id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
678edcb0fa469535b9e18efc | 5 | In this section, we summarize the essentials of PM-DDNN, and we introduce the new functional form for the parametrically managed activation function. Since the discussion of previously published work is just a recap, we refer the readers to ref. 28 A key aspect of the PM-DDNN method is that it treats the DPEM as a sum of two parts, namely, a many-body matrix that is learned by the neural network and a pre-fit diagonal lower-dimensional matrix, ๐ต ! "# , which, like the DPEM, is stored as a one-dimensional array of length ๐, where ๐ is the number of unique elements of the symmetric DPEM as given by |
678edcb0fa469535b9e18efc | 6 | where ๐ !"#"$! is the number of electronic states considered. In current work, ๐ !"#"$! is 14. The first ๐ !"#"$! elements of ๐ต ! "# are (up to a re-ordering of states and aside from an additive constant) the sum of the three adiabatic diatomic potentials as explained below. The remaining elements of ๐ต ! "# are zero. |
678edcb0fa469535b9e18efc | 7 | For an L-layer PM-DDNN, the input layer for geometry n is ๐ % & , the hidden layers are ๐ % ' to ๐ % ()' , the DPEM layer is ๐ % ()& , and the adiabatic potential energy layer is ๐ % ( . The vector ๐ ! $%& has ๐ components. In the current work, it is a 105-component vector; the first 14 elements of ๐ ! $%& are the diagonal elements of the DPEM, i.e., they are the diabatic potentials, and the remaining 91 elements are diabatic couplings. |
678edcb0fa469535b9e18efc | 8 | We take advantage of the nonuniqueness of the diabatic basis to make the diabatic and adiabatic potentials the same in asymptotic regions. Therefore, the first ๐ !"#"$! elements of ๐ต % 23 are the asymptotic energies of O + O2 (which are functions of the O2 internuclear distance but independent of the O-to-O2 distance), and the remaining ๐ -๐ !"#"$! elements are zero. The role of F is to decay ๐ณ % ()' to its asymptotic form, given by ๐ต % , and therefore, in asymptotic regions, ๐ % ()' ๐ ! ./ #!67,"".8 9$:.-/ JโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏโฏL ๐ต % 23 (8) One of the problems with conventional machine-learned PESs is that in regions where there is no data or insufficient data, the model (i.e., the neural network) may predict unphysical results |
678edcb0fa469535b9e18efc | 9 | (because machine learning is much better for interpolation than for extrapolation). A problem with normal machine-learned PESs arises in asymptotic regions because such regions are infinite and therefore are not necessarily improved by adding an additional (necessarily finite) amount of additional data. In PM-DDNN, we circumvent this difficulty by introducing the parametrically managed activation function, and therefore in asymptotic regions, the, and the DPEM reduces to the lower-dimensional DPEM, which is ๐ต % . In that way, as long as we have a physical a lowerdimensional DPEM, the PM-DDNN results will be physical even in asymptotic regions. |
678edcb0fa469535b9e18efc | 10 | For O3, since we want many-body part of the DPEM elements to decay to zero, ๐ D.!!,% should be defined such that when the system becomes O2 + O, in which one of the atoms is far from the other two, ๐ & 9๐ D.!!,% ;๐ ' 9๐ D.!!,% ; approaches zero. We previously 4,28,39 defined ๐ D.!!,% as the second largest of the three diatomic distances. This works because when O is dissociated from O2, two of the three O-O distances should be big. Hence, using the second largest distance is a natural choice to indicate if the system has entered the asymptotic region. However, such a choice makes ๐ D.!!,% a discontinuous function. Therefore, in the present work we have re-defined |
678edcb0fa469535b9e18efc | 11 | Both terms of eq (12) are dominated by the largest diatomic distance. When ๐ is large, eq (12) tends smoothly to the second largest diatomic distance. This motivates the use of this expression with large ๐. We found ๐ = 4 to be sufficienty large, and we use ๐ = 4 in the present work. |
678edcb0fa469535b9e18efc | 12 | Figure compares the present choice of ๐ 7466,! to the previous one. In Figure , the SBW function value is shown as a function of two diatomic distances for geometries at which the O-O-O bond angle is set to 90 degrees, and the f, a, and b parameters in eqs ( ) and ( ) are set to values of 2.0, -2.0,and 4.0 respectively for this comparison. We can observe a discontinuous derivative in panel a of Figure , where ๐ D.!!,% is defined as the second largest diatomic distance, whereas panel b, which uses eq (12), is smooth. Therefore, the definition in eq (12) is preferred to the older one. (12). |
678edcb0fa469535b9e18efc | 13 | Recall that the lower-dimensional DPEM is not the final DPEM. The final DPEM is the sum of the lower-dimensional DPEM (๐ต % ) and a many-body DPEM. The first ๐ !"#"$! elements of ๐ต % 23 are nonzero and are the same as the ๐ !"#"$! adiabatic potential energies except they might be ordered differently. For O3, we use |
678edcb0fa469535b9e18efc | 14 | Note that in a general case the pairwise terms would be different for each A, but in the present case the three diatomics are all the same, so the functions (but not their values) on the right had side are the same for all A. The same ๐ 3N(RS) 9๐ ๐ด,๐ ; is used for all fourteen states and is given elsewhere. The SR functional form is an even-tempered Gaussian function, |
678edcb0fa469535b9e18efc | 15 | where ๐(๐ % ()& ) denotes a conversion from DPEM layer to DPEM (i.e., from the unique elements of the DPEM stored as a vector to the full matrix form), and diag[๐(๐ % ()& )] denotes the diagonalization of the DPEM and saving the eigenvalues into vector ๐ % ( . |
678edcb0fa469535b9e18efc | 16 | where ๐ ,-./"! , ๐ฝ, and ๐ denote the total number of geometries in adiabatic database, diabatic restraint database, and the permutation database, ๐ผ & and ๐ผ ' are weights employed for diabatic restraint and regularization, respectively, ๐ % , and ๐ ^ denote the target adiabatic potential energies for geometry n and s respectively, and ๐ฏ [ denotes the known unique DPEM elements for geometry j in the diabatic restraint database. |
678edcb0fa469535b9e18efc | 17 | The permuted database has ๐ = ๐ _ ๐ ,-./"! geometries, where ๐ _ is the number of nonidentity permutations of the O3 molecule, which is 5. All the geometries in database ๐ are generated by permuting identical nuclei for each geometry in the adiabatic database such that the permuted geometries have the same adiabatic potential energies. |
678edcb0fa469535b9e18efc | 18 | The f, a, and b parameters in eqs ( ) and ( ) are set to values of 0.6 ร
-1 , -2.0 ร
, and 7.0 ร
respectively. The weights in cost function are set to ๐ผ & = 0.01, and ๐ผ ' = 0.0001. The input adiabatic database contains the same 6340 geometries and energies as described previously. We employed a neural network size with four hidden layers; these layers contain 30, 65, 95, and 125 neurons. |
678edcb0fa469535b9e18efc | 19 | We observed that when O approaches O2, the 9th, 10th, and 11th PES exhibit a rapid change in character when O2 bond length is in the range of 1.0~1.2 ร
which can be seen in panel a of Figure . This rapid change of electronic structure is due to interaction with higher states in this short O2 bond length region. States beyond the first 14 are not included in our adiabaticequivalent diabatic representation. With all data weighted equally in 2023 version of the surface, the NN gives unsmooth low-lying PESs; these were not uncovered in our trial dynamics calculations prior to publishing the 2023 surfaces, but they became clear in subsequent dynamics calculations with those surfaces, for which some of the trajectories are unphysical. This problem persisted in the early fts for the present improved surfaces. We finally found that we could make the potential energy surfaces of the low-lying states smoother and reduce the impact of this nonsmoothness by using higher weights for the first 8 states during training. Specifically, the first and last terms in the cost function as described in eq (14) -which account for adiabatic potential energies and permutational restraints -were modified as follows: |
678edcb0fa469535b9e18efc | 20 | The optimized surface has a mean unsigned error (MUE) of 45 meV. Table provides details of the MUEs of the entire set of 532,560 adiabatic energies, breaking them down by electronic state and energy range. The MUE is only 0.7% of the mean value of all the energies in the dataset, which is 6.24 eV. |
66acbd865101a2ffa8eaa181 | 0 | Materials informatics has served as a powerful field of study for the discovery and optimisation of functional materials with engineered properties. For machine learning applications, the representations of materials is an important subfield, where state-of-the-art performances have been achieved through graph representations that incorporate information about both composition and structure . This success has been made evident through the material science specific benchmarking suite Matbench, where graph neural networks (GNNs) dominate for tasks which incorporate structure . |
66acbd865101a2ffa8eaa181 | 1 | In the large data regime, neural network-based models have been able to learn representations of the elements. In addition to these learnt descriptors of the elements, hand-crafted representations of elements and compositions continue to play a role in materials informatics for property prediction . An often overlooked aspect when dealing with traditional element representations is the role of ions. Ions, and the knowledge of oxidation states, play a significant role in both the structural and electronic properties of materials such as electrical conductivity , chemical bonding, and magnetism as a result of their electronic configuration differing from the parent atom . For example, Fe 0 is the building block of ferromagnetic iron metal, Fe 3+ is found in the antiferromagnetic insulator Fe 2 O 3 (hematite), while a mixture of Fe 2+ and Fe 3+ is found in the ferrimagnetic crystals of Fe 3 O 4 (magnetite). |
66acbd865101a2ffa8eaa181 | 2 | For composition-only property prediction (also known as structure-agnostic learning), compositions can be represented as composition-based feature vectors (CBFV) which are derived from element embeddings. Typically, the element embeddings are combined through pooling operations (commonly descriptive statistics) to make a CBFV which can be used as an input for machine learning. This enables material informatics practitioners to make machine-learning property predictions in the absence of explicit structural information. |
66acbd865101a2ffa8eaa181 | 3 | The choice of the underlying element embedding used to make the CBFV does impact the performance of the property prediction task of interest. Depending on the size of the training data, CBFVs which lack domain knowledge (such as random representations or one-hot encoded atomic representations) can perform comparatively to CBFVs built from element properties . |
66acbd865101a2ffa8eaa181 | 4 | A common aspect of structure-agnostic approaches is that they treat element information for compositions. This raises the question of whether there is any utility for having representations or even incorporating knowledge of the oxidation states in structure-agnostic learning. When screening inorganic compositions, oxidation states are typically considered a form of chemical heuristics to achieve charge-balancing for enumerative algorithms , to suggest why one compound may form rather than another or as a measure to check the validity of compositions suggested by generative models . Often, these oxidation states don't see much further use beyond these applications in screening studies, despite their central role in the explanation of properties in inorganic chemistry. |
66acbd865101a2ffa8eaa181 | 5 | In this study, we develop high-dimensional representations of ions and assess their utility for materials informatics tasks. The SkipAtom 30 approach for developing distributed representations for the chemical elements is adapted to develop distributed representations of ionic species. We call our adapted formalism SkipSpecies. The SkipSpecies representation is then benchmarked in a structure-agnostic setting on two regression tasks (formation energy and band gap) and two classification tasks (metallic and magnetic classification) with differing dimension sizes, and different pooling operations. We find that ionic representations can perform better than standard element representations on tasks which are linked to the electronic structure of a material. A periodic table heatmap of the number of oxidation states available for each element in the dataset of 110,160 oxidation-state decorated structures obtained from the Materials Project. This figure was created using Pymatviz. implemented in Pymatgen (version: 2023.7.14) is used to query the oxidation states route of the database to obtain 154,718 non-deprecated structures. |
66acbd865101a2ffa8eaa181 | 6 | We query for the following properties: material_id, structure, formula_pretty, possible_species and method. From this query, we filter out materials which can not be assigned oxidation states by removing entries where the method field is None. The method field in this dataset has three possible values: "Bond Valence Analysis", None and "Oxidation State Guess". "Bond Valence Analysis" refers to materials where the oxidation states can be assigned using the BVAnalyzer class in Pymatgen 33 using the bond valence algorithm which uses element-based parameters based on the ICSD . "Oxidation State Guess" refers to materials where the oxidation states are assigned using the oxi_state_guesses method implemented in the Composition class of the pymatgen.core.composition submodule. This filtering returns 116,363 structures with assigned oxidation states. A further filter is applied to remove structures with noninteger oxidation states, resulting in a final dataset of 110,160 oxidation-state decorated structures. The distribution of the oxidation states of the elements is depicted in fig. . |
66acbd865101a2ffa8eaa181 | 7 | Following the approach in the original paper which introduced SkipAtom 30 , a Voronoi decomposition approach was used to convert the dataset of 110,160 oxidation statedecorated structures into graphs and then from these graphs, another dataset of the co-occurring species pairs is developed. Species which are connected in the graph representation of the structure can be considered to co-occur and make up the training pairs for learning distributed representations. |
66acbd865101a2ffa8eaa181 | 8 | The SkipSpecies approach for learning representations of ionic species is that a "fake" learning task is used to predict the species which co-occur with a target species in a given structure. This task is referred to as "fake" since the aim is not to build a classifier for predicting what species will co-occur with a target species but rather to use the learned matrix of parameters as an embedding table for the species within the dataset of materials. |
66acbd865101a2ffa8eaa181 | 9 | The general model architecture for training SkipSpecies models is adapted from SkipAtom 30 and consists of an input layer of 336 neurons, a hidden embedding layer with d neurons and an output layer with 336 neurons with a softmax activation. The main modification compared to SkipAtom is that the input and output layers contain 336 neurons instead of 86 neurons to account for each unique species in the dataset. For the hidden dimension d, models with the following dimensions are chosen: 30, 86, 100, 200, 300, 400. The input to this model is the one-hot representation of the target species and the "fake" target variable is the one-hot representation of a species that co-occurs with the target. The loss function for this model is the cross-entropy loss between the one-hot representation of the target species and the probabilities produced by the softmax activation on the output layer of the model. The distributed representations are obtained from the embedding layer. |
66acbd865101a2ffa8eaa181 | 10 | Species that are underrepresented in the dataset will receive fewer parameter updates. The resulting representations are likely to be of a lower quality than the representations for the more frequently represented species. This is also a problem within natural language processing (NLP). An NLP solution to this problem is to apply an optional post-processing technique called induction . This involves adjusting the learned representations into a more sensible area of the representation space by using the learned vectors of their most similar species. To achieve this, a quadruple tuple of periodic group table number, row number, electronegativity and oxidation state is used to represent each ionic species. The cosine similarity of this 4-dimensional vector is used to determine the nearest neighbours of the species. From this, the induced representation, รป can be define from the original representation u: |
66acbd865101a2ffa8eaa181 | 11 | To represent an N-ary ionic composition, X, which can be described as containing species x 1 , x 2 . . . x N , let's consider the quaternary case where X = (a, b, c, d). We can apply a pooling operation, f p , to obtain a vector that represents the composition: |
66acbd865101a2ffa8eaa181 | 12 | where n i and v i are the stoichiometry and the vector representation of the species, i, in the composition X, respectively. V X is the CBFV which represents the composition. Typically, the pooling operation, f p , can be a combination of summary statistics such as the mean, variance, sum, maximum and minimum of the vectors. In this study, we create the CBFVs using the mean, maximum, and sum of the element/species vectors. Max pooling takes the maximum value of each component in the vector representations: |
66acbd865101a2ffa8eaa181 | 13 | The max pooling operation returns a vector of the same dimensions as the constituent vector representations, where each component of the vector is the maximum value of that component of the m constituent vectors. Mean pooling involves taking a component-wise sum of the constituent vector representations of the composition and dividing them by the total number of atoms in the composition: |
66acbd865101a2ffa8eaa181 | 14 | To evaluate the performance of the distributed representations of SkipAtom and SkipSpecies, the ElemNet 38 architecture (as chosen in the original SkipAtom paper ) with the CBFVs derived from the learnt representations as inputs is applied to two classification and two regression tasks: metallic and magnetic classification, and formation energy per atom and band gap, respectively. This work focuses on observing the difference in performance of distributed representations for elements versus atoms as SkipAtom has been shown to give superior performance to one-hot, random and Mat2Vec 39 element representations As this dataset contained polymorphs of the same composition, it was further filtered by only keeping the composition of a polymorph with the lowest energy above the convex hull. This reduced the dataset to 71,470 materials. The property datasets are depicted in Figure . |
66acbd865101a2ffa8eaa181 | 15 | The ElemNet model is implemented in TensorFlow and is a 17-layer feed-forward neural network that consists of 4 1024 neuron layers, 3ร512 neuron layers, 3ร256 neuron layers, 3ร128 neuron layers, 2ร64 neuron layers, and 1ร32 neuron layer. All layers use ReLU activation. For the classification tasks, the output layer is a single neuron layer with sigmoid activation, and the loss function is the binary cross-entropy. The regression task uses an output layer with linear activation and the loss function is the Mean Absolute Error (MAE). The models were trained using the following hyperparameters: a maximum number of epochs of 100, a learning rate of 10 -4 , a batch size of 32, and an L2 lambda of 10 -5 . Twice-repeated 5-fold cross-validation is performed to evaluate the performance of the compound representations on the property prediction tasks. The reported metric on each task is the average MAE and the average AUC (Area under the Receiver-Operator characteristic Curve) for the regression and classification tasks, respectively. The error in the metrics is the standard deviation across the twice-repeated 5-fold cross-validation. |
66acbd865101a2ffa8eaa181 | 16 | Figure shows the distribution of unique components (elements/species) per structure in the dataset that has been curated for training the SkipSpecies vectors as described in Section II B. In Figure and b, the one-component materials come from polymorphs of H 2 and N 2 , which from the oxidation states route of the Materials Project , are automatically assigned oxidation states of 0+. From Figure , the mode number of unique elements in the dataset is three reflecting that ternary materials are the most frequent in this dataset though this is very closely followed by quaternary materials. This is reversed in Figure when we consider the unique species within the structures, where the most frequent number becomes four. This distribution shift arises from the presence of many mixed-valence compounds within this dataset. Further evidence of this is shown in that the maximum number of components when we consider species becomes ten, versus the case when we only consider the elements and would thus have at most nine component compounds. |
66acbd865101a2ffa8eaa181 | 17 | Figure further shows the breakdown between the number of components that each structure has when we consider either unique species or elements. With the exception of unary and nonary materials, mixed valency is present in all the materials. What is further illustrated is that the materials in this dataset can have one or more elements where mixed valency is observed. An example of this is the elemental quaternary material Li 10 GeP 2 S 12 , a known lithium superionic conductor , where depending on the crystal structure, it can be an ionic quinary material (mp-696128) where S exhibits mixed valency as S -and S 2 -or an ionic senary material (mp-696138) where P and S both exhibit mixed valency as and P 4+ and P 5+ , and S -and S 2 -, respectively. |
66acbd865101a2ffa8eaa181 | 18 | Various techniques exist to reduce high-dimensional data to make visualisation easier for us to understand. In Figure , we have chosen the uniform manifold approximation and projection (UMAP) and the t-stochastic neighbours embedding (t-SNE) . Visualising the embeddings can provide a qualitative understanding of the quality of the learned representations. It is apparent that we can recover expected chemical trends but also find patterns that we would not intuitively ex- pect. For example, in Figure , we find that the halides with the exception of the chloride ion appear close to each other within this space with Br -and I -close to each other than with F -. All the halides cluster together when t-SNE is used for the dimension reduction as shown in Figure (the iodide and bromide anions overlap each other in the reduced space). |
66acbd865101a2ffa8eaa181 | 19 | The anions can generally be separated from the cations in the reduced space, though in the t-SNE figure, they form a cluster with a few outliers including Te 2+ , C 2 -and C 4 -. For the UMAP figure, we can still observe a cluster though it is more spread out compared to the t-SNE space. |
66acbd865101a2ffa8eaa181 | 20 | In Figure , the error metrics of the property prediction tasks using the induced SkipSpecies representation are depicted over a range of dimensions of the representation, with different curves representing the different choices of pooling to make the CBFV. Independent of the task, the dimension, and the representation, it is evident that creating a CBFV using max-pooling leads to worse performance for the property prediction tasks. This result likely occurs as not all the information that is present from the constituent species vectors is used in the max pooling operation, whereas sum and mean pooling do use information from all the constituent species vectors. It is possible that the max-pooling operation can, in some cases, neglect the element or species that is most important within a particular composition for the prediction of particular properties. To further expand on this rationale, the max-pooling approach is very sensitive to outliers, as species vectors that may have anomalously high components can dominate the components of the CBFV. Mean-pooling would be less sensitive to outliers due to taking the average over each component, making it more robust to outliers. The resulting max-pooled CBFV as such may not fully describe a composition. |
66acbd865101a2ffa8eaa181 | 21 | The mean and sum pooled CBFVs have comparable performances with each other. This can be rationalised as both pooling operations aggregate information from the constituent vectors that make up the CBFV. This aggregation of info preserves information about the species that make up the composition. Mathematically, the sum and mean-pooled CBFVs of a composition X are related as follows: |
66acbd865101a2ffa8eaa181 | 22 | For creating SkipSpecies representations of the chemical species, there is a choice of dimensions for the resulting distributed representations. The effect of the choice in the dimensionality is depicted in Figure . It can be observed that, generally, as the number of dimensions increases, the performance of the models also increases. This trend is more dramatic for the max-pooled CBFVs. For both the sum-pooled and mean-pooled CBFVs, there are marginal increases in performance beyond 200 dimensions. |
66acbd865101a2ffa8eaa181 | 23 | Within the NLP field for training word embeddings, an arbitrary dimension is often chosen. For the word embeddings, setting a higher number of dimensions usually results in higher-quality embeddings up until a saturation point . We have observed this from our property prediction tasks. The performance tends to improve with dimension size because a higher number of dimensions can capture more complex relationships between the co-occurring pairs. However, the effectiveness of increasing dimensions is ultimately constrained by the available data size, which limits the ability to learn meaningful patterns. |
66acbd865101a2ffa8eaa181 | 24 | To better visualise the effect of the choice of representation on the performance of the four property prediction tasks, heatmaps of the sum-pooled representations have been depicted in Figure . For the band gap prediction task, the SkipSpecies representations perform better than the induced SkipAtom representation across all dimensions. While the choice to apply induction to the SkipSpecies representation does appear to offer a slight improvement to the MAE, the error in these values makes it hard to discern if the application of induction is significant in upgrading the performance of the SkipSpecies representation. For the formation energy per atom task, the induced SkipAtom representation outperforms both SkipSpecies representations across all the dimensions. For each of the representations, there is a marginal improvement in the MAE beyond 200 dimensions, if any. The induced SkipSpecies representation performs the best on both the metallic and magnetic classification tasks with the SkipAtom representation performing the worst. |
66acbd865101a2ffa8eaa181 | 25 | To further highlight the effect of the representation on the model performance, we have shown a plot of how the validation metric (MAE for the regression tasks and AUC for the classification tasks) changes during training in Figure . Except for the formation energy task shown in Figure , the element representation SkipAtom performs the worst compared to the ionic SkipSpecies representations. For the other three tasks, the SkipSpecies representation achieves better results from the start of the training process, and this is maintained throughout the 100 epochs. |
66acbd865101a2ffa8eaa181 | 26 | For the band gap task, an ionic representation may offer better performance than a representation based on the neutral atom due to the knowledge of the oxidation states. This can be rationalised by considering that the oxidation state of an ion allows the model to distinguish between the properties of different materials containing the same element. The loss or gain of electrons affects both the effective radius of a species, impacting its local structural environment and electronic configuration, both of which can alter the band gap. |
66acbd865101a2ffa8eaa181 | 27 | As the embeddings are learned such that species which occur within similar environments should be similar, ionic representations may provide the flexibility to describe different types of compounds containing the same element. For example, TiO 2 containing Ti 4+ has a wide band gap above 3 eV, while Ti 2 O 3 containing Ti 3+ has a small band gap closer to 0 eV. This flexibility may not be captured by atom-only representations. |
66acbd865101a2ffa8eaa181 | 28 | The metallic classification task can be rationalised by an important caveat that this task is based on Materials Project 31 data. This classification is based on the band gap calculated with semi-local density functional theory, so it is possible that some of the compounds in this dataset that are labelled as metals could in fact be semiconductors due to a band gap underestimation at this level of theory . Additionally, it is likely that many known metallic compounds are likely to have been excluded from the construction of the dataset. This occurred due to the need to assign oxidation states to materials, which typically leads to the exclusion of many intermetallic com- pounds. |
66acbd865101a2ffa8eaa181 | 29 | The magnetic classification task is less common for property prediction tasks as noted by its absence from MatBench . The boost in performance from considering ionic representations possibly arises from how magnetism arises from the electronic structure of a material through the spins of unpaired electrons centred on atomic sites. The oxidation states of the ions implicitly encode whether particular ions may possess paired or unpaired electrons depending on the crystal environments and species that they co-occurred within during the original training of the distributed representations which could explain the difference in performance between the SkipSpecies and SkipAtom representations. |
66acbd865101a2ffa8eaa181 | 30 | It is important to note that the performances shown will be influenced both by the quality of the representation and the chosen model architecture. Factors that can affect the quality of the representations include how often a particular species is represented in the dataset. In this work, we applied induction to the SkipSpecies representation to compensate for the underrepresented species. The induction appears to offer a small boost in performance based on the error met-rics. |
66acbd865101a2ffa8eaa181 | 31 | The SkipSpecies ionic representations can be used to develop property prediction models with lower errors than comparable atomic representations for predicting properties such as the band gap or classifying compositions as metals and nonmetals, and magnetic or non-magnetic, suggesting that there may be some utility for ionic representations for predicting the properties of compositions. One caveat is that ionic representations are more restrictive compared to element representations, as the oxidation states in the composition have to be known to use them. While tools such as Pymatgen tions, this does introduce an additional step into a workflow to predict properties. These ionic representations may also find use for property prediction in approaches that create or generate compositions alongside knowledge of the oxidation states of the constituent elements, rather than having to decorate an already existing set of compositions where this information is not already known. One example where these ionic representations could be used is within SMACT-based workflows, as the chemical filters used to generate compositional spaces also return both the constituent elements and the oxidation states of the compositions. These representations can be used for both property prediction on these spaces, as well as providing an alternative means to visualise the compositional space as opposed to using elemental representations to visualise this space. |
66acbd865101a2ffa8eaa181 | 32 | Distributed species representations may have applications for crystal structure assignment by analogy through ionic substitutions, as pairwise similarity values can be derived from the vectors using distance or similarity measures. Alternatively, similarity measures can be applied to compositional feature vectors derived from these representations to suggest what known materials are similar to hypothetical compositions as part of synthesisability models. Finally, we note that such representations are not limited to compositional (structure free) models, but could be used, for example, to initialise node vectors on graph-based models of materials structure and properties. |
67b374426dde43c90849f1c1 | 0 | Cancer is the cause of nearly 10 million deaths per year, making it one of humanity's most significant disease burdens, particularly in western societies. It arises from a variety of factors, including genetic mutations and epigenetic alterations, the latter involving chromatin modifications without changes to the DNA sequence. Chromatin remodeling, like ฮต-lysine acetylation of histones, is tightly regulated, as it is responsible for modulation of transcription, DNA repair, replication, and condensation. To reverse the acetylation, histone deacetylases (HDACs) hydrolyze the amide bond, restoring the positive charge of the lysine and resulting in a more compact chromatin. HDACs can be distinguished into four classes: Class I (HDAC1-3 and 8), class IIa (HDAC4, 5, 7 and 9), class IIb (HDAC6 and 10) and class IV (HDAC11) are zinc-dependent deacylases. Depending on the isoform, they are primarily localized in the nucleus or in the cytoplasm and some contribute to multi-protein complexes. In addition, HDACs act on multiple substrates beyond histones and are involved in the hydrolysis of more than just acetyl groups. The up-regulation and high expression of different HDAC isoforms are associated with poor prognosis in cancers such as multiple myeloma and acute myeloid leukemia. Therefore, HDAC inhibition is a promising strategy for tumor therapy. Numerous studies have demonstrated that HDAC inhibitors (HDACi) reduce angiogenesis, cell migration, proliferation, and resistance to chemotherapy. Furthermore, inhibition of HDACs promotes apoptosis and enhances cell differentiation. U.S. Food and Drug Administration (FDA) or European Medicines Agency (EMA) approved HDACi, such as vorinostat, belinostat, and romidepsin, for the treatment of T-cell lymphoma, panobinostat for the treatment of multiple myeloma, and givinostat for treating Duchenne muscular dystrophy. |
67b374426dde43c90849f1c1 | 1 | In this approach, an E3 ligase is hijacked by a molecular glue or PROTAC to polyubiquitinate a protein of interest (POI), leading to its degradation. In Cullin RING E3 ligases, the largest group of E3 ligases, the E3 ligase complex works alongside an ubiquitin-loaded E2 enzyme, which transfers ubiquitin to the substrate or POI. This leads to subsequent degradation by the ubiquitinproteasome system (UPS). This catalytic mode of action provides significant advantages, such as extended pharmacological effects, doses reduction, and potentially minimizing adverse effects. Importantly, degraders can overcome cancer resistance mechanisms, such as target amplification or overexpression, through their catalytic activity. Additionally, they can counteract resistance caused by altered ligand binding sites, as even weak binders can still enable efficient degradation. While molecular glues identification is often serendipitous, PROTACs can be designed more rationally. However, due to recent discoveries in degrader development it becomes more and more difficult to differentiate between molecular glues and PROTACs. This becomes particularly evident in a series of linker-less PROTACs that combine a POI-warhead with a covalent handle for a specific E3 ligase. These compounds, termed monovalent molecular glues or linker -less PROTACs, challenge traditional classifications. PROTACs for targeted degradation of HDACs were introduced in 2018, and since then, over 100 HDAC PROTACs have been published. Despite the use of many different HDAC ligands, only three E3 ligases have been employed according to PROTAC-DB 3.0: Cereblon (CRBN), Von Hippel-Lindau (VHL), and the inhibitor of apoptosis protein (IAP). The first PROTAC for targeted degradation of an HDAC was the crebinostat-derived degrader I (Figure ). It used pomalidomide for CRBN recruitment and despite PROTAC I contains a pan-HDAC ligand, only HDAC6 was degraded. PROTAC II and IV are representatives of the first successful recruitment of VHL and IAP, respectively. PROTAC III contains a benzamide for targeting class I HDACs, and the switch to a CRBN recruiter shows similar degradation selectivity as II, but to a lesser extent. The E3 ligase switch from IV to V yielded a different result: Both contain the pan-HDAC inhibitor dacinostat and a polyethylene glycol (PEG) spacer, but the IAP-recruiting IV resulted in HDAC6 degradation, while V degraded HDAC3 and 8 by VHL recruitment. Figure . Comparison of HDAC PROTACs recruiting different E3 ligases. PROTAC I was the first utilization of CRBN for HDAC degradation, while PROTAC II represents the first VHL-based HDAC degrader. PROTAC III showed similar degradation selectivity as PROTAC II, but to a lesser extent by switching back to a CRBN recruiter. The first successful utilization of IAP for targeted HDAC degradation was PROTAC IV and changing the E3 ligase to VHL, represented by PROTAC V , resulted in a switch of degraded isoforms. |
67b374426dde43c90849f1c1 | 2 | With more than 600 E3 ligases encoded in the human genome, there is a significant need to expand the number of utilized E3 ligases for HDAC degradation. Furthermore, unlocking specific E3 ligases with dedicated tissue or cell-type specificity holds great potential for achieving degradation in the desired tissue. Likewise, targeting tumor-or disease-enriched E3 ligase could enable the development of more refined and selective degraders. Expanding the use of new E3 ligases can also help to target new tumor entities; for example, it has been shown that the sensitivity pattern of tested tumor cell lines to PROTACs changes by switching from CRBN to VHL. However, ligands for both E3s come with some limitations, as thalidomide is associated with teratogenicity and stability issues, while the VHL ligands increase molecular weight and topological polar surface area, which can be challenging for oral bioavailability. Using E3 ligases which degrade tumor suppressor-proteins can have additional anticancer effects. For example, mouse double minute 2 homolog (MDM2) recruitment for TPD leads to a stabilization of p53 and p21 is substrate of the E3 ligase DCAF11. Lately, the range of E3 ligases used in TPD has expanded. For example, ligands for Fem-1 homolog B (FEM1B), Ring Finger Protein 4 (RNF4), Ring Finger Protein 114 (RNF114), DDB1-and CUL4-associated factor 11 (DCAF11), and DDB1-and CUL4-associated factor 16 (DCAF16) have been utilized in TPD strategies. These ligands share an electrophilic warhead to engage the E3 ligase, resulting in a pseudo-binary complex of E3 ligase, covalently bound PROTAC, and POI. This results in simpler kinetics, as the covalent construct of E3 ligase and PROTAC only needs to recruit a new POI molecule for polyubiquitination. |
67b374426dde43c90849f1c1 | 3 | Based on our previously published HDAC6 degrader A6, we started to design DCAF11-recruiting PROTACs (Figure ). In our previous studies, we developed a highly modular approach for the solid-phase synthesis of PROTACs and extended this approach to the synthesis of hydrophobically tagged HDAC inhibitors. Adapting the U-4CR to the solid-phase conditions allowed us to synthesize a set of eleven PROTACs entirely on resin. The subsequent biological evaluation revealed significant degradation of HDAC1, correlating with cytotoxic effects in the multiple myeloma cell line MM.1S. Notably, further investigation of compounds 1j and 2, which share the POI ligand with A6 but differ in linker length and the E3 ligase ligand, revealed enhanced degradation capabilities and both compounds induced pan-HDAC degradation across all isoforms tested. This effect was accompanied by increased cell cycle arrest, apoptosis induction, and longterm antiproliferative activity. |
67b374426dde43c90849f1c1 | 4 | Design and synthesis of DCAF11-recruiting HDAC PROTACs. The design of our PROTACs was based on the best degrader of our previous HDAC degrader study, A6, which combines a vorinostat-like HDAC inhibitor and a thalidomide-based CRBN ligand, fused by a C7 aliphatic spacer (Figure ). Although vorinostat is a pan-HDAC inhibitor, A6 specifically targeted HDAC6 for degradation. The aim of this study was to investigate how switching the E3 ligase affects the HDAC degradation selectivity profile. |
67b374426dde43c90849f1c1 | 5 | To convert the CRBN-utilizing PROTAC into one that recruits DCAF11, we replaced the thalidomide-based ligand (red, A6) with a DCAF11 ligand (orange, 1i, Figure ). We retained the vorinostat-like pan-HDACi ligand and C7 linker, to trace back any possible effects to the exchange of the E3 ligase recruiter. The initial PROTAC 1i, which represents the direct DCAF11-recruiting analog of A6, was complemented by a set of nine degraders, bearing a broad range of different spacers to introduce diversity in regards to chain length, lipophilicity, and rigidity. In detail, two short and rigid cyclic spacers (1a,b, Figure ), three more flexible PEG spacers (1c-e), and some alkyl spacers ranging from C1-C11 (1f-j) were selected. As mentioned before, the line between PROTACs and molecular glues is becoming increasingly blurred, we also designed a compound that can be considered as a spacer-less PROTAC (2). In total, we decided to synthesize eleven PROTACs to study the effect of DCAF11 recruitment on HDAC degradation. |
67b374426dde43c90849f1c1 | 6 | The synthesis of the DCAF11-recruiting PROTACs was completely carried out on solid support and is shown in Scheme 1. For resin modification, loading determination, and amide coupling reactions, our previously published protocols were used. Specifically, the commercially available 2-chlorotrityl chloride (2-CTC) resin was modified with N-hydroxyphthalimide to immobilize the hydroxylamine as the precursor for the hydroxamic acid on the resin (not shown). |
67b374426dde43c90849f1c1 | 7 | The subsequent hydrazine monohydrate-mediated deprotection released the resin-bound hydroxylamine, enabling its coupling with Fmoc-7-aminoheptanoic acid which was facilitated by HATU and HOBt. This step formed the zinc-binding group and linker of the resin-bound HDAC inhibitor 3. After Fmoc-deprotection of the precursor, Fmoc-4-aminobenzoic acid was used to complete the HDACi 4. The last step before the introduction of the DCAF11 ligand was the attachment of various spacers, which was performed in the same manner as previous couplings steps to produce 5a-j. The precursors 4 and 5a-j were then used to install the DCAF11 ligand by an on resin U-4CR. To this end, 3,4-dichlorobenzaldehyde, 2-chloroacetic acid, and benzyl isocyanide were used in the U-4CR. This synthetic approach enabled the introduction of the DCAF11 ligand in one final step. Finally, all compounds were cleaved from the resin and purified by preparative HPLC to >95% purity. .02 and 4.9 (log2) transcription per million). We first examined the effects on HDAC1 and 6 protein levels, as they represent members of the most studied HDAC classes (I and IIb) and are mainly located in opposite cellular compartments (nucleus and cytoplasm). |
67b374426dde43c90849f1c1 | 8 | Figure summarizes the analyzed HDAC1 and 6 levels: HDAC1 was not degraded by PROTACs bearing the cyclic spacers, but the longer and more polar PEG spacer-containing PROTACs showed a significant reduction of up to 51% (1e). The alkyl spacer-based compounds revealed some unexpected results: while the C5 spacer (compound 1h) did not reduce HDAC1 levels, PROTACs with both shorter and longer spacers showed a trend toward significant HDAC1 degradation. Interestingly, also the spacer-less degrader 2 demonstrated significant HDAC1 degradation. The most pronounced reduction of HDAC1 levels was detected for spacer-less PROTAC 2 (55%) and the longest C11 spacer bearing PROTAC 1j (71%). In addition, these two PROTACs were also capable of achieving significant reduction of HDAC6 protein levels, whereas all the other compounds showed no significant effects on HDAC6 protein abundance. Encouraged by the promising reductions in HDAC1 and HDAC6 levels, we further investigated the phenotypic effects of our DCAF11-recruiting PROTACs on cancer cells. Accordingly, we performed a cell viability assay on multiple myeloma MM.1S cells after treatment with the indicated compounds. The results are presented in Table . The rigid cyclic spacer-bearing PROTACs (1a, 1b) demonstrated diminished antiproliferative activity, whereas the spacer-less (2: EC50 = 1.46 ยตM) and the longest alkyl spacer (1j: EC50 = 2.79 ยตM) yielded pronounced effects. |
67b374426dde43c90849f1c1 | 9 | These findings correlate well with the observed degradation of HDAC1 and 6. While 1a and 1b had no effect on HDAC1 and 6 levels, 2 and 1j demonstrated significant degradation of both HDAC1 and 6. In addition, 2 and 1j displayed superior antiproliferative activity in MM.1S cells compared to ricolinostat, an HDACi currently under clinical investigation. Furthermore, we investigated the contribution of the DCAF11 ligand to the antiproliferative activity of the PROTACs. To this end, we synthesized the n-propyl-substituted DCAF11 warhead 6 (see Scheme S1, Supporting Information) and tested a 1:1 combination of vorinostat and 6 in a cell viability assay. Since the EC50 of the combination treatment was similar to that of vorinostat alone, the DCAF11 ligand seems not to add significant cytotoxicity (Figure , Supporting Information). In contrast, the thalidomide-based PROTAC A6 showed no effect on the viability of MM.1S cells, consistent with the previous reports showing little to no effects of A6 depending on the cell line. Notably, the antiproliferative activity of 2 and 1j is reduced by a factor of ~ 3 -6 compared to the parent HDACi vorinostat, which is comparable to the previously published FEM1B-recruiting |
67b374426dde43c90849f1c1 | 10 | Target engagement and HDAC degradation selectivity profile of 2 and 1j. For further investigations, we selected the two most effective HDAC1 degraders from this set (2 and 1j), as they also demonstrated the highest antiproliferative activity. As target engagement is crucial for targeted protein degradation, we next investigated the HDAC inhibition efficiency of the PROTAC hits. To distinguish between their effects on the various isoforms, we evaluated the inhibitory activity on representative isoforms from different HDAC classes, namely HDAC1, 2, 4 and 6 (Table ). Compound 2 demonstrated comparable inhibition results as A6. However, 1j, the most potent degrader, exhibited comparatively weaker inhibition of all tested isoforms. The half-maximal inhibitory concentration (IC50) is consistently ~4-8 times higher than that of 2. |
67b374426dde43c90849f1c1 | 11 | In the next step, we investigated whether this target engagement translates into degradation of multiple HDAC isoforms. HDAC1, 2, 4, and 6 levels were determined by immunoblot analysis of MM.1S cell lysates and the CRBN-recruiting PROTAC A6 was used as control. The immunoblot analysis confirmed the automated Simple Westernโข immunoassay results, as 1j showed most pronounced degradation of all tested HDAC isoforms (Figure ). In detail, the strongest maximal degradation (Dmax) was observed for HDAC1 (90%) by 1j, while the degradation of HDAC2, 4, and 6 ranged from 71-76%. The Dmax of compound 2, although slightly less efficient, was also the strongest for HDAC1 (74%) followed by HDAC2 (54%), while HDAC4 (40%) and HDAC6 (26%) were also degraded, but to a lesser extent (see Table , Supporting Information). In contrast, A6 |
67b374426dde43c90849f1c1 | 12 | selectively degraded HDAC6 without significant effects on the other HDAC isoforms, as previously reported. By utilizing DCAF11 instead of CRBN for HDAC degradation, 1j and 2 are able to broaden the scope of degraded HDAC isoforms, degrading not only HDAC6 but also HDAC1, 2, 4 and 6. In addition, the degradation pattern of 1j was further investigated by its time dependency. This analysis revealed less differences between the DCAF11-recruiting PROTACs and the HDACi. Synthesis and evaluation of the non-degrading control 1j-nc. In the next step, we wanted to design and synthesize a non-degrading control for the most effective PROTAC 1j. It is essential that the non-degrading control is as comparable to the PROTAC as possible, without binding to the E3 ligase. By the absence of degradation, the non-degrading control proves TPD, which requires the E3 ligase for the reduction of protein levels. In addition, it enables to distinguish between inhibitor and degrader effects. In case of 1j the non-degrading control was designed by substituting the electrophilic warhead of the DCAF11 ligand. The replacement of 2-chloroacetic acid by propionic acid led to no successful U-4CR, because of reduced nucleophilicity of the propionic acid. We successfully obtained the U-4CR product bound on resin by switching to formic acid. After cleavage and purification, as described before, 1j-nc was prepared in 38% yield over seven steps (Scheme 2). |
67b374426dde43c90849f1c1 | 13 | The non-degrading control was then used to treat MM.1S cells in direct comparison to the related PROTAC 1j. In comparison to 1j, the non-degrading control 1j-nc was not able to affect the protein levels of HDAC1 and 6 (Figure ), proving that covalent binding to DCAF11 is crucial for degradation. showed first promising effects. The phenotypic cell viability screening demonstrated that both PROTACs exhibited cytotoxicity against all tested cell lines (Table ). However, the effect was diminished in the solid cancer cell lines, particularly for U-87MG, in comparison to the multiple myeloma cell line MM.1S. As anticipated, the non-degrading control 1j-nc demonstrated no activity in MDA-MB-231, indicating that degradation is the underlying cause of cytotoxicity. |
67b374426dde43c90849f1c1 | 14 | the CRBN ligand with a DCAF11 recruiter shifted selectivity from HDAC6-specific to pan-HDAC degradation. This change in the selectivity profile creates new opportunities to investigate and compare HDAC isoform-specific or general effects. Importantly, PROTACs affect not only the enzymatic but also the non-enzymatic functions of their targets. While HDAC6 selective degraders allow for a more detailed assessment of HDAC6 function in pathology and physiology, a pan-HDAC degrader may provide a more general picture of HDACs functions across isoforms and may also reveal isoform synergies in terms of phenotypic effects such as anticancer activity. In addition, both approaches may have potential clinical applications. While pan-HDAC degraders could be developed to treat multiple cancers , selective HDAC6 PROTACs, like selective HDAC6 inhibitors , could be used in neurodegenerative diseases, inflammatory diseases, and some cancers. In the latter case, it would be important to replace the non-selective HDAC inhibitor warhead in A6 with a selective HDAC6 inhibitor scaffold to avoid class I effects resulting from HDAC inhibition. |
63f7c0af32cd591f126166a1 | 0 | SARS-CoV-2 Spike is heavily modified by N-glycosylation (Figure ), and the glycan attachment sites are highly conserved across the different variants of the virus . The abundance of glycans on SARS-CoV-2 Spike, much like the structurally similar HIV-1 Env cell entry protein machine, has inspired the search for possible SARS-CoV-2 therapeutics among lectins previously shown to be effective as HIV-1 infection inhibitors. For example, the mannose-specific lectin griffithsin (GRFT), a known potent HIV-1 inhibitor, has shown antiviral activity against SARS-CoV-2 pseudoviruses and live viruses . |
63f7c0af32cd591f126166a1 | 1 | Here, we sought to deepen our understanding of molecular factors underpinning lectin antiviral function. Using a pseudovirus platform, we found that CV-N and to a somewhat lesser extent GRFT were potent as infection inhibitors, while microvirin (MVN) was essentially inactive. Furthermore, using washout experiments, CV-N and GRFT were shown to irreversibly inactivate SARS-CoV-2 pseudoviruses in a dose-dependent manner. Functional assays with single glycan attachment site SARS-CoV-2 pseudovirus mutants resulted in the identification of two glycan clusters in the S1 subunit of the Spike important for CV-N inactivation of pseudovirus. Analogous clusters were found to be important for GRFT antiviral activity. The results of this work argue that multivalent engagement of glycans in S1 oligosaccharides disrupts spike conformation, causing the inactivation of SARS-CoV-2. Overall, this work helps build the foundation for defining the conformational vulnerability of the SARS-CoV-2 Spike, the capacity of lectins to hijack this vulnerability, and the potential to develop lectin inactivators for possible COVID-19 treatment. |
63f7c0af32cd591f126166a1 | 2 | Recombinant SARS-CoV-2 pseudoviruses were produced by co-transfecting HEK-293T cells with 8.5 ฮผg of pNL4-3ฮEnv-NanoLuc and 2.5 ฮผg of pSARS-CoV-2-Sฮ19 envelope spike plasmid as previously described . PEI (25 kDa linear, Polysciences, Inc., Warrington, PA) was the transfection reagent. At least 8 hours after transfection, the transfected HEK-293T cells were washed twice with DPBS (Thermo Fisher Scientific, Inc, Waltham, MA) and incubated with fresh media. After 48 hours of transfection, the supernatant was harvested and filtered using a 0.45 ฮผm filter. The filtered supernatant was further purified using a 6-20% iodixanol gradient and centrifuged in an Optima LE-80k Ultracentrifuge with SW41 rotor (Beckman Coulter, Brea, CA) at 20,000 rpm for 2 hours at 4ยฐC. Purified pseudoviruses (lower, heavier fractions) were collected and pooled together for infectivity testing and functional assays, including quantification of pseudovirus production by p24 content detection by ELISA. |
63f7c0af32cd591f126166a1 | 3 | HEK-293T/ACE2*cl.13 target cells were seeded at 10,000 cells per well in 96-well flat-bottom plates and cultured overnight. The next day, SARS-CoV-2 pseudoviruses (50 ng of p24 worth of pseudovirus) were mixed with serial dilutions of lectins (or buffer control), incubated at 37ยฐC for 30 minutes, and then added to the target cells. At a minimum of 12 hours after infection, the mixture was removed, and the cells were washed with 200 ฮผL of 1x PBS before a media change. On day 4 of the assay, the medium was carefully removed, and the cells were washed twice with 100 ฮผL of 1x PBS and treated with 50 ฮผL of 1x lysis buffer (Promega, Madison, WI). After 3 freeze-thaw cycles, the plates were shaken at 300 rpm at room temperature for 5 minutes. Since the pseudoviruses encode a NanoLuc luciferase gene, the percentage of infection by the pseudoviruses in the inhibitor's presence was measured in a luciferase reporter assay using a 25 ฮผL mix of NanoGlo Substrate and NanoGlo buffer (Promega, Madison, WI) in 1:50 ratio, per manufacturer's directions. Briefly, 25 ฮผL of the lysed cells and 25 ฮผL of the substrate were combined in a white bottom 96 well plate and incubated in the dark for 5 minutes before reading it in a PerkinElmer 1450 Microbeta Liquid Scintillation and Luminescence Counter (Waltham, MA). Non-specific lectin toxicity was not observed with the combination of CV-N and HEK-293T/ACE2*cl.13 target cells in this assay configuration (Supplemental Figure ). |
63f7c0af32cd591f126166a1 | 4 | The lectin's ability to cause SARS-CoV-2 pseudoviruses' membrane disruption was measured in a cell-free sandwich ELISA assay, where the amount of p24 released by the viruses was quantified . High binding polystyrene ELISA plates (Corning Inc., Corning, NY) were coated with mouse anti-p24 (ab63958, Abcam, Cambridge, UK) and blocked with a solution of 3% Fraction V BSA (Research Products International, Mount Prospect, IL). A serial dilution of inhibitors being tested was mixed with pseudoviruses (50 ng of p24 worth of pseudovirus) and incubated for 2 hours at 37ยฐC before spinning for 2 hours at 15,000 ร g` supernatants were collected and tested for p24 content. The amount of p24 present on the plate was quantified using by 1 hour incubation at room temperature with a primary rabbit anti-p24 antibody (ab63913, Abcam), plus a 1 hour incubation in secondary anti-rabbit IgG conjugated to horseradish peroxidase (HRP, ab205718, Abcam). HRP conjugate binding activity was determined by incubating with o-phenylenediamine dihydrochloride (OPD, p23938, Sigma-Aldrich, St. Louis, MO) for 30 minutes before reading the signal at 450nm using a transmission plate reader (Infinite F50, Tecan Group Ltd., Mรคnnedorf, Switzerland). The signals from all the functional assays were plotted as a function of p24 released where a lysed virus control treated with 1% triton x-100 was used as the 100% value. No virolysis of control viruses was detected in this assay configuration (Supplemental Figure ). |
63f7c0af32cd591f126166a1 | 5 | The extent of the irreversible inactivation of SARS-CoV-2 by CV-N after washout was determined using the previously described infection inhibition assay. HEK-293T/ACE2*cl.13 cells were seeded at 10,000 cells per well in a 96 well plate and incubated overnight. "CV-Ntreated" SARS-CoV-2 pseudovirus samples were created by mixing produced SARS-CoV pseudoviruses (50 ng of p24 worth of pseudovirus) and CV-N at various concentrations, incubating at 37ยฐC for 2 hours, then washing and resuspending the virus 10 times in fresh media 100kDa spin concentrators to remove unbound CV-N. "CV-N-treated" SARS-CoV-2 pseudovirus samples were then added to the HEK-293T/ACE2*cl.13 cells for 24 hour incubation, followed by a media change, and then another 24 hour incubation. Infection was then measured via the NanoLuc reporter as described in the Infection Inhibition Assay method. Wash flowthroughs and recovered viruses (both SARS-CoV-2 pseudoviruses and Spikenegative virus-like particles) were tested for removed and virus-retained CV-N by gel filtration and Western Blot. |
63f7c0af32cd591f126166a1 | 6 | To determine the ability of CV-N to cause shedding of the S protein complex from SARS-CoV-2 spike trimers upon treatment, SARS-CoV-2 WT pseudoviruses (50 ng of p24 worth of pseudovirus) were treated with various concentrations of CV-N and ACE2 (human recombinant ACE2 protein, 10108-H08H, Sino Biological US Inc., Wayne, PA,) for 24 hours at 37ยฐC, before washout of lectins and ACE2 using a similar protocol to that described in the washout experiment. Untreated pseudovirus controls were incubated and washed under identical conditions. After washing, the flow-through from the washes of "CV-N-treated" pseudoviruses and "ACE-2 treated" pseudoviruses were tested for S1 content in each treatment using the western blot protocol as described. |
63f7c0af32cd591f126166a1 | 7 | Western Blot assays were conducted to quantify washout of CV-N from treated SARS-CoV-2 pseudovirus samples . The collected flowthroughs of the "CV-N-treated" samples, post-washes after treatment with various concentrations of CV-N, were prepared with Laemmli buffer. The samples along with the ladder and CV-N control were loaded on SDS page gel and run at 200 volts for 30 minutes using 1x MES buffer. The gels were transferred onto PVDF membranes using 1x transfer buffer and then blocked for 30 min using 5% skim milk in 1x Phosphate Buffered Saline with 0.1% Tween-20 (PBST). After blocking, anti-CV-N primary antibody was added in a 1:2000 dilution (diluted in 5% skim milk) and incubated overnight at 4ยฐC in a rocking shaker. The PVDF membranes were then washed 3-4 times with 1x PBST and secondary anti-rabbit IgG-HRP conjugate (Abcam) in a 1:2000 dilution (diluted in 5% skim milk) and incubated at 4ยฐC for 1 hour in a rocking shaker. After secondary antibody incubation, the membranes were washed 3-4 times with 1x PBST and once with 1x PBS before being read using Western Bright ECL kit (Advansta Inc., San Jose, CA) and Syngene G-box (Syngene USA Inc, Frederick, MD). Further analyses were performed on blot images with ImageJ . |
63f7c0af32cd591f126166a1 | 8 | Envelope plasmids for all the 22 single-site glycan SARS-CoV-2 pseudoviruses mutants were produced in the Bieniasz Lab by expressing C-terminally truncated, human-codon-optimized SARS-CoV-2 spike protein (pSARS-CoV-2ฮ19) as previously described . Using this plasmid as the template, asparagine at N-linked glycosylation sites were replaced by aspartate in an overlap-extension PCR amplification with primers that incorporated the corresponding nucleotide substitutions. Mutations in the O-linked glycosylation sites in the RBD region (S323, T325) were conducted by a similar process by replacement with alanine. The purified PCR products were then inserted into the pCR3.1 expression vector with NEBuilderยฎ HiFi DNA Assembly. Some mutations within or adjacent to the RBD region, including N282D, N331D, N343D, and double mutation S323A/T325A, were also introduced in spike bearing the R683G substitution, the latter of which alters the furin cleavage site . The plasmids were amplified and validated via gene sequencing for confirmation, prior to being used for SARS-CoV-2 pseudovirus production. |
63f7c0af32cd591f126166a1 | 9 | SARS-CoV-2-ฮCT pseudoviruses for various strains were produced using HEK-293T cells transfected with GeneJammer (Agilent) using IgE-SARS-CoV-2 Spike plasmid WT and Omicron variants (Genscript) and pNL4-3.Luc.R-E-plasmid (NIH AIDS reagent) at a 1:1 ratio. Forty-eight hours post transfection, supernatant was collected, enriched with FBS to 12% final volume, steri-filtered (Millipore Sigma), and aliquoted for storage at -80ยฐC. SARS-CoV-2-ฮCT pseudoviruses were titered to yield greater than 20 times the cells only control relative luminescence units (RLU) after 72h of infection. Pseudovirus neutralization assay was set up using D10 media (DMEM supplemented with 10% FBS and 1X Penicillin-Streptomycin) in a 96well format. CHO cells stably expressing human ACE2 were used as target cells (Creative Biolabs, Catalog No. VCeL-Wyb019). For setting up the pseudovirus neutralization assay, 10,000 CHO-ACE2 cells were plated in 96-well plates in 100 ยตl D10 media and rested overnight at 37หC and 5% CO2. The following day, specific amounts of lectin products were incubated with virus for 90 minutes at room temperature. Following incubation, virus-lectin samples were centrifuged at 1500rpm for 10 minutes and 50 ยตl/well of virus containing supernatant were added to the plated CHO-ACE2 cells. The cells and pseudovirus mixture were incubated in a standard incubator at 37หC and 5% CO2 for 72h. Post 72h, cells were lysed using britelite plus luminescence reporter gene assay system (Perkin Elmer Catalog no. 6066769) and RLU were measured using the Biotek plate reader. Neutralization titers (ID 50 ) were defined as the reciprocal serum dilution at which RLU were reduced by 50% compared to RLU in virus control wells after subtraction of background RLU in cell control wells. Data for percent neutralization vs concentration were fitted to nonlinear regression i.e., log(inhibitor) vs. normalized response --Variable slope Least squares fit to obtain an ID50 value. All calculations were done using GraphPad Prism 8. |
63f7c0af32cd591f126166a1 | 10 | L2 murine fibroblasts were seeded in a 6-well plate at a density of 1x10 6 cells/well and cultured overnight in 2 ml per well of Dulbecco's Modification of Eagle's Medium (DMEM) with 4.5g/L glucose, 2 mM L-glutamine and 10 mM sodium pyruvate (Corning, NY), supplemented with 10% ฮณ-irradiated fetal bovine serum (FBS, Cytiva, Marlborough, MA) and 10 ug/mL ciprofloxacin hydrochloride antibiotic (Bio-World, Dublin, OH) at 37 ยฐC with 5% CO 2 . Twenty-four hours after seeding the cells were washed with 1X Dulbecco's phosphate-buffered saline (DPBS) (Corning) and infected with lectin-virus samples. For the sample prep, the lectins were mixed with the prototype of the betacoronavirus group, mouse hepatitis virus (MHV-A59) ( ) at various multiplicity of infection (MOI) and were incubated for 30 mins at 37ยฐC. Following this, the cells were incubated for 1 hour at 37ยฐC for virus adsorption before the addition of highly pure agarose overlay (Invitrogen, USA) and re-incubated for two days at 37ยฐC, 5% CO 2 . To count the plaques, cells were overlaid with neutral red (Sigma-Aldrich, St. Louis, MO) mixed with agarose-2% FBS DMEM and incubated for 6 hours at 37ยฐC, 5% CO 2 . Next, the plaques were counted, and the viral titer was calculated as plaque forming units (PFU)/mL. |
63f7c0af32cd591f126166a1 | 11 | An SPR Biacore biosensor S200 (Cytiva, Sweden) was used to measure the kinetics and affinity of lectins' interactions with SARS-CoV spike recombinant S1.The experiments were conducted at 25ยฐC in 1x PBS running and sample buffer with 0.05% Tween 20. The CM5 sensor chip (Cytiva) was derivatized using an EDC/NHS mediated standard amine coupling reaction . Spike S1 was diluted in 10mM Sodium Acetate (pH 4.0 to 5.5 established in preconcentration tests) to 25 and 50 ฮผg/m and two S1 surfaces were covalently bound to the sensor matrix at 2970RU and 4200RU, respectively, which is equivalent to a surface density of ~40 and 50RU/kDa respectively . Flow cells were then blocked with 1 M ethanolamine. Soluble ACE2 8.23-2000nM was used as a positive control to confirm the functionality of coupled SARS-CoV S1 subunit. Lectin proteins CV-N, CV-N [P51G], and GRFT were injected over the immobilized proteins at concentrations spanning 8.23 to 2000 nM of a 3-fold dilution series, in duplicate. Multiple buffer injections were included to correct the signal noise. The surfaces were regenerated with two 20s pulses of 10 mM glycine, pH 2.5. The sensorgrams generated from the direct binding assays were processed by double referencing using Scrubber 2.0c software. Exported data were fit to a simple 1:1 and alternative kinetic models were fit using CLAMPยฎ (BioLogic Software) which allowed defining additional kinetic parameters. Sensorgrams demonstrating the binding of Spike subunit S1 to CV-N, [P51G]-CV-N, GRFT, and the sACE2 control are given in Supplemental Figure . |
63f7c0af32cd591f126166a1 | 12 | We investigated the inhibition of infection activities of lectins using SARS-CoV-2 pseudoviruses and ACE2-expressing HEK-293T cells at the outset of this project. Three lectins were compared, namely CV-N, MVN and GRFT. CV-N is a bivalent, mannose-binding lectin that is capable of dimerizing to a 4 binding-site form. Griffithsin is a trivalent, mannose-binding lectin, known to be an obligate domain-swapped dimer, in this case to a 6 binding-site form. In contrast, Microvirin is a monovalent mannose-binding lectin, with no propensity for dimerizing, remaining at 1 binding site per molecule. For direct infection inhibition, CV-N and GRFT both exhibited dose-dependent inhibition (Figure ) of SARS-CoV-2 pseudovirus infection, CV-N with an IC 50 of 45.96ยฑ26.17 nM and GRFT with an IC 50 of 631.6ยฑ378.6 nM. Notably, GRFT as tested in our experiment exhibited comparable potency as found in other pseudovirus assays observed in literature (IC 50 = 293-886 nM (4, 16)), but poorer potencies than in live virus assays (IC 50 = 33-63 nM ). In contrast, the monovalent MVN did not inhibit infection at concentrations up to 2000 nM, suggesting that multivalent binding of the S protein is necessary for inhibition. |
63f7c0af32cd591f126166a1 | 13 | Previous HIV-1 studies with lectin-derived entry inhibitor constructs have demonstrated irreversible inactivation of pseudovirus by lysis and membrane disruption , prompting similar evaluations here with the most active lectin, CV-N. SARS-CoV-2 pseudovirus was incubated with serial dilutions of CV-N, pelleted, and the supernatant tested for p24 released from lysed pseudovirus. Detected leakage was limited and dose-dependent (Supplemental Figure ). Of note, in previous studies where HIV-1 lysis was observed, p24 release occurred as an apparent result of disturbing the metastable structure of the HIV-1 Env spike, and causing conformational rearrangement and irreversible inactivation of spike before target cell encounter . We therefore explored this latter possibility for SARS-CoV-2 by examining the viruses after stringent washouts following lectin exposure, before target cell infection. Indeed, we observed that CV-N-treatment irreversibly inactivates SARS-CoV-2 pseudovirus, with an IC 50 of 242.2ยฑ191.6 nM (Figure ). Briefly, pseudovirus and serial dilutions of CV-N were coincubated for 2 hours at 37 ยฐC, then separated from lectin (Figure ) by repeated resuspension and filtration through a 100 kDa centrifugal concentrator (10 wash cycles) before addition of the virus to HEK293T ACE2 cells. Samples from each wash were saved and evaluated by Western Blot to determine the extent of residual CV-N after washes. Although the wash flowthroughs indicated that all unbound lectin was removed from the virus, small amounts of residual lectin were observed in both the washed virus sample and washed spike-negative virus-like particles (Supplemental Figure ). |
63f7c0af32cd591f126166a1 | 14 | We considered the possibility of S1 shedding as an underlying mechanism of irreversible inactivation, analogous to premature gp120 shedding on HIV-1 Env resulting in virus inactivation . SARS-CoV-2 pseudovirus was treated with serial dilutions of either soluble ACE2 or CV-N, co-incubated for 2 hours at 37ยฐC, and then washed once, saving the flowthrough for Western Blot evaluation. As shown, ACE2 treatment caused dose-dependent shedding of S1 recovered in the flowthrough (Figure ), but CV-N treatment did not (Figure ). Instead, S1 remained associated with the virus samples themselves even after 10 wash cycles (Figure ). |
63f7c0af32cd591f126166a1 | 15 | With both S1-shedding and large-scale virus lysis ruled out, we turned to examination of the site(s) on the S protein potentially important for the inhibitory function of lectins. As mentioned earlier, SARS-CoV-2 Spike has 22 highly conserved N-linked glycans in each of the three protomers , distributed between the S1 and S2 subunits (Figure ). To determine if the mechanism of action for irreversible inactivation by CV-N is due to a specific subset of glycans on the SARS-CoV-2 S protein, we tested single glycan attachment-site mutants of SARS-CoV-2 for continued susceptibility to lectin inhibition. After pseudovirus production, the infectivity of the glycan attachment site mutants and wild type (WT) SARS-CoV-2 pseudoviruses were measured. |
63f7c0af32cd591f126166a1 | 16 | Single-site glycan mutant and WT SARS-CoV-2 pseudoviruses were treated with serial dilutions of CV-N and evaluated for infectivity as previously described. A subset of glycans in the S1 subunit showed a reduced sensitivity to CV-N activity, indicating a potential role in CV-Nmediated pseudovirus inhibition: N61, N122, N331, N343, N603, N657 (Table ). This experiment was also performed with His-tagged GRFT against the same panel of single glycan attachment-site mutant pseudoviruses, identifying several glycan-bearing residues overlapping with the CV-N subset, although not identical: N61, N165, N331, N343, N603, N616, N657 (shared required residues underlined). |
63f7c0af32cd591f126166a1 | 17 | From a structural perspective, the identified glycan attachment sites form two clusters (Figure ), one associated with the RBD domain of S (N122, N165, N331, N343); and one close to the furin cleavage site (N61, N616, N657). We cannot exclude the caveat that some of these glycan attachment site mutations could also affect trafficking of the virus including protein processing in cells. The reduced infectivity of certain glycan mutants could reflect more global changes in Spike structure and glycosylation. To further evaluate this subset of identified residues, the ability of CV-N to irreversibly inactivate pseudovirus mutants (N61D, N122D, N331D, N343D, N603D, N657D) was tested using the the same protocol as in previous experiments with WT pseudovirus. The data indicated that N61D, N657D, N122D, N331D, and N343D did not lose their infectivity upon CV-N treatment and washout compared to the WT control (Figure ). As before, western blot analysis confirmed the removal of un-bound CV-N from "CV-N treated" glycan single attachment-site SARS-CoV-2 pseudovirus mutants (Supplemental Figures S8 to S13). |
63f7c0af32cd591f126166a1 | 18 | In order to further validate the cluster hypothesis for optimal CV-N pseudovirus activity, we examined a predominantly monomeric CV-N [P51G] mutant for its ability to inhibit pseudovirus. Wild Type CV-N has two mannose-binding domains and the ability to form metastable dimers (up to 40% by mass in literature , resulting in up to 4 mannose-binding domains per dimer . We observed a similar 33% dimer content for the CV-N used in this study by measuring Western Blot band intensities with ImageJ, In contrast, the [P51G] mutation has been shown to hinder dimer formation, resulting in a strong preference for the monomeric form of CV-N (โค3% dimeric ), with only two mannose-binding domains per functional unit . CV-N [P51G] was evaluated first against WT SARS-CoV-2 (Figure ). Compared to WT CV-N, the CV-N [P51G] was a poorer inhibitor of WT SARS-CoV-2 by an order of magnitude (WT CV-N 45.96ยฑ26.17 nM, CV-N [P51G] 394.7ยฑ118.5 nM), arguing that dimer formation is a significant contributor to the potency of CV-N inhibition of SARS-CoV-2. |
63f7c0af32cd591f126166a1 | 19 | CV-N [P51G] was then tested against the previously established glycan attachment site mutants of SARS-CoV-2 (Table ). The pattern of single-site glycan mutants that resisted inhibition included all the residues identified in the WT CV-N experiment (N61D, N122D, N331D, N343D, N603D, N657D; denoted in bold in Table ), but also two additional residues not observed as contributing to resistance in either WT CV-N or GRFT experiments (N282D, N1134D). The overlap of key residues does suggest that CV-N [P51G] is attaching in a similar manner compared to the WT, but the reduced probability of dimer formation suggests that a higher overall concentration is needed to inhibit pseudovirus infection. Initial SPR binding analyses of S1-lectin interactions showed good sensorgram fits to multivalent binding models (Supplemental Figure ), in agreement with the infection inhibition results obtained with different lectin types (Figures and). |
63f7c0af32cd591f126166a1 | 20 | To ensure that these glycan requirements for lectin-inhibition are relevant to the realities of disease progression and virus evolution, we conducted infection inhibition assays with several variants of SARS-CoV-2 pseudoviruses and CV-N, the most potent of the lectins we tested (Figure ). While CV-N produced dose-dependent inhibition against all of the variants tested, it was most effective against the most recent variant tested, Omicron. |
63f7c0af32cd591f126166a1 | 21 | As a further test of lectins' applicability against coronaviruses in general, and translation to inhibition of fully infectious virus, plaque reduction assays were also performed with CV-N treatment against fully infectious murine coronavirus (MHV-A59) (Figure ). The data showed that CV-N inhibits infection of fully infectious murine coronavirus, suggesting a pan-coronavirus potential for lectin inactivation. |
63f7c0af32cd591f126166a1 | 22 | The SARS-CoV-2 virus spike protein complex (S) that is required for host cell infection is substantially covered by post-translationally added glycans (1, 2), as depicted in Figure and Table . Here, we asked the extent to which the function of S could be suppressed by lectins specific for glycans bearing terminal mannose residues . We found that the multivalent lectins CV-N and GRFT not only inhibited host cell infection by the Wuhan variant of SARS-CoV-2 pseudovirus by targeting Spike, but also caused irreversible inactivation of the virus. In contrast, the monovalent lectin MVN showed no inhibition of SARS-CoV-2, as opposed to the known MVN infection neutralization observed with HIV-1 pseudoviruses . Irreversible inactivation of SARS-CoV-2 pseudovirus was observed upon repeated lectin washout after treatment, which notably did not induce S1 shedding as a potential mechanism for the irreversible inactivation observed with lectins, as might have been analogous to the irreversible inactivation of HIV . |
63f7c0af32cd591f126166a1 | 23 | From these results, we may speculate that the lectin, CV-N, does not mimic ACE2. We further speculate that it instead may exert a conformational constraint on S-trimer structural dynamics (analogous to state-stabilizing inhibitors of HIV-1 ( )), and which may persist after repeated washing. As previously described, after 10 washes, both SARS-CoV-2 pseudovirus and spikenegative virus-like particles retained small amounts of residual CV-N (Supplemental Figure ). While this could represent lectin binding to non-spike glycoproteins inherited from producer cells, the possibility remains that remnant CV-N may stay bound to Spike on the Spikecontaining pseudoviruses and is at least partly responsible for the irreversible inactivation. |
63f7c0af32cd591f126166a1 | 24 | Using pseudoviruses containing single glycan attachment-site mutations in the spike protein, initial infection by the mutant pseudoviruses, lectin inhibition of pseudovirus infection, and the irreversible inactivation by CV-N were found to be altered. The identified mutations were a subset of glycan attachment sites present in the S1 subunit of the Spike protein. The locations of the critical glycan attachment site mutations could be grouped into two clusters, one located near the receptor binding domain (N122, N331, N343) and the second proximal to the furin cleavage site between the S1 and S2 subunits (N61, N657). Some of the glycan attachment site mutant SARS-CoV-2 pseudoviruses, especially mutations within the S1 subunit, exhibited significantly reduced infectivity compared to WT (Supplemental Figure ). The changes in overall infectivity could indicate that changes in the overall structure of Spike or pattern of glycosylation could occur, though the changes in infectivity still allowed measurement of the effects of CV-N inhibition on pseudovirus mutant infection (for mutant infectivity levels, see Supplemental Figure , and for CV-N dose responses, see Supplemental Figure ). |
63f7c0af32cd591f126166a1 | 25 | The combination of irreversible inactivation and multivalent glycan involvement suggests that the binding of lectins at multiple glycan sites causes conformational disruption of the Spike resulting in viral entry inhibition. This is similar to what has previously been reported for CV-N in the inhibition of HIV infectivity where a post-attachment role at a stage after CD4 binding was implicated . In turn, the capacity to cause irreversible inactivation, taken with the finding that lectin-driven inhibition of infection occurs with several SARS-CoV-2 variants, including Omicron, supports the potential of lectins for disease treatment. Indeed, such a usage is currently being evaluated . |
63f7c0af32cd591f126166a1 | 26 | Importance of multiple glycan clusters. The dependence of the SARS-CoV-2 antiviral functions of lectins on a set of functionally key glycans in the Spike S1 subunit was demonstrated using pseudotyped viruses containing single glycan attachment site mutants of the Wuhan variant. Mutation of any of the key glycans caused a loss of sensitivity to lectin-based inactivation as judged by an increase in CV-N and GRFT IC 50 values (Table ). The similar but non-identical sets of glycans needed for CV-N and GRFT inhibition of infection likely reflects differences in the engagement of the high-mannose glycans by these two lectins. The individual mutations of these glycan-attachment sites also caused a loss of the irreversible inactivation effect of CV-N (Figure ), suggesting possible cooperative effects between glycans. However, since several of these key glycan-attachment site mutants exhibited greatly reduced infectivity, we cannot dismiss the possibility of more global Spike disruptions, and deciphering this can be addressed by future structural and functional studies. |
63f7c0af32cd591f126166a1 | 27 | The indication that positions N61, N122 & N603 could be involved with CV-N binding is strengthened by reports showing that these sites often bear high-mannose oligosaccharides (1). It is surprising, however, that the RBD glycan attachment sites (N331 & N343) were implicated in pseudovirus sensitivity to CV-N as these two sites are known to bear predominantly complex oligosaccharides (1). CV-N's oligosaccharide specificity and antiviral activity has been extensively studied (>200 publications) and CV-N has always been reported to bind exclusively to high-mannose oligosaccharides, not complex oligosaccharides . The results here are a potential indicator of more global changes in Spike structure and glycosylation may have resulted from the mutation of oligosaccharide attachment sites, including possible cooperative effects between glycan sites. For example, RBD glycans have recently been reported to not be binding sites for CV-N (Muรฑoz-Basagoiti et al. PNAS, 2023, in press), and CV-N and GRFT binding were shown not to interfere with ACE2 binding . |
63f7c0af32cd591f126166a1 | 28 | Conformational disruption of metastable Spike complex. We hypothesize that the observed irreversible inactivation of SARS-CoV-2 Spike by lectins may be rooted in the intrinsic metastability of the Spike protein trimer. Dynamic equilibrium between multiple conformational states in the spike complex has been observed previously , and enables the quaternary structure of the Spike to undergo large-scale conformational rearrangement, upon receptor engagement, to liberate and utilize the fusion peptide in the S2 subunit to effect virus-cell membrane fusion during infection. Thus, conformational metastability is a necessary property of the virus Spike to enable infection. We speculate that the Spike conformational plasticity integral for infection could be hijacked by the engagement of the spike by CV-N and GRFT, to drive the conformationally dynamic S complex into a nonfunctional state. Intriguingly, irreversible structural changes in the Spike protein complex by ACE2 decoy receptors have been reported recently . While the relatedness of structural changes caused by lectins compared to those caused by decoy receptors remains to be understood, both may well be dependent on spike metastability. An alternative consideration, which cannot be discounted, is that CV-N does remain attached to spike after washout, though perhaps obscured by the same non-spike binding as exhibited by the spike-negative virus-like particles. Instead of forcing a conformational change in spike, attached CV-N may function as an inactive state stabilizer of spike, rendering it unable to rearrange into downstream infection-productive configurations. |
63f7c0af32cd591f126166a1 | 29 | Multivalent glycan engagement for antiviral functions. The overall dependence of lectin antiviral activity has previously been reported to be dependent on multivalent interactions which suggest cross-linking of oligosaccharides. Previously reported studies on the lectin GRFT with HIV-1 showed that obligate monomeric GRFT was ~1000-fold less active than dimeric GRFT, suggesting the need for multiple oligosaccharide engagement and was further investigated by studies of flexible GRFT tandemers which showed that increasing distances and flexibility between GRFT binding faces could improve anti-HIV activity while simultaneously decreasing viral aggregation . |
63f7c0af32cd591f126166a1 | 30 | Our current observations include the lack of activity for the monovalent lectin MVN and the significantly reduced activity of the largely monomeric CV-N [P51G] mutant versus the wild-type CV-N. In contrast, HIV-1 has been shown to be as or more sensitive to CV-N [P51G] (IC 50 = 0.7-4.0 nM) than WT CV-N (IC 50 = 0.9-16 nM), suggesting that while similarities in HIV-1 and SARS-CoV-2 spike protein glycans exist, they also diverge in key aspects of their structures, relative distances between glycan sites, and importance to viral entry . Moreover, while CV-N [P51G] retains two glycan binding sites, MVN only has one, but still inhibits HIV-1 infection significantly (IC 50 =180 nM) . The fact that each CV-N monomer has two glycan binding sites likely explains why the [P51G] mutant has more potent activity against SARS-CoV-2 pseudoviruses, though still less than WT CV-N which has a significant percentage in the form of a 4-binding site CV-N dimer. Consistent with the importance of multivalency for lectin antiviral functions, SPR binding analysis of lectin-S1 interactions done as part of this study showed good fits of sensorgrams with multivalent binding models (Supplemental Figure ). As described above, multiple Spike glycans are used to enable inhibition of infection and irreversible inactivation. While the precise combinations of glycans used by the lectins for antiviral function is not defined in our current work, nor the structural ramifications of the glycosylation attachment site mutations, the preponderance of data reported here, together with previously published work on other enveloped viruses, clearly supports the important role of multivalent interactions for lectin antiviral activity. Thus, it may be speculated that virus inactivating lectins may bind to various relatively unstable conformational states of the Spike with more closely spaced glycans, pulling the equilibrium to these states and leading to irreversible alterations in the Spike. This conjecture will need future testing by determining the three-dimensional structure of lectin-Spike complexes. |
63f7c0af32cd591f126166a1 | 31 | Breadth of function. The abundant glycan modifications on the exposed SARS-CoV-2 Spike surface have been found by sequence comparisons to be highly conserved in homologous amino acid positions in the known virus variants . On this basis, it may be proposed that multivalent targeting of key glycans on SARS-CoV-2 Spike improves the chances of avoiding facile mutational escape by the virus. Indeed, the substantial breadth of lectin antiviral effects observed in this current work and elsewhere for variants of SARS-CoV-2 supports this expectation. Furthermore, the antiviral effects observed with a fully infectious coronavirus that is not in the SARS-CoV-2 line argues for a potential broad spectrum anti-coronavirus activity as has been previously reported for GRFT in an investigation of its activity against SARS-CoV-1 . As several single glycan-attachment site mutants identified in the current work reduced their lectin sensitivity, the possibility that mutational escape to a lectin treatment could occur. However, importantly, the single site variants that were found in the current study to cause loss of lectin antagonism also have greatly reduced infectivity. Thus, it is highly unlikely that such mutants would survive, thereby mitigating against lectin escape via single-site glycan mutations. |
63f7c0af32cd591f126166a1 | 32 | Therapeutic and prophylactic potential. The breadth and striking antiviral function of virus elimination triggered by lectins such as CV-N argues for a potential therapeutic and/or prophylactic utility. The capacity to irreversibly inactivate infectious SARS-CoV-2 by lectins is particularly compelling for its potential use, as well as the possibility of being employed in noninvasive and environmental roles (e.g. mask modifications or air filtration). Further, the precedent for nasal sprays as a prophylactic strategy raises another possibility for the use of lectins in prevention strategies. The modified lectin Q-GRFT is currently in clinical trials as a nasal spray for the prevention of SARS-CoV-2 infection, with two of them successfully completed , and one dose escalation trial currently recruiting . It will be critical to expand the functional and structural analysis for the irreversibility effect described here, to better understand the potential breadth of utility of lectins in antiviral treatment and prophylaxis. |
620a3ad70c0bf0c968e8e97a | 0 | An increased demand for more sustainable approaches in the chemical industry is being driven by considerable socio-economic and environmental changes. As a result, reusable, energy efficient, and selective catalysts have become a key pillar in the principles of green chemistry. By the immobilisation of homogeneous catalysts or equivalent chemical functionality onto inert insoluble frameworks, effective heterogeneous catalysts that adhere to these principles can be realised. Heterogeneous catalysts offer advantages over homogeneous equivalents, such as use in continuous flow processes, simple separation and recovery from reactants and products, and straightforward recycling. However, reduced catalytic activity, high-costs, and poor stabilities can often hold-back the application of such catalysts in industry. Immobilisation requires the synthesis of pre-functionalised frameworks, followed by the anchoring of active sites, often requiring multistep procedures and expensive reagents. Moreover, immobilisation can alter the microenvironment of active sites, leading to decreased activity or accessibility for reactants. To make heterogeneous catalysts competitive with homogeneous equivalents, their active sites need to be dense and accessible, their production should exploit low-cost design, and their implementation has to offer simple operation and compatibility with many facets of modern catalysis. |
620a3ad70c0bf0c968e8e97a | 1 | Porous organic polymers (POPs) are excellent candidates for heterogeneous catalysis due to their high surface areas, tuneable textural properties, and excellent chemical, thermal, and mechanical stabilities. The vast number of synthetic routes to POPs also permit readily modifiable chemical functionality and/or broad bottom-up design. Common examples of POPs include covalent organic frameworks (COFs) and conjugated microporous polymers (CMPs), both of which have shown great potential in separation and catalysis applications. However, many POP synthesis routes require precious metal catalysts, which are typically not recycled, significantly driving up cost and hindering scale-up. Furthermore, specifically polymerisable groups in monomeric material are often required for the formation of POP networks. Such monomers are seldomly commercially available or are expensive. |
620a3ad70c0bf0c968e8e97a | 2 | One approach to HCPs utilises non-functional aromatic compounds (i.e. without specifically polymerisable groups) that are 'knitted' together in a variety of ways, such as using external crosslinkers, 8 solvent stitching reactions, or Scholl reactions. Another approach produces HCPs via self-condensation reactions of chloromethyl or methyl ether moieties with aromatic carbon to form crosslinks. Hypercrosslinking requires only abundant Fe-or Al-based Lewis acid catalysts or simple organic acid catalysts. Owing to their low costs and broad design scope, HCPs are developed for a variety of applications including gas separation and storage, solid state extraction, and energy storage. HCPs also show promise in heterogeneous catalysis, both as inert frameworks or as active materials. HCP-based catalysts were employed for a wide array of transformations, ranging from biomass conversion to the photocatalytic reduction of CO2. High surface areas allow for an abundance of catalytically active sites and hierarchically porous structures are beneficial to the mass transfer of reactants and products. As such, HCPs show excellent ability in catalysis and are ideal candidates for industrial applications in accordance with the principles of green chemistry. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.