id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
621d31c257a9d25c436ea885
| 52 |
We then generated blind test sets from: a study on linear calibration of xTB-sTDA results (MOPSSAM), a study on singlet fission materials (INDT), a database of molecules relevant to green chemistry (VerdeDB), and a dataset to test the broad applicability of the model (PCQC-AL). On these external datasets, the ML calibration models outperformed both raw xTB-sTDA and linear calibration, oftentimes significantly. Averaging the best MAE over all external test sets (both S 1 and T 1 ) excluding PCQC-AL gave an MAE of 0.14 eV, compared to 0.38 eV for xTB-sTDA. Including PCQC-AL gave an average MAE of 0.57 eV, compared to 0.83 eV for xTB-sTDA. If xTB-ML is used as the first step in a highthroughput screening process instead of raw xTB-sTDA outputs, its low error can help ensure that all relevant molecules are selected, and vice versa.
|
621d31c257a9d25c436ea885
| 53 |
After evaluating the performance of xTB-ML, we then used the model for four applications. First was comparing the xTB-ML model against directly predicted energies with ML, showing the xTB-ML model had better accuracy (0.11 eV vs. 0.21 eV MAE). Second was rapidly screening 250k molecules for suitability as sensitizers and emitters for spectral conversion applications. Third was mapping inaccuracies of xTB-sTDA in chemical space, using the ML model to predict errors. This was used to see which regions of chemical space xTB-sTDA has high errors in. S 1 errors were small, with most molecules being within 0.5 eV. There were clear regions where xTB-sTDA overpredicted S 1 , but only a few for underprediction. T 1 energies were generally overpredicted, with most molecules being between 0 and 1 eV below TD-DFT values. Global chemical space mapping provides another method of predicting xTB-sTDA error, by calculating which cluster a molecule belongs to and referencing the MAE of that cluster. Properties of low-error molecules were also evaluated, finding non-aromatic molecules are likely to have higher error. The final application was generalizing the methodology to calibrate xTB-sTDA against coupled cluster theory, and generating a new xTB-CC-ML model. The calibrated xTB-CC-ML values had high accuracy (0.10 eV MAE), out-performing TD-DFT values calculated with PBE0 (0.26 eV MAE) and CAM-B3LYP (0.19 eV MAE). We also generated more general calibration models with transfer learning and using B3LYP as an intermediary.
|
621d31c257a9d25c436ea885
| 54 |
There are several avenues for future work. First is improving the ML model architecture. While Chemprop's MPNN outperformed other ML models, primarily due to its advanced featurization, only the 2D molecular structure and xTB-sTDA energy were provided as input. Since the 3D structure is known, including this information would likely improve performance. Another improvement to the ML workflow would be to conduct a more intensive conformer search. While OpenBabel's gen3D function includes a search for 200 conformers, these may not include the lowest energy conformer, thus reducing the accuracy of the xTB portion of the workflow. Using a conformer searching tool such as CREST would be more comprehensive, although the computation time added may detract from the high-throughput nature of the xTB-ML process. The ML model could also be expanded to calibrate several singlet states instead of just the first, similar to that in Kang et al. Beyond higher level first-principles data, the calibration models could be further extended to experimental data. However, this would be time-consuming due to the requirement of real-world measurements. There have been a few previous studies in calibrating TD-DFT against experimental values, 28,29 but these used only small experimental datasets. There is a potential here to apply techniques such as text mining to extract experimental excited state data from published papers, though the differences in reporting may make this difficult.
|
621d31c257a9d25c436ea885
| 55 |
Lastly, although xTB-ML is significantly faster than firstprinciples methods, it is still too slow to screen millions of molecules. As stated previously, with our setup xTB-ML can calculate excited states of approximately 1500 molecules per hour (parallelized over 4 computer nodes), for molecules with < 50 heavy atoms such as those in PCQC. Therefore it would take over 3 months to calculate all 3.5M molecules in PCQC. While a definite improvement over TD-DFT (over 3 years), this is still slow. Expanding to larger databases with bigger molecules would increase runtime even further. Therefore, as a final direction, an optimized workflow using active learning could be implemented, intentionally searching for molecules with certain desired properties.
|
621d31c257a9d25c436ea885
| 56 |
The raw data and code to reproduce the figures presented in this paper are available in a repository on GitHub. Trained ML models and prediction data are also available on GitHub. The code for the calibration workflow presented here is available in a repository on GitHub. The scripts for running TD-DFT, running xTB-sTDA, training the ML model, and using it for predictions are available.
|
646e230a4f8b1884b735cbe3
| 0 |
T he global chemical industry provides a multitude of ma- terials and products that are essential for our modern life. Meanwhile, it has grown into a major industrial sector in terms of resource consumption and environmental impacts over the past decades. When counting both feedstock and energy consumption, estimates from the early 2010s suggest that the chemical industry contributed 10 % of global endenergy demand (or 30 % of total global industrial end-energy demand) and 7 % of global greenhouse gas emissions (5.5 % when only counting CO2 emissions) . As the size of the global chemical industry is expected to nearly double by 2030 in comparison to 2020 , a timely transition to a sustain-able one is indispensable. This is demonstrated by the latest European Union (EU) Chemicals Strategy for Sustainability aiming to boost innovation for safe and sustainable chemicals .
|
646e230a4f8b1884b735cbe3
| 1 |
To date, only a limited number of data sources for the sustainability assessment of a diverse range of chemicals exists, including public databases such as the ecoinvent or IDEA databases, datasets from industry associations such as PlasticsEurope , or commercial databases such as the GaBi or cm.chemicals databases. These form the data basis for assessing the sustainability of chemicals by scientists, regulators and the industry. Noticeably, most data is provided in an aggregated form without detailed process-specific documentation. Consequently, the assumptions and calculations behind these datasets cannot be evaluated or reproduced by users. In contrast, the ecoinvent database provides detailed documentation per disaggregated gate-to-gate dataset and thus allows for a systematic analysis of the quality of datasets in the chemicals sector.
|
646e230a4f8b1884b735cbe3
| 2 |
The main application of the ecoinvent database is life cycle assessment (LCA), a key instrument that can be used for assessing and comparing the environmental sustainability of chemicals . LCA quantifies the environmental impacts of chemical products and processes along their supply chains. An integral step this method is compiling the life cycle inventories (LCI), the sum of inputs and outputs for each process across the entire life cycle (i.e. technical inputs and outputs as well as emissions and resource consumption). Such data is the basis for the life cycle impact assessment (LCIA), in which impacts on climate change, toxicity and further impact categories are quantified . Due to the intertwined nature of the chemical sector, poor-quality data for one chemical product may severely affect the data and thus the sustainability scores of all downstream chemical products and processes that build on this product. Hence, the errors caused by poor-quality data will penetrate and potentially accumulate along the supply chains. Therefore, robust data of chemicals is a prerequisite for deriving sound conclusions and providing reliable decision support.
|
646e230a4f8b1884b735cbe3
| 3 |
When compiling chemical process data, significant challenges may exist, including an enormous diversity of products, complex production processes, long and dynamically changing supply chains, the confidentiality of process information, and inconsistent modeling principles . Tiered methods have been developed to systematically fill data gaps with proxy data . In such cases, it is critical to assess how sensitive the results are to uncertain proxy data, and to transparently document the methods and their limitations.
|
646e230a4f8b1884b735cbe3
| 4 |
water demands in chemical data and on the evaluation of the approaches used to derive proxy data. Proxy-based datasets are compared with independent data sources to assess deviations and errors in quantifying climate change impacts of the same chemicals. Finally, this study makes recommendations for database managers on how they should collect data for key inventory flows, (2) users to check the data carefully before using it and to do a thorough uncertainty analysis, (3) the chemical industry to provide appropriate data to support for their sector to become sustainable, (4) and policy makers on how they should ensure the availability of a sound data basis for their decisions.
|
646e230a4f8b1884b735cbe3
| 5 |
Origin of the proxy data. In the ecoinvent database, proxy data has been used primarily used for estimating missing data on (1) electricity demands, (2) heat demands, (3) cooling water demands, (4) process water demands, (5) wastewater treatment, (6) nitrogen demands for blanketing, and (7) infrastructure requirements . The gap-filling approach (13) for the first six types of missing data is to use proxy data derived from the environmental reports of one specific medium-sized chemical 20 manufacturing site in Germany (located close to the city of Gendorf) , whereas the approach for missing infrastructure data includes data from Gendorf and a second German chemical production site located in Ludwigshafen .
|
646e230a4f8b1884b735cbe3
| 6 |
Main processes at the Gendorf site include the production of surfactants from fatty acids and fatty alcohols, membrane electrolysis of rock salt, the production of vinyl chloride monomers and polyvinyl chloride (PVC) from ethylene and chlorine, the production of ethylene oxide and ethylene glycol from ethylene, air separation to supply chemical reactions with oxygen and produce liquid nitrogen, and the liquefaction of some of the CO2 that originates from the ethylene oxide production. Minor processes include the production of polymer films from various types of polymers and the production of fluorine chemicals such as polytetrafluoroethylene (PTFE) from chlorodifluoromethane. The ecoinvent proxy approach. In the ecoinvent modeling approach, the data from the Gendorf site was processed via three main steps to then be used as proxy. First, the total production volume of the site was determined by adding up the intermediate and final outputs at the site . The Gendorf site produces chemicals largely from external inputs that are produced elsewhere, and the processes at Gendorf generally involve only a small number of synthesis steps. Thus, the total product output, including the intermediates used for the creation of the ecoinvent proxy approach, is only about 64 % higher than the amount of final outputs of the site . In a second step, the reported utility demands of the Gendorf site (electricity, heat, natural gas, cooling water, nitrogen and wastewater treatment) are divided by the total chemical production volume of the site to yield the average amounts of utility inputs and outputs per production output mass as proxies, and the water balance is closed. In the case of infrastructure requirements, satellite imagery has been used by ecoinvent to identify the occupied land area for the Gendorf and Ludwigshafen sites, which was then divided by the estimated number of production plants at the two sites . Together with the estimated plant lifetimes, the building infrastructure based on satellite imagery and the inventory of an existing chemical plant, the land occupation, land use change and building infrastructure requirements were estimated for one chemical plant. The production volumes and the plant lifetime estimates then linked the land occupation, land use change and building infrastructure requirements with a specific amount of chemical production as proxies.
|
646e230a4f8b1884b735cbe3
| 7 |
Finally, a third step connects the proxy data with the specific inputs and outputs in the ecoinvent database. This involves mapping how the proxy data is linked to the existing data in the ecoinvent database, both for environmental exchanges (inputs from and outputs to nature) and technosphere exchanges (inputs and outputs of man-made products and wastes). For example, the electricity input per amount of chemical is connected to the ecoinvent technosphere exchange electricity, medium voltage, and the uptake of cooling water from river and well is connected to the environmental exchange water, cooling, unspecified natural origin.
|
646e230a4f8b1884b735cbe3
| 8 |
Inconsistent application of energy proxy data. After these three preparatory steps, the proxy data for chemicals production processes from the Gendorf and Ludwigshafen sites are applied to all datasets with gaps in their exchange data. Examples of using the proxy data for the electricity, heat or water requirements of basic chemicals in the ecoinvent datasets are visualized in figure . From this overview, it can be seen that the proxy data has been frequently applied, even for some very common chemicals whose synthesis steps and conditions are well covered in the literature in comparison to more specialized chemicals. This suggests that sometimes proxy data use was not necessary, contributing to the uncertainty of results. In some cases, the proxy data is used for several sequential synthesis steps, e.g. from cumene production to the epoxy resins production. Under such circumstances, the proxy data may prominently determine the environmental impacts of the final product.
|
646e230a4f8b1884b735cbe3
| 9 |
The implementation of the proxy data in the ecoinvent database follows usually one of two approaches: one that was developed for ecoinvent v2 about 17 years ago , and an updated one that was used in recent years . Both approaches differ in the exact numbers they use because different years of the environmental reports from Gendorf have been used as data sources, but they also differ in terms of coverage and methodological details. For example, the ecoinvent database v3 connected the proxy Gendorf natural gas and steam inputs to the ecoinvent datasets heat, district or industrial, natural gas and heat, from steam, in chemical industry, respectively. The earlier approach from ecoinvent v2 instead used either only heat, from steam, in chemical industry (about 55 % of cases) or only heat, district or industrial, natural gas (about 45 % of cases). This distinction is critical, since the ecoinvent dataset heat, from steam, in chemical industry includes coal and oil combustion as heat supply, which makes this dataset more than 50 % higher in terms of the climate change impacts than heat, district or industrial, natural gas in the Region Europe (RER) version. The difference between environmental impacts between these approaches is even more extreme when considering other impact categories such human health impacts from particulate matter. The reasoning why sometimes only natural gas and at other times a range of fuels were used as proxy input in the ecoinvent database v2 remains unclear, but it may be related to the different authors who have implemented the proxy approach in different ways.
|
646e230a4f8b1884b735cbe3
| 10 |
It should be noted that the synthesis routes of some chemi- GaBi background data, and may thus also be, to some degree, subject to influences from the use of the proxy data in ecoinvent. Furthermore, the use of such aggregated datasets adds intransparency and users cannot no longer adapt the supply chains to their own specific needs. In addition, the aggregated datasets may be base on different background data, including using outdated ecoinvent background data from ecoinvent v2. For example, some PlasticsEurope datasets have only been updated infrequently, and thus date back 15 or more years in many cases. Hence, in addition to intransparency, the aggregated PlasticsEurope data themselves may cause major consistency issues within the ecoinvent supply chains for chemicals.
|
646e230a4f8b1884b735cbe3
| 11 |
Among the analyzed ecoinvent datasets, ethylene dichloride production and methyl tert-butyl ether production stand out 25 because they contain no data of water, electricity, heat, blanketing nitrogen or wastewater treatment requirements at all, neither derived from process-specific data nor from a proxy approach. They do, however, contain the proxy data for infrastructure demands, which is the case for nearly all chemical 30 datasets in the ecoinvent database. It is unclear why some of the inputs or outputs have not been added by either using the proxy approach or specific data, as literature suggests that at least some of these inputs and outputs exist . Such inconsistencies in the implementation of the proxy 35 approach have largely been removed in the latest revision of the ecoinvent data, where some of the older datasets have been updated with the more recent approach. Both approaches still co-exist in the current ecoinvent database v3.8. They share the main weakness that the actual input to the chemical site 40 at Gendorf is natural gas, which is subsequently transformed into electricity and heat at the site, while both the old and new proxy approach in ecoinvent assume this input to be heat. This does not only lead to a mismatch in terms of the energy forms and fuels, but also results in the double-counting of transformation losses associated with the combustion of fuels, and is also very specific to the Gendorf site.
|
646e230a4f8b1884b735cbe3
| 12 |
A closer look at individual cases reveals further nuanced deviations in implementing the proxy data for heat inputs that remain unexplained in the documentation. For example, the heat input added to the ecoinvent database v2 dataset for chloromethyl methyl ether production is 1.9 MJ kg -1 , while both the dataset report and the other datasets from that time use a value of 2 MJ kg -1 instead. Another more recent example of inconsistencies in the individual datasets is the case of 55 3-methylpyridine production that receives all heat inputs from the dataset heat, from steam, in chemical industry instead of a split between that input and heat from natural gas like the majority of other recent datasets. Likewise, the electricity demand also experiences some implementation inconsisten-60 cies. For example, ethylene oxide production in the ecoinvent database v2 has been approximated as 0.33 kWh kg -1 of electricity, medium voltage, while the usually applied proxy amount at that time was instead 0.333 kWh kg -1 .
|
646e230a4f8b1884b735cbe3
| 13 |
The earlier approach to implementing the proxy data for heat demands frequently assumed, in cases of exothermal chemical reactions, that the generated heat from the chemical reaction would exactly match the heat demand for any subsequent purification steps and could therefore be neglected (e.g. nitrobenzene production) . Analyzing the literature data, however, shows that the mean energy release by exothermal reactions is around 2.7 MJ kg -1 , while the mean purification energy demand is at about 3.9 MJ kg -1 (17). In addition, the costs for heat recovery from exothermal reactions may in some cases be uneconomically high, resulting in non-implementation or limited use of heat recovery. Hence, neglecting any heat demands for exothermal reactions is in many cases an oversimplification that leads to an unnecessary underestimation of environmental impacts from chemicals production.
|
646e230a4f8b1884b735cbe3
| 14 |
Proxies for water and nitrogen demands. Meanwhile, consistency in determining water proxy values in the ecoinvent datasets has been improved. In the ecoinvent database v2, technical water inputs have sometimes been estimated as 50 % of the proxy cooling water inputs (e.g. ethyl benzene production ( )), at some other times as 25 % of the proxy cooling water inputs (e.g. dimethyl ether production ), or at other times have been neglected completely (e.g. boric oxide production ). The reasoning for choosing either of these approaches remains unclear, but the implementation appears to depend somewhat on the authors. Technical water inputs of 25 to 50 % of the cooling water inputs appear high in comparison to the literature examples and the share in the IDEA database (about 3 % of the cooling water inputs), which has been derived from the empirical Japanese input-output mass balance data . In the updated ecoinvent v3 approach, technical water proxy inputs are derived from the amount of Gendorf water inputs that are neither cooling nor tap water . This leads to much lower technical water inputs compared to the earlier approaches, and is at about 10 % of the cooling water inputs.
|
646e230a4f8b1884b735cbe3
| 15 |
That is finally much closer to the reported average value of 3 % in the IDEA database. Compared to these technical water inputs, only a minor share (about 0.2 %) of water is modeled to be treated in a wastewater treatment plant, because the revised v3 proxy approach of ecoinvent neglects the internal wastewater treatment plant at the Gendorf site. Adding the internally treated wastewater would increase the treated water to about 65 % of the technical water demands and would lead to additional impacts associated with wastewater treatment (e.g. due to chemical demands or energy demands).
|
646e230a4f8b1884b735cbe3
| 16 |
A new addition to the chemical data proxy approach in ecoinvent v3 is the nitrogen demand for blanketing . This nitrogen demand is derived from the captured nitrogen at the Gendorf site . As outlined above, however, the Gendorf site also produces liquid nitrogen as output, which has not been considered in the ecoinvent modeling. Hence, the nitrogen demand for blanketing is likely overestimated. In addition, this nitrogen demand has been modeled with a technosphere exchange of nitrogen, liquid; however, it is unnecessary to use liquid nitrogen for blanketing and the energy demands of nitrogen liquefaction are several times higher than those for producing nitrogen gas , which leads to an overestimation of environmental impacts.
|
646e230a4f8b1884b735cbe3
| 17 |
Frequency of proxy use. Overall, this study has identified 245 ecoinvent v3.8 cut-off datasets (out of 1207 analyzed chemical sector datasets) that use proxy data for heat or electricity. For other purposes, the proxy data is applied even more frequently, but its use is generally very difficult to track due to differences in the documentation and in the implementation of proxies over time. In particular, the datasets from the older ecoinvent versions have seen a variety of individual interpretations of how to implement the proxy data, and these are thus difficult to identify, so the real number of datasets with energy proxy use may be slightly higher. More recently, the energy proxy approach used in the ecoinvent database has also been applied as a core modeling principle in the cm.chemicals database by carbonminds, which is specialized in chemical LCI datasets . Dataset-specific documentation for that database is not available, and customers only receive fully aggregated data; however, the overall documentation suggests that in this database, the datasets of at least 536 out of 1012 chemicals use the energy proxy data from the ecoinvent database (in June 2022) and thus suffer from the same inaccuracy of 20 this approach. GaBi, the third LCI database with a rather broad global chemical dataset coverage, generally does not disclose its input data at that level of detail and therefore may or may not be subject to similar limitations.
|
646e230a4f8b1884b735cbe3
| 18 |
To quantify the influences of using the proxy data on the sustainability metrics of the chemical sector, both gate-to-gate and cradle-to-gate data from the ecoinvent database have been analyzed in more detail. This analysis focuses on energy as a prominent driver of environmental impacts. Comparisons of gate-to-gate electricity and heat inputs in ecoinvent with the IDEA data are presented in figure .
|
646e230a4f8b1884b735cbe3
| 19 |
This comparison shows that both median gate-to-gate electricity and heat demands in ecoinvent are close to the proxy values derived from the Gendorf site data (about 10 % lower for both electricity and heat). In direct comparison, the primary energy demand for electricity in ecoinvent is higher than for heat. Both IDEA and IHS Markit data, however, show a distinctly different pattern: the median primary energy demand for electricity per gate-to-gate dataset is about half of 40 the ecoinvent value in case of IDEA, while the median primary energy demand related to heat is 220 % higher in the case of IDEA. IHS Markit data shows a similar pattern. In addition, the spread of heat demands in both IDEA and IHS Markit is much wider than in ecoinvent, where instead the overall distribution remains close to the proxy value. Some of the differences between the three databases may arise from a different subset of processes or different allocation principles, but even the much smaller dataset presented by shows a similar pattern compared to the IDEA and IHS 50 Markit data and thus deviates from the ecoinvent pattern. Hence it is most likely that the proxy approach in ecoinvent distorts the overall results by narrowing down the variability in electricity and heat demands as the same data is used for many processes and, more importantly, by underestimating heat and overestimating electricity demands on average due to the missing representative production profile (and hence biased energy demand) of the raw data behind the proxy approach.
|
646e230a4f8b1884b735cbe3
| 20 |
In fact, a major part of the discrepancies between ecoinvent on the one side and the IDEA and IHS Markit data sources on 60 the other side can be directly traced back to the characteristics of its raw data. Since the Gendorf chemical site includes an on-site power plant for supplying both electricity and heat which is not distinguished by the ecoinvent approach, it can be expected that the actual heat and electricity supplied to the processes differs from the ecoinvent assumptions. These assumptions consider the entire energy and fuel inputs to the site including the power plant. Replacing the fuel supply for the power plant in the energy proxy with its heat and electricity production would increase the electricity input of the proxy approach and decrease the heat input of the proxy approach because currently, all the fuel input is assumed to be heat. In part, this may be caused by an incomplete energy balance of the Gendorf site since sales of electricity to the grid are not reported, but are possible according to the registration of the on-site power plant (as described in more detail before).
|
646e230a4f8b1884b735cbe3
| 21 |
Moreover, the rather small number and limited diversity of individual chemical processes at the Gendorf site further contribute to an energy demand profile that is likely not representative for the entire chemical sector. A specific contributor to this is the highly electricity-demanding electrolysis of rock salt at a typical electricity consumption of 6 MJ kg -1 per rock salt input , which is much higher than the median electricity consumption per chemical according to IDEA (0.9 MJ kg -1 ) based on the datasets used in figure 2. In addition, the energydemanding breakdown of long-chain molecules, complicated multi-step synthesis processes, or heat-demanding purification processes of chemicals with very similar physical properties are largely absent from the portfolio of the Gendorf site. For example, the steam cracking of hydrocarbons into ethylene is taking place off-site, while the Gendorf site then only uses the ethylene that is supplied by pipeline. Finally, the technically straightforward but electricity-demanding production of liquefied industrial gases takes place at the site and further increases the electricity consumption at Gendorf. Taking into account all these factors as well as the size of the plant, it appears plausible that the raw data for the chemical proxy approach has a limited representativeness for the entire sector.
|
646e230a4f8b1884b735cbe3
| 22 |
The consequences of high electricity demands and low heat demands in the chemical proxy of ecoinvent for climate change impacts are illustrated in figure . The median cradle-to-gate climate change impacts per functional unit (FU) in ecoinvent are about 45 % lower than in IDEA (2.5 kg CO2e instead of 4.4 kg CO2e). The reason is not that straightforward to trace back because the use of black box PlasticsEurope datasets by ecoinvent as part of the chemical supply chains obscures which emission sources are responsible. As there is still a major overlap in coverage between both the ecoinvent and the IDEA database with regards to the bulk chemicals, and PlasticsEurope data mostly plays a role for the polymer supply chains, but not for most inorganic chemicals, this is not sufficient to explain the major mismatch in climate change scores. These can largely be attributed to heat inputs, which are on average about five times lower in ecoinvent in comparison to IDEA.
|
646e230a4f8b1884b735cbe3
| 23 |
To further eliminate the differences in database coverage as causes for different climate change impacts, a product-by-product comparison for a subset of 50 chemicals from four distinct groups of chemicals has been performed as shown in figure . This comparison between ecoinvent and IDEA climate change impacts and their contributions from electricity, heat and other emissions reveals distinct patterns for different groups of chemicals. Also in this the plotting range have been removed from the calculation of the distribution patterns, but they are included in the calculation of median primary energy demands (horizontal bars). The dashed lines represent the primary energy demands for electricity and heat of the proxy data that are commonly used in the ecoinvent database (mean of the old and new versions). IDEA datasets with net heat or electricity outputs have been excluded as these outputs would be treated as allocatable co-products by ecoinvent.
|
646e230a4f8b1884b735cbe3
| 24 |
comparison, the average climate change impact scores calculated with IDEA exceed the ecoinvent scores substantially, but not to the same extent for all groups. Among the polymers, for example, the ecoinvent polymers that are a direct conversion of PlasticsEurope data (e.g. polyamide 6 (PA6), poly(methyl methacrylate) (PMMA), general purpose polystyrene (GPPS), acrylonitrile butadiene styrene (ABS), and styrene-acrylonitrile (SAN)) match the corresponding IDEA datasets in climate change impact scores better than those that are connected to the chemical supply chains within ecoinvent (e.g. polyethylene 10 terephthalate (PET), ethylene-vinyl acetate (EVA) or polyvinyl chloride (PVC)).
|
646e230a4f8b1884b735cbe3
| 25 |
The production of inorganic fertilizers, in contrast, is a sub-sector within the ecoinvent chemical sector, whose supply chains are more separate from the rest of the chemical sector, and which has been updated in 2020 with process-specific data. Accordingly, the results of ecoinvent and IDEA are in better agreement for systematic fertilizers than for other chemical groups (figure ). There are even some nitrogen fertilizer datasets, which show substantially higher climate change impacts scores in ecoinvent than in IDEA. A reason for this difference is possibly the influence of coal-based ammonia production in China, which leads to higher direct CO2 emissions than the conventional natural gas-based ammonia production, but is only covered by ecoinvent and not by IDEA.
|
646e230a4f8b1884b735cbe3
| 26 |
The comparison of climate impacts of other organic chemicals between both databases shows a trend towards higher impacts calculated from IDEA. In part, this is due to differences in a few key intermediate chemicals that are then used in the manufacturing of many derivatives. For example, ethylene and propylene as key intermediates for the entire chemical sector have been adapted in ecoinvent from Plastic-sEurope data, which uses outdated background supply chain data and is based on other modeling principles and allocation assumptions than ecoinvent (or IDEA) . These two chemicals have lower impacts in ecoinvent than in IDEA and 40 thus contribute to lower overall climate impacts of organic chemicals in ecoinvent. In addition, as has been shown earlier in figure , there is also a large share of proxy use among basic organic chemicals in ecoinvent. The influence of this proxy use is visible in the climate change impacts related to electricity and heat. Many organic chemicals in ecoinvent have either the black box datasets for ethylene or propylene production in their supply chain, which obscure part of their heat and electricity impacts. Consequently, the distinguishable climate change impacts in ecoinvent should be lower than in IDEA, tion, the data availability on inorganic chemical production is more limited than for other types of chemicals, so ecoinvent uses proxy data frequently while IDEA again is relying more on process simulations and reported process-specific data. Higher climate change impacts in IDEA compared to ecoinvent arise both from substantially higher electricity and heat demands. That is also in line with literature recommendation by that suggest to use proxy electricity and heat inputs for inorganic chemicals of 1.9 MJ kg -1 and 5.1 MJ kg -1 , respectively, while the current ecoinvent proxy approach would only suggest values of 1.5 MJ kg -1 and 2.35 MJ kg -1 instead (5). Thus, the proxy electricity demands in ecoinvent, that on average overestimate electricity demands and are strongly based on organic chemicals production, are still not high enough of match the even higher electricity demands for many inorganic chemicals in IDEA. In addition, the heat demands for inorganic chemicals production in ecoinvent also tend be substantially lower than in IDEA among the compared substances. A particularly strong mismatch between ecoinvent and IDEA climate change impact scores is observed for ferric chloride (FeCl3), a chemical that is very commonly used as a flocculant in wastewater treatment and drinking water production. Ecoinvent has partly adapted the supply chain of ferric sulfate (Fe2(SO4)3) as a proxy, which may be a main reason for the deviations in climate change impacts and should trigger an investigation about more appropriate alternatives. At the same time, impacts of inorganic chemicals are sensitive to cut-off and allocation choices in their respective LCI databases due to their metal content, which may also need further analysis.
|
646e230a4f8b1884b735cbe3
| 27 |
Recommendations. The proxy data derived from one specific chemical site has been used frequently for compiling LCI data for chemicals in the ecoinvent database, with the intention of supplementing available data and to avoid neglecting electricity, heat, water, nitrogen or infrastructure demands of chemicals production in case of data gaps. However, this study finds strong evidence that the overall proxy electricity demand used in the ecoinvent is too high, while the heat demand is far too low. In tendency, this leads to an underestimation of key environmental impacts of chemical production such as climate change. Due to the high prevalence of using the proxy data (more than 20 % of datasets in case of energy inputs), such errors can accumulate along chemicals supply chains. The cause for the differences compared to sector averages is likely due to the underlying raw data for deriving proxies for energy inputs that comes largely from a single chemical site in Germany with an unrepresentative process and product portfolio. This is concerning as the ecoinvent database has been and are expected to be widely used in the foreseeable future as the major source of sustainability assessment data for chemicals by a wide range of stakeholders. Further use of such erroneous data in the ecoinvent database will continue to perpetuate the errors and mislead decision-making. Therefore, immediate concerted efforts are warranted to correct the identified inconsistencies and errors, and make funding available for doing so.
|
646e230a4f8b1884b735cbe3
| 28 |
As an immediate mitigation measure, we call for replacing proxies as fast as possible by including the processspecific energy and water demand values from published data sources, several of which have been identified in this article . Proxy data should only be used for remaining gaps that cannot be filled by any other means. For P R E -P R I N T that, company-wide environmental reports of major players in the chemical sector with a much larger production volume coverage than the currently used data could help to improve proxy representativeness. An alternative approach with proxies derived from manufacturing process-based data provides the option to additionally distinguish key features of different types of chemical synthesis pathways (like exothermal and endothermal processes, reaction and purification requirements, or different types of chemicals). The latter approach can furthermore help to avoid problems with regards to incomplete company reporting and issues with regards to the exact system boundary, that have proven to be problematic in the approach ecoinvent currently uses. Finding the most meaningful level of complexity for such an approach would need more research, though, and should also consider alternative approaches like using modern cheminformatics, natural language processing or machine learning to unlock the full potential of existing data in the literature.
|
646e230a4f8b1884b735cbe3
| 29 |
Identifying proxy data and tracing its impacts in ecoinvent was overall difficult because different authors have implemented the same proxy approach in different ways, documentation in the datasets was incomplete, and the chemical supply chains in ecoinvent were not fully connected due to the use of aggregated black box chemical data from PlasticsEurope. These issues strongly hindered data analysis, adaptation and plausibility checks. Substituting PlasticsEurope datasets with fully documented gate-to-gate data, which is publicly available, and creating a separate proxy dataset with clear documentation that is connected in cases with data gaps if needed would increase transparency, remove inconsistencies, improve traceability and add value for all users of the ecoinvent data. This approach is the only way to make sure that upstream data improvements also show up in the downstream products, and that decarbonization progress of the chemical industry can be tracked within the ecoinvent database in a timely manner. In addition, the numerous data products that are also based on similar proxy approaches or on the ecoinvent chemical data will need to be updated as well.
|
646e230a4f8b1884b735cbe3
| 30 |
Finally, chemical sector proxy data is frequently used in ecoinvent for bulk chemicals with good data availability, which can easily be replaced by more specific data, but the proxy approach is also applied for more specialized chemicals with poor data coverage. As ecoinvent is, in contrast to most commercial LCI databases, a not-for-profit project that relies on data contributions, the authors of the present study strongly urge any members of the scientific community and from industry to share better data and to thus improve the meaningfulness of ecoinvent data for all its direct and indirect users. Ultimately, the sharing of high quality data and basing the global decisionmaking on such data is at the core of ensuring the effectiveness of the decarbonization efforts of the chemical sector.
|
646e230a4f8b1884b735cbe3
| 31 |
General approach and data sources. The analysis in this study is based on the ecoinvent database v3.8 (cut-off version) . The conclusions are by also largely valid for other ecoinvent versions, as all are derived from the same unlinked LCI datasets with differences coming from different linking rules, e.g. on the issues of co-product allocation, waste treatment burdens and benefits, and calculation of electricity mixes. The ecoinvent datasets were primarily compared against the IDEA v2.3 database (6), with additional comparisons to a subset of 750 IHS Markit Q2 2021 unit process dataset and additional scientific research .
|
646e230a4f8b1884b735cbe3
| 32 |
Information on the use of proxy data was retrieved from the ecoinvent documentation in the respective datasets, several re-65 ports , and industry raw data behind the proxy data . Gate-to-gate analyses of individual production processes were performed on the raw data directly provided by the respective authors. Cradle-to-gate analyses along the chemical supply chains were performed using Simapro 9. Identification of proxy use. The identification of chemical datasets in the ecoinvent database was based on the product and sector classifications provided by ecoinvent in the raw data, except 75 for the cradle-to-gate analyses that used the classification in the Simapro software instead. The analysis generally excluded ecoinvent market datasets as these are supply or demand mixes composed of other production datasets with limited additional data, and Rest of the World (RoW) datasets as these are 80 adapted copies from localized datasets without region-specific production data collection. The identification of relevant datasets in the IDEA database was based on the Japanese product classification system in the raw data. All IHS Markit datasets were chemical sector datasets, but some available 85 datasets were selectively excluded (e.g. utility datasets such as for water supply) to focus on chemicals production processes, to ensure comparability with the other data sources and to avoid strong biases towards individual product groups.
|
646e230a4f8b1884b735cbe3
| 33 |
2. Screening for the absolute values of certain proxy approaches, and then checking the flow-specific documentation of the ecoinvent datasets or the original data sources to confirm proxy-data use. 100 3. Screening for the specific heat-to-electricity ratio of the proxy data (as the unallocated ecoinvent datasets were unavailable to us, and the allocation procedure changes the absolute numbers but leaves the ratio of inputs unchanged), and again confirming with the documentation. 105 4. Manually analyzing the documentation of cases with suspected proxy use in ecoinvent datasets (for example where input values were very close but not identical to the proxy approach values).
|
646e230a4f8b1884b735cbe3
| 34 |
Analysis of the proxy approach. After identifying datasets con-110 taining proxy data, the exact implementation was analyzed by comparing with the industry raw data of the chemical plant site behind the proxy data. Also, additional analyses were conducted to check the consistency of using proxy data, the documentation of relevant process flows, and the appro-115 priateness of using proxy data (e.g. whether process-specific data had been available at the time of dataset creation). Furthermore, a mapping of the use of proxy data in the bulk data from the different databases to evaluate the impacts of using proxy data. Before conducting this comparison, we first eliminated data inconsistencies. Hence, the gate-to-gate analysis of the ecoinvent, IDEA and IHS Markit datasets was performed by merging each of the electricity and heat inputs, and converting them to primary energy demands under the assumptions of 40 % and 80 % net efficiency, respectively. This was done to harmonize the datasets and enable separating out the influence of proxy data on the final results. A small number of the IDEA datasets that used the system expansion principle were excluded from the analysis, as this principle is not employed in the ecoinvent database. Co-product allocation based on prices was additionally carried out for the IHS Markit datasets to transform multi-output processes into single-output processes. The price data for that was directly obtained from the IHS Markit datasets. Thus, economically allocated singleoutput datasets from IHS Markit were derived in line with the main allocation approach of the ecoinvent database for chemicals. Then, the estimated primary energy demands of processes in the ecoinvent datasets were compared to the corresponding gate-to-gate primary energy demands in the IDEA and IHS Markit datasets.
|
646e230a4f8b1884b735cbe3
| 35 |
For the cradle-to-gate analysis, electricity and heat input data was identified in both the ecoinvent and IDEA databases, and their emission intensities were adjusted to 30 0.16 kg CO2e MJ -1 (33) and 0.1 kg CO2e MJ -1 (34) according to the data from PlasticsEurope, respectively. The value for heat input assumed a useful heat content of industrial steam of 2.8 MJ kg -1 . These emission intensity values were selected to mitigate impacts of energy data differences between the ecoinvent and IDEA databases, and also because the ecoinvent database uses a substantial subset of aggregated datasets from PlasticsEurope, within which emission intensities cannot be adjusted. Other greenhouse gas emissions in both databases were kept as they were. The resulting climate change impacts along 40 the chemical supply chains were calculated based on the 2013 Intergovernmental Panel on Climate Change (IPCC) GWP 100a method in the 2013 version and grouped into impacts resulting from electricity, heat and others (the latter includes unspecified emissions from aggregated PlasticsEurope datasets, direct process emissions of chemicals, and emissions from other parts of the supply chains). Finally, the environmental impacts of chemicals were compared between both databases with regards to their overall patterns and per chemical for a subset of 50 substances.
|
646e230a4f8b1884b735cbe3
| 36 |
Proxy data source description. The total output of the site fluctuates around 1.6 Mt a -1 , which is a medium size for chemical plants. The share of a broad range of surfactants to the total production amount is about 45 %, which makes this the main product group from the site, while downstream products related to rock salt electrolysis (PVC, vinyl chloride monomers and caustic soda) represent the second major group of products, with a share around 35 %.
|
646e230a4f8b1884b735cbe3
| 37 |
The Gendorf site has received deliveries of about 1.4 Mt a -1 of educts in recent years, including fatty acids, fatty alcohols, 60 rock salt, and about 0.3 Mt a -1 of ethylene . Further inputs include about 3.6 TJ a -1 of electricity from the grid, 4.5 TJ a -1 of natural gas, and 0.5 TJ a -1 of steam from an adjacent waste incineration plant . Water is taken from the Alz river, a well and the local drinking water network, while oxygen, nitrogen and pressurized air are obtained from the surrounding air.
|
646e230a4f8b1884b735cbe3
| 38 |
The natural gas input is largely used by the power plant of the Gendorf site, which uses a gas turbine, two steam turbines and a heat recovery steam generator to supply up to 72 MW of electrical power and 146 MW of heat in the form of steam . An auxiliary boiler at the power plant is able to produce additional steam . The plant is able to generate surplus electricity and feed this back into the electricity grid , but it is not reported to which extent this happens. Some of the natural gas is also directly used by the chemical processes, but to a smaller extent than for the power plant. An on-site wastewater treatment plant treats about 2.5 Mt a -1 of wastewater from the chemical processes, while a minor amount of about 6 kt a -1 of wastewater is instead treated at the municipal wastewater treatment plant .
|
6707a95c51558a15efafa915
| 0 |
Metal halide perovskites are promising candidates in the field of photocatalytic activity, photovoltaics (PV), light-emitting applications, and lasers due to their versatile composition, crystalline structure, and tunable optical properties. Generally, the metal halide perovskites (HPs) are expressed in the form of ABX3, where A is a monovalent cation (organic or inorganic), B is a divalent cation (Pb 2+ , Sn 2+ , Ge 2+ ), and X is the halide anion (Cl¯, Br¯, or I¯). The dimensionality of the perovskite crystalline structure is defined by the connectivity of BX6 4- octahedra (corners-sharing) and classified from zero to three-dimension (0D-3D), where ABX3 is 3D, and A4BX6 is the 0D. This variation in the crystalline structure significantly affects the optical properties of the perovskite materials. For instance, the band gap value of the 3D CsPbBr3 is about 2.35 eV, and that of 0D C4PbBr6 is 3.90 eV. Also, while the 3D CsPbBr3 possesses a bright green photoluminescent (PL) emission (~ 527 nm), the PL emission from the 0D Cs4PbBr6 is debatable. Some studies state that the green emission from the 0D C4PbBr6 is an inherent property of the materials, whereas others attribute this emission to the contamination by the 3D CsPbBr3 phase and structural defects. Since the halide perovskite-nanocrystals (HP-NCs) have highly active sites, it is favorable for carrying out photocatalytic reactions. However, the stability of HP-NCs is affected by the solvent, surface ligands, and environmental conditions. Numerous studies show HP-NC's sensitivity to oxygen and moisture and that they decompose upon illumination. Various techniques are used to study the decomposition process of HP-NCs. The most common methods are the in-situ PL and absorbance measurements, where quenching of the spectral features is indicative of the decomposition progress. Another technique is calorimetry, where the enthalpy of the dissolution can be extracted. Stability protocols are developed to track the decomposition of perovskite materials in PV devices. Other methods are ultraviolet photoelectron spectroscopy (UPS) followed by X-ray electron spectroscopy (XPS), which can give information on the electronic and chemical properties of the material under consideration. Many reports focus solely on one aspect of instability, such as medium or temperature. Yet, the primary initiator and the actual decomposition process of the HPs remains unknown.
|
6707a95c51558a15efafa915
| 1 |
Understanding the decomposition process of the HP-NCs under illumination, in the presence of oxygen and moisture, is crucial as it will pave the way for finding a proper solution to their stability issues. Although electron paramagnetic resonance (EPR) is an effective technique, this approach has not been used so far to study the decomposition mechanism of HP-NCs.
|
6707a95c51558a15efafa915
| 2 |
EPR is particularly adept at detecting unpaired electrons, making it a powerful tool for studying radicals and their role in material decomposition. Moreover, it allows for direct observation of radical formation and dynamics. This EPR technique has been widely used to investigate the decomposition mechanisms of polymers and biological compounds. However, despite its utility, this approach has not been used to study the decomposition of HPs. Notably, reactive oxygen species like OH • and O2 •-radicals have extremely short half-lives of 10⁻⁹ and 10⁻⁶ seconds, respectively. Spin-trapping EPR enables the detection of short-lived radicals by forming stable adducts. Spin traps like 5-tert-butoxycarbonyl 5-methyl-1-pyrroline N-oxide (BMPO) enable their detection at room temperature, with distinctive EPR spectra measurable for up to 35 min, and have been effectively used to study radical generation in X-ray-irradiated gold nanoparticles. BMPO has been used effectively to study radical generation by gold nanoparticles irradiated with X-rays. Here, we show a unique way to study the decomposition process of HP-NCs. We are monitoring the ability of the 3D CsPbBr3 and 0D Cs4PbBr6 NCs to generate radicals as a response to visible light illumination in a polar solvent. The generated electrons are migrating and confined by the BMPO spin trap. Using the EPR technique, we can spot the moment when the NCs are starting to decompose. Moreover, we follow this process, varying the amount of a polar solvent and the duration of the illumination. The distinctive EPR spectra provide information on the stability of the studied HPs-NCs.
|
6707a95c51558a15efafa915
| 3 |
Our results show that the 3D CsPbBr3 NCs exhibit enhanced dye degradation activity compared to 0D Cs4PbBr6 NCs. However, correlating the results of EPR measurements with the photocatalytic activity suggests that CsPbBr3 NCs undergo decomposition during the photocatalytic process. This phenomenon indicates that CsPbBr3 NCs act as sacrificial agents, thereby enhancing the photocatalytic activity. Our study demonstrates the effectiveness of EPR spectroscopy in probing the decomposition of CsPbBr₃ and Cs₄PbBr₆ NCs, revealing that lightinduced radical formation accelerates their decomposition. This research uncovers radical-driven decomposition pathways, providing new insights into the decomposition mechanisms of HP-NCs. These findings are pivotal for enhancing the stability and efficiency of HP-NCs in both optoelectronic and photocatalytic applications and provide essential guidance for developing more robust materials.
|
6707a95c51558a15efafa915
| 4 |
After the mixture turned orange (~20 min), which indicates the formation of CsPbBr3 microcrystals, the crystals were dissolved by heating to 145 ˚C and stirring for 40 min. Next, 1 mL of DMF was added, and the solution became transparent. This solution was added dropwise into the ligands blend (OA=125 µL, OLAm=125 µL, and ODE=1.25 mL). Lastly, a swift injection of acetone (5 mL) resulted in a gradual color change from white to green solution and the formation of Cs4PbBr6 NCs. The synthesized NCs were centrifuged at 3500 rpm for 5 min and redispersed in 1 mL of toluene.
|
6707a95c51558a15efafa915
| 5 |
For the synthesis of the CsPbBr3 NCs, a hot injection method was involved. In brief, ODE (10 mL) and OA (630 µL) were introduced to Cs2CO3 (203.5 mg) and subjected to degassing at 120 ˚C for 1 hour. At this point, the color of the solution turned to mild yellow, indicating that cesium reacted with OA(Cs-OA). Then, this solution was heated to 150 ˚C under N2 gas for 1 hour.
|
6707a95c51558a15efafa915
| 6 |
Simultaneously, PbBr2 (69.1mg) was added to 5 mL of ODE and stirred under vacuum for 1 hour at 120 ˚C. The mixture of OA: OLAm (0.5 mL:0.5 mL) was injected into the PbBr2 solution and heated to 170 ˚C for 1 hour. Next, 0.4 mL of Cs-OA was injected, and the reaction was quenched in an ice bath. Here, the color of the solution turned green. To eliminate the ligands from the surface of the HP particles, they were washed several times with toluene with the assistance of the centrifuge at 3500 rpm for 5 min. The particles were redispersed in toluene and kept in the solution for further characterization.
|
6707a95c51558a15efafa915
| 7 |
Organic dyes were prepared by dissolving them in Sol.1 with a concentration of 152 µM of MO and BB. Conducting photocatalysis in toluene is impractical because most dyes are soluble in aqueous or polar solvents. Furthermore, photocatalysis becomes unfeasible in high concentrations of EtOH due to the rapid decomposition of the HPs in such an environment. Consequently, we investigated the photocatalytic capabilities of the HPs for dye degradation in Sol. 1.
|
6707a95c51558a15efafa915
| 8 |
Pristine CsPbBr3 and Cs4PbBr6 NCs were dispersed in Sol.1 to a total volume of 20 mL and a final concentration of 52 µM. Then, 10 mL of MO solution was added to the dispersed NCs and kept under constant stirring. The samples were placed 15 cm from the Xenon lamp source with a power of 60 mW/cm 2 and appropriate cutoff filters to utilize the visible light excitation. For CsPbBr3 NCs, the degradation was analyzed with 5-min time intervals. Every 5 min, 2.5 mL of solution was taken and centrifuged at 7000 rpm for 3 min. The supernatant was subjected to absorption measurement to study the photocatalytic dye degradation. Similarly, in the case of Cs4PbBr6 NCs, the degradation was performed at 30 min time intervals.
|
6707a95c51558a15efafa915
| 9 |
To analyze the degradation of BB in the presence of CsPbBr3 and Cs4PbBr6 HP's catalysts in Sol.1 (total volume of 24 mL), 6 mL of BB dye was added under constant stirring. Before adding the photocatalyst, 3 mL of the pristine dye solution was taken as a "0 min" measurement. Then, the photocatalyst was added to this solution and kept under visible light. The degradation of CsPbBr3
|
6707a95c51558a15efafa915
| 10 |
XRD data were collected in a step-scan mode at room temperature using a Rigaku SmartLab SE diffractometer operated at 40 kV (Cu Kα radiation, 10-80° 2θ range, step width 0.01°). The XRD spectra were analyzed using X'Pert HighScore Plus v2.2e software. The International Centre for Diffraction Data (ICDD) PDF-2 Release 2009 database was used for crystalline phase identification. The HP-NCs dispersed in Sol. 1, 2, and toluene were dried in a vacuum oven at 50ºC for 4 hours before XRD measurement.
|
6707a95c51558a15efafa915
| 11 |
The optical properties of the CsPbBr3 and Cs4PbBr6 NCs were studied by analyzing the absorption and PL. The absorption was measured by a Jasco V-750 spectrophotometer equipped with an integrated sphere within the 300-800 nm wavelength range. The PL emission spectra for a 320 nm excitation wavelength in the 450-650 nm range were measured with Jasco (FP-8350).
|
6707a95c51558a15efafa915
| 12 |
The EPR measurements are carried out at room temperature using a Bruker ELEXYS E500 spectrometer operating at X-band frequencies (9.5 GHz) and a Bruker ER4119HS resonator. The spin trap solution (25 mM) was prepared by first dissolving 50 mg of BMPO (MW = 199.25, Dojindo Molecular Technologies, Inc.) in 2.5 mL of absolute EtOH (99.5 %, Baker HPLC Analyzed, J.T.Baker). Then, 7.5 mL of toluene (99.5 %) was added. For the particle's solution, the HPs powder was first dispersed in toluene by sonication for 3 min. For the measurements, 40 μL of HPs particle solution was mixed with 40 μL of the spin trap solution (solution 1 or Sol. 1), 12.5 % ethanol) using a vortex. For the samples with 42 % ethanol, 40 μL of EtOH was added to solution 1 (solution 2 or Sol. 2). Samples were loaded into Vitrocom quartz capillaries, CV1012-Q-100, with a 1 mm inner diameter. The experimental conditions of the EPR spectral analysis were as follows: microwave power of 20 mW, 0.1 mT modulation amplitude, and 100 kHz modulation frequency. The sweep range was 10 mT, and the spectra consisted of 512 points. The samples were irradiated under white light using a SCHOTT KL 2500 LED light source with a maximum light intensity of 1,100 lm. The tested samples were continuously illuminated during the experiment, and the EPR spectra were taken every 3 min. The fitting of the EPR spectra was done with winsim2002 software. The plots with the fitting parameters were done with the assistance of MATLAB R2021a and easy spin-5.2.
|
6707a95c51558a15efafa915
| 13 |
The CsPbBr3 and Cs4PbBr6 NCs dispersed in toluene were examined via transmission electron microscopy (TEM) (Figures and). The average diameter of the CsPbBr3 and Cs4PbBr6 NCs is 8 ± 1.09 nm and 14 ± 3.06 nm, respectively. The calculated particle size distribution is presented in Figures and, Supporting Information (SI). The lattice fringes spacing is 0.26 nm and 0.40 nm, corresponding to the (301) and (104) crystallographic planes of orthorhombic CsPbBr3 and rhombohedral Cs4PbBr6, accordingly (insets of Figures and). The X-ray diffraction (XRD) patterns of the CsPbBr3 and Cs4PbBr6 NCs are presented in Figure . For CsPbBr3 NCs, the diffraction pattern corresponds to the orthorhombic phase. For Cs4PbBr6 NCs, the XRD pattern corresponds mainly to the rhombohedral phase. However, a small amount of the cubic CsPbBr3 phase is present, as indicated by the peaks at 22.42˚. This coexisting phase of CsPbBr3 in Cs4PbBr6 was previously reported. To understand the impact of polar solvents on the structural integrity and morphology of CsPbBr3 and Cs4PbBr6, we disperse the NCs in the EtOH/Toluene mixture. Here, the NCs were dispersed in two different concentrations: EtOH: Toluene (1:7) and (1:1.4) ratio (named solutions 1 (Sol. 1) and 2 (Sol. 2), respectively). The XRD pattern for CsPbBr3 and Cs4PbBr6 NCs in Sol. 1 and 2 remained the same (Figure ). However, the peak intensities for the NCs dispersed in Sol. 2 were significantly lower. This phenomenon can be ascribed to the partial decomposition of the NCs in the presence of a high concentration of EtOH. The TEM images presented in Figure (SI) illustrate that the morphology of the NCs dispersed in Sol. 1 remains consistent with those dispersed in pure toluene. Contrarily, the NCs dispersed in Sol. 2 are mostly decomposed due to the high EtOH concentration, SI Figures .
|
6707a95c51558a15efafa915
| 14 |
To study the optical properties of CsPbBr3 and Cs4PbBr6 NCs, absorption and photoluminescence (PL) measurements were performed on NCs dispersed in toluene (Figure ). The absorption edge of CsPbBr3 NCs appears at 505 nm, whereas for Cs4PbBr6 NCs, we observe absorption edges at 314 and 506 nm. According to the previous reports, the edge at 505 and 314 nm can be ascribed to the optical band gap absorption of CsPbBr3 and Cs4PbBr6 NCs. The additional absorption edge at 506 nm for Cs4PbBr6 NCs can be ascribed to the coexisting phase of the CsPbBr3. This coexisting phase of CsPbBr3 is also supported by the XRD analysis, as presented in Figure . For both NCs, a weak shoulder-like peak can be observed around 360 nm and is assigned to the lead bromide complex in the CsPbBr3 and Cs4PbBr6 lattice. This lead bromide complex was not detected in XRD analysis, possibly due to its low concentration. The band gap was calculated using a Tauc -equation, and the results are presented in Figures and. The calculated band gap for CsPbBr3 and Cs4PbBr6 are 2.38 and 3.53eV, respectively, which coincides with the previous report. The PL emission peak is at 517 and 520 nm for CsPbBr3 and Cs4PbBr6 NCs, respectively (Figures and). The CsPbBr3 and Cs4PbBr6 NCs exhibit an intense and narrower PL emission peak. The CsPbBr3 exhibits an emission peak with the fullwidth half maxima (FWHM) of ~17 nm, indicating a homogeneous size distribution of NCs. The broader FWHM of Cs4PbBr6 (~21 nm) is due to the larger particle sizes and wider size distribution than the CsPbBr3 NCs. The impact of the solvent on the optoelectronic properties of the NCs was studied by the absorption and PL measurement for the NCs dispersed in Sol. 1 and 2 (Figures ). The results of absorption and PL for CsPbBr3 and Cs4PbBr6 NCs in Sol. 1 and 2 are summarized in Table . The absorption edge and PL peak of CsPbBr3 and Cs4PbBr6 NCs in Sol. 1 and 2 exhibited slight shifts compared to toluene. These shifts can be attributed to the change in the excitonic binding energy, which is induced by the change in the dielectric constant of the solution when the EtOH concentration is increased. The change in the polarity of the solution also induces FWHM broadening and a significant decrease in PL intensity, indicating the decomposition of the NCs. Indeed, as the polarity of the solvent increases, the ligands are detached from the NCs more easily, compromising their structural integrity. Hence, when the polarity increases, the NCs are destroyed much quicker.
|
6707a95c51558a15efafa915
| 15 |
Since the PL measurement indicates the decomposition of the NCs in Sol. 1 and 2, we proceeded to examine the stability of CsPbBr3 and Cs4PbBr6 NCs held in the dark and after the illumination. Note that we have not investigated the stability of CsPbBr3 and Cs4PbBr6 NCs in toluene, as previous reports have consistently shown that the HPs-NCs are stable in toluene. For Sol.
|
6707a95c51558a15efafa915
| 16 |
1 and 2, the PL was recorded at 10-min intervals, and the findings are illustrated in Figure . In the dark, the CsPbBr3 NCs decomposed by ~50% in both low and high EtOH concentrations after 30 min. When exposed to light for 30 min, CsPbBr3 underwent significant degradation, reaching 85% and 100% decomposition in Sol. 1 and 2, respectively. Interestingly, Cs4PbBr6 NCs displayed a mere 20% decomposition in darkness across both solutions. Furthermore, when illuminated for 30 min, Cs4PbBr6 decomposed by only 35% and 45% in Sol. 1 and 2, respectively. This observation suggests that Cs4PbBr6 decomposes at a slower rate compared to CsPbBr3 in polar solvents. The decomposition process of the NCs is discussed in detail in the subsequent EPR section. According to the optical absorption, TEM, and XRD, the NCs exhibit similar behavior in Sol. 1 and toluene. Thus, we consider only Sol.1 and 2 (and not toluene) for further discussion.
|
6707a95c51558a15efafa915
| 17 |
To substantiate and better understand the decomposition process, we performed an EPR analysis on CsPbBr3 and Cs4PbBr6 NCs in Sol. 1 and 2 with and without illumination (Figure ). Spin trapping EPR spectroscopy effectively detects reactive oxygen species like • OH and O2 •-in biological and chemical contexts despite their short half-lives. Using a spin trap like BMPO, these short-lived radicals can be detected by forming stable adducts. In this study, BMPO was added to the HP-NCs solution for radical detection. An excess amount of BMPO was maintained in both solutions to ensure all produced radicals were effectively trapped. The measurements were performed in the dark and under illumination (by tungsten lamp) at room temperature. To avoid any misinterpretations, we first conducted a series of reference measurements. Specifically, we assessed the rate of radical production in pure solutions (without NCs), and we examined the solutions with the NCs in the dark to understand the effects of solution polarity versus light on the decomposition of the NCs.
|
6707a95c51558a15efafa915
| 18 |
As a reference, we tested Sol. 1 and 2 without HPs, both in the dark and under illumination, to confirm that solvent reactions do not generate radicals that could influence the decomposition of CsPbBr₃ and Cs₄PbBr₆ NCs (Figures and). Only a small amount of radicals formed after 24 min of illumination, allowing us to attribute our results primarily to radicals produced by the HPs under illumination.
|
6707a95c51558a15efafa915
| 19 |
To isolate the effect of light on the decomposition of HP-NCs, we analyzed samples kept in the dark for 40 min before illumination (Figure ). The NCs exhibited a weak EPR signal in the dark, indicating that HPs produce only a small amount of electrons prior to illumination. However, the EPR signal was significantly stronger under illumination (Figure ). These findings demonstrate an apparent effect of light on the decomposition of the HPs.
|
6707a95c51558a15efafa915
| 20 |
During the first 3 min under light, HPs produce a significant number of electrons that are trapped by BMPO (Figure ). The EPR spectra of HPs in both solutions exhibit typical BMPO/•OH or BMPO/•OOH signals under illumination conditions. As the illumination time increases, we observe a reduction and change in the EPR signal, indicating the formation of unconventional Br, Cs, and Pb-related adducts due to the decomposition of HPs (Figures , Tables ).
|
6707a95c51558a15efafa915
| 21 |
As mentioned above, the EPR signal obtained after 3 min of illumination suggests that it primarily originated from either •OH or •OOH species. In order to distinguish whether the signal is induced by pure •OH, •OOH species, we employ a dimethyl sulfoxide (DMSO) scavenger. The DMSO traps mainly the OH -radicals. Here, we use Sol. 1 with the addition of 10 μL of DMSO aimed to scavenge the OH -radicals (Figure ). At the 3 min of illumination, we note comparable signals measured with BMPO and the DMSO (Figure vs Figure ). The similarities in the main features of the spectra suggest that the species responsible for the EPR signal are primarily
|
6707a95c51558a15efafa915
| 22 |
•OOH adducts. Although the general behavior of the examined HPs-NCs was similar, there are still significant differences in the decomposition rate and intensity of radical formation. These parameters are also influenced by the polarity of the solvent. In the following section, we will compare the behavior of CsPbBr3 and Cs4PbBr6 NCs in Sol 1 and 2.
|
6707a95c51558a15efafa915
| 23 |
Under 3 min of illumination, the CsPbBr3 NCs produced a significant number of electrons trapped by BMPO, resulting in a maximum EPR signal in both solutions. The signal intensities are 22.8 and 19.1 times higher than the reference measurement in the dark for Sol. 1 and 2, respectively (Figure ). After 6 min of illumination, the EPR signal is changed, and its intensity has decreased by ~45%. This significant signal decay may be caused by the decomposition of the NCs.
|
6707a95c51558a15efafa915
| 24 |
The Cs4PbBr6 NCs also produce a strong EPR signal upon illumination, but the signal remains intense up to 18 and 12 min in Sol. 1 and 2, respectively. This finding indicates the increased stability of Cs₄PbBr₆ compared to CsPbBr₃, with Cs₄PbBr₆ being three times more stable than the CsPbBr₃ NCs in Sol. 1. Interestingly, in Sol. 2, both the CsPbBr3 and Cs4PbBr6 NCs signal intensity starts to decrease faster than in Sol. 1. This behavior can be attributed to the faster decomposition of HPs in high concentrations of EtOH under the illumination conditions.
|
6707a95c51558a15efafa915
| 25 |
We performed a fitting of the EPR spectra composed of radical species to understand the effect of the CsPbBr3 and Cs4PbBr6 NCs decomposition products on their shape. For all the fittings for CsPbBr3 and Cs4PbBr6 NCs, we consider a contribution of the BMPO/•OOH adducts. The summarized results, i.e., the spins, hyperfine values, and the areas used for the fitting, are presented in Tables S1-S4. The results of the fitting confirm the presence of Br, Cs, and Pb-related radical species, which indicates the decomposition of the CsPbBr3 and Cs4PbBr6 NCs. However, CsPbBr3
|
6707a95c51558a15efafa915
| 26 |
Notably, the analysis of the fitting results is very technical; thus, in the following, we will only discuss the main findings by analyzing the hyperfine values (a). The hyperfine is the parameter that quantifies the interaction between electron spins with the magnetic nuclei. The value (a) is directly related to the distance between the spectral lines, which was calculated from the EPR spectrum and thus can serve as an indication for the decomposition.
|
6707a95c51558a15efafa915
| 27 |
After 3 min of illumination, the EPR spectra of CsPbBr3 NCs can be fitted with the contribution of Br and Cs radical species in addition to the •O2 -and without the contribution of Pb (Tables and). The hyperfine (a) value after 6 min of illumination indicates that the influence of Cs is significant in both Sol. 1 and 2, i.e., aCs = 1.457 and aCs = 1.393, respectively (Tables and).
|
6707a95c51558a15efafa915
| 28 |
After 6 min of illumination, the fitting has to be done with consideration of the contribution of Pb, which indicates increased decomposition. It is important to note that the nuclear spin of H is like the nuclear spin value of Pb (spin = ½). Hence, there is a possibility that some of the contributions that we see at the beginning of the HPs decomposition are from the conventional BMPO/•OH and BMPO/•OOH adducts and not only from Pb. Similar EPR analysis of the Cs4PbBr6 NCs spectra reveals that the decomposition starts only after 18 min of illumination, and the signals of Cs, Pb, and Br can be detected. Notably, the Br-related radical species contribute substantially to the EPR spectra for both solutions, i.e., aBr is 0.517 for Sol. 1 and 0.677 for Sol. 2 (Tables and).
|
6707a95c51558a15efafa915
| 29 |
The separated contributions of H, Br, Cs, and Pb radical species resulted from the fitting of the measured EPR spectra after 24 min of illumination for both CsPbBr3 and Cs4PbBr6 NCs in Sol. 1 and 2 are presented in Figure . As the illumination time increases, the signal weakens, which indicates that the radicals produced due to the decomposition of HPs diminish. We find that after 24 min, both CsPbBr3 and Cs4PbBr6 NCs are decomposed to a high degree, as indicated by the shape and the intensity of the EPR spectra.
|
6707a95c51558a15efafa915
| 30 |
To further evaluate the decomposition rate, we follow the progress of the HP-NCs decomposition in the kinetics spectra measured for the 3345 G peak. Reduction in the peak intensity is related to a decreased amount of the generated radicals, and a shift of the peak indicates a change in the shape of the EPR spectra. The EPR spectra before and after the kinetics analyses are presented in Figure . We find that the decomposition of CsPbBr3 NCs starts after ~3 min of illumination, and after 18 min, the number of radicals reduces to about 7 % for both solutions (Figure ).
|
6707a95c51558a15efafa915
| 31 |
Interestingly, we can still observe some radicals even after 42 min of measurements (Figures S10a CsPbBr3 NCs. This is indicated by the rapid rise in the number of radicals for CsPbBr3 compared to Cs4PbBr6 NCs (Figure ). Besides the slower radical generation in Cs4PbBr6 NCs, their higher stability over time is reflected in their slower decay rate. activity of CsPbBr3 vs Cs4PbBr6 NCs is supported by the EPR analysis, where we find that the radical formation of CsPbBr3 is faster (Figures and). Moreover, using EPR kinetic analysis, we found that the decomposition of the NCs accompanies the radical formation, and this process is also much faster for CsPbBr3 NCs. Importantly, for both CsPbBr3 and Cs4PbBr6 NCs, the photocatalytic activity can be attributed to their role as sacrificial agents.
|
6707a95c51558a15efafa915
| 32 |
To derive the mechanism of HP-NCs decomposition, let's consider the radical formation upon illumination, as shown by EPR measurements. In HPs, photogenerated electrons in the conduction band (CB) can be absorbed by oxygen, leading to the formation of •O2 free radicals. Similarly, photogenerated holes in the valence band (VB) can convert OH-ions to •OH radicals. Therefore, we can infer the decomposition mechanism is based on photoinduced radical production in HPs, as described in Table .
|
653f76e3a8b423585a8fdd5c
| 0 |
Artificial metalloenzymes (ArMs) represent an avenue to new-to-nature reactions . These systems expand the biocatalytic toolbox by using transition metal complexes tethered to protein scaffolds; notable examples of chemical transformations using ArMs include transfer hydrogenation , hydroformylation , in vivo metathesis , lignin oxidation , Friedel-Crafts alkylation and other cross-coupling reactions .
|
653f76e3a8b423585a8fdd5c
| 1 |
Site-specific coordination of synthetic metal complexes to proteins plays a crucial role in the development of stereoselective artificial metalloenzymes (ArMs), wherein the protein scaffold forms a secondary coordination sphere around the reactive centre providing stereocontrol . To synthesise ArMs, researchers have used many different approaches from simply leveraging native metal-binding activity found in certain proteins to combining synthetic chemistry and proteins, including using variants of natural cofactors, such as the iron-binding protoporphyrin IX, with alternative transition metals . More synthetic approaches to sitespecific incorporation of metal complexes include the utilisation of supramolecular binding, with several different tethered transition metal complex systems explored to date, the most notable being the biotin-streptavidin (Sav) system . Bioconjugation of metal complexes via covalent attachment with unique reactive amino acid residues such as cysteine or azidophenylalanine is another widely used approach to prepare ArMs . In vivo ligand incorporation offers an attractive alternative where metal-binding unnatural amino acids such as (2,2'-bipyridin-5-yl)alanine (BpyAla) can be selectively introduced to the protein scaffold directly using genetic code expansion technologies, allowing for more streamlined enzyme engineering .
|
653f76e3a8b423585a8fdd5c
| 2 |
Within these different approaches to ArM design, several scaffolds have been repeatedly used with various attachment strategies for transformation into novel catalysts. These scaffolds have been described as 'privileged' scaffolds, in a manner analogous with privileged ligands in homogeneous catalysis. The most prominent examples are the LmrR scaffold, which allows supramolecular, bioconjugation, and incorporation of unnatural amino acids 21 , and Sav, which has been explored in dative coordination , alongside the more common supramolecular approach . Despite these advances, few comparative studies have been conducted which included direct comparisons of the modification strategy for site-selective metal coordination. This makes it difficult to discern the role of protein modification upon ArM reactivity and predict the most effective route to ArM assembly for a desired application.
|
653f76e3a8b423585a8fdd5c
| 3 |
The Kamer and Jarvis groups have extensively studied the human steroid carrier protein (SCP-2L) , a single-domain 13.5 kDa protein containing a hydrophobic tunnel, as a scaffold for ArM design. It has been exploited in the design of ArMs for selective hydroformylation using Rhphosphine organometallic complexes , which showed high activity and selectivity in the production of long chain linear aldehydes, under aqueous conditions. Moreover, engineering the scaffold for improved thermostability incurred a five-fold increase in TON compared to the wild type . This protein has also been used in ArMs for the selective oxidation of lignin model compounds , for enantioselective Cu-catalysed Diels-Alder reactions , and artificial photoenzyme design . Other carrier proteins such as adipocyte lipid binding protein (ALBP) , ferric hydroxamate uptake protein component A (FhuA) , and maltose binding protein have been used in the design of ArMs suggesting that proteins with a carrier function may also serve as privileged scaffolds. In this work, the potential of SCP-2L to serve as a privileged protein scaffold is explored, through studying the copper-catalysed 1,4-addition of indole to enones. Bpy is a well-studied ligand for transition metals and is utilised in the coordination sphere of many catalytic complexes 29 . It can be introduced into a protein via the bioconjugation of maleimide or bromomethylbipyridines at cysteine residues (Figure ). Bpy is also one of very few metal ligands that has been incorporated into the genetic code by amber stop codon suppression. Here two methods for site-selective incorporation of bipyridine into the SCP-2L scaffold were compared, and the copper-catalysed 1,4-addition of indole to enones was used to analyse ArM activity. Differences in stereoselectivities were identified with results rationalised using ArM crystal structures and DFT simulations.
|
653f76e3a8b423585a8fdd5c
| 4 |
The steroid carrier protein type 2 like (SCP-2L) domain of the human multifunctional enzyme 2 (MFE-2) has previously been used as a protein scaffold for both regioselective and enantioselective reactions using covalent modification through the introduction of unique cysteine residues . A number of residues in the scaffold's apolar tunnel (V83, A100, Q111) have previously been identified as being amenable to mutation and modification . Their location along the apolar tunnel provides an ideal starting point for the introduction of bipyridine into the protein scaffold to create an active site within a protein pocket. Three different sterol carrier proteins, each containing a unique cysteine at position Q111C, A100C or V83C, were prepared as described previously . Briefly, pEHISTEV plasmids containing the SCP gene inserted after an N-terminal His6 tag with TEV cleavage site were transformed into Rosetta DE3 E. coli cells. His6-tagged proteins were expressed in production broth media and purified by Ni-affinity chromatography with the His6 tag subsequently removed using TEV protease; a further purification step by Ni-affinity chromatography gave the final proteins in good yields (20-70 mgL -1 of media). Bipyridine was introduced into the protein scaffold through bioconjugation of the cysteine residues with 10 equivalents of 5-bromomethyl-2,2'bipyridine 1 in HEPES buffer at pH 8 (Figure ). The optimal conversion was approximately 70-85% modified protein, with minimal secondary modification (<5 % secondary modification; the remaining mass balance is unreacted protein, see Figure and ESIa Fig. ).
|
653f76e3a8b423585a8fdd5c
| 5 |
In comparison to bioconjugation, genetic code expansion using stop codon suppression provides a method of directly incorporating Bpy as the amino acid BpyAla 2 during protein expression (Figure ) . The synthesis of BpyAla was optimised to allow BpyAla production on scales of 10 grams without the need for column chromatography (see ESIa for full details).
|
653f76e3a8b423585a8fdd5c
| 6 |
The SCP-2L gene was codon-optimised for E. coli and prepared by gene synthesis. Amber stop codons were introduced at A100, V83 or Q111 by site-directed mutagenesis. E. coli BL21 (DE3) cells were co-transformed with the pEVOL-BpyAla plasmid, which contains the orthogonal MjTyrRS/MjtRNA Tyr genes , and the pET28 plasmid carrying the SCP gene with a C-terminal TEV cleavable His6 tag. The genes were expressed in the presence of 0.5 mM BpyAla and the resulting proteins purified by Ni-affinity chromatography with the His6 tag removed using TEVprotease. Typical yields were 5-15 mg L -1 in LB media, representing a decrease of 20-80% compared to the yields of SCP-2L without BpyAla. Proteins containing unnatural amino acids are known to have lower expression yields, due to incomplete stop codon supression giving truncated protein. Mass spectrometry on the purified proteins confirmed the successful incorporation of BpyAla at residue positions 100, 111, or 83 (Figure and ESIa Fig. ). The mass of the BpyAla proteins is not 46 Da lower than the SCP CBpy proteins as might be expected due to the absence of the -CH2S-linker, but higher due to the presence of additional amino acids from the TEV site remaining at the C-terminus.).
|
653f76e3a8b423585a8fdd5c
| 7 |
To obtain the Cu-metalloproteins, one equivalent of Cu(NO3)2 was added directly to the Bpycontaining SCP proteins. Copper binding was confirmed by UV-Vis spectroscopy which clearly showed a red shift in the π-π* transition of the Bpy ligand in the presence of Cu(II), consistent with previous reports on copper binding to Bpy-containing proteins . Titration experiments confirmed that, at the concentrations of interest (20-100 µM), Cu(II) bound to the Bpycontaining SCP proteins with ~1:1 stoichiometry (see ESIa, Figures and). Copper binding was also confirmed by ICP-MS analysis (see ESIa).
|
653f76e3a8b423585a8fdd5c
| 8 |
Whilst, the UV-vis spectroscopy and ICP-MS studies showed copper binding to the ArMs (see ESIa), they give no information about the precise environment of the copper. We therefore looked to X-ray crystallography to obtain structural understanding of the newly created active sites within the metalloproteins. Crystals of SCP_Q111CBpy and SCP_Q111BpyAla were soaked in solutions of Cu(II) ions and their structures determined by X-ray crystallography (Figure and). Triton-X-100 was needed for crystallisation, similar to wt SCP-2L (PDB ID: 1IKT) ,the apolar tunnel is occupied by Triton X-100 in the structures of Cu(II)-bound SCP_Q111CBpy and SCP_Q111BpyAla (Figure ). The structure of Cu(II)-bound SCP_Q111CBpy was determined to 1.52 Å resolution, and clearly confirmed the incorporation of 2,2'-bipyridine at C111, which is positioned at the centre of the C-terminal helix,5. The bipyridine rings are sandwiched in a parallel fashion between the amide side-chain atoms of Q108 (one turn up on -helix 5) and Q90 on -helix 4 (Figure ). The outer and inner pyridine rings are 3.4 Å from Q108 and Q90 respectively, a distance that favours formation of stabilising aromatic -amino electrostatic interactions. Helix 5 is slightly shifted relative to its position in the wt SCP_2L structure (PDB ID 1IKT) (Figure in ESIa) and is more flexible than the remainder of the structure, with B-factors ranging from 52-96 Å 2 from the N-to the C-terminus of the helix (N103-L120), compared to 75 Å 2 for the protein overall. The B-factor of the Bpy moiety is 92 Å 2 , suggesting flexibility in its position. Additional electron density was observed in the 2Fo-Fc map around the solvent-exposed face of the Bpy moiety, and a single peak in an anomalous difference map confirmed this is a bound Cu(II) ion (Figure ). The Cu(II) atom lies in the same plane as the two pyridine rings and is 2 Å from the two pyridine ring nitrogens, and thus we assume that the Cu(II) ion adopts an octahedral coordination geometry. The high B-factor (108 Å) of the Cu(II) is consistent with its position in the flexible -helix 5. Only one water molecule could be modelled around the Cu(II) ion, but the positions of Q108 and Q90 suggest that their amide side-chain atoms may play a role in coordinating the other (unmodelled) waters in the coordination sphere.
|
653f76e3a8b423585a8fdd5c
| 9 |
The structure of Cu(II)-bound SCP_Q111BpyAla was determined to 2.51 Å resolution (Figure and) and revealed a different orientation of the bipyridine moiety compared to that seen in the Cu(II)-bound SCP_Q111CBpy structure. Despite the lower resolution, the electron density for the BpyAla bipyridine rings, -helix 5 and the additional C-terminal amino acids (120-128) was very clear, indicative of a more rigid conformation for these parts of the protein.
|
653f76e3a8b423585a8fdd5c
| 10 |
The co-planar BpyAla bipyridine rings are wedged into a shallow pocket formed by the turn between G85 and P89 at the edge of the apolar tunnel. This conformation is stabilised by van der Waals interactions between C1 of BpyAla and both G85 H (3.35 Å) and D88 H (3.36 Å), C2 of BpyAla and H of P89 (2.87 Å: 3.36 Å to the C) and C3 of BpyAla and H of P89 (2.74 Å: 3.26 Å to the C) (Figure ). In addition, the side-chains of D88 and K115 (one turn down on -helix 5) lie on either face of the bipyridine and may stabilise the BpyAla conformation by aromatic- electrostatic interactions.
|
653f76e3a8b423585a8fdd5c
| 11 |
The 2Fo-Fc electron density map showed a single intense peak adjacent to the BpyAla bipyridine corresponding to a Cu(II) ion coordinated to N1 and N2 of BpyAla (Cu-N bond distance of 2 Å) and four water molecules in an octahedral arrangement (Figure in ESIa). The B-factors for the Bpy moiety and the Cu(II) ion (43 Å 2 and 57.91 Å 2 , respectively) were lower in this structure compared to the bipyridine rings in the Cu(II)-bound Q111CBipy structure, indicating that the catalytic environment is more rigid.
|
653f76e3a8b423585a8fdd5c
| 12 |
Taken together the structural analyses confirm the incorporation of Bpy at position 111 and the binding of Cu(II) to the Bpy moieties in both SCP ArMs. They also revealed marked differences in the environments of the Bpy and bound Cu(II) ions in the SCP_Q111BpyAla and SCP_Q111CBpy structures. Moreover, the Bpy moiety at position 111 is more rigid when incorporated as the unnatural amino acid BpyAla than via bioconjugation at a cysteine.
|
653f76e3a8b423585a8fdd5c
| 13 |
The catalytic activities of Bpy-containing SCP metalloenzymes were evaluated in the Cucatalysed Friedel-Crafts alkylation of 5-methoxy-1H-indole 3 with 1-(1-methyl-1H-imidazol-2yl)but-2-en-1-one 4, resulting in the product 5, the benchmark reaction for Cu(II) catalysis (Table ) . Enantioselective product formation was only observed in the presence of the Bpycontaining SCP-metalloenzymes. SCP ArMs with BpyAla showed a higher yield (42-45%) compared to the equivalent Cyscoupled Bpy SCP ArMs. Among the novel ArMs, SCP_Q111BpyAla displayed the greatest enantioselectivity with an e.r. of 20:80 (±1), favouring the S enantiomer, and a yield of 42% (Table , entry 9). The BpyAla moiety in this SCP ArM is situated at the centre of one side of the tunnel, in the middle of -helix 5 (Figure and 3A). By contrast, the ArMs with BpyAla situated at either end of the tunnel showed either no enantioselectivity in the case of SCP_A100BpyAla (Table , entry 7), or lower enantioselectivity with a preference for the R enantiomer (Table , entry 8). We hypothesise that the lower selectivity at the entries to the tunnel could be due to increased flexibility and space leading to less defined binding pocket, compared to position Q111 in the centre of the tunnel. While enantioselectivity towards the R enantiomer was observed with all three Cys mutants, it was limited with the best result observed for SCP_A100CBpy which gave a 66:34 ratio (Table , entries 4-6).
|
653f76e3a8b423585a8fdd5c
| 14 |
Both enantiomers of 5 can be obtained in the Friedel-Crafts alkylation using SCP-ArMs with Bpy attached to residue 111, but by using different attachment strategies (Table , entries 6 and 9). The crystal structures of SCP_Q111BpyAla and SCP_Q111CBpy provide a structural indication for the observed differences in the enantioselectivities of these ArMs [almost 2x higher e.e., from 64:36 (R) for SCP_Q111CBpy to 20:80 (S) for SCP_Q111BpyAla]. The longer linker to the protein backbone in SCP_Q111CBpy, which has an additional -CH2-S-compared to SCP_Q111BpyAla, allows it to extend out from -helix 5 and adopt a more exposed position on the surface of the protein, compared to the BpyAla which is closer to the scaffold protein.
|
653f76e3a8b423585a8fdd5c
| 15 |
In order to gain more information on how the ArMs control enantioselectivity, computational modelling was used to probe the stereochemical outcomes for the Cu-catalysed Friedel-Crafts reaction. The crystal structure coordinates were used for the computations. As these included Triton-X100, the ArM-catalysed Friedel-Crafts reactions were repeated in the presence of 2 eq. Triton-X100 to ascertain if this was likely to change the catalysis and thus if the crystal structures are a reasonable structural starting point. In the case of SCP_Q111BpyAla and SCP_Q111CBpy, the same enantioselectivities were observed as reactions without Triton-X100, with only a slight decrease in yield suggesting that the structures are a valid model to use to rationalise the differences in enantioselectivity. However, for SCP_A100BpyAla and SCP_V83BpyAla there was a significant decrease in activity, which could be caused by Triton-X100 blocking the availability of Cu-Bpy to the substrates as it binds within the protein tunnel (Table , entries 10-13) .
|
653f76e3a8b423585a8fdd5c
| 16 |
The first step of the reaction is the conjugate addition of indole 3 to enone 4 to form intermediate (Int1), via a transition state (TS1) (Scheme S1 ESIb). The second step is the product formation by the protonation reaction via a second transition state (TS2). To investigate the enantioselectivity of the Friedel-Craft alkylation, we looked to prior work using cluster models for metalloenzymes-catalyzed reactions . The same QM-cluster model technique along with density functional theory (DFT) methodology was utilised to study our reaction. Three active site models were created for analysis (Figure ESIb). Model A represents a reaction catalysed by Cu(II)-2,2ꞌ-bipyridine, in the absence of protein, whereas models B and C represent reactions catalysed by Cu(II)-bound SCP_Q111BpyAla and SCP_Q111CBpy, respectively. The computational analysis on Model A (Figure ESIb) showed that the conjugate coupling step is both the rate-determining step as well as the enantioselective step, and the formation of the keto product is most likely. This agrees well with the experimental results (Table ) as well as previous studies , therefore only the first step was modelled going forwards. To create models B and C, the indole 3 and enone 4 substrates, along with Cu(II) ions, were incorporated into the crystal structures of Cu(II)-bound SCP_Q111BpyAla and SCP_Q111CBpy, respectively. The PDB files of the apo SCP_ArMs were prepared as described in the ESIb, and 3 and 4 were docked using AutoDock Vina , with the lowest energy docked poses used for molecular dynamics (MD) simulations (see ESIb for detailed methods and results). Next, the Friedel-Craft (FC) alkylation reaction catalysed by SCP_Q111BpyAla, using ReproS,B and ReproR,B reactants as the starting structures, was calculated. Firstly, the conjugate addition of indole 3 to enone 4, where substrate 3 configuration is either proS or proR was tested. The C-C coupling transition states (TS1proS,B and TS1proR,B) of 3 with 4 leads to an intermediate for each configuration: Int1proS,B and Int1proR,B respectively. The calculated potential energy landscape for the conjugate coupling reaction for Model B is represented, along with the optimized geometric structures of these transition states, in Figure . The energy barrier for TS1proS,B (10.1 kcal mol 1 ) is lower than TS1proR,B (15.5 kcal mol 1 ) (values relative to the stable configuration ReProS,B). The transition states were characterized by an imaginary frequency i308 cm 1 and i319 cm 1 for TS1proS,B and TS1proR,B respectively. Furthermore, in transition state TS1proS,B, indole 3 has strong water bridge interactions with D88 that make it more stable than the TS1proR,B, which has a week hydrogen bonding interaction with M112. Subsequently, the transition states are relaxed to their respective intermediates (Int1proS,B and Int1proR,B), which characterized by lower relative energy values: 1.65 and 7.92 kcal mol 1 , Thus, the proS configuration pathway is energetically favoured over the proR configuration pathway. Taken together these analyses explain the preferred enantioselectivity of SCP_Q111BpyAla (e.r. 14:86, R:S) that we observed experimentally (Table ). In a similar manner, the FC alkylation reaction mechanism catalysed by SCP_Q111CBpy (Model C) was explored. The reactant complexes in both proS and proR configurations were selected as the starting structures. The optimized geometries of these reactant complexes ReproS,C and ReproR,C are represented in Figures and. The free energy of the ReproS,C structure is higher than the ReproR,C by 5.6 kcal mol -1 (∆E + ZPE), indicating that the reactant complex of model C in proR configuration shows the strongest substrate-bound pose and hence signifies the favoured reactant orientation.
|
653f76e3a8b423585a8fdd5c
| 17 |
Finally, the FC reaction pathway of Cu(II)-bound SCP_Q111CBpy was explored, using ReproS,C and ReproR,C reactants as the starting structures. The calculated potential energy landscape for the conjugate coupling reaction for model C is represented, along with the optimized geometric structures of these transition states, in Figure . The lowest energy barrier (13.6 kcal mol -1 ) was obtained for the proR C-C coupling transition state (TS1proR,C) while the energy barrier for the proS transition state (TS1proS,C) was higher (15.4 kcal mol -1 ). The transition are characterized by the presence of an imaginary frequencies i288 cm -1 and i273 cm -1 for TS1proR,C and TS1proS,C respectively. Furthermore, in the TS1proR,C transition state a hydrophobic interaction from M105 was observed with 3, while no such interaction of 3 was observed in TS1proS,C. The system then relaxed to the low energy state intermediates Int1proR,C and Int1proS,C, which are characterized in the local minima state by the presence of all real frequency values indicating stable structures. The Int1proR,C has lower energy (6.7 kcal mol -1 ) than the Int1proS,C configuration (9.8 kcal mol -1 ), which favours the proR configuration pathway over the proS pathway. This is consistent with the experimentally observed preference to produce the R enantiomer by SCP_Q111CBpy (Table ).
|
653f76e3a8b423585a8fdd5c
| 18 |
Overall, the substrates' activation in FC alkylation mechanism with SCP_Q111BpyAla favours the formation of proS product over proR while with SCP_Q111CBpy favours proR product formation over proS. However, the products distribution was achieved via a competitive pathway, so a mixture of both enantiomers of the product is predicted, which agrees well with our experimental results. The larger difference in the relative energies for the SCP_Q111BpyAla TS intermediates matches the experimental observation of improved enantioselectivity when using SCP_Q111BpyAla as the catalyst over the use of SCP_Q111CBpy
|
653f76e3a8b423585a8fdd5c
| 19 |
Engineering SCP_Q111BpyAla the most promising ArM (SCP_Q111BpyAla) forwards, we chose to use a structurebased alanine scanning approach to see if any of the residues identified from either the computational or structural work impacted activity or selectivity . Each amino acid was substituted individually with alanine via site-directed mutagenesis, and the yields and enantioselectivities in Friedel-Crafts reactions catalysed by these mutant SCP ArMs were determined (Table ). These reactions were carried out at pH 5, as slightly improved enantiomeric ratios were observed compared to those performed at pH 6 (Table None of the mutations screened showed substantial variations (i.e. a complete drop in reactivity or selectivity). Indeed, the mutation with the biggest difference was F34A which was chosen due to its location nearby and within the protein hydrophobic pocket (Table , entry 4). SCP_Q111BpyAla F34A exhibited substantial precipitation suggesting that disruption to the protein core reduced stability, leading to the observed low activity and selectivity. A similar rationale was also used for the choice of V82A which gave lower activity and selectivity but not to the same extent (Table , entry 3 and 4). The crystal structure suggested that K115 and D88 may play a role in stabilising the bipyridine position within the protein through aromatic- electrostatic interactions. Mutating D88 to gave no meaningful in enantioselectivity (Table , entry 7), whilst K115A showed a reduction in enantioselectivity (Table , entry 6). The more flexible nature of the lysine side chain means it is less clear if this can be attributed to structural changes as opposed to some role with substrate binding. The computational work revealed no interaction of the substrates with K115. In contrast, a close contact between the side chain of D88 and the copper atom was observed in the modelled transition states and D88 was shown to participate in water bonding networks with the substrates. Both these interactions would be disrupted on mutation of D88 to alanine, suggesting that either these interactions are of minor importance in catalysis or alternative residues such as Q108 could replace the hydrogen bonding interactions with water. Whilst Q108A was shown to reduce the enantioselectivity obtained with SCP-Q111BpyAla experimentally (Table , entry 2), no clear role was observed in the computational work.
|
653f76e3a8b423585a8fdd5c
| 20 |
The only amino acid shown to make direct contact with the substrate during the course of the reaction was M112: the methyl makes a weak hydrophobic interaction with 3 during substrate binding in the proS (Figure ), whilst the amide backbone of M112 makes a weak hydrogen bond with 3 in the proR transition state (Figure ). Mutating M112 to alanine gave small reduction in enantioselectivity (Table , entry 5). Whilst alanine can also make hydrophobic interactions, its reduced chain length would preclude interactions with 3 in this instance and thus M112 could be helping to stabilise the substrates to a small extent. No difference would be expected for the proR pathway as alanine can still participate in amide backbone bonding. The lack of side chain interactions with the substrates and no clear results from the alanine scanning suggests that a rational approach to designing the active site via mutagenesis may not lead to improved enzyme. Indeed, the exposed nature of the active site on the side of the protein suggests that extensive backbone engineering to build up the protein bulk around the substrates may lead to more promising candidates for enantioselective catalysis.
|
653f76e3a8b423585a8fdd5c
| 21 |
Using two different approaches to forming artificial metalloenzymes, a library of successful catalysts for the Cu-catalysed Friedel Crafts alkylation were obtained. Moderate activities and good enantiomeric ratios were observed with SCP_Q111BpyAla giving the highest e.r. of 14:86 in favour of the S enantiomer. In general, lower enantioselectivities were observed for the cysteine-modified catalysts and this is attributed to increased flexibility around the active site due to the additional S-CH2 linker. Changing from genetic incorporation of BpyAla at Q111 to bioconjugation of bipyridine through a cysteine, resulted in an unexpected change in the major isomer observed from S to R, albeit with a reduction in enantiomeric excess. This result is intriguing as a major challenge in enzyme catalysis is accessing the opposite product enantiomer. Crystallographic and computational studies confirm that changes in the structure of the active site led to the stabilisation of indole 3 on different sides of the bound enone 4, with the enantiomeric ratios observed correlating with the difference in the calculated transition state energies for the proR and proS SCP_Q111BpyAla and SCP_Q111CBpy Cu catalyst models. Based on these findings, future work should prioritize determining the most suitable modification method for the desired application of the ArM before proceeding with reaction screening and optimization, as subtle differences in protein structure can substantially change the stabilised transition state.
|
643c03bc73c6563f14fae441
| 0 |
Hydrogen production by solar water splitting (SWS) using abundant and eco-friendly photoelectrode materials is very appealing nowadays. The overall efficiency of SWS reaction is directly related to the photoelectrochemical (PEC) activity including both oxidation and reduction reactions that occur at the interface between photoelectrodes and the aqueous electrolyte. Hematite is the archetype semiconducting material used as photoanode, presenting a band gap of 2.15 eV, perfectly matching the solar spectrum for optimized absorption and thus for direct SWS applications. Compared to state-of-the-art oxide semiconductors used for solar water splitting, pristine hematite presents low efficiency because of the reduced hole mean free path (~ 2 -4 nm) and poor surface kinetics related to a complex oxygen evolution reaction (OER). It was demonstrated that surface kinetics during SWS can be tuned by controlling the surface states nature using various approaches: doping, hetero-and nano-structuring, annealing, catalyst film coating, etc. In particular the synergic effect between Ti-doping and induced oxygen vacancies through thermal treatments during or post growth under oxygen depleted atmosphere has shown its efficiency. Various mechanisms were suggested to contribute to the enhancement of PEC activity. Zhao et al. have shown that by annealing pristine and Ti-doped hematite photoanodes in Nitrogen gas at 600°C the photocurrent is increased by 200 % and 67 % respectively. They assigned this enhancement to oxygen vacancies generated by nitrogen treatment, leading to carrier density increase. Opposite behaviors were found for the charge transfer processes, comparing pristine hematite and Ti-doped hematite, suggesting increased transfer resistance for the later one. To the contrary, Wang et al. demonstrated strong electrocatalytic surface contribution of the induced oxygen vacancies along to the expected improvement of the bulk conductivity. Such divergences arise from the variety of the employed sample preparation methods, leading to different sensitivities to subtle surface effects. These conclusions are obtained from PEC macroscopic measurements only, nanoscale information being absent. A perfect demonstration is given by Zhang et al., in the case of highly ordered attached Ti-modified hematite mesocrystals. Their study presents strongly enhanced PEC properties promoted by a double contribution: first, the formation of interfacial oxygen vacancies yielding high carrier density, and second, shorter depletion width (< 10 nm) over large regions through formation of rutile TiO2 at the mesocrystals surface, as determined by high resolution electron microscopy. Indeed, complex hematite-Ti based heterostructures (Fe2O3/Fe2TiO5 or Fe2O3/FeTiO3) are known to be responsible for PEC activity enhancement owing to strongly increased surface charge separation. This leads to increased number of holes injection at the interface with the electrolyte.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.