Comprehensive in silico analysis and molecular dynamics of the superoxide dismutase 1 (SOD1) variants related to amyotrophic lateral sclerosis

Amyotrophic Lateral Sclerosis (ALS) is the most frequent motor neuron disorder, with a significant social and economic burden. ALS remains incurable, and the only drugs approved for its treatments confers a survival benefit of a few months for the patients. Missense mutations in superoxide dismutase 1 (SOD1), a major cytoplasmic antioxidant enzyme, has been associated with ALS development, accounting for 23% of its familial cases and 7% of all sporadic cases. This work aims to characterize in silico the structural and functional effects of SOD1 protein variants. Missense mutations in SOD1 were compiled from the literature and databases. Twelve algorithms were used to predict the functional and stability effects of these mutations. ConSurf was used to estimate the evolutionary conservation of SOD1 amino-acids. GROMACS was used to perform molecular dynamics (MD) simulations of SOD1 wild-type and variants A4V, D90A, H46R, and I113T, which account for approximately half of all ALS-SOD1 cases in the United States, Europe, Japan, and United Kingdom, respectively. 233 missense mutations in SOD1 protein were compiled from the databases and literature consulted. The predictive analyses pointed to an elevated rate of deleterious and destabilizing predictions for the analyzed variants, indicating their harmful effects. The ConSurf analysis suggested that mutations in SOD1 mainly affect conserved and possibly functionally essential amino acids. The MD analyses pointed to flexibility and essential dynamics alterations at the electrostatic and metal-binding loops of variants A4V, D90A, H46R, and I113T that could lead to aberrant interactions triggering toxic protein aggregation. These alterations may have harmful implications for SOD1 and explain their association with ALS. Understanding the effects of SOD1 mutations on protein structure and function facilitates the design of further experiments and provides relevant information on the molecular mechanism of pathology, which may contribute to improvements in existing treatments for ALS.

Editor: Maria Gasset, Consejo Superior de Investigaciones Cientificas, SPAIN Introduction highest atomic resolution (1.07Å) among the SOD1 structures available in PDB [25]. The complete amino-acid sequence of the SOD1 protein was retrieved from the UniProt database [Uni-Prot ID: P00441] [26]. SOD1 contains an 8-stranded beta-barrel motif and several loops. Two loop elements termed metal-binding (amino-acids  and electrostatic loops (aminoacids 122-143) project from the beta-barrel ( Fig 1A) and are important for the SOD1 enzymatic activity [39]. The electrostatic loop is composed of positively charged amino-acids that guide superoxide radicals into the SOD1 active site where the copper ion is located [41]. The metal-binding loop coordinates the binding of Zinc and Copper ions [42,43], which is structurally important for SOD1 [44]. These loops together shape the active site pocket [43].
Two hundred and thirty-three (233) mutations in SOD1 protein were compiled from the databases and literature consulted (S1 Table). As shown in Fig 1B, SOD1 mutations occur all over the protein structure. Thirty-one (31) mutations were found occurring at the metal-binding loop, while 30 mutations were found at the electrostatic loop of SOD1.

Functional and stability prediction analysis
As shown in Fig 2A, all the algorithms used predicted more than 50% of the SOD1 mutations as deleterious. SNPs&GO was the algorithm with the lowest number of deleterious predictions, while PMUT was the algorithm with the highest number of deleterious predictions (Fig 2A). As shown in Fig 2D, 47% of the analyzed variants were predicted as deleterious by all the ten functional prediction algorithms used, while only 1% of them were predicted as neutral by all the algorithms. Moreover, the SOD1 protein variants presented an elevated rate of deleterious predictions (Fig 2D). This analysis also showed that the mutations with the highest rates of deleterious predictions were concentrated in the metal-binding (amino-acids 51-84) and electrostatic (amino-acids 122-143) loops of SOD1 (S2 Table).
According to the SNPEffect4.0, twenty-one variants were predicted to affect the SOD1 aggregation tendency, ten variants were predicted to affect amyloid propensity, and one variant was predicted to affect chaperone binding (Fig 2B). The variants predicted to affect protein aggregation mostly occur in the N and C-terminal portions of SOD1, while the variants predicted to affect amyloid propensity mostly occur in its N-terminal portion (S3 Table).
As shown in Fig 2C, each algorithm used predicted more than 50% of the SOD1 variants as decreasing protein stability. This analysis also showed that 49% of these variants were predicted to decrease protein stability by I-Mutant and Fold-X, while less than 6% were predicted by both algorithms to increase or not affect this feature (S3 Table). The I-Mutant and Fold-X predictions were concordant for 54.5% of the analyzed variants, but only 2.2% of all predictions were opposed (i.e. increase and decrease for the same variant).
The variants A4V, D90A, H46R, and I113T, which are the most frequent ALS-related mutations in SOD1 [15,39], were predicted as deleterious by 90%, 50%, 100%, and 100% of the algorithms used, respectively (S1 Fig). The variants A4V, D90A, and I113T were also predicted to decrease protein stability by Fold-X and I-Mutant, while the variant H46R was only predicted as destabilizing by Fold-X. Furthermore, the A4V variant was predicted to increase protein aggregation. None of these variants were predicted to affect amyloid propensity and chaperone binding (S3 Table).

Evolutionary conservation analysis
The crystallographic structure of wild-type SOD1 (PDB ID: 2C9V) was submitted to the Con-Surf server, which estimated the evolutionary conservation score (ConSurf-score) of each protein's amino acid. ConSurf performs multiple-sequence alignments to compare a given amino acid sequence to its homologous sequences and estimate their evolutionary conservation. Con-Surf is useful for identifying functionally important regions on proteins, as these are usually conserved throughout evolution [37]. The computed ConSurf-score was projected on the protein's surface ( Fig 3A). The SOD1 amino-acids were colored according to their ConSurf-score following a coloring-code scheme, which varies from (1) cyan and variable to (9) maroon and conserved.
As shown in Fig 3B, 59% of the analyzed SOD1 variants occur in conserved positions (i.e. ConSurf-score � 7), while only 17% of them occur in variable positions (i.e. ConSurfscore � 3). This analysis also showed that variants affecting highly conserved positions (i.e. ConSurf-score = 9) mainly occur at the metal-binding (amino-acids 51-84) and electrostatic (amino-acids 122-143) loops of SOD1. Among the most frequent ALS-related mutations in SOD1 [15,39], the variant I113T was the only variant predicted to affect a variable region. The variants A4V and H46R were predicted to affect conserved amino-acids of SOD1, while the variant D90A was predicted to affect an average conserved position (S2 Fig).

Molecular dynamics simulations
MD is a computational method that predicts the time-dependent motion of an atomic system by solving Newton's equations of motion [45]. The MD simulations can be used to reproduce the real behavior of a protein in its environment, which allows the study of relevant biomolecular processes, including conformational change, protein folding, and ligand binding. These simulations can also be used to predict the effects of perturbations such as mutation, phosphorylation, and protonation on the protein's structure [46]. During an MD, the interatomic interactions are calculated at every simulation step and the resulting atomic positions are registered in a trajectory file, which provided detailed information on changes in protein conformation that can be used to assess several structural parameters [19].
We performed MD simulations of the wild-type SOD1 (PDB ID: 2C9V) and its most frequent ALS-related variants, i.e. A4V, D90A, H46R, and I113T [15,39], to better understand the impact of these amino-acid substitutions on protein structure. The following structural Each SOD1 amino-acid is represented as a surface model and colored according to its ConSurf conservation score. The color-coding bar shows the ConSurf coloring scheme, which varies from (1) cyan and variable to (9) maroon and conserved. The amino-acids colored yellow did not receive conservation scores due to insufficient data. (B) The circle chart indicates the percentage of variants affecting amino-acids predicted as variable to conserved. The circle chart is also colored according to the ConSurf coloring scheme.
RMSD is commonly used in MD simulations to measure the spatial differences between a starting structure and its subsequent coordinates computed over time [47]. This parameter is useful to analyze the time-dependent motion of protein structures and to determine their structural convergence throughout the simulation [19]. RMSD values were calculated from the total number of protein conformations computed in the MD trajectories. As shown in Fig 4, triplicates for wild-type SOD1 and its variants presented a similar behavior throughout the simulation, given their respective means and confidence intervals. A sudden increase in RMSD values was observed at the beginning of all simulations. This was followed by the establishment of a plateau in RMSD values after approximately 150ns, which suggests that the protein structures float around average stable conformations [16]. The initial effects of the trajectories, i.e. those occurring until 150ns, were discounted in further analyses for meaningful comparison [19,34].
RMSF is a measure of the structural displacement of an amino-acid from its average position throughout the simulation [16]. RMSF is useful to assess local flexibility in protein structures, allowing the identification of flexible and rigid regions [48]. The RMSF analysis pointed to a flexibility increase at the metal-binding and electrostatic loops of the variants A4V and D90A, in addition to a flexibility decrease in the same regions of H46R ( Fig 5).
Flexibility can also be assessed during an MD simulation by analyzing the B-factor, a temperature-displacement factor. B-factor is a measure of the structural displacement of an amino-acid around its average position due to thermal vibrations [49]. The B-factor values computed for each SOD1 amino-acid were projected on the protein surface, providing an interesting three-dimensional representation of structural flexibility [19]. The B-factor distribution over a protein structure also provides an important indicator of its dynamics [50]. As shown in Fig 6, the variants A4V and D90A presented increased flexibility at their metal-binding and electrostatic loops, while the variant H46R presented decreased flexibility at those regions when compared to the wild-type SOD1. The B-factor analysis suggested flexibility alterations in the variants A4V, D90A, and H46R at regions similar to those found altered in the RMSF analysis.
PCA is also known as essential dynamics (ED) when applied to analyze MD simulations [19]. ED linear transforms the high-dimensional and complex data contained in a molecular trajectory into a low-dimensional space in which the large-scale protein motions occur [51], reducing the number of dimensions needed to describe protein dynamics. This statistical technique systematically filters the observed motions from the largest to smallest spatial-scale in a molecular trajectory using a covariance matrix constructed with the Cartesian coordinates representing atomic positions [52]. ED allows separating the essential from the remaining protein motions. The essential motions (i.e. the largest-scales motions) are usually biologically relevant movements such as opening, closing, and flexing, while the remaining motions describe small irrelevant local fluctuations [20]. The essential motions are often confined into the first two PCA modes (principal components) [53].
The projections for the MD trajectories of wild-type SOD1 and its variants into the subspace spanned by PC1 and PC2 are shown in Fig 7. The ED analysis suggested that the first two principal components (PC1 and PC2) capture the dominant motions, accounting for 62.8%, 73.3%, 70.7%, 35.3%, and 62.1% of the total variance for wild-type SOD1 and its variants A4V, D90A, H46R, and I113T, respectively. As shown in Fig 7, the variants A4V, D90A, and I113T occupied a higher area in conformational space, while the variant H46R occupied a smaller area. Changes in cluster shape were also observed in the conformational space of all  We also analyzed the RMSF contribution of each protein amino-acid for the first two principal components. RMSF values for the PC1 and PC2 projections are shown in Figs 8 and 9, respectively. The RMSF contribution to PC1 pointed to increased essential mobility at the metal-binding loops of the A4V, D90A, and I113T (Fig 8), while the RMSF contribution for the PC2 pointed to essential mobility alterations at the metal-binding and electrostatic loop of the A4V and H46R variants (Fig 9). As shown in Fig 8, these alterations are particularly high at the metal-binding loop of A4V and D90A variants.
Rg is commonly used to measure the structural displacement of protein atoms from their common center of mass throughout the simulation, thus, providing comprehensive information on protein compactness over time [19]. The Rg values computed for the wild-type SOD1 and its variants pointed to an initial moment of structural instability, which is observed in the first half of all simulations (Fig 10). After this time, all the analyzed protein structures presented steady Rg values, which further suggests stable protein folding [54]. This analysis also SASA is a measure of the exposed surface in protein structures that can be accessible to solvent molecules [55]. SASA analysis thus provides relevant information into the protein's degree of exposure to its environment over time [56]. The SASA values computed throughout the simulations (Fig 11) pointed to an unstable behavior at the beginning of all simulations. After approximately 150ns, the SASA values assumed a steady behavior until the end of the MD trajectories. This analysis indicated that the average SASA values for the wild-type SOD1     We also performed an SS analysis to further investigate the structural impact of SOD1 protein variants. The average number (in percentage) of coils, beta-sheets, and alpha-helices formed throughout the simulations was computed and displayed in barplot graphs. As shown in Fig 12, no alterations in the average number of SS were observed for the analyzed variants when compared to the wild-type SOD1.

Discussion
Here, we compiled 233 missense mutations in the SOD1 protein (S1 Table), including thirty mutations that have not yet been registered in the dbSNP database. Forty-seven percent of these variants were classified as deleterious by all the functional prediction algorithms used (Fig 2D), suggesting that they might be harmful to SOD1. Our findings also suggested that missense mutations in SOD1 are most likely to be harmful than neutral to protein function, given the elevated rates of deleterious predictions computed for the variants (Fig 2D). Furthermore, the SOD1 variants with the highest rates of deleterious predictions in the functional prediction analysis mainly affect its metal-binding and electrostatic loops (S1 Fig), which reinforces the functional importance of the amino-acids within these regions.
The functional prediction algorithms used (S2 Fig) presented high accuracy in detecting the known functional impact of the variants A4V, H46R, and I113T [15,39]. However, these algorithms presented low accuracy in predicting the deleterious potential of variant D90A. Thus, as previously shown by our group, using a variety of algorithms is crucial for predicting the functional impact of missense mutations. Especially because these algorithms apply different strategies to make predictions, and no gold standard method is currently established for this purpose [16].
Our findings also suggested that stability reduction is a common feature in SOD1 protein variants, given the elevated rate of destabilizing predictions (forty-nine percent) in the Mutant3.0 and Fold-X analyses (S3 Table). In vitro studies analyzing the stability of over 100 SOD1 protein variants have shown that most of these mutations are destabilizing [57,58].
The divergent results observed at the stability prediction of some SOD1 protein variants (S3 Table) may occur due to the different strategies applied by FoldX and I-Mutant to make predictions. FoldX applies an empirical force field trained in a database of engineered proteins to estimate free-energy changes (ΔΔG) upon mutations and evaluate the stability impact of missense mutations [59], while I-Mutant 2.0 uses Support Vector Machine and information from a database of experimentally determined structures for this purpose [60].
This analysis also showed that the variants A4V, D90A, and I113T were predicted to decrease protein stability by Fold-X and I-Mutant, which is in agreement with the literature consulted [61][62][63][64][65]. The variant H46R, in turn, was only predicted as destabilizing by Fold-X (S3 Table). The variant H46R is not destabilized relative to the wild-type SOD1 in the nascent state. After the post-translational modifications necessary for SOD1 activity [66], this variant becomes relatively destabilized [64,67].
Although this association is not yet fully understood [68], stability reduction and protein aggregation are considered synergistic risk factors for ALS severity and short survival time [69,70]. All the variants predicted to increase protein aggregation tendency, except for variants D96V and H120L, were also predicted to decrease protein stability by at least one functional prediction algorithm used (S2 Table and S3 Table). This result thereby suggested that these variants may be potentially harmful to SOD1 and, consequently, valuable targets for future investigation.
The SNPEffect analysis (S3 Table) also showed that five variants were predicted to increase SOD1 amyloid propensity (V14A, G16C, G16A, G16S, and A89Y), while 5 variants were predicted to decrease this feature (V14M, Q15R, I17V, I18M, and N19S). Four amyloid-prone segments were identified in silico at the SOD1 sequence by Ivanova et al., 2014: 14-21, 30-38, 101-107, and 147-153. All the mutations predicted here to affect amyloid propensity, except for variant A90Y, occur in one of these regions, i.e. between amino-acids 14 and 21. Amyloidlike aggregates were observed in vitro for SOD1 variants occurring at these amyloid-prone regions [41,71]. SOD1 folding and Cu/Zn acquisition are catalyzed by the copper chaperone for SOD1 (CCS). In the absence of CCS interaction, SOD1 is highly likely to remain inactive and unfolded, being degraded or accumulate in toxic aggregates [72]. The predictive analysis showed that the variant P66R decreases SOD1 chaperone binding (S3 Table). This mutation was previously found in an fALS patient, but the characterization of its effects has not yet been performed [73]. The amino-acid substitution P66R occurs at the metal-binding loop of SOD1 [74], which could hinder binding, preventing CCS action [75][76][77].
Functionally important amino-acids on proteins are usually conserved throughout evolution due to the high selective pressure [19]. Most of the analyzed SOD1 variants occur in conserved positions (Fig 3B), which might be harmful to SOD1 as they affect possibly important amino-acids to protein, especially the variants also predicted as deleterious by all the functional prediction algorithms used (S1 Fig). Mutations affecting conserved amino-acids within SOD1 can cause ALS, which suggests that even minor alterations in this protein may severely affect its structure and function [85].
The variants affecting highly conserved amino-acids mostly occur at the metal-binding and electrostatic loops (S2 Fig). The highly conserved amino-acids within the SOD1 structure are mainly located on these functionally important loops [43,86], particularly at positions directly involved in metal-binding, superoxide guidance, and active site pocket formation [42,43,75].
As predicted by ConSurf (S2 Fig), the amino-acid substitution A4V and H46R occur at known conserved regions of SOD1 [21,35,87]. Although our results indicated that the variant I113T occurs in a variable position, this amino-acid is evolutionarily conserved in mammals [88]. The variant D90A, in turn, was predicted to affect an average conserved position. Unlike most ALS-related mutations in SOD1, D90A does not occur in a conserved amino-acid [65].
High fluctuations of RMSD values (Fig 4) were observed at the beginning of all simulations, which pointed to an initial moment of structural instability. This behavior is further confirmed by the high fluctuations of Rg (Fig 10), and SASA (Fig 11) observed in the first half of simulations. This might be attributed to the initial kinetic shock of molecular systems that occur in MD simulation processes [34]. The RMSD analysis also indicated that the protein structures fluctuate around average stable conformations after approximately 150ns (Fig 4), making sense to analyze local structural parameters, such as RMSF and B-factor [20]. The Rg analysis further suggested stable protein folding for SOD1 wild-type and its variants in the second half of trajectories (Fig 10). No alterations in the protein compactness (Fig 10), accessible surface (Fig 11), and secondary structure formation (Fig 12) were observed for the analyzed variants.
The essential mobility alterations observed in the ED analysis (Figs 7 and 8) occur at regions similar to those found altered in the RMSF (Fig 5) and B-factor (Fig 6) analyses, which mainly affect the metal-binding and electrostatic loops of the variants. Flexibility alterations can lead to strong and non-intuitive consequences for protein binding properties, as structural flexibility is determinant for binding affinity and specificity [89]. Moreover, the biological function of a protein is usually determined by its essential motions, especially those involved in protein interactions and binding to substrates [19]. Thus, the alterations observed during the MD simulations for the analyzed variants may affect SOD1 binding properties and, possibly, functional interactions, particularly at the metal-binding and electrostatic loops.
The ALS-SOD1 pathogenesis was initially believed to arise from the oxidative damage caused by reduced SOD1 activity, nevertheless, studies have further revealed that SOD1-null mice do not develop motor neuron disorder. On the other hand, transgenic mice expressing ALS-SOD1 mutant proteins become paralyzed despite possessing normal/elevated levels of SOD1 activity [90]. These pieces of evidence then suggest that SOD1 mutations lead to motor neuron death through toxic-gain of function rather than loss of enzymatic activity [91,92].
The precise nature of ALS-SOD1 toxicity has not yet been fully elucidated. However, a substantial body of literature suggested that it arises from SOD1 misfolding and aggregation, which are common properties of ALS-SOD1 mutations and central events in ALS-SOD1 pathophysiology [85,90,93,94]. More than 180 SOD1 coding mutations have already been described in the literature as occurring all over the protein and causing the same disease phenotype, ALS. Apparently, through similar toxic mechanisms, even mutations in SOD1 such as G93A, which cause changes as simple as the addition of a methyl group, can lead to the development of ALS [94,95].
Many studies have been carried out to understand the molecular basis of ALS-SOD1 toxicity. The study of Elam et.al. 2003 assessed the structural impact of the S134N and H46R variants in SOD1 using X-ray crystallography [90]. The study of Banci et.al. 2005 evaluated the effect of S134N using nuclear magnet resonance methods (RMN) [35]. The study of Molnar et.al. 2009 analyzed the impact of 13 ALS-SOD1 mutations (A4V, L38V, G41S, H46R, G72Y, D76S, G85R, D90A, G93A, D124V, D125H, and S134N) on protein structure and dynamics using hydrogen/deuterium exchange mass spectrometry [95]. Despite the different methods used by the authors, all of them compared the ALS-related mutants with the wild-type protein and observed structural or dynamics perturbations at the electrostatic loop of the analyzed variants [35,90,95]. Molnar et.al. 2009 also observed alterations at the metal-binding loop of variants A4V, G93A, S134N, G85R, D124V, and D125H. No results were generated for the electrostatic and metal-binding loops of H46R in Molnar's hydrogen/deuterium exchange experiments [95]. Further studies comparing the reactivity of conformation-specific antibodies for wild-type SOD1 and a series of other variants confirmed that SOD1 mutations induce some degree of misfolding at the electrostatic and metal-binding loop [85]. Together with the wet-lab experiments previously described in this paragraph, the molecular dynamics results presented here for A4V, D90A, H46R, and I113T further suggest that dynamics and structural alterations at the metal-binding and electrostatic loops could be a common characteristic of ALS-SOD1 mutations [35,90,95].
The computational methods are important allies of the wet-lab experiments [96], as the silico analyses can complement and overcome some difficulties of in vitro and in vivo assays [97,98]. The MD simulations performed in this study can contribute to a better understanding of the molecular basis of ALS-SOD1 since this method provides detailed information on protein conformation and fluctuations over time [20], which can complement the results of already performed wet-lab experiments and overcome some limitations such as the static images provided by X-ray crystallography and the limited size of peptic fragments that can be analyzed by hydrogen/deuterium exchange [95]. On the other hand, we did not analyze whether SOD1 mutations not related to ALS also cause dynamics and structural alterations at the electrostatic and metal-binding loops, which is a limitation of this study. The alterations observed here and in the wet-lab experiments previously mentioned [35,85,90,95] for ALS-SOD1 mutations could be an effect of SOD1 mutations in general, not necessarily related to the development of ALS.
Although the biological consequences of alterations at the functional loops of ALS-SOD1 variants are not fully understood, a well-accepted hypothesis suggests that this common property could lead to ALS possibly through a similar toxic mechanism involving aberrant interactions with cellular constituents and protein aggregation [68,85,90,93,95]. The study of Elam et.al. 2003 indicated that well-ordered electrostatic and metal-binding loops of SOD1 are critical to prevent protein aggregation. This study suggested that the functional loops adopt nonnative conformations upon H46R and S134N mutations, which deprotects a hydrophobic core of the β-barrel region of SOD1 that serves as a molecular interface for non-native protein interactions. Among the non-native interactions described by Elam et.al. 2003, extensive apolar and hydrogen-bond interactions were formed between the electrostatic loop of a mutant protein and the deprotected β-barrel region of a neighboring SOD1 [90]. Banci et.al. 2005 identified abnormal contacts at the electrostatic loop and increased aggregate formation for the variant S134N in solution, which the authors attributed to the increased mobility and flexibility also observed at this functional loop. Protein surface regions characterized by high mobility and flexibility have been described as an important factor for molecular recognition [93]. Aberrant interactions with dynein and mitochondrial proteins, such as BCL-2, were also reported in the literature as a common feature of ALS-SOD1 mutations. The study of Zhang et.al. 2007 showed that the variants A4V, G85R, and G93A interact with the dynein, while the wild-type SOD1 does not. SOD1 mutant forms with disordered loop regions preferentially interact with the dynein complex, favoring aggregation [99]. Thus, together with these previous studies, our MD simulation results can further support this hypothesis since the flexibility and essential dynamics alterations observed at the functional loops of A4V, D90A, H46R, and I113T may favor or disfavor interactions, possibly leading to abnormal association with cellular constituents [19,89].

Conclusion
Two hundred and thirty-three SOD1 variants were compiled from the databases and literature consulted, which underwent functional and stability predictions. The predictive analysis pointed to an elevated rate of deleterious and destabilizing predictions for these variants, indicating their harmful effects. Among the analyzed variants, those occurring at the electrostatic and metal-binding loops were noticeably more damaging to protein structure and function, which makes them valuable targets for future investigation. The ConSurf analysis indicated that mutations in SOD1 mainly affect conserved and possibly functionally important aminoacids. The MD analyses of variants A4V, D90A, H46R, and I113T pointed to flexibility and essential dynamics alterations at the electrostatic and metal-binding loops that could lead to aberrant interactions triggering toxic protein aggregation. These alterations may have harmful implications for SOD1 and explain their association with ALS. Understanding the effects of SOD1 mutations on protein structure and function facilitates the design of further experiments and provides relevant information on the molecular mechanism of pathology, which may contribute to improvements in existing treatments for ALS.
Supporting information S1  The bar plot shows the ConSurf score for each amino-acid of SOD1 protein affected by mutations. The bar plot was colored according to the ConSurf coloring-scheme, which varies from cyan and variable to maroon and conserved. The mutations affecting highly conserved amino-acids (i.e. ConSurf score = 9) are marked by an asterisk. (A) ConSurf prediction from the variant A2V to L39V. ConSurf did not predict the conservation score of the A2V variant due to the lack of evolutionary information. (B) ConSurf prediction from the variant T40I to A90Y. (C) ConSurf prediction from the variant A90V to V120L. (D) ConSurf prediction from the variant H121Q to A153P. (TIF)