preincubated with HR2-labeled HRP (KPL, Gaithersburg, MD, USA) in 100 ng/mL for 20 min at 20 °C. Then the mixtures were added to each well of the plate and incubated for 1 h at 37 °C. After washing the plate and adding the OPD substrate, the OD values of the plates were determined at 490 nm/630 nm. The percentage inhibition of HR121/HR2 binding was calculated using the following formula: % inhibition = [1 – (E – N)/(P – N)] × 100, where E represents the OD value in the presence of a serum or IgG sample, P represents the OD value in the absence of the serum or IgG sample, and N corresponds to the OD value in the absence of the sample and HR2-labeled HRP. The inhibition data were plotted and log-transformed in GraphPad Prism 8.0.1 (GraphPad Software, Inc., San Diego, CA, USA). The NT50s in the serum and IC50s in the IgG samples were calculated using the “nonlinear regression (curve fit) — log(inhibitor) vs response -- variable slope (four parameters)” model.
To construct VSV-based SARS-CoV-2 pseudoviruses, pCMV3 plasmid containing the full sequence of the codon-optimized spike gene was purchased (Sino Biological, Beijing, China). The spike gene with a C-terminal 18-amino acid truncation of the SARS-CoV-2 Wuhan-Hu-1 strain was constructed by PCR and inserted into the eukaryotic expression plasmid pcDNA3.1(+) between the BamH1 and EcoR1 sites. A Kozak sequence (GCCACC) was introduced in front of the spike gene to generate the recombinant plasmid pcDNA3.1-SARS-CoV-2-SΔ18. Based on pcDNA3.1-SARS-CoV-2-SΔ18 sequence, a panel of plasmids containing the SARS-CoV-2 spike-related mutants of S477N, E484K, A222V, N439K, K417N, D614G, and D839Y were introduced using a Fast mutagenesis system kit (Transgen Biotech, Beijing, China). Another panel of plasmids containing the spike genes of 17 important SARS-CoV-2 variants, including B.1.617, B.1.617.1, B.1.617.2.V2, B.1.429, B.1.525, B.1.526, B.1.1.7, B.1.351, B.1.1.28, B.1.617.2, C.37, B.1.621, B.1.1.529 (Omicron BA.1), Omicron BA.2, Omicron BA.3, and Omicron BA.4/5, were synthesized in codon optimization with the corresponding gene fragments. The primers for the point mutants are listed in Supplementary information, Table S2, and the multiple mutations introduced in the spikes of 13 important SARS-CoV-2 variants are listed in Supplementary information, Table S3. SARS-CoV-2 pseudovirus was prepared using a VSV pseudotyped SARS-CoV-2 S packaging system as described previously. Briefly, 1 × 106 293T cells were seeded on a T75 cell flask in DMEM low-sugar medium (Gibco) at 37 °C overnight. After reaching 80% confluence, the cells were transfected with 10 μg pcDNA3.1-SARS-CoV-2-variant-SΔ18 using jetPRIME transfection reagent (Polyplus-transfection, Illkirch, France). After 24 h, the cells were infected with G*ΔG-VSV-Rluc virus at an M.O.I. = 1. Six hours after infection, the cells were washed three times with PBS containing 1% FBS. Thirty-six hours after infection, the supernatant was collected, centrifuged at 1000× g for 10 min, aliquoted, and stored at −80 °C. The TCID50 of the SARS-CoV-2 pseudotyped virus was determined in 293T-ACE2 cells, as previously described.
To measure the neutralizing activity of sera or IgGs against various SARS-CoV-2 variant pseudotyped viruses, 1 × 104 293T-ACE2 cells (100 μL) were seeded in 96-well plates in DMEM low-sugar medium (Gibco). The next day, the sera from HR121-immunized animals were three-fold serially diluted in another 96-well plate in a volume of 60 μL. Then, 60 μL SARS-CoV-2 pseudovirus (M.O.I. = 0.1) was added to the diluted sera and incubated for 1 h at 37 °C. Thereafter, 100 μL of the mixture was incubated with 293T-ACE2 cells for 24 h at 37 °C, and the supernatant was removed from the cells after incubation. Renilla luciferase activity was determined using a Renilla luciferase assay kit (Promega, Madison, WI, USA). The NT50s of the sera or IC50s of the IgGs were calculated in GraphPad Prism 8.0.1 software as mentioned above.
8 × 105 HPAEpiC cells (200 μL) were seeded into each well of a 48-well plate and incubated at 37 °C overnight. The next day, sera or IgGs from HR121-immunized animals were two-fold serially diluted in another 48-well plate at a volume of 100 μL. Then, 100 μL SARS-CoV-2 (M.O.I. = 1) were added into the diluted sera or IgG and incubated for 1 h at 37 °C. Thereafter, the medium was removed and replaced with the virus-serum or virus-IgG mixture. After incubating at 37 °C for another 1 h, the mixture was removed and washed 3 times with PBS. Subsequently, fresh medium (200 μL) containing the same diluted sera or IgGs was added. The cells were cultured at 37 °C for 48 h, and then the supernatants were collected for viral RNA extraction by kit (Roche Diagnostics, Mannheim, Germany), followed by analysis of viral load (viral genome RNA). The NT50 and IC50 values were calculated using GraphPad Prism 8.0.1 as mentioned above.
RNA was extracted from lung tissues using Trizol (Life Technologies, Carlsbad, CA, USA). The RNA concentration was measured using a NanoDrop 2000 (Thermo Fisher Scientific, USA). Viral gRNA and sgRNA were measured by real-time qPCR using a one-step qRT-PCR kit (QRZ-101, Toyobo, Osaka, Japan) on a ViiA7 Real-Time PCR System (Life Technologies). The primers for gRNA detection were derived from the nucleocapsid (N) gene of SARS-CoV-2, as previously described, including forward, 5′-GGGGAACTTCTCCTGCTAGAAT-3′; reverse, 5′-CAGACATTTTGCTCTCAAGCTG-3′; and probe, 5′-FAM-TTGCTGCTGCTTGACAGATT-TAMRA-3′. The primers for sgRNA detection were derived from the envelope (E) gene of SARS-CoV-2, as previously described, including forward, 5′-CGATCTCTTGTAGATCTGTTCTC-3′; reverse, 5′-ATATTGCAGCAGTACGCACACA-3′; and probe, 5′-FAM-CGAAGCGCAGTAAGGATGGCTAGTGT-TAMRA-3′.
Real-time qPCR was performed to evaluate the mRNA expression of inflammation-related cytokines in the lung tissues of hACE2 mice after SARS-CoV-2 challenge. Primers (for genes including IFNG, IL-2, IL-4, IL-6, IL-10, IP-10, MX2, and TNFA) were designed according to mouse mRNA sequences (Supplementary information, Table S4). cDNA was generated using a PrimeScript RT Reagent Kit with gDNA Eraser (Takara, Beijing, China). Real-time qPCR was performed using a ViiA7 Real-Time PCR System and SYBR Premix Ex Taq II (Takara). Expression levels of the genes of interest were analyzed using the comparative cycle threshold (Ct) method, where Ct is the cycle threshold number normalized to that of ACTB mRNA. The fold change was calculated using the 2−ΔΔCt method by dividing the normalized quantity of post-infection samples by that of healthy hACE2 mouse samples.
Lung tissues of SARS-CoV-2-infected animals were collected and fixed in 4% paraformaldehyde for 3 days, followed by embedding in paraffin and cutting into 3 μm sections. Some sections were stained with H&E for evaluation of lung injury, and other sections were stained with the SARS-CoV-2 nucleocapsid protein (Sino Biological) to evaluate SARS-CoV-2 replication levels in the lung tissues.
SARS-CoV-2 HR1-specific cytotoxic T lymphocytes (CTLs) were evaluated using a murine IFN gamma set (Diaclone Research, Besancon, France) according to the manufacturer’s instructions. Briefly, a MultiScreenHTS IP filter plate (Millipore) was pre-coated with anti-murine IFNγ. Then, 1 × 106 splenocytes isolated from BALB/c mice were co-cultured with a pool of 15-amino-acid overlapped peptides covering the full HR1 sequence in 1 μg/mL (Supplementary information, Table S5, synthesized by Generay Biotech) for 24 h. After incubation, the wells were washed and colored, and images were collected using an ImmunoSpot S6 universal analyzer (Cellular Technology Limited, Cleveland, OH, USA). Using an automated program, the spots were counted with parameters such as size, intensity and gradient. To evaluate SARS-CoV-2 HR121-specific humoral responses, a MultiScreenHTS IP filter plate was pre-coated with HR121 (1 μg/mL), and washed 3 times. Then, 1 × 106 splenocytes isolated from BALB/c mice were added into each well and cultured for 24 h. Using the same method as that for IFNγ, the plate was imaged.
All statistical analyses were performed using GraphPad Prism 8.0.1. P values were labeled in the figures. The titers of the bAbs or nAbs were presented as geometric mean ± geometric SD, and the copies of viral gRNAs or sgRNAs were presented as median ± interquartile range. Two-tailed Mann-Whitney test or Wilcoxon test was used to compare the difference between HR121 and control groups.
Supplementary information, Fig. S1 Supplementary information, Fig. S2 Supplementary information, Fig. S3 Supplementary information, Fig. S4 Supplementary information, Fig. S5 Supplementary information, Fig. S6 Supplementary information, Fig. S7 Supplementary information, Fig. S8 Supplementary information, Table S1 Supplementary information, Table S2 Supplementary information, Table S3 Supplementary information, Table S4 Supplementary information, Table S5 Supplementary information, Video S1 Supplementary Video S1 legend |
PMC9648452 | Zhixia Zhang,Wenyi Yu,Guangyao Li,Yukun He,Zhiming Shi,Jing Wu,Xinqian Ma,Yu Zhu,Lili Zhao,Siqin Liu,Yue Wei,Jianbo Xue,Shuming Guo,Zhancheng Gao | Characteristics of oral microbiome of healthcare workers in different clinical scenarios: a cross-sectional analysis | 10-11-2022 | Healthcare institutes,Coronary care unit,Healthcare worker,16S rRNA,Oral microbiome,Microbial alteration | The environment of healthcare institutes (HCIs) potentially affects the internal microecology of medical workers, which is reflected not only in the well-studied gut microbiome but also in the more susceptible oral microbiome. We conducted a prospective cross-sectional cohort study in four hospital departments in Central China. Oropharyngeal swabs from 65 healthcare workers were collected and analyzed using 16S rRNA gene amplicon sequencing. The oral microbiome of healthcare workers exhibited prominent deviations in diversity, microbial structure, and predicted function. The coronary care unit (CCU) samples exhibited robust features and stability, with significantly higher abundances of genera such as Haemophilus, Fusobacterium, and Streptococcus, and a lower abundance of Prevotella. Functional prediction analysis showed that vitamin, nucleotide, and amino acid metabolisms were significantly different among the four departments. The CCU group was at a potential risk of developing periodontal disease owing to the increased abundance of F. nucleatum. Additionally, oral microbial diversification of healthcare workers was related to seniority. We described the oral microbiome profile of healthcare workers in different clinical scenarios and demonstrated that community diversity, structure, and potential functions differed markedly among departments. Intense modulation of the oral microbiome of healthcare workers occurs because of their original departments, especially in the CCU. Supplementary Information The online version contains supplementary material available at 10.1186/s12903-022-02501-x. | Characteristics of oral microbiome of healthcare workers in different clinical scenarios: a cross-sectional analysis
The environment of healthcare institutes (HCIs) potentially affects the internal microecology of medical workers, which is reflected not only in the well-studied gut microbiome but also in the more susceptible oral microbiome. We conducted a prospective cross-sectional cohort study in four hospital departments in Central China. Oropharyngeal swabs from 65 healthcare workers were collected and analyzed using 16S rRNA gene amplicon sequencing. The oral microbiome of healthcare workers exhibited prominent deviations in diversity, microbial structure, and predicted function. The coronary care unit (CCU) samples exhibited robust features and stability, with significantly higher abundances of genera such as Haemophilus, Fusobacterium, and Streptococcus, and a lower abundance of Prevotella. Functional prediction analysis showed that vitamin, nucleotide, and amino acid metabolisms were significantly different among the four departments. The CCU group was at a potential risk of developing periodontal disease owing to the increased abundance of F. nucleatum. Additionally, oral microbial diversification of healthcare workers was related to seniority. We described the oral microbiome profile of healthcare workers in different clinical scenarios and demonstrated that community diversity, structure, and potential functions differed markedly among departments. Intense modulation of the oral microbiome of healthcare workers occurs because of their original departments, especially in the CCU.
The online version contains supplementary material available at 10.1186/s12903-022-02501-x.
Unique conditions render the microbial composition in healthcare institutes (HCIs) vastly different from that in the external natural environment [1–3]. Hospitalized individuals are at a greater risk of human-related microorganisms or pathogens colonizing their nasal cavity compared with non-hospitalized groups [4]. Additionally, the diversity and compositions of the microbiome change dynamically in different clinical scenarios, which may be correlated with different indoor environmental conditions [5]. A previous study reported that the gut microbiome of intensive care unit (ICU) workers, compared with non-ICU workers, showed a significantly increased abundance of Dialister, Enterobacteriaceae, Phascolarctobacterium, Pseudomonas, Veillonella, and Streptococcus and a marked depletion of Faecalibacterium, Blautia, and Coprococcus [6]. Beyond the aspect of common colonizing microorganisms, potential pathogens in hospital settings cannot be neglected. Thus, different clinical scenarios in HCIs may affect the microbiota of healthcare workers. Microbial ecology in HCIs comprehensively affects the health of healthcare workers. Studies have shown that the adverse microbial characteristics of HCIs could increase the incidence of microbial infections in staff [7–9]. Therefore, identifying the microbial status of healthcare workers is of extreme importance in the control of nosocomial infections. Previous studies have widely focused on stool samples from healthcare workers to elucidate gut microbial properties, while respiratory samples are rarely involved. The respiratory tract is a common route of nosocomial infections [10], and the occurrence and progression of many diseases have been related to alterations in the respiratory microbiome [11–13]. The oral microbiome closely resembles that of the lung [14]. Numerous empirical studies have illustrated the dominance of Firmicutes, Actinobacteria, Proteobacteria, Fusobacteria, Bacteroidetes, and Spirochaetes in a healthy oral cavity, constituting 96% of the total oral bacteria [15, 16]. It has been noted that specific genera or species play roles in oral health and disease, even extraoral sites in systemic diseases [17]. Hence, exploring the extent of the oral microbiome through non-invasive operations may reflect more internal respiratory tract characterizations [18, 19]. Meanwhile, 16 S rRNA gene sequences have potential advantages for detecting the oral microbiome, whether they are cultivated or not [17]. Currently, the composition of the oral microbiome of healthcare workers and the influence of the HCI environment remain unclear. Here, we profiled the microbial community of oropharyngeal swabs from healthcare workers in different clinical scenarios in a hospital based on 16 S rRNA gene sequencing targeting multiple bacterial hypervariable regions. Oral microbial composition and predicted functional characteristics were analyzed to evaluate the impact of the hospital environment on the oral microbiome of healthcare workers.
A total of 65 full-time healthcare workers from coronary care unit (CCU, n = 12), ICU (n = 16), operating room (OR, n = 16) and department of respiratory medicine (RES, n = 21) of Linfen Central Hospital (Shanxi Province, China) were recruited. Subjects were excluded if they had a respiratory tract infection or respiratory tract disease, or were treated with antibiotics in three months prior to sampling, or had worked less than one year in the hospital. Participants were asked to avoid eating and drinking for three hours prior to sampling. All fresh oropharyngeal swabs were collected by one operator within two hours and immediately stored at −80 °C until DNA extraction. Additional information on age, gender, position, seniority, sleeping and dietary habits were obtained through questionnaires.
Total genomic DNA was extracted from oropharyngeal swabs using a TIANamp Micro DNA Kit (Tiangen, China) following the manufacturer’s instructions. We amplified the corresponding hypervariable regions (V2, V3, V4, V6-7, V8 and V9) of the 16 S rRNA with two primer pools in an Ion 16 S™ Metagenomics Kit (ThermoFisher Scientific, UK). A total of six primer pairs amplifying multiple hypervariable regions listed above were split into two pools to avoid possible interference during the amplification reaction. Every DNA template was amplified in two primer pools, and then two tubes of PCR products were combined to obtain complete amplification products from multiple hypervariable regions. XP beads were next used to purify the amplification products and quantified by Qubit4 (ThermoFisher Scientific, USA). Purified amplicons were ligated with barcodes to prepare the libraries. Then, libraries were pooled in equimolar amounts on chip 530 and sequenced to single-end, 250-base-pair reads on an Ion GeneStudio S5 System (ThermoFisher Scientific, USA) based on the Ion Reporter metagenomics workflow (Ion 16 S mNGS), which had 100% sensitivity when accounting for the genus level of detection [20]. All amplified regions were sequenced. Sequencing of multiple variable regions allows for higher resolution.
Quality filtering, trimming and dereplication of raw sequencing reads were conducted automatically on the Ion Reporter metagenomics workflow, relying on default parameters. We next used the UCHIME algorithm [21] to remove the chimeric sequences and used unoise3 [22] to generate denoising amplicon sequence variants (ASVs). Taxonomy assignment was performed based on vsearch [23] referring to the SILVA (V 138.1) [24] and GreenGene database [25] with a threshold of 97%. Reads with an alignment rate below 97% were rejected. We aligned reads from different variable regions with the bacterial reference genome separately. A consensus table was created by summing all read counts from different regions with identical taxonomic rank detection. We rarefied the sequencing data and then evaluated the alpha diversity by the Good’s average index, Chao1 index, abundance-based coverage estimator (ACE) index, Shannon index and Simpson index. Permutational multivariate analysis of variance (PERMANOVA) and analysis of similarity (ANOSIM) based on the Bray‒Curtis distance were used to evaluate the beta diversity. Differential bacterial taxa among groups were obtained using linear discriminant analysis effect size (LEfSe) [26] with the criteria of LDA > 2 (or LDA > 4) and P < 0.05. Microbiome phenotypes were predicted using BugBase [27]. PICRUSt2 [28] was used to identify microbiome-associated pathways from the inferred metagenomes of taxa using the ‘stratified’ option. We applied the Pearson correlation algorithm to identify associations across bacterial genera, representing correlation strength and assigned them to the edges. For the Pearson correlation table, we used the cytoHubba [29] plugin in Cytoscape [30] (V 3.9.1) to find Hubba nodes based on the maximum cross-correlation algorithm. The Hubba nodes represent the taxonomy that had the highest correlation with the other genera. Then, we took the intersection of correlation nodes and the top 10 Hubba nodes and retained nodes whose absolute correlation value was greater than 0.6.
Parametric continuous variables are presented as mean ± standard deviation, and abnormally distributed continuous variables are presented as medians and interquartile range (25th and 75th percentiles). Categorical variables are described as numbers (percentages). Student’s t-test and analysis of variance (ANOVA) with post hoc Tukey HSD test were used to compare continuous parametric data conforming to normal distribution. Abnormally distributed continuous variables were compared using the Mann‒Whitney U test or Kruskal‒Wallis H-test. Categorical variables were analyzed using the chi-square or Fisher’s test. All tests were two-sided. P values were corrected using FDR, and P < 0.05 was considered statistically significant [31]. We performed statistical analyses by using SPSS version 25 software.
We investigated the demographics of the participants, including several indicators that may affect the oral microbiome [32, 33]. The statistical results are presented in Table 1. The mean age was 32.03 years old, and 76.9% were female. A total of 58.5% of the participants were nursing staff, and the remainder were resident doctors. The mean seniority was 7.86 years. The median number of sleeping hours per day and intake times of sweets/desserts per week were 6.50 and 3.00, respectively. There were no significant differences in these parameters among the CCU, ICU, OR, and RES groups. All participants were free of metabolic diseases such as diabetes and hyperlipidemia.
The rarefaction curves of all samples demonstrated that the sequencing coverage was sufficient to represent the microbial composition (Supplementary Material Fig. S1). A total of 13,251 ASVs were identified in all 65 samples. After filtering out ASVs found in less than two samples, 1519 ASVs were reserved for downstream analysis with a minimum relative abundance of 0.05% [34]. We first assessed alpha diversity within individuals. The CCU group showed the lowest level of microbial diversity compared to the other groups, which was reflected in the Shannon and Good’s coverage index (Fig. 1A, Supplementary Material Fig. S2). In terms of beta diversity, PERMANOVA based on Bray–Curtis distance hardly exhibited obvious intergroup clustering of microbiota structure, even if statistically significant (Fig. 1B). Furthermore, ANOSIM analysis for pairwise comparison showed that the bacterial community of the CCU provided the most significant between-group differential components (Supplementary Material Fig. S3). We also tested whether the oral microbiome of healthcare workers is associated with other demographic characteristics. The department proved to be the main grouping factor accounting for the variance of the oral microbiota compared to other confounders, such as age, sex, position, diet, and sleep, although there was also a significant difference between those with 5–10 years of seniority and those with more than 10 years of seniority (Supplementary Material Fig. S3, P = 0.027). Taken together, these findings revealed prominent differences in oral microbial structure and diversity among the departments.
The microbiome profile comprised 13 phyla, 20 classes, 29 orders, 53 families, and 97 genera. Core phyla were defined as those identified in all samples. Nine core phyla, Bacteroidetes, Firmicutes, Proteobacteria, Actinobacteria, Fusobacteria, Candidatus Saccharibacteria, Spirochaetes, SR1, and Synergistetes, are shown in Fig. 1C. We conducted the Kruskal–Wallis test to compare the relative abundance of the nine core phyla across groups, and Spirochaetes was significant (Fig. 1D, CCU versus ICU P = 0.000; CCU versus RES P = 0.001; ICU versus OR P = 0.005; OR versus RES P = 0.043). We further conducted LEfSe analysis to identify significant differences in abundance between departments, considering CCU as the basic group. We identified 66 microbial taxa (19 CCU-enriched versus 47 ICU-enriched) that differed significantly in relative abundance between CCU and ICU, 36 microbial taxa (14 CCU-enriched versus 22 OR-enriched) that differed significantly in relative abundance between CCU and OR, and 40 microbial taxa (10 CCU-enriched versus 30 RES-enriched) that differed significantly in relative abundance between CCU and RES (Supplementary Table S1). To identify more reliable differential taxa, we set a stricter filter with LDA > 4. The relative abundance of Haemophilus in the CCU group was higher than that in the other groups, while Prevotella showed the opposite trend. Moreover, compared with ICU and RES workers, CCU workers showed a higher abundance of taxa belonging to Fusobacteria (Fusobacterium) and Firmicutes (Streptococcus). The relative abundance of Bacilli, a class belonging to Firmicutes, in the CCU suggested a level of depletion to the OR (Fig. 2A-C, Supplementary Material Fig. S4). We further evaluated the effects of various grouping characteristics on the oral microbiome. Factors such as age, sex, position, diet, and sleep could not be used to distinguish the differential microbiota. Notably, when we grouped the participants by seniority, several differential bacterial genera were identified (Supplementary Material Fig. S5), suggesting that oral microbial diversification of healthcare workers is related to seniority. We next performed co-occurrence network analysis and found vital interconnections within the oral microbiome, indicating that these healthcare-worker-altered taxa did not occur independently in the oral environment. Megasphaera, Prevotella, Leptotrichia, Atopobium, and Veillonella may be essential genera that shape the oral microbiome of healthcare workers, accompanied by rich multivariate interrelationships and strong correlations between each other (Fig. 2D). Porphyromonas was negatively correlated with Prevotella, whereas the rest were positively correlated.
We further explored the distribution of critical periodontal pathogens among departments [35–40]. The relative abundance of F. nucleatum increased significantly in the CCU group compared to the OR group, but compared to the ICU group the increase was not significant (Fig. 2E). The relative abundance of P. gingivalis in the ICU was higher than that in the OR or CCU. However, the relative abundance of P. gingivalis was approximately 10 times lower than that of F. nucleatum (Supplementary Material Fig. S6). These results suggest a potential risk of periodontal disease in the CCU and the ICU.
We analyzed the predicted phenotypes based on taxonomic classification using BugBase. BugBase categorized six main bacterial phenotype categories: Gram staining, oxygen tolerance, ability to form biofilms, mobile element content, pathogenicity, and oxidative stress tolerance. Phenotypes were inferred based on experimental data and pathway/gene presence information collected from various databases, such as Integrated Microbial Genomes (IMG) and the PathoSystems Resource Integration Center (PATRIC) [27]. BugBase data between departments were compared using pairwise Mann-Whitney-Wilcoxon test. Facultative anaerobic bacteria were more abundant in the CCU group than in the other groups (CCU versus ICU, P = 0.017; CCU versus OR, P = 0.047; CCU versus RES, P = 0.048; Supplementary Material Fig. S7A). The relative abundance of gram–positive bacteria in the CCU and OR groups was higher than that in the RES group (CCU versus RES, P = 0.000; OR versus RES, P = 0.043), whereas the opposite was true for gram–negative bacteria (CCU versus RES, P = 0.000; OR versus RES, P = 0.043; Supplementary Material Fig. S7B–C). PICRUSt2 was employed to impute MetaCyc pathway abundance from the original taxonomic assignment. In total, 399 pathways were annotated. Metabolic pathway data were compared by two-sided Welch’s t–test and filtered for false discoveries using the Benjamini-Hochberg method. Items with q-values less than 0.05 were considered significant. Sixteen differential pathways were elucidated between CCU and ICU, two of which were responsible for nucleotide biosynthesis (“pyrimidine deoxyribonucleotides de novo biosynthesis II” and “superpathway of purine nucleotides de novo biosynthesis II”) and were enriched in CCU (Fig. 3A). In addition, vitamin B12 synthesis was also upregulated in CCU(“adenosylcobalamin biosynthesis from cobyrinate a,c-diamide I” and “adenosylcobalamin salvage from cobinamide II”) (Fig. 3A). We identified 15 differential pathways between the CCU and OR groups. Nucleotide and vitamin B12 biosynthesis processes (“superpathway of purine nucleotides de novo biosynthesis II,” “pyrimidine deoxyribonucleotides de novo biosynthesis II,” “adenosylcobalamin biosynthesis from cobyrinate a,c-diamide I,” and “adenosylcobalamin salvage from cobinamide II”) were more active in CCU (Fig. 3B). Finally, the functional catalog, including biosynthesis and degradation of nucleotides, amino acids and starch, appeared to be enriched in the RES group compared to that in the CCU group (Fig. 3C). These results suggest that the predicted microbial functions of vitamin, nucleotide and amino acid metabolism were significantly different between the departments.
Healthcare workers are exposed to hospital environments and are constantly in contact with infected patients during daily work. High-risk exposure to transmissible bacteria affects not only the microbiome of the skin surface, but also the respiratory and digestive tracts. To the best of our knowledge, no study has adequately described the characteristics of the oral microbiome in healthcare workers. We demonstrated that the bacterial community diversity, structure, and potential function of staff in the CCU, ICU, OR, and RES departments differed markedly. Since late 2019/early 2020, the COVID-19 pandemic has led to general universal masking in healthcare settings. Our study reflects the differences in the oral microbiota composition of healthcare workers from different HCI environments during the pandemic. We selected a single-center hospital in Shanxi Province, China, for the study, which excluded the influence of diet, the living environment, and cultural background as much as possible [41–43]. It can be speculated that the original departments led to the different compositions of the oral microbial community. In the analysis of beta diversity, our data suggested that the CCU contributed the most significant between-group differences. In the subsequent comparison of oral microbial compositional differences and functional analysis, the CCU also exhibited robust features and stability. We inferred that the oral microbiome of the CCU healthcare workers received characteristic modulations from their departments. Thus, it was reasonable to consider the CCU as the basic object for comparison with other groups. Microbial distribution showed deviation among departments, with an increased abundance of Spirochaetes in the ICU and RES. Numerous empirical studies have shown that oral Spirochaetes cause damage to periodontal tissue by the direct effect of bacterial enzymes and cytotoxic products of bacterial metabolism [44–46]. It seems that the performance of F. nucleatum was more weighted than that of P. gingivalis because of its greater relative abundance. Taken together, these findings indicate that healthcare workers in different departments face specific risks of periodontal disease. At the genus level, the CCU group showed significant differences compared to other groups, with an elevated abundance of Haemophilus and decreased abundance of Prevotella, demonstrating a possible impact on the oral microbiome of healthcare workers in different clinical scenarios. In general, exposure to Haemophilus is most common in the department of respiratory medicine [47, 48]. However, the prevalence of Haemophilus spp. in the CCU was higher in our study. It has been reported that Haemophilus accounts for most gram–negative bacilli causing infective endocarditis [49–51], which is usually treated in the CCU. In contrast, the relatively depleted abundance of Haemophilus in the RES may represent stricter protection for healthcare workers in the Department of Respiratory Medicine. This study had several limitations. First, the study design was not longitudinal, and it lacked long-term tracking and analysis. Second, to explore the stabilizing effects, we selected enrollees working for over one year. More studies are needed to monitor oral microbial changes in short-term healthcare workers. Third, we did not clarify how nonbacterial microbiota (fungi, viruses, and archaea) contribute to the oral microbiome. Finally, despite the observation of different oral microbiomes among departments, we could neither decipher the causal relationships of the differences nor evaluate the influence of such differences on the health of the participants.
In this study, we provide a profile of the oral microbiome of healthcare workers and highlight the essential role of the HCI environment. Workers in the CCU are more likely to exhibit inherent microbiological characteristics, such as reduced diversity, significantly differentiated genera, and higher potential for periodontal diseases. Our study provides a reference for further understanding of the oral microbiological characteristics of healthcare workers. In light of our results, we propose that continuous monitoring of the oral microbiome of healthcare workers in different clinical scenarios should be considered to improve health.
Below is the link to the electronic supplementary material. Additional File 1: Supplementary Table S1 Additional File 2: Supplementary Figure |
|
PMC9648460 | 36357408 | Jeremy B. Ducharme,Zachary J. McKenna,Zachary J. Fennel,Roberto C. Nava,Christine M. Mermier,Michael R. Deyhle | Body fat percentage is independently and inversely associated with serum antibody responses to SARS-CoV-2 mRNA vaccines | 10-11-2022 | Diseases,Risk factors | Vaccination is widely considered the most effective preventative strategy to protect against severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) infection. An individual’s exercise habits, and physical fitness have been shown to impact the immune response following vaccination using traditional vaccine platforms, but their effects are not well characterized following administration of newer vaccination technology (mRNA vaccines). We investigated these effects on the magnitude of antibody responses following SARS-CoV-2 mRNA vaccination while accounting for known covariates (age, sex, time since vaccination, and the type of vaccine administered). Adults of varying fitness levels (18–65 years; N = 50) who had received either the Moderna or Pfizer SARS-CoV-2 mRNA vaccine between 2 weeks and 6 months prior, completed health history and physical activity questionnaires, had their blood drawn, body composition, cardiorespiratory fitness, and strength assessed. Multiple linear regressions assessed the effect of percent body fat, hand grip strength, cardiorespiratory fitness, and physical activity levels on the magnitude of receptor binding domain protein (RBD) and spike protein subunit 1 (S1) and 2 (S2) while accounting for known covariates. Body fat percentage was inversely associated with the magnitude of S1 (p = 0.006, β = − 366.56), RBD (p = 0.003, β = − 249.30), and S2 (p = 0.106, β = − 190.08) antibodies present in the serum following SARS-CoV-2 mRNA vaccination. Given the increasing number of infections, variants, and the known waning effects of vaccination, future mRNA vaccinations such as boosters are encouraged to sustain immunity; reducing excess body fat may improve the efficacy of these vaccinations. | Body fat percentage is independently and inversely associated with serum antibody responses to SARS-CoV-2 mRNA vaccines
Vaccination is widely considered the most effective preventative strategy to protect against severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) infection. An individual’s exercise habits, and physical fitness have been shown to impact the immune response following vaccination using traditional vaccine platforms, but their effects are not well characterized following administration of newer vaccination technology (mRNA vaccines). We investigated these effects on the magnitude of antibody responses following SARS-CoV-2 mRNA vaccination while accounting for known covariates (age, sex, time since vaccination, and the type of vaccine administered). Adults of varying fitness levels (18–65 years; N = 50) who had received either the Moderna or Pfizer SARS-CoV-2 mRNA vaccine between 2 weeks and 6 months prior, completed health history and physical activity questionnaires, had their blood drawn, body composition, cardiorespiratory fitness, and strength assessed. Multiple linear regressions assessed the effect of percent body fat, hand grip strength, cardiorespiratory fitness, and physical activity levels on the magnitude of receptor binding domain protein (RBD) and spike protein subunit 1 (S1) and 2 (S2) while accounting for known covariates. Body fat percentage was inversely associated with the magnitude of S1 (p = 0.006, β = − 366.56), RBD (p = 0.003, β = − 249.30), and S2 (p = 0.106, β = − 190.08) antibodies present in the serum following SARS-CoV-2 mRNA vaccination. Given the increasing number of infections, variants, and the known waning effects of vaccination, future mRNA vaccinations such as boosters are encouraged to sustain immunity; reducing excess body fat may improve the efficacy of these vaccinations.
Vaccination is widely considered the most effective preventative strategy to protect against severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) infection. While traditional vaccine platforms include inactivated or attenuated virus, or recombinant viral proteins, two prominent vaccines to combat SARS-CoV-2 infection, Pfizer (BNT162b2/Comirnaty) and Moderna (mRNA-1273), are both lipid-encapsulated mRNA-based vaccines. This newer vaccination technology has the notable advantage of eliciting higher antibody responses while also being more rapid to produce in large quantities compared to more traditional methods. Although the Moderna and Pfizer SARS-CoV-2 vaccines appear to be highly effective at preventing COVID-19, some individuals who have received the vaccine still contract the disease. For example, of 4,468 patients infected with SARS-CoV-2 in the Houston Methodist health care system between late November 2021 through January 5, 2022, 2,497 (55.9%) had been fully vaccinated (e.g., > 14 days after receiving two doses of the Pfizer [73%] or Moderna [22%] mRNA vaccines or one dose of the J&J [JNJ-78436735] recombinant adenovirus vaccine [5%]). Similarly, previous vaccination attempts for other viral diseases, such as influenza, have also observed breakthrough illnesses in individuals who have been vaccinated. The evolution of SARS-CoV-2 resulting in the advent of multiple variants has contributed to continued SARS-CoV-2 infection, even among vaccinated and previously infected individuals. To combat these variants, it is crucial to understand the wide inter-individual variability that alters the magnitude of protective immunity acquired after vaccination. The individual response to vaccinations varies substantially depending on the type of vaccine administered, age and sex of the individual, and time since vaccination. For example, recent studies have demonstrated that protective immunity can wane within 6 months post vaccination increasing the risk of SARS-CoV-2 infection. Importantly, modifiable factors related to one’s lifestyle such as an individual’s exercise habits, and physical fitness have also been shown to impact the immune response following vaccination. Researchers have demonstrated with previous vaccination programs that individuals who exercise regularly, as well as those that performed an acute bout of exercise within 30 min of vaccination, have improved antibody responses compared to their age-matched controls. Moreover, obesity is associated with impaired immune responses following the influenza vaccine and COVID-19 vaccines, whereas weight loss is associated with improved responses. Together these findings demonstrate that levels of physical activity and body fat can affect the efficacy of vaccination. General lifestyle recommendations including regular exercise to support immune function amid the COVID-19 pandemic have been issued, but the role of lifestyle factors on the prophylactic immune response to SARS-CoV-2 vaccination remains understudied. A further element of uncertainty is that two prominent SARS-CoV-2 vaccines currently being distributed in the United States and other countries are mRNA-based vaccines. Previous researchers have demonstrated that physical activity and obesity can contribute to heterogeneity of the immune response following vaccination with vaccines involving inactivated, attenuated, or recombinant viral protein vaccine platforms. Although a recent study showed that an acute bout of exercise after mRNA vaccination boosts antibody responses, the effect of regular physical activity and body fatness on the antibody response following mRNA vaccination is not known. Additionally, the impact that other components of fitness such as muscular strength and cardiorespiratory (aerobic) fitness may have on prophylactic immune responses to mRNA SARS-CoV-2 vaccines are also unknown. Therefore, the purpose of this study was to address this gap in the literature by investigating the effects of exercise habits and several aspects of physical fitness on the magnitude of the antibody responses following SARS-CoV-2 mRNA vaccination. Given that previous researchers have shown that age, sex, time since vaccination, and the type of vaccine affect the antibody response to vaccinations, the current study aimed to evaluate the effect of exercise habits and components of physical fitness on the antibody response following vaccination at the individual level while accounting for these covariates.
Using a cross-sectional study design, participants who had received their second dose of either the Moderna or Pfizer SARS-CoV-2 mRNA vaccine between 2 weeks and 6 months prior, arrived at the Exercise Physiology Laboratory for a single day of data collection in which they completed health history and physical activity questionnaires, and had their blood drawn before having their body composition, cardiorespiratory fitness, and strength assessed. All data collection was completed between April 2021 and September 2021 meaning that boosters for the Moderna and Pfizer SARS-CoV-2 mRNA vaccines were not yet available for our participants.
Data were collected on 60 participants (men, n = 26; women, n = 34) of varying fitness levels including university students, faculty, staff, as well as local community members. Participant demographics are presented in Table 1. Adults between the ages of 18 and 65 who received their first and second dose of either the Moderna or Pfizer SARS-CoV-2 mRNA vaccine between 2 weeks and 6 months prior to visiting the Exercise Physiology Laboratory for data collection were eligible for the study. Participants were ineligible if they did not meet the inclusion criteria or were (a) experiencing symptoms of COVID-19 (cough, fever, shortness of breath, loss of taste or smell, chills, body aches, sore throat, diarrhea), (b) had a diagnosis of an immune-compromising condition, or (c) taking a medication that alters heart rate. Participants were apparently healthy and had no known diseases or symptoms. All research procedures were approved by the University of New Mexico Institutional Review Board prior to starting this research (IRB # 05321, approved on April 17th, 2021). This research was performed in accordance with the Declaration of Helsinki. Each participant provided written informed consent before beginning the study.
The short International Physical Activity Questionnaire (IPAQ) was used to gather information about the participant’s exercise and physical activity habits. The short IPAQ is a widely used questionnaire that provides a reliable and valid assessment of physical activity habits. The questionnaire consists of seven questions related to the amount of time an individual spends exercising or working at low, moderate and vigorous levels of exertion over a typical 7-day period. When completing the questionnaire, participants were asked to recall their average 7-day level of physical activity that they had engaged in over the previous 6 months. The data gained from the questionnaire was processed accordingly to determine the average self-reported metabolic equivalent of task (MET) minutes/week.
Blood samples were collected through venipuncture of an arm vein into serum separator Vacutainers®. Blood samples were centrifuged at 1600xg for 15 min in 4 °C to separate serum. Samples were stored in 1 mL aliquots in a − 80 °C freezer until later analysis. Serum samples were assayed for antibodies that react with the following four SARS-CoV-2 proteins: receptor binding domain protein (RBD), spike protein subunit 1 (S1), spike protein subunit 2 (S2), and nucleocapsid (N) via MAGPIX multiplexing (Luminex xMAP Technology, San Diego, CA) and were reported according to their median fluorescence intensity (MFI). Because the SARS-CoV-2 vaccines used in this study only deliver messenger RNA that encodes for the spike protein, participants whose samples were positive for N protein-reactive antibodies were deemed to have been previously exposed to the live SARS-CoV-2 virus and were removed from further analyses. Intraassay coefficient of variations for S1, S2, RBD, and N antibodies were 5.35%, 4.53%, 8.95%, and 11.85%, respectively.
Participant height (cm) was measured using a stadiometer (Holtain Limited, Crymych, Dyfed, Great Britain) and body mass (kg) was recorded using a digital weight scale (MedWeight MS-3900, Itin Scale Company, Brooklyn, NY, United States). Participants’ body density was estimated using 3-site skinfold measurements and equations for men (chest, abdomen, and thigh) and women (triceps, suprailiac, and thigh). These measurements were used to estimate participants’ percent body fat (%BF) using the Siri equation.
Maximal cardiorespiratory fitness () was estimated via the Ebbeling et al. sub-maximal walking test on a motorized treadmill (Precor® C966, Woodinville, WA). Participants were fitted with a wireless heart rate monitor (Polar, Oulu, Finland) with the transmitter placed around their chest. Following a two-minute walking warm up at a slow speed (1.5 to 2 mph), the speed of the treadmill was increased to elicit a heart rate between 50 and 70% of the participants’ estimated heart rate maximum (220-age). After the appropriate heart rate was reached, incline of the treadmill was increased to a 5% grade. Heart rate was recorded again at minutes 3 and 4 during this stage. If the heart rate measurements varied by more than 5 beats per minute, the test was extended for another minute and heart rate was measured again. The final two heart rates were used to calculate estimated maximal cardiorespiratory fitness according to the Ebbeling et al. method.
Hand grip strength (HGS) of the dominant hand was measured using a manual spring-loaded dynamometer (A729-300 hydraulic hand dynamometer, Rolyan Ability One, Germantown, WI). The hand grip dynamometer was adjusted relative to the individual’s hand size, so that the bar was between the proximal and middle phalanges. While standing erect, the arm and forearm were positioned such that the shoulder was adducted and neutrally rotated, elbow flexed at 90°, and forearm in a neutral position. Participants were instructed to squeeze the handle of the dynamometer with maximal effort for approximately 3 s with no extraneous body movement. This process was repeated three times with 1-min rest between trials. The highest value was recorded in kilograms and normalized to body mass (kg of grip force ÷ kg of body mass) for analyses.
An a priori power analysis was conducted using G*Power software (version 3.1.9.6). Using a conservative estimate of effect size (f2 = 0.15), with one tested predictor (either %BF, relative HGS, estimated , or MET min/week) and four covariates (age, sex, time since vaccination, and the type of mRNA vaccine received) making a total of five predictors for each model, it was estimated with an α-level of 0.05, and a power of 0.80 (1–β) that 55 participants would be required to detect a statistically significant effect of medium size on the outcome variables (S1, S2, and RBD antibodies following SARS-CoV-2 mRNA vaccination). Differences among participant demographics were evaluated using two-tailed independent sample t tests. Four separate hierarchical multiple linear regression analyses were performed to assess the main effect of an individual’s %BF, relative HGS, estimated , and self-reported MET minutes/week on the magnitude of RBD, S1, and S2 antibodies following SARS-CoV-2 mRNA vaccination. Age, sex, time since vaccination, and the type of mRNA vaccine received (Pfizer or Moderna) were included as covariates due to their known associations with prophylactic immune responses to vaccination. These predictive models were defined as the body composition model (%BF and covariates), strength model (relative HGS and covariates), aerobic fitness model (estimated and covariates), and the physical activity model (MET minutes/week and covariates). Dummy codes were created for sex (0 = women, 1 = men) and type of mRNA vaccine received (0 = Moderna, 1 = Pfizer). The β coefficient for each predictor variable of interest was used to quantify the magnitude of their effect. Standard error (SE) was used as a measure of the statistical accuracy of an effect size. An α of 0.05 was used as a threshold for statistical significance. Analyses were performed using RStudio version 2022.02.2 + 485 "Prairie Trillium" for Windows. Case diagnostics were performed to ensure assumptions of linearity were met and correlations between variables were examined for multicollinearity. In all models, the standardized residuals were examined to confirm that assumptions of normality and homoscedasticity were met. Potential influential outliers in each model were examined by comparing the Cook’s distance values.
Case diagnostics identified two males and one female as potential influential outliers by having large Cook's distance values relative to the rest of the data set. To avoid violating the normal distribution assumption of regression analysis, these data points were excluded from further analyses. An additional seven participants were excluded from analyses as they were deemed to have likely had a recent SARS-CoV-2 infection based on high levels of antibodies against SARS-CoV-2 N (Fig. 1). Assessments of multicollinearity revealed that sex had a strong association with %BF (%BF = − 9.3 (sex) + 25.4, R2 = 0.340, p < 0.001), relative HGS (HGS = 0.14 (sex) + 0.52, R2 = 0.213, p < 0.001) and estimated (=12.5 (sex) + 37.50, R2 = 0.421, p < 0.001). This meant that on average, women had 9.3% more body fat, 0.14 kg less relative HGS, and an estimated that was 12.5 ml/kg/min less than men (Table 1). To avoid violating the assumption of collinearity, sex was removed from the models that investigated the effect of %BF, relative HGS, and estimated on the S1, S2, and RBD antibody response. No association was observed between the sex of the individual and the average amount of physical activity they engaged in each week (MET min/week = − 167.2 (sex) + 4758.7, R2 < 0.01, p = 0.954). Of the participants in the current study, 70% (35/50) received the Pfizer SARS-CoV-2 mRNA vaccine while 30% (15/50) received the Moderna SARS-CoV-2 mRNA vaccine.
To illustrate the range of individual’s antibody responses to vaccination, Fig. 2 shows the magnitude of antibodies present in the serum following SARS-CoV-2 mRNA vaccination. To investigate the effect of modifiable physical fitness variables on these responses multiple models were developed. The body composition model (F4,45 = 2.735, p = 0.040, R2 = 0.195) indicated a significant effect of %BF on the amount of S1 antibody present in the serum (β = − 366.56, SE = 129.2, p = 0.006, Fig. 3A). This meant that for every 1% increase in %BF there was a significant decrease in the amount of S1 antibody regardless of the individual’s age, time since vaccination, or the type of mRNA vaccine received. In the strength model (F4,45 = 0.624, p = 0.648, R2 = 0.052), no significant effect was observed for relative HGS on the amount of S1 antibody present in the serum (β = 1279.56, SE = 6732.23, p = 0.850). For the aerobic fitness model (F4,45 = 1.257, p = 0.301, R2 = 0.100), no significant effect was observed for estimated on the amount of S1 antibody present in the serum (β = − 199.95, SE = 128.09, p = 0.125). Using the physical activity model (F4,45 = 1.003, p = 0.427, R2 = 0.102), no significant effect was observed for physical activity on the amount of S1 antibody present in the serum (β = 0.27, SE = 0.17, p = 0.127). In agreement with the S1 antibody response, the body composition model (F4,45 = 3.201, p = 0.021, R2 = 0.221) indicated a significant effect of %BF when controlling for these covariates on the amount of RBD antibody present in the serum (β = − 249.30, SE = 80.13, p = 0.003, Fig. 3B). Using the strength model (F4,45 = 0.681, p = 0.609, R2 = 0.057), like the S1 antibody response, no significant effect was observed for relative HGS on the amount of RBD antibody present in the serum (β = 1612.46, SE = 4232.33, p = 0.705). Similar to the S1 antibody response, no significant effect was observed in the aerobic fitness model (F4,45 = 0.983, p = 0.426, R2 = 0.080) for estimated on the amount of RBD antibody present in the serum (β = − 92.64, SE = 81.61, p = 0.262). Using the physical activity model (F4,45 = 1.175, p = 0.336, R2 = 0.118), like the S1 antibody response, no significant effect of physical activity was observed on the amount of RBD antibody present in the serum (β = 0.19, SE = 0.11, p = 0.081). The body composition model (F4,45 = 1.299, p = 0.285, R2 = 0.103) indicated no significant effect of %BF on the amount of S2 antibody present in the serum (β = − 190.08, SE = 115.14, p = 0.106, Fig. 3C). Similar to the observation with S1 and RBD antibody levels, no significant effect was observed for relative HGS in the strength model (F4,45 = 0.628, p = 0.645, R2 = 0.053) on the amount of S2 antibody present in the serum (β = − 2342.00, SE = 5680.30, p = 0.682). Using the aerobic fitness model (F4,45 = 1.109, p = 0.364, R2 = 0.090), similar to the S1 and RBD antibody response, no significant effect was observed for estimated on the amount of S2 antibody present in the serum (β = − 153.74, SE = 108.74, p = 0.164). Again, like the S1 and RBD antibody response, the physical activity model (F5,45 = 1.144, p = 0.352, R2 = 0.115) indicated no significant effect of physical activity on the amount of S2 antibody present in the serum (β = 0.23, SE = 0.14, p = 0.120).
The primary aim of this study was to determine the effect of physical activity habits and multiple aspects of physical fitness—strength, aerobic fitness, and body composition—on the magnitude of antibody responses following SARS-CoV-2 mRNA vaccination. Poor responses to vaccination (e.g., lower viral protein specific antibody production) limits vaccine efficacy and may reflect an increased risk for infectious disease morbidity and mortality. The main finding of the present study was that %BF had a significant effect on the antibody response after accounting for the effects of covariates that could impact this response including age, time since vaccination, and the type of vaccine received (Fig. 3). Our statistical models demonstrated that %BF was an independent predictor of S1 and RBD antibodies present in the serum following SARS-CoV-2 mRNA vaccination. Specifically, in people with higher %BF there was a significantly lower amount of both the RBD and S1 antibodies present in the serum. While we lack sufficient evidence to suggest that an individual’s %BF has a statistically significant effect on the amount of S2 antibody present in the serum (p = 0.106), we demonstrated that there was still a negative effect (β = − 190.08) of %BF on the S2 antibody response. Therefore, the effect of %BF on the S2 antibody response may indeed be of practical significance and the directionality of the relationship between %BF and S2 antibody response is in agreement with the responses observed for the S1 and RBD antibodies. The current study lends support that as %BF increases there is a decreased efficacy of SARS-CoV-2 mRNA vaccination, possibly leading to a greater risk of an individual contracting SARS-CoV-2 and developing COVID-19 regardless of their age, type of mRNA vaccine received, and days since vaccination. Therefore, although vaccination is one of our primary strategies against viruses such as SARS-CoV-2, this data suggests that higher %BF could reduce the level of circulating SARS-CoV-2-specific IgG antibodies following SARS-CoV-2 mRNA vaccination. Obesity is characterized by chronic low-grade inflammation and is a risk factor for increased morbidity and mortality amongst SARS-CoV-2 infected patients. In one study, obese high fat-fed mice developed elevated markers of inflammation that were associated with lower levels of influenza-specific neutralizing antibodies as well as defective generation of effector memory CD8 + T cells after vaccination. The authors discussed the possibility that the impaired memory T-cell response could have influenced the poorer maintenance of neutralizing antibodies observed in their study. There is also evidence that obesity interferes with B-cell responses furthering the autoimmune inflammatory response following SARS-CoV-2 infection and increasing the severity of COVID-19. Aside from the mentioned physiological/immunological factors that could contribute to reduced vaccine effectiveness, anatomical factors in people with higher body fatness may also reduce effectiveness. For example, as individuals increase in weight and %BF, longer needle lengths may be required in order to ensure that the vaccine is deposited intramuscularly. Although vaccination amongst obese individuals will likely still result in some vaccine-conferred protection, needles that are too short and not individualized to the individual’s level of body fatness may result in the vaccine being deposited into a fat pad, which could reduce the immunogenic response to the vaccine. In agreement with these observations, previous researchers have demonstrated that excess body fat and obesity are associated with an impaired immune response and poor efficacy of other vaccines such as those against the hepatitis B virus, hepatitis A virus, and the influenza A virus. Importantly, these previous vaccines elicited an immune response via inactive, attenuated, or recombinant viral proteins and are different than the two vaccines evaluated in the current study (Moderna and Pfizer) which are both lipid-encapsulated mRNA vaccines. Therefore, the current study builds upon previous findings by extending them to SARS-CoV-2 mRNA vaccines (Moderna and Pfizer) and demonstrates that the effect of an individual’s %BF on the magnitude of their antibody response is independent of known factors that influence vaccine-induced immunity such as the individual’s age and time since vaccination. Importantly, chronic low level inflammation associated with obesity impairs the generation of antibodies following influenza vaccination. This may explain the inverse relationship between %BF and antibody responses observed in the current study following SARS-CoV-2 mRNA vaccination. Given the obesity epidemic and the fact that obesity is an independent risk factor for severe COVID-19, there is a need to enact strategies for increasing the efficacy of SARS-CoV-2 vaccines amongst this population. Based on the present results, one effective lifestyle-related strategy may be to reduce excess body fat. Although obesity can have a biological component and is therefore not always modifiable, there are behavioral and environmental factors that contribute to the development of obesity that can be modified. Proper nutritional and exercise interventions are often recommended as effective strategies for combating and preventing obesity. Decreasing excess body fat can improve the likelihood that the vaccine is deposited intramuscularly that may increase its immunogenic effect. Evidence supports that decreasing excess body fat can also decrease proinflammatory markers, therefore potentially protecting the generation of effector memory T-cells and neutralizing antibodies following vaccination. Results of the current study might indicate that decreasing excess body fat can positively affect the immunogenic effect of vaccination, a notion supported by recent evidence. Participating in regular exercise and being physical active is recommended for all adults. The negative health consequences of failing to meet adequate levels of physical activity are myriad and include a greater risk of COVID-19 leading to hospitalization and death. Whether physical activity levels impact the efficacy of prophylactic immune response to SARS-CoV-2 vaccination is not clear, but there is growing evidence that regular physical activity increases the efficacy of vaccination. While we did not observe a statistically significant effect of physical activity on the magnitude of S1 (p = 0.127), RBD (p = 0.081), and S2 (p = 0.120) antibodies following SARS-CoV-2 mRNA vaccination, in our a priori power analysis we assumed a moderate effect size (f2 = 0.15) between our predictor and outcome variables which may have overestimated the relationship between physical activity and antibody responses resulting in the current study being slightly underpowered to identify a statistically significant effect. Though not statistically significant, there was a positive effect of the average time spent being physical active on the magnitude of antibody response for S1 (β = 0.27), RBD (β = 0.19), and S2 (β = 0.23) suggesting that physical activity may be of practical significance for improving an individual’s antibody response following vaccination. In support of this, previously sedentary aged adults (~ 70 years) who were randomly assigned to a sham intervention had poorer influenza vaccine responses, while those assigned to a cardiovascular exercise program had a significantly greater seroprotection rate at 24 weeks after vaccination. Also, in a recent preprint, individuals who were more physically active had a greater S1/S2 antibody response following vaccination with an inactivated SARS-CoV-2 virus vaccine (CoronaVac). The current study builds upon these findings by suggesting that increasing an individual’s weekly physical activity may potentially improve the antibody response following mRNA vaccinations as well.
With the increasing number of SARS-CoV-2 infections and variants, along with the known waning effects of vaccination, future mRNA vaccinations such as boosters are encouraged to sustain immunity. Our findings demonstrate that as %BF increases there is a decreased efficacy of SARS-CoV-2 mRNA vaccination, characterized as decreased S1, S2, and RBD antibodies present in the serum, which can lead to a greater risk of contracting SARS-CoV-2 and developing COVID-19 regardless of the individual’s age, type of mRNA vaccine received, and time since vaccination. Therefore, decreasing excess body fat may be an effective strategy to improve the prophylactic immune response following mRNA vaccination. In addition, increased physical activity may be a beneficial strategy to improve the antibody response following mRNA vaccination, but more research is warranted with larger sample sizes to substantiate this effect. These findings highlight several behavioral changes (reducing excess body fat and increasing physical activity) that individuals can take that may improve the efficacy of SARS-CoV-2 mRNA vaccination. |
PMC9648462 | Juliette Delhove,Moayed Alawami,Matthew Macowan,Susan E. Lester,Phan T. Nguyen,Hubertus P. A. Jersmann,Paul N. Reynolds,Eugene Roscioli | Organotypic sinonasal airway culture systems are predictive of the mucociliary phenotype produced by bronchial airway epithelial cells | 10-11-2022 | Biological techniques,Cell biology,Physiology,Medical research | Differentiated air–liquid interface models are the current standard to assess the mucociliary phenotype using clinically-derived samples in a controlled environment. However, obtaining basal progenitor airway epithelial cells (AEC) from the lungs is invasive and resource-intensive. Hence, we applied a tissue engineering approach to generate organotypic sinonasal AEC (nAEC) epithelia to determine whether they are predictive of bronchial AEC (bAEC) models. Basal progenitor AEC were isolated from healthy participants using a cytological brushing method and differentiated into epithelia on transwells until the mucociliary phenotype was observed. Tissue architecture was assessed using H&E and alcian blue/Verhoeff–Van Gieson staining, immunofluorescence (for cilia via acetylated α-tubulin labelling) and scanning electron microscopy. Differentiation and the formation of tight-junctions were monitored over the culture period (day 1–32) by quantifying trans-epithelial electrical resistance. End point (day 32) tight junction protein expression was assessed using Western blot analysis of ZO-1, Occludin-1 and Claudin-1. Reverse transcription qPCR-array was used to assess immunomodulatory and autophagy-specific transcript profiles. All outcome measures were assessed using R-statistical software. Mucociliary architecture was comparable for nAEC and bAEC-derived cultures, e.g. cell density P = 0.55, epithelial height P = 0.88 and cilia abundance P = 0.41. Trans-epithelial electrical resistance measures were distinct from day 1–14, converged over days 16–32, and were statistically similar over the entire culture period (global P < 0.001). This agreed with end-point (day 32) measures of tight junction protein abundance which were non-significant for each analyte (P > 0.05). Transcript analysis for inflammatory markers demonstrated significant variation between nAEC and bAEC epithelial cultures, and favoured increased abundance in the nAEC model (e.g. TGFβ and IL-1β; P < 0.05). Conversely, the abundance of autophagy-related transcripts were comparable and the range of outcome measures for either model exhibited a considerably more confined uncertainty distribution than those observed for the inflammatory markers. Organotypic air–liquid interface models of nAEC are predictive of outcomes related to barrier function, mucociliary architecture and autophagy gene activity in corresponding bAEC models. However, inflammatory markers exhibited wide variation which may be explained by the sentinel immunological surveillance role of the sinonasal epithelium. | Organotypic sinonasal airway culture systems are predictive of the mucociliary phenotype produced by bronchial airway epithelial cells
Differentiated air–liquid interface models are the current standard to assess the mucociliary phenotype using clinically-derived samples in a controlled environment. However, obtaining basal progenitor airway epithelial cells (AEC) from the lungs is invasive and resource-intensive. Hence, we applied a tissue engineering approach to generate organotypic sinonasal AEC (nAEC) epithelia to determine whether they are predictive of bronchial AEC (bAEC) models. Basal progenitor AEC were isolated from healthy participants using a cytological brushing method and differentiated into epithelia on transwells until the mucociliary phenotype was observed. Tissue architecture was assessed using H&E and alcian blue/Verhoeff–Van Gieson staining, immunofluorescence (for cilia via acetylated α-tubulin labelling) and scanning electron microscopy. Differentiation and the formation of tight-junctions were monitored over the culture period (day 1–32) by quantifying trans-epithelial electrical resistance. End point (day 32) tight junction protein expression was assessed using Western blot analysis of ZO-1, Occludin-1 and Claudin-1. Reverse transcription qPCR-array was used to assess immunomodulatory and autophagy-specific transcript profiles. All outcome measures were assessed using R-statistical software. Mucociliary architecture was comparable for nAEC and bAEC-derived cultures, e.g. cell density P = 0.55, epithelial height P = 0.88 and cilia abundance P = 0.41. Trans-epithelial electrical resistance measures were distinct from day 1–14, converged over days 16–32, and were statistically similar over the entire culture period (global P < 0.001). This agreed with end-point (day 32) measures of tight junction protein abundance which were non-significant for each analyte (P > 0.05). Transcript analysis for inflammatory markers demonstrated significant variation between nAEC and bAEC epithelial cultures, and favoured increased abundance in the nAEC model (e.g. TGFβ and IL-1β; P < 0.05). Conversely, the abundance of autophagy-related transcripts were comparable and the range of outcome measures for either model exhibited a considerably more confined uncertainty distribution than those observed for the inflammatory markers. Organotypic air–liquid interface models of nAEC are predictive of outcomes related to barrier function, mucociliary architecture and autophagy gene activity in corresponding bAEC models. However, inflammatory markers exhibited wide variation which may be explained by the sentinel immunological surveillance role of the sinonasal epithelium.
The possibility of making predictions of the lower airways using cells in the sinonasal compartment is extremely attractive as, (1) they require appreciably less intervention and time/cost resource to obtain which relaxes concerns for the participant and clinicians/physicians, (2) this, in turn alleviates the issues surrounding recruiting control participants and widens disease-related inclusion criteria, and (3) in technical terms, sinonasal sampling is more reproducible and, in our experience, returns a higher yield of viable cells. Given the shared structure and function of the upper and lower epithelium lining conducting airways, it is reasonable to presume nAEC share the same omics-profiles and encounter similar exogenous (e.g. air, microbial, occupational) and endogenous (e.g. nutritional, immune, pharmaceutical) exposures as their lower airway analogues. This underlies the united airway hypothesis which considers the entire airway as a continuum due to the coincidence of sinonasal disease with conditions of the lower airways, and observations that treatment of one site can ameliorate symptoms in the other. For example, observations in asthmatics show that up to 80% also suffer from chronic rhinosinusitis, and treating both sites improves clinical outcomes and can reduce exacerbations vs targeting them individually. Of course, this can be explained by (for example) abatement of pro-inflammatory and damage associated molecular factors that originate from the primary site of disease vs phenomena that are transmitted through the airway epithelium. Further, the sinonasal and lower airways have very different gross anatomical structure, distinctions in temperature and humidity, and their relative location means nAEC are the first to encounter environmental challenges. Nevertheless, it would be a major advantage if samples from the sinonasal cavity proved to be diagnostic (and/or prognostic) of lower airway disease. A recent study applied RNA-Seq technology to assess paediatric samples in the context of the unified airway hypothesis, and identified 91% homology of gene transcript profiles between matched bronchial and nasal cell isolates, with gene activity outcomes aligning with wheeze symptom scores. While a similar investigation in adults would likely resolve further differences due to increased incidence of disease and years of environmental/occupational exposures, this study provides evidence that nAEC may be well suited to proxy the lower airways for paediatric patients. Indeed, a study by Imkamp et al. in the context of COPD, identified appreciable overlap in gene expression and quantitative trait loci associated with cigarette-smoke exposure. However, they also noted frequent gene-dependent observations (as did the Kicic et al. 2020 study), that highlighted significant distinctions between the nasal and bronchial compartments. Gene-specific differences have been reported in at least three further transcriptome-based studies; one investigating whether bAEC expression signatures could serve as a predictor of cigarette smoke-induced damage to the lower airways, another for the early diagnosis of lung cancer, and in the context of cystic fibrosis, Ogilvie et al. concluded that nAEC are not reliable surrogates for the lower airways. Discordance among investigations that assess clinical isolates (vs cultures) gives reason to the absence of clinical (or diagnostic) applications that may be used to predict lower airway disease with sinonasal biopsies. Bourdin et al. provides an excellent discussion of the considerations that disconnect the nasal and bronchial epithelium in the context of the unified airway disease hypothesis. Perhaps the most significant being that the nasal airway epithelium encounters and mitigates significantly more environmental exposures than that of the lower airways. Indeed, the immunological consequences leading to COVID align with a high load of SARS-CoV-2 within the sinus epithelium vs the lower airways. Altogether, these reports point towards differences that give rise to heighten resilience and immunological surveillance for nAEC vs their lower airway counterparts. Another consideration is whether nAEC are valid surrogates for bAEC in laboratory models, and thereby provide a generally applicable (and affordable vs commercial sources), ex vivo model of the airways with greater predictive qualities than current options. nAEC are objectively better than bAEC lines which (for example) require artificial genetic aberrations (e.g. unlimited proliferation in non-physiological media), and they retain few characteristics synonymous with normal physiological activity. For example, we have shown that primary nAEC models respond more effectively to microbial exposures and produce distinct secretory profiles compared to bAEC lines purported to be “close to normal”. Despite this, it is difficult to reconcile the diverse outcomes from the paucity of investigations that directly compare nAEC and bAEC models. Mihayola et al. compared undifferentiated models (albeit using commercially sourced primary cells), and convincingly showed differences in RIG1 activity, coupled with distinct interferon and survival responses, that confer heightened antiviral activity in nAEC cultures. Divergent outcomes have also been reported for the production of influential inflammatory factors (e.g. IL-6 and TLR-4) by Comer et al., who cautioned against applying nAEC cultures in the context of COPD. In contrast, McDougall et al. found significant correlation for inflammatory profiles between submerged nasal and bronchial cultures (stimulated with e.g. IL-1β and TNFα), despite using a heterogenous cohort of non/smokers with a range of respiratory conditions. However, even fewer studies apply a tissue engineering approach to assess organotypic models, which is essential to emulate (for example) omics profiles and mucociliary function observed in vivo. To our knowledge, this has been performed rigorously on one occasion using cystic fibrosis paediatric samples differentiated at ALI. While they demonstrate nAEC and bAEC share comparable electrophysiological properties, the authors cautioned that it is less likely that these results inform nAEC as surrogates in conditions that bias lower airway disease, and that their use of conditionally reprogrammed cells may have diminished cell-specific differences in gene activity. A likely scenario is that nAEC and bAEC share a great majority of genetic and biochemical processes, but which become disparate due to modifications brought about by age, exposures and disease. Hence, here we apply a first principles approach and assess organotypic ex vivo models of the sinonasal and bronchial airway epithelium derived from healthy participants. Our findings show that the models are indistinguishable in terms of morphology and barrier function, and this may also be the case in vivo. Conversely, transcript abundance of inflammatory markers demonstrated appreciable variation between the cultures, particularly in contrast to genes that regulate the evolutionarily conserved process of autophagy.
We coupled the ALI culture model with microscopic analysis techniques and assessed the comparative structure of differentiated nAEC vs bAEC epithelia (Fig. 1). Histological examination showed that either progenitor stem cell population, when cultured on transwell membranes, effectively differentiate to form a three dimensional tissue architecture (100% success rate), which enabled the production of the mucociliary phenotype (Fig. 1A). For all samples (and visual fields), hematoxylin and eosin (H&E) staining resolved ciliated cells, and alcian blue/Verhoeff–Van Gieson staining resolved mucus-producing goblet cells which formed a continuous airway surface fluid layer (blue staining). Observations made during routine light microscopy showed that nAEC produced detectable amounts of airway surface fluid earlier (approximately day 8–10, vs 12–14 days for bAEC), whereas ciliogenesis (and cilia beating) was first seen for the bAEC cultures (day 14–16, vs 16–18 for the nAEC). Quantification of the H&E stained sections returned no significant difference for measures related to cell density (nuclei per µm2; nasal = 6.048/μm2 vs. bronchial = 6.281/μm2, 95% CIs [5.520, 6.58], [5.59, 7.00], P = 0.55, n = 5) and the width of the epithelial layer (µm basal to apical cell margin; nasal = 5.585 μm vs. bronchial = 5.11 μm, 95% CIs [3.96, 7.21], [4.25, 5.95]; P = 0.84; n = 5; Fig. 1C). As both nAEC and bAEC-derived epithelia exhibited similar cell density, we measured a fluorescence signal directed to acetylated α-tubulin to compare their respective ciliogenic potential. There was similarity between the frequency of cilia quantified for either culture (Fig. 1A,C; MFI of acetylated α-tubulin; nasal = 0.94 vs. bronchial = 1.25, 95% CIs [0.69, 1.18], [0.70, 1.80], P = 0.41, n = 5), which was consistent with the magnitude of difference observed for cell density. These results also indicate a proportionally similar frequency of ciliated AEC, and cilia per AEC, shared by these cultures. Observations from high sensitivity microscopy (SEM) confirmed that the distribution of cilia was approximately equal for either AEC culture, and that cilia morphology (e.g. consistently > 6–8 microns in lengths) and clustering were indistinguishable between the two models (Fig. 1B).
Critical to the airway epithelium are distinct cell types which interact via tight junction (TJ) complexes to coordinated emergent barrier properties and the mucociliary escalator. Hence, we assessed the formation of TJ by quantifying ion conductance of the paracellular pathway as basal progenitor nAEC and bAEC differentiated to produce apical ciliated and goblet cells. Interestingly both cultures return high trans-epithelial electrical resistance (TEER) values (days 1–14) that converge and stabilise to approximately 625–650 Ω cm2 after the day-16 observation interval (P > 0.05 from day 16–32 when comparing magnitudes for nAEC and bAEC-derived cultures). We were unable to determine whether this phenomena represents an artefact of cellular turnover that restricts ionic passage until differentiation starts to occur, or whether this reflects a physiological response indicative of a distinctions in a wound healing-like process (Fig. 2). With the exception of day 1 (24 h post-airlift), TEER measures before the 16 day interval show that the nAEC cultures are significantly more effective in impeding the flow of ions through the paracellular pathway. This may reflect heightened goblet cell activity for these cultures, as routine light microscopy resolved more secretion products for the nAEC cultures, and which were noticeably more difficult to clear using PBS (e.g. cilia beating was more difficult to resolve unless the PBS wash step was performed). Why the day 1 observation is contrary to this pattern is unclear. Given that the convention is to challenge (and/or apply end-point analyses) to ALI cultures 28–32 after days after air-lift, means that either may be used to predict phenomena related to TJ activity and barrier function.
The relative abundance of individual components of the TJ apparatus was assessed between cultures at the end of the observation interval (day 32). There was no significant distinction identified for either TJ protein examined for the nasal and bronchial-derived epithelial cultures (Fig. 3A). However, the prolonged period (out of the in vivo situation) required to differentiate progenitor AEC into epithelia may attenuate the differences in nasal vs bronchial-derived stem cells beyond the sensitivity afforded by Western blot analysis. To mitigate this, TJ proteins were also assessed using AEC biopsy samples isolated directly from the bronchoscopy brush. A similar result (as the ex vivo model) was resolve for TJ protein abundance from the nasal sinus and bronchus sourced directly from the airways, and a similar pattern of expression was seen for each TJ factor (Fig. 3B). This result suggests that the conditions used to generate differentiated organotypic epithelial cultures does not distort the relative abundance of TJ proteins observed in participant-derived airway samples.
The products of inflammatory-gene activity with major significance to the nasal and bronchial airway epithelium were compared using a rational transcript analysis experimental design. Given inflammatory genes are highly inducible and subject to variation, the evolutionary conserved process of autophagy (present in all kingdoms except bacteria) was assessed simultaneously. Shown in Fig. 4, the relative abundance of inflammation-related and autophagy transcripts were similar for either cell/culture type (indicated with dashed lines in Fig. 4), with the exception of TGFβ and IL-1β, which exhibited a modest increase in the nAEC model (P < 0.05). In contrast, with the exception of LAMP-3, the gene activity for the regulators of autophagy was tightly controlled as indicated by similar effect sizes among the study cohort (Fig. 4), and confined effect-size distribution (i.e. broader uncertainty intervals) observed for the transcripts related to inflammation.
Ethics approval to perform airway brushings was from the Central Adelaide Local Health Network Human Research Ethics Committee. Methods were conducted in accordance with the Declaration of Helsinki and with the understanding and the written consent of each participant. Airway brushing was performed by the physicians at the Royal Adelaide Hospital’s Department of Thoracic Medicine. Nasal and bronchial samples were obtained from consenting participants with no history of chronic respiratory disease and were never smokers (n = 10; FEV1/FVC median 93% ± 12 [SD], three female, median age 42 ± 16 years [SD]; Table 1). Human primary nasal epithelial cells were sampled from the inferior turbinate mucosa via nasal brushings and bronchial epithelial cells were obtained from the right main bronchi using bronchoscopy and catheter brush. Bronchial and nasal AEC isolates were subject to the same culture condition, as previously described. Briefly, AEC were rinsed from the sampling brush into complete RPMI media, pelleted (300 g, 7 min, 10 °C), and resuspended in Bronchial Epithelial Growth Media (Lonza, Mt. Waverley, VIC, Australia). Cell suspensions were propagated in collagen coated T25 flasks (Sigma-Aldrich, Castle Hill, NSW, Australia) to expand basal progenitor AEC. Monolayers achieved 90% growth density after 8–12 days and were transferred to collagen coated (StemCell Technologies, Sunrise Beach, QLD, Australia) transwells (0.4 µm pores, 6.5 mm diameter; Sigma-Aldrich) and grown in ALI Growth Media (Lonza), at a seeding density of 9 × 104 cells per well. Once cells achieved 100% density (4–6 days), media was removed from the apical and basal reservoirs, and bronchial ALI Differentiation Media (Lonza) was added to the basal reservoir. Media was changed every second day and the apical surface was cleared of excess mucus every 4 days with three 30 s washes using PBS. Cultures were included in ex vivo models when mucociliary differentiation was observed, and transepithelial electrical resistance measures (EVOM2; World Precision Instruments, FL. USA) exceeded 500 Ω cm2 on the final culture day (32 days).
Cells cultured for 32 days at an ALI were prepared for staining with the University of Adelaide Histology group. Briefly, cells were fixed in situ in the transwell system with 10% neutral buffered formalin (1 h), after which the membrane/cells were cut from the transwell frame for dehydration (ethanol gradient) and embedded in paraffin wax. Longitudinal sections of the transwell membrane were cut into 0.4 µm sections and mounted onto slides. After xylene dewaxing and ethanol/water rehydration, H&E and Alcian Blue Verhoeff Van Gieson staining was performed by the SA Pathology Immunohistochemistry Department (Royal Adelaide Hospital). The cells were imaged using NDP.view2 software (version U12388-01) and the NanoZoomer automated slide scanner platform (both Hamamatsu Photonics, Higashi-ku, Hamamatsu City, Shizuoka Pref., Japan). NanoZoomer light microscopy images of H&E-stained ALI sections were imported into ImageJ (version 1.52r) and their sizes calibrated (pixel to µm). ImageJ colour thresholding and particle analysis methods were used to count the number of nuclei per unit area. Nuclei density was then reported as nuclei per 100 µm2. Average height for each ALI culture section was determined by first taking the area of the section, and then dividing this by the width of each section to give the average value for the height of the entire section in µm.
Transwell sections (prepared as above) were brought to water using the Histoclear II (National Diagnostics, Atlanta, GA, USA) and a reducing ethanol gradient. Heat mediated antigen retrieval was performed in sodium citrate buffer (10 mM sodium citrate, 0.05% Tween 20; pH 6.0). Samples were permeabilised in TBS-Triton X-100 (0.3% for 15 min) and blocking for 1 h at room temperature in Dako serum-free protein block (SFB; Agilent, Santa Clara, CA, USA) containing 5% normal goat serum (VectorLabs, Burlingame, CA, USA) and 0.1% Tween 20. To detect cilia, sections were incubated with rabbit anti-human acetylated α-tubulin (1:500; Cell Signaling Technologies, Danvers, MA, USA) overnight at 4 °C followed by secondary antibody (goat anti-rabbit Alexafluor 568; 1:700) (ThermoFisher Scientific, Waltham, MA, USA) for 1 h at RT. Sections were mounted with Prolong Diamond antifade (ThermoFisher Scientic, Waltham, MA, USA) containing 4′,6-diamidino-2-phenylindole counterstain to resolve nuclei. Images were acquired using a Nikon Eclipse Ts2 microscope with NIS-Elements imaging software (version 5.20.00; both Nikon, Tokyo, Japan). Five randomly selected microscopy fields were used to determine mean fluorescence intensity of acetylated α-tubulin for the nasal- and bronchial-derived epithelial cultures.
High-resolution topographical epithelial morphology was assessed in situ with a previously described method using scanning electron microscopy (SEM). Briefly, fixative agents were applied directly to the cells on transwell (3 h) at the end of the culture period. The transwell/cells were then excised from the transwell frame, dehydrated and coated with platinum to for ultra-high resolution imaging to discern individual cilia. Coated samples (n = 2 each from nAEC and bAEC-derived cultures) were mounted on to SEM pegs and imaged using the FEI Quanta 200 scanning electron microscope (FEI Australia Pty Ltd, Canberra, ACT, Australia) housed in the University of Adelaide Microscopy Suite.
Electrical impedance (to restrict ionic conductance) imparted by epithelial cultures was measured to quantify barrier function (TJ activity), as previously described. One difference here was the models were assessed during the entire differentiation process from day 1 (24 h after “air-lift”) to day 32. Briefly, transwell ALI cultures were allowed to acclimatise for 30 min on a 37 °C heating platform (Lecia Biosystems, Mt. Waverley, VIC, Australia) before reading electrical impedance using the EVOM2 Ohm meter (World Precision Instruments, Sarasota, FL, USA). Raw ohm values (n = 3 measures per well) were converted to magnitudes of TEER by subtracting the resistance imparted by an empty transwell and dividing by the surface area of the membrane support (0.33 cm2; Ω.cm2). Five transwells were prepared and assessed per participant per culture type.
As previously described, protein was isolated from differentiated AEC cultures in situ from the transwell membrane surface using M-Per mammalian cell protein lysis reagent and Halt® protease and phosphatase inhibitor cocktail (both Thermo Scientific, Victoria, Australia). Protein samples were quantified using the BCA protein assay method (Bio-Rad, Victoria, Australia), and 10 μg electrophoresed using Novex® 4–12% gradient Bis–Tris denaturing gels (Life Technologies, Victoria, Australia) and electroblotted to Trans-Blot® Turbo nitrocellulose membranes (Bio-Rad). Membranes were blocked in 5% diploma skim milk and probed overnight at 4 °C with primary antibodies directed to Claudin-1 (37–4900), Occludin-1 (71–1500), ZO-1 (33–9100) (all Thermo Fisher Scientific, Waltham, MA, USA), or β-actin (A1978; Sigma-Aldrich Co., St Louis, MO, USA). Secondary antibody incubation was 1 h at RT with horse radish peroxidase-labelled secondary antibody (R&D Systems, MN, USA). Chemiluminescent imaging was performed using the LAS-3000 platform and histogram densitometry was performed using Multi Gauge software (V3.1 Fujifilm, Tokyo, Japan).
Quantitative reverse transcription real-time PCR was used to identify differences in transcript abundance in differentiated nasal and bronchial cultures. RNA was extracted and genomic DNA eliminated from cell in situ while on the transwell membranes using the RNeasy Plus column system following the manufacturers protocol (QIAGEN, Chadstone, VIC Australia). Purified RNA was quantified using the NanoDrop One spectrophotometer (Thermo Scientific). RNA (1 µg) was subject to a second gDNA purification and reverse transcribed into cDNA with SuperScript IV VILO ezDNase (Thermo Scientific) using an Eppendorf Mastercycler (Eppendorf, Hamburg, Germany). cDNA samples (and reaction mix) were entered into a custom TaqMan low density array (Thermo Fisher Scientific Australia, Scoresby, VIC, Australia). qPCR was performed with 500 ng cDNA using TaqMan primer/probes technology in conjunction with TaqMan Fast Advanced Master Mix chemistry, in the microfluidics system (all Thermo Scientific). Thermocycling was performed using the ViiA7 7300 Real Time polymerase chain reaction platform (Applied Biosystems, Carlsbad, California, USA). Three internal reference genes were included to normalise outcomes across the experimental groups: RNA18S5 (Hs99999901_s1), HMBS (Hs00609297_m1) and TBP (Hs00427620_m1). The complete set of primer/probes are shown in Supplementary Table S1 online.
Immunofluorescence and IHC outcome results were analysed with R (v 4.0.5) using the ggpubr R package (v 0.4.0) using non-paired Wilcoxon tests. TEER results and Western blot densitometry results were reanalysed using multi-level generalised linear mixed models (GLMM) in Stata v 16 (StataCorp LLC, TX, USA). TEER results were analysed as a linear glmm with measurement times (days) as repeated measures within participants. Results were expressed as predicted marginal means with 95% confidence intervals. Western blots densitometry data was analysed as a gamma (log link) glmm and results expressed as a ratio relative to β-actin using R (v 4.0.5). qPCR analysis was performed in R (v 4.0.5) using the MCMC.qpcr (v1.2.4) package. MCMC.qpcr implements a Bayesian GLMM analysis of multi-gene qPCR data in a single model with multi-gene normalisation to endogenous controls, individual random effects for each sample, and gene-specific variances. Endogenous controls were 18S rRNA, hydroxymethylbilane synthase (HMBS) and TATA-box binding protein (TBP). Results were expressed as log2(Relative Expression).
Written informed consent was obtained by all participants at time of recruitment. Ethical approval was obtained from the Central Adelaide Local Health Network Human Research Ethics Committee.
The airways epithelium is uniquely complex due to the range of activities required to maintain an extensive interface with the external environment and provide the conditions needed for efficient external respiration. Central to this, AEC have a range of functions that govern innate and adaptive immunity. Fundamental to these phenomena is a selectively permeable barrier that protects (and is protected by) the airways via coordinated mucociliary interactions, which also surveille the atmosphere with an array of pattern recognition receptors. Consequently, pulmonary function (ergo immediate life), relies on inter-AEC communication and the relay of environmental stimuli to professional immune cells that are proportional and specific to potentially harmful inhaled particles. Accordingly, the central mechanisms linked to incurable respiratory diseases such as asthma and COPD are frequently found to originate from dysfunction in the homeostatic and regenerative potential of the airway epithelium. Assessing discrete disease-related phenomena in the clinic remains challenging as the airway epithelium exhibits a complex structure–function relationship and can rapidly initiate immune hyperreactions that exacerbate disease. Organotypic models derived from clinical samples can obviate many of these issues and answer fundamental questions that are not amenable to clinic research. However, sampling the lower airways remains a relatively invasive procedure that requires significant resource at the operating theatre. This can greatly limit (or prohibit) the frequency and diversity of samples necessary to realise objective outcomes. Despite the advantages of nasal cell cultures, there is insufficient agreement to support modelling the lower airway using cells from the sinonasal cavity. Here we provide evidence that organotypic systems of the nAEC and bAEC exhibit the same barrier activity within the scope of our model and analysis techniques. However, we add further evidence that there is significant uncertainty as to whether predictions can be made about gene activity in the lower airways using nAEC models. We chose to assess barrier function as most activities governed by the airway epithelium emerge from effective cell-to-cell interactions. Also, it is reasonable to assume that the similarities between the sinonasal and bronchial epithelial architecture arise from analogous underlying regulatory mechanisms, which serve to minimise uncertainty related to these complex processes. Routine laboratory observations showed similar growth and morphological changes during progenitor cell differentiated (into a mature epithelium), which were supported by histological, immunofluorescence and high power microscopy endpoint measures (Fig. 1). Important to find a consensus between the modest number of reports in this area, our measures for growth, morphology and ciliogenesis (for the most part) align with observations made using diverse clinical and laboratory models of the airway epithelium. However, endpoint analyses provide limited information about the processes that potentiate the mucociliary phenotype. In line with this, we observed higher resistance measures in nAEC during the 0–16 day differentiation interval (Fig. 2). This may reflect differences in TJ synthesis that favour nAEC, which is consistent with the heightened resilience that is ascribed to the sinonasal epithelium (e.g.). Conceivably, accelerated TJ formation between nAEC would expedite healing (hence, return of function) and sentinel activity to protect the lower airways. Indeed, rapid TJ synthesis is a likely contributing factor in the maladaptive response observed in chronic rhinosinusitis that promotes nasal polyposis; a phenomenon that is not observed in the lungs. While we were unable to spare cultures to assess the abundance of Claudin-1, Occludin-1 and ZO-1 corresponding to each resistance measure, we found that (in agreement with the outcomes for TEER), the abundance of these essential TJ factors were similar in both cultures (Fig. 3A). Likewise, nAEC and bAEC taken directly from biopsies (i.e. from the cytology brush in vivo) expressed approximately the same amount of TJ proteins (Fig. 3B), and the relative abundance of Claudin-1, Occludin-1 and ZO-1 were similar to those observed in the ex vivo models (Fig. 3). Hence, our results support organotypic models of the sinonasal epithelium as a reliable proxy to predict barrier function in their bAEC counterparts. However, comparisons made prior to full differentiation may be confounded by differences in the rate of TJ synthesis. In contrast, we did observe variation in gene activity between the two epithelia for inflammatory factors commonly assessed as markers of disease at the airway epithelium (Fig. 4). The tendency for dichotomous results favouring increased transcript abundance of TLR-4, IL-6, TNFα, MMP9, and VEGF in nAEC generally agreed with previous reports (e.g.). Further to these, we observed significant elevation (in nAEC cultures) of the influential regulators of airway inflammation TGF-β and IL-1β (Fig. 4). Most of these inflammatory factors participate in early responses to microbial and allergic challenges imparted by the environment. So we cannot rule out that their elevation in nAEC may reflect the sinonasal labyrinth as the front-line defence against randomly encountered airborne factors (although our participants were asymptomatic), rather than any systematic difference. This is underscored by the wide confidence intervals for the majority of the inflammatory factors (with the exception of TSLP), and difficulties inherent to controlling environmental exposures encountered by study participants. Hence, it appears important to consider which genes (or gene networks) best apply to inflammation in the lungs when using nAEC as a surrogate model. To support this recommendation, we also assessed autophagy, which is an ancient and evolutionarily conserved cellular survival process, and unlike inflammation, is a fundamental aspect of all eukaryotic cells. In contrast to the outcomes for inflammation, there was appreciably less variation among autophagy genes (n = 29) vs the genes involved in the immunological response (n = 11). A central reason for this is autophagy is primarily an intracellular process and generally only participates in activities elicited by exogenous stimuli when they become intracellular or they restrict nutrient availability. In contrast, inflammation is orchestrated at an organismal level and requires a variety of cells (to name one intricacy). This involves complex gene-environment interactions that lend to uncertainty, particularly when trying to assign clinical or biological relevance using a data-driven approach. Hence, validation studies are advised for particular biochemical/physiological phenomena when considering the transcriptome profile of nAEC as a means to predict gene-activity in the lungs. Assessing the sinonasal compartment for the presence of SARS-CoV-2 has demonstrated the exceptional utility of sampling the upper airway (vs the lungs or peripheral blood), as a non-invasive means to diagnose respiratory disease. But there remains an urgent need for similar methods that predict the onset of chronic lung diseases before they become irreversible. Here, we have shown that differentiated models of the sinonasal epithelium are an effective proxy to assess barrier activity in analogous bronchial systems. While the abundance of ciliated cells are central for restricting environmental agents, we did not directly assess the important contribution imparted by the basal progenitor cells, which also give rise to (for example) secretory cells that maintain the airway surface liquid. We are now trying to determine whether barrier dysfunction in nAEC is predictive of COPD, which will incorporate a comparison of the basal and secretory cells as they are principal components for this lower airway disease. We also advise that, at present, hypothesis-driven approaches to measure gene activity are an important source of reproducible data, and that autophagy is an example where gene-activity (at least in healthy individuals) is stable at the mRNA level. While a common aetiology remains elusive, there is a clear correlation between the incidence of chronic rhinosinusitis and lung diseases with an significant genetic component such as asthma. Further, there is evidence that points towards the airway epithelium as a conduit for disease-propagation between distant sites of the respiratory tract in conditions that manifest primarily as a result of exogenous factors. Progress in this domain can be well served in controlled laboratory systems, particularly as organotypic ex vivo models are becoming standardised and more reproducible. Outcomes from these efforts could significantly bolster renewed consideration to assess the sinonasal epithelium to expedite innovative strategies for fatal lung conditions. One ideal outcome would be the development of a cogent sinonasal transcriptome signature that enables early diagnosis and personalised treatment strategies for lung disease.
Supplementary Information. |
|
PMC9648470 | Rana Dilara Incebacak Eltemur,Huu Phuc Nguyen,Jonasz Jeremiasz Weber | Calpain-mediated proteolysis as driver and modulator of polyglutamine toxicity | 19-10-2022 | posttranslational modifications (PTMs),proteolytic cleavage,calpains,toxic fragments,Huntington disease (HD),spinocerebellar ataxia (SCA),dentatorubral-pallidoluysian atrophy (DRPLA),spinal and bulbar muscular atrophy (SBMA) | Among posttranslational modifications, directed proteolytic processes have the strongest impact on protein integrity. They are executed by a variety of cellular machineries and lead to a wide range of molecular consequences. Compared to other forms of proteolytic enzymes, the class of calcium-activated calpains is considered as modulator proteases due to their limited proteolytic activity, which changes the structure and function of their target substrates. In the context of neurodegeneration and - in particular - polyglutamine disorders, proteolytic events have been linked to modulatory effects on the molecular pathogenesis by generating harmful breakdown products of disease proteins. These findings led to the formulation of the toxic fragment hypothesis, and calpains appeared to be one of the key players and auspicious therapeutic targets in Huntington disease and Machado Joseph disease. This review provides a current survey of the role of calpains in proteolytic processes found in polyglutamine disorders. Together with insights into general concepts behind toxic fragments and findings in polyglutamine disorders, this work aims to inspire researchers to broaden and deepen the knowledge in this field, which will help to evaluate calpain-mediated proteolysis as a unifying and therapeutically targetable posttranslational mechanism in neurodegeneration. | Calpain-mediated proteolysis as driver and modulator of polyglutamine toxicity
Among posttranslational modifications, directed proteolytic processes have the strongest impact on protein integrity. They are executed by a variety of cellular machineries and lead to a wide range of molecular consequences. Compared to other forms of proteolytic enzymes, the class of calcium-activated calpains is considered as modulator proteases due to their limited proteolytic activity, which changes the structure and function of their target substrates. In the context of neurodegeneration and - in particular - polyglutamine disorders, proteolytic events have been linked to modulatory effects on the molecular pathogenesis by generating harmful breakdown products of disease proteins. These findings led to the formulation of the toxic fragment hypothesis, and calpains appeared to be one of the key players and auspicious therapeutic targets in Huntington disease and Machado Joseph disease. This review provides a current survey of the role of calpains in proteolytic processes found in polyglutamine disorders. Together with insights into general concepts behind toxic fragments and findings in polyglutamine disorders, this work aims to inspire researchers to broaden and deepen the knowledge in this field, which will help to evaluate calpain-mediated proteolysis as a unifying and therapeutically targetable posttranslational mechanism in neurodegeneration.
One of the most impactful posttranslational modifications (PTMs) is proteolytic fragmentation, a process in which proteases cleave and trim their target proteins, thereby directly affecting the structural integrity of the substrate. These changes, which range from removal of single amino acids up to ablation of entire domains, may have major consequences on the function, localization, interactome, and stability of the affected protein, including the availability of sequences for further PTMs. In comparison to other modifications such as phosphorylation, lipidation, or ubiquitination, proteolytic fragmentation is an irreversible process, strongly determining the fate of its substrate. The field of research, which deals with the entirety of these proteolytic processes, both in physiological and pathological contexts is termed degradomics (Rogers and Overall, 2013; Klein et al., 2018). In a more pathological context, proteolytic fragmentation can lead to the generation of detrimental protein species, which might harm affected cells and organs and culminate in medical conditions, as highlighted by the secretase-based processing of amyloid precursor protein (APP) in Alzheimer’s disease (AD; Selkoe and Hardy, 2016). In a further context of neurodegeneration, proteolytic fragmentation of disease proteins in polyglutamine (polyQ) disorders has been identified as pivotal modulators of their molecular pathology (Wellington and Hayden, 1997). Multiple proteases were associated with polyQ protein fragmentation, including apoptosis-associated caspases, lysosomal cathepsins, as well as matrix metalloproteinases. Calpains, a class of calcium-activated proteolytic enzymes, emerged as one of the key players in polyQ disorders (Weber et al., 2014; Matos et al., 2017). Here, we provide a comprehensive overview of the current knowledge on the involvement of calpains in this group of rare and yet incurable diseases, emphasizing the relevance of further investigations for improving our understanding of calpain-mediated cleavage and, thereby, uncovering new therapeutic avenues for neurodegeneration.
Calpains are a family of calcium-dependent, intracellular cysteine proteases implicated in the maintenance of cellular homeostasis and a variety of physiological processes (Ono and Sorimachi, 2012). Their initial discovery can be traced back to the early 1960s when the first calcium-activated neutral proteinase was extracted from rat brain (Guroff, 1964). Several years later, this class of proteases was collectively renamed calpains - combining “cal” as an abbreviation for calcium and reference to the calcium-binding protein calmodulin, with “pain” in a nod to structurally similar cysteine proteases like papain (Murachi et al., 1980). Calpains and their respective homologs are an evolutionarily distinct group of proteases and are found in nearly all unicellular and multicellular eukaryotes, including animals, fungi, and plants, as well as in some bacteria but not archaea (Goll et al., 2003; Sorimachi et al., 2011; Ono and Sorimachi, 2012). The highest diversity of calpain homologs can be found in animals, with the largest number of 15 different enzymes present in mammals, and thus also in the human genome. This circumstance has enabled researchers to investigate the physiological and pathophysiological roles of human calpains in classical rodent models such as mice and rats, being the foundation of a vast number of significant studies in the field (Goll et al., 2003; Sorimachi et al., 2011; Ono and Sorimachi, 2012). Calpains are primarily characterized by their highly conserved calpain-like cysteine protease domain (CysPc), which can be divided into the two core domains PC1 and PC2. Both core units bind one calcium ion each, leading to their conformational rearrangements and activation of the catalytic triad composed of cysteine, histidine, and asparagine residues (Arthur et al., 1995; Moldoveanu et al., 2002). Most calpains feature one central calpain-type β-sandwich (CBSW) domain with calcium-binding properties, followed by an additional C-terminal domain (Tompa et al., 2001; Ono et al., 2016). Based on their structural and domain-wise composition, the known 15 members of the calpain family in humans can be grouped into classical (calpain-1, -2, -3, -8, -9, 11–14) or non-classical calpains (calpain-5, -6, -7, -10, -15, -16; Table 1; Sorimachi et al., 2011; Ono and Sorimachi, 2012). Classical calpains share common structural features, comprising - aside from the aforementioned CysPc domain - a C-terminal penta-EF-hand (PEF) domain with five identical calcium-binding EF-hand motifs, of which four bind calcium ions (Figure 1A; Table 1; Blanchard et al., 1997; Lin et al., 1997). The fifth EF-hand motif is essential for the so-called conventional calpains, calpain-1 and calpain-2, to form inactive stable heterodimers with the regulatory calpain small subunit 1 (CSS1, earlier known as calpain-4). CSS1 is composed of an N-terminal glycine-rich region (GR) and a C-terminal PEF domain which, corresponding to the PEF domain in the large catalytic subunits, also binds four calcium ions and facilitates the interaction with conventional calpains (Figure 1B). It has been reported that, as a part of the calcium-dependent calpain activation process, CSS1 may dissociate from the large subunits. However, other studies suggested that CSS1 remains in the dimer when activated (Ravulapalli et al., 2009; Sorimachi et al., 2011; Ono and Sorimachi, 2012). Interestingly, knockout of CSS1 resulted in destabilization and activity loss of calpain-1 and calpain-2, and was found to result in embryonic lethality (Arthur et al., 2000; Zimmerman et al., 2000). Although all classical calpains are characterized by their C-terminal PEF domain, the function of CSS1 and interaction with other calpains remain unclear (Ono and Sorimachi, 2012). Moreover, the existence of a second small regulatory subunit (CSS2) is known, but its functions are even less understood (Schád et al., 2002; Ma et al., 2004). Unlike classical calpains, non-classical calpains appear in various structural compositions, and lack or have a replacement for typical domains like the PEF domain (Figure 1C, Table 1; Sorimachi et al., 2011; Ono and Sorimachi, 2012; Ono et al., 2016; Nian and Ma, 2021; Spinozzi et al., 2021). The expression of calpain family members ranges from wide distributions to very limited expression patterns. Whereas calpain-1, -2, -5, -7, -10, 13–16, and CSS1/2 are expressed ubiquitously (Sorimachi et al., 2011; Ono and Sorimachi, 2012), the remaining calpains are restricted to a tissue-specific expression (Table 1). For instance, calpain-3 or calpain-12 are solely expressed in skeletal muscle or hair follicle cells, respectively (Sorimachi et al., 1989; Dear et al., 2000). The calpain system does not only include various calpain enzymes and their regulatory subunit but also calpastatin (CAST), the only endogenous and ubiquitously expressed inhibitor of classical calpains (Ono and Sorimachi, 2012). CAST features four repetitive inhibitor domains with distinct specificities, by which it can bind, amongst others, to the proteolytic CysPc and PEF domains, looping out and around the catalytically active cysteine residue, eventually inhibiting calpain activation (Figure 1D; Hanna et al., 2008; Moldoveanu et al., 2008). Calpains are referred to as modulator proteases, as they do not randomly cleave and degrade proteins but perform a limited proteolysis on their substrates. More specifically, they remove distinct motifs and domains from their target proteins, thereby modulating their structure, function, and activity (Sorimachi et al., 2011; Ono et al., 2016). They participate in a variety of vital cellular processes, including cell cycle and proliferation, apoptosis, cell motility, signal transduction pathways (Goll et al., 2003), neurogenesis (Baudry et al., 2021), and synaptic plasticity (Baudry and Bi, 2016). Based on their wide spectrum of cellular functions, disruption, and dysregulation of calpains are associated with various diseases (Ono et al., 2016). Noteworthy, calpainopathies are specified as disorders explicitly caused by mutations in calpain genes. This family of disorders includes forms of limb-girdle muscular dystrophy and autosomal dominant neovascular inflammatory vitreoretinopathy (ADNIV), which are caused by mutations in calpain-3 and calpain-5, respectively (Gallardo et al., 2011; Mahajan et al., 2012; Vissing et al., 2016). For a comprehensive overview of diseases caused by, or linked to, mutations and variants in calpain system-related genes, see Table 1. Moreover, dysregulations of the intracellular calpain system are associated with neurodegenerative disorders, like AD, Parkinson’s disease (PD), amyotrophic lateral sclerosis (ALS), and polyglutamine (polyQ) disorders. Furthermore, calpains were shown to be implicated in other diseases, such as cancer, cardiovascular and ischemic disorders, and diabetes (Ono et al., 2016).
Aberrant proteolytic processing of disease proteins by proteases is a phenomenon observed in several neurodegenerative disorders. About three decades ago, disease-associated proteolysis was described for the first time in AD (Esch et al., 1990; Selkoe, 1994; de Strooper and Annaert, 2000). Here, altered cleavage of APP by secretases results in amyloid beta peptides that form highly insoluble amyloid plaques and have neurotoxic properties (Esler and Wolfe, 2001; LaFerla et al., 2007; Selkoe and Hardy, 2016). Some years later, the so-called toxic fragment hypothesis extended the concept to polyQ disorders by postulating that polyQ stretch-containing fragments of disease proteins are more toxic than their full-length forms (Wellington and Hayden, 1997). Also, in synucleinopathies such as certain forms of Parkinson’s disease (PD) or, for instance, in SOD1- or TDP-43-linked ALS, disease protein cleavage was identified as a disease-modifying factor (Wright and Vissel, 2016; Bluhm et al., 2021; Chhangani et al., 2021). While protein-degrading machineries, such as autophagy and ubiquitin-proteasome system, lead to a full disintegration of proteins to amino acids or peptides, multiple classes of proteolytic enzymes, including caspases, cathepsins, matrix metalloproteinases, secretases, and calpains were associated with the production of disease protein fragments in polyQ disorders (Weber et al., 2014; Matos et al., 2017). Due to the still incomplete picture of their degree of involvement in disease protein cleavage, and thus in the molecular pathogenesis, a reasonable ranking of these proteases’ disease-modifying significance in the context of polyQ disorders is unattainable. However, it is noteworthy that calpains, due to their particular function as modulator proteases, were found to act upstream of proteolytic processes, controlling degradational mechanisms such as autophagy, and regulating caspase activation under apoptotic or degenerative conditions (Yousefi et al., 2006; Gafni et al., 2009; Smith and Schnellmann, 2012; Sorimachi and Ono, 2012; Weber et al., 2019a). Based on these considerations, calpains emerge as important modulators of disease protein toxicity, thus representing excellent targets for therapeutic intervention (Weber et al., 2014; Matos et al., 2017).
Among the large group of inherited neurodegenerative conditions, the clinically and phenotypically heterogeneous class of polyQ disorders is defined by a common type of causative mutation, the expansion of an exonic CAG repeat motif in the affected gene. The CAG base triplet codes for the amino acid glutamine and is translated into an elongated polyQ stretch in the disease protein. So far, the family of polyQ disorders comprises nine rare conditions, namely the spinocerebellar ataxias 1 (SCA1), 2 (SCA2), 3 (SCA3; also known and hereinafter referred to as Machado-Joseph disease, MJD), 6 (SCA6), 7 (SCA7), and 17 (SCA17) as well as Huntington disease (HD), dentatorubral-pallidoluysian atrophy (DRPLA) and spinal and bulbar muscular atrophy (SBMA). The inheritance of these diseases is autosomal dominant, except for SBMA which follows an X-linked recessive pattern (Paulson et al., 2017; Stoyas and la Spada, 2018). The clinical manifestation of the mutation depends on disease-specific thresholds of the highly polymorphic CAG repeat/polyQ lengths, with most diseases featuring a defined intermediate expansion range linked to reduced penetrance. Above that range, the disease shows full manifestation, with both age at onset (AAO) and symptomatic severity negatively correlating with expansion length. Due to an instability of the CAG repeat during meiosis, polyQ disorders are furthermore characterized by the genetic phenomenon of anticipation, leading to longer expansions and increased severity in the next generation (McMurray, 2010). Interestingly, the observed discrepancy in AAO cannot be explained by the CAG repeat length alone, suggesting additional genetic modifiers and environmental influences as major contributing factors (Chen et al., 2018). As their name implies, SCAs are primarily characterized by the occurrence of ataxic symptoms, triggered by the degeneration of the cerebellum and brainstem, whereas HD shows symptoms such as hyperkinetic movements (chorea) and neuropsychological manifestations, caused by damage in the striatum and cerebral cortex. DRPLA features ataxia, epilepsy, and intellectual deterioration based on atrophy of the cerebellum and brainstem, and SBMA is defined by a motoneuron loss-dependent muscle weakness and wasting. All polyQ disorders are highly impairing, life-shortening, and - at the present moment - incurable (Paulson et al., 2017; Stoyas and la Spada, 2018). These pathophysiological and neuropathological differences between polyQ disorders can be largely attributed to the diversity in affected genes and proteins they encode. The giant huntingtin protein (HTT) involved in HD was shown to serve as a scaffold protein for a large number of interacting partners, employing it in multiple pathways and mechanisms such as autophagy, cell division, endocytosis, vesicle trafficking, and transcriptional regulation. The SCA1 protein ataxin-1 is involved in transcriptional repression, a function it shares with the DRPLA protein atrophin-1. Ataxin-2 (Atx2), the disease protein of SCA2, which has been also linked to ALS and PD, regulates RNA metabolism and mRNA translation, whereas MJD protein ataxin-3 (Atx3) is a deubiquitinase, which trims ubiquitin chains on various target proteins, thereby influencing their stability and function. The calcium voltage-gated channel subunit α1A (CACNA1A) acts as a calcium channel, located in the cell membrane. Besides its causative role in SCA6, non-polyQ mutations in its gene were associated with episodic ataxia type 2 and familial hemiplegic migraine. The SCA7 protein ataxin-7 (Atx7) is a scaffolding component of the multifunctional Spt-Ada-Gcn5 acetyltransferase (SAGA) complex. TATA box-binding protein (TBP) of SCA17 is a general transcription factor, a function shared with the sex hormone-dependent androgen receptor (AR) in SBMA (Orr, 2012; Lieberman et al., 2019; Johnson et al., 2022). PolyQ expansions within these proteins were demonstrated to interfere with their physiological role in a gain-of-function or loss-of-function modality, further explaining the phenotypic distinctness of polyQ disorders. Despite its relative heterogeneity in symptoms and affected brain areas, and its diversity in the nature of their disease proteins, polyQ disorders also share common features in their pathomechanisms and disease-modifying pathways. These unifying characteristics include the formation of intracellular inclusion bodies - a histological feature of all polyQ diseases - as well as dysregulation in protein clearance mechanisms, nuclear import, gene expression, mitochondrial function, or solute homeostasis, which in consequence trigger neuronal disturbances and eventually cellular demise (Weber et al., 2014; Lieberman et al., 2019; Bunting et al., 2022). Many of these perturbations are modulated by various disease protein-targeting PTMs including phosphorylation, SUMOylation, ubiquitination, as well as proteolytic fragmentation (Matos et al., 2017; Johnson et al., 2022). The latter type of modification represents an important source of deleterious and aggregation-prone fragments, subsumed under the term toxic fragments, and calpains - together with caspases - appear to be the key players in the proteolysis of many polyQ proteins. Intriguingly, calpains were reported to be pathologically overactivated in polyQ diseases, a circumstance that might be directly linked to the known disbalance of the calcium signaling homeostasis in neurodegenerative disorders (Bezprozvanny, 2009; Weber et al., 2014; Matos et al., 2017).
Proteolytic cleavage as a posttranslational modification of HTT or, more generally, the occurrence of truncated forms of the mutant protein is considered a crucial mediator of polyQ toxicity in the molecular pathogenesis of HD (Wellington and Hayden, 1997; Ehrnhoefer et al., 2011). Different studies demonstrated that N-terminal fragments of HTT containing the polyglutamine expansion were present in the brains of HD patients as well as mouse and cell models (DiFiglia et al., 1997; Tanaka et al., 2006; Schilling et al., 2007; Landles et al., 2010). Importantly, overexpression of such N-terminal HTT fragments in rodent models of HD was sufficient to manifest a progressive disease phenotype (Mangiarini et al., 1996; Davies et al., 1997). Moreover, truncated forms of HTT were reported to translocate into the nucleus, accumulate and form intranuclear aggregates, which eventually induce apoptotic stress and cell death (Hackam et al., 1998; Martindale et al., 1998; Zhou et al., 2003; Landles et al., 2010). Intriguingly, also HTT fragments lacking the polyQ stretch were found to cause cellular dysregulations, e.g., endoplasmic reticulum stress and autophagic perturbations (Martin et al., 2014; El-Daher et al., 2015). Early reports indicated that cleavage by caspases, especially caspase-6, was a relevant molecular modifier of HD pathogenesis, and inhibiting HTT caspase-dependent cleavage ameliorated multiple disease hallmarks in cell and animal models (Wellington et al., 1998, 2000, 2002; Graham et al., 2006). An alternative non-proteolytic source for truncated HTT was found in a mis-splicing event, which was shown to release a short, toxic, and aggregation-prone exon 1 fragment of the polyQ-expanded protein (Sathasivam et al., 2013). Just after demonstrating that caspases are involved in HD pathogenesis, researchers focused on evaluating the potential contribution of calpains to the disease pathways. In primary studies, wild-type, as well as mutant HTT, were shown to be substrates of calpain-dependent proteolysis (Gafni and Ellerby, 2002; Goffredo et al., 2002; Gafni et al., 2004). Moreover, caspase cleavage-derived N-terminal fragments of HTT appeared to undergo further calpain-mediated proteolysis (Kim et al., 2001). Importantly, analysis of post-mortem brains of HD patients and mouse models detected elevated levels of calpain-1, -5, -7, and -10. Along with a strong calpain activation, an increased and altered fragmentation pattern of polyQ-expanded HTT was observed in the brains of HD patients (Gafni and Ellerby, 2002; Gafni et al., 2004). Moreover, in an RNA-silencing-based approach, calpain-10 - together with other proteases - was found specifically accountable for the formation of two short N-terminal HTT fragments (Ratovitski et al., 2011). Calpain overactivation at baseline was detected in various HD cell and rodent models, when investigating known calpain substrate proteins such as α-spectrin and p35 (Gafni et al., 2004; Cowan et al., 2008; Paoletti et al., 2008; Clemens et al., 2015; Weber et al., 2016, 2018). This overactivation was also associated with a compromised N-methyl-D-aspartate (NMDA) receptor signaling and excitotoxic effects in HD mice (Cowan et al., 2008; Gladding et al., 2012, 2014). Mutating identified calpain cleavage sites at amino acid positions T467 and S534 (positions based on UniProt reference isoform; identifier: P42858; Figure 2A) protected polyQ-expanded HTT from calpain-mediated fragmentation, consequently lowering levels of N-terminal fragments, and reducing HTT aggregation and cytotoxicity (Gafni et al., 2004). Interestingly, mimicking the phosphorylation of S534 by substitution to aspartic acid could strongly lower the cleavage of HTT and reduce the polyQ-mediated toxicity, highlighting the vital crosstalk between different PTMs (Schilling et al., 2006). Based on the observed calpain overactivation in HD, many studies focused on the potentially protective effects of targeting calpain activity as a therapeutic approach. Overexpression of CAST lowered calpain activation, calpain-dependent HTT fragmentation and aggregation, whereas CAST depletion was associated with calpain overactivation as well as enhanced calpain-mediated HTT cleavage and aggregation in cellulo (Weber et al., 2018). Furthermore, in HD knock-in mice, genetic CAST ablation worsened molecular and neuropathological features by further triggering calpain activity, inducing polyQ-expanded HTT cleavage, and increasing fragmentation of additional neuronal substrate proteins (Weber et al., 2018). In line with these findings, calpain knockdown and CAST overexpression in vivo showed beneficial effects on mutant HTT aggregation, polyQ toxicity, and HD-related behavior in HTT fragment models, where fragmentation of the disease protein played a subordinate role. These beneficial effects were explained by a calpain inhibition-dependent stimulation of autophagic pathways (Menzies et al., 2015). Interestingly, olesoxime, a neuroprotective cholesterol-like drug candidate which binds the voltage-dependent anion channels (VDACs) and the translocator protein (TSPO) on the outer mitochondrial membrane, was reported to ameliorate disease-related abnormalities in HD rodent models by lowering calpain activity (Clemens et al., 2015; Weber et al., 2016). In these studies, treatment with olesoxime attenuated calpain overactivation, which was accompanied by reduced fragmentation, aggregation, and nuclear accumulation of polyQ-expanded HTT. Importantly, olesoxime administration led to an improvement of the cognitive and psychiatric phenotype as well as reduced brain atrophy in the HD rat model (Clemens et al., 2015). Although the precise mode of action of the molecule remains unknown, it was hypothesized that olesoxime might act beneficially on the mitochondria-endoplasmic calcium coupling in neurons (Weber et al., 2019b). Recently, another small-molecule compound termed CHIR99021, a known glycogen synthase kinase 3 (GSK3) inhibitor, was demonstrated to inhibit calpain overactivation by suppressing proteasomal degradation of CAST in cell and animal models of HD. Notably, CHIR99021 treatment reduced various mitochondrial, neuropathological, and disease-associated hallmarks in HD knock-in mice (Hu et al., 2021). However, this study omitted to investigate the consequences of CHIR99021 administration on HTT fragmentation. On the other side, calpain-mediated cleavage was also reported to have positive effects on the turnover rate of mutant HTT. In an HD cell model, N-terminal HTT fragments of a calpain-independent origin were shown to be further degraded by calpains, and calpain inhibition led to an accumulation of these breakdown products, accompanied by elevated mutant HTT aggregate formation (Ratovitski et al., 2007). Furthermore, Happ1, an intrabody that binds the proline-rich region of the N-terminus of HTT was shown to enhance the degradation of an exon 1 construct of poly Q-expanded HTT in a calpain-dependent fashion and involving cleavage at K15 (Figure 2A). Reciprocally, another intrabody, which binds to this calpain cleavage site, prevented clearance of the mutant HTT constructs (Southwell et al., 2011). Aside from its directly linked effects on HTT cleavage and aggregation, treatment with direct or indirect inhibitors of calpain activity or overexpression of CAST proved effective for ameliorating further pathological hallmarks in a multitude of HD cell and animal models. For instance, the administration of calpain inhibitor I/ALLN to medium-sized spiny neurons (MSNs) isolated from HD mice balanced the loss rate of surface NMDA receptors and reduced NMDA-induced apoptosis (Cowan et al., 2008). CAST overexpression as well as CX295 treatments rescued NF-κB-p65 levels in HD cells, thereby lowering oxidative stress and cell degeneration (Reijonen et al., 2010). Likewise, administration of PRE084, a sigma-1 receptor agonist, elevated CAST levels and exerted neuroprotective effects via NF-κB-p65 signaling in the same cell model (Hyrskyluoto et al., 2013).
Concurrent with the research history of HD, studies investigated the role of proteolytic cleavage and the validity of the toxic fragment hypothesis in the pathological context of MJD. Similar to findings in HD models, early findings in MJD saw correlations between the expression of truncated, polyQ stretch-containing forms of the disease protein Atx3 and increased toxicity, which was accompanied by nuclear mislocalization and aggregation in vitro and in vivo (Ikeda et al., 1996; Paulson et al., 1997; Goti et al., 2004; Haacke et al., 2006). Fragments of polyQ-expanded Atx3 were detected in the brains of MJD mice and patients and associated with the disease progression (Goti et al., 2004). A subsequent study could narrow down the cleavage site presumably responsible for the observed fragments, but without identifying a responsible protease (Colomer Gould et al., 2007). Caspases were the first suspects regarding Atx3 cleavage, and some works demonstrated that caspase-dependent fragmentation was indeed occurring in vitro and in cell models, however, without detecting reported breakdown products in MJD patient brains (Wellington et al., 1998; Berke et al., 2004). Further supportive data for the involvement of caspases was obtained from a Drosophila model of MJD, where caspase cleavage site-resistant Atx3 protected against the polyQ-induced eye degeneration but without effects on disease protein aggregation (Jung et al., 2009). Primary reports on the involvement of calpains in the fragmentation of polyQ-expanded Atx3 were based on the analysis of proteolytic events observed in neuroblastoma cells. Stimulation of cell lysates with calcium or treatment of cells with a calcium ionophore induced fragmentation of Atx3, which was abolished by the administration of calpain inhibitors, whereas blocking caspases and other proteases did not prevent Atx3 cleavage. CAST overexpression, on the other hand, lowered polyQ-expanded Atx3 cleavage, and aggregation (Haacke et al., 2007). In addition, this study delivered the first information on calpain-specific cleavage sites in the Atx3, which were precisely mapped to two main amino positions at D208 and S256 in a later study (positions based on UniProt reference isoform; identifier: P54252-2; Figure 2B; Haacke et al., 2007; Weber et al., 2017). Fragments derived from calpain cleavage at the identified sites showed strongly increased aggregation propensities and cytotoxicity in cell models of MJD and were found to occur in patient-derived fibroblasts, induced pluripotent stem cells, induced cortical neurons (iCNs), and - most importantly - post-mortem MJD patient brain (Weber et al., 2017). Interestingly, N-terminal fragments of Atx3 lacking the polyQ stretch were shown to induce an MJD-like phenotype in mice and led to mitochondrial perturbations in cell models, pointing toward their participation in the molecular pathogenesis (Hübener et al., 2011; Harmuth et al., 2018). Based on these findings, several studies targeted the mapped calpain cleavage sites in Atx3 to render the polyQ-expanded protein calpain cleavage-resistant and evaluate the consequences of these modifications on disease hallmarks. For instance, mutating three amino acids around the cleavage sites D208 and S256 to tryptophan residues efficiently abolished cleavage by calpains (Weber et al., 2017). In an antisense oligonucleotide-based exon skipping approach in MJD patient-derived fibroblasts, the removal of two exons of Atx3 ablated the main recognition sites for calpains and caspases of Atx3 and blocked the formation of potentially toxic polyQ-containing fragments. However, due to the exon removal-induced loss of two functionally important ubiquitin interacting motifs in Atx3, as well as low exon skipping efficiencies, this strategy was deemed non-viable as a therapeutic approach (Toonen et al., 2016). In a lentiviral MJD mouse model, deletion of larger amino acid stretches on adjacent calpain cleavage sites within Atx3 reduced disease protein cleavage and aggregation, and retained its cytoplasmic localization (Simões et al., 2022). Consistent with findings in HD, calpains were not only responsible for the fragmentation of the disease protein but exhibited also an overactivation in respective disease models. A direct link between neuronal specificity of MJD and calpain-mediated cleavage was gained from analyzing the effects of neurotransmitter-induced excitation of MJD patient-derived iCNs. Here, treatment with L-glutamate led to an excitation-dependent calcium influx and, thereby, calpain-dependent cleavage and aggregation of polyQ-expanded Atx3 in iCNs (Koch et al., 2011). Further analysis of MJD patient-derived fibroblasts and MJD animal models delivered additional proof of a pathological calpain overactivation, which caused subsequent proteolytic perturbations in neuronal substrate proteins and might be linked to a described dysregulation of calcium homeostasis by polyQ-expanded Atx3 (Chen et al., 2008; Simões et al., 2012; Weber et al., 2020). Moreover, triggering calpain activation in MJD mice by genetically depleting CAST led to a worsening of the disease-associated molecular and behavioral characteristics, whereas CAST overexpression in a further study ameliorated pathological hallmarks including polyQ-expanded Atx3 cleavage, mislocalization, and aggregation, and neuronal loss (Simões et al., 2012; Hübener et al., 2013). Interestingly, MJD mice harboring a knockout of calpain-1 showed a partially improved phenotype regarding reduced Atx3 cleavage, lowered fragmentation of synaptic proteins, as well as increased body weight and survival, but featured worsened motor symptoms (Weber et al., 2020). These mixed consequences might be explained by the vital role of calpains in neuroprotection and neuronal plasticity and suggest calpain-2 as a more suitable target for therapeutic intervention (Baudry and Bi, 2016). Aside from genetic strategies targeting cleavage sites in Atx3 or the cleavage-executing calpain system, different studies focused on more clinically translatable approaches using calpain inhibitors. Treatment of MJD patient-derived iCNs with ALLN or calpeptin after excitotoxic L-glutamate stimulation reduced polyQ-expanded Atx3, while caspase-specific inhibitors failed to do so (Koch et al., 2011). Administration of BDA-410, an inhibitor with a relatively higher selectivity for calpain-1 over calpain-2, lowered polyQ-expanded Atx3 cleavage and aggregation, and alleviated neuronal loss and motor symptoms in MJD mice (Li et al., 2007; Simões et al., 2014). Interestingly, two treatment studies using calpain inhibitor calpeptin and BLD-2736, a novel inhibitor of calpain-1, -2, and -9, in MJD zebrafish, did not primarily link the observed beneficial effects on Atx3 aggregation and motor phenotype with a reduced fragmentation of Atx3, but with its higher turnover via the autophagic system (Watchon et al., 2017; Robinson et al., 2021), which is known to be modulated by calpain activity (Weber et al., 2019a).
The presence of fragments or protease-dependent cleavage of disease proteins was shown to modulate the pathogenesis of SCA1, SCA2, SCA6, SCA7, DRPLA, and SBMA (Weber et al., 2014; Matos et al., 2017). Some studies demonstrated connections to disease protein fragmentation by caspases, e.g., the role of caspase-7-mediated cleavage of Atx7 in SCA7 (Wellington et al., 1998; Young et al., 2007). However, unlike for HD and MJD, the importance of calpains and calpain-mediated disease protein cleavage has been investigated to a lesser extent in the remaining polyQ diseases, leaving many questions - in this context - unanswered. In the case of SCA17, earlier attempts to associate caspases with the cleavage of the disease protein TBP were inconclusive, despite unequivocal reports on the involvement of truncated forms of polyQ-expanded TBP in the molecular pathology of SCA17 (Wellington et al., 1998; Friedman et al., 2008). In a recent study, new light was shed on TBP fragmentation, detecting calpains as players in the pathogenesis of SCA17 (Weber et al., 2022). In SCA17 cell and rat models, TBP was processed by the overactivated calpain system into prominent C-terminal fragments. Interestingly, in contrast to TBP’s nuclear presence, these arising C-terminal breakdown products were mislocalized to the cytoplasm, suggesting potential negative repercussions on the protein function as a general transcription factor. Importantly, inhibition of calpains by overexpression of CAST or calpain inhibitor administration in SCA17 cells reduced TBP fragmentation, decreased its aggregation, and rescued cell viability impairments, indicating the toxic potential of TBP fragments (Weber et al., 2022). Despite these first significant findings, various aspects of the involvement of calpains in SCA17 remained elusive. Although TBP was suggested to be proteolyzed by calpains C-terminally of the polyQ stretch around amino acid position A110, the exact cleavage site is unclarified (position based on UniProt reference isoform; identifier: P20226-1; Figure 2C; Weber et al., 2022). Moreover, the exact pathological role of calpain cleavage-derived TBP fragments and potential therapeutic interventions on fragmentation and calpain activation by genetic or pharmacologic approaches in vivo still have to be examined. The SCA6 disease protein CACNA1A was found to be cleaved by an unknown protease into a C-terminal fragment, which exhibited increased toxicity and resistance to further proteolysis in cell models (Kubodera et al., 2003). Moreover, this fragment was detected in the post-mortem brains of SCA6 patients and associated with the cytoplasmic aggregation of the disease protein (Ishiguro et al., 2010). Another study, however, showed that shorter, presumably calpain cleavage-independent C-terminal CACNA1A fragments translocate into the nucleus and, thereby, mediate polyglutamine-dependent cytotoxicity (Kordasiewicz et al., 2006). On the other hand, one must consider that any pathological alterations of CACNA1A’s function may affect the cellular calcium homeostasis and consequently induce calpain activation. Fragments of atrophin-1 (ATN1), the disease protein of DRPLA, were found in cell models, as well as mouse and patient brain, and associated with its polyQ-dependent toxicity. Caspases appeared as primary executors of the underlying proteolytic origin (Miyashita et al., 1997; Ellerby et al., 1999; Schilling et al., 1999). However, two other studies showed that pathologically relevant fragments of ATN1 can also arise in a caspase-independent process (Nucifora et al., 2003; Suzuki et al., 2010). Thus, and despite some opposing indications, calpains cannot be ruled out as potentially responsible proteases in the fragmentation of ATN1 in DRPLA. While there is evidence for implications of truncated forms of the androgen receptor (AR) in the molecular pathogenesis of SBMA, and caspases have been reported to cleave the disease protein, information on participation of calpains in the disease is scarce (Butler et al., 1998; Kobayashi et al., 1998; Merry et al., 1998; Wellington et al., 1998). However, in an SBMA-independent context, calpain-mediated cleavage of AR was shown to occur in prostate cancer cells rendering the receptor androgen-independent or leading to its elimination (Libertini et al., 2007; Yang et al., 2008). Thus, an involvement of calpains in the cleavage of polyQ-expanded AR is very likely and demands further scrutiny. Whether disease protein fragmentation must have negative consequences is still debated, with studies challenging - or at least softening - this concept. Contradictory results were found for the SCA2 disease protein Atx2. Here, truncated forms of Atx2 containing the polyglutamine stretch showed high aggregation propensity, and their polyQ flanking regions were crucial for the aggregation process (Nozaki et al., 2001). However, in another study, N-terminally truncated Atx2 fragments did not form aggregates and were less cytotoxic than the full-length protein (Ng et al., 2007). Still, responsible proteases for this process remain to be discovered.
Over the past three decades, a great number of studies corroborated proteolytic fragmentation of disease proteins as highly relevant PTMs, with significant consequences on the molecular mechanisms of neurodegeneration. Based on the findings in HD, MJD, and SCA17 described in this review, the contribution of calpains as the executor of disease protein cleavage within the general mechanisms of polyQ disorders is highly conceivable. A large variety of molecular repercussions may be triggered by the overactivation of calpains and calpain-mediated fragmentation of disease proteins, with detrimental consequences for affected cells as summarized in Figure 3. However, as knowledge of the contribution of calpains in other polyQ diseases - namely SCA1, SCA2, SCA6, SCA7, DRPLA, and SMBA - is fragmentary, additional research efforts are demanded to fully confirm calpain cleavage as a unifying mechanism. A first step and fast way to gauge whether yet experimentally unverified polyQ proteins might represent calpain substrates is utilizing well-established computational approaches. Tools such as Calpacchopper or GPS-CCD can offer a first insight into the potential proteolytic likelihood and give further information on expected fragment sizes and their composition (DuVerle et al., 2011; Liu et al., 2011; duVerle and Mamitsuka, 2019). The potential involvement of calpains and their pathological activation in underreported polyQ disorders appear, furthermore, very likely, as the intracellular calcium homeostasis - a pivotal factor of their functionality - was shown to be dysregulated in many SCAs (Bezprozvanny, 2009; Kasumu et al., 2012). A multitude of pharmacologic treatment approaches for lowering detrimental calpain activation have been evaluated in preclinical studies in vitro and in vivo, with pathology-ameliorating outcomes for polyQ disorders. Novel inhibitors such as BLD-2736, which work at sub-micromolar ranges, showed promising effects in MJD zebrafish (Robinson et al., 2021), and should be further reassessed in additional in vivo models, and also for other diseases such as HD and SCA17. A summary of so far tested genetic and direct or indirect pharmacologic strategies targeting the calpain system can be found in Table 2. The overall findings are encouraging. However, the lack of specificity and efficacy of available calpain inhibitors poses a therapeutic obstacle, despite compounds like ABT-957 and BDL-2660 reaching clinical trials (ClinicalTrials.gov, NCT02220738 and NCT04334460) for treatment of AD and COVID-19. In this context, it should not be forgotten that the activity of calpains serves a physiological purpose in a functional biological system. Thus, a misdirected therapeutic interference with this system may lead to unwanted negative repercussions. This circumstance is highlighted by mutations in the human CAPN1 gene, which reportedly cause spastic paraplegia 76 (SPG76) or a form of cerebellar ataxia (Gan-Or et al., 2016; Wang et al., 2016). Therefore, the development and availability of highly specific inhibitors which only target the desired calpain molecule are essential, as suggested for calpain-2 in the treatment of acute neuronal injury and beyond (Wang et al., 2018). Alternatively, indirect strategies targeting dysfunctional calcium homeostasis, impaired mitochondria, or depletion of CAST levels are auspicious, as demonstrated for compounds like olesoxime and CHIR99021 in HD (Clemens et al., 2015; Weber et al., 2016; Hu et al., 2021). These drugs do not only antagonize overactivated calpains and ameliorate subsequent detrimental effects but also mitigate their underlying disturbances caused by polyQ toxicity. A future translation of these findings to other polyQ disorders and potential clinical trials is necessary to assess their general applicability. Modifying calpains and calpain-mediated cleavage with genetic means has been tested in different experimental set-ups in vitro and in vivo for HD and SCA3, such as by knocking out specific calpain isoforms (Menzies et al., 2015; Weber et al., 2020), overexpressing the endogenous inhibitor CAST (Simões et al., 2012; Menzies et al., 2015) or removing calpain cleavage sites (Gafni et al., 2004; Toonen et al., 2016; Weber et al., 2017; Simões et al., 2022). Although these strategies led to a therapeutic proof of principle and - in most cases - positive repercussions on certain pathological parameters, the translation into clinical approaches appears intricate, as targeting the mutant disease gene still represents the most direct way. The biological function of the wild-type polyQ protein cleavage by calpains, which act in a modulatory way, remains largely unclarified. Both HTT and Atx3 undergo calpain-mediated proteolysis at baseline regardless of their polyQ-expansion, suggesting a natural function behind this process (Weber et al., 2017, 2018). Investigating the consequences of calpain-mediated cleavage on the protein’s and its fragments’ function, intracellular localization, interactome, and stability might generate a better understanding of both physiological and pathological mechanisms in the respective disease context. Only by broadening our perspective and deepening our insights into the role of posttranslational proteolytic events mediated by calpains - not only in polyQ disorders but also in other neurodegenerative diseases - will allow us to understand and harness the full therapeutic potential of calpains.
RI, HN, and JW conceptualized the manuscript. RI and JW wrote the initial version of the manuscript. RI created the illustrations. All authors contributed to the article and approved the submitted version.
HN and JW received funding from the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG; research grant numbers NG 101/6-1 | WE 6585/1-1). We acknowledge support by the Open Access Publication Funds of the Ruhr-Universität Bochum.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648488 | 36170815 | Priya Rangan,Fleur Lobo,Edoardo Parrella,Nicolas Rochette,Marco Morselli,Terri-Leigh Stephen,Anna Laura Cremonini,Luca Tagliafico,Angelica Persia,Irene Caffa,Fiammetta Monacelli,Patrizio Odetti,Tommaso Bonfiglio,Alessio Nencioni,Martina Pigliautile,Virginia Boccardi,Patrizia Mecocci,Christian J. Pike,Pinchas Cohen,Mary Jo LaDu,Matteo Pellegrini,Kyle Xia,Katelynn Tran,Brandon Ann,Dolly Chowdhury,Valter D. Longo | Fasting-mimicking diet cycles reduce neuroinflammation to attenuate cognitive decline in Alzheimer’s models | 27-09-2022 | SUMMARY The effects of fasting-mimicking diet (FMD) cycles in reducing many aging and disease risk factors indicate it could affect Alzheimer’s disease (AD). Here, we show that FMD cycles reduce cognitive decline and AD pathology in E4FAD and 3xTg AD mouse models, with effects superior to those caused by protein restriction cycles. In 3xTg mice, long-term FMD cycles reduce hippocampal Aβ load and hyperphosphorylated tau, enhance genesis of neural stem cells, decrease microglia number, and reduce expression of neuroinflammatory genes, including superoxide-generating NADPH oxidase (Nox2). 3xTg mice lacking Nox2 or mice treated with the NADPH oxidase inhibitor apocynin also display improved cognition and reduced microglia activation compared with controls. Clinical data indicate that FMD cycles are feasible and generally safe in a small group of AD patients. These results indicate that FMD cycles delay cognitive decline in AD models in part by reducing neuroinflammation and/or superoxide production in the brain. | Fasting-mimicking diet cycles reduce neuroinflammation to attenuate cognitive decline in Alzheimer’s models
The effects of fasting-mimicking diet (FMD) cycles in reducing many aging and disease risk factors indicate it could affect Alzheimer’s disease (AD). Here, we show that FMD cycles reduce cognitive decline and AD pathology in E4FAD and 3xTg AD mouse models, with effects superior to those caused by protein restriction cycles. In 3xTg mice, long-term FMD cycles reduce hippocampal Aβ load and hyperphosphorylated tau, enhance genesis of neural stem cells, decrease microglia number, and reduce expression of neuroinflammatory genes, including superoxide-generating NADPH oxidase (Nox2). 3xTg mice lacking Nox2 or mice treated with the NADPH oxidase inhibitor apocynin also display improved cognition and reduced microglia activation compared with controls. Clinical data indicate that FMD cycles are feasible and generally safe in a small group of AD patients. These results indicate that FMD cycles delay cognitive decline in AD models in part by reducing neuroinflammation and/or superoxide production in the brain.
Alzheimer’s disease (AD) is a neurodegenerative disease characterized by the accumulation of amyloid-beta (Aβ) via Aβ oligomers (oAβ) that can be toxic in their fibrillar form (Gong et al., 2003) or aggregate to form amyloid plaques and promote the generation of hyperphosphorylated tau protein (Bloom, 2014). This distinct neuropathology can lead to inflammation and oxidative damage, synaptic degeneration, and neuronal death, ultimately affecting the learning and memory functions of the cerebral cortex and hippocampus (Cline et al., 2018). The 3xTg-AD mouse model (3xTg) exhibits both Aβ and tau pathology, characteristic of the human disease (Oddo et al., 2003; Sterniczuk et al., 2010). In contrast, the EFAD-Tg mouse model (Youmans et al., 2012) can have different human APOE alleles (APOE2, APOE3, APOE4) knocked into the 5xFAD-Tg mice (Oakley et al., 2006), allowing investigation of the role of different APOEs on AD pathology (Lewandowski et al., 2020). The efficacy of drugs thus far approved for AD treatment is limited (Connelly et al., 2019; Schneider et al., 2011). The more recent effort to remove Aβ by an antibody-based intervention appears to be promising and has led to cognitive improvements in AD patients, but the high incidence of amyloid-related imaging Abnormalities (ARIAa), particularly in APOE4 carriers treated with the higher and effective doses, may limit its efficacy and safety (Plotkin and Cashman, 2020). Thus, there is a need for broader-acting but also safer interventions, particularly considering the advanced ages of subjects diagnosed with dementias, possibly to be combined with more targeted therapies. Increasing evidence suggests that different forms of dietary interventions may be effective in limiting AD progression in mouse models. Caloric restriction (CR) generally refers to a 20%–40% reduction in total calorie intake, without lowering micronutrient intake, whereas dietary restriction refers to the restriction of a particular macronutrient (proteins, carbohydrates, or fats) with or without a reduction in calorie intake (Mirzaei et al., 2016). CR studies previously conducted in PS1 mutant knockin mice concluded that an alternate-day fasting (intermittent fasting [IF]) regimen of 3 months reduced excitotoxic damage to hippocampal CA1 and CA3 neurons compared with mice that were fed ad libitum (Zhu et al., 1999), while caloric restriction for 14 weeks in amyloid precursor protein (APP) and PS1 transgenic mice led to a reduction in the accumulation of Aβ plaques and decreased Aβ plaque-associated astrocyte activation (Patel et al., 2005). CR regimens in other AD mouse models slowed the progression of Aβ deposition in the hippocampus and in cerebral cortex (Mouton et al., 2009; Halagappa et al., 2007). A long-term study of 3xTg mice undergoing either CR or IF for either 7 or 14 months concluded that both CR and IF dietary regimens ameliorate age-related behavioral deficits by mechanisms that may or may not be related to Aβ and tau pathologies (Halagappa et al., 2007). However, chronic dietary restrictions are associated with both safety and compliance concerns, particularly in the elderly population, which represents the great majority of AD subjects. For that purpose, our group has investigated the role of alternative and less restrictive interventions in murine AD or cognitive decline models, which may be feasible for human testing including intermittent essential amino acid/protein restriction (Parrella et al., 2013) and periodic fasting and fasting-mimicking diets (Brandhorst et al., 2015). Notably, in human clinical trials, FMD cycles either caused no loss or an increase in lean body mass and function (Wei et al., 2017; Caffa et al., 2020). We previously showed that, after 4 months of protein restriction cycles, alternated with normal feeding, male 3xTg mice exhibited improved behavior performance and reduced phosphorylated tau compared with ad libitum-fed animals, and these improvements were accompanied by reduced IGF-1 signaling during the restricted period (Parrella et al., 2013). Inflammation and oxidative stress play a role in AD pathology, by damaging neurons and contributing to the accumulation of Aβ (Block, 2008). In turn, this leads to the activation of microglia cells, which respond by changing morphology, by upregulating or by synthetizing de novo surface receptors, and by secreting pro-inflammatory cytokines and reactive species of oxygen (ROS), such as nitric oxide (NO) and superoxide (Schlachetzki and Hull, 2009). The NADPH oxidase enzymatic complex (NOX2), the major producer of superoxide in microglia cells, was proposed as an attractive therapeutic target for the development of interventions against AD (Block, 2008). In fact, we have previously shown that Aβ can stimulate both an increase in superoxide production in neurons (Longo et al., 2000) and the generation of highly neurotoxic microglial/NADPH oxidase-derived peroxynitrite (Xie et al., 2002). The fasting-mimicking diet (FMD) is a low-calorie/low-protein but high-unsaturated-fat diet that has an effect on stress resistance/longevity-related markers similar to those caused by water-only fasting, but that minimizes the burden of prolonged fasting as it provides both macro- and micronutrients. Here we investigated the effect of bi-monthly cycles of an FMD, administered at an early age in male and female 3xTg and in female E4FAD mice. We show that FMD cycles improve cognitive performance, reduce AD-associated neuropathology, increase markers of neural stem cell regeneration, and reduce microglial activation and neuroinflammation, including Nox2 (NADPH oxidase) expression levels. Nox2 deficiency or treatment with the NADPH oxidase inhibitor apocynin was sufficient to reduce microglia density/activation and delay cognitive decline, indicating that microglial-dependent toxicity is important for age-dependent cognitive impairment.
We first evaluated the effects of FMD cycles specifically on female E4FAD mice expressing the human ApoE4 isoform, since, in these mice, pathology rapidly develops in the cortex and hippocampus, reaching significant levels by 6 months of age (Youmans et al., 2012; Cacciottolo et al., 2016). Using a variety of AD-relevant readouts (Maldonado Weng, 2019), the hierarchy of pathology in the EFAD mice is male E3FAD < female E3FAD ~ male E4FAD < female E4FAD. Female E4FAD mice were enrolled in the study at approximately 2.5 months of age, during which baseline behavior tests (spontaneous alternation behavior [SAB] and elevated plus maze [EPM]) were conducted. Mice were then assigned to either a standard rodent chow diet group (control) or to a 5-day FMD twice a month starting at 3 months of age, with the standard chow diet fed to the mice between FMD cycles. The dietary regimens were halted at ~7–7.5 months of age once the end point behavioral tests were completed (Figure 1A). At baseline, female E4FAD mice obtained an SAB score of 56.4% ± 10.5% (mean ± SD; Figure 1B). After ~3 months of FMD cycles (at 6 months of age), the FMD group displayed significantly higher SAB scores compared with mice fed a control diet (p < 0.05; Figure 1B). We also analyzed changes in anxiety using the EPM, in which more time spent in the open arms of the maze reflects lower levels of anxiety and increased exploratory behavior (Young, 1996). E4FAD mice that underwent FMD treatment spent more time in the open arms compared with age-matched control (animals from both groups spent comparable times in the open arms at baseline; Figure S1A). We also assessed spatial memory using the Barnes maze in 6.5- to 7-month-old E4FAD females. After the 7-day training period, during the retention test at day 14, FMD-treated mice performed better than those from the control group both in terms of latency (p < 0.0001; Figure 1C) and success (p < 0.001; Figure 1D), with mice from the FMD group taking less time to find the escape box and achieving a higher success in finding the escape box in the time allotted. There was no significant difference between control versus the FMD group in errors made or in deviation from the escape box (Data S1B and S1C). A larger percentage of mice from the FMD group also progressively increased utilization of a spatial strategy versus the random strategy taken by control diet mice (Figures 1E and 1F). Altogether, the behavioral tests suggest that bi-monthly FMD cycles improve visual attention, working memory, and spatial memory; ameliorate anxiety-associated behaviors; and increase exploratory activity in E4FAD female mice.
In the E4FAD model, we assessed hippocampal and cortical Aβ immunoreactivity in 7- to 7.5-month-old female mice after ~4 months of bi-monthly FMD cycles (Figure 2A and S5A). Compared with E4FAD females that were fed an ad libitum diet, those receiving the FMD showed reduced Aβ load in the subiculum (p < 0.01; Figure 2B) and cortex (p < 0.001; Figure 2D), but not in the CA1 region of the hippocampus (Figure 2C). We also measured the levels of Aβ38, Aβ40, and of Aβ42 in Tris-buffered saline (TBS)-soluble cortex extracts, but did not detect changes between the E4FAD control and FMD groups (Figures S1B–S1D). In the triton-soluble (TBS-X) cortex extracts, which were examined to assess the amount of Aβ in a membrane-associated state (Kostylev et al., 2015), we observed a significant reduction in all three peptides in the E4FAD mice treated with periodic FMD (Aβ38, p < 0.05, Figure 2E; Aβ40, p < 0.01, Figure 2F; Aβ42, p < 0.05, Figure 2G). These results indicate that FMD cycles either reduce the generation or contribute to the clearance of different Aβ peptides. Next, we measured neurogenesis markers in female E4FAD mice after 4 months of FMD cycles. Specifically, we measured cells that stained positive for Ki67+ (an endogenous marker of proliferation) in the dentate gyrus (DG) (Khuu et al., 2018; Ma et al., 2014; Kee et al., 2002), or that expressed both Ki67 and Sox2, a neuronal stem cell marker, in the hippocampus of control and of FMD-treated female E4FAD mice (Figure 2H). Sox2+ levels remained unchanged between the two groups (Figure 2I), while there was a significant increase in Ki67+Sox2+ cells in mice receiving the FMD compared with the control animals (p < 0.05; Figure 2J), consistent with adult neurogenesis in FMD-treated mice. To exclude the possibility that the Sox2+ cells may be proliferating astrocytes, immunofluorescence staining was carried out on sections from E4FAD control and FMD mice using Sox2+ and GFAP+ antibodies. There was no significant difference in the number of Sox2+ cells, GFAP+ Sox2+ cells, and the percentage of GFAP+Sox2+ cells/total Sox2+ cells in the DG between the E4FAD control (n = 3) and FMD (n = 4) groups (Figures S1F, S1G, S1I, and S1J). However, there was a non-significant trend for an increase in the number of GFAP+ cells in the DG in the E4FAD FMD mice compared with the E4FAD control mice (p = 0.0715; Figures S1F and S1H). To investigate neuroinflammation in the E4FAD model, we performed a multi-plex analysis on several cytokines in TBS-soluble and triton-soluble cortex extracts from female E4FAD mice treated or untreated for 4 months with FMD cycles (ending at ~7–7.5 months of age). E4FAD females that underwent four FMD cycles displayed a non-significant trend toward an increase in interleukin (IL)-12p70 (a cytokine) (Méndez-Samperio, 2010), and an increase in IL-2 levels compared with age-matched controls (p = 0.0583, Figure 2K; p < 0.05, Figure 2L). IL-2 was previously shown to reduce the levels of AD pathology and to improve cognitive behavior in AD transgenic mouse models (Gao et al., 2017; Alves et al., 2017), while IL-2 knockout mice have an impairment in spatial learning and memory (Petitto et al., 1999). Tumor necrosis factor alpha (TNFα) levels in human AD brains are normally elevated (Dong et al., 2015). In our triton-soluble cortex extracts, we observed a non-significant trend toward TNFα reduction in E4FAD FMD females compared with control mice (p = 0.0737; Figure 2M). Because FMD cycles caused a significant reduction in Aβ peptides in E4FAD FMD females (Figures 2E–2G), we asked whether there could be an interaction between CD11b and Aβ, since CD11b was previously shown to form a complex with Aβ in AD subjects and to facilitate its clearance (Zabel et al., 2013). Our co-immunoprecipitation (coIP analysis) of CD11b and Aβ42 binding did not reveal a significant difference in Aβ42 fold change between the E4FAD control and FMD groups, at least in the triton-soluble cortex extracts, which harbor membrane-associated Aβ (Figure S1E). Taken together, these data suggest that 4 months of bi-monthly FMD cycles in female E4FAD mice mitigate Aβ hippocampal and cortical load, reduce Aβ38/40/42 and increase IL-2 expression in cortex extracts. There is also evidence for a modest increase in neurogenesis after FMD treatment.
We also studied the effects of FMD cycles on 3xTg male and female mice starting at 3.5 months of age. In addition to a group on a standard rodent chow diet (control) and a group receiving FMD treatment for 4–5 days, we studied a group that consumed a 7-day protein-restricted diet (approximately 4% of calories obtained from proteins [4% PR]), based on our previous studies (Parrella et al., 2013; Levine et al., 2014) (Figure 3A). FMD males underwent a 4-day cycle versus a 5-day cycle, since 3xTg males were found to lose weight more rapidly and to a higher extent compared with female mice. 3xTg male and female control mice weighed an average of 40.3 g and 34.7 g respectively at 18.5 months. 3xTg male and female mice in the FMD and 4% PR groups regained most of their weight lost during the diet cycles upon refeeding (Figures S4A–S4F). Midpoint behavioral assays SAB in the Y-Maze and Novel Object Recognition [NOR]) were conducted when mice reached 10.5 months of age (after 7 months of dietary treatment), and approximately half-way through the study, with a final behavioral assay (Barnes maze) conducted toward the end of the study, prior to sacrifice, when mice were approximately 18 months of age (Data S1, 2A to 7D). Survival was monitored until 18 months of age. No significant difference in percent survival was detected between control, FMD, and 4% PR groups at 18 months (Figures 3B and 3C). 3xTg mice receiving FMD or 4% PR diets had higher SAB scores compared with the 3xTg that received the control diet or wild-type (WT) mice on the control diet (3xTg FMD versus 3xTg control, p < 0.01; 3xTg 4% PR versus 3xTg control, p < 0.05; Figure 3D). For the male mice, there was a significant increase in SAB in the 3xTg FMD group (p < 0.05; Figure 3E), but not in the 4% PR group compared with the 3xTg control. No significant difference in the total number of arm entries (a measure of activity) was seen among the groups in both sexes, except when the 3xTg mice groups were compared with a sex-matched WT (**p < 0.01, ***p < 0.001; Figures S2A and S2B), indicating that all groups were similarly active. In the NOR assay, there were no significant differences among the groups, male or female, in the first trial of the test, indicating no significant biases were present in mice activity among the groups (Figures S2C and S2D). In trial 2, during which one of the identical objects was replaced with a novel object, 3xTg mice displayed a major reduction in recognition index (RI), which did not occur in 3xTg mice treated with either FMD or 4% PR cycles, among the male 3xTg groups (3xTg FMD versus 3xTg control, p < 0.01; 3xTg 4% PR versus 3xTg control, p < 0.05; Figure 3G). There were no differences in RI among the female 3xTg groups (Figure 3F). These results suggests that both FMD and protein restriction cycles can improve cognitive performance in 3xTg mice, although FMD cycles appear to have a more consistent effect, which is in most cases not dependent on sex. We also assessed male and female 18-month-old 3xTg mice after ~14.5 months of FMD and 4% PR dietary regimens as well as age-matched controls (Data S1- 2A and 7D). Of the various combinations tested, a non-significant trend for a reduction in latency (time spent to locate escape box) was observed in female 3xTg FMD compared with female 3xTg controls (Data S1- 2D). The female 3xTg FMD group also displayed a non-significant trend for an increased success rate in finding the escape box compared with the 3xTg female controls (Data S1–2E). Together, these behavioral tests suggest that FMD cycles improve visual attention and working and spatial memory in aged 3xTg mice. However, in old mice, the accumulation of pathology appears to minimize these effects of the FMD in 3xTg mice.
Since Aβ accumulation and hyperphosphorylated tau are well-established markers for AD (Lo et al., 2013), we assessed hippocampal Aβ load and the number of hippocampal neurons that are AT8+, a marker for abnormally phosphorylated tau (Parrella et al., 2013), in female and male 3xTg mice at the end of the study. We stained for Aβ in the subiculum and CA1 hippocampal regions of aged female 3xTg mice after ~15 months of dietary regimens in approximately 18.5-month-old mice (Figure 4A, top and middle; Figure S5B, top). We observed that female 3xTg mice treated with the periodic FMD had a reduced Aβ load in the subiculum (p < 0.01; Figure 4B) and in CA1 regions (p < 0.05; Figure 4C) compared with control 3xTg mice. A similar reduction in Aβ load was observed in the subiculum of 4% PR female 3xTg mice compared with the control 3xTg females (p < 0.01; Figure 4B). The number of AT8+ neurons in the combined subiculum and CA1 regions from female 3xTg groups (Figure 4A, bottom; Figure S5B, bottom) was reduced in FMD-treated 3xTg females compared with the control 3xTg mice (p < 0.01; Figure 4D), while there was no significant difference in AT8+ neuron number between the 4% PR 3xTg and the 3xTg control animals (Figure 4D). In the subiculum and CA1 hippocampal regions of aged male 3xTg mice after ~15 months of dietary regimens (18.5 months of age) (Figure 4E, top and middle; Figure S5C, top), there were no significant reductions in subiculum (Figure 4F) or CA1 (Figure 4G) Aβ load between the FMD and control 3xTg groups. A non-significant trend toward a reduction in Aβ load in the CA1 was observed in the 4% PR male 3xTg group with the control 3xTg males (p = 0.0525; Figure 4G). Concerning hyperphosphorylated tau counts, the number of AT8+ neurons in the combined subiculum and CA1 regions for the male 3xTg groups (Figure 4E, bottom; Figure S5C, bottom) was reduced in the FMD 3xTg mice compared with the 3xTg controls (p < 0.05; Figure 4H). A similar reduction in the number of AT8+ neurons was observed in the 4% PR animals and FMD group with the 3xTg controls (Figure 4E, bottom) (p < 0.05; Figure 4H). These results indicate that FMD cycles reduce both Aβ load and/or hyperphosphorylated tau in female and male mice, in part through the temporary reduction of protein intake. Previous work from our group indicates that FMD cycles in aged WT mice can promote neurogenesis (Brandhorst et al., 2015). A greater decline in neurogenesis was previously observed in aging 3xTg mice versus age-matched WT controls (Rodríguez et al., 2008). We measured bromodeoxyuridine (BrdU) incorporation within the subgranular zone (SGZ) and inner third of the granule cell layer of the DG (DG) in 18.5-month-old male and female 3xTg mice to determine whether FMD cycles had any effect on neurogenesis (FMD and 4% PR; Figures 4I and 4J and 4L and 4M and S2E–S2J). Using DAB immunohistochemistry, we observed that, among the female 3xTg groups, the FMD group but not the 4% PR group displayed an increase in the number of BrdU+ cells in the DG compared with that in 3xTg female controls (p < 0.01; Figure 4I) (Figure S2F). Among the 3xTg males, both the FMD and 4% PR groups showed an increase in the number of BrdU+ cells in the DG as well, compared with the male controls (p < 0.01, Figure 4L; p < 0.05, Figure S2H). Because FMD cycles increased BrdU+ DG cells, we evaluated whether BrdU+ cells in these groups were also Sox2+, as evidence for type I and type II neural stem cells (NSCs) (Ming and Song, 2005) (Figures 4J and 4M). In both the female and male 3xTg control and FMD groups, we did not observe changes in the number of Sox2+ cells in the DG (Figures S2I and S2J), although we did see a modest but significant increase in BrdU+Sox2+ levels in the FMD groups compared with the controls (p < 0.05; Figures 4J and 4M). These results indicate that FMD cycles can cause modest increases in the generation of NSCs in agreement with the results for the E4FAD mice. To assess neuroinflammation, we stained for CD11b, a marker for activated microglia and macrophages in the hippocampi of 18.5-month-old WT male and female mice and compared them with the control and FMD 18.5-month-old 3xTg female and male groups (Figure 4K, top; Figure 4N, top). Among females, there was a major increase in CD11b+ cells in the combined subiculum and CA1 regions of the 3xTg mice on a standard diet compared with microglia density in WT controls (p < 0.01; Figure 4K, bottom left). FMD cycles reduced CD11b cells to a level that was no longer significantly higher than that of the WT mice (Figure 4K, bottom left). Female 3xTg mice that were subjected either to the control diet or to the periodic FMD showed a lower proportion of resting state microglia and a higher proportion of activated and amoeboid microglia compared with the age-matched WT mice (stage 3: WT versus control, *p < 0.05; WT versus FMD, **p < 0.01; Figure 4K, bottom right). Similarly, among male mice, there was a major increase in microglia density in the combined subiculum and CA1 regions of the 3xTg mice on a standard diet compared with that in WT controls (p < 0.05; Figure 4N, bottom left). FMD cycles reversed this increase in microglia (Figure 4N, bottom left). The 3xTg mice on the control or FMD diets displayed either a lower or a trend toward a lower proportion of resting state microglia and a higher proportion of amoeboid microglia compared with the age-matched WT mice on the standard diet (stage 1, WT versus control,*p < 0.05; stage 3, WT versus control, ***p < 0.0001; WT versus FMD,*p < 0.05; stage 4, WT versus control and WT versus FMD, *p < 0.05; Figure 4N, bottom right). FMD cycles also reduced the average number of cells expressing CD68+, another microglia marker in the hippocampus of male 3xTg mice (p < 0.05; Figure 4O, bottom). Overall, these data indicate that a long-term regimen of FMD cycles can reduce AD-associated pathology in aged male and female 3xTg mice, possibly by reducing microglia density and by modulating microglia activation state. Notably, others have shown that perivascular macrophages and not microglia may be responsible for oxidative damage and pathology in AD mouse models (Park et al., 2017), raising the possibility that macrophages in addition to or instead of microglia may be responsible for these effects. Stem cell-dependent generation of neurons could also contribute to the positive effects of the FMD on cognition in 3xTg mice.
We also administered FMD cycles over a short-term period to mice ~6.5 months of age, when AD pathology is increasing rapidly and begins to influence cognitive behavior and neuroinflammation (Oddo et al., 2003). Male and female 3xTg mice from the FMD group were administered a 4-day FMD, for five cycles (4 days of diet, 10 days of refeeding), and were sacrificed after the fifth cycle, and before refeeding (Figure 5A). Cognitive behavior was assessed after four cycles, when mice were approximately 8.5 months old and before the fifth cycle. SAB results showed no significant differences among the male or female cohorts (Figures S3A and S3B). However, 3xTg mice did not display a reduction in SAB and NOR performance compared with WT controls, suggesting that the effect of the FMD may represent an improvement even compared with WT control mice, in agreement with our previous studies (Brandhorst et al., 2015). NOR results instead suggest the 3xTg FMD males had a significantly higher average RI score compared with that in 3xTg control male mice (p < 0.05; Figure 5B). No significant differences were observed among the female groups (Figure S3C). There were no significant differences among the groups, female or male, in the total number of arm entries (a measure of activity) in the Y-maze (Figures S3D and S3E) as well as in the first trial of the NOR test, indicating no significant biases were present in mice activity among the groups (Figures S3F and S3G). Among the female groups, no significant changes in Aβ load were observed in the subiculum (Figures 5D and S5D, top), but, in the CA1 region, there was a significant reduction in Aβ in the 3xTg FMD females compared with controls (p < 0.05; Figure 5E). AT8+ hyperphosphorylated tau was significantly reduced in the hippocampus of 3xTg FMD females (p < 0.0001; Figure 5F and S5D, bottom). Aβ load was significantly reduced in the subiculum of 3xTg FMD males compared with that in 3xTg control diet males (p < 0.01; Figures 5H and S5E, top), whereas no significant changes were observed in CA1 Aβ load (Figure 5I). A significant reduction in hyperphosphorylated tau was observed in the 3xTg FMD males compared with that in controls (p < 0.01; Figure 5J and S5E, bottom). We also stained hippocampal tissue for ionized calcium-binding adaptor protein-1 (Iba1), a 17-kDa actin-binding protein that is specifically and constitutively expressed in all microglia (Hovens et al., 2014), counted Iba1 density, and categorized the stages of microglial activation based on criteria established in previous studies (Kreutzberg, 1996; Crews and Vetreno, 2016) (Figures 5K and 5L, top). 3xTg female mice displayed a major increase in microglial number compared with WT controls (p < 0.0001; Figure 5K, bottom left), which was reduced in 3xTg FMD females (p < 0.001; Figure 5K, bottom left). There was a significant reduction in microglia at stage 1 of activation in the 3xTg female controls compared with WT (p < 0.05; Figure 5K, bottom right). Microglia at stage 3 and 4 of activation were significantly increased in the 3xTg control group compared with WT (stage 3, p < 0.01; stage 4, p < 0.05; Figure 5K, bottom right). 3xTg FMD females displayed a non-significant trend for the reduction in microglia at stage 4 compared with the 3xTg control diet group (p = 0.0772; Figure 5K, bottom right). Among the groups in the male cohort, there was no significant increase in microglia number among any of the male groups (Figure 5L; bottom left). Both 3xTg male groups having significantly fewer microglia at stage 1 compared with WT males (p < 0.05; Figure 5L, bottom right). In summary, resting-state microglia were reduced and different forms of activated microglia were increased in 3xTg compared with WT mice. In this set of short-term treatment experiments, the FMD appears to allow a general state of microglial activation while reducing the highly active phagocytic microglia, possibly in part because of the reduction in Aβ and AT8+ hyperphosphorylated tau in the 3xTg FMD-treated mice. In order to investigate the effect of FMD cycles on amyloid pathology, we isolated microglia from primary mixed glia cultures from the whole brains (except the cerebellum) of 8.5-month-old male and female 3xTg mice five cycles of FMD or control diet, and added oligomeric Aβ42 to the microglia to assess whether FMD can enhance the uptake of oligomeric Aβ42 by IBA-1 positive microglia compared with the control diet. We found that the oligomeric Aβ42 localizes to the cytosol of the microglial cells (Figures S7A and S7C). IBA-1 positive microglia isolated from the brain of 8.5-month-old 3xTg male mice after FMD cycles starting at 6.5 months of age internalized significantly more Aβ42 compared with the IBA-1 positive microglia from the control group (p < 0.05; Figure S7B). A possible link between FMD and reduced Aβ accumulation could be via the increase in oligomeric Aβ42 internalization by IBA-1 positive microglia. We next investigated the presence of any changes in the characteristics of microglia after short-term FMD cycles in male and female 3xTg mice. Using quantitative confocal microscopy methodology previously established (Stephen et al., 2015, 2019), representative images of confocal stack immune-reactive for Iba1 microglia in the prefrontal cortex (Figure 5M) and 3D skeletonized microglial projections (Figure 5N) were used to quantify the Iba1 immuno-reactive soma area (Figure 5O) and circularity (Figure 5P). Resting microglia cells tend to be smaller, rounder cells with elaborate ramifications, whereas activated microglia tend to be bigger and more amoeboid-like in shape with retracted processes (Davis et al., 2017). Iba1 soma area was significantly reduced in the 3xTg FMD females compared with 3xTg control diet females (p < 0.01; Figure 5O), while soma circularity was significantly increased in the dietary intervention groups of both sexes compared with sex-matched controls (p < 0.01, 3xTg FMD males versus 3xTg control males; p < 0.001, 3xTg FMD females versus 3xTg control females; Figure 5P). These results are consistent with an effect of FMD cycles in modulating the microglial activation state observed in 3xTg mice. To further test the hypothesis that reduced or modified neuroinflammation and/or microgliosis mediate part of the protective effects of FMD cycles, we performed mRNA sequencing from homogenized cortex samples from hemi-brains of female 3xTg mice ~8.5 months old after only one cycle of a 4-day FMD and before refeeding (Figure 5Q, top left) as well as in male 3xTg ~8.5-month-old mice after four cycles of a 4-day FMD and after 2 days of refeeding (Figure 5R, top left). For 104 genes, expression was altered significantly by the FMD in the male cohort (Figure 5R, right). When the expression of these genes was assessed in the female cohort, a similar pattern emerged (Figure 5Q, right). Relative fold change indicated an effect of FMD in consistently reducing the expression of genes associated with neuroinflammation, microglial activation, tau phosphorylation, and AD pathogenesis, and increasing in the expression of genes associated with anti-inflammatory functions, based on characteristics from the Ensembl genome database (Figures 5Q and 5R, bottom left; Table S1). Of the genes that were downregulated in the FMD groups, Fosb, Gdi1, Hspa8, Smdt1, Nr4a1, Nr4a3, Fosl2, Trib1, Egr1/2/3/4, Tiparp, Fbxo33, Hspa1b, and Prmt1 were all previously found to be linked to increased oxidative stress, pro-inflammatory cytokine secretion, and/or upregulation of gene expression after ischemia (Nomaru et al., 2015; Lopes et al., 2016; Wu et al. (2019b); Bonam et al., 2019; Azevedo et al., 2018; Lyons and West, 2011; Close et al., 2019; Vilkeviciute et al., 2019; Giri et al., 2005; Mengozzia et al., 2012; Marballi and Gallitano, 2018; Wu et al., 2019a; Flood et al., 2004; Clarimón et al., 2003; Liu et al., 2019) (Table S1; Data S1 pages 3–17). Of these genes, Egr1 was also suggested to have a role in stimulating microglial activation (Raj et al., 2015), along with Sertad1, Gadd45b, Ndel1, Hmgcr, Spry2, Arc, Prdx5, Siah2, and Hspa1a (Kuhn et al., 2006; Tamboli et al., 2010; Abels et al., 2019; Rosi, 2011; Sun et al., 2010; Park et al., 2016) (Table S1; Data S1, pages 20–28). On the other hand, the genes that were upregulated in the FMD group, including Cd33, Inpp5d, Stab1, Tia1, and Rsrp1, are associated with anti-inflammatory roles (Malik et al., 2015; Park et al., 2009; Chen and Liu, 2017; Stephens et al., 2019) (Table S1; Data S1, pages 33–37). Mertk, a gene associated with activated microglia that phagocytose dying, stressed, or excess neurons (Nomura et al., 2017), was also upregulated in the FMD groups (Table S1; Data S1, page 29). Thus, these results, taken together, are consistent with an effect of the FMD in decreasing the activation and oxidant production in microglia but possibly also macrophages or other brains cells, including smooth muscle cells and endothelial cells, which have been linked to inflammation and the expression of many genes identified here (Chow et al., 2007; Park et al., 2017). The diet groups also showed reductions in Nedd8, Cacybp, Fos, Phf13, and Junb, with the male cohort showing an additional reduction in Ppme1. (Table S1; Data S1, pages 39–44), which were all previously linked to the localization and/or increased phosphorylation of tau (Mori et al., 2004; Wasik et al., 2013; Anderson et al., 1994; Vázquez-Higuera et al., 2011; Park et al., 2018; Chu et al., 2013; Fang et al., 2016). The expression of Hook1, which encodes the protein that localizes to tau aggregates, was slightly increased in the dieting mice groups (Table S1; Data S1, page 45), although it has been previously demonstrated that the expression of Hook1 proteins is reduced in AD (Herrmann et al., 2015). Rheb, a direct activator of mammalian target of rapamycin (mTOR) (Lafourcade et al., 2013), was significantly downregulated in FMD-treated male 3xTg mice, with a similar trend seen with dieting 3xTg females (Table S1; Data S1, page 46) in agreement with the demonstrated effect of mTOR inhibition in neuroprotection in various in vivo models of neurodegenerative disease, and of the mTOR inhibitor rapamycin in ameliorating tau pathology (Caccamo et al., 2013). Yod1, which is specifically associated with macro-autophagy, and codes for the cofactor YOD1, which binds with p97 to promote lysosomal clearance (Papadopoulos et al., 2017), was significantly increased in the FMD 3xTg male cohort compared with male 3xTg controls, although this effect was not observed in female 3xTg mice (Table S1; Data S1, page 47). Notably, full-length tau is preferentially degraded via macro-autophagy, which involves the activation of AMP-activated protein kinase (AMPK), and in turn reduces mTOR signaling (Zare-shahabadi et al., 2015). Hspa8, Atpif1, and Fez1, genes associated with autophagy, had reduced expression in both males after four FMD cycles and 2 days of refeeding and in females after 4 days of FMD and no refeeding, indicating that chronic autophagy activation is unlikely to be responsible for the clearance of tau or Aβ. Other genes generally associated with AD pathogenesis were also affected by diet. Plcg2 was increased in FMD group males, with no effect in FMD group females (Table S1; Data S1, page 49). It has been suggested that the activation of Plcg2, rather than inhibition, could be therapeutically beneficial in treating AD (Magno et al., 2018). Otud1 encodes a deubiquitinase (DUB) enzyme, a class of enzymes that have been suggested as potential targets for treating AD (Baillie et al., 2017; Yuan et al., 2018). In FMD-treated mice, Otud1 was significantly reduced in males, with a similar trend seen in females (Table S1; Data S1, page 50). Erf, which encodes a transcription factor related to the E26 transformation-specific (ETS) family of proteins, was found to be reduced in FMD-treated mice (Table S1, Data S1, page 51). ETS-domain proteins have been linked to regulating neuronal functions, especially by activating the transcription of early-onset AD genes such as PSEN1 (Pandey et al., 2019). Taken together, these results suggest that short-term FMD cycles reduce inflammation, improve short-term memory, and ameliorate AD-associated pathology in 3xTg mice. These effects may be mediated by wide-acting effects involving the modulation of microglia and possibly other pro-inflammatory cell types, including perivascular macrophages (Park et al., 2017), resulting in reduced inflammation, Aβ, and hyperphosphorylated tau, while allowing microglia to contribute to the scavenging of damaged cells, organelles, and macromolecules.
We have previously shown that O2− and iron contribute to neurotoxicity in Aβ-treated neuronal cell lines (Longo et al., 2000), and our studies in vitro demonstrated that peroxynitrite (ONOO−), formed by the reaction between NO and O2−, is a major mediator of the neurotoxicity promoted by microglia activated by Aβ or lipopolysaccharide (LPS), suggesting that it can mediate the toxicity caused by chronically activated microglia in brains affected by AD (Xie et al., 2002). In fact, targeting Nox2 reduced neurovascular and cognitive dysfunctions in mice overexpressing the Swedish mutation of the APP (Park et al., 2008). Altogether, this suggests that, while microglia cells play a key role in neuronal protection and repair, at the same time, their production of O2− leading to the subsequent production of ONOO− may play a central role in neurotoxicity (Figure 6A). Thus, we assessed whether the expression of Nox2 was affected at different time points of FMD administration. After one cycle of a 4-day FMD to young female 3xTg mice, aged ~8.5 months (Figure 5Q, top left), we observed a significant reduction in Nox2 levels compared with an age- and sex-matched control (p < 0.05; Figure 6B). In female mice, Nox2 levels were significantly reduced in the 3xTg group that received five cycles of the FMD compared with both female WT mice and with the 3xTg control diet group (p < 0.01; Figure 6C). In the male cohort, there was a significant reduction in Nox2 levels in both 3xTg males on the control diet or that received five FMD cycles compared with WT male mice on a standard diet (p < 0.01, WT versus 3xTg FMD; p < 0.05, WT versus 3xTg control; Figure 6D). We did not observe a significant difference in Nox2 between the FMD and control groups after four cycles of FMD and 2 days of refeeding in male 3xTg mice (p = 0.1497; Figure S3H), although we observed a non-significant trend toward a reduction of Nox2 levels after refeeding in FMD-treated female E4FAD mice (p = 0.0549; Figures 6E and 6F). Based on the results of the effects of FMD in mediating Nox2 levels and the potential role of Nox2 as a mediator of AD pathology, we hypothesized that the knockout of NADPH oxidase could protect against cognitive decay and neuropathology in the 3xTg mouse model. Thus, we generated 3xTg/Nox2-KO mice by crossing Nox2-KO (Cybb−/−) mice with 3xTg mice (Figure 6G). Similar to 18.5-month-old male 3xTg mice treated with FMD cycles (Figure 4N), 13.5- to 14-month-old 3xTg/Nox2-KO male mice displayed a partial reversal of the increase in microglia density in the combined subiculum and CA1 regions compared with 3xTg mice. Activated microglia cells as detected by CD11b (Lynch, 2009) were reduced in 3xTg/Nox2-KO mice, compared with 3xTg mice (#p < 0.05; Figure 6H, bottom left), while both groups had significantly elevated levels of microglia in the combined subiculum and CA1 regions compared with the WT controls (****p < 0.0001, 3xTg and ***p < 0.001, 3xTg/Nox2-KO; Figure 6H, bottom left). When microglia activation stages were examined, based on a four-stage classification (Zhang et al., 2011; Parrella et al., 2013), 3xTg mice showed a higher proportion of microglia in the highly activated stages (3 and 4) compared with the WT group (stage 1, *p < 0.05; stage 2, **p < 0.01; stage 3, ****p < 0.0001; stage 4, ****p < 0.0001; Figure 6H, bottom right). In contrast, 3xTg/Nox2-KO mice showed a similar resting and stage 2 activation state, but a lower high activation state compared with 3xTg (stage 3, ##p < 0.01; stage 4, ##p < 0.01; Figure 6H, bottom right). With our progeny of male 3xTg/Nox2-KO mice (Figure 6G), we assessed short-term working memory as well as contextual and tone learning with the Y-maze apparatus and fear-conditioning (FC) behavioral assay, respectively. Mice of strains C57B/6/Nox2-KO, 3xTg, 3xTg/Nox2-KO, and corresponding WT mice (C57B/6 and 129/B6) were assessed in these cognitive tasks. There was a significant reduction in SAB scores with 3xTg male mice compared with 12.5-month-old, age-matched WT mice and the 3xTg/Nox2-KO mice (p < 0.01; Figure 6I), and no significant changes in the number of arm entries were apparent among the groups (Figure S3J). FC tests revealed that memory in the 3xTg/Nox2-KO group improved compared with the 3xTg group (p < 0.05; Figure 6J, left) at both 24- and 48-h post shock (p < 0.05; Figure 6J, right) with no changes in the 3xTg/Nox2-KO group compared with WT. Following Y-maze testing, the mice were tested with FC, based on previously established protocols (Liu et al., 2004). When tested with the FC test on day 1 for baseline measurements, no significant differences in freezing time among the groups was apparent (Figure S3K). On day 2, 24 h post shock, memory in the 3xTg/Nox2-KO improved compared with the 3xTg group (p < 0.05; Figure 6J, left). On day 3, in the novel environment before re-exposure to the tone, no significant changes in freezing time were seen among the WT, 3xTg, and 3xTg/Nox2-KO mice (Figure S3I). Upon re-exposure to the tone, 3xTg mice displayed decreased freezing behavior compared with the WT group (p < 0.05; Figure 6J, right). These results indicate that Nox2 inactivation was able to delay the decline in working memory and associative learning in 3xTg/Nox2-KO mice. Although Aβ accumulation was not modified by the inactivation of NADPH oxidase (Figure S3I), we observed a significant reduction in AT8+ hyperphosphorylated tau in the 3xTg/Nox2-KO mice compared with 3xTg male controls (p < 0.05; Figure 6K), indicating that reduced NADPH oxidase activity and O2−/ONOO− generation represents only part of the effects of FMD/refeeding cycles. Apocynin is extensively used as an inhibitor of NADPH oxidase activity and of the concomitant production of ROS, including peroxynitrite, both in vitro and in vivo. In vivo treatment for periods of 6 months were reported (Figure 6L; Stefanska and Pawliczak, 2008; Simonyi et al. 2012; ‘t Hart et al., 2014). The effect of apocynin treatment on 3xTg cognitive dysfunction was assessed using the Y-maze apparatus and the NOR assay. When we tested these mice at ~12 months of age on the Y maze, 3xTg mice exhibited a significant working memory deficit in comparison with WT mice, whereas SAB performance in 3xTg mice treated with apocynin was similar to that of control mice, in agreement with our results with 3xTg/Nox2-KO mice (p < 0.01, 3xTg vehicle versus WT vehicle; Figure 6M). We did not find significant differences in the number of arm entries among WT and 3xTg groups, suggesting that the drug treatment does not interfere with the activity levels of the rodents (Figure S3M). Similarly, for NOR, 3xTg mice had a significantly lower RI score compared with the WT group, whereas 3xTg apocynin-treated mice did not (p < 0.05, 3xTg vehicle versus WT vehicle; Figure 6N), which is also in agreement with the results obtained with 3xTg/Nox2-KO mice. There were no significant differences among the groups in the first trial of the test, indicating no biases were present in mice activity among the groups (Figure S3N). Although apocynin-treated WT mice were assessed in all of the above parameters, there were no significant differences between apocynin-treated WT and the WT vehicle groups (Figures S6A–S6D). Notably, mouse treatment with apocynin did not modulate Aβ accumulation and did not affect tau hyperphosphorylation in the hippocampus of 3xTg mice (Figures S6E–S6H). Apocynin treatment had minor effects in dampening the increase in amoeboid and phagocytic states observed in the hippocampus of 3xTg mice (stage 4, ***p < 0.001 3xTg vehicle versus WT vehicle, **p < 0.01 3xTg Apocynin versus WT vehicle; Figure 6O). These results indicate that NADPH oxidase activity and probably O2−/ONOO− generation contribute to cognitive decline, but not to Aβ accumulation in Alzheimer’s mouse models.
To begin to assess the feasibility and safety of FMD cycles in patients diagnosed with amnestic mild cognitive impairment (aMCI) or mild AD, we started a phase I/II randomized and placebo-controlled (single-blind) clinical study to test the effects of monthly FMD cycles in 40 patients with aMCI or mild AD and adequate nutritional status (see STAR Methods for the inclusion criteria of this trial). Twenty-eight patients have been enrolled to date (13 males, 15 females; average age 71 years, range 55–80 years; Figure 7A). Patients were diagnosed with aMCI or with early-stage AD (Mini-Mental Status Examination [MMSE] score 18–23) according to the international diagnostic criteria and regularly followed up at the Geriatric Unit of the San Martino Hospital (Genoa, Italy) or of the Santa Maria della Misericordia Hospital (Perugia, Italy). After screening and a baseline assessment addressing cognitive performance, functional status and caregiver burden (same assessments are repeated at 6- and 12-month points of study), 12 patients were randomly included in the FMD (active) group, while the other 16 patients were assigned to the placebo group (Figure 7A). The placebo diet assigned to patients in the control arm consists of replacing lunch or dinner with a meal based on pasta or rice with vegetables for 5 days a month, without supplements, whereas patients in the FMD arm complete FMD cycles that last 5 days, with supplements noted for fasting-mimicking, neuroprotective, anti-inflammatory, and antioxidant properties, including olive oil, coconut oil, algal oil, nuts, caffeine, and cocoa, given to patient in between FMD cycles for 25 days and while on a normal diet (Figure 7A). Patients from the FMD arm and from the placebo control arm have received an average of 5.8 (range 1–12) and of 6 (range 1–12) diet cycles, respectively. Five cases of drop-out were recorded in the FMD arm (41.6%) after an average of four FMD cycles. These were due to poor acceptance of the FMD components (n = 2), worsening of the nutritional status (n = 2), or to personal reasons (n = 1). On the other hand, in the placebo diet arm, five cases of drop-out (31.3%) were also recorded (also after an average of four placebo diet cycles), as a result of worsening of the nutritional status (n = 3), poor acceptance of the prescribed diet (n = 1), or personal reasons (n = 1). The FMD-emergent adverse events (graded according to Common Terminology Criteria for Adverse Events 5.0) that have been observed so far are all mild or moderate. They include fatigue (grade 1 [G1]; n = 5, 41.6%), headache (G1 or G2; n = 4, 33.3%), hypotension (G2; n = 1), irritability (G2; n = 1), autoimmune reaction (G2; n = 1; transient worsening of a pre-existing pemphigus), abdominal pain (G1; n = 2), depression (G1; n = 1), and paresthesia (G1; n = 1). The number of patients enrolled in the clinical trial, the number of FMD cycles, and the adverse events for each patient have been included in Table S2. Overall, diet compliance has also been satisfactory during the periods between FMD cycles, when patients take several supplements during the day. Thus, these initial data suggest that 5-day FMD cycles administered once a month have been feasible and overall safe in a small group of patients with aMCI/early AD. However, further monitoring of the enrolled patients and completing the foreseen patient accrual for this clinical study are both necessary to confirm these conclusions. As more patients are enrolled, and cognitive assessments or secondary endpoints are assessed after multiple FMD cycles, more information will be available to determine whether FMD cycles can slow down cognitive decline, delay the conversion rate to AD (for patients with aMCI), and affect biomarkers of inflammation, oxidative stress, neuronal damage, circulating stem cells, and markers of cellular aging.
Results from two AD transgenic mouse models indicate that FMD cycles reduce the levels of key pathological markers, including Aβ and hyperphosphorylated tau, as well as microglia density and markers for neuroinflammation to improve cognition. The results obtained with the 3xTg/Nox2-KO mice and with apocynin treatment support our hypothesis that FMD cycles induce positive effects on the 3xTg and E4FAD models in part by modulating the activity of microglia and possibly brain macrophages allowing it to perform protective functions, including the scavenging of Aβ, while at the same time reducing production of toxic O2−/ONOO−. We show that, in males, there is a significant improvement in oligomeric Aβ42 uptake by microglia isolated from FMD-treated mice. In females, we instead only observed a trend for this effect (Figure S7A–S7D). A recent study showed that microglia activation is higher in AppNL –G-F female mice compared with male mice in response to amyloidosis (Biechele et al., 2020). There is also a higher proportion of activated response microglia (ARMs) cells in 6-month-old and older AppNL–G-F female mice compared with male mice, indicating that the female microglia in an amyloid model of AD are activated earlier than those in males (Sala Frigerio et al., 2019). This could explain the differences in the effects of FMD on the uptake of oligomeric Aβ42 in male versus female microglia. Based on current studies but also on our past studies with periodic essential amino acid restriction, we believe that FMD cycles can affect Aβ levels, but that they protect AD mice by altering both Aβ-dependent and Ab-independent effects, including those on hyperphosphorylated tau and neuroinflammation. Increasing evidence suggests that the NADPH oxidase complex has a crucial and specific role in modulating microglia activation status, and that ROS production by NADPH oxidase and other sources is involved in neurotoxicity and Aβ-dependent microglia proliferation through the release of pro-inflammatory cytokines (Jekabsone et al., 2006). Together with our previous results showing that both O2− and ONOO− promote Aβ- and microglial-dependent neurotoxicity (Longo et al., 2000; Xie et al., 2002), these results suggest that part of the protective effects of the FMD cycles are associated with reduced microglial and Nox2 activation/levels and therefore reduced O2− and ONOO− generation and toxicity, which may contribute to cognitive decline by damaging neurons, increasing tau phosphorylation but potentially also by interfering with synaptic plasticity (Figure 7B). In fact, O2− and other ROS are known to promote synaptic plasticity raising the possibility that a high and continuous stimulation of long-term potentiation (LTP) could eventually interfere with learning and memory (Massaad and Klann, 2011). Notably, because microglia and perivascular macrophages share key markers, and considering that both can have central roles in superoxide production and neuroinflammation related to AD (Park et al., 2017), further studies are needed to determine the relative role of brain microglia and macrophages but also of endothelial cells, astrocytes, and neurons in the effects described here, and specifically in the contribution of NO and O2−. The fact that the great majority of significantly downregulated genes in the FMD 3xTg male and female groups compared with 3xTg on the standard diet were in the neuroinflammation, microglial activation, tau phosphorylation, and AD pathogenesis groups, and that the major set of upregulated genes was in the anti-inflammatory group, is consistent with our model (Figure 7B). One notable finding was that the family of Egr(1–4) genes were all downregulated in FMD-treated 3xTg mouse cortex compared with 3xTg controls (Table S1; Data S1, pages 11–13, 31). In mammals, Egr1 expression is upregulated by the master energy sensor AMPK (Benboubker et al., 2014; Berasi et al., 2006; Andrade et al., 2013), which itself is activated in low-calorie conditions, as well as by drugs such as metformin (Hardie, 2011; Draznin et al., 2012; Blagosklonny, 2009). Interestingly, we previously investigated the role of mammalian EGR1 in cellular protection and its link to glucose and PKA/AMPK, finding that glucose restriction protects cardiomyocytes through AMPK/EGR1 activation (Di Biase et al., 2017). In the context of neurological disorders and injury, it has been suggested that Egr1 is expressed during microglial activation (Raj et al., 2015) and that its inhibition through either siRNA or pharmacological agents may reduce microglial activation and possibly ameliorate the neuroinflammation imposed by AD (Giri et al., 2005). Thus, starvation conditions appear to modulate EGR1 expression differentially in different cell types and tissues, but, in the nervous system, its downregulation is associated with reduced inflammation and pathology. The downregulation of Egr genes that is related to neuroinflammation could also be affected by Egr expression changes in neurons not necessarily related to neuroinflammation. Notably, early growth response 1 (Egr1) gene expression was upregulated in the brain endothelial cells of aged mice (Zhao et al., 2020), and the brains of mice lacking the transcription factor Egr-1 had significantly lower levels of Aβ and β-secretase 1 (BACE-1) compared with WT mice. It was found that Egr-1 promotes Aβ synthesis via transcriptional activation of BACE-1 (Qin et al., 2016). In previous studies, we showed that FMD cycles alternated with normal diet refeeding can promote stem cell-dependent regeneration in the nervous and other systems (Brandhorst et al., 2015; Choi et al., 2016; Rangan et al., 2019; Cheng et al., 2017). Here, we also show that FMD cycles increase the expression and generation of NSCs (as indicated by BrdU+Sox2+ expression), but the contribution of NSCs to the cognitive improvements or connection with Nox2 remains to be investigated. In summary, we propose a role for periodic FMD cycles in reducing the expression and activity of neuroinflammatory and toxic microglial/macrophage genes/proteins, including Nox2, and consequently in reducing O2−/ONOO− while preserving the ability of microglia and possibly other cell types to remove Aβ and other accumulated or damaged proteins or cellular components, consistent with studies in a different AD mouse model (Bruce-Keller, et al., 2011). Our ongoing, placebo-controlled, randomized clinical trial in patients with MCI or early-stage AD provides initial evidence indicating that FMD cycles were feasible and safe overall in the 12 patients assigned to the FMD arm thus far. The enrollment of all patients in this trial will provide more conclusive data on the portion of the patient population for which FMD cycles are feasible and safe, but may also provide initial evidence of their effect against cognitive decline and AD progression.
It is important to note that a limitation of the RNA sequencing data (Table S1; RNA sequencing [RNA-seq] supplemental file) is that it was obtained from the analysis of RNA from hemi-cortex and not from specific cell types such as neurons or microglia. Thus, in future studies, it will be important to identify the source of high Egr gene(s) expression in AD to determine whether their expression is upstream or downstream of neuroinflammation, pathology, and/or cognitive impairment.
Further information and requests for reagents may be directed to and will be fulfilled by the lead contact, Valter D. Longo ([email protected]).
This study did not generate new unique reagents.
All data reported in this paper will be shared by the lead contact upon request. RNA-seq data have been deposited at SRA and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Barnes maze data have been deposited at Mendeley data and are publicly available as of the date of publication. The DOI is listed in the key resources table. This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request ([email protected]).
All animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Southern California. Animals were maintained in a pathogen-free environment, housed in clear shoebox cages in groups of up to five animals per cage with constant temperature, humidity and 12 h/12 h light/dark cycles, and unlimited access to water. Body weight and food intake of individual animals was measured routinely: every day during diet weeks and at least once in the middle of a refeeding week. Mice that appeared weak and/or showed signs of illness were not included in any experiment.
3xTg breeding pairs were obtained from Frank LaFerla, PhD (UC Irvine) under an MTA agreement. For the long-term study using 3xTg mice, male and female 3xTg mice were divided into the ad libitum-fed control (Control) group, the FMD group, and a 7-day protein-restricted diet (approximately 4% protein in composition relative to carbohydrates and fat) group (4% PR) starting at 3 months of age, until approximately 18.5 months of age. Wildtype mice (C57BL/6) were purchased from Jackson Laboratory (Bar Harbor, ME) at 6 months of age and aged to 18 months at the University of Southern California.
For the short-term study using 3xTg mice, male 3xTg mice were divided into the ad libitum-fed control (Control) group and the FMD group starting at ~6.5 months of age, until approximately ~8.5 months of age. FMD cycles were 4 days long followed by 10 days of refeeding. Male 3xTg mice were sacrificed 2 days after refeeding, after the 4th FMD cycle. Female 3xTg mice were divided into the ad libitum-fed control (Control) group and the FMD group starting at 8.5 months of age to assess 1 cycle of FMD and sacrificed after 4 days of FMD prior to refeeding. In another short-term study, male and female 3xTg mice were given FMD starting at 6.5 to ~8.5 months for a total of 5 cycles and the mice were euthanized after the 5th cycle before refeeding. Wildtype mice (C57BL/6) were purchased from Jackson Laboratory (Bar Harbor, ME) at 8.5 months of age.
E4FAD breeding trios were obtained from Mary Jo LaDu, PhD (University of Illinois) under an MTA agreement. Female E4FAD were divided into the ad libitum-fed control (Control) group or the FMD group starting at 3 months of age, until approximately 7–7.5 months of age.
The following male mice were used in this study: 3xTg, Nox2–KO, 3xTg/Nox2-KO and corresponding wildtypes [WT: C57BL/6 (B6), C57BL/6/129S (129/B6) and C57BL/6/129S /C57BL/6 (129/B6/B6), respectively]. For simplicity, in reference to this study, only the following groups are represented in the data of this manuscript: 129/B6 (referred to as Wildtype), 3xTg, and 3xTg/Nox2-KO. Nox2-KO mice present a null allele of the X-linked Cybb gene which encodes the Nox2 subunit of the NADPH oxidase complex and lack phagocyte superoxide production (Pollock et al., 1995). Nox2-KO mice and corresponding WT (C57BL/6) were purchased from Jackson Laboratory (Bar Harbor, ME). Colonies of 3xTg mice and related WT (C57BL/6/129S) were established and bred at the University of Southern California. 3xTg/Nox2-KO mice have been generated in our laboratory as described below: 3xTg males were crossed with females KO for Cybb, the gene coding for the Nox2 subunit of NADPH oxidase. Nox2-KO mice and corresponding wildtype (C57BL/6) were purchased from Jackson Laboratory. Since the Cybb gene is X-linked, all the male progenies were hemizygous for the genes related to AD and lacked Nox2. The F1 male offspring were back-crossed to homozygosity for APP/Tau and PS-1 genes. The correct genotype was determined by PCR (Guoet al., 1999; Pollock et al., 1995) or qPCR by comparing ΔCt values of each unknown sample against known homozygous and hemizygous controls. Similarly, in order to obtain the correct 3xTg/Nox2-KO background, C57BL/6/129S males (129/B6, 3xTg background) were crossed with C57BL/6 (B6, Nox2-KO background). To equalize the genetic background the F1 offspring were backcrossed for four generations. To detect possible side effects caused by Nox2 deletion, 3xTg/Nox2-KO mice were constantly monitored for abnormal changes. Although we did not observe apparent signs of distress in 3xTg/Nox2-KO mice, animals showing first signs of skin lesion were immediately excluded from the study.
A phase I/II randomized and placebo controlled (single-blind) clinical study of a Fasting Mimicking Diet (FMD) in 40 patients diagnosed with amnestic Mild Cognitive Impairment (aMCI) or mild Alzheimer Disease (AD) was designed as follows: The inclusion criteria are: 1) Written informed consent; 2) Age 55–80; 3) Diagnosis of aMCI or mild AD according to the international definition criteria (Petersen, 2004; McKhann et al., 2011); 4) Adequate nutritional status according to the Mini Nutritional Assessment - MNA® Elderly (12–14 points) 5) Body Mass Index (BMI) ≥ 20 kg/m2. 5) Phase Angle (PhA) > 4.7° at the bioimpedance analysis; 6) Normal organ function. The exclusion criteria are: 1) Age >80. 2) Diabetes Mellitus. 3) Liver or renal failure. 4) Food allergies to the components of FMD; 5) Therapy with vitamin K antagonists; 6) Patients who live alone or are not adequately supported by the family context; 7) Other experimental therapies. The primary endpoints of the study are the feasibility and safety of 12 monthly cycles of the FMD in patients with aMCI or mild AD. Feasibility is defined as the consumption of at least one FMD cycle every two months with the possibility of admitting the consumption of only 50% of the planned diet and/or a maximum consumption of 10 Kcal/kg body weight of food not provided in only one of the 5 days of each cycle. FMD-emergent side effects are monitored regularly at each visit. Secondary endpoints are: 1) conversion rate to AD (for patients with aMCI); 2) effect on cognitive performance, functional and emotional status, quality of life (QoL), stress of the caregiver; 3) effect on circulating inflammatory and oxidative stress markers, neuronal damage markers (Neurofilament Light, NfL), circulating stem cells and markers of cellular aging (e.g. evaluation of the telomerase activity of lymphocytes). The FMD tested in this study (ProlonAD TM) is a plant-based dietary regimen that has fasting-mimicking properties, while ensuring an adequate supply of macro- and micronutrients. Every FMD cycle lasts 5 days and is followed by 25 days of a normal diet. The composition of ProlonAD TM is equal to that of the ProlonTM diet (L-Nutra, Los Angeles, CA, https://www.prolon.it) and provides ~43–47% carbohydrates, ~ 44–46% fats and ~ 9–11% proteins per day in the form of vegetable soups, broths, bars, olives, crackers, herbal teas, with the addition of specific supplements (like olive oil, coconut oil, algal oil, nuts, caffeine and cocoa) that substantially increase the daily calories (300–500 kcal) compared to ProlonTM and seem to have neuroprotective, anti-inflammatory and antioxidant properties. The placebo diet assigned to patients in the control arm will simply consist in taking a meal based on pasta or rice with vegetables to replace lunch or dinner for 5 days a month, without supplements. Patients, which are all at low nutritional risk, receive rigorous monitoring of their nutritional status every cycle (monthly) through the collection of anthropometric data, evaluation of body composition by bioimpedance analysis and assessment of muscle strength with the handgrip strength test; moreover, they are invited to perform a light/moderate daily physical exercise to promote protein anabolism and maintenance of muscle mass. In each visit FMD-related side effects are recorded, and a venous sampling is made for the evaluation of markers of inflammation, neuronal injury, cellular aging and oxidative stress. The cognitive performance, functional status and caregiver burden is evaluated at the beginning of the study, at 6 months and at 12 months with a complete neuropsychological and geriatric assessment [Free and Cued Selective Reminding Test (FCRST), Addenbrook’s Cognitive Examination- Revised (ACE-R), Stroop Test, Babcock Test, Digit Span, Trail Making Test A and B, Raven Test, Token Test, Barthel Index, Rockwood’s Frailty Index, CDR-Sum of the boxes, Activities of Daily Living (ADL), Instrumental Activities of Daily Living (IADL), Center for Epidemiologic Studies Depression Scale (CESD-R), Caregiver Burden Inventory, Neuropsychiatric Inventory (NPI)].
Mice were fed ad libitum with PicoLab Rodent Diet 20 (LabDiet). The FMD is based on a nutritional screen that identified ingredients that allow nourishment during periods of low-calorie consumption (Brandhorst et al., 2015). The FMD formulation fed to the AD mouse models consists of two different components designated as day 1 diet and day 2–4/5 diet (in the long-term study, male 3xTg mice consumed 4 days of diet and female 3xTg mice consumed 5 days of diet): The day 1 diet consists of a mix of various low-calorie broth powders, a vegetable medley powder, extra virgin olive oil, and essential fatty acids; day 2–4/5 diet consists of the same ingredients as in day 1, with the addition of glycerol. The diet formulations also contain hydrogel (ClearH2O) to achieve binding and to allow the supply of the food in the cage feeders. Day 1 is composed of approximately 55% fat, 36% carbohydrate and 9% protein, with mice consuming anywhere between 12.36 kJ-17.31 kJ. Day 2–4/5 is composed of approximately 43.3% fat, 47.4% carbohydrate, and 9.3% protein, with mice consuming anywhere between 8.17kJ – 11.44 kJ a day. The 4% protein-restricted diet is based on previous amino acid restriction studies with 3xTg mice (Parrella et al., 2013). A 7-day diet composed of approximately 75.67% carbohydrate, 3.90% casein protein, and 20.43% fat was fed to 3xTg male and female mice in the 4% PR group, with mice consuming anywhere between 19.36 kJ – 27.10 kJ per day. Mice consumed all the supplied food on each day of the FMD regimen and showed no signs of food aversion. At the end of either diet, we supplied PicoLab Rodent Diet 20 chow ad libitum for 10, 9, or 7 days before starting another FMD or 4% PR cycle. Prior to the FMD, animals were transferred into fresh cages to avoid feeding on residual chow and coprophagy.
5′-bromo-2′-deoxyuridine (BrdU; Sigma-Aldrich) was administered to male and female 3xTg mice in the long-term study at a 50 mg/kg dose for 7 consecutive days before mice were sacrificed. BrdU powder was dissolved in boiling saline and prepared fresh each day of injection.
Male 3xTg and corresponding wildtype (129/B6 background) mice were housed 3–5 per cage and, starting at 8 months of age, they were treated with 1 mg/mL apocynin (Sigma-Aldrich) dissolved in drinking water or apocynin-free water for 6 months. Dose and route of administration of the drug were chosen according to previous reports (Harraz et al., 2008). The mice were assigned to the experimental groups by simple randomization. The time point of drug treatment was chosen based on previous studies showing that at 8 months of age 3xTg rodents display detectable memory deficits and their brains show Aβ and tau neuropathology (Mastrangelo and Bowers 2008; Oddo et al., 2003). Apocynin powder was dissolved into hot (~60°C) sterile water and then allowed to cool to room temperature before being given to the mice. The drug was replaced every week since between 5-7 days any decay of apocynin is detectable (Harraz et al., 2008). Considering the average daily water intake of approximately 4 mL/mouse ((Bachmanov et al., 2002) and Figure S6I) the mg/kg dose was estimated approximately 150 mg/kg body weight/day. In order to detect possible adverse effects related to the apocynin treatment, the animals were weighed weekly and monitored for abnormal changes during the treatment period. During the final three weeks of treatment, while they were still under treatment, the mice were tested with the indicated behavioral tests. At the end of the treatment the animals were euthanized as described previously. Since we did not find significant differences between wildtype mice treated with apocynin or vehicle (apocynin-free water) in any of the parameters analyzed, the results obtained with apocynin-treated wildtype mice are not shown.
Short-term working memory was examined by a spatial novelty preference task in the Y-maze (Parrella et al., 2013). The Y-maze was made of black plexi-glass and was designed with three identical arms (50 × 9 × 10 cm), radiating from a central triangle (8 cm on each side) and spaced 120° apart from each other. Mice were placed at the bottom of one of the arms of the maze at the beginning of the test and allowed to freely explore the environment for 8 min. The total number of arm entries and arm choices were recorded. Arm choices are defined as forepaws fully entering the arm. SAB was calculated as the proportion of alternations, defined as an arm choice different from the previous two choices, to the total number of alternations.
The novel object recognition task was used to assess the ability of rodents to recognize a novel object in a familiar environment (Parrella et al., 2013). The test includes a habituation phase (5 min on day one) and trial phases (5 min each on the second day) for each mouse. Briefly, in the habituation phase, the mouse was placed into a rectangular cage (50 x 50 x 40 cm) made of black plexi-glass for 5 min on day one without any objects. The following day, mice underwent two test trials, the duration of each trial being 5 min. Mice were always placed in the apparatus facing the wall at the middle of the front segment. Exploration of the objects is defined as any physical contact with an object (whisking, sniffing, rearing on or touching the object) as well as positioning its nose toward the object at a distance of less than 2 cm. Sitting or standing on top of the object is not counted toward the exploration time. In trial 1, mice explored the arena with two identical objects in presence. After the first exploration period, mice were placed back in their home cage. To control for odor cues, the open field arena and the objects were thoroughly cleaned with 70% ethanol, dried, and ventilated for a few minutes between mice. After a 5-min* delay interval (related to Figure 5 et al., 2003), mice were placed back in the apparatus for the second trial, but with one of the objects from the first trial placed with a new one. Recognition index (RI) was calculated as time the animals spent exploring the novel object to the total time spent exploring both the objects (Parrella et al., 2013).
The mice treated with apocynin or apocynin-free water were tested for short-term spatial memory using a maze consisting of an opaque plastic box measuring 61 cm (length) x 36 cm (width) x 30 cm (height). Briefly, on the first day of test (habituation day) the mice were placed in the box and allowed to explore the field for 5 min. Twenty-four hours later (test day) habituated mice were placed again into the box in the presence of two identical, non-toxic objects and were allowed to freely explore them for 5 min (trial 1). The time spent exploring the objects was recorded considering exploration as any physical contact with an object and/or approach with obvious orientation to it within 5 cm. At the end of the trial 1 the animals were returned to their home cage. After 3 min the mice were returned to the testing field where one of the familiar objects was replaced by a novel object. The mice could explore the arena for 5 min and the time exploring the objects monitored again. Recognition index (RI) was calculated as time the animals spent exploring the novel object to the total time spent exploring both the objects (Parrella et al., 2013).
The Barnes maze protocol used here was based on previous protocols (Barnes, 1988; Michán et al., 2010). The maze consists of a platform with 20 holes (San Diego Instruments) and 20 boxes underneath each hole. A nestlet was placed in one box (escape box, “EB”), which was big enough for the mouse to enter/hide. All three walls around Barnes maze have different reference cues to learn the position of the escape hole. A unique position for the EB based on the position of the reference cues was randomly assigned to each mouse and this position was always located underneath the same hole for a specific animal. All mice were trained once daily on days 0 to 7. During training sessions, mice could freely explore the maze until either entering the EB or after 2 min time elapsed. On Day 0, the mice were either identified as active or inactive based on whether the mice move around the center of the maze or explored the holes at the periphery of the maze. If the mouse did not enter the EB by itself, it was gently guided to and allowed to stay in the EB for 30 s. After the training session, mice were tested twice daily for 7 days. Testing was like training, but if after 2 min the mouse did not find the EB, it was directly returned to its cage. Mice were acclimated to the behavior room an hour before the test began each day. The buzzer with a noise level of 80 dB was always switched on when the mice are in the behavior room except when the mouse finds the escape box. Success rate (100%, finding the escape box [EB] within 2 min; 0%, not finding the EB within 2 min), latency (time to enter the EB), number of errors (nose pokes and head deflections over false holes), deviation (how many holes away from the EB was the first error), and strategies used to locate the EB were recorded and averaged from two tests to obtain daily values. Search strategies were classified as random (crossings through the maze center), serial (searches in clockwise or counterclockwise direction), or spatial (navigating directly to the EB with both error and deviation scores of no more than 3). Retention was assessed by testing once on day 14. Mice that did not move in any of the trials on any day were excluded from the study.
The Elevated Plus Maze is used to assess anxiety behaviors, being based on rodent exploratory behavior and by a rodent’s natural aversion against open space. The avoidance of elevated open arms is an indication of the intensity of anxiety (Carroll et al., 2010). The maze has the shape of a cross formed by two alternate open and two alternate closed arms extending from a central platform, each arm measuring 30 cm in length, 5 cm in width, and 15 cm in height. During the test the mouse was placed onto the center field and allowed to freely explore the maze for 5 min, and the time spent in the open arms, corresponding to lower anxiety levels, was measured (Parrella et al., 2013).
The FC test was used to measure contextual and tone learning in the 3xTg/Nox2-KO mice. The apparatus and the protocol used for FC test have been described previously (Liu et al., 2004). Both the initial training and the contextual testing were conducted in the same conditioning chamber, while tone testing was performed in a novel chamber. The conditioning chamber (27 x 28 x 30.5 cm) was built of plexi-glass (front and back) and aluminum (top and sides), whereas the floor consisted of stainless-steel rods connected to a shock generator (Precision Controlled Animal Shocker; Coulbourn Instruments, Allentown, PA). The chamber was equipped with a speaker and a house light. For tone testing in a novel environment, the new chamber consisted of an aluminum box (27 x 21.5 x 17 cm). A house light and the speaker were placed on top of the box. Briefly, on day 1 of training (training day), the animals were placed in the conditioning chamber. After a 3 min baseline period, the mice were exposed to 3 mild tone-footshock pairings (tone, 20 s, 80 dB, 2 kHz; footshock, 1 s, 0.5 Ma at the termination of the tone; intertrial interval, 1 min). The delivery of the footshock was controlled by software (LabLinc Operant Control Software; Coulbourn Instruments). One minute after the third shock, mice were returned to their home cages. Twenty-four hours later (day 2), contextual learning of the mice was assessed by returning the mice to the conditioning chamber for 8 min (context test). 48 h later, mice were evaluated for cued fear conditioning to the tone in a novel context. Mice were placed in a new chamber and, following 3 min of exploration (day 3, first 3 min), the same tone delivered on day 1 was played for 8 min (day 3, final 8 min). The chamber was cleaned with 70% ethanol between individual sessions on day 1 and day 2 and with 5% ammonium hydroxide on day 3. Mice behavior was monitored using a video camera mounted in front of the test chambers. Freezing behavior, defined as the lack of all body movement except for that associated with respiration, was scored during the three days of tests by two observers blind to mouse identity.
Adult mice were anesthetized with isoflurane, punctured in the heart for serum, followed by intracardial perfusion with 4% paraformaldehyde (PFA). The tissues were removed immediately and post-fixed in 4% PFA for 24 h (long-term 3xTg mice study) and stored in 0.05% sodium azide. For E4FAD and short-term 3xTg studies, mice were perfused with saline. Brains were removed immediately and cut in half. One half was post-fixed in 4% PFA for 48 h and stored in 0.05% sodium azide afterwards, while the other half was dissected for the cortex and hippocampus, both of which were flash-frozen at −80C until further processing. Brains from 3xTg/Nox2 mice were collected, immersion-fixed in fresh 4% paraformaldehyde/0.1 M PBS for 48 h and then stored at 4°C in 0.1 M PBS/0.2% sodium azide until further processing. Hemi-brains were cut in a sagittal direction (40 μm) using a vibratome VT1000S (Leica) and stored in 0.05% sodium azide solution until ready for immunohistochemistry (IHC). Every 6th section was immunostained in the subsequent protocols.
For 3,3′-diaminobenzidine (DAB)-based protocols, sections were washed in TBS, pre-treated for antigen retrieval when necessary, incubated in methanol and H2O2 and blocked in 2% BSA for 1 h before incubating with primary antibodies overnight. The following day, sections would be washed in TBS, incubate with secondary antibodies, ABC Vector Elite (Vector Laboratories), and DAB kits (Vector Laboratories) before finally being mounted on slides and air-dried. Primary antibodies used for this protocol were beta amyloid (1:300; Thermo Fisher Scientific), Phospho-Tau (Ser202, Thr205) Antibody (AT8) (1:1000; Thermo Fisher Scientific), BrdU (1:500; Novus Biologicals), CD11b (1:500; Bio-Rad),Iba1 (1:500; Wako) and CD68 (1:500; Bio-Rad) . Secondary antibodies used for this protocol were biotinylated goat anti-rabbit IgG (for beta amyloid, 1:500; Vector Laboratories), biotinylated horse anti-mouse IgG Antibody, rat adsorbed (for AT8 and Iba1, 1:500; Vector Laboratories), biotin-SP-AffiniPure donkey anti-rat IgG (H + L) (for BrdU, 1:250; Jackson ImmunoResearch); donkey anti rat IgG (H + L), biotin (for CD11b, 1:250; Thermo Fisher Scientific) and biotin-SP-AffiniPure donkey anti rat IgG (H + L) (for CD68, 1:500; Jackson ImmunoResearch).
For fluorescent IHC, sections were rinsed 3 times in phosphate buffered saline (PBS) for 5 min and denatured in 2N HCl at 37°C for 20 min. Sections were neutralized with 0.1 M boric acid for 10 min and blocked with 2% Normal Donkey Serum (NDS; Jackson ImmunoResearch) for 1 h at room temperature. Sections were incubated with primary antibodies overnight in a solution containing 2% NDS and 0.3% triton at 4°C. The following day, sections were rinsed 3 times in PBS for 10 min, followed by incubation with secondary antibodies. After subsequent washes in PBS, nuclei were stained with Hoechst 33342 (Thermo Fisher), washed in PBS, and mounted on slides (except for GFAP + Sox2+ staining as the mounting medium contained DAPI). Slides were cover-slipped with antifading polyvinyl alcohol mounting medium with DABCO (Sigma-Aldrich) or VECTASHIELD® HardSet™ Antifade Mounting Medium with DAPI (Vector Laboratories) immediately after. Primary antibodies used for this protocol were BrdU (1:200; Novus Biologicals), Ki67 (1:200; Thermo Fisher Scientific), Sox2 (1:200; Abcam) and GFAP (1:200; Cell Signaling). Secondary antibodies used for this protocol were donkey anti-rat-488 (for BrdU, Ki67 and GFAP, 1:400; Thermo Fisher Scientific) and donkey anti-rabbit-594 (for Sox2, 1:400; Thermo Fisher Scientific). Co-expression was confirmed by fluorescent microscopy. For quantification, stereological counting methods were used.
To enhance Aβ immunoreactivity (IR), sections were rinsed for 5 min in 99% formic acid. Aβ IR was calculated as load values. Briefly, selected fields of nonoverlapping immunolabeled sections of hippocampus (one-two fields for subiculum and two-three for CA1–Cornu Ammonis area 1) were captured at 20× and digitized using a video capture system coupled to a microscope (Olympus BX50 microscope and Olympus DP73 camera). Using ImageJ software (NIH), software, images were converted into binary/negative data and the positive pixels (equivalent to IR area) were quantified (Carroll et al., 2010).
AT8-immunoreactive neurons were defined as cells showing strong AT8 immunolabeling over most of the cell surface. The positive cells were counted within the hippocampal CA1 and subiculum regions. Images were captured with a video capture system coupled to a microscope (Olympus BX50 microscope and Olympus DP73 camera) at 20×.
CD11b-immunoreactive (ir), Iba1 positive and CD68 positive microglia cells were defined as cells covered by CD11b, Iba1 or CD68 immunostaining over the cell body and processes. Sections were either treated with 10 mM EDTA (pH 6) or Citric Acid Buffer (pH 6) at 95°C for 10 min before proceeding with staining. CD11b-ir or Iba1 cells were counted in the subiculum and CA1 hippocampal regions. There was no pre-treatment with 10 mM EDTA (pH 6) or Citric Acid Buffer (pH 6) for required the CD68 Antibody.CD68+ cells were counted in the subiculum and CA1 hippocampal regions. The stage of cells activation was identified by their morphology. Briefly, we defined four stages of microglia activation based on criteria established in previous studies (Zhang et al., 2011; Kreutzberg, 1996; Crews and Vetreno, 2016): Stage 1: Resting microglia. Rod-shaped soma with many long thin ramified processes. Stage 2: Activated ramified microglia. Elongated cell body, the processes are thicker. Stage 3: Amoeboid microglia showing a marked cellular hypertrophy and short and thick processes. Stage 4: Phagocytic cells. Round cells, processes are not detectable. Cells in the different activation stages were counted and plotted as percentage of the total CD11b-ir or Iba1 cell number. Images were captured with a video capture system coupled to a microscope (Olympus BX50 microscope and Olympus DP73 camera) at 20×.
The levels of BrdU (DAB and fluorescent), Sox2, and Ki67 were quantified in the dentate gyrus. Images were captured at 20× and captured with a video capture system coupled to a microscope (Olympus BX50 microscope and Olympus DP73 camera) or with the BZ-X710 All-in-One Fluorescence Microscope (Keyence) and analyzed with ImageJ (NIH). Individual BrdU+, Sox2+, Ki67+ and GFAP+ cells, as well as BrdU+Sox2+, Ki67+Sox2+ and GFAP+Sox2+ co-stained cells, were counted if they were located within the subgranular zone (SGZ) and up to the inner third of the granule cell layer. Every sixth section of the mouse hemi-brain containing the hippocampus was stained. The numbers quantified are the averages taken from the total number of sections stained for each hemi-brain.
Soluble Aβ peptides (Aβ38/40/42) were extracted from the cortex of hemi-brains of female E4FAD 7–7.5 months old, after undergoing 4 months of FMD treatment. A serial extraction process previously described was used to produce TBS-soluble and TBS plus triton-soluble fractions (Youmans et al., 2011). The amount of soluble Aβ per isoform per sample in the resulting extracts were evaluated using the V-PLEX Plus Aβ Peptide Panel 1 (4G8) (Meso Scale Diagnostics). Plate results were read using the MSD Technology Platform and associated software.
Magnetic Dynabeads (Invitrogen) were re-suspended and 50 μL was transferred to 8 tubes. Beads were separated from supernatant on a magnet and the supernatant was removed. 2.5 μL of rat anti-mouse CD11b antibody (1 μg/μL, Bio-Rad, MCA711GT) was diluted in 200 μL of Ab Binding and Washing Buffer and the immunoprecipitation procedure was carried out using the Dynabeads™ Protein G Immunoprecipitation Kit (Thermo Fisher Scientific, Waltham, MA, USA) according to manufacturer’s instructions. 500 ug of target antigen (Aβ42) was added to the magnetic bead antibody complex. After elution with elution buffer, samples were mixed with 10 μL of pre-mixed NuPAGE™ LDS Sample Buffer and NuPAGE Sample Reducing Agent (Invitrogen) to ensure the resuspension of the magnetic bead-Ab-Ag complex. Samples were heated for 10 min at 70°C. The tubes were placed on a magnet, the supernatant was transferred to a clean tube and prepared for western blot as described in the Western Blot section.
The cortex from E4FAD mouse brains was homogenized in Eppendorf tubes (kept on ice), using 300 ul of RIPA buffer (Thermo Fischer Scientific) with freshly added protease and phosphatase inhibitor cocktails. The samples were placed on an orbital shaker at 4°C overnight and spun the next day at 12000 rpm for 20 min at 4°C. The supernatant from each sample was collected and the protein concentration was assayed using BSA as working standard. Equal amounts of protein (40 μg) were heat-denaturized in NuPAGE LDS sample- loading buffer (Invitrogen) for 5 min at 95°C, resolved by SDS-PAGE and transferred to PVDF membranes (Sigma-Aldrich, MO, USA). The membranes were blocked with Tris-buffered saline (TBS) containing 0.05% Tween and 5% non-fat dry milk (NOX2, Aβ42) and Tris-buffered saline (TBS) containing 0.05% Tween and 2% BSA (Vinculin) and then incubated overnight with antibodies directed against NOX2 (Rabbit monoclonal, 1:5000, Abcam, ab129068), Aβ42 (Rabbit monoclonal, 1:500, Abcam, ab201060) or Vinculin (Mouse monoclonal, 1:1000, Millipore, MAB3574). Peroxidase conjugated IgG was used as secondary antibody. Membrane-bound immune complexes were detected by HyGLO™ Chemiluminescent HRP Detection Reagent (Denville Scientific). Protein loading was normalized according to Vinculin expression. Quantification was performed by densitometric analysis using Imagelab 6.0 software (Bio-Rad).
Nox2 activity in cortex lysates was measured using a Rat CYBB / NOX2 / gp91phox ELISA Kit (LSBio, Seattle, WA, USA, LS-F39030) and performed according to manufacturer’s instructions.
Inflammatory cytokines in the brain were evaluated using aliquots of the female E4FAD hemi-brain TBS-soluble and TBS plus triton-soluble extracts used in the Aβ protein extraction experiment. The amount of IL-10, IL-12p70, IL-1β, IL-6, TNFα, IL-2, KC/GRO, IL-4, and IL-5 cytokines per sample in the resulting extracts were evaluated using V-PLEX Proinflammatory Panel 1 (mouse) Kit (Meso Scale Diagnostics). Plate results were read using the MSD Technology Platform and associated software.
Images were captured using a confocal microscope (Zeiss Laser Scanning Microscope-780 upright microscope) with Zeiss ZEN imaging software. Laser and detector settings were unchanged across acquisition sessions. 3D images of sections labeled with anti-Iba1 were collected with optimal resolution z-stack depths. Stack section depth for low magnification images was 2.2 μm and 0.4 μm for high magnification images. A 63x oil immersion objective (1.4 NA) was used to acquire region of interest (ROI) stacks (192.8 μm x 192.8 μm, 512 x 512 pixels, 16 bit). Analysis was conducted in a researcher-blinded manner with a custom ImageJ plugin (Stephen et al., 2015) and de-noised using background subtraction rolling ball radius of 50 pixels. Average intensity stack projections were used for analysis (unless otherwise stated). Microglial circularity, a measure of roundness/activation, was calculated using the formula 4π x area / (perimeter)2. Data represents average of all cells per group. Microglial soma size was measured by manually identifying, thresholding, outlining, and measuring Iba1-immuno-reactive cell bodies (~ 30 cells/group) using ImageJ.
Murine-derived hemi-cortex samples were prepared as lysates using the RNeasy Lipid Tissue Mini Kit (Qiagen) and RNase-Free DNase Set (Qiagen), according to manufacturer’s protocols. RNAseq libraries were prepared using KAPA mRNA HyperPrep kit (Roche-Sequencing) on a custom-built Janus G3 liquid handler (Perkin-Elmer). The quality of the final libraries was assessed with the D1000 assay on a 2200 TapeStation (Agilent Technologies) and the pooled samples were sequenced 1 × 50 bp on an Illumina HiSeq3000 system, yielding 24–52 M reads per sample for 31 samples. Reads were quality checked with FastQC v0.11.8 (Andrews, 2010), filtered with Trimmomatic v0.38 (Bolger et al., 2014), and aligned to the Mus musculus GRCm38 reference genome with GENCODE M22 annotations using STAR v2.6.1d (Dobin et al., 2013) with a single-pass approach, retaining uniquely mapped reads with less than 5 mismatches. Gene expression was quantified using HTSeq (Anders et al., 2015) over exons in union mode, and analyzed using the R package EdgeR (McCarthy et al., 2012; Robinson et al., 2010). Four samples that appeared to be technical outliers in the FastQC and gene expression analyses were discarded. Gene expressions for the remaining 18 male and 9 female samples were separately TMM-normalized and between-diet group differences were assessed using EdgeR’s GLM approach. Gene ontology analyses were performed using the R packages gProfileR v0.6.8 (Reimand et al., 2007) and GO.db (Carlson, 2019). GO terms for the significantly expressed genes are listed in Table S1.
Mixed glia were cultured from the brains of 8.5 month old 3xTg male and female mice administered either a control diet or 5 cycles of FMD with 4 days of refeeding (n = 1–2 mice/group). The mice started the diet at 6.5 months till 8.5 months of age. The brains were perfused using DPBS and the whole brain (except the cerebellum) from each of the 5 mice (3 female mice-2 FMD and 1 Control and 2 male mice- 1 FMD and 1 Control) were placed in 5 mL of DMEM/F12 (Gibco) and stored on ice. Whole brains were mechanically and enzymatically dissociated by first mincing with a sterile razor blade and then incubated with 2 mL of serum free DMEM/12 media containing 10 units/mL papain for 25 min followed by the addition of 40 ul of Dnase (10 mg/mL stock, final 0.2 ug/mL) for 5 min at 37°C. Enzymatic activity was quenched by addition of 500 μL of fetal bovine serum followed by addition of 2 mL of serum free media at room temperature. Once the serum free media was added, large clumps of tissue were broken up by pipetting using a 1 mL pipette. After the cell pellets settle for two minutes, the supernatant was transferred to a fresh falcon tube. This step was repeated twice more. Isotonic Percoll was added to the cell suspension (final concentration: 20%). Isotonic Percoll (ISP) was prepared by mixing nine parts of Percoll (Percoll PLUS, Cytiva) with one-part 10× HBSS. To collect the glial-enriched fraction, the cell suspension was centrifuged for 20 min at 700×g in a 20% ISP/cell suspension. The supernatant was aspirated using a 5 mL stereological pipette, and the cell pellets were re suspended in 8 mL of DMEM/F12 supplemented with 10% FBS and a penicillin G and streptomycin cocktail. The resultant suspension was transferred through a 100 μm cell strainer using a 5 mL stereological pipette on petri plates coated with 10 μg/mL of Poly-D Lysine. Cultures were maintained at 37°C in a humidified incubator for 10 days with room air supplemented with 5% CO2. The media was replaced 3 times per week. After the primary mixed glial cultures were incubated for 10 days, 2500 microglial cells per well were seeded on 96 well plates coated with 10 ug/mL of Poly-D Lysine and incubated for two days at 37°C in a humidified incubator with room air supplemented with 5% CO2. HiLyte™ Fluor 555-labeled Amyloidβ 42 (Aβ42) (AnaSpec, Liège, Belgium) was dissolved in 100% 1,1,1,3,3,3-Hexafluoro-2-propanol and subsequently dried followed by evaporation. The peptide film was dissolved in 10 mM NaOH and neutralized by adding pH 7.4 PBS (final concentration of Aβ42: 20 μM). After the sample was vortexed for 30s, Aβ42 was incubated at 4°C for 24 h for oligomerization and was used for assay immediately where it was added to the seeded cells for an hour to quantify the levels of Amyloid beta uptake in FMD compared to the control group. The microglia were fixed with 4% PFA and stained with IBA-1 (1:200; Wako) to confirm the cell type. The Secondary antibody used for this protocol was donkey anti-mouse-488 (for IBA-1, 1:500; Thermo Fisher Scientific). Images are taken with the BZ-X710 All-in-One Fluorescence Microscope (Keyence) and analyzed with ImageJ (NIH). The area of Amyloid beta 42 to IBA-1 area was quantified to identify the percentage of Aβ 42 uptake in each group.
The software used for statistical analysis was GraphPad Prism v.8. The figure legends describe the statistical tests used, value of n for each experimental group, and what n represents for each experiment. All statistical analyses were two-sided and p values <0.05 were considered significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). Differences between the means of two groups were tested with unpaired 2-tailed student’s t-test comparison, and one-way or two-way ANOVA followed by Bonferroni, Tukey’s, or Fisher’s post-test among multiple groups. Unless otherwise specified in figure legends, all data are expressed as the mean ± SEM. All samples represent biological replicates. No samples were excluded from analysis, and animals were only excluded if they fell ill based on protocol. Sample size estimates were not used. Animals were randomly assigned to experimental groups.
Clinical trial registry (NCT05480358) for the human FMD trial data described in the manuscript can be found here. (description:https://clinicaltrials.gov/ct2/show/NCT05480358?term=NCT05480358&cntry=IT&draw=2&rank=1). |
|
PMC9648508 | 35705209 | Yaw-Syan Fu,Ning Kang,Yanping Yu,Yan Mi,Jialin Guo,Jingyi Wu,Ching-Feng Weng | Polyphenols, flavonoids and inflammasomes: the role of cigarette smoke in COPD | 15-06-2022 | COPD is predicted to become the third leading cause of morbidity and mortality worldwide by 2030. Cigarette smoking (active or passive) is one of its chief causes, with about 20% of cigarette smokers developing COPD from cigarette smoke (CS)-induced irreversible damage and sustained inflammation of the airway epithelium. Inflammasome activation leads to the cleavage of pro-interleukin (IL)-1β and pro-IL-18, along with the release of pro-inflammatory cytokines via gasdermin D N-terminal fragment membrane pores, which further triggers acute phase pro-inflammatory responses and concurrent pyroptosis. There is currently intense interest in the role of nucleotide-binding oligomerisation domain-like receptor family, pyrin domain containing protein-3 inflammasomes in chronic inflammatory lung diseases such as COPD and their potential for therapeutic targeting. Phytochemicals including polyphenols and flavonoids have phyto-medicinal benefits in CS-COPD. Here, we review published articles from the last decade regarding the known associations between inflammasome-mediated responses and ameliorations in pre-clinical manifestations of CS-COPD via polyphenol and flavonoid treatment, with a focus on the underlying mechanistic insights. This article will potentially assist the development of drugs for the prevention and therapy of COPD, particularly in cigarette smokers. | Polyphenols, flavonoids and inflammasomes: the role of cigarette smoke in COPD
COPD is predicted to become the third leading cause of morbidity and mortality worldwide by 2030. Cigarette smoking (active or passive) is one of its chief causes, with about 20% of cigarette smokers developing COPD from cigarette smoke (CS)-induced irreversible damage and sustained inflammation of the airway epithelium. Inflammasome activation leads to the cleavage of pro-interleukin (IL)-1β and pro-IL-18, along with the release of pro-inflammatory cytokines via gasdermin D N-terminal fragment membrane pores, which further triggers acute phase pro-inflammatory responses and concurrent pyroptosis. There is currently intense interest in the role of nucleotide-binding oligomerisation domain-like receptor family, pyrin domain containing protein-3 inflammasomes in chronic inflammatory lung diseases such as COPD and their potential for therapeutic targeting. Phytochemicals including polyphenols and flavonoids have phyto-medicinal benefits in CS-COPD. Here, we review published articles from the last decade regarding the known associations between inflammasome-mediated responses and ameliorations in pre-clinical manifestations of CS-COPD via polyphenol and flavonoid treatment, with a focus on the underlying mechanistic insights. This article will potentially assist the development of drugs for the prevention and therapy of COPD, particularly in cigarette smokers.
COPD has an increasing prevalence and is anticipated to become the third main cause of morbidity and mortality in the global population by 2030 [1]. Vitally, COPD can be caused by prolonged exposure to harmful particles, particulates and gases, infection with acute or chronic inflammatory conditions, or by injuries that include the airways, pulmonary vasculature and lung parenchyma. COPD is categorised by persistent airflow obstruction instigated by exposure to irritants; for instance, cigarette smoke (CS), dust and fumes [2]. Pulmonary injury comprises the phases of initiation (by exposure to CS, pollutants, infectious pathogens and agents), progression and consolidation. The involvement of complex interactions among oxidative stress, inflammation, extracellular matrix proteolysis, autophagic and apoptotic cell death leads to tissue damage [3]. The inflammatory response to progressive inflammation and tissue oxidative stresses leads to airway remodelling, which limits airflow and causes airway obstructions and tissue destruction, resulting in a decrease of forced expiratory volume [4]. In other airway obstructions, a major feature of COPD is pulmonary emphysema, which is described as structural changes and destruction of the alveolar air sacs that lead to decreased gas-exchanging surfaces and impaired gas exchange [5]. From a physio–pathological viewpoint, COPD is concomitant with chronic inflammation influencing the lung parenchyma and peripheral airways that cause irreversible and progressive airflow limitations. This inflammation is indicated by augmented numbers of alveolar macrophages, neutrophils, T-lymphocytes (cytotoxic T-cell type 1, T-helper cell (Th) 1 and Th17 cells) and intrinsic lymphoid cells engaged from the circulation [6]. These cells, as well as structural cells encompassing epithelial and endothelial cells and fibroblasts, secrete a variety of pro-inflammatory mediators, including cytokines, chemokines, growth factors and lipid mediators [7]. Interleukin (IL)-18 is a pro-inflammatory cytokine that was first described as an interferon (IFN)-γ-inducing factor. Similar to IL-1β, IL-18 is synthesised as an inactive precursor that requires processing by caspase-1 into an active-form cytokine [8]. Moreover, several studies used a mouse model to show that IL-18 over-expression results in emphysematous lesions. The data prompts the hypothesis that IL-18 induces an extensive range of COPD-like inflammatory and remodelling responses in the murine lung and induces mixed IL-1, IL-2 and IL-17 responses. IL-18 has been identified as a latent target for future COPD therapeutics to restrain the damaging and remodelling processes arising in COPD lungs [9].
The functions of innate immune responses act as a first-line defence upon exposure to potentially deleterious stimuli. The innate immune system has evolved various extracellular and intracellular receptors that carry out surveillance for possibly detrimental particulates. Inflammasomes are defined as intracellular innate immune multiplex proteins that are formed and activated upon subsequent interaction with these stimuli [7]. The canonical inflammasomes can be elicited by pathogen-associated molecular patterns (PAMPs) or damage-associated molecular patterns (DAMPs) [10]. Two families of pattern-recognition receptors (PRRs), including the nucleotide-binding domain, leucine-rich repeat containing proteins (NLRs) and absent in melanoma 2 (AIM2)-like receptors, act as sensor components to stimulate the inflammasome [11]. The NLR family consists of several subclasses, including nucleotide-binding oligomerisation domain-like receptor family, pyrin domain containing protein (NLRP) 1, NLRP3, NLRP6, NLRP12 and NLR family CARD domain-containing protein 4 (NLRC4; also known as interleukin converting enzyme protease activating factor (IPAF)). In fact, the NLR family has two common features: the first one is a nucleotide-binding oligomerisation domain, which is bounded by ribonucleotide-phosphates, and the second one is the C-terminal leucine-rich repeat (LRR), which functions as a ligand-recognition domain [12]. In addition, the AIM2-like receptor (ALR) family encompasses two subclasses: AIM2 and interferon gamma inducible protein (IFI) 16. The common feature of the ALR family is PRRs with one or two hematopoietic, IFN-inducible and nuclear (HIN) domains at the C-terminal and a pyrin domain (PYD) at the N-terminal that can bind with cytoplasmic double-strand DNA (dsDNA) to initiate the inflammasome pathway [13]. In mammals, four major types of inflammasome, including NLRP1, NLRP3, NLRC4 and AIM2, have been confirmed by numerous studies of the signalling mechanisms and their functional effects. The activation of different inflammasomes may contribute to important immune response functions or cause cell pyroptosis [14]. The major domains of different inflammasome protein components can be classified as: caspase recruitment domain (CARD), domain with function to find (FIIND), HIN-200/IF120x domain (HIN), LRR, NAIP (neuronal apoptosis inhibitory protein), CIITA (MHC class II transcription activator), HET-E (incompatibility locus protein from Podospora anserina) and TP1 (telomerase-associated protein) (NACHT), and PYD. In general, an unoligomerised inflammasome may consist of two or three protein subunits. The unoligomerised NLRP1 inflammasome is composed of NLRP1 and caspase-1 (CASP1) protein subunits [15]; the unoligomerised NLRP3 inflammasome is composed of NLRP3, ASC (apoptosis-associated speck-like protein containing a caspase recruitment domain) and CASP1 protein subunits [16]; the unoligomerised IPAF (NLRC4) inflammasome is composed of IPAF and CASP1 protein subunits [17]; and the unoligomerised AIM2 inflammasome is composed of AIM2 and CASP1 protein subunits [18]. The NLRP1 protein comprises five domains: PYD, NACHT, LRR, FIIND and CARD; the NLRP3 protein encompasses three domains: PYD, NACHT and LRR; the ASC protein includes two domains: PYD and CARD; the IPAF protein contains three domains: CARD, NACHT and LRR; the CASP1 protein consists of three domains: CARD, p20 and p10. All four major inflammasomes possess one unique CASP1 protein subunit, and the CASP1 can link to the other inflammasome components via CARD–CARD interaction. The activation on PRRs of inflammasome may cause the autocleavage of CASP1 to release the active-form caspase-1 [10]. Inflammasomes are the intracellular multiprotein complexes that can directly or indirectly trigger one or more caspase signalling pathways to process and secrete various pro-inflammatory cytokines, including IL-1α, IL-1β, IL-18 or transforming growth factor-β (TGF-β), and subsequently engage innate immune responses [19, 20]. During bacterial infection, NLRP1 may be responsible for pathogen recognition and resistance. The NLRP1 inflammasome pathway could be activated by various stimuli, i.e. pathogen lethal toxin, muramyl dipeptide or a reduction in intracellular ATP level [15]. Upon cytoplasmic infection by Gram-negative bacteria, NAIPs act as upstream sensors that bind with bacterial ligands and then co-assemble with NLRC4 to constitute NLRC4 inflammasome, which leads to the release of active-form caspase-1 [21, 22]. Of note, AIM2 inflammasome, a cytosolic dsDNA-sensing inflammasome, takes part in host defences and immune responses to DNA virus infections as well as intracellular bacterial infections [23, 24]. During viral infection, mitochondrial misfunction or loss of membranous integrity, dsDNA can appear in the cytoplasm, and that may elicit AIM2 inflammasome pathways [25]. Some investigations have revealed the involvement of AIM2 inflammasome expression in subsequent accelerated COPD development. The redistribution of AIM2 from nucleus to cytoplasm and co-localisation with the cleaved IL-1β indicates that AIM2 inflammasome can be prompted in the airway epithelium and macrophages of COPD patients and CS-exposure mice [26]. An in vitro study showed that the expression of AIM2 in COPD-derived CD14+ peripheral blood mononuclear cells is higher than that in nonsmokers and smokers [26, 27]. The activation of AIM2 inflammasome in a caspase-1- and caspase-4-dependent fashion can cause IL-1α release but not IL-1β, and that can further promote TGF-β release as pro-fibrotic processes [28]. An in vivo study exhibited that the recruited macrophages or dendritic cells in smoking mice increased the AIM2 expression associated with caspase-11 (analogue with human caspase-4) but not caspase-1. CS exposure still caused an induction of alveoli enlargement in caspase-11 dysfunctional (knockout) mice, but the deposition of collagen in the bronchioles was lower than that in CS-exposure normal mice [27]. NLRP3-knockout mice have protective effects on the CS-induced COPD model, whereas IPAF- or AIM2-knockout mice have no protected potential in the same CS-challenge COPD model [29]. According to the limited available evidence, the pathologic mechanisms of AIM2 inflammasome in CS-induced COPD are different from the NLRP3 inflammasome pathway, and it are permeated by caspase-4 and TGF-β release to promote bronchial tissue fibrosis and exacerbated COPD. In all inflammasomes pathways, the NLRP3 inflammasome could be activated in response to the various types of stimuli by PAMPs and DAMPs, such as toxins, extracellular ATP, cytokines, ion fluxes, endoplasmic reticulum (ER) stress and oxidative stress [30, 31]. The NLRP3 inflammasome entails a cytosolic PRR, an adaptor molecule ASC and the protease precursor procaspase-1. Assembly of the NLRP3 inflammasome complex causes the cleavage and activation of caspase-1 that triggers processing, which releases IL-1β and IL-18 cytokines and causes cell death via pyroptosis [32]. Some inducers, i.e. lipopolysaccharide (LPS), act on the canonical NLRP3 inflammasome pathways to increase cytoplasmic active-form caspase-1, and the synchronous trigger of noncanonical inflammasome pathways to activate caspase-4/11, caspase-5 or caspase-8 as well, and that may modulate cell death by apoptosis or necrosis [33–35]. Macrophages sense danger signals such as bacterial toxins or extracellular ATP derived from tissue damage or infection and initiate the activation of an intracellular molecular complex inflammasome. Pyroptosis is an inflammatory form of programmed cell death (PCD) that is found to be involved in the development of chronic inflammation, i.e. COPD, which is activated by the inflammatory caspase cleavage of gasdermin D (GSDMD) and apoptotic caspase cleavage of gasdermin E [20, 23]. It is critical that the activation of the inflammatory response and inflammasome are tightly regulated [36]. In the immunopathology of COPD, immune and nonimmune inflammatory alterations with oxidative stress imbalances are found, as well as changes in the protease/anti-protease ratio caused by direct and indirect genetic and epigenetic-environmental defects. COPD leads to irreversible tissue damage and chronic inflammation with aberrant tissue repairs, which induces chronic obstruction of the airway, bronchitis and systemic damage [2, 9]. Evidence from a previous review suggested the role of inflammasomes in the pathogenesis of numerous chronic respiratory diseases and acute lung injuries, such as asthma, transfusion-related acute lung injury (ALI), ventilator-induced lung injury, COPD and pulmonary fibrosis [37]. From a physiological viewpoint, the respiratory system provides gas exchange from the external environment for metabolism. The airway and lung are incessantly exposed and challenged to a variety of inhaled pathogen agents, harmful particulates and internal self-derived dangerous signals during their whole lifespan. The innate immune reaction plays a vital role in protecting the pulmonary system from infection and disease, and inflammasome pathways act on the initiation of innate immune responses. Continuously and seriously activated inflammasomes in epithelial cells, macrophages or other immune cells may cause an increasing release of several pro-inflammatory cytokines and the occurrence of cell pyroptosis, which subsequently result in pulmonary tissue remodelling, fibrosis and COPD. Thereby, this article comprehensively reviews and discusses the effects of polyphenols and flavonoids on the inflammasome pathways in CS-induced COPD to provide a deep insight into the benefits of natural medicinal products as pharmaceutics.
Plants that harbour secondary metabolites such as alkaloids, polyphenols, flavonoids, terpenoids and other specific natural compounds have gained a great deal of attention for the treatment of several clinical complications. Considering the promising results obtained using medicinal plants, with few/no side effects, as well as their easy attainment, comprehensive research on herbal plants to treat disease should be anticipated [38–40]. Polyphenols are natural composites found in many plants. Mostly, the polyphenols are categorised in three major classes: flavonoids, lignans and nonflavonoids (stilbenes and phenolic acids). Flavonoids comprise a large class of food components encompassing the flavone, isoflavanone, flavanone, flavonols catechin and anthocyanin subclasses [41–43]. Remarkably, several biological properties of polyphenols have been scientifically demonstrated, including antiallergic, antiviral, antibacterial, anticarcinogenic, anti-inflammatory, antithrombotic, vasodilatory and hepatoprotective effects [44, 45], as well as benefits in age-associated diseases including cancer, diabetes, Alzheimer's, osteoporosis and Parkinson's [46].
With cigarette smoking (active or passive) being one of the chief causes of its occurrence, about 20% of cigarette smokers develop COPD. Furthermore, estimates suggest that from 25% to 33% of COPD patients are nonsmokers. Notably, CS exposure has been observed to result in irreversible damage and sustained inflammation of the airway epithelium, which might lead to COPD, although the exact pathophysiology remains elusive. CS induces the gathering of inflammatory cells (macrophages, neutrophils and lymphocytes), cytokine production, the triggering of inflammasome components (NLRP3, ASC, caspase-1), depression-related behaviours, and the enhancement of glucocorticoid receptor (GR) signalling. One in vivo study investigated the relationship and underlying molecular mechanism of CS-exposure, COPD and depression, and the evidence indicated that glucocorticoid resistance was manifested during central nervous system inflammation due to CS exposure, and a potential crosstalk-underlined mechanism between the brain and lung was found [47]. Recent studies have revealed that mitochondria are engaged in innate immune signalling that shows critical features in CS-induced inflammasome activation, pulmonary inflammation and tissue-remodelling responses. Mitochondrial dysfunction might increase intracellular oxidative stresses, decrease intracellular ATP level and cause mitochondrial DNA leakage into cytoplasm to trigger different inflammasome pathways. It has been revealed that mitochondria play a crucial role on the pathogenesis of CS-induced COPD [48]. While earlier studies have explored the character of membrane-bound Toll-like receptors in CS-induced inflammation, scarce information is available about the role of cytosolic NLRs in modulating CS-mediated inflammatory responses. The key role of the P2X7–NACHT, LRR and PYD domain-containing protein 3–ASC–caspase-1/11–IL-1β/IL-18 axis in CS-induced airway inflammation highlights this pathway as a probable therapeutic target for the treatment of COPD [49]. Triggering receptors expressed on myeloid cells 1 (TREM-1) is a crucial signalling receptor that can amplify pro-inflammatory innate immune responses. The activation of TREM-1 can aggravate the formation of inflammation by activating NLRP3 inflammasome pathways, and blocking TREM-1 may inhibit oxidative stress and decrease pro-inflammatory cytokine expression in order to alleviate ALI [50]. There are two proteins involved in the intercellular communication of epithelial cells: connexins 40 and 43 (Cx40, Cx43). Several studies have indicated that oxidative stress, described as the overproduction of 4-hydroxy-2-nonenal (4-HNE), is a crucial factor in pulmonary fibrosis [51]. CS exposure might induce the activation of nuclear factor-κB (NF-κB) and increase the formation of 4HNE-Cx40 adducts that may cause lung epithelium losses to intercellular junctions and remodelling. After exposure to CS extract (CSE), 16 human airway epithelial (16HBE) cells increase lactate dehydrogenase release, upregulate the transcription and translation of NLRP3 inflammasome, and augment caspase-1 activity; enhanced IL-1β and IL-18 cytokine release has also been observed. In addition, NLRP3 is required to activate caspase-1. The results suggest that 16HBE cells treated with NLRP3 small interfering RNA or a caspase-1 inhibitor to silence NLRP3 expression can cause a decrease of IL-1β release and cell pyroptosis. CSE-induced inflammation contributed to pyroptosis via the reactive oxygen species (ROS)–NLRP3–caspase-1 pathway in the 16HBE cells. Thereby, NLRP3 inflammasome participated in CSE-induced cell damage and pyroptosis, which affords a new insight into COPD [52]. In general, the activation of NLRP3 inflammasome and subsequent cytokine secretion is a life-threatening step for innate immune responses. The activation of NLRP3 inflammasome is triggered by several internal and external factors and consequently results in inflammatory cytokine secretion. Inflammasome formation and activity play risky roles in several pathologies of diseases, such as cardiovascular, digestive, lung, metabolic, renal and central nervous system diseases [53]. CS and CSE exposure may induce inflammation and contribute to epithelial cell pyroptosis through the ROS–NLRP3–caspase-1 pathway in airway and lung tissues [52]. CS exposure leads to acute and chronic injury in lung tissues, as well as acting on the endothelial cells of the cardiovascular system to cause cell pyroptosis and the shrinkage of tunica intima/endocardium and epithelium in the bladder. The activation of NLRP3 inflammasome in endothelial cells by inducible nitric oxide synthase (iNOS) following CS exposure may be mediated by the soluble guanylyl cyclase–cyclic guanosine monophosphate–protein kinase G–tumour necrosis factor-α (TNF-α)-converting enzyme (TACE)–TNF-α pathway [54] as well as CS-induced pyroptosis of bladder tissues by activated the ROS–NLRP3–caspase-1 signalling pathway through the NLRP3 inflammasome activation relative pathways [55]. The nicotine in CS can decrease cell viability and induce ROS generation by triggering the NLRP6 inflammasome and provoke ER stress in human kidney cells to cause chronic kidney disease [56]. GSDMD undergoes proteolytic cleavage with caspase-1 to release its N-terminal fragment, which in turn mediates IL-18 and IL-1β secretions and causes pyroptosis. Pyroptosis is an inflammatory form of PCD, defined as being caspase-GSDMD-dependent. The NLRP3 inflammasome plays an indispensable role in mediating GSDMD activation. Asthma or infections may induce airway inflammation and cell pyroptosis via NLRP3 signalling pathways to cause capase-1 activation. Activated caspase-1 catalyses pro-IL-1β and pro-IL-18 to form IL-1β and IL-18; however, it cleaves GSDMD to form GSDMD N-terminal fragment (GSDMDNterm), which assembles into GSDMDNterm membrane pores and releases pro-inflammatory cytokines, IL-1β and IL-18, and subsequently causes pyroptosis [57, 58]. Current evidence has demonstrated that the NLRP3 inflammasome signalling axis could functionally mediate CS-induced inflammation, cytokine release, mucus production and pyroptosis in airway mucosa, which might be a critical mechanism associated with CS-induced airway remodelling. The involvement of signal pathways or targets in CS-induced lung injury or COPD from current investigations is summarised in table 1. Moreover, the signal pathways or targets of the inflammasomes activated in COPD are illustrated in figure 1.
Currently, there is great interest in the role of inflammasomes in chronic inflammatory lung diseases and their targeting potential for therapy [7]. As alternative therapies, phytocompounds have been broadly used for destruction of inflammatory responses before. Selected phytochemicals have shown inhibitory properties on NLRP3 inflammasome activity in in vitro and in vivo tests [53]. Some flavonoids perform anti-inflammatory effects through the obstruction of NF-κB and NLRP3 inflammasome, suppression of pro-inflammatory cytokine IL-1β, IL-2, IL-6, TNF-α and IL-17A production, down-regulation of chemokines, and decrease of reactive nitrogen species and ROS. The most effective flavonoids for inflammation and modified immune responses are apigenin, quercetin and epigallocatechin-3-gallate (EGCG), although other compounds are still under investigation and cannot be omitted. The promising future of these compounds in different therapies has been discussed [72]. EGCG is the most plentiful catechin in green tea and provides protection against oxidative stress, lipid peroxidation and inflammatory responses caused by CS. Treatment with EGCG can decrease ROS production and suppress the lipid peroxidation induced by CSE and 4-HNE–protein adduct formation in airway epithelial cells. Further, it can inhibit CSE-induced NF-κB activity to decrease pro-inflammatory gene expression (cyclooxygenase-2 (COX-2), IL-6, IL-8, NADPH oxidase 4 (NOX4), iNOS, intercellular adhesion molecule 1, matrix metalloproteinase-9, cyclin-D1) via inhibition of the CSE-induced mitogen-activated protein kinase (MAPK) pathway [73, 74]. EGCG can restore superoxide dismutase (SOD) and catalase activities in the lungs and decrease CS-induced goblet cell hyperplasia and emphysema [75]. CS promotes mucus secretion and has a strongly effects the synthesis and secretion of MUC5AC [76]. Theaflavins are flavonoids extracted from black tea and are major antioxidants with protective effects. These effects are ascribed to the following major polyphenols: theaflavin, theaflavin-3-gallate, theaflavin-3′-gallate and theaflavin-3,3′-digallate [77]. Oral treatments with theaflavins showed the inhibition of epidermal growth factor receptor (EGFR) activation by CS and relieved airway mucous hypersecretion to reduce the levels of mucin MUC5AC [78]. Alternatively, treatment with theaflavin-3,3'-digallate significantly promoted the anti-oxidation capacity of lung tissues and attenuated CSE-induced emphysema and lung injury. Theaflavin-3,3'-digallate can decrease the generation of ROS and the expression levels of TNF-α, IL-1β and IL-6 of BEAS-2B (human bronchial epithelial) cells and inhibits necroptosis via the mediation of p38 MAPK–receptor-interacting serine–threonine-protein kinase 3 (RIPK3)–mixed lineage kinase domain-like pseudokinase (MLKL) signalling pathways. The pro-inflammatory cytokine IL-1 is catalysed by capase-1 from activated NLRP3 inflammasome. The protective effects of theaflavin-3,3'-digallate might inhibit NLRP3 pathways [79]. Flavonoids, e.g. (-)-epicatechin, increase tripartite motif-containing protein 25 and nuclear factor erythroid 2-related factor 2 (Nrf2) protein expression to promote ubiquitin-mediated Keap1 degradation and inhibit the activation of NLRP3 inflammasome by CS stimulation [80]. In the CS-induced airway inflammation of rats, baicalin treatment can modulate the histone deacetylase 2 (HDAC2)–NF-κB–plasminogen activator inhibitor type-1 signalling pathway to attenuate inflammation [81]. In the CS-induced COPD model, the TREM-1 level was increased, which activated the NLRP3 inflammasome pathway to aggravate tissue inflammation and lung injury [68]. TREM-1 is a crucial signalling receptor that can amplify pro-inflammatory innate immune responses. The activation of TREM-1 can aggravate the formation of inflammations by activating NLRP3 inflammasome pathways. Blocking TREM-1 may inhibit oxidative stress and decrease pro-inflammatory cytokine expression in response to ALI [50]. TREM-1, as the macrophage's cell surface receptor, is involved in the spread of the inflammatory response to bacterial infections or LPS challenges. Pre-treatment with quercetin, resveratrol and tea polyphenols can inhibit the secretion/shedding of soluble TREM-1 [82–84]. Quercetin is one of polyphenols that functions as an ROS scavenger and adenosine monophosphate-activated protein kinase (AMPK) activator. Peripheral blood mononuclear cells from COPD patients treated with quercetin can activate AMPK and promote the expression of Nrf2, which was found to reverse CSE-induced corticosteroid insensitivity [85]. Quercetin can protect and reduce the oxidative stress of macrophages from CSE exposure via reducing the levels of leukocytes, oxidative stress, histological pattern alterations of pulmonary parenchyma and lung function alterations [86]. In vitro CSE-induced results showed that several fatty acid esters of quercetin-3-O-glucoside had a protective potential against nicotine-induced cell death, membrane lipid peroxidation, and ameliorated the inflammation biomarkers COX-2 and prostaglandin E2 expression [87]. An in vitro study has shown that jaboticabin and 3,3'-dimethyellagic acid-4-O-sulphate are polyphenols originating from jaboticaba and exhibit anti-inflammatory activities induced by CSE [88]. Extracts of Tussilago farfara have a high polyphenol content and excellent antioxidant and good antimicrobic properties [89]. CS-induced lung inflammation can be attenuated by using ethanol extract from T. farfara flower buds through mediating the NLRP3 inflammasome, NF-κB and Nrf2 pathways [90]. N-Acetylcysteine has shown the potential to improve the immune state of COPD patients and a COPD animal model by downregulating pro-inflammatory and inflammatory cytokines, including IL-1β, IFN-γ, TNF-α and IL-18, through modulating and suppressing the NLRP3 inflammasome pathways of macrophages [91]. Magnesium isoglycyrrhizinate (MgIG), glycyrrhizic acid, is an anti-inflammatory agent originally used for treating hepatitis. Animals treated with MgIG experienced reduced inflammatory cell infiltration and accumulation in broncho-alveolar lavage. Decreased IL-6 and TNF-α production was revealed in the serum of CS/LPS-induced COPD rats through the inhibition of NLRP3 pathways [80]. For protection against epithelial injury by CS exposure, corilagin can reduce CS-induced intercellular junction breakdown on epitheliums by inhibiting 4HNE-Cx40 adduct formation and NF-κB activation to diminish CS-induced epithelial changes [92]. Gallic acid (GA) is a natural phenolic composite with high free radical scavenging potential to act as an antioxidant and anti-inflammatory agent [93]. GA treatment can restrain elastase-induced neutrophil infiltration and elevate myeloperoxidase (MPO) activity. In addition, production of the pro-inflammatory cytokines IL-6, TNF-α and IL-1β is suppressed by phosphorylation of p65NF-κB and IκBα with the down-regulation of IL-1β/TNF-α/keratinocyte chemoattractant/macrophage inflammatory protein-2/granulocyte colony stimulating factor gene expression. GA also suppressed the influx of neutrophils and macrophages induced by CS and dampened the expression of TNF-α/MIP-2/KC genes. This could significantly restore the levels of glutathione reductase and catalase, as well as reduce xanthine oxidase activity in lung tissues to attenuate and modulate the tissue oxidative stress and inflammation induced by CS [94, 95]. Exposure to particulate matter (PM) from air pollution decreases Nrf2 expression; GA can restore antioxidant status through activating the Nrf2 pathway to prevent and attenuate lung injury from PM exposure [96]. In CS-induced acute and chronic lung injuries, supplements with pomegranate juice can significantly protect lung tissue from injuries and decrease the expression of inflammatory mediators, apoptosis and oxidative stress [97]. In CSE and porcine pancreatic elastase (PPE)-induced animal models, spray instillations with salvianolic acid B significantly improved the bioavailability about 30∼100 fold over oral administration. The anti-emphysema effects of salvianolic acid B can reduce CSE-induced apoptosis and lipid peroxidation, elevate phosphor-signal transducer and activator of transcription 3 and vascular endothelial growth factor (VEGF) expression, and stimulate lung cell proliferation to enable the reversal of alveolar structural destruction in vivo [98]. CSE may cause inflammation and increase the tissue oxidative stress that leads to decreased elastin formation of fibroblasts by inhibiting the enzyme activity of lysyl oxidase [99]. The bioavailability of isorhapontigenin by oral administration is high and it has good anti-inflammatory effects and pharmacokinetic properties [100]. In vitro cells treated with isorhapontigenin can effectively reduce the levels of ROS and display cytokine-suppressing effects through the inhibited phosphoinositide 3-kinase (PI3K)/protein kinase B (Akt) pathway, which is insensitive to corticosteroids [101]. Silymarin has been reported to attenuate the chronic inflammation and oxidative stress in a CS-induced COPD model and the anti-inflammatory effect of silymarin may thoroughly suppress the activity of the extracellular signal-regulated kinases (ERK)/MAPK pathway [102, 103]. Murine orally received isoflavones led to reduced CS-induced pro-inflammatory cytokine gene expression, decreased inflammatory cells in the bronchoalveolar lavage fluid (BALF) and shrunk pulmonary emphysema [104]. Curcumin is a naturally existing polyphenolic compound present in the rhizome of Curcuma longa belonging to the family Zingiberaceae. The biological activity of curcumin to regulate the function of multiple signal pathways or transductions is linked with the attenuation of acute and chronic diseases. Numerous pre-clinical and clinical studies have demonstrated that curcumin modulated several mediators in cell signal transduction pathways including Akt, AMPK, activator protein 1, β-catenin, enhancer binding protein α, ERK5, p38/MAPK, mammalian target of rapamycin (mTOR), MyD-88, NLRP3 inflammasome, Notch-1, Nrf2, P21 (RAC1) activated kinase 1, PI3K, peroxisome proliferator-activated receptor γ (PPARγ), Rac1, Shh, STAT3, TGF-β, Toll-like receptor (TLR)-4 and Wnt [105–107]. The use of adjuvants such as piperine, curcumin nanoparticles, liposomal curcumin and the curcumin phospholipid complex showed the enhanced bioavailability and therapeutic potential [105]. In vitro, curcumin treatment can activate the PPARγ pathway and inhibit the NF-κB pathway to attenuate CSE-induced cell death and inflammatory cytokine expression. In vivo CS exposure can down-regulate the PPARγ+ cell number in lung tissues; while curcumin treatment can restore it and hinder TNF-α and IL-6 levels of serum to mitigate CS-induced inflammation [106]. Oral curcumin administration can increase the antioxidant gene expression in alveolar macrophage and lung tissue, such as oxygenase-1, glutamate-cysteine ligase and glutathione reductase, increase MPO activity, and decrease the neutrophils and macrophages in BALF to decrease intratracheal PPE- and CS-induced pulmonary inflammation and emphysema in mice [107]. Effective anti-inflammation corticosteroids are widely used in clinical therapies for asthma and COPD treatment. The suppression of pro-inflammatory gene expression by corticosteroids is based on the recruitment of the transcriptional co-repressor HDAC2 into an activated GR complex [108]. Chronic oxidative stress in COPD lungs is impeded HDAC2 activity [109], which blocks steroid efficacy or causes desensitisation [110]. Curcumin can restore and promote HDAC2 expression and activity, which may reverse the steroid insensitivity of asthma or CS-induced COPD [111]. In ALI induced by a serious infection, e.g. sepsis, curcumin treatment can hinder NF-κB, NLRP3 and pyroptosis-related protein expressions and increase sirtuin 1 (SIRT1) expression to demonstrate the protective effects of curcumin on septic ALI through up-regulation of SIRT1 to modulate the NLRP3 inflammasome pathway [112]. The major limiting factor of curcumin medical applications is its low bioavailability. However, there is a new treatment formulation using dry powder inhalation of curcumin for COPD that can provide good solubility and dissolution [113]. There are several classes of polyphenols present in apple juice, including flavanols, dihydrochalcones, flavan-3-ols and phenolic acids. Apple phenolics function as strong anti-inflammatory agents by shielding tissues from inflammatory injury [114]. Animals that received the oral administrate of apple polyphenols had significantly decreased oxidative stress and expression of pro-inflammatory factors in the lungs, reduced CS-induced accumulation of inflammatory cells in airways and lung, which might be partially reduced by inhibiting the P38/MAPK signalling pathway [115]. Increasing studies have illustrated the favourable therapeutic effects of resveratrol against lung diseases by preventing ageing, inflammation, oxidative stress, fibrosis and cancer in vitro and in vivo. Treatment with resveratrol can hinder inflammatory cytokine IL-1β and stimulated anti-inflammatory cytokine IL-8 release from alveolar macrophages in COPD [116]. Additionally, resveratrol treatment attenuates CSE-mediated glutathione (GSH) depletion in alveolar epithelial cells by promoting GSH synthesis and Nrf2 activity [117]. Corticosteroids are inefficient for reducing IL-8 and granulocyte-macrophage colony-stimulating factor (GM-CSF) release from alveolar macrophages, and inflammatory mediators released from the airways smooth the muscle cells of COPD patients. Resveratrol treatment can reduce the release of IL-8 and GM-CSF and restore VEGF release from human airway smooth muscle cells in COPD [118]. The combinative treatment of resveratrol and dexamethasone can significantly reduce all inflammatory parameters [119]. Resveratrol treatment can upregulate SIRT1 and proliferator-activated receptor-γ coactivator-1α (PGC-1α) expression to decrease the level of malondialdehyde and increase activity of SOD [71, 120]. In an in vivo CS-induced or LPS treatment COPD model, oxidative stress and inflammation caused injury to respiratory and cardiovascular systems. Treatment with resveratrol could reduce cardiac oxidative damage, prevent left ventricular remodelling and restore the expression of SIRT1 in the hearts of aged rats with emphysema [121] and may have a protective effect against CSE-induced apoptosis of human bronchial epithelial cells [122]. In an elastase-induced pulmonary emphysema COPD model, the expression of VEGFA, Nrf2, manganese SOD and airspace volume were significantly augmented in the lung tissues of the resveratrol treatment group [123]. In a combined LPS/CE-induced COPD model, resveratrol could significantly decrease the inflammatory cells in BALF and the level of inflammatory-IL-17, IL-6, IL-8, TNF-α and TGF-β cytokines in the lungs [124]. In CS exposure-induced chronic bronchitis, naringin treatment can significantly reduce the concentrations of IL-8, leukotriene B4 and TNF-α in BALF and improve SOD activity in lung tissues [125]. Naringin is a flavanone glycoside found in citrus fruits and grapefruit that is characterised by the following effects: antioxidation, anti-peroxidation and anti-inflammation [126]. Naringin treatment can significantly reduce ovalbumin-induced coughs and airway hyper-responsiveness and inhibit inflammatory cell infiltration and IL-4, IL-5 and IL-13 in BALF [127]. Naringin can restore the IL-10 level, prevent CS-induced neutrophil infiltration and decrease the activation of MPO and MMP-9 with suppressed inflammatory cytokine release, such as TNF-α and IL-8, in a CS-exposed animal model [128]. Liquiritin apioside is a flavonoid from Glycyrrhiza uralensis that can protect lung epithelial cells from CS-induced injury by increasing the levels of anti-oxidative GSH and obstructing the expressions of TGF-β and TNF-α [129]. Additionally, some flavonoids and polyphenols have been found to perform or moderate the AIM2 inflammasome pathway. Quercetin treatment was found to impede NLRP3 inflammasome and the expression of pro-caspase-1 as well as suppressing the expression of AIM2 in a dose-dependent fashion in in vitro and in vivo investigations [130, 131]. Treatment with EGCG failed to decrease AIM2 expression, but it could down-regulate the expression of active caspase-1 and IL-1β in epidermal keratinocytes [132]. Several flavonoids have been demonstrated to have multiple inhibiting effects on different inflammasomes. Glycyrrhizin, one of the triterpenoids, has hindering effects on both NLRP3 and AIM2 inflammasomes to decrease the formation of active-form caspase-1 and IL-1β release [133]. Isorhamnetin can impede AIM2 and NLRP3 inflammasome to decrease several pro-inflammatory cytokine expressions, such as IL-1α, β, IL-18, TGF-β and hyperoside; it suppresses the activation of AIM2 and NLRC4 inflammasome but does not affect cytokine expression [134]. In vitro, THP-1 cells treated with apigenin significantly inhibit NLRP3 and AIM2 inflammation to reduce caspase-1 and IL-1β production [135]. The activation of AIM2 inflammasome expression in A549 and H460 cells could be suppressed by luteolin in a concentration-dependent manner [136]. Some flavonoids can serve as activators of inflammasome, i.e. icariside I and bavachin. Icariside I can promote and regulate bone remodelling and is considered to be a candidate compound for osteoporosis treatment [137]. Bavachin can increase the mRNA levels of oestrogen-responsive genes and serve as the activator of oestrogen receptor [138]. A recent study showed that bavachin and icariside I could selectively exacerbate NLRP3 inflammasome but not NLRC4 or AIM2 to promote the expression and activity of caspase-1, TNF-α and IL-1β and that may cause idiosyncratic hepatocyte toxicity [138, 139].
Hydrogen sulphide played a protective role in PM-induced mice emphysema and airway inflammation by hampering the formation of NLRP3 inflammasome and apoptosis via the Nrf2-dependent pathway [140]. Recently, low-molecular-weight anti-inflammatory agents, including antioxidants, inflammasome inhibitors, kinase inhibitors, modulators of inflammatory mediators, phosphodiesterase-4 inhibitors, protease inhibitors, and other agents have shed light on the development of COPD treatment. The molecular docking results of low-molecular-weight agents and targeted proteins provide new insights for targeted COPD treatments, particularly for small-molecule agents [141]. Melatonin showed protection against COPD in vitro and in vivo tests. It activated the intracellular antioxidant thioredoxin-1 (thereby inhibiting the thioredoxin-interacting protein–NLRP3 pathway) and suppressing the impaired mitophagy-mediated inflammasome activation (upregulating PTEN-induced kinase 1, Parkin and LC3B-II expression). Melatonin also enhanced the overall anti-oxidative status of a COPD lung via Nrf2–haem oxygenase-1 axis restoration [142]. Melatonin attenuated airway inflammation via SIRT1-dependent inhibition of NLRP3 inflammasome and IL-1β in rats challenged with CS and LPS [143]. MCC950 is a potent and selective inhibitor of NLRP3 inflammasome and reduced the mRNA expressions of IL-1β, IL-8, TGF-β and MMP-9. At 24 h after LPS instillation, MCC950 could also reduce the protein levels of IL-1β, IL-18 and caspase-1 in mice [144]. Lipoxin receptor agonist (BML-111) may prevent the activation of NLRP3 inflammasome and hinder ROS production via upregulation of Nrf2 in COPD model mice [145]. Forskolin and isoforskolin (ISOF) are polyphenols isolated from Coleus forskohlii and have been reported to have anti-inflammatory and anti-oxidative effects [146]. Forskolin inhibited the activation of NLRP3 inflammasome and decreased the secretions of IL-1β and IL-18 from macrophages [147, 148]. ISOF alleviated the acute exacerbation of COPD by prompting pulmonary function and mitigating inflammation via downregulation of pro-inflammatory cytokines (TNF-α, IL-1β, IL-6), monocyte chemoattractant protein-1, monokine induced by interferon-γ, IP-10, C-reactive protein, Th17/IL-17A and NF-κB/NLRP3 pathways in CS-induced mice [149]. ISOF also tempered IL-1β and IL-18 secretion by NLRP3 inflammasome activation in human macrophages [150]. Treatment with ISOF can downregulate the mTOR level in lung tissues and increase intracellular cyclic adenosine monophosphate levels to relax the histamine-induced contraction of the lungs and trachea smooth muscles. Pre-treatment with ISOF significantly ameliorated the pathological damage to lung tissue and improved pulmonary function in COPD and airway hyper-responsive rats [149, 151]. The differential impact of macrolides-azithromycin on the inflammasome–IL-1β axis may be of consequence in inflammasome-driven diseases, e.g. COPD and asthma [152]. Histidine repressed the activation of NLRP3 inflammasome in both PPE- and LPS-induced COPD mouse models and in vitro mouse alveolar macrophage MH-S cells [153]. Coixol inhibited MAPKs, NF-κB pathways and the activation of NLRP3 inflammasome in LPS-induced RAW 264.7 cells [154]. After long-term PM-exposure-induced lung inflammation and fibrosis, resveratrol intervention improved these adverse effects by preventing autophagy-related NLRP3 inflammasome activation [155]. CS-challenged mice were treated with flavonoids isolated from loquat (Eriobotrya japonica) leaves, the results revealed that flavonoids have a protective effect and putative mechanism of the action of total flavonoids resulted in the inhibition of inflammation and oxidative stress through the regulation of transient receptor potential vanilloid 1 and related signal pathways in lung tissues [156]. Table 2 lists the mediating roles of polyphenols and flavonoids on the amelioration of CS-induced NLRP3 inflammasome in COPD. Additionally, figure 2 illustrates the acting target molecules of the polyphenols and flavonoids on the signal pathways of CS-induced NLRP3 inflammasome. From this diagram, it is obvious that this has been separated into three approaches: the first, CS/CSE, causes the involvement of CS-induced inflammasome signal pathways (shown in black boxes) and the second is the attenuating mediators of regulating signals (shown in red boxes). Thirdly, the acting targets of polyphenols and flavonoids are displayed in green boxes (figure 2). Moreover, quercetogetin suppressed mitophagy-dependent apoptosis by inhibiting the expression of cleaved caspase-3, -8 and -9, and downregulating caspase activity in human lung bronchial epithelial cells (BEAS-2B cells) exposed to CSE [157].
Inflammation plays a vital role in the development of COPD. Pyroptosis, an inflammatory form of cell death, may be implicated in the pathogenesis of COPD. The NLRP3 inflammasome enhances inflammatory cell recruitment and mediates immune responses in the lungs. Further, this is involved in numerous diseases characterised by induced lung disease by focusing on pathways causing chronic respiratory epithelial cell injury, cell death, alveolar destruction and tissue remodelling affiliated with the progress of ALI and COPD. The experimental findings may afford mechanistic insights into the immunosuppression in smokers via the CS-activated ROS–NLRP3 axis and induced epithelial cells pyroptosis. According to the abovementioned section, polyphenols and flavonoids may have therapeutic potential in CS-COPD and present exclusive opportunities to develop an approach or strategy to modulate immune functionality via mediating NLRP3 inflammasome. Corticosteroids are the most-used anti-inflammatory agent for asthma and COPD. However, people with severe asthma or COPD show poor response to the anti-inflammatory effects of corticosteroids. The Corticosteroid resistance is a main therapeutic challenge to the management of COPD. Prominently, resveratrol, icariin and quercetin could restore the sensitivity of corticosteroid, and, in combination with corticosteroids, have the potential to form novel treatments for COPD. A panel of clinically effective drugs has displayed potential in restoring steroid- resistance in experimental models. It is highly plausible that some of these phytocompounds or molecules can be efficaciously repositioned for clinical use in the management of COPD. One clinical trial has reported that 21 COPD patients (forced expiratory volume in 1 s (FEV1) 53±15% predicted; age 67±9 years; body mass index 24.5±3.3 kg·m−2) received resveratrol (150 mg·day−1) or placebo for 4 weeks, the results failed to ratify the previously designated positive effects of resveratrol on the mitochondrial function of skeletal muscle in patients with COPD [158]. In mild-to-severe lung disease, COPD patients with FEV1 ranging between >35% and <80% were recruited and supplemented with either quercetin (500∼2000 mg·day−1) or placebo for 1 week. The patients showed no drug-related severe adverse manifestations based on blood tests, which contained complete blood counts and an evaluation of a comprehensive metabolic panel [159].
Remarkably, underlying mechanisms and some possible causes have led to inconsistent results in observational studies and supplementation trials because investigations in this field are mostly limited to animal models or small clinical trials. More forthcoming cohorts and well-designed clinical trials are needed to support the introduction of individualised phytochemical interventions into health policy.
Recently, NLRP3 inflammasome has been confirmed to play a significant role in the pathogenesis of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and is involved in the cytokine storm formation of ALI associated with coronavirus disease 2019 (COVID-19). Up to date, there are more than 1300 scientific articles that have looked into the link between COPD and COVID-19, and more than 280 reports have explored the inflammasome mechanisms in the pathogenesis of COVID-19. Notably, there is accumulating evidence that the inducers, components, pathways and effectors of inflammasome signal pathways are attributed to the pathogenesis in COPD and SARS-CoV-2 infection [160, 161]. On the other hand, SARS-CoV-2 infection may cause more severity in COPD patients [162]. A clinical report found that curcumin can mediate and down-regulate the NLRP3 inflammasome relative intracellular signal transduction pathways that are involved in inflammation of COVID-19 patients [163]. This is an electrifying outcome for combating COVID-19 and creates a new hope for promoting drug development from the medicinal perspective of curcumin. |
|
PMC9648513 | 36268582 | Noëlle K J Bittner,Katya L Mack,Michael W Nachman | Shared Patterns of Gene Expression and Protein Evolution Associated with Adaptation to Desert Environments in Rodents | 21-10-2022 | adaptation,desert,RNA-Seq,genomics,rodents | Abstract Desert specialization has arisen multiple times across rodents and is often associated with a suite of convergent phenotypes, including modification of the kidneys to mitigate water loss. However, the extent to which phenotypic convergence in desert rodents is mirrored at the molecular level is unknown. Here, we sequenced kidney mRNA and assembled transcriptomes for three pairs of rodent species to search for shared differences in gene expression and amino acid sequence associated with adaptation to deserts. We conducted phylogenetically independent comparisons between a desert specialist and a non-desert relative in three families representing ∼70 million years of evolution. Overall, patterns of gene expression faithfully recapitulated the phylogeny of these six taxa providing a strong evolutionary signal in levels of mRNA abundance. We also found that 8.6% of all genes showed shared patterns of expression divergence between desert and non-desert taxa, much of which likely reflects convergent evolution, and representing more than expected by chance under a model of independent gene evolution. In addition to these shared changes, we observed many species-pair-specific changes in gene expression indicating that instances of adaptation to deserts include a combination of unique and shared changes. Patterns of protein evolution revealed a small number of genes showing evidence of positive selection, the majority of which did not show shared changes in gene expression. Overall, our results suggest that convergent changes in gene regulation play an important role in the complex trait of desert adaptation in rodents. | Shared Patterns of Gene Expression and Protein Evolution Associated with Adaptation to Desert Environments in Rodents
Desert specialization has arisen multiple times across rodents and is often associated with a suite of convergent phenotypes, including modification of the kidneys to mitigate water loss. However, the extent to which phenotypic convergence in desert rodents is mirrored at the molecular level is unknown. Here, we sequenced kidney mRNA and assembled transcriptomes for three pairs of rodent species to search for shared differences in gene expression and amino acid sequence associated with adaptation to deserts. We conducted phylogenetically independent comparisons between a desert specialist and a non-desert relative in three families representing ∼70 million years of evolution. Overall, patterns of gene expression faithfully recapitulated the phylogeny of these six taxa providing a strong evolutionary signal in levels of mRNA abundance. We also found that 8.6% of all genes showed shared patterns of expression divergence between desert and non-desert taxa, much of which likely reflects convergent evolution, and representing more than expected by chance under a model of independent gene evolution. In addition to these shared changes, we observed many species-pair-specific changes in gene expression indicating that instances of adaptation to deserts include a combination of unique and shared changes. Patterns of protein evolution revealed a small number of genes showing evidence of positive selection, the majority of which did not show shared changes in gene expression. Overall, our results suggest that convergent changes in gene regulation play an important role in the complex trait of desert adaptation in rodents.
Finding genes underlying the convergent basis of evolution to extreme environments allows us to understand both how organisms have adapted to these environments and what paths are accessible to evolution. While most previous research has been limited to individual species, this study expands the scope by comparing RNA-Seq data from multiple rodent families to identify changes at the gene expression and protein level associated with adaptation to desert environments.
The repeatability of adaptive evolution at the molecular level remains an open question. In situations where the mutational target is small and constraints exist due to epistasis or pleiotropy, the molecular paths available to adaptation may be highly limited (Weinreich et al. 2006; Karageorgi et al. 2019). Indeed, there are a number of excellent examples of convergent molecular evolution underlying simple traits (e.g., Stewart and Wilson 1987; Mundy 2005; Zhen et al. 2012). For highly polygenic traits, however, convergence may be less expected simply because the mutational target is large and multiple paths may be available on which selection can act. Nonetheless, several studies have found evidence for convergence at the molecular level even for complex traits (e.g., Marcovitz et al. 2019; Sackton et al. 2019). Convergent phenotypic evolution may be due to changes in gene regulation, to changes in protein structure, or both, yet these processes are rarely studied together in the context of complex adaptive traits (but see Hao et al. 2019). There is evidence that gene expression divergence and amino acid sequence divergence are correlated between paralogs following gene duplications (Gu et al. 2002; Makova and Li 2003), and more generally that rates of gene expression and rates of protein evolution are coupled in some lineages (e.g., Nuzhdin et al. 2004; Lemos et al. 2005). These observations raise the possibility that changes in both gene expression and protein sequence may contribute to the repeated evolution of complex adaptive traits. Adaptation to desert environments in rodents provides an opportunity to study repeated evolution in both gene expression and protein sequence for a complex trait. Desert ecosystems present the challenge of extreme aridity and low or seasonally absent water, yet multiple lineages of rodents have independently evolved the ability to survive in these unusually harsh environments (reviewed in Degen 1997). Rodents have solved these challenges in myriad ways, including dietary specialization on plants that are high in water content or modifications to reduce evaporative water loss (Schmidt-Nielsen and Schmidt-Nielsen 1952; Schmidt-Nielsen 1964; Degen 1997). A common feature of most desert rodents is a modified kidney capable of producing highly concentrated urine (MacMillen and Lee 1967; Beuchat 1990; Al-kahtani et al. 2004; Donald and Pannabecker 2015). The production of hyperosmotic urine has evolved independently multiple times across rodents, and maximum urine concentration has been found to be correlated with environmental aridity in mammals (Rocha, Brito, et al. 2021). Final excreted urine concentration depends on the development and maintenance of a corticomedullary osmotic gradient within the kidney. Studies have shown that many aspects of kidney morphology and physiology have been modified in different lineages to produce hyperosmotic urine (Bankir and de Rouffignac 1985; Donald and Pannabecker 2015). The genetic basis of desert adaptation has been studied independently in a handful of species, and individual genes and pathways which may underlie this adaptive phenotype have been identified (Marra et al. 2012, 2014; Wu et al. 2014; MacManes 2017; Giorello et al. 2018; Tigano et al. 2020; reviewed in Rocha, Godinho, et al. 2021). Here, we study three phylogenetically independent lineages of rodents that have all converged on a common phenotype, ultra-high urine concentration associated with desert living, to identify shared molecular changes associated with habitat type. We compared gene expression and protein sequence divergence in the kidney, the organ responsible for sodium and water homeostasis, between a desert and a non-desert species in three comparisons representing transitions to desert living in three different rodent families (Heteromyidae, Dipodidae, and Muridae). Desert species were chosen based on their high urine concentration, a proxy for increased osmoregulatory capacity (fig. 1). Within Muridae, we compared the Australian Spinifex Hopping Mouse, Notomys alexis, the mammal with the highest known urine concentration and well studied for its modifications to desert life (MacMillen and Lee 1967; Macmillen and Lee 1969; Baudinette 1972; Donald et al. 2012), to the house mouse (Mus musculus), a widespread generalist. Within Dipodidae, we compared the desert-dwelling Lesser Egyptian Jerboa, Jaculus jaculus, previously studied for its kidney modifications associated with granivorous desert living (Schmidt-Nielsen and Schmidt-Nielsen 1952; Khalil and Tawfic 1963), to the Western Jumping Mouse, Zapus princeps, a North American species found in riparian environments. Within Heteromyidae, we compared the Rock Pocket Mouse, Chaetodipus intermedius (Bradley et al. 1975; Altschuler et al. 1979), native to the North American Sonoran desert, to the Desmarest's Spiny Pocket Mouse, Heteromys desmarestianus, a neotropical species found in mesic areas that cannot survive without free water (Fleming 1977). We sequenced kidney mRNA from these pairs of taxa and assembled and annotated de novo transcriptomes for three desert-mesic species pairs spanning ∼70 million years of evolution. Assembled transcriptomes were used to analyze rates of evolution in single-copy orthologs to identify genes putatively under selection across desert lineages. We also performed mRNA sequencing on multiple individuals within each species to study gene expression divergence between desert and non-desert species. Global patterns of gene expression recapitulated the phylogeny of these six species. However, we also discovered a significantly greater number of shared changes in gene expression than expected by chance between desert and non-desert species. In contrast, shared changes in amino acid sequence were identified in a smaller proportion of genes.
We generated on average ∼123 million reads per sample for the assembly of de novo kidney transcriptomes in each species. For M. musculus, five smaller libraries were concatenated for assembly. After read correction, quality filtering, and adapter trimming, each library had an average of ∼103 million reads which were used for the assembly. Each assembly contained 965,227 transcripts on average. We reduced the number of redundant transcripts in the assembly to improve accuracy of downstream analyses by clustering similar transcripts together using CD-HIT-EST. This decreased the number of transcripts by ∼20% per sample to an average of 793,887 transcripts (supplementary table S2, Supplementary Material online). We used Benchmarking Universal Single-Copy Orthologs (BUSCO) to check assembly completeness to determine how many of the 6,192 orthologs found in the Euarchontoglires odb9 were present in our assembled transcriptomes. The six assemblies ranged in completeness from 80–87% (supplementary fig. S1, Supplementary Material online). This level of completeness reflects a single tissue (kidney) taken at one developmental time point. After open reading frame (ORF) prediction, we annotated each transcript to known M. musculus proteins. We were able to assign transcripts to 395,029 putative ortholog groups.
To identify patterns of differential gene expression, we sequenced kidney mRNA from additional individuals in each of the six species for an average of ∼27 million reads per individual. We retrieved 13,305 genes in C. intermedius, 11,749 genes in H. desmarestianus, 14,891 genes in J. jaculus, 14,380 genes in Z. princeps, 18,622 genes in N. alexis, and 19,913 genes in M. musculus for which we were able to quantify expression levels. These genes were annotated using M. musculus transcripts so, as expected, the number of genes we were able to annotate in more divergent species is more limited. Gene expression profiles largely recapitulated the known phylogenetic relationships of these six species (fig. 2A). Individuals within each species form well-defined clusters (apart from a single H. desmarestianus individual), and the different genera within each family share expression profiles that are more similar to each other than they are to genera in different families. Further, Muridae and Dipodidae are more similar to each other in expression profiles than either is to Heteromyidae, reflecting the known evolutionary relationships of these families. Thus, the overall expression patterns reflect evolutionary history more than habitat type. These patterns are also seen in a principal component analysis (PCA) based on expression-level co-variance (fig. 2B), where PC1 (accounting for 33% of the variance) largely reflects phylogeny. Despite the overall phylogenetic pattern of gene expression, consistent differences in expression were seen between desert and non-desert species within each family. In particular, PC4 captures this variation, separating desert from non-desert taxa (explaining 11% of the variation; fig. 2C).
We quantified differential expression (DE) between desert and non-desert species within each family. In pairwise contrasts between desert and non-desert species in Heteromyidae, Dipodidae, and Muridae, we identified >4,000 genes in each comparison with evidence of significant DE (supplementary table S3, Supplementary Material online, FDR < 0.01). DE between desert and non-desert species in each of the three families was associated with several GO categories, including cellular metabolic processes and nitrogen metabolic processes (supplementary table S4, Supplementary Material online). We identified a total of 654 genes that showed significant DE in all three species pairs (supplementary fig. S2, Supplementary Material online), with 145 of these genes showing shifts in the same direction in each comparison. To identify shared shifts in gene expression associated with desert living, we also modeled gene expression as a function of species pair (i.e., family), habitat, and their interaction. Shared changes were identified as those for which there was a significant effect of habitat (FDR < 0.01), but no interaction between species pair and habitat (FDR > 0.05; see Materials and Methods; Parker et al. 2019). We identified 702 genes with shared shifts in desert rodents relative to the mesic comparison (fig. 3A). This set includes all the 145 genes identified above in pairwise tests. Thus, 8.6% (702/8,174) of genes showed shared shifts in expression in desert rodents compared with their non-desert relatives. Shared shifts in gene expression can be a consequence of selection in response to shared environmental pressures or stochastic processes. These shared environmental pressures are not limited to aridity but may extend to other differences in habitat type, diet, condition, and more. To ask if the observed number of genes with shared DE was more than expected by chance, we performed a permutation test in which we took each gene and randomly switched habitat assignment within species pairs, while always maintaining the same label for all biological replicates within a species, to create 10,000 permuted data sets (fig. 3B; see Materials and Methods; Parker et al. 2019). Permuted data sets never identified more genes than the observed set of shared genes, suggesting an enrichment of shared DE associated with habitat type under the simplifying assumption that expression changes at genes are independent (see Discussion). Shared changes in gene expression could be due to convergent evolution or to a similar plastic response in similar xeric environments. To disentangle these, we analyzed RNA-seq data from a previous study in which M. musculus from a relatively mesic environment in Canada were subjected to three days of water deprivation in the laboratory (Bittner et al. 2021). We compared the identity of genes showing a plastic response in that study to the identity of genes showing an evolved response between N. alexis and M. musculus in the present study. In the former study, 1,354 genes were differentially expressed between control and treatment (water deprivation) groups, while in the present study, 8,008 genes were differentially expressed between N. alexis and M. musculus. The overlap between these sets was 547 genes, and this overlap is not more than expected by chance (hypergeometric test, P = 0.99). This suggests that the vast majority of differentially expressed genes (7461/8008 = 93%) between N. alexis and M. musculus reflect evolved differences rather than plastic responses to xeric conditions. We also compared the 1,354 genes showing a plastic response in Bittner et al. (2021) with the 145 genes in this study in which all three species pairs showed expression changes in the same direction. The overlap was nine genes, and this is not more than expected by chance (hypergeometric test, P = 0.99). Thus, the majority of genes showing shared shifts in gene expression in this study ([145 − 9]/145 = 94%) do not show a plastic response to water stress in M. musculus under laboratory conditions. From these analyses, we argue that most shared changes in gene expression reflect convergent evolution rather than phenotypic plasticity. The magnitude of expression differences between individual desert-mesic pairs was often modest in one or more contrasts between species pairs (supplementary fig. S3, Supplementary Material online); only 208 genes with shared expression shifts showed an average of greater >0.5 log2-fold change difference between each desert-mesic species pair. The number of genes showing higher expression in desert rodents compared with non-desert relatives (335 genes, shown in blue in fig. 3A) was slightly fewer than the number of genes showing lower expression in desert rodents compared with non-desert relatives (367 genes, shown in red in fig. 3A). Additionally, across all genes, fold changes between individual desert-mesic species were found to be significantly correlated in two of the three comparisons of species pairs (Spearman's rank correlation rho, C. intermedius/H. desmarestianus vs. J. jaculus/Z. princeps, P = 0.0062, rho = 0.03; C. intermedius/H. desmarestianus vs. N. alexis/M. musculus, P < 2.2e-16, rho = 0.10; N. alexis/M. musculus vs. J. jaculus/Z. Princeps, P = 0.12, rho = −0.017). To identify genes and pathways of interest, we divided the shared set of differentially expressed genes into those that are upregulated with respect to the desert taxa in all comparisons and those that are downregulated with respect to the desert taxa in all comparisons and performed phenotype and gene ontology (GO) term enrichment tests on these (see Materials and Methods). Genes upregulated across desert rodents were enriched for several GO terms related to gene regulation, including regulation of RNA metabolic process (q = 2.55 × 10−5), regulation of gene expression (q = 1.34 x 10−5), and regulation of RNA biosynthetic process (q = 3.87 × 10−5). Genes downregulated in desert rodents were enriched for GO terms related to metabolic processes, including metabolic process (q = 1.56 × 10−3), organic substance metabolic process (q = 3.93 × 10−3), and cellular metabolic process (q = 3.54 × 10−3). Genes with shared patterns of expression included those with mouse mutant phenotypes related to kidney development and physiology or homeostasis (supplementary table S5, Supplementary Material online). For example, Aquaporin 11 (Aqp11) is expressed at a lower level in all desert species compared with non-desert species in all three comparisons (fig. 3C). This gene is part of a family of genes encoding membrane-integrated channels responsible for water transfer across membranes throughout the body. This set also includes genes associated with human phenotypes related to kidney and renal diseases (supplementary table S6, Supplementary Material online); for example, mutations in the gene Col4a5, which is downregulated in desert species, have been associated with Alport syndrome, a disease characterized by kidney inflammation (Köhler et al. 2019).
Next, we tested for evidence of selection on protein-coding sequences using well-aligned single-copy orthologs found in all desert-mesic species pairs. We performed three analyses using the 1,474 single-copy orthologs with high-quality alignments present in all species. First, we searched for genes showing signatures of selection shared by all desert species by calculating the ratio of non-synonymous substitutions per non-synonymous site to synonymous substitutions per synonymous site (ω) at each locus to estimate rates of evolution along each branch implemented using a branch model for codeml in PAML (Yang, 1998, 2007). When testing for evidence of selection shared only by desert species compared with non-desert species, we uncovered 39 genes (39/1,474 = 2.6%) for which ω was greater in all three desert lineages (supplementary table S7, Supplementary Material online, FDR < 0.1). This group of genes is enriched for phenotypes related to multiple aspects of the immune response as well as to hearing/vestibular/ear phenotypes and other aspects of osteology (supplementary table S8, Supplementary Material online). Immune genes are some of the fastest evolving genes in the genome and are disproportionately found to be under selection in many studies (Hurst and Smith 1999; Schlenke and Begun 2003; Nielsen et al. 2005). One gene of particular interest is FAT atypical cadherin 4 (FAT4) (q = 0.018). FAT4 has been implicated in human kidney diseases (Alders et al. 2014) and is involved in normal kidney development through modulating the RET signaling pathway in mouse models (Mao et al. 2015; Zhang et al. 2019). FAT4 homozygous knockout mice have smaller kidneys with cysts in renal tubules when compared with wild-type mice and die within a few hours of birth (Saburi et al. 2008). These phenotypes in laboratory mice make this an interesting candidate gene for future studies in desert rodents. We found three genes that showed evidence of positive selection and also showed shared shifts in gene expression (Rows 1–3 in supplementary table S9, Supplementary Material online); however, they are not known to be associated with phenotypes of interest. This amount of overlap is no more than expected by chance (hypergeometric test, P = 0.64). We then tested for lineage-specific selection by allowing two values of ω on the tree. We calculated ω for each of the three desert lineages individually and compared this to the value of ω for the five remaining taxa for each ortholog. This analysis was implemented as above but restricts the comparison to one species instead of all three. We identified 23 genes in C. intermedius, 19 in J. jaculus, and 18 in N. alexis where ω was significantly elevated (at FDR <0.1; supplementary table S10, Supplementary Material online) compared with all other taxa. These genes are candidates for underlying lineage-specific adaptations. In C. intermedius, enriched phenotypes were related to immunity and morphological traits including kidney size, while in J. jaculus and N. alexis, enriched phenotypes were related to behavioral and electrophysiological traits (supplementary table S11, Supplementary Material online). In the C. intermedius comparison, Dusp4 is of some interest as it has been associated with aberrant circulating solute levels in mouse models. Deletion of this gene has been associated with increased excreted protein and altered kidney structure in diabetic mice (Denhez et al. 2019). It was also found in the set of genes showing shared DE. Overall, the amount of overlap (hypergeometric test, P > 0.06 in all comparisons) between any of these lists and differentially expressed genes between lineage pairs is no more than expected by chance (supplementary table S9, Supplementary Material online). In the third analysis, we employed a branch-site model to identify genes in which specific codons may be under positive selection. In this approach, genes for which specific codons have a ω > 1 in the “foreground” branches (defined to include all three desert species) compared with the “background” branches are identified. Seven genes were identified (supplementary table S12, Supplementary Material online) with codons under selection in all three desert lineages, including Coro2b, a gene implicated in abnormal renal glomerulus morphology (Schwarz et al. 2019) and urine protein level (Rogg et al. 2017) and Bloc1s4, which is implicated in abnormal renal physiology (Gwynn et al. 2000). Again, there was no significant overlap with the genes identified in the DE analysis (P = 0.47; supplementary table S9, Supplementary Material online).
The molecular basis of convergent evolution has been well studied for a number of simple traits but has been less studied for complex traits. Even fewer studies have compared convergence in both gene expression and protein evolution for complex traits. Here, we studied shared changes in gene expression and amino acid sequence in three species of desert rodents and their non-desert relatives from across the rodent tree, representing ∼70 million years of evolution to identify candidate genes for convergent evolution underlying this complex adaptive trait. Despite the long evolutionary timeframe and the fact that most expression patterns tracked phylogeny (fig. 2), we identified hundreds of genes (702/8,174 = 8.6%) that showed shared shifts in gene expression (fig. 3A). This number is more than expected by chance under a model of independent gene evolution (fig. 3B). It is important to recognize that the number of genes showing shared expression changes does not reflect the number of causative changes (i.e., mutational events in evolution), as many of these shared changes in expression might reflect downstream consequences of a smaller number of changes at upstream regulators that govern networks of co-regulated genes. Nonetheless, the large number of shared changes in expression suggests that a measurable amount of desert adaptation is mediated by a large set of shared changes in gene regulation, whether at the level of individual genes or through sets of co-regulated genes. Future studies aimed at understanding the precise mechanisms mediating expression levels in each lineage would help quantify the degree to which these patterns are the result of cis-regulatory changes at the same locus or shared upstream trans-regulators that mediate shifts in expression across sets of co-regulated genes. Our analysis identified specific genes of interest known to be involved in osmoregulation and kidney function. Of note, Apq11 was found to be expressed at lower levels in all desert taxa compared with their non-desert counterparts. Aquaporins have been repeatedly implicated in studies of desert adaptation across rodents (Marra et al. 2012, 2014; Pannabecker 2015; Giorello et al. 2018). Mouse knockouts have demonstrated that Aqp11 is necessary for proximal tubular function and the formation of healthy kidneys (Morishita et al. 2005; Tchekneva et al. 2008). In addition, Aqp11 plays a role in salivary gland development (Larsen et al. 2010). This paper provides further evidence that aquaporins are a common evolutionary target in desert adaptation. In addition to these shared changes in gene expression, we identified a large number of species-specific changes in each species pair. Perhaps not surprising given the long evolutionary timescales and complexity of osmoregulatory function, much of the evolutionary response appears to be specific to individual lineages. In principle, phenotypic plasticity could also mediate some of the observed expression differences as most of the animals were caught in the wild and therefore experienced different environments, presumably with differences in hydration status. However, several observations suggest that evolved changes, rather than plastic changes, underlie most of the observed expression differences. First, comparison of the plastic expression response of M. musculus when deprived of water (Bittner et al. 2021) to the expression differences observed between M. musculus and N. alexis (this paper) revealed little gene overlap, suggesting that >93% of the differences seen between M. musculus and N. alexis reflect evolved changes. Second, one of the desert-adapted species (J. jaculus) was from a laboratory colony in which animals were reared with free access to fresh vegetables. Comparisons between J. jaculus and Z. princeps identified many differentially expressed genes, including many that were shared among the other comparisons, indicating that most expression divergence does not reflect expression plasticity due to differences in recent water availability. Third, apple was offered to most of the wild-caught species for a period before they were sacrificed, mitigating the effects of short-term water stress. Fourth, many of the desert-adapted species will not accept free water if offered. Finally, we note that the overall patterns of expression (fig. 2) recapitulate the known phylogeny of these species (fig. 1), consistent with the idea that most expression variation is due to evolved differences. While not discounting that some of the observed differences in expression may reflect phenotypic plasticity, these observations strongly support the inference that most shared changes are due to convergent evolution. In contrast to the fairly long evolutionary timescales in the present study, we documented differences in kidney gene expression between desert and non-desert populations of M. musculus separated by only a few hundred generations of evolution (Bittner et al. 2021). In that study, we identified 3,935 differentially expressed genes of which 99 were found in the list of genes with shared DE patterns in all three desert lineages in the present study. The lack of significant overlap (hypergeometric test, P = 0.99) suggests that over long evolutionary timescales, adaptive responses to xeric conditions may be different from the evolved changes in gene expression over short evolutionary timescales. In contrast to the number of shared changes in gene expression, we observed few genes that showed evidence of positive selection on amino acid sequences (39/1,474 = 2.6%) among desert species. These proportions are not directly comparable, as the methods used to detect shared expression changes and positive selection are quite different. Nonetheless, our analyses suggest that the phenotypic convergence seen in urine concentration is reflected at the molecular level more in patterns of gene regulation than in patterns of protein evolution. In this study, we took advantage of RNA-seq data to also study protein evolution in non-model organisms whose genomes have not yet been sequenced. However, this approach has limitations. By restricting our analyses to single-copy orthologs found in all transcriptomes, we were only able to survey a small proportion of the protein-coding genes that are known to be expressed in the kidneys. Future analyses based on whole-genome sequences would provide a clearer picture of protein evolution. Although expression evolution and amino acid sequence evolution have been found to be correlated in some cases (Nuzhdin et al. 2004; Lemos et al. 2005), we did not find significant overlap in the number of genes showing shared patterns of gene expression and shared signatures of selection. The small amount of overlap might reflect differences in the selection pressures on these two classes of changes. For example, cis-regulatory changes in gene expression are often controlled in a tissue-specific and developmental-stage-specific manner, and as such are expected to be less pleiotropic and thus less constrained in evolution (e.g., Wray 2007). Protein-coding changes, on the other hand, affect all tissues and developmental stages in which the protein is expressed and thus may be more pleiotropic and consequently more constrained. The small amount of overlap might also reflect both statistical and methodological limitations of our study. First, the analytic methods used to detect shared expression changes and shared signals of selection are quite distinct and likely have different false-negative and false-positive rates. Second, we studied gene expression in adults, yet gene expression varies considerably during kidney development (Schwab et al. 2003) and early expression is undoubtedly important in establishing morphological differences between desert and non-desert kidneys. Third, kidneys have a heterogenous cellular composition, and changes in cellular composition between species are likely to affect measures of gene expression in bulk preparations. Future studies of gene expression in single cell preparations at early developmental time points might uncover additional signals of expression variation associated with desert conditions. It would also be valuable to study expression changes in both males and females as the demands of water balance may be especially acute in lactating females. Despite these caveats, we identified a number of potential candidate genes associated with desert living, including some that showed both shared gene expression and shared amino acid sequence evolution. The target available to selection in a trait as complex as desert adaptation is likely large and constrained along each lineage to a different degree by other aspects of the organism's morphology and physiology. Nonetheless, an interesting outcome of our analysis is that a number of the genes and pathways identified here are similar to those identified in other studies of rodent and mammalian desert adaptation (Marra et al. 2012, 2014; Wu et al. 2014; MacManes 2017; Giorello et al. 2018; Tigano et al. 2020). It is clear that gene families such as aquaporins, which are responsible for facilitating water transport across membranes, and solute carriers, may play a role in mitigating water loss across multiple systems and therefore underlie convergent evolution at the genetic level to desert environments. Altogether, our results demonstrate the power of gene expression studies in a phylogenetically controlled manner for identifying shared changes associated with habitat. Incorporating whole-genome (or whole exome) sequences of these and other taxa could provide a more complete comparison of protein coding and gene expression changes in facilitating adaptation to deserts.
Five adult male mice for each species, apart from H. desmarestianus, for which only four samples could be obtained, were included in this study. Chaetodipus intermedius, Zapus princeps, and Mus musculus were caught by N. Bittner using Sherman live traps set overnight following the guidelines of the American Society of Mammalogists (Sikes and Gannon 2011) and an ACUC protocol approved by UC Berkeley (AUP-2016-03-8536). Both desert and non-desert animals were offered apple after capture to avoid acute dehydration in traps. Mice were euthanized by cervical dislocation, and kidney and liver were removed and preserved in RNAlater according to manufacturer's instructions. C. intermedius were trapped near Tucson, AZ, USA, Z. princeps were trapped at Sagehen Creek Field Station near Truckee, CA, USA, and M. musculus were trapped near Berkeley, CA, USA. Heteromys desmarestianus were collected in Chiapas, Mexico by Beatriz Jimenez, and Notomys alexis were collected by Kevin Rowe in Northern Territory, Australia. Mice collected by N. Bittner were prepared as museum specimens (skins and skulls) and deposited in the collections of the UC Berkeley Museum of Vertebrate Zoology. Animals collected by K. Rowe were prepared as museum specimens and deposited at Museums Victoria. The collecting localities, collector's numbers, and museum catalog numbers for each specimen are provided for all wild-caught animals in supplementary table S1, Supplementary Material online. Samples from J. jaculus were provided by Kim Cooper at UC San Diego from an outbred lab colony.
To target loci underlying adaptation to desert environments, we focused on genes expressed in the kidney. RNA was extracted from kidney preserved in RNAlater using the MoBio Laboratories Powerlyzer Ultraclean Tissue & Cells RNA Isolation Kit. Remaining DNA was removed with DNAse-1 followed by a Zymo RNA Clean and Concentrator column clean-up. Due to the lower quality of some samples (RIN scores <5), a ribosomal RNA depletion step was performed with a KAPA Riboerase Kit before libraries were prepared with the KAPA HyperPrep Kit. Libraries were pooled and sequenced across two lanes of 150 bp PE Illumina NovaSeq (one lane of S1 and one of SP) at the Vincent J. Coates Genomics Sequencing Center at UC Berkeley. One library from each species (except M. musculus; see below) was sequenced at greater depth for transcriptome assembly; these were sequenced to a target of 100M read pairs while the remaining 24 libraries, intended for expression analysis, were sequenced to a target of 20M read pairs (see supplementary file S1, Supplementary Material online).
For each of the five 100M-read-pair libraries, reads were examined for quality metrics with FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and then corrected by removing erroneous k-mers using rCorrector (Song and Florea 2015). Adapters and poor-quality sequence were trimmed using Trim Galore! (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). As FastQC revealed a large quantity of duplicates within the sequenced libraries, which is likely in part due to rRNA contamination, we chose to remove all reads that mapped to known rodent rRNA from NCBI using bowtie2 (Langmead and Salzberg 2012). We ran Trinity v2.1.1 (Grabherr et al. 2011) to generate a transcriptome assembly for each species. Because transcriptome-depth (i.e., 100M read pairs) sequencing was not done for M. musculus, reads from all five individuals (approximately equal to the sequencing depth for transcriptome individuals) were combined to assemble the transcriptome of a local individual as above (supplementary table S2, Supplementary Material online). To remove redundant transcripts from the Trinity assembly, transcripts with ≥95% sequence identity were clustered with CD-HIT-EST (settings: -c 0.95 -n 8; Li and Godzik 2006) to create representative transcripts before use in downstream analysis (supplementary table S2, Supplementary Material online). This was done to collapse transcript isoforms as well as to remove transcripts created by assembly errors (chimeras, duplicates, misassembled transcripts, and the like). Transrate (Smith-Unna et al. 2016) was used to calculate assembly statistics. To assess assembly completeness, we used BUSCO (Manni et al. 2021) to look for the 6,192 orthologs found in the Euarchontoglires odb9 database and thus expected to exist in the taxa studied here.
To identify coding regions within our assembled transcripts for downstream analyses, we utilized TransDecoder v5.5.0 (http://transdecoder.sourceforge.net). We identified the longest ORF and searched for matches to both the Pfam protein domain database (Bateman et al. 2004) and mouse-specific SwissProt database (Bairoch and Apweiler 2000) to retain ORFs based on homology. As high-quality gene annotations were available for M. musculus, we used the curated RefSeq protein database for this species. Orthologous gene groups across all six taxa were identified using OrthoFinder v2.3.3 (setting: -S diamond; Emms and Kelly 2015). To minimize the number of alternate isoforms used in the analysis, we only used the longest ORF identified per gene.
Raw reads from all libraries were examined for quality with FastQC. Adapters and poor-quality sequence were trimmed using Trimmomatic v0.36 (Bolger et al. 2014). The five libraries that were generated for transcriptome assembly were subsampled to the average read number of the libraries generated for expression (27,787,405 reads). Reads were mapped to transcriptomes generated for each species with Salmon v 0.14.1 (Patro et al. 2017). To compare across genera, transcripts were annotated using BLASTn to the RefSeq cdna database for M. musculus. Read counts were summed across transcripts for each annotated gene.
DESeq2 (Love et al. 2014) was used to normalize for differences in library size and to call DE between species within each family and across all samples. As transcripts between species can differ in length, a length correction was applied. Reads were subsequently transformed with a variance stabilizing transformation for PCA. We used DESeq2 to identify shared patterns of DE between desert and non-desert species across all three families using an approach similar to that used by Parker et al. (2019). We fit a generalized linear model for gene expression as a function of habitat (desert vs. non-desert), family (species pair), and their interaction. Genes were classified as shared and differentially expressed in cases where there was a significant effect of habitat (desert vs. non-desert, FDR < 0.01) but no interaction effect of species pair by habitat (FDR > 0.05). This analysis was restricted to genes with greater than an average of 20 reads per sample for each species, resulting in a total of 8,174 genes. P-values were adjusted for multiple testing using a Benjamini and Hochberg (1995) correction. Adjusted P-values (FDR) are reported in the results. DE within each species pair was identified using pairwise contrasts. For pairwise contrasts, genes with a mean of fewer than ten reads per sample were removed from the analysis. Shared expression differences among pairs of desert and non-desert species could be due to convergent evolution in response to similar selection pressures, a similar plastic expression response in similar environments, or chance. To assess the contribution of phenotypic plasticity to observed expression differences between species, we analyzed a previously published RNA-seq data set from M. musculus in which mice were deprived of water for three days (Bittner et al. 2021). Comparisons between control mice (water ad libitum) and treatment mice (water deprived) were used to identify expression changes in the kidney that reflect plastic responses to xeric conditions. This set of genes was then compared with the set of genes showing expression divergence between M. musculus and N. alexis and with the set of genes showing expression divergence across all three xeric-mesic species pairs. These analyses suggest that only a small fraction of the expression differences between species is due to phenotypic plasticity (see Results). In addition, we note that one desert-adapted species (J. jaculus) was reared in the laboratory with access to water. Therefore, expression differences between J. jaculus and Z. princeps cannot be attributed to differences in the availability of water. To assess whether more genes showed shared DE shifts by habitat type than expected by chance, we used permutation tests as described in Parker et al. (2019). For each gene, read counts were randomly assigned to habitat within each species pair. All biological replicates (i.e., all five individuals) in each species were assigned to the same habitat. This process was used to create 10,000 permuted data sets. The number of shared differentially expressed genes in these data sets was compared with that of the observed data set. We note that this approach, while used in other studies, treats genes as independent and so may not fully account for shared expression responses in sets of co-regulated genes.
Using the single-copy ortholog groups generated by OrthoFinder for all six species, we aligned these using Guidance2 (Sela et al. 2015) with the standard parameters for a MAAFT alignment for amino acids. This provided alignment quality scores for all 1,855 genes and we removed those for which the alignment quality score was poor (mean column score <0.8) resulting in a set of 1,474 genes with protein-coding alignments for subsequent analyses. This reduced data set reflects the fact that to be included, a gene must be expressed and well assembled in the transcriptomes of all six species. To identify genes in desert lineages displaying evidence of selection, we calculated the ratio of nonsynonymous substitutions per nonsynonymous site to synonymous substitutions per synonymous site (ω) at each locus to estimate rates of evolution along each branch. We used a maximum likelihood approach in a phylogenetic context by implementing the codeml package in PAML v 4.9 (Yang 1997). The following three analyses were implemented using alignments for the 1,474 single-copy orthologs with high-quality alignments present in all species with an unrooted species tree: (M. musculus, N. alexis), (Z. princeps, J. jaculus), (H. desmarestianus, C. intermedius). In the first analysis, we tested for evidence of selection shared by and exclusive to all three desert lineages for each ortholog. We reason that genes with elevated rates of evolution in organisms with shared habitat types may relate to adaptation to that environment. To this end, we used a two-ratio branch model (model 2) in which we allow ω to vary along the branches using the F3 × 4 codon model. For each gene, we estimated ω along the three desert branches and estimated a separate value for ω along the three non-desert branches (Yang 1998, 2007). While traditionally a value of ω > 1 is considered evidence of positive selection, this is generally considered too strict for a gene-wide analysis, as selection is thought to be acting on individual codons rather than on the entire gene. Therefore, a significantly greater value of ω in desert lineages compared with non-desert lineages is considered evidence of positive selection, while the converse is considered evidence of purifying selection. A greater value of ω in desert lineages compared to non-desert lineages could also be caused by a relaxation of selective constraint in desert lineages. Significance is measured by comparing the value with a null model using a χ2 test. P-values were adjusted for multiple testing using a Benjamini and Hochberg (1995) correction. These adjusted P-values (FDR) are reported. This analysis was intended to identify genes underlying desert adaptation common to all three species. In the second analysis, we utilized the framework as in the first analysis but executed it for each desert species individually. We used a two-branch ratio model and compared C. intermedius, J. jaculus, and N. alexis individually to the other five species in the tree. This analysis was intended to identify species-specific adaptations. We hypothesize that while genes under selection in all three desert lineages are intriguing candidates for desert selection, many lineage-specific modifications may contribute to adaptation to extreme environments. Values of ω were compared at each locus with a null model as above to assess for evidence of selection. For the third analysis, we performed a branch-site model which allows ω to vary both across sites in the protein and across branches on the tree to allow detection of specific codons under selection. The first two analyses (above) calculate averages of locus-wide estimates of ω. By interrogating each codon, a more localized signal of selection may be identified. For this analysis, we tested codons for values of ω > 1 along branches for multiple values of Ω (0.1, 0.5, 1.5) and compared this to a null model, as described above.
For gene sets of interest, GO category enrichment tests were performed with GOrilla (Eden et al. 2009) implemented with their online platform. For the sequence-based analyses, we compared the subset of genes with evidence of selection against the remainder of genes used in the analysis to determine whether the subset was enriched for GO categories of interest. We retained those with an FDR corrected P-value generated by GOrilla <0.05 for further analysis. For the expression-level analyses, we compared all significant genes regardless of direction of expression difference against the remainder of genes for which we had expression information. Phenotype enrichment tests were performed with modPhea with their online platform (Weng and Liao 2017) using the same framework.
Click here for additional data file. |
PMC9648514 | 36263788 | Karl Dyrhage,Andrea Garcia-Montaner,Daniel Tamarit,Christian Seeger,Kristina Näslund,Tobias C Olofsson,Alejandra Vasquez,Matthew T Webster,Siv G E Andersson | Genome Evolution of a Symbiont Population for Pathogen Defense in Honeybees | 20-10-2022 | defensive symbionts,evolution,Apilactobacillus kunkeei,plasmids,mobile elements,transposons | Abstract The honeybee gut microbiome is thought to be important for bee health, but the role of the individual members is poorly understood. Here, we present closed genomes and associated mobilomes of 102 Apilactobacillus kunkeei isolates obtained from the honey crop (foregut) of honeybees sampled from beehives in Helsingborg in the south of Sweden and from the islands Gotland and Åland in the Baltic Sea. Each beehive contained a unique composition of isolates and repeated sampling of similar isolates from two beehives in Helsingborg suggests that the bacterial community is stably maintained across bee generations during the summer months. The sampled bacterial population contained an open pan-genome structure with a high genomic density of transposons. A subset of strains affiliated with phylogroup A inhibited growth of the bee pathogen Melissococcus plutonius, all of which contained a 19.5 kb plasmid for the synthesis of the antimicrobial compound kunkecin A, while a subset of phylogroups B and C strains contained a 32.9 kb plasmid for the synthesis of a putative polyketide antibiotic. This study suggests that the mobile gene pool of A. kunkeei plays a key role in pathogen defense in honeybees, providing new insights into the evolutionary dynamics of defensive symbiont populations. | Genome Evolution of a Symbiont Population for Pathogen Defense in Honeybees
The honeybee gut microbiome is thought to be important for bee health, but the role of the individual members is poorly understood. Here, we present closed genomes and associated mobilomes of 102 Apilactobacillus kunkeei isolates obtained from the honey crop (foregut) of honeybees sampled from beehives in Helsingborg in the south of Sweden and from the islands Gotland and Åland in the Baltic Sea. Each beehive contained a unique composition of isolates and repeated sampling of similar isolates from two beehives in Helsingborg suggests that the bacterial community is stably maintained across bee generations during the summer months. The sampled bacterial population contained an open pan-genome structure with a high genomic density of transposons. A subset of strains affiliated with phylogroup A inhibited growth of the bee pathogen Melissococcus plutonius, all of which contained a 19.5 kb plasmid for the synthesis of the antimicrobial compound kunkecin A, while a subset of phylogroups B and C strains contained a 32.9 kb plasmid for the synthesis of a putative polyketide antibiotic. This study suggests that the mobile gene pool of A. kunkeei plays a key role in pathogen defense in honeybees, providing new insights into the evolutionary dynamics of defensive symbiont populations.
The contribution of the beehive microbiota to the health of honeybees is poorly understood. The bacterium Apilactobacillus kunkeei is highly abundant in the honey crop and the honeybee food products. By comparing the genomes and mobilomes of more than 100 novel A. kunkeei isolates, we found that some strains contained plasmids for molecular defense systems. Our in vitro studies confirmed that strains that contained the plasmid-encoded biosynthetic gene cluster for enzymes involved in the synthesis of kunkecin A inhibited growth of the bee pathogen Melisococcus plutonius. This plasmid was stably maintained in strains obtained from one of four beehives throughout the summer months. We propose that A. kunkeei is a defensive symbiont of honeybees and that its mobilome provides a dynamic protection system against M. plutonius.
Symbiotic relationships between bacteria and eukaryotes are common in nature, affecting the physiology, development, behavior, and growth habitat of the host. Many insects house nutritional symbionts, which serve as small bacterial factories for the production of amino acids, vitamins, or other compounds lacking in the host diet (reviewed in Dale and Moran 2006; Toft and Andersson 2010; Bennet and Moran 2015). Most obligate nutritional symbionts have undergone reductive genome evolution and massive gene loss, whereas genes that serve host–beneficial functions have evolved under strong selective pressures. Defensive symbionts, which provide protection against parasites and pathogens, are also widespread in insects (reviewed in White and Torres 2009; Clay 2014; Ford and King 2016; King 2018; Vorburger and Perlman 2018). These bacteria offer protection from infectious diseases by secreting antimicrobial compounds and/or by stimulating the host immune system (reviewed in Van Arnam et al. 2018). The symbiotic relationships are more complex and dynamic than those of obligate nutritional symbionts, and the genomes of defensive symbionts are large. However, one-third of the protein coding genes in the 7 Mb genome of the vertically inherited actinobacterial symbionts of solitary beewolf wasps contain frameshift mutations, suggesting that it is an early stage of genome reduction (Nechitaylo et al. 2021). In Lagria beetles, one strain of Burkholderia gladioli has a genome of about 2.3 Mb, whereas other coexisting strains of the same species have genomes of 8.5 Mb (Floréz et al. 2018). Advanced stages of genome reduction are very rare but has been observed in Candidatus Profftelia armature which lives inside bacteriocytes in its psyllid host and has a drastically reduced genome of only 465 kb, 15% of which contain genes for the synthesis of a polyketide toxin (Nakabachi et al. 2013). The molecular defense systems may be encoded by horizontally acquired chromosomal genes, as in Candidatus Profftelia armature (Nakabachi et al. 2013), or associated with mobile elements, such as plasmids in Pseudonocardia symbionts of fungus-growing ants (Gerardo and Parker 2014; Van Arnam et al. 2015), or phages, as in the Hamiltonella symbionts defending aphids against parasitoid wasps (Degnan and Moran 2008; Oliver et al. 2009). Yet, very little is still known about the mobility of host-selected bacterial defense systems, and the evolutionary dynamics of defensive symbiont populations have rarely been explored beyond comparisons of 16S rRNA and genome sequence identities. Honeybees provide an attractive model system for evolutionary studies of host-associated microbial communities and their influence on host health (Engel et al. 2016; Kwong and Moran 2016; Zheng et al. 2018; Nowak et al. 2021). Honeybees are social organisms who share their food sources (nectar and pollen) and nurse the next generation of bees in a collaborative manner. They establish colonies comprising 30,000–80,000 adult workers. As in human societies, their social life style makes the colonies vulnerable to infections by viruses, parasites and bacterial pathogens. The most common infectious disease agents of honeybees are microsporidians and other fungi, RNA viruses transmitted by the Varroa mite, and bacteria such as Paenibacillus larvae and Melissococcus plutonius, which causes larval foulbrood disease (reviewed in Funfhaus et al. 2018; Li et al. 2018a, 2018b). Altogether, these infections pose severe threats to honeybee colonies worldwide and are of great economic concern for farmers and the global commercial honeybee industry. Honeybees use a variety of defense mechanisms to tackle the threats from infectious diseases, including changes in social behavior, establishment of physical barriers in the gut and induction of apoptosis and immune response systems (Doublet et al. 2017; Li et al. 2018a, 2018b). Some protection against infectious diseases may also be obtained from the gut microbiome, which consists of about 108 to 109 bacterial cells per worker bee. The core bee gut microbiome consists of only a few bacterial genera, including several species of Snodgrasella, Gilliamella, Lactobacillus and Bifidobacterium (Martinson et al. 2011; Engel et al. 2012, 2014; Moran et al. 2012; Sabree et al. 2012; Jones et al. 2017; Ellegaard and Engel 2019). A few additional, but less prevalent species have also been identified, such as Frischella perrara, which solely infects the pylorus (Engel et al. 2013). The establishment of the core lineages in the gut coincided with the diversification of eusocial corbiculate bees from solitary bees (Kwong et al. 2017a, 2017b). The composition of the gut microbiome in worker bees has been shown to be influenced by infections of M. plutonius (Erban et al. 2017), as well as by the administration of the honeybee gut bacterium Snodgrassella alvi and the protozoan parasite Lotmaria passim (Schwarz et al. 2016). Comparative studies of bees with and without a microbiota provided support for the hypothesis that the gut microbiome exerts an effect on the host immune system (Kwong et al. 2017a, 2017b). Frischella perrara in particular has a strong effect on the immune system, possibly indicating that it is a pathogen rather than a mutualist (Emery et al. 2017). However, the mechanisms involved and the impact of the individual members of the gut microbiome on the health of individual worker bees are still poorly understood (discussed in Engel et al. 2016). Apilactobacillus kunkeei (formerly named Lactobacillus kunkeei) is a fructophilic lactic acid bacterium (Endo et al. 2012) and the dominant bacterial species in the honey crop (Vasquez et al. 2012), which is the first stomach (or sac) in the gut used for storage and transport of nectar to the hive. Apilactobacillus kunkeei has also been found on flowers as well as in the honeybee food products, such as fresh honey, bee bread, and royal jelly (Olofsson and Vasquez 2008; Vasquez and Olofsson 2009; Anderson et al. 2013; Endo and Salminen 2013; Tamarit et al. 2015; Anderson and Ricigliano 2017). Because of its broad prevalence in the food sources of honeybees, A. kunkeei has often been viewed as an environmental hive bacterium rather than a core gut symbiont (Anderson et al. 2013; Anderson and Ricigliano 2017; Kwong et al. 2017a, 2017b). Yet increasingly, it is recognized that A. kunkeei may be a key player in the defense against pathogens and parasites in honeybee colonies. For example, it has been shown that A. kunkeei can inhibit the growth of the bacterium P. larvae (Forsgren et al. 2010; Butler et al. 2013; Kacaniova et al. 2020), the microsporidian Nosema ceranae (Arredondo et al. 2018), as well as the fungus Ascosphaera apis (Iorizzo et al. 2020). Growth inhibition of P. larvae was also observed when cell-free supernatants of A. kunkeei were used, suggesting that the antimicrobial effects were caused by secreted metabolites and/or proteins (Lamei et al. 2019). Recently, a strain of A. kunkeei FF30-6 that exhibits antibacterial activity (Endo and Salminen 2013) was found to contain a plasmid (pKUNF330) that codes for enzymes involved in the synthesis of a novel antimicrobial compound, named kunkecin A (Zendo et al. 2020). This metabolite has a narrow antimicrobial spectrum with high activity against M. plutonius (Zendo et al. 2020), suggesting that A. kunkeei strains hosting this plasmid serve a role as defensive symbionts of honeybees. The few genomes sequenced until now, only one of which is closed (Asenjo et al. 2016), indicate extensive gene content diversity among strains despite near 16S rRNA identity (Tamarit et al. 2015). Notably, the A. kunkeei genome is functionally structured such that genes encoding metabolic functions are located near the origin of replication, whereas genes for basic information processes are located around the terminus (Tamarit et al. 2015). Despite the many advances that the genomic studies have offered, it is not known if the plasmid for kunkecin A biosynthesis is widespread in the A. kunkeei population and if additional plasmids with defensive functions are circulating in the honeybee growth niche. In this study, we compare the closed genomes and the associated mobilomes of 102 new A. kunkeei isolates along with re-sequenced genomes of A. kunkeei Fhon2 and Apilactobacillus apinorum Fhon13. The novel strains were sampled from the honey crop of honeybees in the south of Sweden (Helsingborg), as well as from bees at two geographically isolated islands in the Baltic Sea (Åland and Gotland). The Gotland bees are a mixture of European ancestries caused by importation from other parts of Europe and fairly typical of managed honeybees all over Sweden (Wallberg et al. 2014). However, whereas the Gotland bees belong to a bee population that lives in contact with the Varroa mite and has developed a natural resistance to it, the bees from Åland come from a Varroa mite-free area (Fries et al. 2006; Lattorff et al. 2015; Thaduri et al. 2018). Thus, the sampled honeybee populations have evolved under distinct ecological settings. We present an overview of the mobile gene pool in the A. kunkeei population and discuss its role in food preservation and health in honeybee colonies.
Bacterial colonies were isolated from the honey crop of honeybees sampled from four beehives (fig. 1). In total, 102 novel bacterial isolates were obtained, of which 61 isolates were from Helsingborg, 17 from Åland and 24 from Gotland. At all sites, 1–7 bees were collected per beehive and sampling time point, of which 1–14 bacterial colonies per bee were isolated (supplementary table S1, Supplementary Material online). Samples from beehives in Helsingborg were taken as a time series from May to August. The generation times of the new A. kunkeei isolates in fMRS media ranged from 34 to 61 min, with an average of 48.3 min (calculated from the mean of the mean growth rate for each isolate) (supplementary tables S2 and S3, Supplementary Material online). The genomes of the novel isolates as well as the re-grown strains A. kunkeei Fhon2 and A. apinorum Fhon13, altogether 104 genomes, were sequenced and assembled into single contigs with sizes ranging from 1.49 to 1.64 Mb (mean 1.57 Mb) (supplementary table S3, Supplementary Material online). The genomes contained 5 rRNA operons, 65–72 tRNA genes and 1,345–1,504 (mean 1,430) protein coding genes (supplementary table S3, Supplementary Material online). Plasmids and phage–plasmids of 20–40 kb were identified in 40 isolates (supplementary table S3, Supplementary Material online). The ratios of sequence read coverage over these elements compared with the chromosome were on the average equal to or lower than one for each type of plasmid and phage-plasmid (supplementary fig. S1, Supplementary Material online).
Pairwise 16S rRNA sequence identities were calculated for each of the 102 novel isolates in comparisons to A. kunkeei Fhon2 and A. apinorum Fhon13. The values ranged from 99.4% to 100% to A. kunkeei strain Fhon2 and from 98.4% to 99.2% to A. apinorum strain Fhon13 (supplementary table S3, Supplementary Material online). Moreover, a maximum likelihood phylogenetic tree (fig. 2; supplementary fig. S2, Supplementary Material online) based on a concatenated alignment of 682 proteins (supplementary table S4, Supplementary Material online) showed that all isolates obtained in this study clustered with the previously described A. kunkeei strains to the exclusion of A. apinorum Fhon13. The tree topology indicated that 95% of the novel A. kunkeei isolates belonged to three main clades, corresponding to the previously suggested phylogroups A, B, and C (fig. 2; supplementary fig. S2, Supplementary Material online). The remaining four isolates formed two deeply diverging lineages, here named phylogroups E and F. The average nucleotide identity (ANI) values for strain comparisons between phylogroups A, B and C were about 95–96%, while pairwise comparisons of these strains to those classified into phylogroups E and F were in the range of 90–93% (supplementary table S5, Supplementary Material online). Thus, based on the 16S rRNA analysis alone, all novel isolates should be classified as strains of the A. kunkeei species, while the ANI values suggest that the four isolates from Åland in phylogroups E and F should be considered as belonging to a different species. For now, we have named them as novel strains of the A. kunkeei species, while awaiting taxonomic evaluation. The large majority of isolates obtained from honeybees from the island Gotland belonged to phylogroup A, while a few represented a rare variant in phylogroup B obtained from a single bee that was sampled more deeply. Samples from the island Åland contributed isolates from most phylogroups and the sampled bees contained mostly isolates from two and sometimes even three different phylogroups, indicative of a more diverse strain co-occurrence profile. The A. kunkeei isolates obtained from Helsingborg were evenly distributed between the A and C phylogroups in addition to a few isolates affiliated with phylogroup B. However, there was a bias such that the A-group strains were more abundant in isolates cultivated from beehive 3, whereas the C-group strains dominated among isolates obtained from beehive 4. The community profiles of the isolates from Helsingborg were largely specific for each of the two beehives and remarkably stable during the summer months (May–August).
The pan-genome structure of the A. kunkeei population was examined based on 2,656 protein families (supplementary table S4, Supplementary Material online) that contained proteins encoded by chromosomal genes in the 104 isolates with closed genomes (fig. 3A). In addition to the 1,134 protein families shared by all isolates, the “soft-core” proteome (shared by 95–99% of the strains) consisted of 97 families, the “shell” proteome (shared by 15–95% of the genomes) contained 315 families, and the “cloud” proteome (shared by 0–15% of the genomes) included 1,110 families, of which 620 were specific for a single strain. The cumulative number of families continued to increase with the addition of new isolates, showing no sign of flattening (fig. 3B). The core proteins encoded by genes present in all genomes were mostly involved in information processes and metabolic systems, while defense systems and mobile genetic elements were over-represented among the shell and cloud proteins (fig. 3C; supplementary table S4, Supplementary Material online). Thus, the sampled A. kunkeei community retained an open pan-genome structure despite being represented by over 100 genomes, which suggests high ecological plasticity in the species.
Most genomes contained a prophage of 35–50 kb located at either of two sites around the terminus of replication (between positions ∼600,000 and ∼1,000,000, respectively) (fig. 4; supplementary fig. S3, Supplementary Material online). Furthermore, all genomes contained a phage defense island, which encoded either a CRISPR-Cas system or a restriction-modification system, or a prophage as in the MP2 genome (supplementary fig. S4, Supplementary Material online). On average, we identified 52 transposase genes per genome (ranging from 2–90 per genome), many of which were found in pairs and belonged to the IS3 family (subgroup IS150) (fig. 4; supplementary fig. S3, Supplementary Material online). A plot of the average number of transposases along 34 reference genomes (selected to represent groups of genomes with ANI values of 99.9%) revealed multiple transposon-dense areas (fig. 4A). In the chromosomal half that flanks the origin of replication, these were often associated with chromosomal segments that were variable in gene content (fig. 4B; supplementary fig. S3, Supplementary Material online), suggesting that much of the genomic variability in the A. kunkeei population is due to transposon-mediated activities. Indeed, we identified genomic heterogeneity within some of the isolates in the form of transposon-containing contigs that were homologous to chromosomal segments. One such contig in strain H3B2-03M contained genes that were homologous to genes for enzymes involved in exopolysaccharide synthesis and biofilm formation located in a chromosomal segment of 30.3 kb flanked by transposons. This strain segregated upon repeated culturing and plating into colonies with different growth kinetics (supplementary fig. S5 and supplementary table S2, Supplementary Material online). Resequencing showed that the extrachromosomal contig was retained in one colony, but lost in two colonies of which one had also lost the 30.3 kb chromosomal segment (supplementary table S3, Supplementary Material online).
We identified three plasmids and two phage–plasmids in the 20–40 kb size range that contained a gene for the replication protein RepA located next to a gene for the plasmid partitioning protein ParA (fig. 5; supplementary table S6, Supplementary Material online). The three plasmids contained genes for accessory mobile elements, such as IS3 family transposases and the Tn3 family resolvase, as well as genes for adhesins or enzymes putatively involved in the synthesis of antibiotics, as detailed below. The two phage–plasmids contained phage genes for the production of phage particles, host cell lysis and infection in addition to the repA and parA genes (supplementary table S6, Supplementary Material online). The phage proteins showed homology to phage protein families pLP39 and p48, respectively, which have been identified previously in other Lactobacillus species (Pfeifer et al. 2021). In addition, we identified smaller replicons in the 7–10 kb size range, some of which contained a gene for the rolling-circle initiator protein RepB as well as genes coding for macrolide export and ESX secretion systems or DNA methylases and DNA restriction enzymes. The 19.5 kb pKUN plasmid contained a cluster of eight consecutive genes for the synthesis of kunkecin A, a nisin-like antibiotic (fig. 5A; supplementary table S6, Supplementary Material online). The cluster included genes for the kunkecin A precursor peptide, the leader peptidase, proteins involved in the modification and transport of the antibiotic and proteins conferring self-immunity, as in the plasmid described in (Zendo et al. 2020). Downstream of this gene cluster, we identified a gene coding for a protein with a histidine kinase domain and a gene for a protein with a helix-turn-helix motif. We suggest that these two genes code for homologs to the NisR/NisK two-component regulatory system, which were previously thought to be missing from the kunkecin A biosynthetic gene cluster (Zendo et al. 2020). The 32.9 kb plasmid was named pPKS because it contained a cluster of genes for polyketide synthesis (fig. 5B; supplementary table S6, Supplementary Material online). This plasmid contained three genes for the beta-ketoacyl synthetase subunits KSα and Ksβ and the acyl carrier protein, which constitute the minimal polyketide synthase complex. The cluster also contained genes for enzymes involved in methyl transfer, decarboxylation and cyclization of polyketide antibiotics. Additionally, the plasmid contained three genes coding for proteins with high sequence similarity to ATP-grasp domain proteins in Streptococcus and Lactobacillus species (1e−95). The 33 kb plasmid was named pLPX because it contained a 4 kb gene coding for a surface protein with the cell-wall anchoring LPxTG motif (fig. 5C; supplementary table S6, Supplementary Material online). The gene for the LPxTG-containing protein was located next to a gene for a serine recombinase involved in site-specific DNA inversions, suggesting that its expression may be controlled by phase variation. The gene coding for the protein with the LPxTG motif was homologous to more than ten chromosomal genes in the A. kunkeei population coding for proteins of variable sizes, of which four to six different gene variants were present in each genome (fig. 6A). Different proteins containing the LPxTG motif were located at the same chromosomal site in different strains, indicative of replacement events within and across phylogroups (fig. 6BandC). Next, we compared the presence of plasmids with the ability of the strains to inhibit growth of M. plutonius. For this analysis, we selected a set of 47 strains that represent the genetic diversity of all isolates (see fig. 4), including 31 strains that did not contain the pKUN plasmid and another 16 isolates that contained the plasmid. Fifteen of the 16 isolates that contained the pKUN plasmid showed growth inhibition of M. plutonius (fig. 7; supplementary fig. S6 and supplementary table S7, Supplementary Material online). Strain H1B1-05A was the sole isolate that contained the pKUN plasmid but was unable to inhibit growth of M. plutonius. Notably, this was also the only isolate in which one gene in the biosynthetic gene cluster was truncated. None of the A. kunkeei reference strains without the pKUN plasmid were able to inhibit growth of M. plutonius (supplementary fig. S7 and supplementary table S7, Supplementary Material online). Based on these results, we conclude that the plasmid-encoded biosynthetic gene cluster is responsible for growth inhibition of M. plutonius. Finally, we examined the phyletic distribution patterns of the extrachromosomal replicons (fig. 8; supplementary table S6, Supplementary Material online). The pKUN plasmid was solely identified in phylogroup A strains from Helsingborg, 13 of 16 of which was sampled from hive 3 (fig. 8; supplementary table S6, Supplementary Material online). The pPKS plasmid was solely identified in phylogroup B and C strains from Helsingborg, 12 of 13 of which were sampled from hive 4. Likewise, the distribution pattern of the two phage–plasmids was phylogroup-specific: one was found in isolates of phylogroup A, while the other was identified in isolates from phylogroups BC and E. Additionally, three phylogroup A isolates from Helsingborg and three phylogroup B isolates from Gotland contained the pLPX plasmid for surface attachment. In contrast to these distinct distribution patterns, the smaller 7–10 kb replicons were broadly present in isolates from all phylogroups and beehives. Generation times were highly variable across isolates irrespectively of their mobile element repertoires (fig. 8; supplementary tables S2 and S3, Supplementary Material online). No PCR tests were performed to test for infections in the hives, but all bees from which bacterial samples were obtained were healthy at the time of sampling. The identification of antibiotic-producing A. kunkeei isolates in samples obtained from the two beehives in Helsingborg may indicate previous exposure to M. plutonius or other pathogens, in contrast to the lack of such isolates from the mite-free and mite-resistant bees at Åland and Gotland, respectively.
Insects use bacterial symbionts for the supply of nutrients that are lacking in their diet or as antimicrobial defense systems to tackle the threats from infectious disease agents. There is much hope that understanding the ecology and evolution of the honeybee gut microbiome will help prevent the dramatic losses of managed honeybee colonies in recent years. Due to its identification in the honeycrop and food products in the beehive, A. kunkeei is considered to be a core component of the beehive and has been suggested to defend the bees against microbial pathogens. In this paper, we compared the genomes of 104 A. kunkeei isolates, including 102 novel genomes and their associated mobilomes. Growth inhibitory effects of the larvae-infecting bee pathogen M. plutonius were observed for 15 of these isolates, all of which contained a plasmid for the synthesis of kunkecin A. We also identified a novel plasmid coding for enzymes putatively involved in the synthesis of a polyketide antibiotic. As such, the results strongly suggest the hypothesis that A. kunkeei is a defensive symbiont of honeybees. The comparison of isolates from four different beehives showed that each hive contained a unique composition of strains. Importantly, the same phylogroup strains and type of plasmids were repeatedly sampled from each of the two beehives in Helsingborg during the summer months, despite a continuous turnover of bees and bacterial cells in the hive. This suggests that A. kunkeei, like the stable members of the honeybee gut microbiome (Powell et al. 2014; Kwong and Moran 2016), is mainly transmitted through contacts within the hive. Likewise, antibiotic-producing Actinobacteria associated with fungus-farming ant lineages are vertically transmitted across ant colony generations (Li et al. 2018a, 2018b). In contrast, environmental symbionts like Burkholderia gut symbionts of stinkbugs are obtained from the environmental microbial pool for every generation (Kikuchi et al. 2011). Nevertheless, previous studies have shown that A. kunkeei strains do not co-speciate with their hosts, suggesting that horizontal transmission of bacteria occurs across sites and host species in the long-term (Tamarit et al. 2015). Likewise, the long-term evolutionary history of the ant-Actinobacteria symbiosis includes horizontal transmission, multiple losses of symbionts, and convergent anatomical adaptations to support the interactions of hosts and symbionts (Li et al. 2018a, 2018b). Thus, although closely related defensive symbionts may show co-divergence with their hosts due to vertical transmission, co-divergence is typically not seen for distantly related organisms due to losses and sporadic horizontal acquisition events (Vorburger and Perlman 2018). Chemical warfare is a common mechanism whereby symbionts protect their hosts from infections, and the plasmid-encoded molecular defense systems are likely to be adaptive for both A. kunkeei and the honeybees. The pKUN plasmid for kunkecin A biosynthesis was mostly associated with phylogroup A strains from Beehive 3 in Helsingborg, while the pPKS plasmid was mostly associated with phylogroups B and C strains from Beehive 4 in Helsingborg. None of the plasmids for antibiotic synthesis were identified in samples from either Gotland or Åland. It is thus likely that selection for plasmids involved in the defense against bee pathogens has driven strains associated with phylogroups A, B and C to high abundances in the beehives located at Helsingborg, while the more diverse bacterial community sampled from the beehive at Åland may have evolved under no or less selection for host-resistance against said pathogen. The role of geography and phylogeny for horizontal gene transfer and how conditions of rampant gene exchange affect bacterial speciation processes are issues of much debate (Arevalo et al. 2019; Greenlon et al. 2019; VanInsberghe et al. 2020). Local emergence of infectious disease agents may generate geographic structures in defensive symbiont populations as observed for A. kunkeei, although most genotypes may be present at most sites. Indeed, selective sweeps driven by antibiotic-producing plasmids may explain the overall abundance of phylogroup A and C strains, as well as the geographic distribution pattern of bacterial isolates and plasmids. Although all bees were healthy at the time of sampling, it is tempting to speculate that the beehives at Helsingborg may have been exposed to bacterial pathogens more recently than the beehives at Åland or Gotland. We do not know if the measures taken to limit the spread of mite infections at Åland have also limited the spread of bacterial pathogens, but it is intriguing to note that the A. kunkeei isolates from this island were genetically more diverse than isolates from the other beehives. In the long-term, host-selected expansion of microbial defense systems could lead to segregation of populations into distinct gene-flow units and thereby drive niche-specialization and speciation. Apilactobacillus kunkeei may thus be used as a model system for studies of niche-specialization and speciation processes in defensive symbiont populations. It is also of interest to note that A. kunkeei circulates among several microhabitats in the beehive and growth of the bacterial population may not necessarily occur in the same habitat as antibiotic synthesis. Such a separation of growth from antibiotic synthesis has been observed for example in the defensive symbiont of beewolf digger wasps, which grows in the antennae of adult female wasps but produces antibiotics in the brood cells and larvae (Kaltenpoh et al. 2010; Kroiss et al. 2010). Furthermore, some polyketides, such as actinorhodin produced by Streptomyces coelicolor (Wright and Hopwood 1976) are synthesized under oxygen-rich conditions at high pH while the compound itself is most active at low pH (Mak and Nodwell 2017). It is thus tempting to speculate that the antibiotic synthesized by enzymes encoded by the pPKS plasmid may be most active in microhabitats of low pH, such as in the fresh honey and royal jelly. Restricting the activity of the antimicrobial compounds to the honeybee food products may be a mechanism to target pathogens that infect the larval food without harming other members of the honeybee microbiome, provided that the antibiotics are not stable enough to persist in the bee food. Bombella apis is also found in multiple and similar microhabitats within the hive, such as the nurse crop, nectar, larval diet and royal jelly. It is shown that B. apis can synthetize all essential amino acids and secrete lysine, suggesting that this bacterial species may serve as a nutritional symbiont of honeybee larvae (Parish et al. 2022). Thus, B. apis and A. kunkeei may jointly support larval development by complementing the larval diet and protecting it against infections. Social insects provide a model for how the transmission of infectious disease agents may be reduced in animal communities with high population densities by incorporating beneficial microbes into the host–parasite co-evolutionary arms race. At the more advanced stages of these interactions, the defensive symbionts may take over the interactions with the pathogens, thereby relaxing selection on the host-derived immune system. Taken together, the results presented in this study suggest that the A. kunkeei community serves an important role in bee health by protecting the larvae and their diet against infections. Future studies should be targeted towards characterizing the putative antibiotics, the microhabitat in which they are synthesized and their mode of action on known bee pathogens. We have currently no evidence to suggest that the honeybees regulate the mobilome and other activities of A. kunkeei, nor that the adoption of defensive symbionts has had an impact on the honeybees own immune system against infectious disease agents, but this is an interesting avenue for further studies. Alternatively, selection may simply have favored bees containing the best fitted A. kunkeii strains. The availability of more than 100 A. kunkeei isolates with different gene complements and plasmids now enable studies to address the long-standing debate about whether horizontally transferred genes are mostly neutral, deleterious or adaptive (discussed in Rocha 2018). Importantly, the results suggest that the complex interactions of hosts, defensive symbionts and pathogens cannot be adequately understood unless the dynamics of the mobile gene pool of the symbionts are incorporated into the models. Understanding these interactions will be important for the design of engineered strains and plasmids to improve the health of honeybees.
Honeybees from the subspecies Apis mellifera were collected from beehives located on the islands Åland and Gotland as well as near Helsingborg, Honeybees from Åland and Gotland were stored at −80 °C. Prior to dissection, honeybees were thawed and placed individually in 2 ml ethanol. Honeybees from Helsingborg were dissected immediately after collection as described in (Olofsson and Vasquez 2008; Vasquez and Olofsson 2009). The honeycrops were extracted and homogenized in PBS and spread on MRS agar (Sigma Aldrich), supplemented with 0.5% fructose and 1.5% glycine (samples from Åland and Gotland) or 2% fructose 0.1% cysteine (samples from Helsingborg). Bacterial isolates were obtained after incubation at 35 °C and 5% CO2 for 2–3 days and re-isolated to obtain pure isolates in the same growth media. The isolates were examined by PCR and Sanger sequencing using universal 16S rRNA primers and MALDI-TOFMS protein profiling as described in (Olofsson et al. 2014). For estimates of generation times, the A. kunkeei strains were first grown on MRS agar plates supplemented with 0.5% D-Fructose (fMRS agar) overnight at 35 °C, 5% CO2. To obtain biological replicates, several single colonies were picked per strain to inoculate liquid MRS medium (+0.5% D-Fructose, fMRS medium) and two rounds of pre-cultures were incubated for 18 h at 35 °C, 5% CO2 after which the concentration of the cell suspensions was adjusted to OD600 = 0.005 in fresh fMRS in multi-well plates and the absorbance at λ = 600 nm was monitored in 10 min intervals for up to 24 h on a Bioscreen C MBR at a temperature of 35 °C (Oy Growth Curves Ab Ltd). Cultures were kept on ice during sample preparation. Biological triplicates of A. kunkeei strain A1401 were included as a positive control in every experiment. Non-inoculated fMRS medium served as negative control. Raw data was log-transformed, background corrected, trimmed after 10 h and the growth dynamics were determined in R (R Core Team) using the growthcurver package (Sprouffske and Wagner 2016). To test the inhibitory potency of A. kunkeei against M. plutonius (DSM29964), cell-free supernatants were isolated from 47 A. kunkeei strains. The strain collection included 33 representative strains shown in figure 4B, excluding A. kunkeei MP2, and, in addition, 14 A. kunkeei strains with predicted pKUN plasmids. Apilactobacillus kunkeei strains H1B1-05A and H3B1-04J with predicted pKUN plasmids were part of the representative strain collection. To obtain the cell-free supernatant, the bacterial cells were cultivated as described for the growth analysis. After the final batch cultivation in fMRS medium for 18 h at 35 °C, 5% CO, cells were pelleted by centrifugation (4,500 × g, 10 min, 4 °C) and the supernatant was passed through 0.2 µm membrane filters. Melissococcus plutonius (DSM29964) was cultivated in DSM 1582 medium at 30 °C for 3 days under anaerobic conditions. Anaerobic growth conditions were created in an Anaerocult jar (Millipore) using Anaerocult A bags and monitored using Anaerotest Strips. For the inhibition studies, M. plutonius was diluted to a cell concentration corresponding to OD600 of 0.05–0.2 and evenly spread on DSM 1582 agar plates. In a spot-on-lawn assay, 10 µl of A. kunkeei cell-free supernatants were added upon the M. plutonius cell lawn and plates were incubated anaerobically at 30 °C for at least 2 days. Inhibition was assessed if a clear inhibition zone was observed after cultivation. Experiments were performed with biological triplicates, cell-free supernatants from strains H3B1-09M and H3B1-10M were used as positive control and fMRS medium was included as a negative control.
The genomes of the 102 novel A. kunkeei isolates as well as the genomes of the previously isolated A. kunkeei Fhon2 and A. apinorum Fhon13 strains were sequenced with PacBio RS II and PacBio Sequel technologies. The reads from each sequencing run were assembled into closed genomes with HGAP3 or HGAP4 (Chin et al. 2013), as detailed in (supplementary table S1, Supplementary Material online). MUMmer v3.23 (Kurtz et al. 2004) was used to rule out mis-assemblies. Pairwise ANI values between chromosomes were calculated using FastANI v1.2 with default parameters. The manually annotated genome of A. kunkeei Fhon2 (Tamarit et al. 2015) was used as reference by the Prokka v1.14.6 annotation pipeline (Seemann 2014) and complemented by searches against all bacterial sequences in UniProtKB (The UniProt Consortium 2020). The eggNOG-mapper v2.1.2 (Huerta-Cepas et al. 2016) and InterProScan v5.51-85.0 (Jones et al. 2014) were applied to the genomes using default parameters. All genomes were scanned for transposable elements, deduced from prokka, eggNOG, and InterProScan functional predictions. Proteins were sorted into Clusters of Orthologous Groups (COGs) using BLASTP against the COG2020 database (Galperin et al. 2021) and discarding overlapping hits. Prophages were predicted using PHASTER with Prophage/Virus database as of December 22, 2020 (Arndt et al. 2016). Extrachromosomal elements were assembled with Flye v2.8.3-b1725 (Kolmogorov et al. 2019) with the –plasmid parameter and –asm_coverage set to 100, 150, 200, and 500. The resulting non-chromosomal contigs were grouped based on a combination of shared gene content, reciprocal blast hits, and ANI values. For each strain containing elements assigned to a given group, the contig with the length closest to the median length of that group was selected as the reference assembly. The two phage–plasmids were classified by taking protein sequences from one representative each and using protein BLAST against all phages described in (Pfeifer et al. 2021) and counting hits. Plasmids and phage–plasmids of 20 kb or more were ordered using the repA, parE, and repB genes. Extrachromosomal contigs solely present in a single strain often contained transposons and homologs to chromosomal genes. These elements were not further examined in this study, except for the transposon-containing contig identified in the assembly of isolate H3B2-03M.
The previously published closed genome of A. kunkeei strain MP2 (Asenjo et al. 2016) and the non-closed genomes of 15 A. kunkeei strains (Djukic et al. 2015; Porcellato et al. 2015; Sun et al. 2015; Tamarit et al. 2015) (supplementary table S8, Supplementary Material online) were re-annotated using the Prokka v1.14.6 annotation pipeline (Seemann 2014) for consistency with the 102 novel genomes. The inferred proteomes of the 102 novel A. kunkeei genomes, the A. kunkeei Fhon2 and A. apinorum Fhon13 genomes, and the 16 previously sequenced A. kunkeei genomes were sorted into protein families using OrthoMCL v2.0 (Li et al. 2003). For the phylogenetic analysis, we selected a subset of core protein families present in all A. kunkeei strains and A. apinorum Fhon13 after exclusion of protein families with multiple members in any single genome, proteins shorter than 100 amino acids, and proteins inferred to be recombinant by all tests in the software PhiPack v1.1 (Bruen et al. 2006). The selected proteins were individually aligned with mafft-linsi v7.453 (Katoh et al. 2002) allowing regions with gaps (option –leavegappyregion). Sequences were trimmed using trimal v1.4rev15 (Capella-Gutiérrez et al. 2009) with automatic detection of optimal thresholds (-gappyout) and concatenated. A phylogenetic tree was inferred with IQ-Tree v1.6.10 (Nguyen et al. 2015), using the LG + F + R5 substitution model, selected using ModelFinder (Kalyaanamoorthy et al. 2017). A total of 1,000 ultrafast bootstrap (option -bb 1000) and SH-like (option -alrt 1000) pseudoreplicates were performed to assess branch support and stability. Apilactobacillus apinorum strain Fhon13 was used to root the tree. For the pan-genome analysis, we selected protein families containing proteins encoded by chromosomal genes in at least one of the 104 A. kunkeei isolates with closed genomes. The protein families were classified as core, soft-core, shell, and cloud depending on whether they contained proteins in all 104 isolates (100%), ≥99 (95%), ≥16 (15%), or ≤15 isolates, respectively.
Click here for additional data file. |
PMC9648531 | Claire Shin,Saeed Ali,Sana Hussain,Itishree Trivedi,Yubo Gao,Asim Shuja | Epidemiology of irritable bowel syndrome in hospitalized patients with inflammatory bowel disease: Nationwide Inpatient Sample analysis from 2007-2016 | 17-10-2022 | Irritable bowel disease,inflammatory bowel disease,epidemiology | Background Despite effective treatments for inflammatory bowel disease (IBD), patients in remission may still suffer from gastrointestinal symptoms attributable to overlying irritable bowel syndrome (IBS). In this population-based cohort study, we investigated the epidemiology of IBS in hospitalized IBD patients and explored the differences between hospitalized IBD-IBS vs. IBD patients to distinguish this patient population. Methods Using the Nationwide Inpatient Sample database from 2007-2016, we identified patients with a primary or secondary discharge diagnosis of IBD, with or without IBS, using ICD-9 and ICD-10 codes. We extracted information on demographics, psychological comorbidities, IBD complications, cost and duration of stay of each group, from either discharge records or diagnosis codes. These were analyzed using SAS version 4.0. Results There was a rise in the prevalence of IBS among inpatients with ulcerative colitis (P=0.025) and Crohn’s disease (P=0.0014) over the study period. This study revealed that IBD patients with IBS tend to be female, younger, are less likely to be morbidly obese and have higher rates of psychological disorders (P<0.001) compared to IBD patients with no IBS co-diagnosis. They also have fewer IBD-specific complications, such as strictures, obstruction, fistula and abdominal abscess (P<0.001). Shorter hospital stays (P<0.001) and lower hospital charges (P<0.001) were also noted in these patients. Conclusions IBD patients with IBS are significantly different from other IBD patients, and are associated with less severe disease, a shorter hospital stay and lower hospital expenses. Early and accurate classification of this patient population may prevent unnecessary treatment and hospitalization in the future. | Epidemiology of irritable bowel syndrome in hospitalized patients with inflammatory bowel disease: Nationwide Inpatient Sample analysis from 2007-2016
Despite effective treatments for inflammatory bowel disease (IBD), patients in remission may still suffer from gastrointestinal symptoms attributable to overlying irritable bowel syndrome (IBS). In this population-based cohort study, we investigated the epidemiology of IBS in hospitalized IBD patients and explored the differences between hospitalized IBD-IBS vs. IBD patients to distinguish this patient population.
Using the Nationwide Inpatient Sample database from 2007-2016, we identified patients with a primary or secondary discharge diagnosis of IBD, with or without IBS, using ICD-9 and ICD-10 codes. We extracted information on demographics, psychological comorbidities, IBD complications, cost and duration of stay of each group, from either discharge records or diagnosis codes. These were analyzed using SAS version 4.0.
There was a rise in the prevalence of IBS among inpatients with ulcerative colitis (P=0.025) and Crohn’s disease (P=0.0014) over the study period. This study revealed that IBD patients with IBS tend to be female, younger, are less likely to be morbidly obese and have higher rates of psychological disorders (P<0.001) compared to IBD patients with no IBS co-diagnosis. They also have fewer IBD-specific complications, such as strictures, obstruction, fistula and abdominal abscess (P<0.001). Shorter hospital stays (P<0.001) and lower hospital charges (P<0.001) were also noted in these patients.
IBD patients with IBS are significantly different from other IBD patients, and are associated with less severe disease, a shorter hospital stay and lower hospital expenses. Early and accurate classification of this patient population may prevent unnecessary treatment and hospitalization in the future.
Inflammatory bowel disease (IBD), which includes ulcerative colitis (UC) and Crohn’s disease (CD), is characterized by inflammation or ulceration in the intestinal tract from the activation of the mucosal immune system [1,2]. Irritable bowel syndrome (IBS), on the other hand, is a functional gut disorder with an altered gut-brain axis causing recurrent abdominal pain and changing bowel habits [3]. Despite the fact that they are separate disease entities, there is evidence that these 2 conditions frequently overlap. From 25-46% of IBD patients in remission suffer from IBS-like symptoms, a higher prevalence compared to the general population without IBD [4,5]. The association between IBS and IBD has been extensively investigated for decades, but the etiology of IBS-type symptoms in patients with IBD is a source of ongoing research [6]. Some have suggested that IBS and IBD represent a spectrum of disease, and that IBS could represent subclinical inflammation and a prodrome of IBD; evidence to support this includes inflammation with elevated colonic mucosal inflammatory cell infiltrate, and mucosal tumor necrosis factor-a mRNA protein expression in patients with IBS symptoms [7,8]. Others have argued that chronic inflammation from IBD may progress to IBS-like symptoms by changing the enteric nervous system and intestinal wall, which consequently leads to alteration of motility and visceral hypersensitivity [9,10]. Yet another argument considers IBS and IBD to be unrelated entities, given the poor correlation between IBD disease activity and IBS-like symptoms [11]. Because of the overlap of symptoms in IBD and IBS, the diagnosis of IBS is only made when patients have persistent gastrointestinal (GI) symptoms in the setting of clinical and endoscopic remission of IBD. The current American Gastroenterological Association guideline recommends using fecal calprotectin, endoscopy with biopsy, and cross-sectional imaging to rule out active inflammatory activity and determine remission in IBD patients. The standard of care of IBS in IBD patients does not differ from the current treatments for IBS, consisting of a combination of dietary changes, psychological therapy, antispasmodic, laxative or hypomotility agents, and probiotics [12]. The economic impact of IBD patients is large. A study by Xu et al revealed that the annual estimated costs of IBD hospitalization are $11,345 for CD and $13,412 for UC, and the total costs has been annually increasing by 3-4% for both CD and UC, of which a certain percentage may be due to IBS overlap and potentially avoidable [13]. As of yet, there has not been a large inpatient epidemiology study that closely looks at IBD patients with an additional IBS diagnosis. The aim of our study was to assess the prevalence and describe the demographic, complications, comorbidities, and hospital utilization of the IBD inpatient population with a co-diagnosis of IBS, in order to better understand this subgroup.
This study cohort was obtained from the Nationwide Inpatient Sample database from 2007-2016 [14]. That database contains discharge records from over 7 million hospital visits representative of discharges from a random 20% stratified sample of United States (US) hospitals per year. It provides a useful representation of national US statistics and includes patient demographics, admission status, up to 30 primary and secondary discharge diagnoses, and 15 procedures coded using ICD-9 and ICD-10. Because of the de-identified nature of this publicly available data, our study did not require Institutional Review Board approval.
All subjects aged 18 years or older with a primary or secondary discharge diagnosis of IBD (including UC and CD), according to ICD-9 and ICD-10 codes, were identified. The diagnosis codes used for UC were 5560 to 5569, K5180, K5120, K5130, K5140, K5150, K5100, K5180 and K5190, while the diagnosis codes used for CD were 5550, 5551, 5552, 5559, K5000, K5010, K5080 and K5090. They were divided into 2 groups based on the presence or absence of IBS codes, including 5641, K58, K580, K581, K582, K588, K589. The sex, age, body mass index (BMI), race, insurance, hospital location, psychological comorbidities and IBD complications were directly collected through either discharge records or diagnosis codes.
Descriptive statistics, such as total numbers, means, frequencies and percentages, were calculated using appropriate procedures. Categorical and continuous variables were compared using the chi-square test or a 2-group t-test, respectively. A 2-tailed P-value of 0.05 was considered as statistical significance. Standardized residuals were calculated for chi-square values. The yearly prevalence was analyzed using Pearson’s correlation coefficient. All analyses were performed using the SAS software (version 4.0, Cary, North Carolina).
A total of 577,576 discharges of patients with a diagnosis of UC or CD hospitalized across the US based on the Nationwide Inpatient Sample database from 2007-2016 were included in the study. There were 212,318 discharges of UC patients, of which 6412 (3.0%) also had a co-diagnosis of IBS. Similarly, of 365,258 discharges of CD patients, 10,717 (2.9%) had IBS as a co-diagnosis. Overall, there was an annual rise in the prevalence of IBS in inpatient IBD patients over the study period (P=0.025 for UC and P=0.0014 for CD), as shown in Fig. 1. The demographic characteristics, length of stay and hospital charges per admission are summarized in Table 1. Complications and comorbidities of IBD-IBS vs. IBD groups are summarized in Tables 2 and 3, respectively. Types of admission, primary payer, and hospital location are summarized in Supplementary Table 1. As shown in Table 1, most patients with IBS in the UC and CD cohorts were young, female, Caucasian and obese. When the IBD-IBS group and IBD group were compared side by side, the IBD-IBS patient was more likely to be female (P<0.001) and younger (P<0.001) than the IBD patient. The IBD-IBS group was more likely to be of Caucasian race, while less likely to be Black or Other, compared to the IBD group (P<0.001). Most patients in the study were morbidly obese, with a BMI of 40 kg/m2 or greater (60.0-71.5%); however, when the 2 groups were compared the IBD-IBS group had overall lower BMI (P<0.001) compared to the IBD group. IBD patients who had IBS had lower hospital expenses ($40418 vs. $48414 for UC and $33713 vs. $39018 for CD, P<0.001) and shorter hospital stays (5.6 vs. 6.0 for UC and 5.1 vs. 5.3 for CD, P<0.001) than patients who had IBD alone. The IBD-IBS group had a lower prevalence of complicated IBD. In particular, complications specific to CD, such as fistulation, strictures, intra-abdominal abscess, perianal disease and malnutrition, were significantly less common in the CD-IBS group compared to the CD group (Table 2). Among comorbidities, anxiety, depression, bipolar disorder and opioid abuse were significantly more frequent in the IBD-IBS group than the IBD group (Table 3). While 17.1-21.1% of the IBD-IBS group had anxiety, only 9.7-11.8% of the IBD group had the diagnosis of anxiety.
This is the first large-scale study that has attempted to determine the frequency and study the epidemiology of IBS among hospitalized IBD patients. In this national cohort of hospitalized patients, the estimated age-adjusted frequency of IBS among IBD patients increased over the years, which underlines the importance of understanding this subpopulation (Fig. 1). In our study, the average prevalence of IBS co-diagnosis in hospitalized IBD patients was 3%. As of today, there has been no study that evaluated the incidence and prevalence trend of IBS in IBD inpatients. However, a recent meta-analysis that assessed IBS-like symptoms in a non-hospitalized IBD population stated that the prevalence was about 35% [5]. Our theory regarding the large disparity between the rates of IBS in inpatients vs. outpatients with IBD is that IBS-IBD patients tend to have relatively low symptom severity, which does not lead to frequent hospital admission. Our data seem to further support this theory, showing a shorter length of inpatient stay, a lower hospital cost, a less complicated IBD disease course, and fewer elective admissions in IBD-IBS patients. Our findings are similar to those reported by Gracie et al in 2018, that hospitalizations due to disease activity were significantly fewer in IBD-IBS patients compared to IBD patients without IBS [15]. Another possible explanation of the shorter hospital stays and lower costs in the IBD-IBS group, despite the overall increasing frequency of hospitalization, could be that these patients are hospitalized to rule out possible IBD flare, but are soon discharged with an unremarkable inflammatory laboratory workup and without undergoing GI procedures, which can be both costly and time consuming. No prior studies have revealed any significant age difference between IBD-IBS and IBD patients [16]. However, studies that attempted to investigate any association of age with IBS diagnosis among IBD patients found that diagnosis of IBD at a younger age increased the risk of developing IBS [17,18]. We also found that, while patients in both groups were mostly obese, IBS patients had relatively lower body weights. Recent studies have shown a positive association between BMI and symptom severity in both IBS and IBD, and patients with a BMI >30 kg/m2 experienced more frequent flare ups requiring hospitalizations, which is consistent with our inpatient demographic data of both groups [19,20]. The IBD-IBS group had a lower prevalence of complicated IBD, which was anticipated as IBS in IBD patients is diagnosed by ruling out active inflammation. CD patients with a co-diagnosis of IBS were less likely to have complications secondary to CD, such as bowel obstruction, strictures, fistulas, intra-abdominal abscess, perianal disease and malnourishment (Table 2). Based on our analysis of the demographics, comorbidities and complications of this population, the IBD-IBS group differs significantly from the IBD-alone group, which supports the theory that they may be mutually exclusive [11]. The IBD-IBS group had higher rates of major depressive disorder, bipolar disease and opioid use disorder than the IBD group. The current literature suggests that anxiety is more common, and the quality of life significantly lower, in IBD-IBS patients than in IBD patients [21,22]. More importantly, generalized anxiety disorder was found to be twice as prevalent in the IBD-IBS group. Considering that the prevalence of anxiety disorder and depressive disorder for IBS and IBD are similar, we can hypothesize that the higher rates of anxiety and depressive disorder in our inpatient IBD-IBS group may have been due to cumulative psychological effects from both IBD and IBS [23]. The strong association of depression and anxiety with both IBS and IBD has been well established in the literature, as IBS is a functional disease that is influenced by psychological factors, whereas IBD is a disease of autoimmune inflammation with psychiatric problems developing as a consequence and extraintestinal manifestation [24,25]. Studies have shown that IBD patients are at increased risk of developing anxiety and/or depression, and that these patients have more severe disease than those without anxiety and depression [26,27]. Further studies that examine psychological risk factors in the IBD-IBS group would be helpful to evaluate potential contributing factors to the high prevalence of psychological disorder in these patients. Understanding and identifying the IBS-IBD population could alert physicians to the need for a more appropriate means of workup, including reassurance for patients, psychological counseling, and treatment of any underlying psychiatric disorder. Consequently, appropriate management of these patients could reduce the exposure of IBS patients to unnecessary hospitalization and surgical interventions; a study by Perera et al revealed that IBD patients with IBS had greater healthcare utilization, such as frequent hospital visits, and surgeries [17, 28-30]. A limitation of our study is that as a large-scale retrospective study, the use of ICD codes to select the study population and collect data is inevitably less reliable than a small prospective study that has less room for study errors. The large inpatient dataset is a great dataset that allows us to study the inpatient prevalence and epidemiology of this population in the US; however, given it is a large cohort study, we are unable to carry out individual chart reviews to verify the rationale or the timeline for the diagnosis. For example, even though ICD codes for IBD-related adverse events were rare in IBD-IBS patients, this could be evidence for a study error, a possible misdiagnosis of IBS, or a mistaken ICD code as the discharge and billing diagnosis by a provider. Based on current guidelines, in order to diagnose IBS in IBD patients, imaging, endoscopy or stool biomarkers should be obtained to exclude active disease as the etiology of symptoms in a patient with IBD [12]. Educating providers regarding IBS and IBD would be necessary to reduce the error. To our knowledge, this is the first large inpatient population cohort study to assess the demographics, complications, comorbidities and hospital utilization of patients with a co-diagnosis of IBS and IBD. This study revealed that IBD patients with IBS tend to be female, younger, of lower weight, have higher rates of psychological disorders, and have less IBD complications compared to IBD patients without an IBS co-diagnosis. Shorter hospital stays and lower hospital charges were also noted in these patients. Our study highlights that these patients differ significantly from IBD-only patients, as IBS is correlated with fewer complications and a shorter length of stay, which indicates that this population has less severe disease. The study’s findings may help clinicians and patients gain new insight into our understanding of the disease entity, which will further increase our awareness and lead to better competency in the diagnosis and management of this population.
What is already known: Despite treatments for inflammatory bowel disease (IBD), patients in remission may still suffer from gastrointestinal symptoms attributable to overlyingirritable bowel syndrome (IBS) IBD patients with IBS had greater healthcare utilization, such as frequent hospital visits, and surgeries No one has yet explored the differences between IBD-IBS and IBD inpatients to better distinguish this patient population What the new findings are: The estimated age-adjusted frequency of IBS among IBD patients has increased over the years IBS-IBD patients tend to be female, younger and have lower weight, with more psychological disorders and fewer IBD-specific complications They also have shorter hospital stays and lower hospital charges
Click here for additional data file. |
|
PMC9648539 | Xi Xiao,Jianpeng Li,Shun Wan,Mingzhe Wu,Zonglin Li,Junqiang Tian,Jun Mi | A novel signature based on pyroptosis-related genes for predicting prognosis and treatment response in prostate cancer patients 10.3389/fgene.2022.1006151 | 27-10-2022 | prostate cancer,pyroptosis,tumor microenvironment,immune checkpoint inhibitor,treatment response | Background: Pyroptosis is a form of programmed cell death accompanied by specific inflammatory and immune responses, and it is closely related to the occurrence and progression of various cancers. However, the roles of pyroptosis-related genes (PRGs) in the prognosis, treatment response, and tumor microenvironment (TME) of prostate cancer (PCa) remain to be investigated. Methods: The mRNA expression data and clinical information of PCa patients were obtained from the Cancer Genome Atlas database (TCGA) and the cBioPortal for Cancer Genomics website, and the 52 PRGs were obtained from the published papers. The univariate, multivariate, and LASSO Cox regression algorithms were used to obtain prognostic hub PRGs. Meanwhile, qRT-PCR was used to validate the expression of hub genes between PCa lines and normal prostate epithelial cell lines. We then constructed and validated a risk model associated with the patient’s disease-free survival (DFS). Finally, the relationships between risk score and clinicopathological characteristics, tumor immune microenvironment, and drug treatment response of PCa were systematically analyzed. Results: A prognostic risk model was constructed with 6 hub PRGs (CHMP4C, GSDMB, NOD2, PLCG1, CYCS, GPX4), and patients were divided into high and low-risk groups by median risk score. The risk score was confirmed to be an independent prognostic factor for PCa in both the training and external validation sets. Patients in the high-risk group had a worse prognosis than those in the low-risk group, and they had more increased somatic mutations, higher immune cell infiltration and higher expression of immune checkpoint-related genes. Moreover, they were more sensitive to cell cycle-related chemotherapeutic drugs and might be more responsive to immunotherapy. Conclusion: In our study, pyroptosis played a significant role in the management of the prognosis and tumor microenvironment of PCa. Meanwhile, the established model might help to develop more effective individual treatment strategies. | A novel signature based on pyroptosis-related genes for predicting prognosis and treatment response in prostate cancer patients 10.3389/fgene.2022.1006151
Background: Pyroptosis is a form of programmed cell death accompanied by specific inflammatory and immune responses, and it is closely related to the occurrence and progression of various cancers. However, the roles of pyroptosis-related genes (PRGs) in the prognosis, treatment response, and tumor microenvironment (TME) of prostate cancer (PCa) remain to be investigated. Methods: The mRNA expression data and clinical information of PCa patients were obtained from the Cancer Genome Atlas database (TCGA) and the cBioPortal for Cancer Genomics website, and the 52 PRGs were obtained from the published papers. The univariate, multivariate, and LASSO Cox regression algorithms were used to obtain prognostic hub PRGs. Meanwhile, qRT-PCR was used to validate the expression of hub genes between PCa lines and normal prostate epithelial cell lines. We then constructed and validated a risk model associated with the patient’s disease-free survival (DFS). Finally, the relationships between risk score and clinicopathological characteristics, tumor immune microenvironment, and drug treatment response of PCa were systematically analyzed. Results: A prognostic risk model was constructed with 6 hub PRGs (CHMP4C, GSDMB, NOD2, PLCG1, CYCS, GPX4), and patients were divided into high and low-risk groups by median risk score. The risk score was confirmed to be an independent prognostic factor for PCa in both the training and external validation sets. Patients in the high-risk group had a worse prognosis than those in the low-risk group, and they had more increased somatic mutations, higher immune cell infiltration and higher expression of immune checkpoint-related genes. Moreover, they were more sensitive to cell cycle-related chemotherapeutic drugs and might be more responsive to immunotherapy. Conclusion: In our study, pyroptosis played a significant role in the management of the prognosis and tumor microenvironment of PCa. Meanwhile, the established model might help to develop more effective individual treatment strategies.
Prostate cancer (PCa) is the world’s second most frequent male malignancy, and it causes significant health problems for men (Sung et al., 2021). In the United States, the number of new cases in 2021 is expected to be around 248,530, with around 34,130 fatalities (Siegel et al., 2021). Although PCa has a higher overall survival rate than some other cancers, it has a very high recurrence rate. Many patients will experience disease progression and eventually develop castration-resistant prostate cancer (CRPC), which is incurable and may become drug resistant (De Angelis et al., 2014; Fujita and Nonomura, 2019; Howard et al., 2019). Certainly, Individualized chemotherapy and immunotherapy have a good prospect of promise for improving the prognosis of PCa patients (Dudzinski et al., 2019). However, immunotherapy has a low response rate in unselected PCa patients (Sandhu et al., 2021). Fortunately, genetic testing is becoming increasingly beneficial for treating patients with PCa (Merseburger et al., 2021). That is, because identification of target genes can guide patients to assess cancer risk, conduct, precision medicine treatment (such as individualized chemotherapy and immunotherapy), and manage disease prognosis (Giri et al., 2018). Therefore, further studies into the molecular mechanisms of PCa, and the development of effective biomarkers, are required to improve patient prognosis and quality of life. Pyroptosis is a novel mechanism of programmed cell death triggered by some inflammasomes. Pyroptosis causes cell swelling, plasma membrane lysis, chromatin breakage, and cell content release via particular pathways, resulting in a potent inflammatory response. Pyroptotic cells are unique in maintaining nuclear integrity (Shi et al., 2015; Ding et al., 2016; Kovacs and Miao, 2017; Fang et al., 2020). Generally, there are three pathways to activate pyroptosis: the canonical pathway, the noncanonical pathway, and a new-found pathway. In the canonical pathway, some inflammasomes recruit and bind to apoptosis-associated speck-like protein containing a caspase recruitment domain (ASC), resulting in the formation of the ASC complex which recruits procaspase-1 and activates caspase-1. Caspase-1 is involved in the cleavage and maturation of proIL-18/1β, as well as the cleavage of gasderminD (GSDMD). The released N-terminal fragment of GSDMD (GSDMD-NT) causes pore formation in the plasma membrane, leading to secretion of IL-18/1β and water influx, which results in cell swelling and osmotic lysis (Liu et al., 2016; Fang et al., 2020). In the noncanonical pathway, bacterial-derived lipopolysaccharide (LPS) recognizes and activates caspase-4/5/11 to induce pyroptosis by cleaving GSDMD (Khanova et al., 2018; Rathinam et al., 2019). The new-found pathway is achieved by the cleavage of gasderminE (GSDME), which depends on the activation and participation of caspase-3 (Rogers et al., 2017; Wang et al., 2017). Pyroptosis appears to play a significant role in tumor progression and is linked to proliferation, migration, cell cycle, and treatment resistance in various of cancers, according to accumulated evidence (Heo et al., 2019; Yu et al., 2019; Tan et al., 2021). Recent studies have found that pyroptosis-related genes (PRGs) have satisfactory predictive abilities in the prognosis of PCa and could be used as novel tumor biomarkers (Fu et al., 2022; Hu et al., 2022; Wang et al., 2022). Meanwhile, its relationship with PCa immunity may provide assistance in the treatment of PCa (Li et al., 2022; Zhang et al., 2022). However, systematic evaluation of the relationship between differentially expressed PRGs and the prognosis, immune microenvironment, and treatment response of PCa is still worth further exploration. Therefore, our study aims to develop a novel prognostic signature based on PRGs to systematically explore the relationships between the signature and clinicopathological characteristics and disease progression in PCa patients. In addition, we further investigated its correlation with the tumor microenvironment (TME), mutation profiles, and the patient’s response to immunotherapy and chemotherapy in PCa. This study provides new insights into the role of pyroptosis in PCa.
Gene expression data (FPKM value) for 495 prostate cancer samples and 52 normal samples were obtained from the TCGA official website (https://portal.gdc.cancer.gov/). The log2 transformation is used to normalize the TCGA-PRAD cohort. The clinical information for TCGA-PRAD was obtained from the cBioPortal for Cancer Genomics website (http://www.cbioportal.org/), as were the gene expression data and clinical information for the MSKCC/GSE21032 dataset. Patients who did not have survival information were excluded from our analysis. The clinical information of patients was shown in Supplementary Table S1. PRGs were gathered from the Molecular Signatures Database (MSigDB) (http://www.gsea-msigdb.org/gsea/msigdb/search.jsp) and previous reports (Liberzon et al., 2015; Wu et al., 2021a). We got a total gene set of 52 PRGs after deleting duplicate genes, found in Supplementary Table S2.
First, we used the R package “limma” to investigate the differential expression of PRGs between PCa tissues and adjacent nontumorous samples (Ritchie et al., 2015), and then we created a heat map with the R package “pheatmap” and a bar graph with the R packages “ggplot2” and “ggpubr” (Kolde, 2019). The “spearman” method was used to calculate the correlation coefficients of the differentially expressed pyroptosis-related genes (DE-PRGs) in PCa, and correlation plots were created using the R package “corrplot” (Wei and Simko, 2017). The STRING website (https://cn.string-db.org/) was used to calculate and generate the interaction network of DE-PRGs. Additionally, based on the DE-PRGs, we utilized the R package “ConsensusClusterPlus” for unsupervised clustering analysis of PCa samples (Wilkerson and Hayes, 2010), as well as the R package “survival” for survival analysis, to see whether the DE-PRGs were associated with patient differences (Therneau, 2020). For DE-PRGs, we utilized univariate Cox regression analysis to screen for genes associated with disease-free survival (DFS), and p < 0.05 was considered the cut-off value. LASSO regression was applied to lessen the risk of overfitting by R package “glmnet” (Simon et al., 2011). Finally, the multivariate stepwise Cox regression analysis was used to identify the hub genes, which were most associated with the prognosis of PCa.
We obtained three PCa cell lines (LNCap, PC3, DU-145) cultured in RP1640 medium (Gibco) and one normal prostate epithelial cell line (RWPE-1) cultured in DMEM medium (Gibco) from the Second Hospital of Lanzhou University. Meanwhile all cells were cultured in a humidified incubator at 37°C and 5% CO2 with 10% fetal bovine serum added to every medium. Then we extracted the total RNA from the cells using TRIzol (AG21101; Hunan, China) reagent according to the manufacturer’s instructions, followed by reverse transcription. In addition, we measured the mRNA relative expression levels of the hub genes by real-time quantitative PCR, which were quantified by 2–ΔΔCT. The primer sequences of the hub genes and the internal reference gene could be found in Supplementary Table S3. Finally, we obtained immunohistochemistry (IHC) correlation data of hub genes from the Human Protein Atlas (HPA) (https://www.proteinatlas.org/) and further validated them by the protein expression levels of the genes (Uhlen et al., 2017).
We utilized the training set (TCGA cohort) and the validation set (MSKCC cohort) to construct and validate the risk model, and both datasets calculated the risk score according to the formula: (expgene1 × coefgene1) + (expgene2 × coefgene2) + (expgene3 × coefgene3) +(expgene4 × coefgene4) +(expgene5 × coefgene5) +(expgene6 × coefgene6). The median risk score was the cut-off value to separate patients into high and low risk groups. Kaplan-Meier (KM) survival analysis with log-rank test and time-dependent subject work characteristics (ROC) analysis were used to assess the risk model’s correctness. We then utilized univariate and multivariate analyses to explore whether the risk score compared to clinicopathological characteristics of PCa was an independent prognostic factor. In addition, Wilcoxon and Kruskal-Wallis tests were used to examine the relationship between risk score and clinicopathological characteristics of PCa (age, T-stage, N-stage, Gleason score, and PSA value).
Based on the independent prognostic factor risk score and Gleason score, we employed the R packages “rms” (Harrell, 2021) and “survival” (Therneau, 2020) to generate a nomogram to forecast the probability of DFS at 1, 3, and 5 years, and estimated the nomogram prediction scores for each patient. To evaluate the accuracy of the nomogram, we utilized the “calibration” function of the R package “rms” for calibration curve analysis and the R package “timeROC” for ROC analysis (Blanche et al., 2013; Harrell, 2021).
To better elucidate the biological function of FRGs in PCa, we obtained EDGs between high and low risk groups by the R package “limma” using p-value < 0.05 and log2 foldchange (log2FC) > 0.585 (Ritchie et al., 2015). The Gene Ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were then carried out using the R packages “clusterProfiler” and “org.Hs.eg.DB,” with a critical value of p < 0.05 (Carlson et al., 2019; Wu et al., 2021b).
The TME is comprised of tumor cells and non-tumor components such as blood vessels, immune cells, adipocytes, and tumor-associated fibroblasts (Binnewies et al., 2018). Hence, we analyzed the infiltration of immune cells in PCa samples by single sample gene set enrichment analysis (ssGSEA) using the R package “GSVA” (Hänzelmann et al., 2013) and the correlation of immune cells with risk score by the “Spearman” method using the R package “reshape2” (Wickham, 2007), and visualized by the R package “ggplot2” (Wickham, 2016). Then, we used the R package " estimate " to perform stromal score, immune score, and estimate score of PCa samples and to elucidate their relationship with high and low risk groups (Kosuke et al., 2016). The tumor mutation burden (TMB) data of PCa samples was collected from the TCGA database. The samples were separated into two groups based on the risk model, and the TMB score was computed using the R package “maftools” and displayed as a waterfall chart via the R package “ggplot2” (Wickham, 2016; Mayakonda et al., 2018). We also used the R package “reshape2″ to examine the link between risk score and TMB, followed by survival analysis using the R packages “survival” and “survminer” (Kassambara et al., 2021).
The different expression of common immune checkpoint-related genes in high and low-risk score groups was achieved by the Wilcoxon test, and the “spearman” method was used to determine the correlation between immune checkpoint-related genes and risk score using the R package “reshape2” (Wickham, 2007), and visualized using the R package “ggplot2” (Wickham, 2016). In previous studies, the Immunophenoscore (IPS) was used to predict tumor response to immunotherapy with CTLA-4 and PD-1 blockers (Charoentong et al., 2017). Furthermore, we used the Wilcoxon test to compare IPS in high and low-risk groups after downloading IPS data for PCa from the Cancer Immunome Atlas (TCIA) (https://TCIA.at/home). In addition, the sensitivity of prostate cancer patients in high and low-risk groups to commonly used cell cycle chemotherapy drugs was computed using the R package “pRRophetic”, which was based on the Genomics of Drug Sensitivity in Cancer (GDSC, https://www.cancerrxgene.org) database (Geeleher et al., 2014).
R software (version 4.1.2) and GraphPad Prism (version 9.0) were used for data analysis, statistics, and graphs in this study. The hub genes were discovered by univariate Cox regression, LASSO regression, and multivariate stepwise Cox regression analysis on DE-PRGs. The Wilcoxon test, Kruskal-Wallis test, and Dunnett’s test were used to compare differences between two or more groups as appropriate. The “Spearman” or “Pearson” approach was used to explore the relationship between distinct variables. The log-rank test of Kaplan-Meier analysis was used to perform the survival analysis. The above statistical methods produced significant results at p < 0.05.
First, we analyzed the expression of 52 FRGs in 495 tumor samples and 52 normal samples from the TCGA cohort, finding that 35 PRGs were expressed differently in normal and tumor tissues. From the heat map and boxplot, it can be seen that 12 genes, BAK1, CASP6, CYCS, PLCG1, TP53, CHMP2A, CASP8, GPX4, BAX, CHMP4C, GSDMB, and GSDMA, are highly expressed in tumor tissues, and the remaining 23 genes are highly expressed in normal tissues (Figures 1A,B). The PPI analysis revealed that these 35 PRGs had abundant interactions (Figure 1C). Meanwhile, the correlation analysis of these 35 genes in the TCGA cohort showed that they had a high correlation, such as GPX4 and CHMP2A (Figure 1D). Furthermore, we used these 35 genes to divide the TCGA cohort into two clusters (Figure 1E) and performed survival analysis, finding that patients in cluster 2 had a worse DFS (Figure 1F), implying a solid link between PRGs and patient differences. As a result, it was necessary for us to investigate the prognostic PRGs further.
The univariate Cox regression analysis was used to analyze the above 35 DE-PRGs, and 13 genes were found to be associated with the DFS of PCa (Figure 2A). We then performed LASSO regression analysis with tenfold cross-validation on these 13 genes to mitigate the overfitting effect (Figure 2B). Subsequently, we performed multivariate stepwise Cox regression analysis to find 6 hub genes with the best prognostic value (Figure 2C). Before establishing the prognostic model, we conducted individual survival analyses on these 6 genes and discovered that genes with low expression had superior DFS (Figure 2D). In addition, we used qRT-PCR to compare the expression of hub genes (CHMP4C, GSDMB, NOD2, PLCG1, CYCS, GPX4) between prostate cancer cell lines (LNCap, PC3, DU-145) and the normal prostate epithelial cell line (RWPE-1). The results showed that, in comparison to RWPE-1, the 6 hub genes were generally more highly expressed in LNCaP, PC-3, and DU-145 cells (Figures 3A–F). Meanwhile, the IHC data obtained from HPA showed that the protein expression levels of the six genes were also higher in the prostate tumor tissues (Figures 4A–F). Therefore, we established a prognostic model using these six genes. Risk score = (0.2985 * expCHMP4C) + (0.5625 * expCYCS) + (0.6243 * expGPX4) + (0.3102 * expGSDMB) + (1.0209 * expNOD2) + (0.9242 * expPLCG1). Then, we divided PCa patients from the TCGA cohorts into high and low risk groups, with the median risk score as the cut-off value (Figure 5A). As shown in Figure 5B, PCa patients in the high-risk group had a higher likelihood of disease progression, which occurred earlier. According to the KM survival analysis, patients in the low-risk group had a better DFS than those in the high-risk group (Figure 5C). Furthermore, the area under the receiver operating characteristic curve (AUC) of the 1, 3, and 5-year DFS for the TCGA cohort was 0.685, 0.735, and 0.729 (Figure 5D), respectively, demonstrating that our risk models have a relatively high degree of accuracy. The heat map and box plot showed that all six hub genes had higher expression in the high-risk group of patients than in the low-risk group (Figures 5E,F). In the validation set (MSKCC cohort), the risk score was generated using the same formula, and PCa patients were classified into two groups: high and low risk, with the median risk score (Figure 6A). Although the expression of hub genes in the high and low-risk groups was slightly different from the TCGA cohort, the overall tendency was for the high-risk group to have higher gene expression (Figures 6E,F). Consistent with the results of the TCGA cohort, PCa patients in the high-risk group in the MSKCC cohort also had faster disease progression (Figure 6B). The results of KM survival analysis showed that patients in the low-risk group had a more favorable DFS (Figure 6C). And the AUC for 1, 3, and 5-year DFS were 0.734, 0.645, and 0.619, respectively (Figure 6D). These results indicated that the risk model in the MSKCC cohort could also play an important role in the prognosis of PCa.
To evaluate the prognostic significance of different clinical features of PCa patients and to evaluate whether the risk model can be used as an independent prognostic factor for PCa, we used univariate and multivariate Cox regression analyses on risk scores and different clinical features of PCa patients in the TCGA and MSKCC cohorts, respectively. In the TCGA cohort, the risk score and the Gleason score had p < 0.05 (Figures 7A,B) in both univariate and multivariate analyses, indicating that they were both independent prognostic factors for PCa. Meanwhile, they got the same results in the validation set (MSKCC cohort) (Figures 7C,D). Furthermore, we evaluated the relationship between risk score and clinical features of PCa, finding that patients older than 55 years old had a higher risk score than patients younger than 55 years old, and that the risk score of patients increased as Gleason score, T-stage, and N-stage increased (Figures 8A–E). These results were also verified in the MSKCC cohort (Figures 8F–I).
We utilized the risk score and Gleason score to construct a prognosis nomogram based on the TCGA cohort because they are independent prognostic factors for PCa. The sample “TCGA-KK-A6E6” was chosen as an example simultaneously. The result showed that this patient’s probability of disease recurrence was 8.57%, 24.6%, and 36.4% at 1, 3, and 5 years, respectively (Figure 9A). Furthermore, the 1, 3, and 5-year calibration curves in the TCGA and MSKCC cohorts were all near the standard curve (Figures 9B,C). The AUC of the time-dependent ROC of the nomogram at 1, 3, and 5-year were all greater than 0.75 (Figures 9D,E), indicating that the prognostic nomogram we developed has high accuracy and validates its utility in predicting patient prognosis.
The volcano diagram shows that there are 856 DEGs between high and low risk groups (FDR <0.05, |log2FC| ≥ 0.585), 684 of which are up-regulated genes and 172 of which are down-regulated genes (Figure 10A). The functional enrichment of these 856 genes was then performed using GO and KEGG enrichment analysis. Nuclear division, mitotic nuclear division, chromosomal segregation, mitotic sister chromatid segregation, and other cell cycle-related functions were mostly represented in the GO enrichment analysis (Figure 10B). In addition, KEGG enrichment analysis suggested that the genes were mainly associated with the cell cycle, cytokine-cytokine receptor interaction, ECM-receptor interaction, primary immunodeficiency. (Figure 10C).
According to the results of functional enrichment, the risk score was closely related to the cell cycle process, extracellular matrix, and cytokines. These factors play essential roles in the tumor microenvironment, tumor genetic alterations, and the treatment of tumors (Quail and Joyce, 2013). The following study discovered that the high-risk group had a larger infiltration of immune cells in the TCGA cohort (Figure 11A). The results of correlation analysis showed that the risk score was significantly positively correlated with the activated CD8 T cell, CD56dim natural killer cell, effector memory CD8 T cell, activated CD4 T cell, myeloid derived suppressor cell, regulatory T cell, plasmacytoid dendritic cell and macrophage. And the risk score was significantly negatively correlated with neutrophil, monocyte, mast cell and type 17 T helper cell (Figure 11C). Furthermore, the stromal score, immune score, and estimate score all exhibited higher expression in the high-risk group (Figure 11B), indicating that the tumor and non-tumor components of PCa in the high-risk group had a more complex relationship. We further analyzed the TMB of PCa, finding that there was a significant difference in TMB score between high and low-risk groups, with the high-risk group having the higher score (Figure 12A), and correlation analysis also revealed that risk score increased with increasing TMB score (Figure 12B). Additionally, according to the optimal TMB threshold, PCa patients were separated into two groups: H-TMB and L-TMB, and the results of survival analysis revealed that patients in the low TMB group had a better DFS (Figure 12C). Similarly, when we combined the TMB and risk score groups, we found that the L-TMB + low-risk group had the best DFS (Figure 12D). Furthermore, no significant relationship was found between the risk score and mRNAsi (Figures 12E,F). Finally, there is a distinction between the high and low-risk groups in terms of the tumor somatic mutation. The overall mutation rate in the high-risk group is higher (63.25%) than in the low-risk group (52.36%). The mutation rate of “TP53” is highest in the high-risk group, while “SPOP” is highest in the low-risk group (Figures 12G,H).
Immunotherapy for tumors has entered a new era with the continuous development of immune checkpoint and chimeric antigen receptor (CAR) T cell therapies (Yang, 2015). The immune checkpoint blocking therapy was crucial in the immunotherapy of some malignancies (Grapin et al., 2019). We analyzed the association of PCa immune checkpoint-related genes PD-1 (PDCD1), PD-L1 (CD274), CTLA4, PD-L2 (PDCD1LG2), IDO1, and VTCN1 with the risk score and hub genes (Figure 13A) and discovered that PD-1, CTLA4, and IDO1 were highly expressed in the high-risk group (Figure 13B), and the risk score was significantly positively correlated with PD-1, CTLA4, and IDO1 (Figure 13C). Furthermore, Figure 13A showed that NOD2 was the hub gene with the strongest association to immune checkpoint-related genes, and NOD2 was significantly positively connected to these 6 genes (Figure 13D). Subsequently, we further downloaded IPS for PCa from the TCIA database. We analyzed the relationship between IPS and high and low-risk groups, finding that the four components of negative or positive responses for PD1 and CTLA4 were not significantly different in high and low-risk groups (Figure 14A). Fortunately, patients with high NOD2 expression had the higher IPS than those with low expression (Figure 14B). The risk score was closely related to the cell cycle progression of PCa according to the above functional enrichment analysis, and we further analyzed the response of PCa patients in the TCGA cohort to eight common cell cycle-related chemotherapy drugs (Docetaxel, Gemcitabine, Paclitaxel, Doxorubicin, Cisplatin, Etoposide, Mitomycin, and Methotrexate). The results revealed that these drugs had lower half maximal (50%) inhibitory concentration (IC50) in patients of the high-risk group (Figure 14C), implying that these patients may be more sensitive to these drugs.
PCa is a common male urological malignancy. In Asia, the 5-year survival rate for PCa is above 60% (Hassanipour et al., 2020). Between 2001 and 2016 in the United States, the 10-year survival rate for localized stage PCa approached 100% (Siegel et al., 2020). However, a large proportion of PCa patients might experience disease progression, even to the CRPC stage, which increases the risk of PCa-specific death. From 2011 to 2016, the 5-year survival rate for distant stage PCa in the United States was only 32.3% (Siegel et al., 2020). Therefore, there is an urgent need to identify novel prognostic signatures for PCa to improve precise treatment and health management. Pyroptosis, a new type of programmed cell death that involves the release of inflammatory factors and some immunological responses, is closely related to the occurrence and development of tumors (Du et al., 2021). At present, much research has explored the role of pyroptosis in various tumors, establishing some effective models for predicting prognosis and treatment response and analyzing the potential role of pyroptosis in the tumor microenvironment (Wu et al., 2021a; Li et al., 2021; Shao et al., 2021). A recent study explored the correlation between pyroptosis and PCa patients, resulting in a new signature for predicting PCa patients’ prognosis (Hu et al., 2022). However, the relationship between the members of PRGs and PCa still remains worthy of research. In this study, we first obtained 35 PRGs that were differentially expressed between tumor and normal tissues in the TCGA-PRAD cohort. Following that, six hub genes (CHMP4C, NOD2, GSDMB, PLCG1, GPX4, CYCS) were found to be strongly associated with the DFS of PCa using univariate cox regression, LASSO regression, and multivariate stepwise Cox regression analysis. In the TCGA-PRAD cohort, we established a risk model of PRGs using these six hub genes, and patients were separated into high and low-risk groups based on their median risk score, with the patients in the high-risk group being found to be more likely to experience a worse DFS. These findings were validated in the MSKCC external validation dataset. According to previous research, these hub genes are closely related to the occurrence and development of various diseases. CHMP4C, an ESCRT-III subunit, is involved in the abscission checkpoint (NoCut) in response to mitotic problems. Dysregulation of abscission by CHMP4C may act in concert with oncogene-induced mitotic stress to promote genomic instability and tumorigenesis (Sadler et al., 2018). It has been reported that CHMP4C may play an important role in aggressive prostate cancer and may be a potential therapeutic target (Fujita et al., 2017). NOD2 is an intracellular pattern recognition receptor that senses bacterial peptidoglycan conserved motifs in the cytosol and stimulates the host immune response (Ferrand et al., 2019). It has been reported that NOD2 has been linked to the innate immune response of prostate epithelial cells and the occurrence and progression of prostate cancer (Kang et al., 2012). GSDMB, a member of the Gasdermin family, is a downstream effector protein in the pyroptosis pathway (Li et al., 2020), and it has been related to the development of bladder and stomach malignancies in multiple studies (Zhou et al., 2020; He et al., 2021). PLCG1 is a member of the phosphatidylinositol-specific phospholipase C (PLC) family that hydrolyzes phosphatidylinositol 4,5-bisphosphate (PIP2) to generate inositol 1,4,5-trisphosphate and diacylglycerol (DAG), which is associated with the proliferation and invasion of tumor cells. Aberrant expression and regulation of PLCG1 have been linked to the development of various cancers, including breast, lung, pancreatic, gastric, prostate, and ovarian cancers (Mandal et al., 2021). GPX4 is an enzyme that explicitly reduces phospholipid hydroperoxides to repair oxidative lipid damage (Gaschler and Stockwell, 2017). GPX4 is not only a negative regulator of ferroptosis and has been associated with numerous cancers (Riegman et al., 2020), but it also helps to attenuate lipid peroxidation, inflammasome activation, and pyroptosis in the context of sepsis (Kang et al., 2018). CYCS, or cytochrome c, has been implicated in numerous regulated cell death forms in addition to being an electron carrier in the mitochondrial respiratory chain (Bock and Tait, 2020), such as the release of cytochrome c into the cytoplasmic matrix upon stimulation by Bax to activate caspase-3, which leads to pyroptosis by triggering GSDME cleavage (Zhou et al., 2018). Meanwhile, a previous study indicated that cytochrome c may impact the sensitivity of the PCa cell line (PC3) to chemotherapeutic agents (Grayson et al., 2021). Therefore, these hub genes might be potential therapeutic targets for PCa. The following study investigated the association between risk score and clinicopathological characteristics, and discovered that the high-risk group had a higher degree of malignancy. Meanwhile, in our study, only the risk score and the Gleason score were independent prognostic factors for PCa, showing that the risk model had a strong prognostic value. Furthermore, a nomogram with two independent prognostic factors can assist clinicians in predicting patient prognosis and provide a more trustworthy reference for health management than a single routine clinical parameter. To explore the functional mechanisms of the risk model, we first obtained 856 differentially expressed genes between the high and low-risk groups, and functional enrichment analysis revealed that these genes were mainly closely related to cell cycle processes. And there were several cell cycle related drugs in chemotherapy for PCa, such as Docetaxel, Gemcitabine, Paclitaxel, Doxorubicin, Cisplatin, Etoposide, Mitomycin, and Methotrexate. We then calculated their estimated IC50s in different patients. The estimated IC50s of these drugs were all lower in the high-risk group than in the low-risk group, indicating that patients in the high-risk group were more sensitive to these drugs. There is mounting evidence that cell cycle processes are not only linked to tumor development (Liu et al., 2022), but also play a role in immune escape and immunotherapy (Bednarski and Sleckman, 2019). In the subsequent study, we discovered that the high-risk group had more immune cell expression than the low-risk group, and a majority of the immune cell infiltration was positively correlated with the risk score, suggesting that there may be more abundant immune effects in the high-risk group. Later, immunotherapy-related markers such as TMB, mRNAsi, and IPS were incorporated into further studies. The results showed that the risk score was positively correlated with the TMB score, and the total somatic mutation rate in the high-risk group (63.25%) was higher than that in the low-risk group (52.36%). However, there was no obvious link between risk scores and mRNAsi. In addition, although the immune checkpoint-related genes PD-1 and CTLA4 were significantly higher expression in the high-risk group, the IPS analysis revealed no significant difference in the response of patients in the high and low-risk groups to PD1 and CTLA4 immune checkpoint inhibitors. Fortunately, the hub gene NOD2 was significantly and positively correlated with the expression of immune checkpoint-related genes, and patients with high NOD2 expression also had the higher IPS than those with low expression. These results point to a complex relationship between the PRGs and the immune microenvironment of PCa, which could be helpful for future research into PCa immunotherapy, particularly the function of the hub genes. We constructed a risk model of PCa using PRGs and analyzed the relationship between the risk model and PCa from multiple perspectives, which may have good clinical significance. However, our study also has certain limitations. The sample size from the TCGA and MSKCC databases may not be sufficient and more data needs to be collected. At the same time, further in vitro experimental research and clinical trials are required to confirm our findings.
In conclusion, our study demonstrates that pyroptosis plays a vital role in PCa prognosis and that pyroptosis has some effects on the regulation of the TME in PCa. Meanwhile, we provide new insights into PCa prognostic research and assist in developing more effective individual treatment strategies. |
|
PMC9648559 | 36288798 | Ming Zhou,Tianzhen Wu,Yue Chen,Shixia Xu,Guang Yang | Functional Attenuation of UCP1 as the Potential Mechanism for a Thickened Blubber Layer in Cetaceans | 27-10-2022 | cetaceans,UCP1,blubber,thermoregulation,pseudogenization | Abstract Uncoupling protein 1 (UCP1) is an essential protein in the mitochondrial inner membrane that mediates nonshivering thermogenesis (NST) and plays an important role in thermoregulation and fat deposition. However, the relationship between the evolution of UCP1 and fat deposition in the blubber layer in cetaceans remains unclear. Here, frameshift mutations, premature termination, and relaxed selection pressure (ω = 0.9557, P < 0.05) were detected in UCP1 in cetaceans, suggesting that UCP1 was inactivated during cetacean evolution. By time estimation, it was found that the inactivation of UCP1 in cetaceans occurred between 53.1 and 50.2 Ma. However, combined with findings from immunohistochemical analysis of the blubber layer of the Yangtze finless porpoise and in vitro functional assays, a premature termination of cetacean UCP1 resulted in a reduction of UCP1-mediated NST capacity (about 50%) and lipolytic capacity (about 40%), both of which were beneficial to maintain blubber layer and body temperature without excessive fat consumption. This study provides new insights into the molecular mechanisms of the blubber thickening in cetaceans and highlights the importance of UCP1 attenuation in cetaceans for secondary aquatic adaptation. | Functional Attenuation of UCP1 as the Potential Mechanism for a Thickened Blubber Layer in Cetaceans
Uncoupling protein 1 (UCP1) is an essential protein in the mitochondrial inner membrane that mediates nonshivering thermogenesis (NST) and plays an important role in thermoregulation and fat deposition. However, the relationship between the evolution of UCP1 and fat deposition in the blubber layer in cetaceans remains unclear. Here, frameshift mutations, premature termination, and relaxed selection pressure (ω = 0.9557, P < 0.05) were detected in UCP1 in cetaceans, suggesting that UCP1 was inactivated during cetacean evolution. By time estimation, it was found that the inactivation of UCP1 in cetaceans occurred between 53.1 and 50.2 Ma. However, combined with findings from immunohistochemical analysis of the blubber layer of the Yangtze finless porpoise and in vitro functional assays, a premature termination of cetacean UCP1 resulted in a reduction of UCP1-mediated NST capacity (about 50%) and lipolytic capacity (about 40%), both of which were beneficial to maintain blubber layer and body temperature without excessive fat consumption. This study provides new insights into the molecular mechanisms of the blubber thickening in cetaceans and highlights the importance of UCP1 attenuation in cetaceans for secondary aquatic adaptation.
Cetaceans have evolved adaptively in morphology and physiology since their return from land to sea. The thermal conductivity of water is about 25 times that of air, and maintaining the central body temperature of cetaceans at 37 °C is a tremendous challenge (Parry 1949; Scholander et al. 1950; Wright et al. 2021). To cope with this challenge, the blubber layer is about 20–30 cm thick in cetaceans, more than ten times thicker than in other even-toed ungulates (Pond 1978; Struntz et al. 2004; Ellis 2009). The blubber is a reservoir of abundant fat and energy, playing an important role in ecology, reproduction, and survival, as well as being a major source of energy and freshwater balance during the fasting season (Ellis 2009). More critically, it effectively reduces heat conduction and heat loss, which are central to the entire process of thermoregulation in cetaceans, and acts as an insulator (Ellis 2009; Hashimoto et al. 2015). Therefore, it is particularly important for cetaceans to maintain adequate blubber thickness in order to adapt to the aquatic environment. However, the evolutionary mechanisms by which cetaceans effectively regulate and maintain their thickened blubber have not been well studied. In mammals, brown adipose tissue (BAT) is a specialized nonshivering thermogenesis (NST) tissue that consumes white adipocytes to increase heat production and regulate body temperature during cold stress (Golozoubova et al. 2001; Cannon and Nedergaard 2004). As a core protein mediating NST in BAT, UCP1 (uncoupling protein 1) plays an essential role in NST and is mainly expressed in the mitochondrial inner membrane (Nicholls and Locke 1984); when activated, UCP1 promotes proton leakage into the mitochondrial matrix (Heaton et al. 1978), but without a concomitant production of adenosine triphosphate (ATP), it would lead to a futile cycle of enhanced substrate oxidation and thermogenesis (Ledesma et al. 2002; Cedikova et al. 2016; Nowack et al. 2017). In addition to thermoregulation, increased BAT activity could be effective against diet-induced metabolic diseases such as obesity, type 2 diabetes, and hyperlipidemia (Hamann et al. 1995; Poher et al. 2015; Kim and Plutzky 2016). For example, mice with increased UCP1 mRNA stability or BAT function were more resistant to high-fat diet-induced obesity (Qiang et al. 2012; Dempersmier et al. 2015; Takahashi et al. 2015). Alternatively, after ectopic expression of UCP1 in pigs, UCP1-KI pigs showed increased thermoregulation and significantly reduced body fat and backfat thickness (Zheng et al. 2017). On the other hand, it has been reported that mice lacking UCP1 or BAT-deficient mice show significantly increased fat deposition and obesity susceptibility under high-fat diet conditions (Lowell et al. 1993; Kopecky et al. 1995; Feldmann et al. 2009; Wang and Seale 2016; Luijten et al. 2019). In humans, increased obesity was significantly associated with decreased BAT activity (Vijgen et al. 2011, 2012). Previous studies revealed that mammals adapted to extremely cold environments, such as cetaceans, sirenians (Trichechus manatus), and woolly mammoth (Mammuthus primigenius), tend to show inactivation of UCP1 (Gaudry et al. 2017). Notably, these species exhibit a remarkable convergence in significant thickening of subcutaneous fat, especially in cetaceans. Therefore, it is interesting to explore the evolution of the UCP1 gene and its potential role in promoting the thickening of the subcutaneous fat layer associated with secondary aquatic adaptation. In this study, we identified inactivating mutations in the cetacean UCP1 gene through a comparative analysis of the UCP1 gene in 27 cetacean species. Furthermore, immunohistochemical (IHC) and in vitro functional analyses revealed reduced expression, stability, and uncoupling activity of the cetacean UCP1 protein. These findings may provide new insights into the molecular mechanisms of blubber thickening in secondary aquatic adaptation.
To investigate, the role of the UCP1 gene in cetacean fat consumption and metabolic thermogenesis, we identified UCP1 sequences in 42 cetartiodactylan species with high-quality genome assemblies, including 27 cetacean species. Numerous variations in the exons of UCP1 in cetaceans were widely observed, whereas large fragment deletions were found in the toothed whales, such as Delphinidae and Hyperoodontidae (complete deletion of UCP1), Phocoenidae and Monodontidae (deletion of exons 1–4), fin whales (Balaenoptera physalus) and boutu (Inia geoffrensis) (deletion of exons 3 and 4), and franciscana (Pontoporia blainvillei) (deletion of exon 6) (fig. 1A). Although the complete coding sequence (CDS) of UCP1 was conserved in most baleen whales (except fin whale), many frameshift mutations were identified. Five shared mutations were found in all cetaceans, including splice site mutations in exon 1 (AG→AT) and exon 3 (GT→GG), a 4-bp deletion in exon 4, and a 2-bp deletion and premature termination (CGA→TGA) in exon 6 (fig. 1A). All of these mutations were verified in the Sequence Read Archive (SRA) database (fig. 1B). Mutations in the Yangtze finless porpoise, baiji, minke whale, and sperm whale were verified and corrected by polymerase chain reaction (PCR), and the 1 bp deletion of exon 2 in baiji, the deletion of exon 2 in Yangtze finless porpoise, and the 2 bp deletion of exon 2 in sperm whale were found to be due to sequence errors (supplementary fig. S1, Supplementary Material online). In addition, UCP1 is associated with ADP/ATP carriers, and is a member of the mitochondrial carrier family (MCF) consisting of three ∼100-amino acid homologous domains. The MCF domains function to interact to form a pseudo-symmetric salt bridge network to maintain stability, which is essential for UCP1 protein function. A comparison of UCP1 protein domains revealed that sperm whale conserved three MCF domains and maintained the UCP1-specific motifs after pseudogeneity, whereas other cetaceans had only one or two MCF domains (fig. 2). Since these domains are essential for the UCP1 protein to perform its function, we speculated that pseudogenized UCP1 may still have some function in cetaceans.
Pseudogenes are generally under a relaxed selective pressure. To explore the evolutionary pattern of the UCP1 gene in cetaceans, we firstly used one-ratio CODEML model to detect an overall selection pressure for all species. In this model, mammalian UCP1 exhibited an overall selection constrained dN/dS ratio (ω = 0.57822, P < 0.05; table 1). Notably, in the branching model, the cetacean UCP1 gene showed a nearly neutral evolutionary rate (ω = 0.9557, P < 0.05), whereas the intact UCP1 branch showed purifying selection (ω = 0.14593, P < 0.05, table 1). Furthermore, the relaxed selection pressure on cetacean lineages was supported by the RELAX results using cetaceans with inactivating mutations as test branches (K = 0, P < 0.0001, supplementary fig. S2, Supplementary Material online). Based on the measured dN/dS ratios and the divergence time of cetaceans, the UCP1 gene in cetaceans was estimated to have been inactivated at 50.2–53.1 Ma, shortly after the ancestral cetaceans diverged from the hippopotamuses. On the other hand, complete deletion of UCP1 occurred in the middle Oligocene in Hyperodontidae, and in the early Miocene in Delphinidae, Phocoenidae, and Monodontidae (fig. 3).
The increased number of mitochondria and miniaturization of lipid droplets in adipocytes are important markers of enhanced UCP1-mediated NST in adipocytes. The inner, middle, and outer layers were defined by the location of the blubber layer relative to skeletal muscle (fig. 4A), and hematoxylin–eosin (HE) staining revealed that adipocytes in the inner layer were small unilocular fat droplets (90–170 μm), significantly smaller than those in the middle and outer layers (190–290 μm; fig. 4B). This suggested that adipocytes in the inner layer of blubber had the potential to undergo fatty acid β-oxidation. Western blotting (WB) analysis of extracts from the blubber layer showed that UCP1 protein expressed in the inner layer was significantly increased compared with the outer and middle layers of the blubber (fig. 4C). Furthermore, IHC analysis using mouse anti-UCP1 antibody showed that UCP1 protein was highly expressed in the inner layer of the blubber. Additionally, voltage-dependent anion channel protein 1 (VDAC1) was the most abundant protein in the outer mitochondrial membrane and played an important role in regulating mitochondrial energy metabolism, and the amount of VDAC1 protein can be a marker of mitochondrial energy metabolism and quantity. WB and IHC staining results revealed that VDAC1 protein was significantly enriched in the inner layer of the blubber (fig. 4DandE), suggesting that adipocytes in the inner layer of the blubber contained more mitochondria. Taken together, these results suggested that the pseudogene UCP1 was highly expressed in the inner layer of blubber, and its inner layer adipocytes were similar to brown adipocytes with UCP1-mediated NST function.
Pseudogenization of a gene could affect the expression, translation, and function of the targeted gene and its downstream-related genes. WB analysis showed that, although pseudogenic UCP1 is normally translated into protein, total protein expression and stability were significantly reduced (fig. 5AandB). When the UCP1 protein performs its uncoupling function, mitochondrial membrane potential (MMP) is reduced, and changes in MMP can be as a characterization of UCP1 protein function. As a result, a significant increase (∼1-fold) in MMP was detected in pseudogenic UCP1 compared with intact gene copy (fig. 6; supplementary fig. S3, Supplementary Material online). Furthermore, pseudogene UCP1 was found to significantly increase triglyceride (TG) content in 3T3-L1 cells (fig. 5C) and significantly decrease expression levels of thermogenic genes (TFAM and CD137) and lipolytic genes (ATGL and CPT1A) in C2C12 cells (fig. 7). These results suggest that pseudogenized UCP1 in cetaceans may significantly attenuate uncoupling and lipolytic functions.
The blubber is recognized as the primary site of fat and energy storage in cetaceans. It is central to the entire process of thermoregulation and maintenance of body streamliner and is an essential guarantee for cetaceans to adapt to the aquatic environment (Ellis 2009). However, the mechanism of how cetaceans effectively balance blubber thickness and energy metabolism has remained unclear. In this study, we combined bioinformatics analysis and in vitro functional assays to explore the role of the UCP1 gene in cetacean aquatic adaptation, and provided new insights into blubber thickness and body temperature regulation in cetaceans. Comparative analysis of 27 cetacean UCP1 genes identified frameshifting insertions and deletions, splice site disruption mutations, and in-frame stop codon mutations as signals of gene inactivation mutations in the cetacean UCP1 gene. This is similar to the results of Gaudry et al. (2017), but we corrected some of these putative pseudogene signals using PCR (supplementary fig. S1, Supplementary Material online). This suggested that the UCP1 gene in cetaceans may have been pseudogenized. However, further analysis of the sequence alignment implied that the pseudogenized cetacean UCP1 gene seemed to still retain its function. For example, these mutations appeared in nonfunctional domains or splice sites. Other studies have shown that such a kind of mutations do not necessarily lead to functional changes or inactivation of the protein (Gaudry and Campbell 2017; Huang et al. 2020; Manger et al. 2021). Although two premature stop codons were identified in cetacean UCP1, both were located in the C-terminal region, indicating that the MCF domains were not significantly disrupted, and that retention of the MCF domain is important for UCP1 protein to perform its function (Nelson et al. 1998; Pebay et al. 2003; Ruprecht et al. 2014; Crichton et al. 2017). However, the number of MCF domains varied among cetaceans, and this variation might influence the stability of the pseudo-symmetric salt bridge network, the formation of the central cavity, and the rate of substrate exchange (Robinson et al. 2008; Divakaruni et al. 2012; Crichton et al. 2015). Furthermore, considering that the TGA stop codon showed significantly lower termination efficiency than other stop codons such as TAG and TAA (Cridge et al. 2018; Huang et al. 2020), the two shared TGA stop codons from the frameshift mutation of the cetacean UCP1 gene may have had a lower effect on transcriptional termination of this gene. In summary, the cetacean UCP1 gene may still retain some function, although it shows some signals of pseudogenization. The retention of a functional UCP1 gene with a premature stop codon in cetaceans was supported by further functional assays. For example, the UCP1 gene in cetaceans can be expressed and translated into UCP1 protein (fig. 5AandB) and showed uncoupling activity (supplementary fig. S3, Supplementary Material online), indicating that the pseudogene UCP1 is functional. This was further confirmed by IHC analysis, which showed that UCP1 protein is highly expressed in the inner blubber layer of Yangtze finless porpoise (fig. 4D). Indeed, several previous studies have demonstrated that parts of pseudogenes are transcribed or expressed and fully or partially functional as intact genes in humans and mice (Zheng et al. 2007; Tarn et al. 2008; Poliseno et al. 2010; Han et al. 2011; Muro et al. 2011; Cheetham et al. 2020). For example, the pseudogene CX43 could not only be transcribed normally, but also encode the same cell growth inhibitory protein as the parental functional CX43 gene (Kandouz et al. 2004). UCP1, as a core protein of BAT thermogenesis, can activate the sensitive lipase (HSL) and lipid TG lipase (ATGL) to promote lipolysis in adipocytes (Michurina et al. 2021). Surprisingly, when the UCP1 gene was pseudogenized, expression levels of two lipolysis rate-limiting enzymes (ATGL and CPT1A) were detected to be significantly reduced (fig. 7), suggesting a mechanism to avoid excessive lipolysis. ATGL is a rate-limiting enzyme that catalyzes TG hydrolysis. It has been reported that TG hydrolysis activity is reduced and fat deposition is accelerated in ATGL knockout mice (Haemmerle et al. 2006; Yamaguchi 2010; Wong et al. 2011). Similarly, CPT1A, the rate-limiting enzyme for fatty acid β-oxidation, can regulate TG accumulation in the body by mediating fatty acid β-oxidation (Gagnon et al. 2014). Suppression of CPT1A gene expression may promote the expression of fatty acid synthase and acetyl-core carboxylase (ACCα) genes, resulting in TG accumulation (Bhuiyan et al. 1994). For these reasons, the pseudogenized UCP1 gene may decrease lipid degradation in adipocytes and promote TG accumulation in cetaceans. This was further evidenced by experiments at the cellular level. The pseudogenized UCP1 gene significantly increased (40% increase) TG accumulation in 3T3-L1 cells (fig. 5C). On the other hand, it was found specific convergent amino acid substitutions in several genes (i.e., NFIA, SEMA3E, and MFN2) in fully aquatic marine mammals, and these genetic changes probably contribute to the development of blubber and the formation of a counter-current heat exchange system, that can effectively limit heat loss rather than increasing UCP1-mediated NST to maintain body temperature (Yuan et al. 2021). Notably and interestingly, inactivating mutations were also detected in UCP1 of sirenians and woolly mammoths with thick subcutaneous fat, but the mutations occurred at different amino acid sites. Although all these mutations occurred independently and no mutations were shared with cetaceans, the pseudogenized UCP1 gene in these species suggesting that these species may have evolved a functionally convergent mechanism to reduce heat dissipation in cold environments to maintain subcutaneous fat thickness. The present study suggested that the pseudogenized UCP1 gene identified in cetaceans has weakened the uncoupling function of this gene (fig. 6; supplementary fig. S3, Supplementary Material online) and may have significantly reduced UCP1-mediated NST. This may be another indirect mechanism by which cetaceans may no longer need to increase lipolysis for heat production, to prevent unrestricted lipolysis and maintaining the blubber layer for aquatic adaptation. In conclusion, it was found that pseudogenization of the UCP1 gene in cetaceans may effectively reduce lipolysis in adipocytes and promote fat deposition. Therefore, pseudogenization and reduced function of the UCP1 gene in cetaceans may be an evolutionary mechanism that aids secondary aquatic adaptation experienced by cetaceans during the transition from land to sea. This is supported to some extent by the fact that the period of inactivation of the UCP1 gene in cetaceans (∼53.1–50.2 Ma) coincided with the period when they returned to the sea (∼56–53 Ma).
In this study, we combined evidence from bioinformatics, immunohistochemistry, and functional analysis to comprehensively examine the adaptive evolution of the UCP1 gene in cetaceans for the first time. Bioinformatics analysis showed that the cetacean UCP1 gene exhibited pseudogenization signals, but functional evidence suggested that cetaceans still retain partial functions that may help them effectively prevent excessive lipolysis and maintain adequate blubber thickness. This may provide new insights into the evolutionary mechanisms of secondary aquatic adaptation in cetaceans.
Muscle samples of sperm whale (Physeter catodon), minke whale (Balaenoptera acutorostrata), Yangtze finless porpoise (Neophocaena asiaeorientalis), baiji (Lipotes vexillifer), melon whale (Peponocephala electra), and Blainville's beaked whale (Mesoplodon densirostris) were stored in a −20 °C freezer of the Jiangsu Key Laboratory for Biodiversity and Biotechnology, Nanjing Normal University. Whole-genomic DNA was extracted according to the conventional phenol-chloroform protocol (Sambrock and Russel 2001) and transferred to −20 °C for long-term storage. To obtain CDS of the UCP1 gene, six specific primer pairs for PCR amplification were designed (supplementary table S1, Supplementary Material online). PCR profiles were 95 °C/3 min, 95 °C-15 s, 55 °C-15 s, 72 °C-20 s, and 72 °C/8 min cycles, repeated 32 times. PCR products were directly sequenced using an ABI 3730 DNA sequencer (Sangon Biotech [Shanghai] Co., Ltd).
All samples examined in this study were taken from individuals estimated to die within 48 h prior to sampling. Full-depth samples, including skin, blubber, and a portion of the muscle, were taken from the mid-thoracic aspect of each individual, and the blubber layer was separated into three regions (inner, middle, and outer layers). All samples from the different sections were immediately fixed in 4% paraformaldehyde (PFA) for histological analysis or frozen at −80 °C until use. At least three tissue blocks were taken from each sampling region.
HE staining was performed according to standard methods (Hashimoto et al. 2015). Briefly, blubber was fixed in 4% PFA at 4 °C for 24 h. After dehydration, blubber tissue was embedded in paraffin and sectioned to 5 μm. Sections were stained with HE. Within each layer, the cytoplasmic size of adipose droplets and adipocytes was measured in three randomly selected fields using Image-ProPlus 6.0. For IHC staining, deparaffinized sections were incubated with H2O2 and reacted with mouse anti-UCP1 antibody (Proteintech, 23673-1-AP, 1:100) or mouse anti-VDAC1 antibody (Proteintech, 55259-1-AP, 1:100) overnight at 4 °C. The sections were incubated with 1× phosphatidylinositol (1:100), washed three times with 1X phosphate-buffered saline, and then incubated with secondary antibodies (goat anti-mouse IgG, SA00001-1, 1:10,000; goat anti-rabbit IgG, SA00001-2, 1:10,000) for 1 h at room temperature. Nuclei were stained with 4′,6-diamidino-2-phenylindole.
Sequences encoding mouse UCP1 and its mutants were amplified from mouse cDNA sequences using specific primers (supplementary table S1, Supplementary Material online) and cloned into pEGFP-C1 or pcDNA 3.1(+) plasmids. All sequences were confirmed by Sanger sequencing. C2C12 cells and 293T cells were purchased from the American Type Culture Cell Collection. 3T3-L1 adipocytes were donated by Zhu Li of the Farm Animal Genetic Resources Exploration and Innovation Key Laboratory of Sichuan Province, Sichuan Agricultural University, Chengdu, Sichuan, China. Mouse C2C12, 293T, and 3T3-L1 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (P/S) in a culture vessel maintained at 37 °C and 5% CO2 for 48 h. Then, LipoD3000 DNA in vitro transfection reagent (Ver. II; SignaGen Laboratories), plasmid pEGFP-C1 or pcDNA 3.1(+) (negative control) and recombinant plasmids were separately transfected into C2C12, 3T3-L1, or 293T cells. After 12 h, the medium was replaced with DMEM supplemented with 10% FBS and 1% P/S. For adipocyte differentiation, 3T3-L1 and C2C12 cells were allowed to reach 100% confluency and treated for 48 h with an induction medium containing DMEM, 10% FBS, 1% P/S, 5 μg/ml insulin (Solarbio), 0.5 mM isobutylmethylxanthine (Sigma), 1 μM dexamethasone (Solarbio), 1 μM triiodothyronine (T3, Solarbio), and 1 μM rosiglitazone in induction medium for 48 h. The cells were then maintained in a differentiation medium supplemented with 5 μg/ml insulin, 1 μM T3, and 1 μM rosiglitazone for another 6 days. During differentiation, the medium was changed every 2 days. To induce the expression of thermogenes, C2C12 cells were incubated with 10 μM forskolin for 4 h.
Triglyceride concentration in mature adipocytes was detected using a TG assay kit (Solarbio Science & Technology, Beijing, China) according to the manufacturer's instructions. Briefly, mature adipocytes were collected in six-well plates, reacted with reagents supplied with the TG content detection kit, and the OD value was measured by a microplate reader at a wavelength of 420 nm and the TG concentration was calculated based on a standard curve.
Mitochondrial membrane potential was detected using the JC-10 Assay Kit (Solarbio Science & Technology) according to the manufacturer's protocol. Briefly, mature adipocytes were seeded in six-well plates, stained with JC-10 dye for 20 min in an incubator maintained at 37 °C, washed twice with JC-10 staining buffer, and 2 ml of cell culture medium was added. Fluorescence was detected in a microplate reader at wavelengths of 490 nm (excitation), 530 nm (emission), 525 nm (excitation), and 590 nm (emission), respectively. All experiments were performed at least three times.
Total RNA was isolated using Trizol Extraction Reagent (Vazyme, Nanjing, China) according to the manufacturer's instructions. Extracted total RNA (2 μg) was converted back to cDNA using Hifair III first Strand cDNA Synthesis SuperMix qPCR (YEASEN, Shanghai, China) and analyzed with Roche LightCycler 480 II instrument (Roche Applied Science, Mannheim, Germany) for real-time PCR system and mRNA expression was quantitatively measured using SYBR Green (YEASEN) with β-tublin as an internal normalization control. Primer sequences are shown in supplementary table S1, Supplementary Material online. For WB, cells and blubber tissue were lysed at 4 °C in radioimmunoprecipitation assay buffer (Solarbio) mixed with protease and phosphatase inhibitors. Protein samples were incubated with primary antibodies against UCP1 (Proteintech, 23673-1-AP, 1:800), 3× FLAG-tag (Proteintech, 20543-1-AP, 1:3,000), or β-tublin (Proteintech, 10094-1-AP, 1:3,000) at 4 °C overnight. Proteins were detected by incubation with horseradish peroxidase-conjugated secondary antibodies and visualization was detected with a chemiluminescence system (Fusion Solo S; Vilber Lourmat, France). Quantitative analysis of WB bands was performed using Image-ProPlus 6.0 software. For protein stability assays, cycloheximide (CHX; 100 μg/ml; Santa Cruz Biotechnology) was added to cultures for the desired treatment time (0, 1, 3, and 5 h). All cells were collected 36 h after transfection, and lysates were assayed by WB. Quantitative analysis of WB bands was performed using Image-ProPlus 6.0 software.
To detect the adaptive evolution of the UCP1 gene in cetaceans, a total of 42 cetartiodactylan species (including 27 cetaceans) with high-quality genomes were used in this study (supplementary table S2, Supplementary Material online). First, CDSs of the UCP1 genes of cattle (Bos taurus), common bottlenose dolphin (Tursiops truncatus), and minke whale were downloaded from the NCBI database (https://www.ncbi.nlm.nih.gov/). The UCP1 sequence for the bowhead whale (Balaena mysticetus) was downloaded from http://www.bowhead-whale.org/. BLAST alignments of other mammals were performed using ∼100 bp upstream and downstream of each exon of these species as reference sequences. Sequence alignment was performed in MUSCLE in MEGA6 (Edgar 2004; Tamura et al. 2013). Frameshift mutations and overlapping regions were also aligned manually. We followed the method described by Sharma et al. (2020) to detect the inactivation of the UCP1 gene in 42 mammals. To verify the presence of inactivating mutations in the ancestors of cetaceans, the ancestral sequences were reconstructed by adding the “-showanc” parameter to PRANK (Löytynoja and Goldman 2010). In addition, validation of putative inactivating mutations was performed using raw sequence reads in the SRA data in the NCBI online database (https://www.ncbi.nlm.nih.gov/sra/term). In addition, the UCP1 sequences of Yangtze finless porpoise, baiji, sperm whale, and minke whale were amplified using PCR with the primer sequences shown in supplementary table S2 (Supplementary Material online).
To assess whether the UCP1 CDS was subjected to different evolutionary selection pressures across species, the CODEML program in PAML 4 (Yang 2007) was used to calculate the nonsynonymous substitution rate (dN)/synonymous substitution rate (dS) ratio (ω = dN/dS). All frameshift insertions/deletions and nonsense mutations in the alignments were removed prior to analysis; the phylogenetic tree at PAML runtime was downloaded from the TimeTree Database (http://www.timetree.org/). Five models were used to determine the selection pressure in different branches of the UCP1 gene. Briefly, model A assumes that all branches have the same ω; model B is the same as model A, but ω is fixed at 1 (ω = 1); model C assumes that pseudogenized branches (cetacean branches) have the same ω2 and intact branches (noncetacean branches) have the same ω1; model D is the same as model C, but ω2 is fixed at 1 (ω2 = 1); and model E assumes that each branch has ω independently. The five models are then compared pairwise (A vs. B, C vs. A, D vs. C, and E vs. C). A likelihood ratio test with a χ2 test was performed on the nested likelihood models. The mean ω for each rank or super rank was calculated according to the results of the free likelihood model. In addition, RELAX is a hypothesis testing framework that asks whether the intensity of natural selection has relaxed or strengthened along a given set of test branches. It is the most common tool for identifying trends or shifts in the severity of natural selection for a given gene (Wertheim et al. 2014). In this study, the program RELAX, available on the Datamonkey website (http://www.datamonkey.org/), was used to further examine relaxed selection.
To assess when UCP1 was inactivated in cetaceans, we applied the methods described by Chou et al. (2002) and Mu et al. (2021). Genes evolve under the same selection pressures (Ks) as other species until they are inactivated. Once inactivated, genes are assumed to evolve neutrally (Kn = 1) and accumulate nonsynonymous and synonymous mutations at a neutral rate. The dN/dS values (K) estimated for the entire branch are the average dN/dS values of the branch under selection (Ks) and the dN/dS values of the branch under neutral evolution (Kn = 1). The genes were then weighted by the proportion of time they were evolving under selection (Ts/T) and neutrally (Tn/T) (Chou et al. 2002; Mu et al. 2021).where T is the time since divergence from the last common ancestor; upper and lower confidence intervals for species divergence time T were obtained from the TimeTree Database (http://www.timetree.org/); Ts represents the time for a gene to evolve under selection pressure; Tn is the time for a gene in neutral evolution.
All data are presented as mean ± SEM. Significance was analyzed by using Student's t-test or one-way analysis of variance; P < 0.05 was considered statistically significant. Data analysis was completed by applying SPSS 20.0 (SPSS Inc., Chicago, IL, USA).
Click here for additional data file. |
PMC9648563 | Marta Hernández-Meneses,Jaume Llopis,Elena Sandoval,Salvador Ninot,Manel Almela,Carlos Falces,Juan M Pericàs,Bárbara Vidal,Andrés Perissinotti,Francesc Marco,Carlos A Mestres,Carlos Paré,Cristina García de la María,Guillermo Cuervo,Eduard Quintana,José M Tolosana,Asunción Moreno,José M Miró, | Forty-Year Trends in Cardiac Implantable Electronic Device Infective Endocarditis | 14-10-2022 | 40 years,CIED infective endocarditis,device removal,epidemiology,prognosis | Abstract Background Studies investigating cardiac implantable electronic device infective endocarditis (CIED-IE) epidemiological changes and prognosis over long periods of time are lacking. Methods Retrospective single cardiovascular surgery center cohort study of definite CIED-IE episodes between 1981–2020. A comparative analysis of two periods (1981–2000 vs 2001–2020) was conducted to analyze changes in epidemiology and outcome over time. Results One-hundred and thirty-eight CIED-IE episodes were diagnosed: 25 (18%) first period and 113 (82%) second. CIED-IE was 4.5 times more frequent in the second period, especially in implantable cardiac defibrillators. Age (63 [53-70] vs 71 [63–76] years, P < .01), comorbidities (CCI 3.0 [2–4] vs 4.5 [3–6], P > .01), nosocomial infections (4% vs 15.9%, P = .02) and transfers from other centers (8% vs 41.6%, P < .01) were significantly more frequent in the second period, as were methicillin-resistant coagulase-negative staphylococcal (MR-CoNS) (0% vs 13.3%, P < .01) and Enterococcus spp. (0% vs 5.3%, P = .01) infections, pulmonary embolism (0% vs 10.6%, P < .01) and heart failure (12% vs 28.3%, p < .01). Second period surgery rates were lower (96% vs 87.6%, P = .09), and there were no differences in in-hospital (20% vs 11.5%, P = .11) and one-year mortalities (24% vs 15%, P = .33), or relapses (8% vs 5.3%, P = 0.65). Multivariate analysis showed Charlson index (hazard ratios [95% confidence intervals]; 1.5 [1.16–1.94]) and septic shock (23.09 [4.57–116.67]) were associated with a worse prognosis, whereas device removal (0.11 [.02–.57]), transfers (0.13 [.02–0.95]), and second-period diagnosis (0.13 [.02–.71]) were associated with better one-year outcomes. Conclusions CIED-IE episodes increased more than four-fold during last 40 years. Despite CIED-IE involved an older population with more comorbidities, antibiotic-resistant MR-CoNS, and complex devices, one-year survival improved. | Forty-Year Trends in Cardiac Implantable Electronic Device Infective Endocarditis
Studies investigating cardiac implantable electronic device infective endocarditis (CIED-IE) epidemiological changes and prognosis over long periods of time are lacking.
Retrospective single cardiovascular surgery center cohort study of definite CIED-IE episodes between 1981–2020. A comparative analysis of two periods (1981–2000 vs 2001–2020) was conducted to analyze changes in epidemiology and outcome over time.
One-hundred and thirty-eight CIED-IE episodes were diagnosed: 25 (18%) first period and 113 (82%) second. CIED-IE was 4.5 times more frequent in the second period, especially in implantable cardiac defibrillators. Age (63 [53-70] vs 71 [63–76] years, P < .01), comorbidities (CCI 3.0 [2–4] vs 4.5 [3–6], P > .01), nosocomial infections (4% vs 15.9%, P = .02) and transfers from other centers (8% vs 41.6%, P < .01) were significantly more frequent in the second period, as were methicillin-resistant coagulase-negative staphylococcal (MR-CoNS) (0% vs 13.3%, P < .01) and Enterococcus spp. (0% vs 5.3%, P = .01) infections, pulmonary embolism (0% vs 10.6%, P < .01) and heart failure (12% vs 28.3%, p < .01). Second period surgery rates were lower (96% vs 87.6%, P = .09), and there were no differences in in-hospital (20% vs 11.5%, P = .11) and one-year mortalities (24% vs 15%, P = .33), or relapses (8% vs 5.3%, P = 0.65). Multivariate analysis showed Charlson index (hazard ratios [95% confidence intervals]; 1.5 [1.16–1.94]) and septic shock (23.09 [4.57–116.67]) were associated with a worse prognosis, whereas device removal (0.11 [.02–.57]), transfers (0.13 [.02–0.95]), and second-period diagnosis (0.13 [.02–.71]) were associated with better one-year outcomes.
CIED-IE episodes increased more than four-fold during last 40 years. Despite CIED-IE involved an older population with more comorbidities, antibiotic-resistant MR-CoNS, and complex devices, one-year survival improved.
Longevity in developing countries has increased remarkably in recent decades. In Spain, the average life expectancy in 1981 was 72 years and is 83.6 years in 2019, among the highest in the world [1]. The toll of rising life expectations is the growth in comorbidities, primarily cardiovascular diseases; therefore, the number of people requiring cardiac implantable electronic devices (CIEDs) has increased. The technological development of cardiac medical devices has been noteworthy in recent years, increasing the use of last-generation pacemakers (PPMs), implantable cardioverter-defibrillators (ICDs), and cardiac resynchronization therapy (CRT) [2]. Although the reported incidence of CIED infections varies notably among different studies, the increase in implantations has augmented the overall device infection rate [3, 4]. Contemporary authors have published a prevalence ranging from 0.68% to 5.7% [3–6], and the risk seems to be higher in CRT than in ICDs and PPMs [5]. Infective endocarditis (IE) related to CIEDs (CIED-IE) is the most severe complication, representing 10% of overall IE [5]. CIED-IE's global characteristics and evolution over the years are poorly studied. Stratification of risk depending on the type of the device (PPM, ICD, CRT), the clinical profile according to the time of presenting symptoms (early or late), or the etiology could guide the diagnosis and management of CIED-IE. It is recognized that removing the entire device is the key to managing these infections [7, 8]. However, the main problem is that CIEDs might have been implanted a long time ago in older, comorbid, and fragile patients; thus, combined with a high risk of complications during extraction surgery, CIEDs can sometimes not be removed. In those cases, patients may require lifelong oral antibiotic suppression treatment, decreasing their quality of life and increasing morbidity and mortality in the short and medium terms [9, 10]. Changes in the CIED-IE paradigm due to the growth in comorbidities, age, and implantation rate of overall devices have resulted in more complex infections, elevated surgical risk, and more patients without complete device removal. Chronic oral suppression therapy indication has been poorly reviewed, and whether these variations might have overcome increasing mortality for CIED-IE has not been reported. This study investigates the historical evolution of consecutive CIED-IE episodes and defines changes in epidemiology, clinical presentation, outcomes, and 1-year mortality during the last 4 decades.
This was an observational retrospective study of prospectively followed CIED-IE at Hospital Clinic de Barcelona (HCB), a referral cardiovascular surgery center for IE and cardiovascular infections. Cases were followed since 1979, when the HCB IE database was created. The first pacemaker was implanted at our center in April 1980, and the first CIED-IE was diagnosed in January 1981. In addition, the first ICD was implanted in 1991 and the first CRT in 1999. Thus, data were collected during the index hospitalization between January 1981 to December 2020. All patients had 1-year follow-up. The study ended on 31 December 2021.
We included 138 consecutive patients with definite CIED-IE. The management of all patients was discussed at weekly endocarditis team meetings since 1986 [11]. The final diagnosis was accomplished by consensus of the IE team.
Only patients with definite CIED-IE using the modified Duke criteria for IE and presented in the IE team meetings were included [4, 12]. All patients, with or without local signs of pocket infection, had valve vegetations in either valve or lead of the CIED and positive blood cultures and/or positive lead culture and/or 16S ribosomal RNA (rRNA) gene sequencing positive. Due to the aim of this study, we used only the first episode of CIED-IE for each patient. Patients with no definite criteria for IE were excluded.
CIED-related pocket infection was defined by local signs of inflammation at the generator of the device, including erythema, warmth, fluctuance, wound dehiscence, tenderness, purulent drainage, or erosion of the generator or lead through the skin and/or positive pocket swab or positive device or subcutaneous lead cultures or 16S rRNA gene sequencing positive. CIED-IE was considered in patients who met the Duke criteria for IE. All patients presented positive blood cultures and/or lead, and/or valve cultures and/or 16S rRNA gene sequencing positive, and lead or valve vegetations in echocardiography. Echocardiographic diagnosis was achieved by transthoracic echocardiography between 1981 and 1990, whereas, since January 1991, most cases have undergone transesophageal echocardiography (TEE). Any mass seen on a lead and/or valve in echocardiography in the context of bacteremia was assumed to be vegetation. A second investigator validated all echocardiography studies and discrepancies were sorted out by adopting the most prevalent opinion when consulting a third member of the endocarditis team. 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18FDG-PET/CT) was included in our center in 2014 and was not considered as a diagnostic CIED-IE criterion for this study; we recorded all 18FDG-PET/CT data of CIED-IE patients on whom it was performed. Microbiological diagnosis included microorganisms detected by blood cultures or cultures of cardiac device lead and/or 16S rRNA gene sequencing positive. 16S rRNA polymerase chain reaction (PCR) was implemented since 2015.
Devices included in this study were PPMs, ICDs, and CRT.
Healthcare-associated IE was defined in outpatients with extensive healthcare contact as reflected by any of the following: (1) received intravenous therapy, wound care, or specialized nursing care at home within 30 days before admission; (2) attended a hospital or hemodialysis clinic or received intravenous chemotherapy within 30 days before diagnosis; (3) was hospitalized in an acute care hospital for 2 or more days within 90 days before admission; or (4) resided in a nursing home or long-term care facility. Nosocomial IE was defined as an infection diagnosed after 72 hours of admission in an outpatient [13].
Early CIED-IE was defined as signs and symptoms within 6 months of the most recent CIED procedure. Signs and symptoms occurring >6 months after surgery were described as late CIED-IE [13, 14]. The Charlson Comorbidity Index (CCI) was used to assess patient morbidities. The CCI consists of 19 different disease categories with varying numerical weights (1, 2, 3, or 6 points based on adjusted 1-year mortality relative risk) allotted to specific diseases [15]. It has been previously validated as a predictor of mortality in many clinical contexts, including patients with permanent CIED implantation.
The following systemic complications were recorded: heart failure (HF), central nervous system complications, pulmonary embolisms, acute renal failure, persistent bacteremia, and septic shock. Persistent bacteremia was defined as positive blood cultures yielding the causative microorganism after 7 days of effective antibiotic therapy [13].
We analyzed indication of device removal, type of device removal procedure, cause of surgery rejection, and length of antimicrobial treatment. Patients without a complete device removal underwent oral antibiotic suppression therapy. The duration of oral suppression treatment was recorded, and antimicrobial susceptibilities were used to guide the definitive oral antimicrobial therapy. Relapse was defined as the isolation of the same microorganism in blood cultures within 180 days after the end of antibiotic treatment. Reinfection was described as a new episode of IE caused by a different microorganism or by the same microorganism ≥180 days after the end of the antibiotic treatment. Cardiac surgery and mortality were classified into in-hospital and 1-year surgery/mortality.
Primary endpoints were in-hospital and 1-year mortality, and secondary endpoints were device removal and relapses. We compared the prevalence, epidemiology, clinical characteristics, and outcomes between 1981–2000 and 2001–2020. We also compared the clinical characteristics and outcomes of CIED-IE according to etiology (coagulase-negative staphylococci [CoNS] vs no CoNS), timing of diagnosis (early-presenting [≤6 months from device implantation] vs late-presenting symptoms [>6 months]), and device type (PPM versus ICD/CRT). Data are presented as median (interquartile range [IQR]) for continuous variables and as frequencies (percentages) for categorical variables. As appropriate, continuous variables were compared using Student t test or the Mann-Whitney U test. Categorical variables were compared using the χ2 or Fisher tests, as appropriate. Predicted factors of 1-year mortality were also studied. Risk factors for in-hospital and 1-year mortality were analyzed using a logistic regression model with comparisons reporting odds ratios (ORs) or hazard ratios (HRs) with 95% confidence intervals (CIs), as appropriate. Variables found to have a simple association with mortality (P < .10) were considered for the final models. The 1-year mortality multivariate analysis was calculated considering just the related survival clinical variables. Age, diabetes, and chronic renal failure were excluded from the model, as they are included in the CCI. For all tests, statistical significance was determined at the P = .05 level. Survival analysis was performed using the Kaplan-Meier method. All statistical analyses were performed using Stata statistical package version 14 (StataCorp LLC).
The Institutional Review Board of the Hospital Clinic of Barcelona approved the implementation of this study (ERB number HCB/2018/0538). The study's retrospective nature waived the requirement for informed written consent. Patient identification was encoded, complying with the needs of the Organic Law on Data Protection 15/1999.
One hundred thirty-eight CIED-IE episodes were included in the study. We compared them according to 2 periods (1981–2000 versus 2001–2020) and between the last 2 decades (2001–2010 versus 2011–2020). The characteristics of the 4 groups are depicted in Table 1. The first (1981–2000) and the second (2001–2020) periods included 25 (18%) and 113 (82%) CIED-IEs, respectively. In the recent period, age (median, 63 years [IQR, 53–70] vs 71 [IQR, 63–76]; P < .01), comorbidities (median CCI score, 3.0 [IQR, 2–4] vs 4.5 [IQR, 3–6]; P < .01), nosocomial acquisition (4% vs 16%; P = .02), and referral from other centers (8% vs 41.6%; P < .01) were significantly more frequent. The performance of 18FDG-PET/CT was described only for the second period, as it was only introduced in 2014; 29 patients underwent 18FDG-PET/CT to complement the diagnostic approach, with 24 receiving positive results (82.8%). Specifically, 16 of 24 (66.7%) had a positive 18FDG-PET/CT at the pocket ± subcutaneous, 11 of 16 (68.75%) had a pocket and subcutaneous pathological uptake, and only (33.3%) showed endovascular involvement. Although in the second period there was a trend toward more patients on oral antibiotic suppression therapy (4% vs 10.6%; P = .18), there were fewer, but not significantly so, complete removals of device systems (96% vs 87.6%; P = .09) and no differences in the rates of in-hospital mortality (20% vs 11.5%; P = .11) or relapse (8% vs 5.3%; P = .65) between the 2 periods. In the recent period, complex infections due to methicillin-resistant CoNS (0 vs 13.3%; P < .01) and Enterococcus spp (0 vs 5.3%; P = .01) were more frequent, as were complications, for example, pulmonary embolism (0 vs 10.6%; P < .01) and HF (12% vs 28.3%; P < .01). Figure 1A summarizes the proportion of CIED-IE compared with overall IE episodes over the 4 decades. Figure 1B compares changes in the proportion of CIED-IE episodes according to the type of device (PPM and ICD/CRT). Between the 2 defined periods, the cumulative number of CIED-IE episodes was 4.5-fold higher in the second period (25 vs 113 cases), especially in ICD (2 vs 21 cases). Focusing on the comparison of the last 2 decades (2001–2010 vs 2011–2020), in the most recent period, there was a tendency for greater age (median, 73 years [IQR, 64–78] vs 69 years [IQR, 61–76]; P = .15) and significantly more comorbidities (median CCI score, 5 [IQR, 4–6.5] vs 4 [IQR, 3–5]; P < .01) and CRT (22.8% vs 8.9%; P = .04). Diagnostic tests—for example, 18FDG-PET/CT (47.4% vs 3.6%; P < .01) and molecular biology (45.6% vs 7.1%; P < .01)—were statistically more frequent in the most recent decade. In the 2011–2020 period, patients were less likely to undergo device removal (78.9% vs 96.4%; P < .01), so there were more patients on oral chronic suppression therapy (17.5% vs 3.6%; P = .01). Complicated CIED-IE cases were more frequent in the 2011–2020 period (73.7% vs 46.4%; P < .01). However, in terms of in-hospital and 1-year mortality, there were no differences between periods (14% vs 8.9%, P = .39 and 17.5% vs 12.5%, P = .45, respectively) although significantly more patients had relapses (10.5% vs 0%; P = .01) and underwent late surgery (8.8% vs 0%; P = .02) in the latter period.
Of the overall cohort, 62 episodes were due to CoNS and 76 due to other microorganisms (Table 2). CIED-IE due to CoNS had significantly more concomitant pocket infections (50% vs 31.6%; P = .03) and fewer comorbidities (median CCI score, 4.0 [IQR, 2.0–5.0] vs 5.0 [IQR, 4.0–7.0]; P < .01). Patients with CoNS CIED-IE had a larger valve vegetation size (18.0 mm vs 9.0 mm; P < .01) and were significantly more likely to undergo removal of the cardiac device system (96.8% vs 82.9%; P < .01); consequently, there were significantly more reimplants (76.76% vs 60.3%; P = .04). Oral antibiotic suppression therapy in patients without removal of the cardiac device system was significantly higher in CoNS CIED-IE than in the other etiologies (14.5% vs 3.2%; P = .01).
Considering the timing of diagnosis, early CIED-IE had tended toward local signs of infection predominancy (51.3% vs 35.4%; P = .09); meanwhile, fever was significantly the typical clinical manifestation of late-presenting CIED-IE (73.7% vs 53.8%; P = .03). Community-acquired (64.6% vs 41%; P = .01) and polymicrobial infections (7.1% vs 0%; P < .01) were significantly more frequent in late CIED-IE, as was the presence of vegetations in any valve (see summarized data in Supplementary Table 1). Thus, peripherical embolisms were more prevalent in late CIED-IE (11.1 vs 2.6%; P = .04). Regarding the type of device, PPM-IE represented 114 episodes from 138 (82.6%), whereas 24 episodes were on ICD/CRT-IE. PPM-IE patients were older (median age, 72 [IQR, 63–77] vs 62.5 [IQR, 54–68] years; P < .01), had significantly higher proportion of females (19.3% vs 4.2%; P < .01), and had more late IE (70.2% vs 41.7%; P = .01), as presented in Table 3. PPM-IE episodes more frequently had mitral valve vegetation (7% vs 0, P < .01). There were no differences between PCM-IE and ICD/CRT-IE regarding device removal, reimplantation rate, antibiotic suppression therapy, relapses, and hospital or 1-year mortality. Vegetation involvement is analyzed in Supplementary Table 2. A comparison between CIED-IE with isolated lead vegetations, CIED-IE with tricuspid valve vegetations (right-side), and CIED-IE with left-side valve vegetations (with or without lead vegetations) was performed. CIED-IE with lead and left-side involvement was significantly found in older patients than others (median age 74.5 vs 67 vs 71 years; P = .04), with a tendency for more comorbidities and earlier infection (50% vs 9.7% and 30.3%; P = .03), whereas CIED-IE with only lead involvement had more concomitant pocket infection (50.6% vs 22.6% and 16.7%; P = .02). CIED-IE with left-side and right-side valve involvement presented more complications (77.8% and 74.2% vs 42.7%; P = .04), for example, HF and central nervous system embolism, and was more likely to result in open surgery for device removal (66.7% vs 26% vs 12.8%; P < .01). There were no differences between in-hospital and 1-year mortality (Figure 2C), 1-year surgery, or relapses among the 3 groups.
From the overall cohort, 112 CIED-IE patients were alive and 23 died (16.7%) at 1 year of follow-up. Supplementary Table 3 compares the main differences between patients who were alive or had died at 1 year. Survivors had more concomitant pocket infections (43.8% vs 21.7%; P = .03), were more likely to have been transferred (39.3% vs 17.4%; P = .01), had fewer comorbidities (CCI score, 4.0 vs 5.0; P < .019), were more likely to have polymicrobial infections (6.3% vs 0; P < .01) and removal of cardiac device systems (93.8% vs 69.6%; P = .01). Conversely, complications (49.1% vs 82.6%; P < .01) such as HF (17.9% vs 65.2%; P < .01) and septic shock (4.5% vs 43.5%; P < .01) were more frequent in patients who had died at 1 year. Figure 2 shows the Kaplan-Meier survival curve for 1-year mortality in the overall cohort of patients with CIED-IE (Figure 2A) and the comparison of survival curves between the 2 studied periods (1981–2000 vs 2001–2020) (Figure 2B), among the 3 groups of valve vegetations (Figure 2C), and in patients with and without device removal (Figure 2D). Results of the 1-year survival multivariate analysis are shown in Table 4. CCI (HR, 1.44 [95% CI, 1.11–1.88]) and septic shock (HR, 13.12 [95% CI, 2.16–79.47]) were associated with a worse prognosis, whereas device removal (HR, 0.14 [95% CI, .02–.76]), being transferred from another center (HR, 0.13 [95% CI, .02–.95]), and a 2001–2020 period diagnosis (HR, 0.13 [95% CI, .02–.71]) were associated with lower 1-year mortality.
This is the largest historical cohort focused on CIED-IE over 40 years of study and managed by a single IE team in a referral center. As our IE team was created in 1985, all cases have been evaluated with uniform diagnostic and medical and surgical management criteria [11]. Several works have tried to define the epidemiological profile of CIED infections in recent years. For example, Dai et al [5] described another large cohort of CIED infections from the last 3 decades; however, they included overall CIED infections and did not incorporate the assessment of an IE team. All recent studies did factor in rising device implantation rates, likely related to a significant increase in PPM indication and lifetime use, more elderly patients, and higher ICD implantation, for sudden death prevention [5, 12–16]. Our study has also demonstrated fundamental changes in the epidemiology: an increase in median age, more comorbidities, and new types of CIED. We also reported new diagnostic techniques and greater resistance to antimicrobials in isolated pathogens. Despite all of these changes, in-hospital mortality did not significantly increase (20% during 1981–2000 vs 11.5% during 2001–2020, and 8.9% during 2001–2010 vs 14% during 2011–2020), and neither did 1-year mortality (24% during 1981–2000 vs 15% during 2001–2020; and 12.5% during 2001–2010 vs 17.4% during 2011–2020). However, the proportion of patients with unremovable CIEDs has notably increased over time (4% vs 12.4%; P = .09), mainly in the last decade (21.1%; P < .01) due to the population aging and the increase of comorbid conditions and complexity of devices. The cause of the higher number of infections, despite a decrease in overall device-related complications, is not clear [14, 17, 18]. One possibility is the accumulative numbers of ICDs and CRT, whose longevity is lower than PPMs, requiring more complex procedures and battery exchanges, which are strongly associated with risk of infection [13]. TEE plays an essential role in the diagnosis of CIED-IE when it is suspected in patients. However, it may prove challenging to differentiate vegetations from lead strands or small-adhered thrombi. George et al [8] described a case-control retrospective observational study showing how TEE could not distinguish the general characteristics of vegetations obtained from blinded TEE reports unless there was knowledge of clinical and microbiological parameters. In our cohort, incorporating 18FDG-PET/CT and molecular biology had a significant impact in the second period, having a sensitivity of 82.8% and 52.7%, respectively. However, this study was not designed to evaluate the diagnostic yield of these methods. Our analysis revealed a 4.5-fold increase in ICD/CRT-IE compared with PPM-IE when analyzing the cases from the 2 different periods. In the second period, the demographic and clinical characteristics of PPM-IE<.01 compared with those of ICD/CRT-IE were entirely different. Patients who received ICD/CRT were younger, predominantly male, and had more ischemic cardiomyopathy, diabetes, and HF. Greenspon et al [13] showed the nonvariation of the 4 significant comorbidities (renal failure, respiratory failure, HF, and diabetes) over almost the 2 last decades, but, similarly, there was a substantial increase in infection rate, mostly in ICDs (ICDs represented 35% of all devices, and CIED infection rates reported increased by 2.1% to 2.41% in 2008; P < .001). The etiology of CIED-IE was characterized by a predominance of staphylococcal infections, as is reported in our cohort, and fairly described by other investigators [5, 6, 14, 17–20]. However, interestingly, we identified an increase of Enterococcus spp infections in the second period, probably due to aging and more frequent comorbidities. In their study of the MEDIC cohort, Oh et al [21] conducted a descriptive analysis and reported 4.8% of enterococcal CIED infections from the whole database of 433 patients. Although they found no significant increase in enterococcal CIED infections over time, we did find a significant increase (up to 5.3%) in the second period of our study (P = .01). However, both studies consistently reported the profile of an elderly (median age, 70 years) combined with multiple underlying comorbidities (median CCI score, 6) and late infections. In our cohort, CoNS were the primary cause of CIED-IE, and methicillin resistance was expanding, in line with numerous medical reports [22], as were the CoNS factors of virulence and their presence in infections related to medical devices [23]. The medical and surgical approach has not changed between the 2 periods, and removing the entire device is mandatory [24, 25]. In the second period, the population was overall older and presented more frequent comorbidities; the proportion of nonremoval of the devices also increased, but mortality did not. The number of patients receiving antibiotic suppression therapy also increased. Other authors have also reported the increasing use of suppression therapy to manage CIED-IE when device removal is not possible [26, 27]. Since CIED-IE has low in-hospital mortality rates when compared to left-sided IE, we have calculated variables associated with survival at 1 year, given the greater perspective on the global management of these patients obtained over that length of time [28]. We identified CCI as an independent prognostic factor for 1-year mortality, as has been observed by other authors over the years [14]. In our analysis, we excluded age, chronic renal failure, and diabetes mellitus, because they are contained in CCI, although they are well-known risk factors for IE-related death [28, 29]. Septic shock was also associated with a worse prognosis, as has been broadly reported in other studies [5, 14, 19]. Our study identifies patient transfer from community centers as an independent protective factor. It was also more frequent in the second period. This finding may be explained by the tendency to transfer patients with better prognoses and fewer comorbidities for device removal [28, 30]. Complete device removal is the most important protective factor as has been shown in many studies [14, 24]. Finally, despite aging and greater patient complexity, the latter period was associated as a protective factor. This may be explained by improvements in diagnosis and medical and surgical management. Indeed, more accurate microbiological diagnosis using molecular techniques (eg, 16S rRNA PCR) [31, 32], and imaging diagnosis (eg, 18FDG-PET/CT) [32], in addition to improved surgical removal techniques, may support these results. Our study has several limitations. The first stems from the retrospective design. Nevertheless, the prospective homogenous diagnostic and therapeutic management provided by an IE team assessing the cases over 4 decades has allowed us to overcome this issue. Second, a selection bias might have partially influenced our temporal perspective of the profile of CIED-IE cases, because we are a referral center for cardiovascular surgery, and the characteristics of episodes managed at community noncardiac surgery centers are lacking. Third, although we included a large population-based cohort with long-term follow-up, this is a single-center study. A multicenter study may be more appropriate for obtaining a better population sample and render the study more broadly applicable. However, studies of this nature are unfeasible, because few sites maintain databases including patients over such long periods. Fourth, we were unable to accomplish the degrees of tricuspid valvular regurgitation in all CIED-IE episodes, and we did not record the notations of functional device failure-to-capture during CIED-IE episodes in our analysis. Finally, we randomly selected the 2 comparison periods considering the division by decades, and these small sample–sized subgroups might have hindered some statistical comparisons, so our findings should therefore be interpreted carefully. In conclusion, CIED-IE episodes have increased >4-fold over the last 40 years and more frequently presented infections caused by methicillin-resistant CoNS and Enterococcus spp. One-year survival significantly has improved over the last 2 decades compared to the last 20 years of the 20th century, despite increasing age and comorbidities among patients, who also now present more complex infections. Further studies are needed to clarify the upcoming challenges in diagnosing and managing CIED-IE when device removal is precluded in a growing high-risk population.
Click here for additional data file. |
|
PMC9648564 | Yannuo Li,Ioannis P. Androulakis | Light-induced synchronization of the SCN coupled oscillators and implications for entraining the HPA axis | 27-10-2022 | SCN,circadian,cortisol,HPA,seasonal | The suprachiasmatic nucleus (SCN) synchronizes the physiological rhythms to the external light-dark cycle and tunes the dynamics of circadian rhythms to photoperiod fluctuations. Changes in the neuronal network topologies are suggested to cause adaptation of the SCN in different photoperiods, resulting in the broader phase distribution of neuron activities in long photoperiods (LP) compared to short photoperiods (SP). Regulated by the SCN output, the level of glucocorticoids is elevated in short photoperiod, which is associated with peak disease incidence. The underlying coupling mechanisms of the SCN and the interplay between the SCN and the HPA axis have yet to be fully elucidated. In this work, we propose a mathematical model including a multiple-cellular SCN compartment and the HPA axis to investigate the properties of the circadian timing system under photoperiod changes. Our model predicts that the probability-dependent network is more energy-efficient than the distance-dependent network. Coupling the SCN network by intra-subpopulation and inter-subpopulation forces, we identified the negative correlation between robustness and plasticity of the oscillatory network. The HPA rhythms were predicted to be strongly entrained to the SCN rhythms with a pro-inflammatory high-amplitude glucocorticoid profile under SP. The fast temporal topology switch of the SCN network was predicted to enhance synchronization when the synchronization is not complete. These synchronization and circadian dynamics alterations might govern the seasonal variation of disease incidence and its symptom severity. | Light-induced synchronization of the SCN coupled oscillators and implications for entraining the HPA axis
The suprachiasmatic nucleus (SCN) synchronizes the physiological rhythms to the external light-dark cycle and tunes the dynamics of circadian rhythms to photoperiod fluctuations. Changes in the neuronal network topologies are suggested to cause adaptation of the SCN in different photoperiods, resulting in the broader phase distribution of neuron activities in long photoperiods (LP) compared to short photoperiods (SP). Regulated by the SCN output, the level of glucocorticoids is elevated in short photoperiod, which is associated with peak disease incidence. The underlying coupling mechanisms of the SCN and the interplay between the SCN and the HPA axis have yet to be fully elucidated. In this work, we propose a mathematical model including a multiple-cellular SCN compartment and the HPA axis to investigate the properties of the circadian timing system under photoperiod changes. Our model predicts that the probability-dependent network is more energy-efficient than the distance-dependent network. Coupling the SCN network by intra-subpopulation and inter-subpopulation forces, we identified the negative correlation between robustness and plasticity of the oscillatory network. The HPA rhythms were predicted to be strongly entrained to the SCN rhythms with a pro-inflammatory high-amplitude glucocorticoid profile under SP. The fast temporal topology switch of the SCN network was predicted to enhance synchronization when the synchronization is not complete. These synchronization and circadian dynamics alterations might govern the seasonal variation of disease incidence and its symptom severity.
Virtually all organisms possess an endogenous time-keeping system that entrains the biological and behavioral rhythms to a circadian 24 h period induced by light/dark cycles (zeitgeber). In mammals, the circadian clock forms a hierarchically organized network with the primary circadian pacemaker located in the suprachiasmatic nucleus (SCN), which converts environmental photic cues into hormonal, metabolic, and neuronal signals which subsequently synchronize the peripheral clocks and behavioral activities (1–3). The circadian timing system resembles a clock shop rather than a single clock due to the widespread existence of clocks throughout the body (4). As a result, the robust rhythmicity of such a system is indispensable for temporally coordinating organ functions, while the disruption or misalignment of the rhythms is strongly associated with various widespread diseases (5). Due to the revolution of the earth, physiological adjustments also correspond to seasonal changes to increase survival and reproductive success by relying on signals such as temperature, rainfall patterns, food availability, and duration of daily illumination (6). The latter, which is also defined as photoperiods, has been identified as the most pervasive factor influencing the properties of circadian entrainments, such as phase, amplitude, and degree of synchronization (7). The photoperiod varies across seasons, especially in high latitudes, with photoperiods longer in the summer and shorter in the winter. Accordingly, the SCN has evolved to adapt to the different photoperiods and exhibits distinct behaviors in long photoperiods (LP) and short photoperiods (SP). The heterogeneous SCN network consists of ~ 20,000 neurons coupled through neurotransmitters. In vivo, SCN neurons must synchronize to environmental cycles and each other to generate a coherent circadian output signal to entrain the downstream clocks (8). While rhythms of electrical activity for individual cells remain comparable across seasons, the amplitude of activity at the ensemble level is dependent on the degree of synchronization among neurons (9). Experimental evidence indicates that the degree of synchronization between SCN neurons in mice is lower under LP, and the neuronal phases are more widely distributed (3). Moreover, the peak of the SCN’s output is during the day, and the trough is during the night in both diurnal and nocturnal species (10), and transitions in behavioral patterns between rest and activity occur at half-maximal levels of the SCN’s activity rhythm. Therefore, the increased ensemble amplitude of the SCN in SP due to the narrowing of the neuronal phase distribution results in the physiology and behavior of organisms being shorter during SP than during LP (9, 11). Although much is known about the mechanisms of light entrainment of the circadian timing system, relatively little is known about the internal synchronizing strategies of the SCN (8, 12). Multiple studies have highlighted that the SCN is typically organized into the ventrolateral “core” region (VL) and the dorsomedial “shell” region based on anatomical and neurochemical content differences. Experimental evidence suggests that light-induced gene and protein expression occurs mainly in the VL core region in rats (13, 14), hamsters (15), and mice SCN (16), with the precise peptidergic identity of directly retinol-recipient cells species-dependent (17). In rats, the ventrolateral core mainly contains vasoactive intestinal polypeptide (VIP)-expressing neurons, whereas the dorsomedial shell predominantly contains arginine vasopressin (AVP)-expressing neurons. Other neurotransmitters such as gastrin-releasing peptide (GRP) and Neuromedin S (NMS) also exist and contribute to the activity of the SCN (18). Moreover, almost all SCN neurons are γ-amino butyric acid (GABA)-ergic (19). Although the specific mechanisms of different neurotransmitters might be different, those neurotransmitters are speculated to regulate neurons through a mutual coupling mechanism (20, 21), where they are released in a circadian fashion and exhibit feedback on the clock genes. The SCN’s function is dependent on the collective activity of the neurons and the inter-neuronal coupling (12, 22). Evidence for the SCN heterogeneity has led to various mathematical models of its network (21, 23, 24). Thus, several common network topologies have been examined in modeling the SCN’s structure, including all-to-all connectivity networks, regularly connected networks, Newman-Watts (NW) small-world networks, and scale-free networks (25). These models, while varying considerably in their complexity and underlying assumptions, all converge on the inclusion of a population of heterogeneous oscillators coupled with inter-neuronal communications. Theoretical works suggest that the response of the SCN to photoperiod changes can be effectively regulated by the number of connections between the core and the shell (2, 9, 26, 27). However, due to experimental limitations, the core-shell coupling nature remains unclear (28). The SCN consists of both distance-dependent and probability-dependent connections (2, 29). Specifically, the SCN rhythms of VIP knockout mice were restored by the coculture of WT SCN, implying that small diffusible (distance-dependent) neuropeptides secreted from the WT SCN are critical for the rhythm synchronization (30). On the other hand, optogenetic-assisted circuit mapping observed functional long-range connectivity between VIP hypothalamic neurons and AVP hypothalamic neurons (31), implying the importance of the neuronal network circuitry mediated by probability-dependent synaptic interactions. The above indicates that both the distance-dependent coupling and probability-dependent coupling can contribute to the mechanisms behind the core-shell communication, which are therefore both considered by the present work. Regulated by the SCN, glucocorticoid (cortisol in humans, corticosterone in rodents) levels secreted by the HPA axis also exhibit significant seasonal rhythmicity, with levels peaking during SP months and reduced during the LP months (32). The humoral glucocorticoid signals synchronize the expression of peripheral molecular clock components, which in turn regulate biological processes including immune function (33), glucose homeostasis (34), and steroidogenesis (35). Abnormal glucocorticoid circadian patterns are associated with chronic conditions such as rheumatoid arthritis (RA) (36), type 2 diabetes (37), metabolic disorders (38), and behavioral disorders such as post-traumatic stress disorder and depression (39). Therefore, the HPA axis plays a central role in response to the photoperiod variations transduced by the SCN, enabling the host to actively re-establish homeostasis in different seasons (40). Understanding the dynamics of circadian rhythms at a systematic level motivates the need for characterizing the SCN network, interactions between the SCN and the HPA axis, and synchronization variations of the system as photoperiods change. The present study investigates how SCN network topology influences system behavior and circadian dynamics by incorporating a precise single-cell oscillator model into a heterogeneous network of linked oscillators with two possible coupling mechanisms (distance-dependent and probability-dependent) across the core and the shell. By quantifying the synchronization efficiency of two alternative coupling mechanisms, our findings identify the probability-dependent core-shell coupling network as an “ideal” design capable of producing coherent signals with a reduced wiring cost. We further investigated the SCN and HPA axis response under different photoperiod schedules. Our results show that (1) neuronal communications induce systemic oscillations within the SCN even without an external zeitgeber; (2) the robustness and plasticity of the oscillatory network are negatively correlated (3) HPA rhythms were strongly entrained to the seasonally varying SCN rhythms with distinct day-length differences; Specifically, our results suggest that the pro-inflammatory high-amplitude glucocorticoids profile occurs during the shorter winter days, which is associated with enhanced immune responses and disease activities; (4) the temporal topology switch of the SCN network increases synchronization when the synchronization is not complete. Thus, our work sheds light on the adaptation strategies of the circadian timing system by examining complicated physiological entrainment structures between the SCN and the HPA axis under different photoperiods.
To mimic the organization of the circadian timing system that consists of the SCN and the HPA axis, our system is mathematically constructed with three compartments: the SCN core, the SCN shell, and the HPA axis ( Figure 1 ). The core and the shell are represented as two ensembles of molecular circadian oscillators. At a single-cell level, a detailed molecular model was used to describe the intracellular activities of clock genes and proteins, along the lines of our earlier work (41, 42). At the ensemble level, the cells are coupled by the released neurotransmitters. The HPA axis consists of a single cell model which incorporates the glucocorticoid dynamics as a synchronization outcome of the neurotransmitter AVP secreted from the shell (43). The schematic of the model is shown in Figure 1 . The detailed molecular model of the SCN includes a positive and negative feedback loop, which leads to an autonomous oscillation (44). In the positive feedback loop, the PER and CRY proteins translocate into the nucleus (Equations 4 and 12) following their expression (Equations 3 and 11) and activate Bmal1 mRNA transcription (Equations 5 and 13). After the translation of BMAL1 protein (Equations 6 and 14) and its translocation to the nucleus (Equations 7 and 15), PER/CRY protein indirectly increases the expression of CLOCK/BMAL1 heterodimer (Equations 8 and 16). In the negative feedback loop, nucleus PER/CRY protein inhibits their own mRNA’s transcription by binding to the E box enhancer of the CLOCK/BMAL1 heterocomplex (Equations 2 and 10). As included in Supplementary Table 1 , the parameters for the core and the shell are adapted from our previous work (45), the suffix “m” denotes parameters in the core compartment, and the suffix “e” denotes parameters in the shell compartment. In the SCN, neurons communicate with each other via the neurotransmitters (46). The neurotransmitters VIP and AVP release is activated by cytosolic PER/CRY proteins in the core and the shell (21, 47) (Equations 9 and 17). The model includes two types of inter-neuronal couplings: (a) intra-subpopulation coupling, which includes the communication within the core and the shell; (b) inter-subpopulation coupling, which includes the coupling signals from the core to the shell. The intra-subpopulation coupling is accomplished by the neurotransmitter secreted within its subpopulation (i.e., VIP for the core and AVP for the shell). The inter-subpopulation coupling is achieved by VIP signals secreted from the core. Both neurotransmitters are hypothesized to induce activation effects on Per/Cry mRNA transcription (Equations 2 and 10). The intra-subpopulation coupling is modeled as a constitutive activator term (21, 48). The inter-subpopulation coupling is introduced by an additional term in Equation 10, which accounts for an independent neurotransmitter-driven transcription process apart from the CLOCK/BMAL1 driven transcription (49). Light entrainment of the core is modeled by an additional term in Equation 2 to describe the independent photic activation effects on Per/Cry mRNA transcription (45). Although light signals are mainly transduced by the core, it is strongly speculated that the core is not the only light-responsive subregion. Studies have shown that the entire mouse SCN receives dense innervation from retinal ganglion cells (50). However, due to the dense retinal innervation pattern, deciphering the neuron identity innervated by retinal ganglion cells has been difficult, and little is known about which cell types in the SCN receive the retinal input (51, 52). Considering the focus of the present study, we modeled the light transduction pathway in a conventional manner, where the core receives the optic input, and the shell receives input from the core (2, 53). The model of the HPA axis is adapted from our prior works (43, 45, 54), where its entraining effects are achieved as AVP’s (ensemble average secreted from the SCN shell, Equation 18) down-regulation on the CRH secretion (Equation 19). CRH further regulates the release of ACTH (Equation 20) and eventually induces glucocorticoid (CORT) secretion (corticosterone in this work, Equation 21). Following the transcription and translation of glucocorticoid receptors (Equations 22-23), CORT binds to its receptor (GR) in the cytoplasm of target cells, forming the receptor/CORT complex (DR) (Equation 24). DR subsequently translocates to the nucleus, resulting in the formation of the nuclear-activated receptor, glucocorticoid complex (DR(N)) (Equation 25). DR(N) exhibits receptor-mediated inhibition on CRH and ACTH (Equations 19 and 20) and accounts for the negative feedback loop of the HPA axis. Admittedly, the current HPA model does not provide a complete representation of all the underlying dynamics regulating seasonal changes in the circadian timing system. For example, seasonal photoperiod acts on the pineal gland to secret different levels of melatonin, resulting in seasonal changes in the HPA axis receptor, which may be the pathophysiological basis for the onset of seasonal affective disorder (55). Moreover, the ultradian oscillation of the HPA axis is neglected (56). However, considering the focus of the present work, our model will exclusively investigate the circadian interaction between the SCN and the HPA axis. Studies have shown that many essential properties of glucocorticoid rhythms can be explained using circadian limit cycle oscillators of the HPA axis (39, 57). Therefore, we expect the critical features of our simulation results to be preserved when incorporating the system with more complex components. Single cell model in the core: Single cell model in the shell: HPA axis: Glucocorticoid receptor pharmacodynamics
The SCN consists of a heterogeneous ensemble of cells whose free-running periods and phases vary significantly (58). To account for the heterogeneity of the SCN, we simulated cell-cell variability using Sobol sampling (± 3% ) of the parameters of Equations 2-17, which reflect the stochastic expression of intracellular activities. While alternative approaches such as the Gillespie or Chemical Langevin equations exist (41, 42, 59), in the present study, we chose parameter sampling as a simple method to generate a distribution of heterogeneous cells. The nominal parameter sets for the core, the shell, and the HPA axis are adopted from our previous work (45), with their values and descriptions included in Supplementary Table 1 . To simulate the experimentally observed SCN topology (47, 60), we randomly distributed the heterogeneous core and shell populations in a lobe-shaped area where the shell surrounds the core. The light-receptive core region contains approximately 25% of the neurons in the SCN, and the light-insensitive shell region processes the remaining 75% of the neurons (61). Consistent with such a ratio, we created N and 3N neurons for the core and the shell, respectively, as N represents the total number of neurons in the core. We acknowledge that the two-dimensional geometry and the distinction of core/shell region are oversimplified since the distribution of three-dimensional SCN neurons shows a more complex spatial organization (62, 63). However, since our study is a first attempt to simulate the interplay between the coupled SCN network and the HPA axis, we aim at a manageable representation of the SCN compartment. Moreover, this simplification strategy was also suggested by several previous SCN models (9, 21, 27).
To investigate the effects of coupling topologies on SCN synchronization, both the distance-dependent and probability-dependent networks have been modeled. For the distance-dependent coupling model, all the neuronal connections (including core-core, shell-shell, and core-shell) are diffusible, and neurons receive coupling signals depending on their location. The network topology is defined by a binary coupling term A ij . If there is a directed link between neuronal node j and i , A ij =1 ; otherwise, A ij =0 . We define A ij =1 if the physical distance between nodes i and j is smaller than a predefined threshold distance d , while A ij =0 if the distance is larger than d . For the probability-dependent coupling model, the intra-subpopulation connections (core-core and shell-shell) are described as distance-dependent, as we assume the released neurotransmitters to have a more significant effect on adjacent cells within the core and shell subpopulations. The inter-subpopulation connections (core-shell), on the other hand, are determined based on a probability P PT which describes the likelihood that two nodes are connected. To capture the stochastic feature of the core-shell coupling, we consider randomly changing inter-subpopulation topologies during individual simulations. We define the interval under which the topology changes as a “switching period”. For instance, a 12 h switching period means that the core and shell connections are randomly re-organized every twelve hours. Intra-subpopulation neurotransmitter signals received by neuron i were computed using Equations 26-29. For simplicity, we only consider unidirectional coupling for all the core-shell connections, following the experimental findings that the core projects densely to the shell while the shell projects sparsely to the core (64). The inter-subpopulation couplings sensed by each shell neuron are described in Equation 30. To model the experimentally observed differences in synchronization degree of the SCN neuronal oscillators under LP and SP, we set the threshold distance across the core and the shell (D PT , PT stands for photoperiod) in the diffusible coupling model and the probability (P PT ) in the probability-dependent coupling model as two tunable parameters dependent on photoperiod. Equations 32 and 34 link the two parameters to the photoperiod changes. As D PT or P PT rises with decreasing photoperiod, more core-shell connections will exist during the short photoperiod. It should be noted that while the results of two models run in different seasons will not be quantitatively identical due to the way the model equations were constructed, the overall trends of changes in synchronization degree, phase, and amplitude predicted by the two models are qualitatively similar. Accordingly, the binary coupling terms in the distance-dependent and probability-dependent models are calculated by Equation 31 and Equation 33, separately. The sum of local concentration of neurotransmitters is scaled using the factor , which represents the number of signals received by neuron i . N and 3N describe the total cell numbers in the SCN core and shell sub-populations. d core and d shell are constants defined as the threshold distances inside the core and the shell. Distance-dependent network: Probability-dependent network:
In the SCN, the synchronization of heterogeneous oscillations through inter-neuronal communication is essential for generating a consistent physiological rhythm (8). To study how the SCN topology and photic inputs affect the state of the SCN neurons, we evaluate the degree of synchronization among the SCN neurons. We incorporated an order parameter R syn for Per/Cry mRNA, representing the ratio of the mean-field variance over the mean-variance of each oscillator (65). In Equation 35, y kj is the time vector for cell j , component k , n is the total cell number (n=N for the core, n=3N for the shell), is the average time vector across a population of n cells, while <y k,j > represents the temporal averages of the oscillator. All R syn values in the present work have been estimated over a 50-day period once the system reaches a steady-state. When the cells are entirely desynchronized, R syn has a value of zero, and when the cells are fully synchronized, R syn has a value of one (24). The peak time is used to identify the period and phases. We calculated the mean time gap between two consecutive peaks to estimate the period and the time gap between the peak time at steady-state and 12 am (reference phase) to estimate the phase. The phase on angular coordinates was computed using the formula Φ=(2π.Δt)/T , where Δt is the time lag between the peak and 12 am and T is the period of the calculated component.
The network schematics of distance-dependent and probability-dependent networks are represented in Figures 2A, B , respectively. Within the core and the shell, the existence of intra-subpopulation couplings depends on the physical distance between two neurons, as the blue and red dashed lines denote, respectively. In the distance-dependent network, neurons in the shell receive VIP signals only if they are within the threshold of the core. While in the probability-dependent network, the inter-subpopulation couplings are randomly distributed across the core and the shell. Before proceeding with the model predictions, it is critical to understand the role the number of neurons in each SCN compartment plays. To better describe the influence of population size, we first analyzed the synchronization at the SCN as a function of population size in both the core and shell. For small population sizes, the degree of synchronization was significantly dependent on the number of cells in both the core and the shell, as shown in Figure 3 . Based on our model, cell entrainment is dependent on the entraining signal as well as the cell-cell communications. Since the core is directly entrained by the zeitgeber, the number of cells plays a weaker role. As such, a very small critical number of core neurons of about N≈20 cells is enough to produce an entrained population of cells. The picture in the shell is somewhat more complicated. Shell neurons receive the weaker in strength, core-produced VIP signal which acts as a “global” entrainer, as well as signals from other shell neurons. We observe, in Figure 3 (right), that R syn is dependent on the nature of core-shell communication. Compared to the distance-dependent network, the probability-dependent network is less sensitive to population size changes, as R syn establishes a slower increase as a function of the population size. This observation is likely due to the fact that in the probability-dependent network, a more coherent fraction of shell neurons receive entraining signals from the core, with their location being distributed all over the shell. Shell cells that receive signals from the core in the distance-dependent network, on the other hand, are only distributed within the area close to the core. Therefore, a larger cellular population is required for the probability-dependent network to achieve a similar level of synchronization compared to the distance-dependent network. Compared to an abrupt increase in the core, the degree of synchronization increases more gradually in the shell since its corresponding network receives entraining information indirectly through the core. The standard deviation in R syn computed from 50 network realizations for each N was relatively large for shell cell numbers smaller than 180. The average degree of synchronization was nearly independent of population size for N core >20 and N shell >300 cells. R syn values in the two topologies show a similar trend of increase as cell number increases. Experimentally, electrical recordings of the SCN population pattern usually show a near sinusoidal pattern (66). Recordings of varied population sizes revealed that roughly 50 neurons are required to produce the measured population waveform. With fewer neurons in the recording, the pattern is not entirely representative of the actual neuronal distribution within the SCN (11). In accordance with such findings, our results indicate larger cell populations enhance the phase synchronization of the heterogeneous neurons and are necessary for SCN to generate a robust outcome. Therefore, our simulation predictions should not be inadvertently biased when cell populations are sufficiently large. To compare the synchronization efficiency of the two proposed networks, we calculated changes in the average number of connections per shell neuron as a function of two adjustable parameters: normalized distance threshold of the distance-dependent model and probability P PT of the probability-dependent model. In Figure 4A , the average connection number increases monotonically as the distance threshold or probability increases. While the average connection number increases linearly with probability, the average connection number exhibits a sigmoid increase with the distance threshold, affected by the locations of the SCN core and shell neurons. We calculate R syn under different distance thresholds and probabilities. Compared to the diffusible network, R syn increases rapidly in the probability-dependent network, as shown in Figure 4B . Specifically, an abrupt increase in the degree of synchronization was observed for 0<p<0.2 . An asymptotic R syn value was achieved at p=0.4 in the probability-dependent model, while a similar level of synchronization was achieved for a normalized distance of 0.6 in the distance-dependent model. We further plotted R syn versus average connections per cell for two networks, as shown in Figure 4C . R syn increases rapidly in the probability-dependent network and shows nearly complete synchronization under a relatively low connection number compared to the distance-dependent network. The predicted asymptotic R syn of 0.86 was close to the value calculated from SCN slice data (67). It should be noted that the lack of full synchronization (R syn =1 ) is due to the cellular heterogeneity of the population rather than the insufficient coupling of the neurotransmitters. Our results indicate that a minimum number of core-shell connections is necessary and sufficient to mediate synchronization in the heterogeneous population in a probability-dependent coupling network. To investigate the underlying reason for the probability-dependent network being more efficient in synchronization, we calculated the degree distributions of the connecting numbers with different average connection numbers of two networks, as shown in Figure 5 . In the diffusible network, the number of connections follows a relatively uniform distribution when the connections are not the highest or the lowest (in our case, the total node number is 60, so the highest connection is 60 while the lowest connection is 0). When the average connection number increases in the distance-dependent network, the highest distribution frequency shifts from the lowest connection (0) to the highest connection (60), with the frequencies in between nearly unchanged. In the probability-dependent network, the stochastic nature of coupling induces a normal distribution, with more cells receiving at least one connection from the core. As the average connection number increases, the distribution of the probability-dependent network shifts toward the right, with the distribution pattern unchanged. Taken together, our findings show that the distribution of connections plays a critical role in a population’s degree of synchronization. Compared to the biased distribution pattern induced from the distance-dependent network, the distribution of the probability-dependent network results in more cells receiving a moderate level of the signal, which leads to a higher synchronization. According to our simulation results, a fundamental advantage of the probability-dependent network is that it promotes synchronization across cell populations with a smaller number of average connection degrees per cell. Therefore, we consider the probability-dependent network a low “energy budget” network (67). To character the SCN network in a more energy-efficient manner, we will focus our discussion on the simulation result obtained by the probability-dependent coupling network.
The existence of neuronal communication within and across the two subpopulations of the SCN is an important and novel aspect of the model. Understanding the system’s inherent synchronization processes is crucial to uncovering underlying physiological relationships. Since at the ensemble level, VIP and AVP, secreted by the core and the shell respectively, serve as the SCN’s output that entrains the downstream compartments, we use VIP and AVP signals as our representative variables for the core and the shell compartments. To understand the stochastic nature of the system independent of entrainment, we investigated the distribution of single-cell phases and periods of the core and the shell in the absence of coupling. In Figures 6A, D , we examine single-cell phase distribution when the coupling strengths (v c1 , v c2 , k c in Equations 2 and 10) are set to 0 for all cells. In both core and shell, the clocks maintain their oscillatory characteristics displaying a broad distribution of periods and phases with the degree of synchronization values close to zero ( Supplementary Table 2 ), indicating little to no synchrony among the SCN neurons. Cell phases in both the core and shell are uniformly distributed through the entire regime from 0 to 2π (blue curves in Figures 6H, J ). In terms of periods, the stochastic dynamics of individual neurons produce a normally distributed pattern of values with a mean period of 24.21 h for the core and 23.84 h for the shell (blue curves in Figures 6G, I ). Because of this asynchrony, dampened, low-amplitude ensemble average profiles for VIP and AVP are observed. This result indicates that the stochasticity introduced by sampling leads the individual cells to adopt random phases and periods, thus exhibiting a damped ensemble AVP and VIP averages in the absence of an entraining signal. To explore the entraining impact of the intra-subpopulation coupling induced by neurotransmitters, simulations were performed with the non-zero values for inter-neuronal coupling strengths v c1 and v c2 . In Figures 6B, E , the intra-subpopulation communication within the core and the shell induced by VIP and AVP signals significantly augmented the entrainment level indicated by higher R syn values of VIP and AVP ( Supplementary Table 2 ). The distributions of phase and amplitude were more concentrated, with an intrinsic period of 24.05 h for the core and 23.69 h for the shell (red curves in Figures 6G–J ). The slightly shorter (1.5% ) intrinsic period of the shell than that of the core is in accordance with experimental findings (68). Figures 6C, F investigated the synchronization effects of inter-subpopulation (core-shell) couplings. The coefficient k c , which quantifies the VIP coupling force of the shell, is set to its non-zero value, imposing entrainment of VIP secreted by the core on the shell subpopulation. VIP affects the shell ensemble in two ways: (1) it induces a more robust pattern with a higher ensemble amplitude (compare Figures 6E, F ); (2) it imposes the period of the core onto the shell clocks so that a larger amount of the shell population shares the same period with the core (compare the red and yellow curves in Figure 6I ). As a result, the standard deviation of cell phases in the shell gets smaller than in the case when the inter-subpopulation coupling forces are absent ( Figure 6J ), indicating that the shell maintains a higher degree of synchronization with projections from the core. Taken together, these results provide further evidence that the connections between the core and the shell in the probability-dependent model promote synchronization among sparsely connected shell populations in the absence of light. Our simulations suggest that the observation that separating the dorsal and ventral SCN abolishes synchrony in the dorsal SCN (12) could be related to the role of core-shell connections in increasing the synchronization of the shell oscillators, possibly by selectively cutting VIP connections projecting from the ventral SCN to the dorsal SCN. To better understand the interplay between intra- and inter-subpopulation coupling, we tested how R syn in the shell varies as a function of its intrinsic coupling strength both with and without the external VIP signals. As shown in Figure 7 , when AVP (i.e., v c2 ) is the only coupling force of the shell ensemble (i.e., k c =0 ), R syn increases as the AVP coupling strength increases, reflecting an increase in the robustness of the intrinsic shell ensemble. We calculated the synchronization degree when the internal and external coupling forces both exist in the system, with different VIP inter-subpopulation coupling strengths. When the VIP coupling strength is relatively high (k c equals four times its nominal value), the synchronization of the system is almost complete; thereby, no significant change was observed in R syn as the AVP coupling strength increases. When the VIP coupling strength is moderate (k c equals two-three times its nominal value), an increase followed by a decrease in R syn was observed as AVP coupling strength rises. At the beginning stage, a weak intrinsic coupling facilitates the system to get entrained by an external signal. However, as the intrinsic coupling strength grows, it can become a resistant force that contradicts the external coupling strength. When the VIP coupling strength is relatively weak (k c equals its nominal value), we observed another increase in R syn following its previous increase and decrease phases, indicating that the intrinsic coupling became the principal driving force. Therefore, our results reveal an antagonistic correlation between the intrinsic and external coupling forces. Circadian timing systems are both robust to external changes in the environment and plastic to adapt to environmental changes such as temperature and nutrient conditions (69). Our results reflect a complex relationship between the robustness and flexibility of the multi-cellular oscillatory system. More details regarding this trade-off will be included in “Discussion”.
We further tested the response of the SCN when entrained by different photoperiods (PP) in a way mimicking seasonality (54, 70). Figure 8 depicts the photoperiod-dependent circadian profiles for the SCN core and shell under LP (16L/8D) and SP (8L/16D). Entrained by L/D cycles, the synchronization of the SCN is significantly increased with the ensemble oscillations of both the core and the shell adopting periods (~ 24 h ) equivalent to the external photoperiod. Under LP, both the core and the shell ensembles showed smaller degrees of synchronization compared to that of SP, indicated by the broader phase distributions of VIP and AVP profiles. Therefore, the model consistently identified a greater asynchrony level under LP, which agrees with the experimental findings (71). To obtain additional insights into the ways photic signals propagate from the core and shell to the HPA under different PP, we examined the circadian profiles of three compartments (core, shell, and HPA axis) as photoperiods gradually shift from SP to LP (PP = 8, 10, 12, 14, 16 h ). As depicted in Figure 9 , the amplitude of three representative components, VIP (ensemble), AVP (ensemble), and CORT, decreased as the photoperiod increased. Our model predicts a higher AVP ensemble oscillation inducing higher corticosterone amplitude in SP. Such an increase of CORT in winter (shorter PP) was also reported in both experimental and modeling works (32, 70). Moreover, the peak time of the core exhibits phase delay as photoperiod increases, while that of the shell and the HPA axis firstly exhibit a phase delay and then show slight phase advance, with the most delayed phase achieved at equinoctial zeitgebers (PP = 12 ). The peak time of CORT during SP is slightly delayed compared to LP. This delay is manifested as the peak phase shift to the right during SP for the CORT oscillations shown in Figure 9F . Experimental studies reported a phase advance of cortisol under long-photoperiod summer compared to winter (72). Moreover, previous modeling results also suggested a phase delay occurred in the HPA axis and a phase advance in the SCN populations during SP (9, 70). Taken together, our model predicted glucocorticoid phase shifts under different photoperiods in accordance with previous findings.
Stochastic network reorganization is constantly observed in physiological systems (73), complicating the functional understanding of the SCN since the network changes over time (8). To investigate the interplay between the spatiotemporal architecture reorganization of the SCN and the synchronization of the physiological features of the circadian system in our probability-dependent coupling model, we also considered the possibility of changing topologies during the course of the simulations. We calculated R syn under different topology switching periods which describe the interval under which the core-shell connection changes (e.g., a 2 h switching period means that the core-shell couplings are reorganized every two hours). Figure 10 depicts how the degree of synchronization changes as a function of the switching period under different photoperiods. For each switching period under a certain photoperiod, 50 network realizations were calculated to account for statistical network behaviors. For a short photoperiod (PP=8 ), 50 realizations of the probability-dependent coupling produced an almost complete degree of synchronization with a slight standard deviation under all the simulated switching periods. This indicated that switching topologies in a highly synchronized system will not significantly affect its synchronization behavior. Moving to larger photoperiods (PP=10, 12, 14 ), an overall drop in R syn and increase in standard deviation was observed as the value of the switching period increases. These results indicate that when the synchronization is not complete, a fast-switching topology process in the SCN probability-dependent network helps achieve a higher degree of synchronization with a minimum connection number. Moreover, as the switching period increases, the synchronization status calculated from different realizations tends to differ more from each other. The synchronization processes in longer photoperiods are less consistent considering the heterogeneities in various network realizations. Under long photoperiod (PP=16), when R syn is the lowest with the highest standard deviation, changing shifting frequencies in the network failed to impose any significant effect on the degree of synchronization. The neural network circuitry and synchronization behavior are currently understudied experimentally, and additional research is required to examine the SCN network thoroughly (73). Our model enabled us to study the spatial-temporal network of the SCN circadian oscillators. The results suggest a potential mechanism for the system to increase its synchronization efficiency. More experimental investigations on this in silicon observation are anticipated to be done in the future.
The synchronization of the hierarchical circadian timing system is critical for maintaining homeostasis. The central SCN pacemaker is a heterogeneous organization consisting of different subpopulations classified by their anatomical connections and neurotransmitter identity (17). The entrainment signals from the SCN are conveyed to the periphery through the central effector (HPA axis) that controls organismal behavior. Dysregulation of the system through alterations of the timing of zeitgebers (jet lag, shift work, disruptions of feeding time, etc.) can lead to the incidence of a variety of pathologies, including metabolic syndrome, diabetes, obesity, tumorigenesis, and auto-immune disorders (74, 75). Since the light/dark ratio has been identified as one of the factors that affect the synchronization of circadian entrainment, the complexity of the dynamic response of the circadian timing system under different photic conditions requires a systematic approach to investigate a multitude of interactions. As an extension of our previous work (45), the present study integrated a multi-compartment SCN model with the downstream HPA axis, taking into account the flexible inter-SCN coupling with varying photoperiods. Utilizing the model, we tested the synchronization efficiency of two potential SCN inter-subpopulation coupling mechanisms, the temporal topology changes of the SCN, as well as the circadian dynamics of the HPA axis under different LD ratios. While the mechanisms underlying interactions between neuronal nodes in the SCN network remain unclear, mathematical modeling can help elaborate on the effects of potential topologies and examine the collective behavior of such complex networks (28). We modeled both distance-dependent and probability-dependent coupling topologies for the SCN to adjust its phase distributions according to photoperiod changes since both diffusible and synaptic signals can contribute to the coupling mechanisms behind the core-shell communication (29). By comparing the two mechanisms, we hypothesize that the probability-dependent network requires fewer connections to achieve the same degree of synchronization compared to the distance-dependent model ( Figures 4 , 5 ). Wiring distant neurons across the brain is a costly mechanism that depends on the network coupling volume (76). Considering the “energy budget” (numbers of connections required to generate a certain level of synchronization) of connecting neuronal cells throughout the brain due to the signal transmission (77), probability-dependent coupling networks are therefore predicted to provide more efficient connectivity patterns while maintaining the same performance in terms of rhythm generation and cellular synchronization of the SCN. Our model provides alternative explanations for the distinct synchronization of the SCN neurons under different photoperiods by considering the diffusible distance threshold or the connection probability across the core and the shell ( Figure 2 ). Both the distance threshold and connection probability are hypothesized to decrease as the photoperiod increases (Equations 32 and 34). Thus, fewer cells in the shell can receive signals from the core, therefore becoming less synchronized. While the real SCN network is more complex than the proposed models and can include both mechanisms at the same time, our results suggest that the probability-dependent coupling might be more realistic for the SCN network to diminish its energy budget. By exploring the intrinsic dynamics of the SCN inter-neuronal coupling, our results indicate that autocrine neurotransmitter activation is sufficient to sustain oscillations in the SCN topology ( Figure 6 ). The intra-neuronal coupling strength induces a mild increase in R syn , which agrees with the fact that the behavioral and physiological rhythms regulated by endogenous circadian clocks persist even without environmental cues such as light and food (78). In rodents, the transcription of Cry1 showed a robust and sustained circadian rhythm without damping circadian amplitude in both LD and DD (79), meaning that the SCN has topographically structured coupling mechanisms and can maintain synchronization even in complete darkness. In accordance with such findings, our model predicts highly synchronized SCN network can be achieved by the coupling effects of its neurotransmitters. This self-sustained synchronization is essential for the coherent circadian output of the SCN to regulate the downstream activities. Biological systems exhibit both robustness to environmental changes and plasticity to adapt to external conditions (69). By comparing the synchronization degree of the shell under different combinations of inter-subpopulation and intra-subpopulation coupling strengths, we identified a trade-off between robustness/rigidity and plasticity/flexibility of the SCN network induced by the antagonistic effects of the intrinsic and external coupling forces (intrinsic force: AVP intra-subpopulation coupling; external force: VIP inter-subpopulation coupling, Figure 7 ). When the intrinsic or the external force is the principal driver, the degree of synchronization increases as the intrinsic coupling force rises. During the range when the two forces play a relatively equivalent role, the increase of the intrinsic force increases the rigidity of the system, making it more difficult to get entrained by the external signal. This indicates that the SCN oscillator must be strong enough to withstand external noise while also being flexible enough to perform as an entrainable clock under varying photoperiods (e.g. seasons). Experiments reveal that compared to oscillators in the periphery, the intercellular communication among SCN cells through the neuronal and diffusible networks is the unique feature of the SCN (2). On the other hand, almost no coupling exists in the periphery, which ensures that they stay flexible when responding to signals from the SCN (80). Therefore, by strengthening its intrinsic coupling strength, the SCN clock developed rigidity to act as a noise filter to the external entraining signals. By modifying the network structure in terms of inter-subpopulation neuronal connections, we modeled the system’s response to photoperiod changes. Our results show a non-monotonical pattern for phases in the shell and HPA axis to shift as the photoperiod progresses from LP to SP ( Figure 9 ). Therefore, our model hypothesizes a strategy where the response of the HPA axis under different photoperiods is regulated by the distribution of the AVP ensemble outcome. In line with average ensemble amplitudes, R syn metrics of the core and shell populations further testify to the seasonal differences in synchrony, with PP 8 exhibiting the most synchronous oscillations and PP 16 the most asynchronous. Interestingly, as the photoperiod increases, synchronization levels in the shell first show a slight decrease and exhibit an abrupt drop between PP 12 and 16, indicating that the spring, autumn, and winter seasons are more synchronized than the summer season, in accordance with results reported previously (54). Consequently, the higher degree of synchronization of the shell under short photoperiods results in a higher level of CORT. Seasonal changes can cause misalignment of the physiological systems. For instance, seasonal affective disorder (SAD) is characterized by a cyclical pattern of depression that occurs during short photoperiods and then fades during longer photoperiods (81, 82). Different individuals react differently to seasonal changes and can develop differential severity and symptoms of SAD. For example, a fraction of SAD patients exhibits a variety of secondary symptoms, such as sleep disturbances, lethargy, and carbohydrate cravings, while others are less sensitive to seasonal changes (83). SAD also affects men and women in different ways, with women three to five times more likely than males to be affected (84). We anticipate that these individualized differences can be accounted for by the parameters in Equations 32 and 34, which affect the plasticity of the system during seasonal changes. The plasticity of their SCN networks is hypothesized to be lower in people who are more resilient to seasonal variations. Further theoretical simulations regarding this will be conducted in the future. Studies show that individual SCN neuron dynamics are not static. For example, wild-type SCN neurons can randomly switch from rhythmic to nonrhythmic and vice versa (2). Furthermore, even mutant SCN neurons occasionally display infrequent, intermittent PER2 oscillations, suggesting that rhythmicity is a stochastic event (85). Inspired by the stochastic nature of the circadian oscillators, we hypothesized that the coupling network which connects the core and the shell could change its topology over time. By modeling the temporal topology shifts of the core-shell coupling ( Figure 10 ), we predict a positive correlation between the topology switching speed and the degree of synchronization when the synchronization is not complete. The fundamental rationale might be that by changing the inter-subpopulation topology quickly, more cells in the shell will be able to receive signals from the core over time. Therefore, our model suggests a potential mechanism for the SCN to achieve a higher degree of synchronization when the probability-dependent connection numbers between two subpopulations are insufficient. In summary, mathematical modeling can provide substantial insight into physiological systems (86, 87). Our model attempts to provide explanations of the variations in SCN and HPA axis responses as photoperiod changes. While testing two possible topologies for the SCN core-shell coupling networks, the probability-dependent network was suggested as a low “energy budget” architecture. We also identified the antagonistic effects of inter-subpopulation and intra-subpopulation coupling on the SCN network’s resilience and stiffness. Moreover, the temporal topology switch of the network has been predicted to facilitate multi-cellular synchronization for the first time. Since our model incorporates the essential hierarchical structure of the SCN, the HPA axis, and their response under different photoperiods, it should be a valuable tool for multiple additional studies, including the investigation of the individualized seasonal response of diseases; the inclusion of the downstream entraining of peripheral clocks; and the simulation of seasonal jetlag.
The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.
YL developed the model, designed and conducted the calculations, analyzed the results, and prepared the manuscript. IA conceived the study, analyzed the results, edited manuscript. All authors contributed to the article and approved the submitted version.
The authors acknowledge financial support from NIH GM131800.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648570 | Liu Yang,Wen Li | Giant inflammatory myofibroblastic tumor of the hypopharynx: an unusual cause of acute upper airway obstruction in a 6-year-old child | 13-02-2022 | otorhinolaryngologic diseases,diagnostic imaging,emergency service,hospital,pathology,surgery,plastic | Giant inflammatory myofibroblastic tumor of the hypopharynx: an unusual cause of acute upper airway obstruction in a 6-year-old child
Acute upper airway obstruction (UAO) in children is a common disease in the emergency department. It is life-threatening and usually caused by laryngospasm or laryngeal edema owing to infection, allergy, or a foreign body.1 In addition, pediatric UAO can also be the result of laryngopharyngeal space-occupying lesions, such as laryngeal papilloma, lymphangioma, hemangioma, or cyst. Malignancy is uncommon. Inflammatory myofibroblastic tumor (IMT) is a mesenchymal tumor consisting of differentiated spindle fibroblast cells with plasma cells and/or lymphocytes infiltration. The most common sites of IMT are the lungs, gastrointestinal tract, urogenital tract, abdominal cavity, viscera, central nervous system, upper respiratory tract, and soft tissue. IMT in the head and neck region accounts for only 14%–18% of extrapulmonary lesions, of which the orbit and nasal sinus are common locations.2–4 According to the reported cases of laryngeal IMT (fewer than 50 cases), it is usually confined to the glottic region. A lesion with supraglottic and hypopharyngeal involvement is extremely rare and only five cases have been reported in the hypopharynx to date. These cases occurred in adults, including only one female patient.5–9 We describe an extremely rare case of giant IMT in the hypopharynx of a female child with respiratory distress as the initial symptom. A 6-year-old girl was referred to the emergency department of our hospital with progressive dyspnea, stridor, cyanosis, and dysphonia. Hoarseness was not obvious, but a change in her voice and snoring had been present for 4 months. Acute upper respiratory infection, anaphylaxis, and airway foreign body were excluded following detailed postadmission medical inquiries. Compared with children of the same age, she showed stunted growth. Oxygen therapy did not completely relieve her dyspnea, and transient loss of consciousness was observed; therefore, tentative intubation was conducted but failed. A huge lump blocking the glottis was found during attempted intubation; thus, an emergency tracheotomy was performed. Subsequent laryngoscopy showed an exophytic tumor located in the right piriform sinus partially covering the laryngeal entrance (figure 1A). Contrast-enhanced CT revealed a 30 mm×20 mm×20 mm mass with evident homogeneous enhancement occluding the entrance to the larynx (figure 1B). Routine laboratory tests showed no abnormalities suggestive of infection or immune disorders. The preliminary diagnosis was malignancy of the hypopharynx. The patient underwent surgical exploration and an ill-defined tumor was found occupying the right piriform sinus and esophageal inlet. The cricoarytenoid joint was also involved. The tumor and the involved larynx were excised with the help of frozen sections to determine the safe margin (figure 1C). Ipsilateral enlarged cervical lymph nodes were also observed. The patient’s postoperative recovery was uneventful. Macroscopically, the tumor was granular and fragile, covered by incomplete mucosa, with a white to tan gross appearance (figure 1D). Microscopically, spindle cell proliferation and inflammatory cell infiltration were observed in the tumor, while reactive hyperplasia in the lymph nodes was noted (figure 2A, B). Immunohistochemically, the spindle cells were positive for smooth muscle actin, myo-specific actin (figure 2C, D), and transducin-like enhancer of split 1 (weakly positive), and negative for anaplastic lymphoma kinase (ALK), epithelial membrane antigen, desmin, H-caldesmon, myogenin, P63, CD34, S-100, and Epstein-Barr virus-encoded RNA in situ hybridization. The proliferation marker protein Ki-67 (ki-67) index was 5%. A hypopharyngeal IMT was finally diagnosed. Three months after surgery, decannulation was achieved. She was able to communicate in a hoarse voice after decannulation. There was no evidence of recurrence during the 4-year follow-up period and the quality of the voice gradually improved. Acute UAO usually manifests as dyspnea and stridor and is more likely to occur in pediatric patients. In this condition, intrinsic and extrinsic etiologies should be differentiated so that immediate effective treatment can be given to the patient. Anaphylaxis, airway foreign body, acute epiglottitis, and laryngotracheitis should be considered initially.1 In addition, upper airway space-occupying benign lesions, such as laryngeal papilloma, hemangioma, lymphangioma, and retropharyngeal abscess, should also be considered. Although malignancy is uncommon, it can also be fatal if undiagnosed. A detailed medical history and careful physical examinations are important for diagnosis. Imaging and endoscopic examinations should be carried out as soon as possible if conditions permit. Early identification of airway stability or lability is vital in the management of children with acute UAO.10 Pediatric tracheotomy should be carried out as an elective procedure in extreme emergent situations, such as in patients with progressive aggravation of hypoxia after routine oxygen therapy, hormonotherapy, and intubation have been performed. In the present case, the tumor occluded the laryngeal entrance and made routine oxygen intake impossible. IMT was previously named inflammatory pseudotumor, plasma cell granuloma, pseudosarcomatous inflammatory lesion, and so on. It is now recognized as a real tumor with uncertain behavior and etiology. Recurrence and metastasis of IMT have been reported in several isolated cases, and clonal cytogenetic aberrations have also been detected.4 9 IMT is very rare in the upper aerodigestive tract, and the larynx is the most common site.4 Hoarseness is the most common symptom in most reported laryngeal IMTs with glottic involvement, while acute UAO is extremely rare except in subglottic locations. In pharyngeal IMTs, dysphagia and globus sensation may be relevant mild symptoms. Unlike adults, IMT is very rare in children, especially in the head and neck region. Because pediatric IMTs usually mimic malignancy, such as sarcoma or lymphoma, surgery is the main therapeutic approach to cure the disease or to achieve a final diagnosis.11 The differential diagnosis of IMT should include spindle cell tumors with inflammation, such as IgG4-related diseases (IgG4-RD), inflammatory myomatous polyps (IFPs), inflammatory leiomyomas, etc. Approximately 16% of IgG4-RD cases may present as an inflammatory pseudotumor and these cases are usually accompanied by elevated levels of IgG4 in serum and tissues.7 Corticosteroid therapy is usually effective in these cases. IFP is characterized by eosinophil infiltration and immunohistochemistry shows CD34 positivity, which can differentiate it from IMT. Leiomyomas with lymphocytic infiltration are most easily confused with IMT. The spindle cells in leiomyoma are smooth muscle cells rather than myofibroblast cells. The degeneration of interstitial mucus is not obvious, and inflammatory cells are mainly small lymphocytes, whereas in IMT plasma cells and neutrophils are the main cells. Hypopharyngeal IMT is extremely rare and only five case reports (four English articles and one Chinese article) were found through a PubMed search (table 1). The five cases were in one female and four male patients aged 49–74 years (mean 57.8 years). One malignant case was confirmed to harbor the 3a/b variants of the EML4-ALK fusion gene by genetic testing.9 The disease course is from 2 months to 2 years. The lesion usually occurred in the posterior wall of the pharynx. Globus sensation was the most common local symptom, followed by progressive dysphagia. Three patients presented with general symptoms, including weight loss and general malaise. Pharyngeal malignancy was considered in almost all patients prior to pathological results. Three of five patients underwent total resection transorally, and the patient with IgG4 IMT was treated with corticosteroids.8 The case of malignant transformation only received palliative surgical resection and subsequently succumbed to disease progression at 11 months postdiagnosis.8 From the reviewed literature, we concluded that hypopharyngeal IMTs occur mostly in adults and rarely affect respiratory function, and total resection is the preferred treatment for IMT. Although there is no consensus on the treatment of IMTs, complete excision is critical for cure and can result in a favorable prognosis in children. Anti-inflammatory treatment, chemotherapy, steroids, or immunomodulators as potential remedial and effective management strategies are recommended for adult patients who may not be able to undergo possible complete excision or in ALK-negative cases with a higher frequency of metastasis and invasiveness.12 In conclusion, in this child, the initial symptom of a large IMT was acute UAO. This report highlights the importance of differential diagnosis in a child with acute UAO. Following relief of acute UAO, wide excision of the tumor with safe margins should be carried out. |
||
PMC9648586 | Yue Wu,Xiaosi Jin,Yuhao Zhang,Jing Zheng,Rulai Yang | Genetic and epigenetic mechanisms in the development of congenital heart diseases | 29-04-2021 | cardiovascular system,child health,genetics,pediatrics | Congenital heart disease (CHD) is the most common of congenital cardiovascular malformations associated with birth defects, and it results in significant morbidity and mortality worldwide. The classification of CHD is still elusive owing to the complex pathogenesis of CHD. Advances in molecular medicine have revealed the genetic basis of some heart anomalies. Genes associated with CHD might be modulated by various epigenetic factors. Thus, the genetic and epigenetic factors are gradually accepted as important triggers in the pathogenesis of CHD. However, few literatures have comprehensively elaborated the genetic and epigenetic mechanisms of CHD. This review focuses on the etiology of CHD from genetics and epigenetics to discuss the role of these factors in the development of CHD. The interactions between genetic and epigenetic in the pathogenesis of CHD are also elaborated. Chromosome abnormalities and gene mutations in genetics, and DNA methylations, histone modifications and on-coding RNAs in epigenetics are summarized in detail. We hope the summative knowledge of these etiologies may be useful for improved diagnosis and further elucidation of CHD so that morbidity and mortality of children with CHD can be reduced in the near future. | Genetic and epigenetic mechanisms in the development of congenital heart diseases
Congenital heart disease (CHD) is the most common of congenital cardiovascular malformations associated with birth defects, and it results in significant morbidity and mortality worldwide. The classification of CHD is still elusive owing to the complex pathogenesis of CHD. Advances in molecular medicine have revealed the genetic basis of some heart anomalies. Genes associated with CHD might be modulated by various epigenetic factors. Thus, the genetic and epigenetic factors are gradually accepted as important triggers in the pathogenesis of CHD. However, few literatures have comprehensively elaborated the genetic and epigenetic mechanisms of CHD. This review focuses on the etiology of CHD from genetics and epigenetics to discuss the role of these factors in the development of CHD. The interactions between genetic and epigenetic in the pathogenesis of CHD are also elaborated. Chromosome abnormalities and gene mutations in genetics, and DNA methylations, histone modifications and on-coding RNAs in epigenetics are summarized in detail. We hope the summative knowledge of these etiologies may be useful for improved diagnosis and further elucidation of CHD so that morbidity and mortality of children with CHD can be reduced in the near future.
Congenital heart disease (CHD) is a group of disorders attributed by abnormalities in fetal heart and large vessels that lead to actual or potential impairment of cardiac function in infants. CHD is the most common type of congenital defect worldwide and is the most common and the most life-threatening class of birth defects in infants, affecting approximately 1% live births annually worldwide.1 Epidemiological investigations indicate that the overall incidence of CHD varies across countries and continents, and the prevalence of CHD in Asia is higher than that in North America.2 Despite benefits from the remarkable progress in therapeutic strategies of surgery and catheter intervention, CHD is still the principal source of mortality in infants. However, owing to medical, surgical and technological evolutions during the past decades, more than 90% of CHD infants now survive to adulthood.3 Improvement in surgical intervention techniques and perioperative care has dramatically changed the management of these populations with CHD. However, CHD is still a bothersome question owing to its undesirable outcomes and expensive healthcare costs, which bring substantial physiological, emotional and socioeconomic challenges to patients, families and society. According to the final anatomical and pathophysiological complexities, CHD can be classified as mild, moderate or severe. Detailed classification is shown in box 1.4 The prognosis, morbidity and mortality vary with the severity of the anomalies. Despite the rapid advances in medical care and detection technology, the etiology of most CHD remains poorly understood. It is therefore imperative to improve our understanding of the disease mechanisms to reduce the frequent occurrence of CHD. During the past decades a consensus has emerged that both genetic (eg, chromosomal abnormalities, smaller copy number variants and point mutations) and environmental (extrinsic factors, such as teratogen exposure and nutrient deficiencies; intrinsic factors, including maternal disease, illness and viral infection)5 factors are related to the occurrence of CHD. Progress in molecular genetic diagnosis has provided a valuable opportunity to investigate the genetic factors of CHD. Furthermore, a multitude of animal models (eg, mouse, zebrafish, frog and fruit fly) have witnessed the significant effects of genetic etiology of CHD. These in vivo studies on animal models, in turn, have resulted in the identification of numerous structural genes, transcriptional regulators and signaling molecules that are critical for normal cardiac morphogenesis.6
Isolated congenital aortic valve disease and bicuspid aortic disease. Isolated congenital mitral valve disease (except parachute valve and cleft leaflet). Mild isolated PS (infundibular, valvular and supravalvular). Isolated small ASD, VSD or PDA. Repaired secundum ASD, sinus venosus defect, VSD or PDA without residuae or sequellae, such as chamber enlargement, ventricular dysfunction or elevated pulmonary artery pressure.
Anomalous pulmonary venous connection (partial or total). Anomalous coronary artery arising from the PA. Anomalous coronary artery arising from the opposite sinus. AS subvalvular or supravalvular. AVSD, partial or complete, including primum ASD (excluding pulmonary vascular disease). Coarctation of the aorta. Double-chambered right ventricle. Ebstein anomaly. Marfan syndrome and related HTAD and Turner syndrome. PDA, moderate or large unrepaired (excluding pulmonary vascular disease). PPS. PS (infundibular, valvular and supravalvular), moderate or severe. Sinus of Valsalva aneurysm/fistula. Sinus venosus defect. TOF repaired. Transposition of the great arteries after arterial switch operation. VSD with associated abnormalities (excluding pulmonary vascular disease) and/or moderate or greater shunt.
Any CHD (repaired or unrepaired) associated with pulmonary vascular disease (including Eisenmenger syndrome). Any cyanotic CHD (unoperated or palliated). Double-outlet ventricle. Fontan circulation. IAA. Pulmonary atresia (all forms). Transposition of the great arteries (except for patients with arterial switch operation). Univentricular heart (including double inlet left/right ventricle, tricuspid/mitral atresia, hypoplastic left heart syndrome and any other anatomic abnormalitywith a functionally single ventricle). Truncus arteriosus. Other complex abnormalities of atrioventricular and ventriculoarterial connection (ie, crisscross heart, heterotaxy syndromes and ventricular inversion). AS, aortic stenosis; ASD, atrial septal defect; AVSD, atrioventricular septal defect; CHD, congenital heart disease; HTAD, heritable thoracic aortic disease; IAA, interrupted aortic arch; PA, pulmonary artery; PDA, patent ductus arteriosus; PPS, peripheral pulmonary stenosis; PS, pulmonary stenosis; TOF, tetralogy of Fallot; VSD, ventricular septal defect.
To our knowledge, although numerous literatures have discussed the genetic mechanisms of CHD, few have comprehensively elaborated the genetic and epigenetic mechanisms of CHD. In this review, we focus on CHD origin from the etiology of genetics and epigenetics. Chromosomal abnormalities and gene mutations in genetics, and DNA methylations, histone modifications and on-coding RNAs in epigenetics are summarized in detail. Moreover, we expect that rapidly emerging data could provide a further understanding of genetics and epigenetics in the development of CHD and also a basis for further exploring the early diagnosis and individualized therapy of CHD.
Chromosome abnormalities refer to abnormal chromosome numbers and structural aberrations including aneuploidies and copy number variations (CNVs), respectively. Conventional chromosome anomalies associated with CHD were identified half a century ago. A study by Pierpont et al 7 suggested that ~30% of children with a chromosomal abnormality would suffer from CHD. In the following section, we summarize the involvement of chromosomal aneuploidies and CNVs in CHD in detail.
Chromosomal aneuploidy is the earliest recognized genetic cause of CHD and accounts for a great proportion of CHD (table 1). Approximately 50% of individuals born with trisomy 21 have the phenotypes of CHD, ranging from atrial septal defect (ASD)/ventricular septal defect (VSD) to atrioventricular canal lesions.1 2 The prevalence of CHD in newborns with trisomy 13 and trisomy 18 increases to 80%, and the major phenotypes of CHD are heterotaxy, laterality and septal defects.8 9 CHD is observed in approximately 33% of females with Turner syndrome or monosomy X, and the cardiac malformations are usually diagnosed as VSD, coarctation of aorta (CoA), bicuspid aortic valve and hypoplastic left heart.1 Although abnormal X chromosome numbers are rare, they also could result in CHD. For example, males with Klinefelter syndrome or 47, XXY have a 50% chance of CHD with the phenotypes of patent ductus arteriosus (PDA) and ASD.7 Moreover, sporadic 49, XXXXX cases with the phenotypes of ASD and vascular malformations have also been reported.10 At present, chromosomal G-banded karyotype analysis has been applied to detect the anomalies of chromosomes in spite of its limitation for the restrictive base resolution in exploring the tiny abnormalities of chromosomes.
Conventional chromosomal microscopy in clinic could only detect the alterations of structure, numbers of chromosomes and abnormalities of large fragments, causing incomplete diagnosis of CHD. Numerous detection technologies include fluorescent in situ hybridization and multiplex ligation-dependent amplification. Chromosomal microarrays have been applied to explore the submicroscopic chromosomal anomalies and to elucidate the pathogenic mechanisms of CHD, which may become a better approach for diagnosis. CNVs refer to structural aberrations consisting of deletions or duplications, which are too small to be detected by routine karyotype analysis. A few CNVs can alter one or more contiguous genes and then inappropriately affect their expression, leading to the development of CHD.11 12 CNVs can occur de novo in sporadic cases, or they can be inherited familiarly causing complex congenital heart malformations. However, the underlying mechanisms of CNVs in CHD still need to be elucidated. One of the most common CNVs syndromes in CHD, 22q11.2 deletion syndrome (DiGeorge syndrome or velocardiofacial syndrome), is caused by a microscopic deletion on chromosome 22q11.2. Variable phenotypes, such as craniofacial abnormalities, neurocognitive disabilities, palate abnormalities, hypocalcemia and immunodeficiencies, can be observed in this syndrome. The main cardiac malformations of this syndrome contain VSD, arch abnormalities and tetralogy of Fallot (TOF).13 To date, more than 30 gene deletions have been identified in the 22q11.2 locus by sequencing. Deletions in the T-box transcription factor TBX1 account for the major molecular basis of these cardiac malformations.14 Further work revealed that distal 22q11.2 deletion presented atypical clinical features of DiGeorge syndrome; therefore, this deletion was proposed as one of the etiologies of CHD.15 Williams-Beuren syndrome, which is characterized by supravalvar aortic stenosis (SVAS), peripheral pulmononic stenosis, coronary artery stenosis, pulmonary artery sling, developmental delays, typical elfin facies, infantile hypercalcemia and cognitive disability, has been reportedly caused by microdeletion of over 25 genes in the 7q11.23 region.16 It has been known that cardiovascular abnormalities, such as SVAS, can be induced by haploinsufficiency of the elastin gene (ELN).17 Some CNVs encompass previously identified CHD genes, or genes known to be implicated in heart development. For example, 8 p deletion syndrome and mutations in GATA4, a cardiac transcription factor, are reported to be associated with CHD.18 The 8p23.1 delete region that overlaps with the locus of GATA4 could elucidate this causal outcome. However, Kumar et al reported an interesting case of 8p23.3p23.1 deletion and 8p23.1p11.1 interstitial duplication syndrome that a male toddler with global developmental delay, dysmorphic facies, seizures and large doubly committed VSD occurred without the GATA4 gene involvement.19 Duplications at the 8p23.1 locus have also been identified in CHD, including ASD, VSD and TOF.20 21 Other studies have confirmed that the CNVs at chromosome 11q23 were associated with Jacobsen syndrome.22 23 Moreover, several CNVs have been identified from larger cohorts of patients with CHD, including 1q21.1,24–26 4p16.3,27 4q22.1,26 28 9q34.326 29 and 15q11.2.28 All of these CNVs mentioned above are detailed in table 1 for their relationship to CHD in human.
Numerous mutations are implicated in the development of CHD (table 2). Some mutations are identified in pedigrees of CHD, while others are initially observed in sporadic cases of CHD.2 Currently, mutations in more than 50 genes have been found to be associated with CHD by the application of high-throughput sequencing of whole-genome and whole-exome, and many of these affected genes have been confirmed to be involved in transcriptional regulation, signal transduction and cardiac development.
Heart development is regulated by several transcriptional circuits that are members of a core group of transcription factors, including NKX2.5, GATA4 and TBX5.30 31 Therefore, transcription factors have been considered as the prime inducer of CHD. NKX2.5 is the earliest known marker of myocardial progenitor cells in all species.32 The mechanisms of NKX2.5 regulation and its interaction with other transcription factors in early cardiac development have been studied extensively. It has been found that mutations in NKX2.5 could result in variable types of CHD, including ASD, VSD, TOF, hypoplastic left heart syndrome, CoA, transposition of the great arteries, double outlet right ventricle (DORV), interrupted aortic arch and cardiac outflow tract defects.33 Furthermore, NKX2.5 mutations are the most common cause of ASD in individuals with defects of conduction system.34 35 However, some individuals with CHD caused by the mutations of NKX2.5 can manifest an individual phenotype of ASD and/or conduction defects.36 To date, approximately 80 different mutations have been identified in NKX2.5, including missense mutations [eg, c.44A>T (p.K15I), c.232A>G (p.N19S), c.673C>A (p.N188K) and c.1089A>G (p.S305G)], synonymous mutations [eg, c.543G>A (p.Q181Q), c.677A>G (p.E167E), c.902C>G (p.G242G) and c.1142A>G (p.R322R)] and nonsense mutations [eg, c.1149T>C (stop→Gln)].33 The location of some of these mutations is depicted in figure 1A. Holt-Oram syndrome, which is characterized by CHD (eg, ASD, VSD and atrioventricular conduction system disease) and upper limb malformations, could be caused by the loss-of-function mutations in TBX5, which is notably expressed in the upper limbs and heart.37 38 Nonsense or frameshift mutations of TBX5 may be responsible for this syndrome.12 Functional deficiency of the conserved DNA-binding motif in the transcription factor encoded by TBX5 may be the etiology of Holt-Oram syndrome.39 GATA4, encoding one of the GATA zinc-finger transcription factors, is a deeply studied gene and is essential for cardiogenesis. Mutations in GATA4 are reported to be implicated in cardiac septal defects.31 Over 100 mutations in GATA4 coding region have been identified in patients with CHD, and more mutations will be identified with intensive research.40 Among these mutations, 11 sites (two synonymous mutations, seven missense mutations and two frameshift mutations) have been studied in familial cases, which highlight the significance of these sites in the development of CHD (figure 1B). Multiple mutations identified from other transcription factors, including NKX2.6, 41 GATA5, 42–44 GATA6, 45 TFAP2β, 46 TBX1, 47 TBX20 48 49 and ZIC3, 50 have also been reported to be associated with the incidence of CHD.
Signaling pathways involved in the occurrence of CHD are widely studied. Genes involved in these different signaling pathways can converge into a large and sophisticated regulatory network that plays an important role in cardiac development and pathogenesis of CHD. Recent studies have also suggested the potential contributions of vascular endothelial growth factor-A (VEGF-A), Notch signaling, Wnt signaling, transforming growth factor-β (TGF-β), bone morphogenic protein (BMP) signaling and cpathway to the occurrence of CHD.51–57 In these pathways, some are essential for the formation of cardiac septum, valves and the construction of cardiac outflow tracts, while others are associated with the asymmetric development of the heart. Gene mutations in renin–angiotensin system mitogen-activated protein kinase (RAS-MAPK) signal transduction pathway can lead to Noonan syndrome with the typical phenotype of pulmonary valve stenosis and hypertrophic cardiomyopathy.58–61 Six missense variants (COL6A1, COL6A2, CRELD1, FBLN2, FRZB and GATA5), acting in the VEGF-A pathway, were found to be damaged in individuals with complete atrioventricular septal defect (AVSD),51 suggesting that rare variants in the VEGF-A pathway might play a role in the development of AVSD. In addition, mutations in VEGF-A have been reported to be associated with congenital left ventricular outflow tract obstruction.62 Notch signaling is a highly conserved pathway involved in developmental process of heart. JAG1 encodes a ligand in the Notch signaling pathway, which leads to localization of Notch to the nucleus and downstream activation of target genes. Mutations in JAG1 have been found in over 90% cases of Alagille syndrome.63 64 Some cases (~2%) that have mutations in NOTCH2, a NOTCH receptor gene, are also correlated with Alagille syndrome.65 However, variants of NOTCH1 that belongs to the Notch signaling pathway have been identified to be associated with Adams-Oliver syndrome.15 Mind bomb 1 (Mib1) is a vital protein that promotes ubiquitination, endocytosis and subsequent activation of Notch ligands to activate the Notch signaling pathway. Mutations in MIB1 have been identified to be associated with cardiac deformity such as ASD, AVSD and VSD, through a lower level of JAG1 ubiquitination and Notch signaling induction.66 In addition, mutations in other genes of Notch signaling pathway include MAML1, 67 DLL4 68 and GALNT11 69 are also involved in CHD. AXIN2 is involved in the regulation network of cardiac valve formation and elongation, and its expression product is a negative regulator of Wnt/β-catenin signaling pathway.70 71 It has been found that mutations in AXIN2 can result in CHD with the phenotype of congenital valve defect.72 HHEX, a member of the Homeobox gene family, is an important cardiac determinant and controls the early differentiation, migration and development of cardiomyocytes.73 Foley et al 74 reported that mutations in HHEX could lead to a phenotype of abnormal developmental endogenous cardiac or ectopic heart, which was similar to the antagonistic effect of Dickkopf-1 to Wnt signaling pathway. Aberrant Wnt signaling pathways implicated in CHD have been summarized in a previous excellent literature.75 We describe the aberrant expression of genes associated with CHD within the Wnt signaling pathways in figure 2. TGF-β signaling pathway has important role in the development and remodeling of cardiovascular system. Aberrant TGF-β signaling pathway is involved in the pathogenesis of several human cardiovascular diseases through the epithelial-to-mesenchymal transition (EMT) of resident fibroblasts, circulating progenitors, pericytes, epithelial cells and/or the endothelial-mesenchymal transdifferentiation of endothelial cells.76 77 For example, mutations in TGF-β1, one major subtype of TGF-β family, have been reported to be associated with CHD in pediatric patients.78 In general, BMP synergizes with TGF-β signaling to activate the downstream genes, such as Smad1, Smad5 or Smad8, with the changed transcription of target genes. Mutations in the genes that encode transducers of the TGF-β and BMP signaling pathway have been identified in the pathogenesis of cardiovascular diseases, such as Marfan syndrome and Loeys-Dietz syndrome.55 79 Nodal signaling pathway is involved in the left–right patterning and development of the heart and in abnormal gene products throughout the pathway that are clearly associated with CHD. Roessler et al 56 previously demonstrated that reduced nodal signaling strength via mutation of FOXH1 was linked to human heart defects. SHH, one morphogen of hedgehog (HH) family proteins, is involved in a remarkably wide variety of process, including cardiovascular development. Previous evidence demonstrated the mutations in genes of HH signaling were implicated in the occurrence of CHD with a phenotype of ASD, VSD or AVSD, and the responsible mutated genes include SHH, Gli3 and MKS1.80–83 Noonan syndrome, one of the most common genetic syndromes of CHD, is caused by mutations in genes of the RAS-MAPK pathway.84 At present, several genes have been verified to be responsible for the development of Noonan syndrome and other disorders. Mutations in genes, encoding molecules implicated in the RAS-MAPK signaling pathway, account for approximately 90% of affected CHD cases. These genes include: PTPN11, SOS1, KRAS, RAF1, BRAF, SHOC2, NRAS, HRAS, CBL, MEK1, MEK2, PPP1CB, RIT1 and SOS2.15 58 61 84 Currently, most studies have only stayed on the basic exploration of a single gene, which cannot fully explain the pathogenesis of CHD. The interactions of various signaling pathways involved in heart development are still incompletely understood. Some responsible mutated genes have been proposed to clarify the interactions of the signaling pathways implicated in CHD (figure 2). More researches are needed to help elaborate the pathogenic mechanisms of CHD for the intricate network of signaling pathway.
Several studies use targeted whole-genome sequencing to investigate the genes that encode cardiac structural proteins to elucidate the monogenic cause of CHD. Some rare missense mutations or premature termination mutations in myosin heavy chain 6 (MYH6), a marker gene of myocardial cell, could result in ASD.85 86 However, mutations in MYH7 have been confirmed in bilobal aortic valve.87 CHD with apical hypertrophic cardiomyopathy, left ventricular non-compaction and ASD could be caused by the mutations in ACTC, which is a cardiac actin gene and is essential for cardiac contraction.88 89 MYH11, encoding smooth muscle myosin heavy chain, is another gene that has been reported to be involved in CHD with the phenotype of dominant thoracic aortic aneurysm.90 91 Moreover, loss-of-function mutations in the elastin gene (ELN) could result in CHD with the presence of Williams-Beuren syndrome and non-syndromic SVAS.17
Epigenetics, which refers to the mechanisms of changed gene expression that are independent of DNA sequence, provides a new way to understand the pathogenesis of CHD. To date, three canonical mechanisms of epigenetics include DNA methylation, histone modification and non-coding RNAs. The increasing evidence suggests that the aberrant regulation of gene expression by epigenetics is a key factor in the development of cardiovascular diseases, which have attracted attention to focus on the role of epigenetics in CHD.
DNA methylation, the most widely studied epigenetic mechanism, refers to the formation of a methyl group (−CH3) in the 5′ carbon of cytosine (CpG islands), which induces an alteration of the structure of DNA. The methylation process is catalyzed by DNA methyltransferases (DNMTs), comprising DNMT1, DNMT3A and DNMT3B. The dysregulation of DNA methylation during different stages of development could lead to the transcriptional repression and functional inactivation of tissue-specific genes, resulting in increased risk for several diseases, including cardiac malformation. Alterations of DNA methylation, especially in CpG islands, which are close to core transcription factors and genes in signaling pathway, have been reported in patients with cardiac malformations.92 DNA methylation plays a critical role in the development of the heart. The expression of hyaluronan synthase 2 (Has2) is necessary for the formation of the heart valves, septa and epicardium. However, Has expression was found to be downregulated via DNA methylation in the heart at E14.5 embryos.93 Furthermore, gene knockout model indicated that expression of Has2 is downregulated via DNA methyltransferase 3B (DNMT3B), which was coexpressed with Has2 in the region of cardiac valve, suggesting that changes in DNA methylation might be involved in the regulatory function of Has2 enhancer. Aberrant methylation of CITED2 may play an important role in the development of VSD, ASD and TOF.94 Aberrant methylations of CITED2 could decrease its mRNA expression and be regarded as the prime cause of CHD.95 The presence of aberrant hypomethylation in the CpG region of BRG1 was found to exist in patients with ASD and interventricular septal defect.96 It has been shown that the promoter region of CX43 plays an essential role in the development of the heart outflow tract, and the aberrant hypomethylation of this enhancer region of CX43 could be considered as one of the etiologies of CHD.97 98 In another study, two encoding transcription factors, zinc-finger in cerebellum 3 and nuclear receptor subfamily 2 group F member 2, were found to be hypermethylated in monozygotic twins with DORV, suggesting that the differential methylation of these transcription factors could be regarded as a potential pathogenesis of the diseased twin.99 Hypermethylation of the promoter region of SCO2, a cytochrome oxidase, has been found in patients with TOF and VSD. The methylation of CpG islands located in the promoter of SCO2 result in reduced expression of SCO2, which may be the mechanism of the occurrence of these diseases.100 Further studies discovered that hypermethylation of CpG islands of NKX2.5, HAND1 and RXRα was also essential for CHD, including VSD and TOF.101 102 Multiple differential methylated genes have been identified from cardiomyocytes of newborn and adults, and adult failing hearts have revealed a highly dynamic of DNA methylation under specific developmental or pathological conditions. DNA methylation is tightly regulated during cardiac differentiation and maturation.103 Taken together, these findings highlight the importance of DNA methylation in cardiac morphogenesis and CHD formation.
Histones, including H1, H2A, H2B, H3 and H4, and DNA constitute the nucleosome, which is the basic structural unit of chromatin in eukaryotic cells.104 Altering the histone-DNA contacts effectively via histone, post-translational modifications could loosen or tighten the chromatin architecture to control the availability of gene transcription or expression. Histone post-translational modifications could be modified by the catalysis of histone-modifying enzymes, such as histone methylases, demethylases, acetylases and deacetylases, ubiquitin enzymes and phosphorylases. Aberrant expression and mutation of the histone modifiers during the development of heart can influence the response of heart to pathological stresses.105 Moreover, increasing researches show that the interplay between these different cardiac transcription factors and histone modifiers plays a significant role in heart development. Numerous studies have shown that methylation and acetylation of histone is an emerging epigenetic mechanism for the regulation of gene transcription. Therefore, we focused only the mechanism of histone methylation and acetylation in CHD. Methylation mainly occurs at the core histones H3 and H4, which is performed by the catalysis of histone methyltransferases. H3K4, H3K36 and H3K79 methylation leads to the transcriptional activation, whereas transcriptional repression can be induced by H3K9, H3K27 and H4K20 methylation.106 CHD associated with Wolf-Hirschhorn syndrome exists deletion of Wolf -Hirschhorn candidate protein 1 (Whsc1), which encodes the H3K36me3-specific methyltransferase. Nimura et al 107 found that the pathogenic role of Whsc1 was associated with the transcriptional activation of Nkx2.5 in its target sites through the model of Whsc1-knockout murine. Another studies revealed that the interactions of JARID2 and SETDB1, which was an H3K9me3-specific methyltransferase, could elucidate the role of Jarid2 in the occurrence of VSD, DORV and impaired ventricular compaction induced by hypertrabeculation.108 109 TBX1, one member of T-box transcription factor family, was shown to interact with H3K4 and H3K27 via two domains of T-box to regulate gene expression, and aberrant expression could result in CHD including TOF, VSD and aortic arch interruption.110 111 A later study further identified the interactions of TBX1 and BAF chromatin remodeling complex regulate Wnt5a expression, and the insufficient expression of TBX1 or Wnt5a could result in the phenotypes of hypoplastic right heart.112 DPF3, an evolutionarily conserved protein, binds methylated and acetylated lysine residues of histone 3 and 4 to regulate gene expression. Dpf3 is expressed in the heart during development and has been found as significantly upregulated in the right ventricular myocardium of patients with TOF.113 Histone acetylation and deacetylation are always in a dynamic state, which is associated with transcriptional activation or transcriptional repression, gene silencing and cell cycle. Histone acetylation and deacetylation have been studied extensively in recent years.114 It has been shown that some histone acetylases, such as EP300, KAT2A and hNAT1, are associated with the development of the heart. Aberrant expression of these acetylases could lead to CHD with ASD, VSD, AVSD and valve dysplasia.115 116 Previous studies have revealed that histone deacetylases (HDACs) are also implicated in CHD. For example, HDAC3 regulates the cellular acetylation level affecting cardiogenesis, and the absence of HDAC3 could cause PDA and TOF.117 In addition, down-regulation of HDAC5 and HDAC9 simultaneously could result in CHD with VSD.118 119 Abnormal regulation of SIRT1 has been reported to be involved in ventricular hypoplasia.120 Park et al 121 revealed that histone methyltransferase SMYD1 was associated with ventricular hypoplasia. Some other histone modifiers have been shown to be important for the development of the heart through numerous studies, such as G9a, Ezh2/PRC2, Baf60c/Brg1, Jmjd3, UTX and MLL2,122–130 and the aberrant modification of them could yield critical CHD phenotypes. The activity of cardiac transcription factor could be changed by histone modifiers, which means these modifiers could be strong candidates for CHD etiology and therapeutic targeting.
Non-coding RNA is another type of epigenetic modification involved in the control of gene expression by post-transcriptional regulation. Among the various non-coding RNAs, microRNAs (miRNAs) and long non-coding RNAs (LncRNAs) are the two best studied groups. miRNAs are capable of regulating gene expression by interacting with mRNA transcript 3′ untranslated regions (UTRs) to repress translation, whereas LncRNAs are able to regulate transcription by directly interacting with chromatin remodeling complexes.131 Emerging evidence that indicates the impact of non-coding RNAs to cardiac development was previously unappreciated but is becoming valuable. A number of miRNAs have recently been shown to function in the heart.132 It has been shown miR-1 is important in cardiac development and is regarded as the most relevant miRNA leading to CHD.133 134 Li et al found that expression of miRNA-1 decreased in patient with VSD.135 Molecularly, miRNA-1 binds to its target gene GJA1 and SOX9, regulating the formation of cardiac valves and septa, and therefore miRNA-1 dysregulation could result in CHD in human.135 Further studies have demonstrated that myocyte enhancer factor 2 (Mef2) could upregulate the expression of miR-1, which suppress the cardiac transcription factor Hand2 and HDAC translation.133 136 Hand2 is known as a target of miRNA-1 and is involved in the growth of the embryonic heart; thus, Hand2 could be associated with CHD.133 Overexpression of miRNA-27b could repress the expression of Mef2c and affect the development of myocardial, leading to cardiac hypertrophy.137 A recent study has confirmed that differentially expressed frataxin (FXN) regulates the development of CHD and that differential expression of FXN is under the control of miRNA-145.138 Further investigation of this study demonstrated that overexpression of miRNA-145 could regulate apoptosis and mitochondrial function by repressing the expression of FXN, leading to the development of CHD.138 In recent years, increasing evidence has confirmed that alterations of miRNAs expression are associated with human cardiovascular diseases, including CHD (reviewed by Nagy).139 Furthermore, miRNAs are extensively studied as clinical biomarkers for their stability in blood, urine and other biological fluids and their ability to evade RNA degrading enzymes. The researchers demonstrated that miRNAs in maternal serum could be used as candidate biomarkers for prenatal detection of fetal CHD in early pregnancy.140 141 These authors identified miR-19b and miR-29c significantly upregulated in patients with VSD, while miR-19, miR-22, miR-29c and miR-375 upregulated in patients with TOF. In recent years, increasing new miRNAs have been found to be associated with CHD, suggesting a potential value of miRNAs as diagnostic markers in human cardiovascular diseases. LncRNA regulates cell growth, differentiation, cell proliferation and apoptosis by controlling gene expression. To date, little information of the functional role of lncRNAs in CHD has been reported, and only scant evidence has demonstrated the involvement of lncRNAs in CHD, particularly on VSD and TOF. Jiang et al 142 reported the increased expression of SNHG6 in fetal cardiac tissues of VSD patients and suggested SNHG6 might be involved in VSD by the mechanistic link between SNHG6 upregulation, miR-100 downregulation, Wnt/β-catenin activation and the formation of VSD. The author also identified that expression of HOTAIR was increased in right atrial biopsies of CHD patients with ASD and VSD and postulated HOTAIR as a biomarker for CHD. Another study identified a polymorphism of MALAT1 associated with ASD and VSD.143 LncRNA TUC40 has been reported to reduce the expression of PBX1 and to affect the differentiation of cardiomyocytes, which may be a potential pathological etiology of VSD.144 In addition to the importance of LncRNAs in cardiac septal defects, the role of LncRNAs in the development of cyanotic heart disease such as TOF has been reported. Wang et al 145 indentified that high expression of HA117 was associated with adverse outcomes in TOF patients, although the mechanism of HA117 in TOF remained unclear. In a recent study by Gu et al,146 the circulating plasma LncRNAs have been implicated in cardiovascular diseases for its potential as new biomarkers of diagnostic and prognostic in clinical treatments. The aberrant expression of the following LncRNAs: ENST00000436681, ENST00000422826, AA584040, AA709223 and BX478947 are associated with CHD. These specific LncRNAs identified from the plasma of pregnant women with typical fetal CHD may play an important role in the development and prenatal diagnosis of fetal CHD.
Normal development of the heart is a complex process involving many regulatory factors, including genetics and epigenetics. Cardiac malformations are congenital developmental disorders that can be induced by the dysregulations of genetic and epigenetic factors. Abundant seminal studies have found that conventional genetic factors could not elucidate the pathogenesis of CHD alone. Epigenetic regulation is also an important aspect of normal cardiac development as well as of defective function in disease situations such as CHD. The interactions at multiple levels provide insights of the combinatorial regulation into the morphogenesis of heart and suggest they can partially compensate each other’s function. Phenotypic heterogeneity and incomplete penetrance of CHD complicate our understanding of the interactions between genetics and epigenetics in CHD. Future studies to focus on elucidating the epigenetic signals of genes associated with cardiac development pathways could throw light on the genetic and epigenetic mechanisms in the development of CHD. With the great progress in basic researches carried out in model systems (eg, in vivo: animal models; in vitro: cell and tissue engineered models) and with the improving limits of detection, the study of CHD has entered a new era for clearer understanding of its etiology. The wide application of these new detection technologies would provide an effective method for the prevention, diagnosis and treatment of CHD, opening up new avenues of individualized care for patients with CHD. 10.1136/wjps-2020-000196.supp1 |
|
PMC9648611 | Jianyu Hu,Xue Zhang,Huan Tao,Yongqian Jia | The prognostic value of Epstein−Barr virus infection in Hodgkin lymphoma: A systematic review and meta-analysis | 27-10-2022 | Epstein – Barr virus,Hodking's lymphoma,Meta - analysis,prognosis,Virus infection | Introduction Epstein−Barr virus (EBV) contributes significantly to the development and occurrence of B-cell lymphomas. However, the association between EBV infection status and clinical outcomes in Hodgkin lymphoma (HL) patients has long been controversial. Therefore, we aimed to estimate the prognostic significance of EBV infection in HL survival. Methods We searched PubMed, Embase, Web of Science, and the Cochrane Library for relevant cohort studies from the date of their inception to February 20, 2022. Hazard ratios (HRs) and 95% confidence intervals (CIs) for overall survival (OS), Failure-free survival (FFS), Progression-free survival (PFS), Event-free survival (EFS) and disease-specific survival (DSS) were extracted from the studies or calculated. Subgroup analyses were conducted independently on the five survival outcomes to investigate the source of heterogeneity. Results A total of 42 qualified studies involving 9570 patients were identified in our meta-analysis. There was an association between EBV positivity and significantly poorer OS (HR=1.443, 95% CI: 1.250-1.666) and DSS (HR=2.312, 95% CI: 1.799-2.972). However, the presence of EBV in HL showed no effect on FFS, PFS or EFS. In subgroup analyses of OS, DSS and FFS stratified by age groups, EBV positivity was associated with poorer prognosis in elderly patients. Meanwhile, in children and adolescents with EBV-positive HL, we also observed a trend toward a better prognosis, though the results were not statistically significant. Conclusions EBV-positive status is associated with poor OS and DSS in HL patients. EBV infection should therefore be considered a valuable prognostic marker and risk-stratifying factor in HL, especially in older patients. Systematic Review Registration https://www.crd.york.ac.uk/PROSPERO/ , identifier CRD42022328708. | The prognostic value of Epstein−Barr virus infection in Hodgkin lymphoma: A systematic review and meta-analysis
Epstein−Barr virus (EBV) contributes significantly to the development and occurrence of B-cell lymphomas. However, the association between EBV infection status and clinical outcomes in Hodgkin lymphoma (HL) patients has long been controversial. Therefore, we aimed to estimate the prognostic significance of EBV infection in HL survival.
We searched PubMed, Embase, Web of Science, and the Cochrane Library for relevant cohort studies from the date of their inception to February 20, 2022. Hazard ratios (HRs) and 95% confidence intervals (CIs) for overall survival (OS), Failure-free survival (FFS), Progression-free survival (PFS), Event-free survival (EFS) and disease-specific survival (DSS) were extracted from the studies or calculated. Subgroup analyses were conducted independently on the five survival outcomes to investigate the source of heterogeneity.
A total of 42 qualified studies involving 9570 patients were identified in our meta-analysis. There was an association between EBV positivity and significantly poorer OS (HR=1.443, 95% CI: 1.250-1.666) and DSS (HR=2.312, 95% CI: 1.799-2.972). However, the presence of EBV in HL showed no effect on FFS, PFS or EFS. In subgroup analyses of OS, DSS and FFS stratified by age groups, EBV positivity was associated with poorer prognosis in elderly patients. Meanwhile, in children and adolescents with EBV-positive HL, we also observed a trend toward a better prognosis, though the results were not statistically significant.
EBV-positive status is associated with poor OS and DSS in HL patients. EBV infection should therefore be considered a valuable prognostic marker and risk-stratifying factor in HL, especially in older patients.
https://www.crd.york.ac.uk/PROSPERO/, identifier CRD42022328708.
Hodgkin lymphoma (HL) is a malignant neoplasm derived from B lymphocytes, accounting for approximately 10% of all human lymphomas (1). HL is one of the most frequent neoplasms in young individuals aged 20 to 40 years, accounting for nearly one-third of all new diagnoses (2). After the advent of combination chemotherapy, HL is now a highly curable malignancy. Optimal treatment selected according to standard staging has led to a cure rate exceeding 90% for limited stage disease and 80% for advanced disease as the norm (3). Nevertheless, the implication behind this rather impressive success rate is inevitability over- and undertreatment of at least 10-20% of patients in all stages of the disease. The challenge under such circumstances is to maximize cure rates while minimizing long-term toxicity, such as the induction of a second malignancy, dysplasia, or cardiac dysfunction (4, 5). Therefore, the identification of factors indicating different survival outcomes is critical in guiding risk-adapted therapy for HL. Currently, commonly used prognostic systems for HL are based mainly on clinical parameters such as Ann Arbor staging and tumor size (6). Clearly, it is necessary to improve the traditional prognostic factors in combination with immunological, biological, and functional imaging data (7). Research on immunohistochemical markers for HL prognosis is currently ongoing, with studies in which the expression of the anti-apoptotic protein B-cell lymphoma-2 (Bcl2), the tumor suppressor protein p53 and topoisomerase IIα are associated with poorer prognosis (8, 9). It is accepted that EBV has transforming potential and that latent infections contribute to the pathogenesis of HL (10). EBV-positive HL is defined as the presence of EBV in tumor cells, not in bystander reactive lymphocytes (11). Currently, EBV-encoded mRNA (EBER) in situ hybridization (ISH) is considered to be the “gold standard” for EBV status. Meanwhile, some studies have shown that immunohistochemistry with LMP-1 antibodies can also reliably indicate EBV infection in HL (12, 13). To date, a large number of studies have reported the correlation between Epstein−Barr virus (EBV) infection and the prognosis of HL, but the results of the studies have been inconsistent (positive, negative or no association) (11, 14–54). The differences in the results of the studies may be explained by different population distributions, patient selection, statistical analysis techniques and outcome measures. Therefore, we conducted a meta-analysis of all eligible published studies to quantify the prognostic value of EBV infection in HL patients.
We followed the PRISMA Statement guidelines to conduct and report this systematic review and meta-analysis (55). The study was registered in PROSPERO (Record Number CRD42022328708).
We systematically searched PubMed, Embase, Web of Science and the Cochrane library for articles published from the date of their inception to February 20, 2022. We identified studies by using the following terms: (“Epstein−Barr Virus Infections” or “EBV Infections” or “Epstein−Barr Virus” or “Human Herpes Virus 4 Infections” or “HHV 4”) and (“Hodgkin Disease” or “Hodgkin Lymphoma” or “Hodgkin’s Disease” or “Hodgkin’s Granuloma”) and (“prognosis” or “prognostic factor” or “survival”). The reference lists of the identified articles were also searched manually to ensure that no studies were overlooked.
Two investigators (J. Y. Hu, X. Zhang) independently screened each study based on titles and abstracts. When the studies met our inclusion criteria, the full text of the articles was retrieved. We resolved disagreements through discussions or negotiations with a third investigator (H. Tao). Studies that met the following criteria were included: (1) discussed the prognosis of EBV infection in HL whose infection status was detected by EBER in situ hybridization and/or LMP-1 immunohistochemistry; (2) outcomes were survival-related; (3) sufficient survival data were provided; (4) articles were published in English; and (5) cohort design. If the same author or institution published multiple articles, we selected the most informative article. Studies were excluded if (1) they were reviews, letters, case reports, conference abstracts, or unpublished articles; (2) study subjects were animals; or (3) the study population was human immunodeficiency virus-associated lymphoma.
The data we extracted from selected articles included the following: (1) baseline characteristics (first author, publication year, country, number of patients, median/mean age, histology, etc.); (2) EBV detection method and EBV status; (3) survival outcomes (including overall survival [OS], failure-free survival [FFS], progression-free survival [PFS], event-free survival [EFS], disease-specific survival [DSS]), definitions of the five survival endpoints are summarized in Table S1 ; and (4) statistical evaluations, including Cox regression analysis hazard ratios (HRs), 95% confidence intervals (CIs), and P values. When HR and 95% CIs were absent from the original article, we used the software designed by Tierney et al. (56) to indirectly estimate from Kaplan−Meier curve.
The quality of each study was assessed independently by two investigators (J. Y. Hu, X. Zhang) using the Newcastle−Ottawa Scale (NOS) (57). This scale is an eight-item instrument used to assess the selection of participants, study comparability, and ascertainment of the outcome. The NOS scores ranged from 0 to 9, and high-quality studies were defined if the score was more than 6.
We used the HRs and corresponding 95% CIs to investigate the associations between EBV infection and HL survival outcomes (OS, FFS, PFS, EFS and DSS). For a more accurate estimation of the effect of EBV infection, we selected the results of the multivariate model when both multivariate and univariate Cox regression analyses were reported in the same article. Heterogeneity was assessed by the Cochran’s Q test and I2 index, which describes the percentage of total variation across studies that is due to heterogeneity rather than chance (58). Statistically significant heterogeneity was defined as I2 statistic>50% and/or P value < 0.10 of Cochran’s Q test. When I2>50% and/or P<0.10, the random-effects model was used to estimate pooled HRs (59); otherwise, a fixed-effects model was used (60). To explore a potential source of heterogeneity, subgroup analyses were conducted based on variables including continent, histologic subtype, age, detection method, and whether a multivariate or univariate Cox regression was used. Sensitivity analyses were performed to assess the stability of pooled HR by sequentially excluding each study. Publication bias was evaluated by visual inspection of the symmetry of the funnel plot and assessment with Begg’s and Egger’s tests (P<0.05 was deemed strong publication bias) (61). All statistical analyses were performed using Stata Version 15.1. (Stata, College Station, TX, USA), and P<0.05 was considered statistically significant.
Figure 1 illustrates a flowchart describing the study inclusion process. We initially identified 4538 articles. After the removal of duplicates and screening of titles and abstracts, the full text of the 176 potentially qualified articles was reviewed. Finally, after excluding those with a duplicated study population (n=6), nonsurvival analysis data (n=78) and unable to obtain HR (n=50), 42 articles (11, 14–54) studying 9570 patients were included in our meta-analysis.
Table 1 shows a summary of the characteristics of the 42 included studies, most of which were retrospective cohort studies. The studies were conducted in Asia (37.7%), Europe (33.3%), North America (6.6%), South America (8.8%), Australia (2.2%) and Africa (6.6%) and published between 1997 and 2022. The sample size per study ranged from 47 to 922. The reported mean or median age for studies differed widely; five studies only included patients younger than 18 years old (16, 30, 33, 41, 48), and one study only included the elderly (20). Thirty studies (11, 14, 15, 17, 18, 20–25, 28–30, 32–36, 38, 41, 44–46, 48–51, 53, 54) reported the median or mean follow-up time, ranging from 25 to 130 months. In terms of methodological quality, all included studies scored more than six stars on the NOS. Details of the risk of bias assessment are shown in Additional file: Table S2 .
Thirty-three studies (11, 14, 16–19, 22–26, 28, 29, 31, 34, 35, 38–54) (corresponding to 36 sets) were included to analyze the impact of EBV infection on OS. Our meta-analysis showed that EBV positivity in HL was correlated with unfavorable outcomes for OS (HR=1.443, 95% CI: 1.250-1.666, P<0.001; Figure 2 ). Moderate heterogeneity was found across the studies (I2 = 43.7%, P=0.003) by employing a fixed effects model. Therefore, to explain the heterogeneity, we conducted subgroup analyses according to continents, histology, age groups, detection method, data source and data extraction ( Table 2 ). In the subgroup analysis by disease distribution on six continents, the African subgroup showed that EBV-positive patients had a borderline better OS (HR=0.408, 95% CI: 0.147-1.129; P =0.084). For age distribution, some articles (16, 18, 25, 28, 41, 48, 50, 52) had sufficient age-stratified survival data, so we combined their HR by a random effects model (I2 = 62.9%, P=0.003), which was 1.080 (95% CI: 0.657-1.776; P=0.762). In the subgroup of children and adolescents, the pooled HR showed that EBV positivity in HL was correlated with a favorable outcome for OS (HR=0.296, 95% CI: 0.085-1.034, P=0.056), while a significantly poorer OS was associated with EBV positivity in studies covering older adults (HR=1.905, 95% CI: 1.380–2.629; P<0.001). This may partly explain the heterogeneity observed when examining EBV infection as a prognostic factor in HL patients.
Nine studies (11, 21–24, 32, 37, 41, 53) were included to analyze the impact of EBV infection on FFS. The pooled estimate showed no significant association between EBV positivity and FFS (HR=1.030, 95% CI: 0.832–1.274, P=0.788; Figure 3 ). No significant heterogeneity was found across the studies (I2 = 0, P=0.556). In the subgroup analysis, we found that EBV positivity was strongly associated with poorer FFS (HR=3.726, 95% CI: 1.649–8.419, P=0.002) in older adults. In addition, the prognostic results of the three subgroups grouped according to detection method, data source and data extraction were similar, and all had no effect on FFS ( Table 2 ).
Five studies (39, 47, 50–52) were included to analyze the impact of EBV infection on PFS. There was significant between-study heterogeneity (I2 = 54.5%, P=0.066), and the pooled estimate by the random-effects model showed that no significant association was found between EBV positivity and PFS (HR=1.302, 95% CI: 0.881–1.926, P=0.186; Figure 4 ). Due to the lower number of analyzable studies, subgroup analysis was not performed for PFS.
Ten studies (29–31, 33–35, 38, 42, 44, 48) were included to analyze the impact of EBV infection on EFS. The pooled estimate showed that no significant association was found between EBV positivity and EFS (HR=0.962, 95% CI: 0.755-1.227, P=0.756; Figure 5 ). No significant heterogeneity was found across the studies (I2 = 0, P=0.543). The prognostic effects were similar between the four predefined subgroups according to continent, detection method, data source and extraction ( Table 2 ).
Six studies (15, 20, 25, 26, 28, 36) were included to analyze the impact of EBV infection on DSS. The pooled HR of 2.312 (95% CI: 1.799-2.972) was calculated on the basis of a fixed-effects model ( Figure 6 ), which showed a worse DSS among EBV-positive patients than EBV-negative patients. In subgroup analysis, a fixed-effects model was used for the age subgroup meta-analysis due to the heterogeneity among studies (I2 = 1.2, P=0.415). Interestingly, EBV-positive older adults had poorer DSS (HR=2.308, 95% CI: 1.607-3.312; P<0.001) than EBV-negative adults, whereas studies involving children, adolescents and young adults yielded no association between EBV infection and DSS ( Table 2 ).
We conducted a sensitivity analysis of the association between EBV infection and survival outcomes and demonstrated that the results were robust after omitting any of the included studies ( Figure S1 ). Funnel plots with Begg’s test and Egger’s test were used to assess publication bias, and no evidence of bias was found in our meta-analysis of the selected studies. The p values were all >0.05, and details of the survival outcome publication bias can be seen in Figure S2 .
The prognostic significance of EBV infection in HL patients remains controversial. Here, we conducted a meta-analysis involving 9570 patients from 42 studies to systematically explore the prognostic value of EBV infection in HL. Our results demonstrated that EBV positivity predicted short DSS and OS, but it had no significant effect on FFS, PFS or EFS. Moreover, subgroup analysis showed that in children and adolescent HL patients, EBV positivity allowed some survival advantage compared with the outcomes of EBV-negative patients, although the difference was not statistically significant. In contrast, EBV-positive elderly patients with HL have strongly poorer survival outcomes than EBV-negative patients. To our knowledge, this study is an update of two meta-analyses published in 2014 (62) and 2015 (63) on EBV infection and HL OS. Chen, Y. P. et al. (63) found no significant association between EBV infection and overall survival, but their age-specific subgroup analyses showed that OS was significantly shorter when patients’ median/mean age was ≥40 years. In addition, they found that EBV positivity had a tendency for worse OS in patients in Europe and North America. Similar to the findings of Chen and colleagues, a meta-analysis by Lee, J. H. et al. (62) also failed to reveal an association between EBV infection status and cHL patient survival. The reason for the different results between the two meta-analyses above and ours may be that we included more studies published in recent years; moreover, the type of disease was not limited to cHL, and the detection of EBV infection was not restricted to LMP1. For the survival endpoints of OS and DSS, our results are in line with 7 studies (20, 28, 36, 39, 46, 47, 49), and other reports describing clinical outcomes in relation to EBV status are conflicting. Many studies have not demonstrated that EBV status has an impact on prognosis (35, 38, 40–42, 44, 45, 51–54), whereas some studies have shown that EBV-positive status is associated with a favorable clinical outcome (14, 16, 17, 23, 48, 50). The discrepancies observed in these studies were generally due to the heterogeneous nature of the disease and the selection bias of study subjects, age groups, EBV detection, treatment regimens and the different outcome measures used. Additionally, since the distribution of EBV varies widely in the population, it may reflect racial or ethnic differences. In fact, our subgroup analyses showed that the HR of OS and DSS was not influenced by whether nodular lymphocytic predominant Hodgkin’s lymphoma (NLPHL) patients were excluded, whether the data were extracted from the KM curve and different detection methods. Only the pooled HR from Africa had a tendency to improve OS; interestingly, two of the three included studies enrolled populations younger than 50 years old (16, 50). This was in good agreement with the findings obtained from our subgroup analysis of different age groups. The effect of EBV status on OS and DSS is age dependent, and older adult patients with EBV-positive HL had a particularly poor prognosis, which was consistent with the findings of some population-based studies (18, 25, 28, 38, 50). Our study showed better survival trends for children and adolescents, although this trend was not statistically significant. The abovementioned differential effects on outcomes with respect to age and geography may be attributed to the following reason. EBV infection rates in patients with HL were significantly higher in African and South American countries than in other regions, according to an epidemiological survey (62); therefore, children had a relatively high risk of early exposure to a wide range of infectious agents. As LMP1 has antigenicity, LMP1 could activate cytotoxic T lymphocytes (CTLs) more effectively, resulting in a stronger antitumor immune response (64), which may in turn limit disease progression. However, cytotoxic T-cell responses have been observed to decline with age (65, 66), and another possibility is immunosenescence, in which the impaired immune system is unable to respond effectively to viral infection, allowing EBV reactivation and oncogenic transformation (67). In summary, the younger group has a beneficial EBV-specific immune response to the tumor cell population, whereas in older patients, this response may be less effective or other negative prognostic factors may outweigh any beneficial effect EBV may have. For example, elderly patients have poor treatment tolerance, with a subset unable to tolerate enough chemotherapy or combined radiotherapy and chemotherapy. Furthermore, elderly patients may have had complications that harmed their chances of survival (68). Herling et al. (22) considered that selection of the study endpoint may be an important factor affecting EBV status and prognosis, and compared to OS and DSS, FFS is a better survival endpoint. OS and DSS are both affected by salvage management after relapse, which was not even mentioned in most studies. Meanwhile, the frequency of disease-unrelated deaths is relatively high in the elderly population, and the natural limitation of life expectancy, these deaths may obscure disease effects in older adult patients. Our meta-analysis concluded that EBV infection status did not affect FFS in the entire population, which is consistent with the results of many previous studies (11, 21, 22, 24, 32, 37, 41, 53). However, Wang, C. et al. (53) and Diepstra, A (32). illustrated that the prognosis was significantly worse for EBV-positive than EBV-negative patients when patients were older than 50 years. Because our age subgroup analysis included the same two papers as well, the results were similar. Only one article (23) reported that EBV-positive patients have a longer FFS than EBV-negative patients; this is the sole study from Australia, which contradicts my results and is probably related to geographical differences. As with FFS, EFS and PFS were also unaffected by EBV infection status. The ending endpoint was PFS in a small number of articles (23, 39, 47, 50–52), and we did not perform further subgroup analyses. A child-based study from India (48) showed that EBV-positive children have longer EFS than EBV-negative children, which contradicted our finding and may be explained by the high prevalence of EBV infection in Indian children and the greater chemotherapy and radiotherapy sensitivity of infected tumor cells (17). At present, the mechanism by which EBV acts on HL is still unclear (69). In EBV-positive HL, viral infection of malignant tumor cells is characterized by the consistent expression of three EBV-associated viral proteins (EBNA1, LMP1, and LMP2A) and two noncoding RNAs (EBERs and BARTs) (67), which are believed to play important roles in tumorigenesis, including the regulation of proliferation, metastasis, immune escape, and apoptosis (70, 71). EBNA1 enhanced the activity of the AP‐1 transcription factor, triggering the induction of VEGF and IL‐8 (72); meanwhile, this protein can inhibit the antigenic peptide bound to major histocompatibility complex 1 (MHC-1) to evade recognition by CTL (73). LMP1 stimulates the proliferation of B cells by activating nuclear factor-kappa B (NFκB) and the transcription factor AP-1 (74). Moreover, LMP1 can also immortalize resting B lymphocytes and turn them into latently infected lymphoblastoid cell lines (75, 76). Collectively, these mechanisms may explain why EBV positivity is associated with poor clinical outcomes in HL patients. There are certain limitations that must be considered when interpreting the results of our study. First, there was some heterogeneity across these included articles, and despite the use of subgroup analysis, it was not feasible to explore all of the variability. Treatment regimens are not clearly indicated in some studies, and many articles do not conduct age-stratified analysis. These limitations prevented us from fully tracing the origin of heterogeneity. Second, the quality of published data for our study was relatively low, and most of the included studies were retrospective in design. Third, the age cutoff between children, adults and elderly varied according to the published studies; thus, to include as many studies as possible, 15-18 years old and 45-50 years old were used as a vague distinction dividing patients into children and adolescents, young adults and older adults. To obtain more meaningful results, more research involving the unified age cutoff is needed. Finally, because this study was limited to studies published in English, publication bias cannot be ruled out. The prevalence of EBV is higher in developing countries, but our study embraces only a small number of studies in Africa and South America. In addition, although some studies have shown that EBV-DNA can be used as a prognostic marker for EBV-associated HL, the choice of compartments of peripheral blood and cut-off copies of EBV-DNA is different in various studies (37, 51, 77, 78). Given the different criteria in the related original studies, we did not include the studies of using PCR method to detect EBV infection.
Our findings suggest that EBV-positive status is associated with poor OS and DSS in HL patients. EBV infection should therefore be considered a valuable prognostic marker and risk-stratifying factor in HL, especially in older patients. More studies in the future should include a larger number of children and young adults to investigate the combined effects of age and EBV status with other prognostic factors to improve the therapeutic applicability of these findings.
The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.
JH and YJ formulated the research questions and designed the study; JH, XZ, and HT conducted the literature search, selected the articles, and extracted the data; JH and HT analyzed the data and drafted the manuscript; and YJ critically revised the article. All authors contributed to the article and approved the submitted version.
This study was supported by the 1·3·5 project for disciplines of excellence–Clinical Research Incubation Project, West China Hospital, Sichuan University (grant number 2020HXFX020).
Thanks to the authors of all the included articles that were used as data sources for this article.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648667 | Shuhui Liu,Ruiying Yin,Ziwei Yang,Feili Wei,Jianhua Hu | The effects of rhein on D-GalN/LPS-induced acute liver injury in mice: Results from gut microbiome-metabolomics and host transcriptome analysis | 27-10-2022 | rhein,acute liver injury,gut microbiota,metabolomics,transcriptomics | Background Rhubarb is an important traditional Chinese medicine, and rhein is one of its most important active ingredients. Studies have found that rhein can improve ulcerative colitis by regulating gut microbes, but there are few reports on its effects on liver diseases. Therefore, this study aims to investigate these effects and underlying mechanisms. Methods Mice were given rhein (100 mg/kg), with both a normal control group and a model group receiving the same amount of normal saline for one week. Acute liver injury was induced in mice by intraperitoneal injection of D-GalN (800 mg/kg)/LPS (10 ug/kg). Samples (blood, liver, and stool) were then collected and assessed for histological lesions and used for 16S rRNA gene sequencing, high-performance liquid chromatography-mass spectrometry (LC-MS) and RNA-seq analysis. Results The levels of ALT and AST in the Model group were abnormal higher compared to the normal control group, and the levels of ALT and AST were significantly relieved in the rhein group. Hepatic HE staining showed that the degree of liver injury in the rhein group was lighter than that in the model group, and microbiological results showed that norank_o:Clostridia_UCG-014, Lachnoclostridium, and Roseburia were more abundant in the model group compared to the normal control group. Notably, the rhein treatment group showed reshaped disturbance of intestinal microbial community by D-GalN/LPS and these mice also had higher levels of Verrucomicrobia, Akkermansiaceae and Bacteroidetes. Additionally, There were multiple metabolites that were significantly different between the normal control group and the model group, such as L-α-amino acid, ofloxacin-N-oxide, 1-hydroxy-1,3-diphenylpropan-2-one,and L-4-hydroxyglutamate semialdehyde, but that returned to normal levels after rhein treatment. The gene expression level in the model group also changed significantly, various genes such as Cxcl2, S100a9, Tnf, Ereg, and IL-10 were up-regulated, while Mfsd2a and Bhlhe41 were down-regulated, which were recovered after rhein treatment. Conclusion Overall, our results show that rhein alleviated D-GalN/LPS-induced acute liver injury in mice. It may help modulate gut microbiota in mice, thereby changing metabolism in the intestine. Meanwhile, rhein also may help regulate genes expression level to alleviate D-GalN/LPS-induced acute liver injury. | The effects of rhein on D-GalN/LPS-induced acute liver injury in mice: Results from gut microbiome-metabolomics and host transcriptome analysis
Rhubarb is an important traditional Chinese medicine, and rhein is one of its most important active ingredients. Studies have found that rhein can improve ulcerative colitis by regulating gut microbes, but there are few reports on its effects on liver diseases. Therefore, this study aims to investigate these effects and underlying mechanisms.
Mice were given rhein (100 mg/kg), with both a normal control group and a model group receiving the same amount of normal saline for one week. Acute liver injury was induced in mice by intraperitoneal injection of D-GalN (800 mg/kg)/LPS (10 ug/kg). Samples (blood, liver, and stool) were then collected and assessed for histological lesions and used for 16S rRNA gene sequencing, high-performance liquid chromatography-mass spectrometry (LC-MS) and RNA-seq analysis.
The levels of ALT and AST in the Model group were abnormal higher compared to the normal control group, and the levels of ALT and AST were significantly relieved in the rhein group. Hepatic HE staining showed that the degree of liver injury in the rhein group was lighter than that in the model group, and microbiological results showed that norank_o:Clostridia_UCG-014, Lachnoclostridium, and Roseburia were more abundant in the model group compared to the normal control group. Notably, the rhein treatment group showed reshaped disturbance of intestinal microbial community by D-GalN/LPS and these mice also had higher levels of Verrucomicrobia, Akkermansiaceae and Bacteroidetes. Additionally, There were multiple metabolites that were significantly different between the normal control group and the model group, such as L-α-amino acid, ofloxacin-N-oxide, 1-hydroxy-1,3-diphenylpropan-2-one,and L-4-hydroxyglutamate semialdehyde, but that returned to normal levels after rhein treatment. The gene expression level in the model group also changed significantly, various genes such as Cxcl2, S100a9, Tnf, Ereg, and IL-10 were up-regulated, while Mfsd2a and Bhlhe41 were down-regulated, which were recovered after rhein treatment.
Overall, our results show that rhein alleviated D-GalN/LPS-induced acute liver injury in mice. It may help modulate gut microbiota in mice, thereby changing metabolism in the intestine. Meanwhile, rhein also may help regulate genes expression level to alleviate D-GalN/LPS-induced acute liver injury.
Rhein is an anthraquinone compound extracted from the rhubarb plant. Multiple studies have found rhein to have various biological effects such as anti-inflammatory (1), lipid-lowering (2), antioxidant (3), and anti-tumor (4) effects. In addition, many studies have also shown that rhein can improve the treatment of various diseases such as diseases of the nervous system (5), blood (6), digestive system (7, 8), metabolism (9), and kidney (10). In the study of liver injury, rhein has been found to help alleviate liver fibrosis caused by CCL4 (11), and for chronic liver injury caused by methotrexate (12) (MTX), rhein has been shown to inhibit the elevation and remission of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) through the Nrf2-HO-1 pathway. Acute liver injury is usually caused by viruses, drugs (13, 14), or autoimmune disorders and frequently leads to hepatocyte death. This type of injury is characterized by no underlying liver disease, rapid onset, and high mortality. At present, the main treatment for acute liver injury is liver transplantation. Although the survival rate of acute liver injury has been improved in recent years, its mortality rate remains high. Studies have found that acute liver injury caused by drugs (such as acetaminophen) (15) has a significantly higher survival rate without transplantation, and a large number of researchers have begun to conduct extensive investigations into nontransplantation treatment of acute liver injury. In recent years, research on the liver-gut axis has gradually deepened, and the development of gut microbiome technology has led to the result that gut microbes are closely related to the occurrence and development of acute liver injury. Although probiotic strains primarily colonize the intestine, they can also interact with the liver at distant sites via the gut-liver axis and its associated metabolism (16). Studies have found that increasing intestinal probiotics such as Bacillus cereus (17), Bifidobacterium longum R0175 (18), and Lactobacillus reuteri DSM 17938 (19), can effectively reduce D-GaIN/LPS-induced increases in plasma AST and ALT, improve the abnormality of liver tissue, and regulate intestinal dysbiosis. Furthermore, studies have also found that the gut microbiome may regulate the level of acute liver injury through certain signaling pathways, such as with Lactobacilli (20), which can activate the transcription factor Nrf2 which has a protective effect on oxidative liver injury. Furthermore, Lactobacillus (21) is able to immunoregulate the recruitment of canonical dendritic cells (cDCs) to the liver to produce IL-10 and TGF-b via TLR9 activation, preventing further liver inflammation. However, studies have also shown that the gut microbiome can modulate the MYC pathway and exacerbate liver damage (22). Moreover, the metabolites of gut microbes such as 1-phenyl-1,2-propanedione (23) have been found to be present in increased proportions in the metabolites of gut microbiota in liver-injured mice, but Galactose, Myo-inositol and Oleic Acid (24, 25) metabolites have been found to be significantly lower in the gut microbiota of liver-injured mice. In addition to the above, rhein has been reported to be able to modulate gut microbiota diversity and community composition, reduce obesity, and improve glucose tolerance in high-fat diet-fed rats (2). Not only that, but rhein can also regulate gut microbiota metabolism and relieve ulcerative colitis (26). We,therefore,hypothesize that gut microbiota is the “targets” for rhein in the treatment of acute liver injury.
Fifteen specific pathogen-free conventional male C57BL/6 mice weighing 18-20 g, were used for this study. After 1 week of adaptive feeding, we randomly divided the mice into three groups: the normal control group (NC group), the D-GaLN/LPS model group (Sigma-Aldrich; St. Louis, USA; cat: G0500/cat: L2630), and the rhein (Abmole Bioscience Inc: Catalog No: M3998) gavage treatment group, with five mice in each group. After 1 week of adaptive feeding, for the 8th to 15th days of the experiment, we gave rhein to the treatment group (100 mg/kg) by intragastric administration, and the other two groups were intragastrically administered the same dose of normal saline. On the 15th day of the experiment, the mice in the model group and rhein group were intraperitoneally injected with D-GalN/LPS (800mg/kg, 10ug/kg), and the normal control group was intraperitoneally injected with the same dose of physiological salt.
After injection of D-GalN/LPS, all animals were anesthetized with ether 5 hours later, blood was collected from the eyeball for AST and ALT analysis, part of the liver was fixed with 4% paraformaldehyde for histomorphological analysis, and part of the liver was snap-frozen in liquid nitrogen for transcriptome analysis. Cecal contents were snap-frozen in liquid nitrogen and immediately stored at -80°C for further 16s rRNA and metabolomic analysis. During the experiment, all tissue samples were kept frozen for as long as possible.
Plasma alanine aminotransferase (ALT) and aspartate aminotransferase (AST) were measured using an automatic biochemical analyzer (Chemray 800, rayto, China). We also excised 2 × 2 ×1 cm of liver tissue from the left lobe, fixed it with 4% formaldehyde, embedded it in paraffin, sectioned it, and stained it with HE. Pathological liver tissue damage was assessed using the HAI score (27).
The total genomic DNA of gut microbiota was extracted from feces using the E.Z.N.A.® Soil DNA Kit (Omega Bio-Tek, USA). Then, DNA concentration and integrity were measured by NanoDrop2000 and agarose gel electrophoresis before sequencing preparation. The 27F_1492R region of the 16S rRNA gene of the microbiota was amplified using barcoded primers (forward primer: 27F: AGRGTTYGATYMTGGCTCAG; reverse primer: 1492R: RGYTACCTTGTTACGACTT) and sequenced using the Illumina MiSeq platform. A thermal cycler was performed on a PCR system (GeneAmp 9700, ABI, USA), and the mixture of PCR products was subsequently purified using an AxyPrep DNA Gel Extraction Kit. After purification, amplicons were combined in equal amounts and subjected to sequencing library preparation according to the manufacturer’s manual. Eligible libraries were sequenced on Pacbio Sequel II (Illumina, USA).
To analyze the metabolites of the gut microbiota, we first added 400 μL of extraction solution (methanol:water=4:1(v:v), containing 0.02 mg/mL internal standard (L-2-chlorophenyl alanine) to 50mg fecal samples, ground them up, and extracted them by low-temperature ultrasonic extraction for 30 min (5°C, 40 KHz), let them stand at -20°C for 30 min, centrifuged them for 15 min (13000 g, 4°C), and pipetted the supernatant into a sample vial with an inner cannula for analysis on the computer. Next, we performed LC-MS analysis and chromatographic separation on an ultra-high performance liquid chromatography-tandem Fourier transform mass spectrometry (UHPLC-Q Exactive HF-X) system. The chromatographic conditions were as follows. The column was ACQUITY UPLC HSS T3 (100 mm × 2.1 mm i.d., 1.8 μm; Waters, Milford, USA), the column temperature was 40°C, and the flow rate was 0.4 mL/min. The mass spectrometry conditions were that the sample was ionized by electrospray and that the mass spectral signals were collected in positive (3500V) and negative ion (3500V) scanning modes, respectively, with a scanning range of 70-1050 (m/z), a sheath gas flow rate of 50 (arb), and an auxiliary gas flow rate of 13 (arb), a capillary temperature of 325°C, a heating temperature of 425°C, an S-Lens voltage of 50V, collision energy parameters of 20eV, 40eV, 60eV, and a resolution of 6000 (Full MS)/7500 (MS2). ProgenesisQI (Waters Corporation, Milford, USA) was used to perform baseline filtering, peak identification, integration, retention time correction, and peak alignment in order to screen for metabolic biomarkers that showed significant differences between different treatment groups. Finally, we matched the obtained precursor and fragment ions with metabolic databases: the human metabolome (database http://www.hmdb.ca/), and the Scripps database (https://metlin.scripps.edu/).
Total RNA was extracted from tissue samples using the Zymo Quick-RNA™ Miniprep Kit (zymo), and we were able to isolate AT base pairings with polyA using magnetic beads with Oligo (dT). After this,we isolated mRNA isolated from total RNA, and this enriched mRNA was randomly broken into small fragments of 300bp. Next, the mRNA was used as a template to reverse synthesize one-strand cDNA using random primers, followed by two-strand synthesis. After adding adapters, the final library was obtained by purification and amplification, and qualified library 2X150bp duplex sequencing was performed on Illumina Novaseq. After filtering the raw data, high-quality sequencing data (clean data) was obtained for subsequent analysis. The original data after quality control and cleaning (reads) were then compared to the reference genome to obtain the number of read counts of genes, and then we carried out the expression difference analysis of genes between samples to identify genes that were differentially expressed between samples.
Statistical analyses were performed using GraphPad 8.0. Data are presented as the Mean ± SEM. The differences between two groups were analyzed by Student’s t-test. Multiple group comparisons were analyzed using one-way analysis of variance (ANOVA) with Bonferroni correction. All results were considered statistically significant at P < 0.05. The differential metabolites were filtered by variable influence on projection (VIP) selection according to the PLS-DA and the filtering conditions VIP > 1 and P < 0.05. Spearman’s correlation values were computed with the R version 3.3.1. Metabolites were tentatively assigned by molecular formula matching and related information obtained from online databases such as the Human Metabolome Database (HMDB, http://www.hmdb.ca/spectra/ms/search) (28). Pathway analysis was performed on the KEGG website (29) (http://www.genome.jp/kegg/).
Compared to the normal control group, the plasma ALT and AST levels of the mice in the D-GalN/LPS group were significantly higher. Plasma ALT [NC vs Model, P < 0.01, Model vs Rhein, P < 0.01] and AST [NC vs Model, P < 0.01, Model vs Rhein, P < 0.001] levels were significantly relieved in the rhein group, however ( Figure 1A ). The histopathological results showed that there was no abnormal difference in the histological changes of the normal mouse liver; the hepatic lobules were clear, and no hepatocyte degeneration or necrosis was found ( Figures 1B, C ). In contrast, D-GalN/LPS treatment resulted in severe liver damage in the mice, but the rhein group showed significantly less liver damage and lower histological changes.
Fecal samples were analyzed by 16s rRNA high-throughput sequencing technology to elucidate the regulatory role of rhein on gut microbiota, and we used the β-diversity analysis method to evaluate the diversity differences between different groups. The structure of gut microbiota was different in different treatment groups. Specifically, in our Bray–Curtis distance-based principal coordinate analysis (PCoA), the NC group showed a separation from the model group and the rhein group, and the gut microbiota structure of the model group showed distinct deviation from the rhein group (explaining 37.42% of the variation), indicating that the core microbiota changed significantly after treatment ( Figure 2A ). Using a diversity index to evaluate the diversity (Shannon index and Simpson index) of the microbial community ( Figure 2B ), we found that compared with the normal control group, the diversity of the microbiota of the mice in the model group decreased, and the intestinal microbes of the rhein intervention group decreased significantly. The microbial community composition of mouse cecal contents showed that the microorganisms primarily consisted of Firmicutes (58.36%), Proteobacteria (24.92%), and Verrucomicrobiota (15.53%) at the phylum level ( Figure 2C ). Compared to the model group, the rhein intervention group had higher levels of Verrucomicrobiota (P<0.01) and Proteobacteria (P<0.05). Additionally, the most abundant families included Lachnospiraceae (42.52%), Muribaculaceae (22.20%), Akkermansiaceae (17.02%), Lactobacillaceae (7.62%), Ruminococcaceae (2.41%), Bacteroidaceae (1.74%), and Oscillospiraceae (1.35%). The amount of Norank_o:Clostridia_UCG-014 in the model group was higher than that in the normal control group and rhein group (P<0.05), but Akkermansiaceae (P<0.01) and Bacteroidaceae (P<0.05) in the rhein treatment group were much higher than those in the model group. The most abundant genera were norank_f:Muribaculaceae (23.96%), Akkermansia (18.88%), Lachnospiraceae_NK4A136_group (17.79%), norank_f:Lachnospiraceae (14.73%), Lactobacillus (8.45%), Bacteroides (1.94%), Lachnoclostridium (1.54%), Roseburia (1.46%), Lachnospiraceae_UCG-006 (1.09%), and Mucispirillum (1.07%). Compared to the normal control group and the rhein group, the relative abundances of Lachnoclostridium and norank_o:Clostridia_UCG-014 in the model group were higher (P<0.05). Furthermore, the abundance of Roseburia in the model group was also higher than that in the other two groups, but the difference was not statistically significant. Finally, compared to the model group, the abundance of Bacteroides in the rhein group was higher (P<0.05). Our LEfSe analysis further showed the abundance of differential taxa, and a histogram with logarithmic LDA > 4.0 and a cladogram is shown in Figure 2D . This analysis showed that acute liver injury was accompanied by higher amounts of o:Clostridia_UCG-014, f:norank_o:Clostridia_UCG014, and g:norank_f:norank_o:Clostridia_UCG-014 and that the rhein intervention group was more abundant in p:Verrucomicrobiota, c:Verrucomicrobiae, o:Verrucomicrobiales, f:Akkermansiaceae, and g:Akkermansia, :Johns:Akkeracactii.
The complex interactions between the host and gut microbiota are closely related to the host-microbe metabolic axis. Hence, we next performed untargeted metabolomics of stool samples using liquid chromatography-mass spectrometry (LC-MS). Overall there were 997 and 869 metabolites identified in feces, under the negative and positive modes, respectively. The score plots of principal component analysis PCA (38.80%) and partial least squares discriminant analysis OPLS-DA (38.40%) showed that the metabolome profiles of the NC group, Model group, and rhein group were clustered separately ( Figure 3A ). There were significant differences in metabolic profiles between the three groups. In analyzing the different metabolites among the three groups, we detected a total of 818 differential metabolites. We conducted KEGG pathway enrichment analysis for all differential metabolites ( Figure 3B ) and found that compared to the normal group, the most significant metabolic pathways in the model group were taurine and hypotaurine metabolism, sphingolipid metabolism, starch and sucrose metabolism, the citrate cycle (TCA cycle), tropane, piperidine and pyridine alkaloid biosynthesis, glycerophospholipid metabolism, alanine, aspartate and glutamate metabolism, caffeine metabolism, arginine and proline metabolism, galactose metabolism, and arginine biosynthesis. Pathways of differential metabolites in the rhein treatment group and acute liver injury group had higher amounts of alanine, aspartate and glutamate metabolism, sphingolipid metabolism, arginine and proline metabolism, arginine biosynthesis, monoterpenoid biosynthesis, glycerophospholipid metabolism, flavonoid biosynthesis, tropane, piperidine and pyridine alkaloid biosynthesis, pyruvate metabolism, histidine metabolism, glycine, serine and threonine metabolism, aminoacyl-tRNA biosynthesis, pyrimidine metabolism, carbon fixation in photosynthetic organisms, and galactose metabolism. This suggests that the rhein group may have partially reversed some of the side effects of D-GalN/LPS through these metabolisms. Next, to discover the possible biomarkers of rhein treatment, we used Student’s t-test to compare metabolite variation in acute liver injury between the three groups. We found that 224 metabolites were significantly changed between the normal control group and the model group (VIP > 1, P < 0.05, FDR < 0.05). Among them, 126 metabolites gradually returned to normal after rhein treatment, and 36 metabolites had statistical significance (P<0.05). Among these, the rhein group up-regulated 11 metabolites reduced by D-GalN/LPS and down-regulated 25 other metabolites ( Figure 3C ). Our spearman correlation analysis of microbiota and metabolites found correlations between the top 40 significantly altered differential fecal metabolites and the top 10 most abundant gut microbes ( Figure 3D ). We found that Dehydrocyanaropicrin, Homoveratric acid, Glutamylalanine, 3-(4-hydroxyphenyl)-N-(4oxobutyl)prop-2-enimidic acid, {[3-(4,5-dihydroxy-2,3-dimethoxyphenyl)prop- 2-en-1yl]oxy}sulfonic acid, L-4-Hydroxyglutamate semialdehyde, furocoumarinic acid glucoside, corchoionoside B, Ofloxacin-N-oxide, (+)-cis-5,6-Dihydro-5-hydroxy-4 -methoxy-6-(2-phenylethyl)- 2H-pyran-2-one, and 1-hydroxy-1,3-diphenylpropan-2-one were all positively correlated with Bacteroidesfen but negatively correlated with Lachnoclostridium, norank_o:Clostridia_UCG-014. dehydrocyanaropicrin, glutamylalanine, {[3-(4,5-dihydroxy-2,3-dimethoxyphenyl)prop-2-en-1yl]oxy}sulfonic acid, and ofloxacin-N-oxide were also positively correlated with Akkermansia but negatively correlated with Roseburia Related. Finally, 5-hydroxymethyl-4-methyluracil, PC(15:0/0:0), Tetranor, 12-HETE, 9(S)-HODE, Heptadecanoyl, Carnitine, Paln, litoleoyl Ethanolamide showed positive correlation with Lachnoclostridium, norank_o:Clostridia_UCG-014, and Roseburia Positive correlation but negative correlation with Bacteroidesfen and Akkermansia.
We used transcriptome analysis to determine whether the gene expression profiles of mouse livers were similar between different treatment groups. Principal component analysis showed that the genes of the mice had significant segregation (41.12%), and compared to the distance between the rhein group and the normal group, the mice in the model group and the normal group had more obvious segregation ( Figure 4A ). We then identified differentially expressed genes (FC >2 or less than 0.5, Padj < 0.05) by pairwise comparison of groups. Compared to the normal group, the number of differentially expressed genes (DEGs) in the ALF group was 5,220 (DEG1 2663/2557; up-regulated and down-regulated DEGs and DEGseq), and compared to the ALF group, the number of differentially expressed genes (DEGs) in the rhein group was 2,503 (DEG2, 1499/1004; up-regulated and down-regulated DEGs and DEGseq) ( Figure 4B ). Among these, rhein vs. ALF and NC vs. ALF intersected for 1,907 genes (DEG3), indicating that rhein may alleviate acute liver injury through these genes. Next, DGE3 underwent KEGG annotation and Gene Ontology (GO) term annotation analysis, and our KEGG annotation of significantly differentially expressed genes indicated that most genes were annotated to amino acid metabolism, carbohydrate metabolism, energy metabolism, lipid metabolism; folding, classification and degradation, transcription, translation, signal transduction, signaling molecules and interactions, cell growth and death, cell movement, cell communities, and eukaryotes. Our GO annotations primarily involved molecular functions, cellular organization, and biological processes ( Figure 4C ). In order to study the expression of rhein vs. ALF and NC vs. ALF for the intersection gene DEG3 in each group, we generated related expression heatmaps and further aggregated these into 10 cluster trend maps ( Figure 4D ). Compared to the NC group and rhein group, subcluster1 and subcluster6 in the ALF group showed an activated expression pattern, while subcluster2, subcluster4, subcluster7, subcluster8, and subcluster10 in the ALF group were inhibited compared to the NC group and rhein group. Next, we further screened the activated and inhibited gene groups in the model group (5<Log2FC<5, Padjust<0.05). In the model group, 9 genes were down-regulated ( Table 1 ), and 102 genes were up-regulated ( Table 2 ), which were recovered after rhein treatment (P<0.05). For our correlation analysis between transcriptome genes and microbiota, we selected the top 30 genes with the largest differences in the above-mentioned gene table with significant differences for correlation analysis with the top 10 microorganisms in the gut microbiota ( Figure 4E ). These results showed that Mfsd2a and Bhlhe41 genes were positively correlated with Bacteroides (P<0.001; P<0.05) and that they were also positively correlated with Akkermansia, but this correlation was not statistically significant. The other 28 genes, including Cxcl2, S100a9, Tnf, Ereg and IL-10 were activated genes in the model group and were associated with Lachnoclostridium, Roseburia, norank_f:norank_o:Clostridia_UCG-014, Lactobacillus, norank_f:Lachnospiraceae, and norank_f:Muribaculaceae, which were all positively regulated.
Our main finding is that the group pretreated with rhein had substantially better outcomes from liver damage after D-GalN/LPS injection. Additionally, we observed several other beneficial effects in the rhein group, including attenuation of microbial dysbiosis, improvement of metabolic profile, and modulation of certain gene levels. We also summarized the results of our 16S rRNA gene sequencing metabolomic and transcriptomic analysis and discussed the relationship between rhein and acute liver injury in light of the association of gut bacteria with metabolic biomarkers and liver tissue genes. Serum ALT and AST levels have been shown to be the major biomarkers for liver injury (30), and this study showed that the rhein group had significantly lower D-GalN/LPS-induced AST, ALT elevation, and liver tissue damage than the other two groups. The changes in these two functional indices, along with the improvement in HAI score, indicated that the hepatocyte injury in the rhein group was alleviated compared to the model group. We also applied fecal microbiome sequencing to identify changes in the microbiota and found that Lachnoclostridium, norank_o:Clostridia_UCG014 and Roseburia were higher in the gut of mice during acute liver injury, whereas Bacteroides and Akkermansiacea were lower in the acute liver injury group. Among these, Lachnoclostridium has been found to be significantly more abundant in the gut of high-fat diet rats (31, 32) but was significantly inhibited after our treatment. Roseburia has also been shown to increase in abundance in high-fat diet mice and has been found to be positively correlated with increased deoxycholic acid in the plasma, liver tissue, and feces of high-fat diet mice as well (33). Studies have shown that deoxycholic acid is a cytotoxic bile acid that can activate oxidative stress and promote hepatocyte apoptosis, causing various diseases of the liver (34, 35). Additionally, norank_o:Clostridia_UCG-014 is associated with ulcerative colitis, and it was significantly higher in the mice in our model group. In models with a tendency to self-heal, this bacteria has shown a downward trend (36). However, our rhein group showed signs of regulated DGalN/LPS-induced gut flora disturbance and suppressed abundance of Lachnoclostridium, norank_o:Clostridia_UCG-014 and Roseburia, and higher amounts of Bacteroides and Akkermansiacea. We further validated these results by LEfSe analysis and found that two species, Akkermansia_muciniphila and Lactobacillus_johnsonii, dominated the rhein-treated group. Akkermansiacea and Bacteroidesfen have been found to be lower in the feces of patients with cirrhosis and nonalcoholic fatty liver, respectively, and both have been shown to be negatively correlated with elevated calprotectin concentration in the feces of patients with cirrhosis (37). Akkermansia_muciniphila has recently been recognized as a next-generation probiotic strain for the treatment of obesity-related diseases (38), and studies have found that supplementation with Akkermansia_muciniphila can reduce the levels of blood markers related to liver dysfunction and inflammation (39) and improve oxidative stress-induced intestinal Apoptosis (40), reduce neutrophil infiltration (41), maintain intestinal barrier function, and promote of short-chain fatty acid (SCFA) secretion (42) as well, thereby remodeling the composition of gut microbiota. It also has preventive effects on fatty liver (40), alcoholic liver disease (41), and hepatic fibrosis induced by acetaminophen or carbon tetrachloride (42, 43), and it has s extracellular vesicles (EVs), which are the cell membranes of Gram-negative and Gram-positive bacteria that can interact not only with host cells but also with other microbiota. Akkermansia_muciniphila has also been found to improve intestinal permeability, modulate inflammatory responses, and prevent liver injury in HFD/CCl4-administered mice (44, 45). Lactobacillus_johnsonii is a type of lactobacillus that adheres to intestinal cells (46). Studies have shown that it can inhibit the cell adhesion of toxic bacteria in the intestine and help maintain intestinal microecology. In addition, Lactobacillus_johnsonii has been shown to improve bacterial translocation in cirrhotic rats with ascites, and bacterial translocation is closely related to mucosal oxidative damage and impaired intestinal permeability (47, 48). Therefore, we speculate that rhein can regulate the intestinal microbiota, induce a more favorable composition of intestinal microbiota, improve the intestinal barrier, and improve inflammatory and oxidative stress responses to prevent and treat acute liver injury. In the primary differential metabolic-pathway-amino-acid metabolism, a number of differential metabolites were screened in this study, among which Tryptophanol, N-acetylputrescine, L-Glutamine, N-Aarbamoylsarcosine, 2-Hydroxycinnamic acid, Ornithine, Citrulline, Maleic acid, L-Proline and L-Aspartic acid were down-regulated in the rhein treatment group. In addition, the rhein group showed up-regulated 3-Indoleacetic acid, Formiminoglutamic acid, L-Arogenate, Vanillylmandelic Acid, Stizolobate, Allysine, Oxoglutaric acid, Indole Acetaldehyde, Gentisic acid, 5hydroxyindoleacetaldehyde, 2-Isopropylmalic acid, Pipecolic acid, Imidazole acetic acid, acid riboside, L- 4-hydroxyglutamate semialdehyde, Citric acid, and Phenyl lactic acid. We consider these to be biomarkers that can be used to assess the effects of rhein treatment on fecal metabolites in mice. There were multiple metabolites that were significantly different between the normal control group and the model group but that returned to normal levels after rhein treatment. Among the metabolites up-regulated in the rhein group, L-4-hydroxyglutamate semialdehyde in the above-mentioned biomarkers is an organic compound of L-alpha-amino acids, and studies have found that amino acid metabolism disorders play an important role in the pathological process of drug-induced liver injury. In vitro experiments (49, 50) have found that L-alpha-amino acids can participate in the consumption of TAMa• free radicals, which are stable carbon-centered free radicals that are mostly consumed by oxidative metabolism in liver microsomes. Experimental results have shown that the consumption of TAMa• free radicals by L-alpha-amino acids is similar to that of glutathione. Abnormal amino acid metabolism, such as L-tyrosine and taurine, has been found to be associated with hydrazine-induced liver injury in vivo (51). In our experiment, two antibacterial components, ofloxacin-N-oxide and 1-hydroxy-1,3-diphenylpropan-2-one, in the up-regulated metabolism of the rhein group were also significantly higher in the fecal metabolites of these mice compared to the model group. Ofloxacin-N-oxide, a metabolite of Ofloxacin, has antibacterial effects on 150 pathogens such as Enterobacteriaceae and Haemophilus influenzae, and can produce antibacterial activity against some pathogens in patients with severe diseases (52). Similarly, 1-hydroxy-1,3diphenylpropan-2-one has shown antibacterial activity against 13 strains of methicillin-resistant Staphylococcus aureus (MRSA) (53). In this study, the rhein group down-regulated multiple metabolites related to liver diseases, such as tetranor 12-HETE, which is significantly associated with nonalcoholic fatty liver fibrosis and can be used as a noninvasive biomarker of liver fibrosis (54) and 9(S) –HODE, which is an endogenous fatty acid (PPAR) gamma agonist and is closely related to hepatic steatosis. After dietary intervention in obese adolescents, 9-HODE, ALT, triglyceride, and cholesterol levels have been found to be significantly reduced (55–57). So the metabolites may jointly improve acute liver injury through some direct and indirect pathways, including inhibiting harmful flora in the gut, regulating the disorder of metabolites such as amino acids, and participating in the oxidative metabolism of free radicals. We also analyzed liver gene expression in different groups of mice. The model group mice had different gene expression profiles compared to the normal group mice, and the rhein treatment group had significantly altered liver gene expression by up-regulating and down-regulating specific gene groups, with more similar gene expression to the normal control group. The GO and KEGG pathways that were significantly enriched in the rhein pretreatment group were similar to those of the metabolomics results and were mostly those responsible for regulating amino acid, lipid, and carbohydrate metabolism. This was also consistent with the metabolomic results from the mouse feces. Research has shown that rhein is a potential treatment for inflammatory diseases (58, 59), and cancer, and we found evidence for its anti-inflammatory, anti-oxidative (60), and anti-cancer effects in this study as well. The model group was enriched with genes that were positively related to Lachnoclostridium and norank_o:Clostridia_UCG-014, and the genes’ functions were mostly proinflammatory, pro-apoptotic, and cancer-promoting, aggravating the degree of liver damage. For example, CXCL2 can recruit neutrophils to help with immunity, induce immunosuppression, and promote HCC progression (61), and S100A9 levels have been shown to play a role in liver necroinflammation and necroptosis (62, 63). Furthermore, the Tnf gene can induce multiple mechanisms to initiate hepatocyte apoptosis, leading to subsequent liver injury (64), and Ereg and IL10 are up-regulated in acute liver injury and hepatocellular carcinoma, respectively (65, 66). however, the mice in the rhein group not only down-regulated the above genes that have a positive effect on liver injury, but also the genes that are positively related to enriched gut flora, that have anti-cancer properties, and that promote liver regeneration. For example, MFSD2A is known to help maintain the blood-brain barrier (67). Recent studies have found that it may also act as a new tumor-suppressing gene in regulating the cell cycle, and it plays an important role in matrix attachment as well (68). Experimental results have also shown that the mRNA and protein levels in cancer tissues are significantly lower than those in adjacent normal tissues (69, 70). Other studies have shown that MFSD2A+ is expressed in many tissues (especially in the liver) and is not only significantly downregulated in hepatocellular carcinoma but also able to repopulate the liver during hepatocyte regeneration (71). In addition, researchers have found that Bhlhe41 is negatively correlated with the transcriptional repressor capicua (CIC) (72) and that CIC is involved in immune regulation. When CIC is inhibited, it can promote follicular helper T (Tfh) and liver-resident memory-like CD8+. The differentiation of T cells (73, 74), both of which are important cell groups in human immunity, maintain the immune balance of the body, and studies have found that there is a causal relationship between the occurrence of immunity and cancer (75). Rhein may up-regulate anti-cancer and liver regeneration-promoting genes, and down-regulate pro-inflammatory, pro-apoptotic, and pro-oncogenes through intestinal flora, and may also alleviate acute liver injury caused by D-GalN/LPS.
Our results suggest that our rhein treatment alleviated D-GalN/LPS-induced acute liver injury in mice, improved intestinal flora disturbance, and modulated metabolic abnormalities and gene expression. From the perspective of gut microbes, we find that rhein may be able to help prevent and treat acute liver injury. Bacteroides and Akkermansiacea may have certain therapeutic effects on acute liver injury. Lachnoclostridium, norank_o:Clostridia_UCG-014 and Roseburia may have some exacerbating effects of acute liver injury. In addition, we described the relationship between microbiota and metabolites and microbiota and gene expression and found that gut microbiota is correlated with a variety of amino acid metabolites and gene expressions for immunity, apoptosis, and cancer. In the future, we will conduct further experimental studies on the mechanism of rhein in alleviating acute liver injury. And we will continue to improve the investigation of the correlation between rhein and intestinal flora and carry out experimental verification such as flora transplantation, to clarify whether the intestinal flora and metabolites regulated by rhein have an certain protective effect on acute liver injury.
The data presented in the study are deposited in the NCBI repository, accession number PRJNA891652, PRJNA892154.
The animal study was reviewed and approved by Capital Medical University.
SL, JH, and FW conceived and designed the experiments. JH, FW, SL, RY, and ZY were involved in the experimental study design, preparation, and review of this manuscript. All authors contributed to the article and approved the submitted version.
This research was funded by the Key medical major of Beijing sailing plan, severe liver disease with integrated traditional Chinese and Western medicine (No. zylx201819).
The authors thank AiMi Academic Services (www.aimieditor.com) for the English language editing and review services.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648668 | Steven Santino Leonardi,Eileen Yiling Koh,Lei Deng,Chenyuan Huang,Lingjun Tong,Jiong-Wei Wang,Kevin Shyong-Wei Tan | The synthesis of extracellular vesicles by the protistan parasite Blastocystis | 27-10-2022 | Blastocystis,gut microbiome,extracellular vesicles (EVs),cytokines,ultracentrifugation | Blastocystis is a genus of single-celled protist belonging to the stramenopile group. Prior studies have shown that isolates of Blastocystis subtype 7 (ST7) induced higher levels of intestinal epithelial cell damage and gut microbiota dysbiosis in comparison to other subtypes in in vivo and in vitro settings. Prior research has shown a link between gut dysbiosis and exposure to extracellular vesicles (EVs) produced by pathogenic microorganisms. This study demonstrates a protocol for the isolation of EVs from Blastocystis ST7 via ultracentrifugation. Nanoparticle tracking analysis and transmission electron microscopy were used to assess EV size and morphology. The protein content of isolated EVs was assessed by mass spectrophotometry and the presence of EV markers were evaluated by Western blotting. Finally, the EVs were cocultured with prominent human gut microbiome species to observe their effect on prokaryote growth. Our data shows that Blastocystis ST7 secretes EVs that are similar in morphology to previously characterized EVs from other organisms and that these EVs contain a limited yet unique protein cargo with functions in host-parasite intercellular communication and cell viability. This cargo may be involved in mediating the effects of Blastocystis on its surrounding environment. | The synthesis of extracellular vesicles by the protistan parasite Blastocystis
Blastocystis is a genus of single-celled protist belonging to the stramenopile group. Prior studies have shown that isolates of Blastocystis subtype 7 (ST7) induced higher levels of intestinal epithelial cell damage and gut microbiota dysbiosis in comparison to other subtypes in in vivo and in vitro settings. Prior research has shown a link between gut dysbiosis and exposure to extracellular vesicles (EVs) produced by pathogenic microorganisms. This study demonstrates a protocol for the isolation of EVs from Blastocystis ST7 via ultracentrifugation. Nanoparticle tracking analysis and transmission electron microscopy were used to assess EV size and morphology. The protein content of isolated EVs was assessed by mass spectrophotometry and the presence of EV markers were evaluated by Western blotting. Finally, the EVs were cocultured with prominent human gut microbiome species to observe their effect on prokaryote growth. Our data shows that Blastocystis ST7 secretes EVs that are similar in morphology to previously characterized EVs from other organisms and that these EVs contain a limited yet unique protein cargo with functions in host-parasite intercellular communication and cell viability. This cargo may be involved in mediating the effects of Blastocystis on its surrounding environment.
Blastocystis is a genus of unicellular stramenopilic gut parasites (Silberman et al., 1996), classified into over 20 subtypes (STs) (Stensvold et al., 2007). Human and animal infection by this parasite is ubiquitous, however it is limited by region and host species on a per-ST basis (Parkar et al., 2010; Alfellani et al., 2013). Infection occurs via the fecal-oral route, and is far more prevalent in developing nations (Tan, 2008). Variations in virulence have been observed both between and within Blastocystis subtypes (Wu et al., 2014b; Ajjampur and Tan, 2016). This paper focuses on ST7, isolates -B and -H. This subtype, in particular these isolates, have been observed to exhibit more virulent characteristics in comparison to the other commonly-studied human-infecting subtypes ST1 and ST4 (Yason et al., 2016; Yason et al., 2019). The process of Blastocystis colonization remains unclear, though research has identified legumain and cathepsin B-like proteases on the surface of the Blastocystis cell as critical virulence factors (Wu et al., 2010; Wawrzyniak et al., 2012; Nourrisson et al., 2016). The process by which the parasite attaches to intestinal epithelial cells is of particular importance, as it is immotile and cannot actively travel to the site of infection (Boorom et al., 2008). Blastocystis infection has been associated with both positive and negative health outcomes in the host. It has been correlated with increases in the biodiversity of the microbiome (Scanlan et al., 2014; Audebert et al., 2016) as well as inflammatory disorders such as irritable bowel syndrome (IBS) (Kesuma et al., 2019) and ulcerative colitis (Stensvold et al., 2013), cell damage-induced disorders such as colorectal cancer (Kumarasamy et al., 2017; Sulżyc-Bielicka et al., 2021), and dysbiosis of the gut microbiome (Beghini et al., 2017; Tito et al., 2018; Billy et al., 2021). Koch’s postulates remain unproven, however, and a definitive causal link between the parasite and a disease has yet to be found. ‘Extracellular vesicles’ (EVs) is a collective term for three types of biogenesis-dependent vesicle that are secreted by living cells (Shao et al., 2018; Huang et al., 2021). Bacteria, mammalian cells, and unicellular parasites have been shown to secrete EVs implicated in cellular communication, regulation, and virulence pathways (Twu et al., 2013; Olmos-Ortiz et al., 2017; Liu et al., 2018; Reclusa et al., 2020). EVs are understood to contain species-dependent proteins, lipids, and nucleic acids, however their function and generation is poorly understood (Twu et al., 2013; Artuyants et al., 2020; Elzanowska et al., 2020). EVs produced by both pathogenic and commensal microbiota species have been implicated in the alteration of gut homeostasis (van Bergenhenegouwen et al., 2014; Chang et al., 2020; Cuesta et al., 2021). In particular, both Entamoeba and Giardia EVs were shown to activate host immune and cytokine responses in an in vitro model (Evans-Osses et al., 2017; Sharma et al., 2020). Host-parasite interactions are commonly considered in gross molecular terms, while information exchange between the host and the parasite is poorly understood. The study of EVs is an intriguing putative avenue for this information exchange. Understanding of Blastocystis-produced EVs remains in a nascent stage, despite first being reported in 2001 (Tan et al., 2001). In this study, we report that Blastocystis subtypes 7B and 7H secrete EVs with similar physical characteristics and protein components to other microorganismal systems, and show that these EVs are capable of influencing in vitro host cells and microbiota species.
To determine whether Blastocystis ST7 secretes EVs, axenic isolates were maintained in IMDM supplemented with 5% horse serum at 37°C in anaerobic conditions. Conditioned medium was collected and subjected to a series of centrifugation cycles adapted from Théry et al. (Théry et al., 2006) ( Figure 1A ). The recommendations of the International Society for Extracellular Vesicles (ISEV) were used as a framework for vesicle characterization (Thery et al., 2018). Visualization using transmission electron microscopy (TEM) showed cup-shaped morphology typical to EVs (Rikkert et al., 2019) ( Figure 1B ). Blastocystis ST7B secreted an average of 1.70*1011 particles mL-1, while ST7H averaged 1.05*1012. Using nanoparticle tracking analysis (NTA), average EV size was measured at 230.4 ± 69.5 mm and 161.6 ± 72.9 mm for 7B and 7H respectively ( Figure 1C ). This data demonstrates that the collected media contained vesicles with a size and shape similar to those produced by other protists (Twu et al., 2013; Sharma et al., 2020).
Mass spectrophotometry was used to determine the protein content of ST7B and ST7H EVs and whole cell lysates (WCLs). Proteins with two or more peptides aligned to the UniProt database were included for downstream analysis. A total of 341 and 321 proteins were identified from ST7B and ST7H EV, along with 1503 and 1517 proteins from ST7B and ST7H WCLs respectively. Gene ontology (GO) analysis using the PANTHER database (Mi et al., 2021) showed a reduction in proteins associated with the molecular GO term ‘structural molecule activity’ in both assessed strains of ST7 EV relative to WCLs ( Figure 2A ). There was also a relative increase in proteins with transporter activity in ST7 EVs relative to WCLs. In both strains of WCL, the top four biological classification GO terms covered the vast majority (≥99%) of detected proteins. In EVs, the proportion of proteins comprising other biological GO terms outside of these four increased to 14 and 16% of the total.
The ST7 EV proteomes were filtered. Identified proteins with a coverage of >2% were excluded. The filtered EV proteomes were then compared with the Entamoeba histolytica EV proteome and the Top 100 most common EV proteins as listed on the Vesiclepedia database at time of analysis (Kalra et al., 2012; Sharma et al., 2020). Nine proteins were found to be shared between ST7 EVs and the Top 100 out of a total of 154 for ST7H and 159 for ST7B. This represents ~6% of the total number of observed proteins in each subtype EV. Nineteen proteins were shared between ST7 and Entamoeba EVs. ST7H EVs expressed more proteins in common with Entamoeba than ST7B. A significant portion of the assessed proteins, 46 (28.9% of 7B total) and 36 (23.4% of 7H total) were unique to that specific strain of Blastocystis ST7.
The tetraspanin family of proteins, known pan-EV markers (Yáñez-Mó et al., 2015) were absent from the ST7 EV proteome. This absence has also been observed in other parasite-derived EVs (Tzelos et al., 2016; Arab et al., 2019; Hansen et al., 2019). NCBI BLAST validation against an existing Blastocystis genome reference (Denoeud et al., 2011) identified no homologs to the tetraspanin family within the genetic code.
Western blotting was employed to confirm a lack of WCL contamination in existing EV samples ( Figure 2C ). The proteins ALIX (programmed cell death 6 interacting protein), TSG101 (tumour susceptibility gene 101), and RAB5C (a member of the RAS oncogene family), which were identified in our mass spectrophotometry were likewise present in our western blotting. APOA1 (apolipoprotein A1) was used as an EV purity control, as it has been previously in the literature as an indicator of WCL contamination (Karimi et al., 2018; Thery et al., 2018). APOA1 was not observed in our EV samples.
ST7 EVs were labeled using CFSE. HT29 cells were then incubated in the presence of varying concentrations of labelled EVs for one hour at 37°C. Confocal imaging showed uptake and internalization of the labelled EVs at all concentrations ( Figure 3A ). Some cells showed visible signs of viability loss, particularly membrane blebbing. The cells were stained with propidium iodide (PI) and assessed via flow cytometry ( Figure 3B ). HT29 cells incubated with ST7 EVs showed increased PI fluorescence. 10.8% and 11% of cells were positive for PI when incubated with ST7B and ST7H respectively, in comparison with <1% of control cells ( Figure 3C ).
In a 2019 study, our lab showed that coculture with Blastocystis ST7 can influence the growth of some gut bacteria species (Yason et al., 2019). Here, we investigated whether similar effects are observed during coculture with ST7 EVs. We cocultured E. coli, B. longum, and L. brevis alternately with a standard number of EVs (107), as well as the concentration of EVs required to observe the effect in the previous experiment (80 µg). B. longum growth was reduced when cultured with ST7 EVs, while E. coli growth increased when cultured with ST7B EVs. L. brevis remained unaffected. These results are similar to those observed in the above study.
A 2014 study by Teo et al. (Teo et al., 2014) showed that Blastocystis ST7, including isolate ST7B, is capable of altering NF-κB levels in THP1-Blue cells. We used qPCR to assess whether ST7B EVs can induce changes in three NF-κB-associated cytokines: IL-1β, IL-6, and TNF-α. When cocultured with ST7B EVs, IL-1β expression significantly increased, while IL-6 expression decreased.
This study is the first to identify and characterize the EVs synthesized by Blastocystis, drawing attention to their putative role in parasite-host interactions. The two ST7 isolates (ST7B and ST7H) used in this study have been reported to be more virulent (Wu et al., 2014a; Wu et al., 2014b), display increased resistance to antiparasitic drugs (Mirza et al., 2011), and induce host immune responses to a greater extent (Yason et al., 2016) compared to other Blastocystis subtypes. In particular, Blastocystis ST4 has been shown to have a beneficial effect on the host (Deng et al., 2021; Deng and Tan, 2022). Currently, the divergence in observed pathogenicity of these and other subtypes has not been explained, and differences in EV characteristics offer a potential solution. We showed that ST7 EVs conform to typical EV size and structure as characterized elsewhere (Twu et al., 2013; Evans-Osses et al., 2017). We also established the number of EVs synthesized by a consistent number of Blastocystis; these will be useful points of comparison in future studies ( Figure 1 ). We defined the ST7 EV proteome via mass spectrophotometry and used western blotting to confirm the presence of canonical EV proteins ALIX, TSG101, and RAB5C ( Figure 2 ). This confirms that we were able to successfully isolate EVs originating from Blastocystis. The absence of the pan-EV marker tetraspanin was noted, however this may be another instance of the absence of tetraspanins from parasite-produced EVs (Tzelos et al., 2016; Arab et al., 2019; Hansen et al., 2019). Gene Ontology analysis of the identified EV proteins showed an enrichment of biological classification GO terms not common in the WCL. Proteins not described by the top four biological GO terms comprised approximately less than 1% of the ST7 WCL, compared to ~15% of the ST7 EV proteome. Similarly, 169 of the 203 filtered ST7 EV proteins assessed against Vesiclepedia in Figure 2B were found to be unique to Blastocystis. A further 82 of those proteins were unique to either ST7B or ST7H. This suggests that Blastocystis EVs can possess a unique protein cargo that diverges even between isolates of a single subtype. EVs in general have been noted to exhibit significant variability, so further research must be performed to confirm these results (Tiruvayipati et al., 2020; Newman et al., 2021). Fluorescence microscopy demonstrated that ST7 EVs can be taken up by a human in vitro model cell line, and that they have a negative effect on the viability of those cells as represented by an increase in propidium iodide uptake ( Figure 3 ). We demonstrated effects of ST7 EVs on gut microbiota ( Figure 4 ) consistent with those previously observed during Blastocystis – prokaryote coculture experiments (Yason et al., 2019) – specifically, an inhibition of the beneficial gut species B. longum (Wong et al., 2019) when incubated with a high concentration of EVs. Finally, we characterized the influence of ST7B EVs on inflammatory cytokines within differentiated THP-1 cells, showing that IL-1β expression increases with an increasing concentration of ST7B EVs ( Figure 5 ). IL-1β is a pro-inflammatory cytokine associated with gastrointestinal cancer and T-cell activation (Lu et al., 2005; Ben-Sasson et al., 2013). The effects of ST7 EVs on E. coli growth and IL-6 expression were statistically significant but not dose-dependent, necessitating further research to validate the observed effects. Our results show evidence for Blastocystis EVs inhibiting the growth of a beneficial gut prokaryote, reducing the viability of a human gastrointestinal cell model, and increasing the expression of a pro-inflammatory, cancer-inducing cytokine in that same model. These correlate with dysbiosis, inflammation, and damage to the host gut epithelium – all features of diseases associated with Blastocystis ( Figure 6 ). This suggests that Blastocystis EVs are likely to play a role in the deleterious effects the parasite has been observed to induce in the gut.
The avenues by which beneficial and pathogenic Blastocystis subtypes affect host cells and host microbiota remain poorly characterized. Understanding these avenues will help shed light on the source of Blastocystis virulence and the reasons for highly divergent behaviour between Blastocystis subtypes. This paper presents EVs as a likely mediator of interactions between Blastocystis, the microbiome, and the host. Future work can use our experiments as embarkation points to investigate the results shown here in greater depth, with an eye on establishing a causal relationship between EVs and effects, or investigating more complex model organisms. Of importance is the need to determine whether these effects are observable in other Blastocystis subtypes, and if so, whether their degree of severity correlates with the virulence of the subtype and isolate.
Human Blastocystis isolates were acquired from patients at the Singapore General Hospital in the early 1990s, before the establishment of the Institutional Review Board at the National University of Singapore (NUS). Both the Blastocystis ST7 isolates -B and -H are maintained at the microbial collection at the Department of Microbiology and Immunology, NUS. Previously axenized cultures of both Blastocystis ST7B and ST7H (Ho et al., 1993) were maintained in pre-reduced Iscove’s Modified Dulbecco’s Medium (IMDM) (Gibco, New Zealand) supplemented with heat-inactivated 5% horse serum (Gibco, New Zealand) These cultures were incubated anaerobically with Anaerogen gas packs (Oxoid, USA) at 37°C in jars and subcultured on a weekly basis.
Human colorectal adenocarcinoma cell line HT-29 (ATCC Cat. No. HTB-38) cells were maintained in T-75 flasks (Corning, USA) in a humidified incubator with 5% CO2 at 37°C. The Dulbecco’s Modified Eagle’s Medium (DMEM; Thermo Scientific, USA) was supplemented with 10% heat-inactivated FBS (Gibco, USA) and 1% each of penicillin-streptomycin, non-essential amino acids (Gibco, USA) and sodium pyruvate (Gibco, USA), and designated as complete medium. HT-29 cells were seeded at approximately 1 x 105 cells mL-1 per well and incubated for 48 hours to achieve at least 80% confluency, followed by synchronization of cells with serum-free DMEM for 24 hours. The HT-29 cells were then used for confocal imaging supplemented with EVs at 80 µg and/or ~1 x 107 Blastocystis ST7 (whole) cells mL-1, respectively. HT-29 cells in culture medium without EVs were included as control.
Escherichia coli (ATCC 11775), Bifidobacterium longum (ATCC 15707) and Lactobacillus brevis (ATCC 14869) were cultured and maintained on Luria-Bertani (LB), Bifidus Selective Medium (BSM) and deMan, Rogosa, Sharpe (MRS) medium (all from Sigma-Aldrich, USA), respectively, in broth and agar formats. All cultures were incubated in an anaerobic jar with AnaeroGen™ gas pack (Oxoid, USA) at 37°C. Absorbance readings of bacterial broth cultures prior to experiments were done using the spectrophotometer at 600 nm wavelength.
An ultracentrifugation protocol established by Théry et al. (Thery et al., 2018) was used to isolate EVs, where serum-free cell culture medium (CCM) was centrifuged at 300 x g for 10 minutes at 4°C to remove larger cells by pelleting. The supernatant was transferred to a new tube and centrifuged at 2000 x g for 10 minutes at 4°C to pellet dead cells, followed by another round of centrifugation at 10,000 x g for 30 minutes at 4°C to collect the cell debris. Clarified CCM was then centrifuged in an Optima™ L-90 ultracentrifuge (Beckman Coulter, USA) at 100,000 x g for 70 minutes at 4°C with a SW41Ti swinging-bucket rotor (Beckman Coulter, USA) to pellet EVs, which are washed with 1xPBS and re-centrifuged for another 70 minutes at 4°C at 100,000 x g. The resulting EV pellet was resuspended in 100μL of 1xPBS and stored at -80°C until usage.
Freshly isolated EVs were prepared for visualization by transmission electron microscopy (TEM) by fixation with 2.5% glutaraldehyde at 4°C for 1 hour. A volume of 20 µL per sample were incubated for 5 minutes on a Formvar Film 200 mesh, CU, FF200-Cu grid (Electron Microscopy Sciences, USA). Negative staining was then performed by incubating with 2.5% gadolinium triacetate for 1 minute. Fixed samples were viewed on the FEI TECNAI SPIRIT G2 (FEI Company, USA) at room temperature. Microscopy images were acquired from three independent experiments with three technical replicates (n=9).
EVs were diluted in filtered phosphate-buffered saline to the optimal concentration for data acquisition. Five 60-second video clips were recorded at room temperature. These video clips were subsequently analyzed with detection threshold of 4-6 with NanoSight NTA software v3.2 (Nanosight, United Kingdom). Data were acquired from three independent experiments with three technical replicates (n=9).
Proteins were extracted from the UC- isolated EVs by homogenization with lysis buffer (50mM triethylammonium bicarbonate, 8M urea, 1% sodium deoxycholate) at a 1:2 pellet volume to lysis buffer ratio. Blastocystis ST7B and ST7H cell lysates were included as controls. The mixture was incubated at room temperature for 20 minutes with vortex every 5 min, followed by centrifugation at ≥21,000 x g at 4°C for 30 minutes. Resultant supernatant was transferred to new microcentrifuge tube and analyzed. EVs and cell lysate protein samples were analyzed in two biological replicates over two independent technical runs. Trypsin digestion of samples were performed before using the TripleTOF® 5600 System (Sciex, USA). The protein sequences were searched against the UniProt protein database using ProteinPilotTM Software v5.0 (Sciex, USA). The Paragon search algorithm was used with the following parameters set to default values: trypsin specificity, cys alkylation, thorough false discovery rate (FDR). Blastocystis hominis was used as the reference organism (Denoeud et al., 2011). Only peptides occurrence of ≥ 2 at 95% confidence levels (CI) with FDR of 1% were taken into consideration. Venn diagram of Blastocystis ST7 EV proteins that were unique, shared and common to Entameoba histolytica EVs (Sharma et al., 2020) and the Top 100 commonly identified EV proteins (Kalra et al., 2012) were generated using the online version of Venny v2.1.0 (Oliveros, 2007). Functional pathways associated with EV proteins identified from Blastocystis ST7 were inferred by KEGG analysis (Kanehisa Laboratories, Japan), and Gene Ontology (GO) annotation results were obtained using the PANTHER database (Mi et al., 2021).
Blastocystis-derived EVs were lysed with RIPA (Invitrogen, USA) containing Protein Inihibitor (100x) on ice for 30 minutes. Proteins were quantified with the PierceTM BCA Protein Assay Kit (Thermo-Fisher Scientific, USA) on the Infinite® 200 PRO NanoQuant (TECAN, USA). Approximately 200 µg of protein from each sample were lysed in 4x loading buffer (Invitrogen, USA) by heating for 10 minutes at 98°C; thereafter loading of the samples onto an SDS-PAGE gel for protein bands separation at 110 V for approximately 1 hour. The Spectra Multicolor Broad Range Protein Ladder (Thermo-Fisher, USA, Cat. No. 26623) was used for reference. Samples were transferred to nitrocellulose membranes and blocked with 5% BSA in Tris-buffered saline + 1% Tween20 (TBST), followed by overnight incubation with primary antibodies: rabbit monoclonal anti-ALIX (Abcam, USA, Cat. No. ab186429), rabbit monoclonal anti-RAB5C (Abcam, USA, Cat. No. ab199530), rabbit monoclonal anti-TSG101 (Abcam, USA, Cat. No. ab125011) and rabbit monoclonal anti-Apo-A1 (Abcam, USA, Cat. No. ab52945). Membranes were washed three times with TBST and mouse anti-rabbit HRP-conjugated secondary antibody incubation was carried out for 60 minutes at room temperature. Protein bands were visualized using the Electrogenerated Chemiluminescence (ECL) Western Blotting Substrate Kit (Millipore, USA, Cat. No. 1825002) and imaging was performed on a ChemiDoc (Bio-Rad Laboratories, Inc., USA). Data were acquired from three independent experiments.
Purified Blastocystis-derived EVs were labeled with Vybrant® CFDA SE Cell Tracer Kit (Invitrogen, USA; hereafter referred to as CFSE) as previously described (Morales-Kastresana et al., 2017). In brief, 100 µL of CFSE (200 µM) was added to 80 µg of EVs and the mixture were incubated for 15 minutes at 37°C. Cells were re-pelleted by centrifugation, resuspended in fresh pre-warmed PBS and incubated for a further 30 minutes to ensure complete modification of the probe. The CFSE-stained EVs were then washed once with PBS by centrifugation, followed by resuspension in serum-free culture medium. For live confocal examinations of EV uptake and intracellular localization, 105 cells mL-1 HT-29 cells were seeded into an 8-well µ-Slide (ibidi GmbH, Germany) and cultured as described above before the addition of CFSE-stained EVs at either 200 µg or 500 µg for co-incubation at 37 °C for 1 hour. The supernatant was subsequently replaced with a mixture consisting of both the Hoescht 33452 and CellMask™ Plasma Membrane Stain Deep Red (Thermo-Fisher, USA) dyes, with incubation at room temperature for 15 minutes before live imaging. All images were taken on the Olympus FV3000 Confocal Laser Scanning Microscope (Olympus Corporation, USA) with oil immersion (n=3), with at least three images taken per sample. Negative control consisting of only stained HT-29 cells without CFSE-stained EVs were included.
The CFSE-stained EV uptake by HT-29 cells was quantified on a Beckman Coulter CytoFLEX Flow Cytometer (Beckman Coulter, USA). Cells were prepared as described for confocal microscopy, and were trypsinized with 0.25% EDTA-trypsin for 10 minutes at 37°C in the 5% CO2 incubator. The detached cells were pelleted by centrifugation followed by supernatant removal; a volume of 0.5mL of 1xPBS was used then to resuspend the cell pellet followed by the addition of 0.5 µL of PI (0.1mg mL-1) prior to flow cytometry. Data were calculated from three independent experiments with two technical replicates per sample (n=6).
L. brevis, B. longum and E. coli cells were washed twice in PBS at 1000 x g for 10 minutes. Prior to the co-incubation, an aliquot of the cells was enumerated using the drop-plate method outlined in Yason et al. (Yason et al., 2019) to determine the number of initial cells. A concentration of 80 µg EVs were then incubated with 1mL of PBS-washed bacterial cells for 24h at 37°C in pre-reduced PBS under anaerobic conditions. After 24 hours, the bacterial numbers were counted and the bacterial colony-forming units (CFUs) mL-1 was determined when the colonies appeared on the agar plates. Controls containing only bacterial cells or co-cultured with 107 Blastocystis cells mL-1 in pre-reduced PBS were included. The number of each bacterial cell type(s) were enumerated from three independent experiments with three technical replicates (n=9) and presented as % relative to control.
THP1 cell line (Invivogen, USA) were maintained in RPMI-1640 (Gibco, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco, USA), 100 U mL-1 penicillin, and 100 μg mL-1 streptomycin (ThermoFisher, USA). Cells were seeded onto a T-75 flask at a density of 2.5 × 106 cells in a humidified incubator at 37°C and 5% CO2. To differentiate THP1 cells into macrophages, cells were stimulated with 25 ng/mL phorbol 12-myristate 13-acetate (Sigma-Aldrich, USA) for 48 h. PMA-differentiated THP1 cells were then co-incubated with Blastocystis ST7-derived EVs (50, 100, and 200 ng mL-1, respectively) or PBS (negative controls) for 48 h.
Total RNA was extracted from Blastocystis ST7-derived EVs and PBS-treated THP1 cells using RNAzol RT (Sigma-Aldrich, USA) according to the manufacturer’s protocols. Complementary DNA was synthesized using the iScript cDNA kit (Bio-Rad, USA). The SsoAdvanced™ Universal SYBR Green Supermix (Bio-Rad, USA) was used in all qPCR amplifications on an Applied Biosystems 7500 Fast Real-Time PCR System (Applied Biosystems, USA). qPCR reaction was carried out in a total volume of 10 μL, which comprised of the master mixture and 2 μL of cDNA template. The former contained 5 μL SsoAdvanced™ Universal SYBR Green Supermix (2×), 0.3 mM of each primer, made up to 10 μL with nuclease-free water. Actin was used as a house-keeping gene, and three pro-inflammatory cytokines TNF-α, interleukin (IL)-6, and IL-1β were detected by qPCR. Fold change was determined by the 2−(ΔΔCt) method (Zhang et al., 2013).
Data were presented as mean ± standard error of the mean (SEM) of triplicate experiments. Either two technical replicates (n=6) or three technical replicates (n=9), as described in the respective method sub-sections, were done. Acquired data were statistically analyzed using GraphPad Prism v8.0 (Swift, 1997). The analysis of variance (ANOVA; comparisons of more than two groups) were computed, with p values of <0.05 taken to be of statistical significance.
The original contributions presented in the study are publicly available. This data can be found here: https://data.mendeley.com/datasets/4r7xgpdrj4.
KT and EK conceived the study and participated in its design and coordination. EK conducted experiments and organized collaborations with CH, LT, and J-WH pertaining to Figures 1 and 2 . CH performed NTA analysis for Figure 1C , while LT performed protein work for Figure 2C . J-WH coordinated their work. DL conducted experiments pertaining to Figure 5 . SS wrote and edited the manuscript, with KT finalizing the submission. All authors read and approved the manuscript. All authors contributed to the article and approved the submitted version.
This research work is supported by the National University of Singapore, under its Department of Surgery Seed Grant (N-176-000-100-001), and Yong Loo Lin School of Medicine ‘Blastocystis under One Health’ grant (NUHSRO/2022/074/NUSMed/Blastocystis/LOA). SS acknowledges the support of the SINGA PhD Research Scholarship. EK was supported by the Ministry of Education – Post-Doctoral Fellowship from the National University of Singapore. CH was supported by the Ministry of Education research scholarship. LT was supported by the China Scholarship Council (202006330097). JWW would like to acknowledge support by the NUS NanoNASH Program (NUHSRO/2020/002/NanoNash/LOA) and the NUS Yong Loo Lin School of Medicine Nanomedicine Translational Research Program (NUHSRO/2021/034/TRP/09/Nanomedicine).
The authors acknowledge the excellent administrative and technical support of Ms. Geok Choo Ng. The authors also thank the following personnel for their excellent technical assistance and advice: Mr. Cyrill Kafi Salim for his technical advice on EV uptake imaging protocols, Ms. Shu Ying Lee for confocal microscopy and Ms. Xiaoning Wang for flow cytometry at the NUS Medicine Flow Cytometry Unit, and Dr. Qifeng Lin and Mr. Teck Kwang Lim at the SINGMASS unit at the National University of Singapore, Singapore.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
|
PMC9648670 | Guanzhou Zhou,Nana Zhang,Ke Meng,Fei Pan | Interaction between gut microbiota and immune checkpoint inhibitor-related colitis | 27-10-2022 | immune checkpoint inhibitor,gut microbiota,colitis,diarrhea,microbiome | Immune checkpoint inhibitors (ICIs) have become a promising therapeutic strategy for malignant tumors, improving patient prognosis, along with a spectrum of immune-related adverse events (irAEs), including gastrointestinal toxicity, ICI-related colitis (IRC), and diarrhea. The gut microbiota has been suggested as an important regulator in the pathogenesis of IRC, and microbiota modulations like probiotics and fecal microbiota transplantation have been explored to treat the disease. This review discusses the interaction between the gut microbiota and IRC, focusing on the potential pathogenic mechanisms and promising interventions. | Interaction between gut microbiota and immune checkpoint inhibitor-related colitis
Immune checkpoint inhibitors (ICIs) have become a promising therapeutic strategy for malignant tumors, improving patient prognosis, along with a spectrum of immune-related adverse events (irAEs), including gastrointestinal toxicity, ICI-related colitis (IRC), and diarrhea. The gut microbiota has been suggested as an important regulator in the pathogenesis of IRC, and microbiota modulations like probiotics and fecal microbiota transplantation have been explored to treat the disease. This review discusses the interaction between the gut microbiota and IRC, focusing on the potential pathogenic mechanisms and promising interventions.
Immune checkpoint inhibitors (ICIs) have received great attention as they have rapidly altered the treatment landscape for multiple tumors, including lung cancer, metastatic melanoma, and urinary epithelial carcinoma. ICIs block inhibitory molecules, such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1) and its ligand 1 (PD-L1) and enhance anti-tumor T-cell activity providing clinical benefits in many patients with advanced cancers (1–3). Yet, multiple organs like skin, lung, liver, and digestive tract are susceptible to the unrestrained immune response activation by the utility of ICIs, which developed to the immune-related adverse events (irAEs) ultimately, including ICI-related colitis (IRC) and diarrhea, which are major causes of ICI discontinuation (4–6). Studies have suggested that the occurrence of diarrhea and colitis is associated with the ICI used. For example, Tandon et al. performed a meta-analysis to evaluate the risk of colitis and diarrhea in patients with advanced melanoma treated with ICIs (anti-PD-1 or anti-CTLA-4 therapy) and concluded that diarrhea and colitis are more frequent in patients treated with CTLA-4 inhibitors (7). Another study showed that patients treated with anti-CTLA-4 therapy have a higher rate of diarrhea (31.8% in anti-CTLA-4 alone versus 10.5% in anti-PD-1 alone) and colitis (7.7% in anti-CTLA-4 alone versus 0.8% in anti-PD-1 alone); also, diarrhea seems to be more common in patients treated with dual ICI therapy than in those with a single-ICI agent (8). One possible explanation for this preference is that the CTLA-4 receptor is often expressed on the surface of CD4+ and CD8+ cells, subsets of B cells and thymocytes, resulting in inhibition at the initial step in an immune response while the PD-1 and its ligand blockades aim at late T-cell proliferation, causing a more localized immune reaction (9, 10). Yet, the mechanisms of IRC are still not fully understood and several key aspects have been proposed: (a) the cross-reactivity of the common antigens on tumor and healthy tissues; (b) activation of humoral immunity like elevated pre-existing autoantibodies level; (c) modulation of pro (anti)-inflammatory cytokines; (d) enhanced complement-mediated inflammation; (e) regulation of effector or suppressor immune cells (10, 11). Moreover, different management is proposed based on the IRC severity. Mild or moderate IRC is closely observed and applied with supportive treatment. Higher-grade toxicities cases may discontinue the ICI course and receive corticosteroids or immunosuppressive therapies such as tumor necrosis factor-α (TNF-α) inhibitors (e.g., infliximab) and anti-integrin agents (e.g., vedolizumab) (11). Recent studies have highlighted an indispensable role of the gut microbiota in the communication between ICI and patients. The anticancer immunotherapy relies on the immunization with some species like Bacteroides fragilis (12). Bifidobacterium and Faecalibacterium promote ICI efficacy with augmented dendritic cell function and T cell accumulation in the tumor microenvironment (13, 14). Fecal microbiota transplantation (FMT) has also demonstrated the ability of overcoming resistance to anti-PD-1 therapy in melanoma patients (15). Besides, emerging evidence emphasizes the critical involvement of gut microbiota in the pathogenesis of IRC, patients vulnerable to IRC development seem to have a distinct microbiota profile ( Table 1 ) (16–22) and the microbiota modulation offers a novel alteration for the treatment. This review discusses the interaction between the gut microbiota and ICI-related colitis, focusing on the potential pathogenic mechanisms and promising interventions.
Accumulating studies indicate that the gut microbiota signature has a strong link with IRC. Chaput et al. (16) collected fecal samples from twenty-six metastatic melanoma patients before the ICI therapy and analyzed the gut microbiota 16S rRNA gene sequencing data. According to the characteristics of baseline microbiota composition, patients were divided into 3 clusters. There was a high proportion of Faecalibacterium and other Firmicutes in the microbiota composition of patients belonging to Cluster A. Cluster B was enriched in Bacteroides, and Cluster C, Prevotella. At the phyla level, patients in Cluster A were prone to develop colitis, with a preference of Firmicutes, while patients without colitis had more Bacteroidetes (like Cluster B). Specifically, Bacteroides vulgatus, and Faecalibacterium prausnitzii A2-165 were detected as potential biomarkers for colitis absence during ICI therapy, whereas several OTUs in Firmicutes phylum, and Gemmiger formicilis ATCC 27749 were detected to be with increased risk of colitis. Meanwhile, there is an overlap that gut microbiota composition associated with IRC also promotes ICI clinical response. For example, Faecalibacterium magnifies systemic immune response mediated by up-regulated antigen presentation and intensified effector T cell function. These overactive immune cells not only infiltrate in tumor microenvironment, strengthening ICI anti-tumor effect, but attack normal intestinal mucosal and induce IRC. In another study of advanced-stage melanoma patients undergoing ICI, stool samples were collected before, during, and after the treatment. Two natural gut microbiome clusters with distinct profiles were identified, and patients with a high proportion of Bacteroides dorei in gut microbiota had high risk of irAE, while the Bacteroides vulgatus was identified as a specific dominance strain in the low-risk cluster (18). Apart from the specific strain, it is inferred that the IRC is associated with decreased diversity of gut microbiome. The low richness of abundance in gut microbiota often refers to a fragile immune homeostasis, which are easily perturbed by ICIs intervention as observed in IRC patients. Mao et al. (22) displayed that ICI-treated hepatobiliary cancer patients with severe diarrhea tends to have lower phylogenetic diversity of gut microbiota. They also recognized several enriched taxa with significant differentiation between the severe and mild diarrhea groups. The enrichment of Dialister genus, which belongs to the Firmicutes phylum, was observed in the mild group. Notably, severe diarrhea patients had a higher abundance of Prevotellamassilia timonensis, which has been suggested as valuable biomarker. Overall, it could be speculated that a higher diversity of gut microbiome may be a protective factor against IRC.
Patients with malignant tumor tend to experience infection due to their impaired immune system, causing higher exposure to antibiotics. In clinical practice, about 70% cancer patients receive antibiotics during the ICI treatment, how they affect IRC deserves exploration (23). Epidemiological studies emphasized that antibiotic therapy weakens ICI efficacy and shortens patient survival across malignancies (24). Antibiotics alter the composition of gut microbiota, leading a decreased bacterial-mediated secondary bile acids production and an increased inflammasome signaling, thus promotes a pro-inflammatory state, susceptible to IRC (25). As a result, the history of antibiotic use may be an indicator of IRC. Researchers established an ICI-related colitis mice model by combining dextran sulfate sodium (DSS) and anti-CTLA-4 to simulate the inflammation condition. Compared to the control group (with ICI isotype and DSS), mice with anti-CTLA-4 pretreatment showed higher mortality, more body weight loss, and worse histopathological scores, thus declaring that preprocess of ICI exaggerates the DSS-induced inflammation in mice. Moreover, pretreatment with vancomycin provoked an even more severe, largely fatal form, indicating that a Gram-positive component of the microbiota had a mitigating effect on colitis (26). Due to the limitation of mice models, they generally do not develop colitis after ICI treatment, unlike malignancy patients, in the absence of chemical damage or genetic defects. Therefore, the potential influence of additional DSS process requires to be further explored. A clinical observational study including 832 patients with ICI treatment exhibited that antibiotic exposure is strongly correlated to grade 3 or 4 irAEs (20). Mohiuddin et al. (19) investigated 568 patients with stage III and IV melanoma receiving immunotherapy. Patients treated with antibiotics within 3 months prior to the first infusion of ICI had significantly worse overall survival and a greater incidence of colitis. The incidence and severity of colitis varies according to some factors. Anaerobic antibiotics were associated with expanded immunosuppressant use, hospitalization, intensive care unit admission due to IRC, and elevated severity grades. At the onset of colitis, the empirical antibiotic group had a higher recurrence rate and colitis severity than the group receiving antibiotics when there was positive evidence of infection. Antibiotic therapy changed the microbiome taxonomic diversity profoundly, inducing a loss of protective bacteria and an impaired immune homeostasis, thus with a worse prognosis. Therefore, it provides an implication for clinical practice that antibiotic use should be taken into consideration carefully in cancer patients.
The species and diversity of gut microbiota influence the development of IRC; yet, the underlying mechanism is still unclear. Deciphering the biological mechanisms is critical for optimizing patient outcome. Multiple results highlighted the involvement of gut microbiota in IRC pathogenesis, not only through direct effect of bacteria, but also through indirect mechanisms like regulating metabolites, cytokines and immune cells. It provides a better understanding of the disease and some novel targets for intervention. This part depicts early evidences and hypothetical scenarios, then discusses the potential mechanisms of the interaction between gut microbiota and ICI-related colitis ( Figure 1 ).
Mounting evidences illustrated that the bacteria exert direct effect via extracellular enzymes, lipopolysaccharide (LPS) and others in their interaction with IRC. Higher levels of Stenotrophomonas have been found in severe diarrhea patients receiving ICI treatment (21). Stenotrophomonas is considered an environmental bacterium commonly found in the respiratory or digestive tract. It often causes pulmonary diseases like Stenotrophomonas maltophilia pneumonia and diarrhea or enteritis in some cases (27, 28). Malignancy patients with impaired immunity are predisposed to this strain and tend to experience severe diarrhea or IRC if infected (27). A range of extracellular enzymes by Stenotrophomonas maltophilia, including DNase, RNase, lipases, protease, and elastase, may be key factors in pathogenesis. Assisted with these enzymes, the strain breaks down the tight junction, decomposes mucin, invades tissue and causes IRC. Bacterial enzymes also play a critical role in the pathogenesis of Prevotellmassilia timonensis, a subspecies strain of Prevotella, which is associated with severe diarrhea in ICI-treated patients. It secretes sialidase, breaks sialic acid and degrades the mucin, increasing the intestinal barrier permeability (29). Dendritic cells (DCs) are also involved in its pathogenic mechanism (30). Endotoxin-like lipopolysaccharide (LPS) is another virulence factor promoting the inflammation. It actives immune cells through the toll-like-4 receptors, synthesizes and releases a variety of cytokines and inflammatory mediators, causing inflammation (31, 32). Compared to the control group, the LPS level was reduced in serum and feces of mice fed with B. vulgatus, which has a strong correlation with few irAEs, indicating a potential protective mechanism via LPS reduction (33). Microbial anti-inflammatory molecules (MAMs) have same favorable effects, which contain a series of proteins produced by Faecalibacterium prausnitzii. In animal models, MAMs exhibit anti-inflammatory effect by blocking the NF-κB pathway and inhibiting the pro-inflammatory Th1 and Th17 immune responses. It also consolidates the gut barrier by upregulating the tight connection gene Zo-1 (34, 35). Therefore, Faecalibacterium prausnitzii could prevent patients from IRC and act as a biomarker for colitis absence.
The gut microbiota consumes carbohydrates and produces variable bioactive molecules, modulating the host immune system differently (36). SCFAs are one of the most extensively characterized classes of microbial metabolites (37, 38). Bacteria break complicated carbohydrates into simple fatty acids like acetate, propionate, and butyrate. These small molecules supply energy for intestinal epithelial cells and exert diverse effects on immune cell function and cytokine production (39). The anti-inflammation characteristic of butyrate is partly attributed to inhibiting the NF-κB activation and its downstream pathway, which in turn reduces the pro-inflammatory cytokines, such as IL-8, and increases anti-inflammatory factors like IL-10. The butyrate also induces tight connection protein expressions in the mucosa and consolidates the gut barrier (40). Indeed, a higher abundance of butyrate-producing Faecalibacterium prausnitzii A2-165 was detected in colitis-absent patients with ICI therapy compared to those who experienced colitis (16). On the contrary, the reduction of SCFAs cannot supply the cell with enough energy, resulting in an impaired gut barrier and immune system. Some species of Prevotella genus aggravate local and systemic inflammation via reduction of SCFAs and IL-18 (41), which may explain their enrichment in feces of severe diarrhea patients receiving ICI treatment for malignancy.
Dubin et al. (17) demonstrated that bacteria belonging to the Bacteroidaceae, Rikenellaceae, Barnesiellaceae family are enriched in patients resistant to IRC. Furthermore, according to the shotgun sequencing and metabolic pathway reconstruction, genetic pathways involved in vitamin B biosynthesis and polyamine transport are correlated with an absence of colitis. Vitamins are necessary micronutrients generated by plants and bacteria. The gut microbiota can metabolize vitamins for humans through its relevant enzymes and transporters (42). Vitamin B1 (thiamine) is essential in energy metabolism, especially in the tricarboxylic acid (TCA) cycles (43). Accumulating evidence proved an energy supply balance between glycolysis and the TCA cycle for immune cells. Generally, quiescent or regulatory-type cells (e.g., naive T cells, Treg cells, and M2 macrophages) use the TCA cycle for energy generation, whereas activated or pro-inflammatory cells (e.g., Th1, Th2, Th17, and M1 macrophages) rely on glycolysis (44, 45). Therefore, thiamine regulates the immune cell balance and poses a potential effect on the IRC. Vitamin B2 (riboflavin) and its active forms (flavin adenine nucleotide (FAD) and flavin mononucleotide (FMN)) are cofactors in enzymatic reactions in the Krebs cycle and fatty acid oxidation (43). The oxidation process is involved in the activation, differentiation, and proliferation of immune cells via producing acetyl-CoA for TCA cycles and energy generation, while riboflavin deficiency inhibits acyl-CoA dehydrogenase activity in the process (46). It is speculated that riboflavin modulates immune function through fatty acid oxidation. Moreover, in the presence of NADPH oxidase 2, riboflavin induces reactive oxygen species (ROS) production, which is an essential effector and signaling molecule in inflammation and immunity (47). Pantothenate, also known as vitamin B5, is a precursor of coenzyme A (CoA). Similar to thiamine and riboflavin, pantothenate has a crucial effect on immunity via cell energy consumption as coenzyme A is an indispensable cofactor for the TCA cycle and fatty acid oxidation (43). Polyamines are small cationic amines exported from bacterial cells via the spermidine and putrescine transport systems (pot A, B, C, and D). It resists inflammation partly by promoting colonic epithelial cell proliferation to maintain the epithelial barrier (48). Spermine, produced by amino acid decarboxylation, reduces colonic IL-18 levels and inhibits NLRP6 inflammasome assembly (49). It also suppressed the secretion of pro-inflammatory cytokines like TNF-α and lymphocyte function-associated antigen-1 (LFA-1), which is a regulator of immune cell adhesion and migration (50).
Conjugated linoleic acid (CLA) is a group of 18 carbon conjugated dienoic acids. It is reported to benefit local immunomodulatory activity through up-regulating anti-inflammation factors, inhibiting pro-inflammation factors, and improving the tight junctions. Some studies displayed that human commensal bacteria like Bifidobacterium possess CLA-production ability and exhibit anti-inflammation ability (51). Wall et al. (52) found that some isomers of CLA are elevated in murine fed with Bifidobacterium breve NCIMB 702258, meanwhile, some pro-inflammatory cytokines like TNF-α and IFN-γ were decreased. Another subtype of Bifidobacterium breve ameliorated mice colitis through CLA accumulation, along with advanced tight conjunction, elevated mucin and decreased IL-1 and IL-6 (53).
Cytokines are a series of small molecules mainly produced by immune cells. They modulate cell growth, differentiation, development and apoptosis, regulate immunity and contribute greatly to multiple bio-active responses including inflammation. Microorganisms induce human cell to generate considerable cytokines, which mainly consists of two types, the pro-inflammatory and anti-inflammatory cytokines. For unfavorable bacteria, they promote the level of inflammation-promotion cytokines like IL-6, TNF-α and IL-1β, exaggerating the IRC. Meanwhile, some favorable bacteria support the anti-inflammatory production like IL-10, beneficial for IRC.
IL-6 is one of the most essential and well-studied pro-inflammatory cytokines, enabling B cells to proliferate, differentiate, and secrete antibodies, and inducing a series of acute-phase reaction proteins such as C reactive protein, serum amyloid A, thrombopoietin, and complement C3. In mice models, pretreatment of ICI process enhanced the susceptibility of DSS-Induced colitis, accompanied by exacerbated hyperplasia and ulceration. It also raised inflammatory leukocyte infiltration in colonic sections, as well as the levels of inflammatory cytokines, IL-6, TNF-α, and IFN-γ in the circulation (54). Mounting evidences highlight the strong association among IL-6, bacteria and colitis. The relative abundance of Streptococcus in feces has a positive correlation with serum IL-6 level in mice models of colitis and with colonic mucosal TLR2 receptor expression in ulcerative colitis patients, respectively (55, 56). Moreover, another study manifested elevated levels of IL-6 and TNF-α in the serum of mice infected with Streptococcus via a TLR2 receptor-dependent pathway (57). Compared to control mice, Bacteroides-treated mice exhibited suppressed inflammation response and significantly lower plasma levels of pro-inflammatory cytokines, such as IL-6, IFN-γ, and TNF-α (33). Overall, it is believed that IL-6 meditates pathogenicity of bacteria on colitis and the reduction of IL-6 might contribute to the resistance to the IRC.
Apart from IL-6, another possible pathogenic mechanism of Streptococcus on IRC is TNF-α induction, meditated by primary bile acid and its receptors (58). TNF-α regulates multiple cellular responses such as vasodilation, edema formation, and leukocyte-epithelial cell adhesion. It also meditates blood coagulation and promotes oxidative stress, causing fever and inflammation indirectly (59). Conversely, the reduction of TNF-α contributes to recovery from colitis. In children with active distal ulcerative colitis, rectal infusion of Lactobacillus reuteri reduces TNF-α mucosal expression (60). The bacteria decompose dietary L-histidine to generate histamine, stimulate intracellular cAMP production through H2 receptors, inhibit TNF-α production in a PKA-MEK/ERK-MAPK-dependent pathway and relieve mucosal inflammation effectively (61).
As a key pro-inflammatory cytokine, IL-1β is engaged in various autoimmune inflammatory responses and cellular activities, including cell proliferation, differentiation, and apoptosis. It is confirmed that Prevotella aggravates the colitis via meditating the maturity of IL-1β (62). Bacteroides intestinalis was also proved to induce IRC via up-regulating IL-1β mucosal transcription (63). This cytokine activates the release of other pro-inflammatory cytokines like IL-6 and induces the differentiation of the Th17 cells. It also promotes monocytes differentiation to conventional DCs and M1-like macrophages and supports the activated B lymphocytes to proliferate and differentiate into plasma cells (64, 65). Meanwhile, the inhibition of IL-1β might contribute to the anti-inflammatory effect of Bifidobacterium breve through the interaction with TLR2 receptor and NF-κB pathway blocking (66).
As for anti-inflammation cytokines, IL-10 suppresses the expression of major histocompatibility complex II (MHC II) on the surface of monocytes, restrains its antigen presentation, impairs the activity of T lymphocytes, and prohibits the activation, migration, and adhesion of inflammatory cells. Moreover, it strongly depresses the synthesis of IL-1, IL-6, IL-8, TNF-α, granulocyte-macrophage colony-stimulating factor (GM-CSF), and granulocyte colony-stimulating factor (G-CSF) at the transcriptional level, leading an anti-inflammatory effect (67, 68). IL-10 also antagonizes the IL-17 and increases the proportion of Foxp3+ Treg cells in CD4+ T cells (69). The special cytokine contributes greatly to bacteria protection against colitis. After supplementation with Bifidobacterium breve for mice, the expression of IL-10 and IL-10Ra expanded in Treg cells in the lamina propria of the intestinal mucosa, which prevents effector T cell proliferation. However, the colitis-relieving effects of B. breve were reduced after IL-10 receptor knockout in mice, emphasizing the role of IL-10 in the anti-inflammatory effects of B. breve (70). The strain activates intestinal CD103+ DCs through the TLR2/MyD88 pathway to generate IL-10 and induce IL-10-secreting type 1 regulatory T cells in the colon, which in turn induces IL-10 and TGF-β, weakening Th1 and Th2 cells function and ameliorating the colitis (71). Other studies pointed out that F.prausnitzii A2-165 attenuates mice colitis induced by 2,4,6-trinitrobenzene sulfonic acid (TNBS) or dinitrobenzene sulfonic acid (DNBS) and modulates the T cell response via inducing IL-10 in human and murine dendritic cells (72–74). Increased IL-10 levels were also observed in mice fed with Lactobacillus reuteri, accompanied by inflammation remission and IL-17 and IL-23 reduction (54). In the future, the level of serum IL-10 may predict patients’ risk for IRC and reflect the efficacy of treatment.
Normally, immune checkpoint inhibitors raise the T cell activity against antigen presented in tumor. Sometimes, the activated immune cells target healthy tissues which have the same antigen causing inflammation like IRC. In general, the enrichment of pathogenic bacteria in IRC patient is usually accompanied with effector T cell accumulation. For those favorable strains for IRC, the immunosuppressive properties of Treg cell enable them to exert fundamental impact on anti-inflammation, partly contributing to their protection. Treg cells are necessary component of immune cells, responsible for maintaining self-tolerance and avoiding excessive immune response damage to the body. Treg cells moderate immunity partly by blocking the induction of IL-2 production in responder T cells and that both IL-10 and TGF-β are engaged in the process (75). Another mechanism of regulation is cytolysis of target cells mediated by Treg cells, which relies on granzyme A and B in human (76). Wang et al. (26) found that the supplementation of bifidobacterium mixture reduces the IRC inflammation and this effect seems to be dependent on Treg cells. Further research identified the effective specific strain, Bifidobacterium breve, and proved that the immune modulation of the strain on IRC has a close association with Treg cell energy metabolism (70). After gavage with B. breve, the circulation level of suberic acid in mice was significantly increased, reflecting the enhanced mitochondrial activity, along with elevated mitochondrial volume and stress level of Treg cells in the lamina propria. Consistent with this finding, multiple genes related to mitochondrial structural components and function were obviously upregulated (70). The relative increase in the proportion of Treg cells within the colonic mucosa was also presented in a refractory IRC patient who achieved recovery after receiving FMT therapy (77). Therefore, the relative abundance of Treg cells could be a predictor for colitis absence and a therapy target in the future.
The gut microbiota occupies a substantial place in the pathogenesis of IRC, which presents an applicable therapy through modulating its composition. Recently, probiotic supplementation has been recommended for IRC. B.breve exhibited anti-inflammatory effect in mice models, it ameliorates their immunopathological condition and rescues them from weight loss without apparent influence on anti-tumor immunity. Lactobacillus reuteri and Lactobacillus rhamnosus GG both abrogated IRC by inhibiting group 3 innate lymphoid cells (ILC3s) or regulating T cells (54, 78). FMT was introduced into the management as it manipulated the gut microbiota of recipients from donor microorganisms and small molecules like SCFAs. Recently, the therapy has been utilized on two refractory IRC patients (77). Two patients both received systemic corticosteroids, infliximab, and vedolizumab but had no settlement of symptoms. After the transfusion from an unrelated donor, they achieved marked improvements both in clinical symptoms and on endoscopic evaluation, with reduced inflammation and resolved ulcerations. Further analyses of patient's microbial composition revealed a tendency towards that of donor. The proportion of immune cells infiltrated in the colonic mucosa changed after the transplantation, such as the reduction in CD8+T cells, providing a plausible explanation of FMT treatment on ICI-related colitis. Additional cases encouraged the idea that FMT appears to be a promising option for ICI-related colitis patients resistant to corticosteroids and monoclonal antibody therapies (79, 80). Besides, a clinical trial is undergoing about FMT in treating ICI induced-diarrhea or colitis in genitourinary cancer patients (NCT04038619). However, further investigations are required to verify the efficacy and safety of FMT on ICI-related colitis, like the donor selection and transplant frequency.
Alterations and dysbiosis of gut microbiota have strong association with immune-related adverse events caused by ICIs, particularly the ICI-related colitis. Several strains have been proposed as valuable biomarkers of IRC. Studies have also suggested that microbiome dysbiosis caused by antibiotics may be an indicator of IRC. Moreover, multiple factors have been identified as involved in this pathogenesis, including metabolites, cytokines, and immune cells. Until now, there is no consensus about the exact role of one strain on IRC and different results are presented based on small sample studies. Therefore, studies with large sample and detailed mechanism are required. Regarding potential treatments, microbiota modulations such as probiotics and fecal microbiota transplantation have been explored as a promising therapy for ICI-related colitis.
FP contributed to the conception and design of the review. The first draft of the manuscript was written by GZ. GZ created all the Figures and tables. NZ, KM, and FP revised the manuscript. All authors contributed to the article and approved the submitted version.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. |
Subsets and Splits