Modeling the Aggregation Propensity and Toxicity of ... - DTU Orbit

215

Journal of Alzheimer’s Disease 47 (2015) 215–229 DOI 10.3233/JAD-150046 IOS Press

Modeling the Aggregation Propensity and Toxicity of Amyloid-␤ Variants Manish K. Tiwari and Kasper P. Kepp∗ Technical University of Denmark, DTU Chemistry, Kongens Lyngby, Denmark Handling Associate Editor: George Acquaah-Mensah

Accepted 11 April 2015

Abstract. Protein aggregation is a hallmark of many neurodegenerative disorders. Alzheimer’s disease (AD) is directly linked to deposits of amyloid-␤ (A␤) derived from the amyloid-␤ protein precursor (A␤PP), and multiple experimental studies have investigated the aggregation behavior of these amyloids. The present paper reports modeling of the aggregation propensities and cell toxicities of genetic variants of A␤ known to increase disease risk. From correlation to experimental data, and using four distinct experimental structures to test structural sensitivity, we find that the Spatial Aggregation Propensity (SAP) formalism can describe the relative experimental aggregation propensities of A␤42 variants (R2 = 0.49 and 0.70, p ∼ 0.02 and 0.002, for ˚ Our analysis finds correlation between the reduction in hydrophilic 1IYT and 1Z0Q conformations using a probe radius of 10 A). surface and experimental aggregation propensities. Finally, we show that experimental cell toxicities of A␤ variants are well described by computed SAP values, suggesting direct interplay between aggregation propensity and cell toxicity and providing a step toward a first computational estimator of A␤ toxicity. The present study contributes to our understanding of amyloid aggregation and suggests a method to predict aggregation propensity and toxicity of A␤ variants, and potentially to reduce aggregation propensities of amyloids by molecular intervention directed toward specific conformations of the peptides. Keywords: Alzheimer’s disease, amyloid-␤, hydrophilic surface, protein aggregation, structure-activity relations

INTRODUCTION During protein aggregation, protein parts or peptides self-assemble to produce small soluble oligomers or fibrillar aggregates, typically cross-linked ␤-sheets [1–4]. As a central characteristic of many human diseases such as Alzheimer’s disease (AD), Parkinson’s disease, and Huntington’s disease, prion diseases, and type II diabetes [2, 3, 5], research into aggregation processes have attracted substantial attention [5–8]. Pathological conditions are thought to emerge from conformational changes in the normal, native states of the peptides that lead to gain-of-toxic-function or loss-of-normal-function [5, 9]. ∗ Correspondence to: Kasper Planeta Kepp, Department of Chemistry, Technical University of Denmark, DK 2800 Kongens Lyngby, Denmark. Tel.: +45 45 25 24 09; Fax: +45 45 88 17 99; E-mail: [email protected].

AD is recognized as a major epidemic and one of the major challenges of the current century [10] with steadily increasing prevalence [10, 11]. Extracellular aggregates of amyloid-␤ (A␤) peptides, “senile plaques”, are a primary pathological hallmark of AD [12, 13]: These peptides are cleaved from the amyloid␤ protein precursor (A␤PP) found in membranes of cells and intracellular organelles [14, 15]. The plaques consist of A␤ isoforms of variable length, primarily A␤40 and A␤42 , that are post-translationally modified by oxidation and metal binding [11, 16]. AD is predominantly sporadic in nature, i.e., with family history seen in only a small minority of cases [2, 11] with multiple contributing risk factors from genes, environment, and lifestyle [17, 18]. However, genetic variations are known to give rise to some forms of early-onset familial AD (FAD) [19]. These variations are found mainly in A␤PP [20, 21] and in presenilin (PSEN) [22, 23], which is the catalytic unit

ISSN 1387-2877/15/$35.00 © 2015 – IOS Press and the authors. All rights reserved

216

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

of the ␥-secretase enzyme complex that produces A␤ upon proteolytic cleavage of A␤PP [24, 25]. Due to this genetic evidence and the amyloid deposits found in brains of patients, A␤ is acknowledged as having a central role in the disease [26–28]. It is now known that soluble oligomers of A␤ are the more toxic species [29–31]. This toxicity can be aggravated by chemical aging processes [32]. Furthermore, interactions with cell membranes during aggregation are important in modulating A␤ toxicity [33–36]. A two-step membrane disruption mechanism was identified recently [34–36]. First, ion-selective channels are formed, then disruption and fragmentation of the membrane occurs during fibril formation, a process that is aggravated by gangliosides [34]. Membrane disruption has also been reported for other proteins such as islet amyloid polypeptide [37, 38]. Among the isoforms of A␤ [39], the longer tend to be more hydrophobic and thus more prone to aggregation [2, 40], and of the two major isoforms, A␤42 (typically 10% of the total amyloid load, but the major form in deposited plaques) is more cytotoxic than the A␤40 isoform [41]. Consistent with this, several A␤PP mutations likely disturb ␥-secretase activity to increase the A␤42 /A␤40 ratio [42]. Despite these advances, the activeconformationsandchemicalpropertiesofthenormal A␤ monomers that serve as precursors in oligomer formation are currently unknown and hard to study experimentally due to their intrinsic disorder [43]. Although recent clinical trials have failed to avert A␤-mediated pathogenesis [44, 45], new promising strategies use molecular structural information to target A␤ toxicity [46, 47]. The need to delineate the fundamental chemical properties of A␤ is underlined by the role of post-translational modifications, notably truncations, metal ion binding, and oxidations of amyloids, and the generation of reactive oxygen species [11, 47–51]. Computational studies have been directed towards the molecular structure and dynamics of A␤ [52], A␤ fibrils [53], protofibrils [54], oligomers [55], and various polymorphisms [56]. Computational efforts have also shown promise toward design of inhibitors of A␤ aggregation [57, 58] and understanding of metal ion binding to A␤ [59, 60]. Still, correlation between molecular properties and experimental aggregation propensities of A␤ remains unexplored. It is thus of major interest to understand the properties and structures that determine A␤ aggregation propensities and whether these propensities can be accurately predicted by computations. To approach these challenges, we set out to model the aggregation propensities of the genetic A␤ variants

that are known clinically to increase risk of disease. We correlated computed and experimentally determined aggregation propensities, using several distinct A␤ NMR structures that represent a spectrum of structures relevant to in vivo conditions. Computation of protein aggregation in relation to neurodegenerative disorders has been a long-standing research goal [61]. However, as far as we know, we present here the first study that systematically compares computed and experimental aggregation propensities of known genetic A␤ variants related to disease. Our findings provide new structural and chemical detail to amyloid aggregation, suggest central structural features that serve as likely precursors, and show options for future computational estimation of aggregation propensities and toxicities of amyloid variants. The structures and properties identified as important for aggregation and toxicity may also potentially be targeted in rational design of amyloid-modifying drugs. MATERIALS AND METHODS Structural models of amyloids Monomers of A␤ are all mixtures of short helices and disordered regions, in contrast to insoluble extracellular fibrils formed upon aggregation, which are largely of ␤-sheet character [62, 63]. We have considered four distinct experimental structures of wild-type (WT) A␤, two for each of the most abundant isoforms; A␤42 : 1IYT [64], 1Z0Q [65] and A␤40 : 1BA4 [66], 2LFM [67] (Fig. 1A–1D). To compare these structures, we calculated the ensemble root mean square deviation (RMSD) of the NMR structures for A␤42 and A␤40 (Supplementary Tables 1–4). Compared to other reported full-length (A␤42 or A␤40 ) WT apo monomer structures (1BA6 [68], 1AML [69]), 1IYT [64], 1Z0Q [65], 1BA4 [66], 2LFM [67], the four structures employed in this study have well-defined conformational ensembles, with conformations within each ensemble showing strong resemblance; this renders a structure-property analysis meaningful. The four structures represent a spectrum of chemical environment and, for the same reason, different helix character, as seen in Fig. 1. Mutant modeling Structural modeling of A␤42 and A␤40 mutants was carried out as described recently [70]. Mutant structures were obtained by replacing side chains using the “build mutant” module [70–72] of Discovery Studio

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

217

Fig. 1. WT and mutant structures of A␤42 and A␤40 . A, B) The NMR coordinates of WT structures of A␤42 (PDB: 1IYT and 1Z0Q) with 15 mutant residues shown in sticks. C, D) The NMR coordinates of WT structures of A␤40 (PDB: 1BA4 and 2LFM) with 13 mutated residues. Colors range from yellow (negative hydrophobicity) to blue (positive hydrophobicity), via green (zero hydrophobicity) (the figure was generated using D.S. 4.0 visualizer).

4.0 (DS4.0) [73]. Briefly, ten conformations of each mutant were generated and evaluated based on conformational energy scoring functions, to produce four WT and 56 mutant structures. These were subsequently modeled as described in detail in the Supplementary Material. Modeling of surface properties and aggregation propensities In addition to the total solvent accessible surface area and the hydrophobic and hydrophilic surfaces [70], we computed the spatial aggregation propensity (SAP) for all the 60 structures. The SAP values (Supplementary Tables 5–8) were obtained using the method described by Chennamsetty et al. [74] as implemented ˚ and R = 10 A ˚ to test senin DS4.0, using both R = 5 A sitivity to surface resolution. This method uses the spatial conformations of the residues and their Black and Mould hydrophobicity index [75] to calculate a

structure-dependent aggregation propensity for each atom using the CHARMm force field, and the total aggregation propensity is obtained as the sum of its atomic aggregation scores [74]. Total solvent accessible surface area was computed using the default parameters of DS4.0, notably a ˚ and specific residue solvent probe radius of 1.4 A, accessibilities were also collected. The hydrophobic and hydrophilic surfaces of each A␤ species were determined using DS4.0 default parameters [70]. Supplementary Tables 9–12 show the computed total, hydrophilic, and hydrophobic surfaces for all conformations of all the mutants and WT amyloids. Decomposition into hydrophilic and hydrophobic surfaces was based on the Kyte and Doolittle scale [76] for individual residues. To provide further validation, we also calculated the same surface areas for the 60 structures using the POPS (Parameter OPtimised Sur˚ the results faces) server [77] with probe radius 1.4 A; can be found in Supplementary Tables 9–12.

218

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

Experimentally derived data set for correlation studies For the known genetic variants of A␤ which relate to disease, we have compiled a table of all relevant clinical and biochemical data on aggregation propensities [78–81] and cytotoxicities (EC50 ) [79, 80] (Supplementary Table 13). This database serves a useful role in pinpointing important chemical properties of amyloids but also, as done in this work, for testing property calculation methods and for understanding the chemical origins of, e.g., aggregation propensity. In the database, to put numbers from different experiments on the same scale, we normalized WT values to 1. As shown from the statistical correlations below, this normalization makes the data commensurable despite heterogeneities in lab protocols. Thioflavin T (ThT) based fluorescence is the most common way to quantitatively estimate the aggregation of peptides. The technique does not distinguish various aggregated or oligomerized forms such as prefibrillar and fibrillar aggregates [82]. ThT itself may also interact with exogenous compounds to affect the measured ThT fluorescence [83]. Yet, this technique is the one that researchers have used and thus forms the basis of the experimental data. However, to validate the use of these data, two independent data sets of aggregation propensities were considered. One data set was derived from the aggregation propensities reported by Murakami et al. [79], Zhou et al. [81], and Hori et al. [84] for A␤42 based on ThT fluorescence intensities after eight hours of incubation time. These data sets provide good numerical spread in aggregation propensities allowing a statistically meaningful correlation of variant properties and aggregation propensities. A second set of aggregation propensities was derived from Betts et al. [78] where ThT intensity was measured on agitated amyloid samples rather than a conventional static ThT binding assay, reporting the reverse time of maximal ThT binding in minutes. Times of half maxima have been compiled as 110 (WT), 130 (A21G), 18 (E22G), 80 (E22K), 60 (E22Q), and 50 (D23N) where the reverse of these numbers were computed and normalized to WT = 1 in this study (Supplementary Table 13). As shown below, both data sets provide statistically significant correlations versus the computed aggregation propensities and the trends are in both cases meaningful, in terms of the changed chemical properties. Thus, despite known shortcomings of the ThT assay, the comparison and analysis done here shows that these experiments can be meaningfully compared and used together.

Two aggregation propensities for E22G have been reported by Nilsberth et al. [85] and Hori et al. [84] (0.3 and 4.4: value normalized against WT). Similarly, two aggregation propensities for D23N have been reported by Nilsberth et al. [85] and Van Nostrand et al. [86] (0.7 and 3.0: value normalized against WT). The aggregation propensities reported for E22G by Hori and coworkers and for D23N reported by Van Nostrand and coworkers; are consistent with the procedure and values of other data and were thus used (Supplementary Table 13). Statistical correlation between amyloid properties We investigated the linear regressions between the computed properties described above, including the SAP, and relative, normalize experimental aggregation propensities and EC50 values for all WT and mutant A␤ structures. The data sets used for these correlation analyses are tabulated in the Supplementary Tables 14–17. A comprehensive list of all the statistical values (R2 and p) is provided in Supplementary Tables 18–21. In the discussion, we have considered values of R2 > 0.3 and p < 0.05 significant for discussion, i.e., the discussed relations are unlikely ( 0.3 significant for discussion, and a traditional choice of p-value < 0.05 was applied. This implies that the correlation observed is less than 5% likely to have occurred by coincidence, and this confidence basis was achieved in several correlations, and thus used as threshold [90]. Figures 2A–D show that the computed aggregation propensities A␤42 genetic variants are in good agree˚ (1IYT; ment with experimental data using both R = 5 A 2 2 R = 0.39, p = 0.04; 1Z0Q R = 0.37; p = 0.05) and ˚ (1IYT; R2 = 0.49, p = 0.02; 1Z0Q; R2 = 0.70; R = 10 A p = 0.002). Somewhat weaker correlations were found for the A␤40 structures 1BA4 and 2LFM both using ˚ (1BA4; R2 = 0.35; p = 0.05; 2LFM; R2 = 0.23; R=5A ˚ (1BA4; R2 = 0.34; p = 0.06; p = 0.13) and R = 10 A 2 2LFM; R = 0.32; p = 0.07) (Supplementary Material, Supplementary Fig. 2A–D). This fits well with the fact that the experimental aggregation propensities were measured on the A␤42 isoform, since the last two

residues both increase absolute aggregation propensities and could also influence the relative aggregation propensities of the different A␤ variants. Thus, the SAP method gives encouraging results ˚ For the specific isoform in particular using R = 10 A. A␤42 represented by experimental data, R2 = 0.49 and R2 approaches the accuracy of specifically fitted models such as the Chiti-Dobson model [91]. The trend in aggregation propensities are produced with high statistical significance at 99% confidence, showing that computed SAP values can provide substantial descriptive and interpretative power to the challenge of understanding molecular causes of amyloid aggregation. Figure 3 shows the same analysis performed using the second experimental data set by Betts et al. [78]. The computed aggregation propensities were in even better agreement with these experimental data and ˚ compusignificant at the 95% level for all R = 10 A tations. Trends obtained were strong both with 1IYT

˚ Fig. 3. Correlations between computed and experimental aggregation propensities from the second data set [78]: A) using 1IYT and R = 5 A; ˚ C) using 1Z0Q and R = 5 A; ˚ D) using 1Z0Q and R = 10 A. ˚ B) using 1IYT and R = 10 A;

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

˚ Fig. 3A, (R2 = 0.80, p = 0.02 using either R = 5 or 10 A, ˚ and B) and 1Z0Q (R2 = 0.58; p = 0.08 using R = 5 A ˚ (Fig. 3C, D). For R2 = 0.81; p = 0.02 using R = 10 A) 1BA4 and 2LFM, correlations were likewise significant at the same levels (Supplementary Figure 3). The strongest correlations were generally obtained ˚ and these results were obtained when using R = 10 A for both data sets. Furthermore, the 1Z0Q structure revealed the highest overall correlation and is likely to represent the most realistic structure in terms of the aggregation experiments; it is a mixed helix-disorder structure. Thus, the computed aggregation propensities ˚ and the 1Z0Q structure as of the variants using 10 A basis gave correlations of R2 = 0.70; p = 0.002 (Fig. 2D) and R2 = 0.81; p = 0.015 (Fig. 3D), which is as accurate as linear 3-parameter models with specific parameterization [91]. The computations thus show that one can obtain statistical significant prediction of aggregation propensities for amyloid variants using the applied

221

methodology, but they also show that aggregation propensities are quite conformation-dependent; an observation that resonates well with recent experimental findings that cytotoxicity of amyloids is likely to be structure-dependent [43, 87]. 2LFM (Supplementary Fig. 2C, D) is the least structured conformation among the four structures we studied and showed the weakest correlation to experimental aggregation propensities, suggesting that this A␤40 structure differs substantially from the A␤42 conformations measured experimentally. Reducing hydrophilic surface is a main determinant of amyloid aggregation To understand the drivers of amyloid aggregation in more detail, we further tested whether simple properties of amyloid variants correlated with experimental aggregation propensities. We found that experimental aggregation propensities were not correlated

Fig. 4. Correlation between experimental aggregation propensities from the first data set [79, 81, 84] and computed hydrophilic surfaces of amyloid variants: A) using structure 1IYT; B) using structure 1Z0Q; C) using structure 1BA4; D) using structure 2LFM.

222

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

Fig. 5. Correlation between experimental aggregation propensities from the second data set [78] and computed hydrophilic surfaces: A) using 1IYT; B) using 1Z0Q; C) using 1BA4; D) using 2LFM.

significantly with computed hydrophobic surface and total solvent surface for the first data set (Supplementary Material, Supplementary Figs. 4 and 5), although the trends show that they have some importance. The same conclusions were obtained using the data by Betts et al. [78] (Supplementary Figs. 6 and 7), i.e., hydrophobic surface contributes but is a partial descriptor for all 8 regressions shown in Supplementary Figs. 4 and 6. Figures 4 and 5 show the linear regression plots for aggregation propensities and hydrophilic surfaces of all the A␤ species. Figure 4 shows the correlation between experimental aggregation propensities reported by Murakami et al. [79], Zhou et al. [81], and Hori et al. [84], and computed hydrophilic surfaces of the A␤ species: This property showed some, but not significant at the 95% confidence level, correlation with reported aggregation propensities for the structures 1IYT (R2 = 0.34; p = 0.06, Fig. 4A), 1BA4

(R2 = 0.32; p = 0.07, Fig. 4C), and 2LFM (R2 = 0.32; p = 0.07, Fig. 4D). Significant correlation was found when using the 1Z0Q structure (R2 = 0.40; p = 0.05, Fig. 4B). Figure 5 shows a similar regression analysis using the experimental aggregation propensities reported by Betts et al. [78]. In this case, computed hydrophilic surface produced highly significant correlations: 1IYT gave R2 = 0.87; p = 0.007 (Fig. 5A), 1Z0Q gave R2 = 0.64; p = 0.06 (Fig. 5B), 1BA4 had R2 = 0.82; p = 0.01 (Fig. 5C), and 2LFM had R2 = 0.83; p = 0.01 (Fig. 5D). For data in Fig. 4, the A␤ variants D678N (D7N; a normalized propensity versus WT of 3.4) and A692G (A21G; 0.07) are outliers in the scatter plots. For the second data set in Fig. 5, outliers are less substantial. From these correlation coefficients and p-values (see Supplementary Tables 18–21 for numerical values of these parameters), we conclude that reduced

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

hydrophilic surface is a determinant in describing the aggregation propensity of the known disease-related A␤ variants. The negative correlations are physically meaningful as they imply that peptide aggregation is determined largely by reduced hydrophilic surface, and reduced hydrophilic surface as seen in some variants strongly correlates with increased aggregation propensity.

Experimental toxicities correlate with structure-dependent aggregation propensities The reported cytotoxicities of A␤ variants [79, 80], as quantified by their EC50 values, are thought to relate to aggregation of amyloids [44], although no statistically significant correlation from analysis of general available data has yet been described. To investigate whether we could identify a simple cause of cytotoxicities of genetic A␤ variants, we correlated known EC50 values for these variants against computed properties, i.e., hydrophobic surface, hydrophilic surface, total surface, and SAP. None of these other properties showed any significant correlation by themselves (Supplementary Figs. 8–10). Figure 6 shows the correlation plots of computed aggregation propensities vs. normalized experimental EC50 values, with the left panels again showing ˚ and the four right panels showing R = 10 A. ˚ R=5A The correlations were again dependent on structure ˚ radius, 1IYT gave R2 = 0.36, p = 0.05 used: With 5 A (Fig. 6A); 1Z0Q gave R2 = 0.55; p = 0.009 (Fig. 6C); 1BA4 gave R2 = 0.60; p = 0.02 (Fig. 6E), and 2LFM ˚ gave R2 = 0.20; p = 0.23 (Fig. 6G).With R = 10 A, 1IYT gave R2 = 0.52, p = 0.01 (Fig. 6B); 1Z0Q gave R2 = 0.31; p = 0.08 (Fig. 6D); 1BA4 gave R2 = 0.50; p = 0.04 (Fig. 6F); and 2LFM gave R2 = 0.44; p = 0.05 (Fig. 6H). From all figures, it can be seen that smaller experimental EC50 value, corresponding to higher toxicity, correlates positively with computed aggregation propensity. This correlation is robust against variations in structure, providing theoretical insight into the relationship between aggregation and cell toxicity of amyloids. With this correlation at hand, one can identify chemical properties important not only for aggregation propensity but also for toxicity. From this analysis, we conclude that experimentally reported toxicities of the various amyloid variants correlate significantly to structure-dependent aggregation propensities of the species. Five of the eight correlations in Fig. 6 are significant at the 95% confidence level, and the direction of the correlation is physi-

223

cally consistent with aggregation contributing to cell toxicity, the first such link supported by statistical significance from regression analysis of available data. Even with heterogeneity in lab protocols and with the variations in chemical structure of the amyloids that necessarily create noise in the data sets, these correlations persist in the majority of cases studied and are thus very unlikely (as also seen from the p-values) to have occurred coincidentally.

Chemical causes for the high aggregation propensity of Aβ variants Increase in hydrophobic exposure is considered one of the major driving forces causing A␤ variants to aggregate [92–94]. For all the A␤ structures, the computed hydrophobic surface area provides some correlation to aggregation propensity but is not statistically significant by itself (Supplementary Figures 4 and 6), consistent with general view that the hydrophobicity only constitutes part of the chemical driving force toward aggregation [91, 94]. Previous experimental investigations of the aggregation propensities of A␤ variants (Supplementary Table 13) have suggested that E22G and D7N variants are most aggregation prone compared to the WT, whereas the A21G variant was found to have lower aggregation tendency compared to the WT, i.e., E22G > D7N > D23N > E22Q ≥ H6R > A2V = D7H > E11K > E22K > WT > A21G (Supplementary Table 13). The latter data point for A21G is an outlier in the regression and has an unusually low aggregation propensity versus WT. While this variant has reduced hydrophobic surface by mutation to the small, hydrophilic glycine, the error in the regression could be due to introduction of glycine, which can produce problems in computed structures due to real co-localization of water not present in the WT-based structure [95]. The E22G variant exhibits the highest aggregation tendency both experimentally and in our computations, and is the second most toxic variant so far characterized, as measured from EC50 values [79] (Supplementary Table 13). A link to the reduction of hydrophilic surface (Figs. 7 and 8) can be seen numerically from Supplementary Tables 9 to 12, irrespective of structures and methods used in the computations. The hydrophilic sur˚ 2 in the WT (Fig. 7A) to face area is reduced from 1257 A 2 ˚ in the E22G variant (Fig. 7B) when being in the 1174 A 1IYT conformation (Supplementary Table 9). When in ˚2 a 1Z0Q-like conformation, this change is from 1189 A

224

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

˚ and 10 A: ˚ A) using Fig. 6. Correlation between experimental EC50 [79, 80] and computed aggregation propensities with probe radii (R) of 5 A ˚ B) using 1IYT and R = 10 A; ˚ C) using 1Z0Q and R = 5 A; ˚ D) using 1Z0Q and R = 10 A; ˚ E) using 1BA4 and R = 5 A; ˚ F) using 1IYT and R = 5 A; ˚ G) using 2LFM and R = 5 A; ˚ H) using 2LFM and R = 10 A. ˚ 1BA4 and R = 10 A;

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

225

Fig. 7. Hydrophilic surface area of WT and mutants of A␤42 : A) and B) using 1IYT structures; C) and D) using 1Z0Q. The mutated residue is shown in sticks while the affected molecular surface is represented in green color. Secondary structure is represented with blue N-terminus and red C-terminus. The picture was made using D.S. 4.0 visualizer.

Fig. 8. Hydrophilic surface of WT and E22G mutants of A␤40 : A) and B) using 1BA4C) and D) using 2LFM. The mutated residue is shown in sticks while the affected surface is represented in green. Secondary structure is represented with blue N-terminus and red C-terminus. The picture was made using D.S. 4.0 visualizer.

226

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids

˚ 2 (Fig. 7C, D; Supplementary Table 10); with to 1133 A ˚ 2 to 1091 A ˚ 2 (Fig. 8A, 1BA4 the change is from 1163 A B; Supplementary Table 11), and with 2LFM the ˚ 2 to 1350 A ˚ 2 (Fig. 8C, D; Supreduction is from 1429 A plementary Table 12). E22G gave the highest computed ˚ and 10 A ˚ aggregation propensity using both R = 5 A across all the structures. While the hydrophilic surface area seems relevant to amyloid aggregation, the molecular mode of toxicity of amyloids remains under debate [96, 97], but it has been suggested that membrane interactions are involved [36]; such a molecular mode of toxicity is consistent with an increase in structure-dependent aggregation propensity of specific conformations of the amyloids, either directly in the monomer forms or in oligomers that have been formed from monomers with high aggregation propensity, as studied in this work. Importantly, the reduced hydrophilicity that we find correlate with aggregation propensity of genetic variants enables prospects for uniting sporadic and FAD, since metal ion binding and other modifications of the hydrophilic N-terminal observed in sporadic AD would likely lead to similar post-translational reduction in hydrophilicity as that caused by genetic mutation contributing to FAD/ cerebral amyloid angiopathy, as studied here. In conclusion, we have compiled experimental data for aggregation propensities and cell toxicities of genetic variants of A␤ known to increase disease risk. By comparison to computed chemical properties for the same mutants, we have enabled a significance study of how variations in chemical properties of amyloids affect their toxicity. The results show that the experimental aggregation propensity of a given amyloid species can be accurately modeled in some cases (p = 0.002), but that some structural conformations are more valid than others, reflecting likely structural features required for aggregation. In addition, the conformation-dependent hydrophilic surfaces display strong correlation with the experimental data. Interestingly, experimental cellular toxicities of amyloid variants can be computed from aggregation propensities, consistent with a strong relation between these properties. The conformation-dependence points to specific conformational features of the monomers likely reflecting experimentally observed precursors of aggregation and toxicity. These findings should be relevant both in the fast screening of aggregation propensity and toxicity of new modified amyloid variants, and possibly also in therapeutic approaches that selectively control the hydrophilic surface area.

ACKNOWLEDGMENTS The authors acknowledge the Technical University of Denmark for providing a Hans Christian Ørsted (HCØ) fellowship to MKT. Authors’ disclosures available online (http://j-alz. com/manuscript-disclosures/15-0046r2).

SUPPLEMENTARY MATERIAL The supplementary material is available in the electronic version of this article: http://dx.doi.org/ 10.3233/JAD-150046.

REFERENCES [1] [2] [3] [4] [5]

[6]

[7] [8]

[9] [10] [11] [12] [13]

[14]

[15]

[16]

[17]

Anfinsen CB (1973) Principles that govern the folding of protein chains. Science 181, 223-230. Chiti F, Dobson CM (2006) Protein misfolding, functional amyloid, and human disease. Annu Rev Biochem 75, 333-366. Dobson CM (2003) Protein folding and misfolding. Nature 426, 884-890. Dyson HJ, Wright PE (2005) Intrinsically unstructured proteins and their functions. Nat Rev Mol Cell Biol 6, 197-208. Knowles TP, Vendruscolo M, Dobson CM (2014) The amyloid state and its association with protein misfolding diseases. Nat Rev Mol Cell Biol 15, 384-396. Tycko R (2004) Progress towards a molecular-level structural understanding of amyloid fibrils. Curr Opin Struct Biol 14, 96-103. Murphy RM (2002) Peptide aggregation in neurodegenerative disease. Annu Rev Biomed Eng 4, 155-174. Selkoe DJ (2004) Cell biology of protein misfolding: The examples of Alzheimer’s and Parkinson’s diseases. Nat Cell Biol 6, 1054-1061. Dobson CM (2002) Protein-misfolding diseases: Getting out of shape. Nature 418, 729-730. Alzheimer’s, Association (2013) 2013 Alzheimer’s disease facts and figures. Alzheimers Dement 9, 208-245. Kepp KP (2012) Bioinorganic chemistry of Alzheimer’s disease. Chem Rev 112, 5193-5239. Zhang YW, Thompson R, Zhang H, Xu H (2011) APP processing in Alzheimer’s disease. Mol Brain 4, 3. Lee HG, Casadesus G, Zhu X, Joseph JA, Perry G, Smith MA (2004) Perspectives on the amyloid-␤ cascade hypothesis. J Alzheimers Dis 6, 137-145. Tanzi RE, Bertram L (2005) Twenty years of the Alzheimer’s disease amyloid hypothesis: A genetic perspective. Cell 120, 545-555. Marlow L, Cain M, Pappolla MA, Sambamurti K (2003) ␤-secretase processing of the Alzheimer’s amyloid protein precursor (APP). J Mol Neurosci 20, 233-239. O’Brien RJ, Wong PC (2011) Amyloid precursor protein processing and Alzheimer’s disease. Annu Rev Neurosci 34, 185-204. Barnes DE, Yaffe K (2011) The projected effect of risk factor reduction on Alzheimer’s disease prevalence. Lancet Neurol 10, 819-828.

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids [18] [19]

[20]

[21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

[29]

[30]

[31]

[32]

[33]

Eisenstein M (2011) Genetics: Finding risk factors. Nature 475, S20-S22. Karlawish J (2011) Addressing the ethical, policy, and social challenges of preclinical Alzheimer disease. Neurology 77, 1487-1493. Goate A (2006) Segregation of a missense mutation in the amyloid beta-protein precursor gene with familial Alzheimer’s disease. J Alzheimers Dis 9, 341-347. Goate A, Chartier-Harlin MC, Mullan M, Brown J, Crawford F, Fidani L, Giuffra L, Haynes A, Irving N, James L, Mant R, Newton P, Rooke K, Roques P, Talbot C, Pericak-Vance M, Roses A, Williamson R, Rossor M, Owen M, Hardy J (1991) Segregation of a missense mutation in the amyloid precursor protein gene with familial Alzheimer’s disease. Nature 349, 704-706. Levy-Lahad E, Wasco W, Poorkaj P, Romano DM, Oshima J, Pettingell WH, Yu CE, Jondro PD, Schmidt SD, Wang K, Crowley AC, Fu Y-H, Guenette SY, Galas D, Nemens E, Wijsman EM, Bird TD, Schellenberg GD, Tanzi RE (1995) Candidate gene for the chromosome 1 familial Alzheimer’s disease locus. Science 269, 973-977. Sherrington R, Rogaev EI, Liang Y, Rogaeva EA, Levesque G, Ikeda M, Chi H, Lin C, Li G, Holman K, Tsuda T, Mar L, Foncin JF, Bruni AC, Montesi MP, Sorbi S, Rainero I, Pinessi L, Nee L, Chumakov I, Pollen D, Brookes A, Sanseau P, Polinsky RJ, Wasco W, Da Silva HA, Haines JL, PerkicakVance MA, Tanzi RE, Roses AD, Fraser PE, Rommens JM, St George-Hyslop PH (1995) Cloning of a gene bearing missense mutations in early-onset familial Alzheimer’s disease. Nature 375, 754-760. De Strooper B, Saftig P, Craessaerts K, Vanderstichele H, Guhde G, Annaert W, Von Figura K, Van Leuven F (1998) Deficiency of presenilin-1 inhibits the normal cleavage of amyloid precursor protein. Nature 391, 387-390. Wolfe MS, Xia W, Ostaszewski BL, Diehl TS, Kimberly WT, Selkoe DJ (1999) Two transmembrane aspartates in presenilin-1 required for presenilin endoproteolysis and gamma-secretase activity. Nature 398, 513-517. Hardy JA, Higgins GA (1992) Alzheimer’s disease: The amyloid cascade hypothesis. Science 256, 184-185. Acquaah-Mensah GK, Agu N, Khan T, Gardner A (2014) A regulatory role for the insulin- and BDNF-linked RORA in the hippocampus: Implications for Alzheimer’s disease. J Alzheimers Dis 44, 827-838. Karran E, Mercken M, De Strooper B (2011) The amyloid cascade hypothesis for Alzheimer’s disease: An appraisal for the development of therapeutics. Nat Rev Drug Discov 10, 698-712. Yankner BA, Dawes LR, Fisher S, Villa-Komaroff L, OsterGranite ML, Neve RL (1989) Neurotoxicity of a fragment of the amyloid precursor associated with Alzheimer’s disease. Science 245, 417-420. Kayed R, Head E, Thompson JL, McIntire TM, Milton SC, Cotman CW, Glabe CG (2003) Common structure of soluble amyloid oligomers implies common mechanism of pathogenesis. Science 300, 486-489. Ono K, Condron MM, Teplow DB (2009) Structureneurotoxicity relationships of amyloid ␤-protein oligomers. Proc Natl Acad Sci U S A 106, 14745-14750. Cohen E, Bieschke J, Perciavalle RM, Kelly JW, Dillin A (2006) Opposing activities protect against age-onset proteotoxicity. Science 313, 1604-1610. Sun XD, Mo ZL, Taylor BM, Epps DE (2003) A slowly formed transient conformer of A␤1-40 is toxic to inward chan-

[34]

[35]

[36]

[37]

[38]

[39]

[40] [41]

[42]

[43]

[44]

[45]

[46]

[47]

[48]

[49]

[50]

227

nels of dissociated hippocampal and cortical neurons of rats. Neurobiol Dis 14, 567-578. Sciacca MF, Kotler SA, Brender JR, Chen J, Lee DK, Ramamoorthy A (2012) Two-step mechanism of membrane disruption by A␤ through membrane fragmentation and pore formation. Biophys J 103, 702-710. Ramamoorthy A, Lim MH (2013) Structural characterization and inhibition of toxic amyloid-␤ oligomeric intermediates. Biophys J 105, 287-288. Kotler SA, Walsh P, Brender JR, Ramamoorthy A (2014) Differences between amyloid-␤ aggregation in solution and on the membrane: Insights into elucidation of the mechanistic details of Alzheimer’s disease. Chem Soc Rev 43, 6692-6700. Brender JR, Heyl DL, Samisetti S, Kotler SA, Osborne JM, Pesaru RR, Ramamoorthy A (2013) Membrane disordering is not sufficient for membrane permeabilization by islet amyloid polypeptide: Studies of IAPP(20–29) fragments. Phys Chem Chem Phys 15, 8908-8915. Brender JR, Salamekh S, Ramamoorthy A (2012) Membrane disruption and early events in the aggregation of the diabetes related peptide IAPP from a molecular perspective. Acc Chem Res 45, 454-462. Atwood CS, Obrenovich ME, Liu T, Chan H, Perry G, Smith MA, Martins RN (2003) Amyloid-␤: A chameleon walking in two worlds: A review of the trophic and toxic properties of amyloid-␤. Brain Res Brain Res Rev 43, 1-16. Rauk A (2009) The chemistry of Alzheimer’s disease. Chem Soc Rev 38, 2698-2715. Hardy J, Selkoe DJ (2002) The amyloid hypothesis of Alzheimer’s disease: Progress and problems on the road to therapeutics. Science 297, 353-356. De Jonghe C, Esselens C, Kumar-Singh S, Craessaerts K, Serneels S, Checler F, Annaert W, Van Broeckhoven C, De Strooper B (2001) Pathogenic APP mutations near the ␥secretase cleavage site differentially affect A␤ secretion and APP C-terminal fragment stability. Hum Mol Genet 10, 16651671. Deshpande A, Mina E, Glabe C, Busciglio J (2006) Different conformations of amyloid ␤ induce neurotoxicity by distinct mechanisms in human cortical neurons. J Neurosci 26, 60116018. Karran E, Hardy J (2014) A critique of the drug discovery and phase 3 clinical programs targeting the amyloid hypothesis for Alzheimer disease. Ann Neurol 76, 185-205. Rosenblum WI (2014) Why Alzheimer trials fail: Removing soluble oligomeric beta amyloid is essential, inconsistent, and difficult. Neurobiol Aging 35, 969-974. DeToma AS, Salamekh S, Ramamoorthy A, Lim MH (2012) Misfolded proteins in Alzheimer’s disease and type II diabetes. Chem Soc Rev 41, 608-621. Savelieff MG, Liu YZ, Senthamarai RRP, Korshavn KJ, Lee HJ, Ramamoorthy A, Lim MH (2014) A small molecule that displays marked reactivity toward copper- versus zincamyloid-␤ implicated in Alzheimer’s disease. Chem Commun 50, 5301-5303. Bayer TA, Wirths O (2014) Focusing the amyloid cascade hypothesis on N-truncated Abeta peptides as drug targets against Alzheimer’s disease. Acta Neuropathol 127, 787-801. Jawhar S, Wirths O, Bayer TA (2011) Pyroglutamate amyloidbeta (A␤): A hatchet man in Alzheimer disease. J Biol Chem 286, 38825-38832. Barnham KJ, Bush AI (2008) Metals in Alzheimer’s and Parkinson’s diseases. Curr Opin Chem Biol 12, 222-228.

228 [51]

[52]

[53]

[54]

[55]

[56]

[57]

[58]

[59]

[60]

[61]

[62]

[63] [64]

[65]

[66]

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids Sch¨oneich C, Williams TD (2002) Cu(II)-catalyzed oxidation of ␤-amyloid peptide targets His13 and His14 over His6: detection of 2-oxo-histidine by HPLC-MS/MS. Chem Res Toxicol 15, 717-722. Fandrich M, Schmidt M, Grigorieff N (2011) Recent progress in understanding Alzheimer’s ␤-amyloid structures. Trends Biochem Sci 36, 338-345. Bertini I, Gonnelli L, Luchinat C, Mao J, Nesi A (2011) A new structural model of A␤40 fibrils. J Am Chem Soc 133, 16013-16022. Kahler A, Sticht H, Horn AH (2013) Conformational stability of fibrillar amyloid-beta oligomers via protofilament pair formation – a systematic computational study. PLoS One 8, e70521. Ahmed M, Davis J, Aucoin D, Sato T, Ahuja S, Aimoto S, Elliott JI, Van Nostrand WE, Smith SO (2010) Structural conversion of neurotoxic amyloid-␤1-42 oligomers to fibrils. Nat Struct Mol Biol 17, 561-567. Colletier JP, Laganowsky A, Landau M, Zhao M, Soriaga AB, Goldschmidt L, Flot D, Cascio D, Sawaya MR, Eisenberg D (2011) Molecular basis for amyloid-␤ polymorphism. Proc Natl Acad Sci U S A 108, 16938-16943. Eskici G, Gur M (2013) Computational design of new peptide inhibitors for amyloid beta (A␤) aggregation in Alzheimer’s disease: Application of a novel methodology. PLoS One 8, e66178. Wang Q, Liang G, Zhang M, Zhao J, Patel K, Yu X, Zhao C, Ding B, Zhang G, Zhou F, Zheng J (2014) De novo design of self-assembled hexapeptides as ␤-amyloid (A␤) peptide inhibitors. ACS Chem Neurosci 5, 972-981. Faller P, Hureau C, Berthoumieu O (2013) Role of metal ions in the self-assembly of the Alzheimer’s amyloid-␤ peptide. Inorg Chem 52, 12193-12206. Hureau C, Dorlet P (2012) Coordination of redox active metal ions to the amyloid precursor protein and to amyloid-␤ peptides involved in Alzheimer disease. Part 2: Dependence of Cu(II) binding sites with A␤ sequences. Coord Chem Rev 256, 2175-2187. Redler RL, Shirvanyants D, Dagliyan O, Ding F, Kim DN, Kota P, Proctor EA, Ramachandran S, Tandon A, Dokholyan NV (2014) Computational approaches to understanding protein aggregation in neurodegeneration. J Mol Cell Biol 6, 104-115. Chimon S, Shaibat MA, Jones CR, Calero DC, Aizezi B, Ishii Y (2007) Evidence of fibril-like ␤-sheet structures in a neurotoxic amyloid intermediate of Alzheimer’s ␤-amyloid. Nat Struct Mol Biol 14, 1157-1164. Serpell LC (2000) Alzheimer’s amyloid fibrils: Structure and assembly. Biochim Biophys Acta 1502, 16-30. Crescenzi O, Tomaselli S, Guerrini R, Salvadori S, D’Ursi AM, Temussi PA, Picone D (2002) Solution structure of the Alzheimer amyloid ␤-peptide (1-42) in an apolar microenvironment. Similarity with a virus fusion domain. Eur J Biochem 269, 5642-5648. Tomaselli S, Esposito V, Vangone P, van Nuland NA, Bonvin AM, Guerrini R, Tancredi T, Temussi PA, Picone D (2006) The ␣-to-␤ conformational transition of Alzheimer’s A␤-(1-42) peptide in aqueous media is reversible: A step by step conformational analysis suggests the location of ␤ conformation seeding. ChemBioChem 7, 257-267. Coles M, Bicknell W, Watson AA, Fairlie DP, Craik DJ (1998) Solution structure of amyloid ␤-peptide(1-40) in a watermicelle environment. Is the membrane-spanning domain where we think it is? Biochemistry 37, 11064-11077.

[67]

[68]

[69]

[70]

[71]

[72]

[73] [74]

[75]

[76]

[77]

[78]

[79]

[80]

[81]

[82]

[83]

Vivekanandan S, Brender JR, Lee SY, Ramamoorthy A (2011) A partially folded structure of amyloid-beta(1-40) in an aqueous environment. Biochem Biophys Res Commun 411, 312-316. Watson AA, Fairlie DP, Craik DJ (1998) Solution structure of methionine-oxidized amyloid ␤-peptide (1-40). Does oxidation affect conformational switching? Biochemistry 37, 12700-12706. Sticht H, Bayer P, Willbold D, Dames S, Hilbich C, Beyreuther K, Frank RW, Rosch P (1995) Structure of amyloid A4(1-40)-peptide of Alzheimer’s disease. Eur J Biochem 233, 293-298. Tiwari MK, Kepp KP (2015) Pathogenic properties of Alzheimer’s ␤-amyloid identified from structureproperty patient-phenotype correlations. Dalton Trans 44, 2747-2754. Tiwari MK, Kalia VC, Kang YC, Lee J-K (2014) Role of a remote leucine residue in the catalytic function of polyol dehydrogenase. Mol BioSyst 10, 3255-3263. Tiwari MK, Singh RK, Singh R, Jeya M, Zhao H, Lee JK (2012) Role of conserved glycine in zinc-dependent medium chain dehydrogenase/reductase superfamily. J Biol Chem 287, 19429-19439. Accelrys Software Inc. (2013) Discovery Studio Modeling Environment, Release 4.0. Accelrys Software Inc., San Diego. Chennamsetty N, Voynov V, Kayser V, Helk B, Trout BL (2009) Design of therapeutic proteins with enhanced stability. Proc Natl Acad Sci U S A 106, 11937-11942. Black SD, Mould DR (1991) Development of hydrophobicity parameters to analyze proteins which bear posttranslational or cotranslational modifications. Anal Biochem 193, 72-82. Kyte J, Doolittle RF (1982) A simple method for displaying the hydropathic character of a protein. J Mol Biol 157, 105132. Cavallo L, Kleinjung J, Fraternali F (2003) POPS: A fast algorithm for solvent accessible surface areas at atomic and residue level. Nucleic Acids Res 31, 3364-3366. Betts V, Leissring MA, Dolios G, Wang R, Selkoe DJ, Walsh DM (2008) Aggregation and catabolism of disease-associated intra-A␤ mutations: Reduced proteolysis of A␤A21G by neprilysin. Neurobiol Dis 31, 442-450. Murakami K, Irie K, Morimoto A, Ohigashi H, Shindo M, Nagao M, Shimizu T, Shirasawa T (2003) Neurotoxicity and physicochemical properties of A␤ mutant peptides from cerebral amyloid angiopathy: Implication for the pathogenesis of cerebral amyloid angiopathy and Alzheimer’s disease. J Biol Chem 278, 46179-46187. Ono K, Condron MM, Teplow DB (2010) Effects of the English (H6R) and Tottori (D7N) familial Alzheimer disease mutations on amyloid ␤-protein assembly and toxicity. J Biol Chem 285, 23186-23197. Zhou L, Brouwers N, Benilova I, Vandersteen A, Mercken M, Van Laere K, Van Damme P, Demedts D, Van Leuven F, Sleegers K, Broersen K, Van Broeckhoven C, Vandenberghe R, De Strooper B (2011) Amyloid precursor protein mutation E682K at the alternative ␤-secretase cleavage ␤’-site increases A␤ generation. EMBO Mol Med 3, 291-302. Reinke AA, Abulwerdi GA, Gestwicki JE (2010) Quantifying prefibrillar amyloids in vitro by using a “thioflavin-like” spectroscopic method. Chembiochem 11, 1889-1895. Hudson SA, Ecroyd H, Kee TW, Carver JA (2009) The thioflavin T fluorescence assay for amyloid fibril detection can be biased by the presence of exogenous compounds. FEBS J 276, 5960-5972.

M.K. Tiwari and K.P. Kepp / Modeling Aggregation and Toxicity of Amyloids [84]

[85]

[86]

[87]

[88]

[89]

Hori Y, Hashimoto T, Wakutani Y, Urakami K, Nakashima K, Condron MM, Tsubuki S, Saido TC, Teplow DB, Iwatsubo T (2007) The Tottori (D7N) and English (H6R) familial Alzheimer disease mutations accelerate A␤ fibril formation without increasing protofibril formation. J Biol Chem 282, 4916-4923. Nilsberth C, Westlind-Danielsson A, Eckman CB, Condron MM, Axelman K, Forsell C, Stenh C, Luthman J, Teplow DB, Younkin SG, Naslund J, Lannfelt L (2001) The ‘Arctic’ APP mutation (E693G) causes Alzheimer’s disease by enhanced A␤ protofibril formation. Nat Neurosci 4, 887-893. Van Nostrand WE, Melchor JP, Cho HS, Greenberg SM, Rebeck GW (2001) Pathogenic effects of D23N Iowa mutant amyloid ␤-protein. J Biol Chem 276, 32860-32866. Ladiwala ARA, Litt J, Kane RS, Aucoin DS, Smith SO, Ranjan S, Davis J, Van Nostrand WE, Tessier PM (2012) Conformational differences between two amyloid ␤ oligomers of similar size and dissimilar toxicity. J Biol Chem 287, 2476524773. Miao J, Xu F, Davis J, Otte-Holler I, Verbeek MM, Van Nostrand WE (2005) Cerebral microvascular amyloid ␤ protein deposition induces vascular degeneration and neuroinflammation in transgenic mice expressing human vasculotropic mutant amyloid ␤ precursor protein. Am J Pathol 167, 505515. Weller RO, Preston SD, Subash M, Carare RO (2009) Cerebral amyloid angiopathy in the aetiology and immunotherapy of Alzheimer disease. Alzheimers Res Ther 1, 6.

[90] [91]

[92]

[93]

[94] [95]

[96]

[97]

229

Fay DS, Gerow K (2013) A biologist’s guide to statistical thinking and analysis. WormBook 1-54. Chiti F, Stefani M, Taddei N, Ramponi G, Dobson CM (2003) Rationalization of the effects of mutations on peptide and protein aggregation rates. Nature 424, 805-808. Hubin E, van Nuland NA, Broersen K, Pauwels K (2014) Transient dynamics of A␤ contribute to toxicity in Alzheimer’s disease. Cell Mol Life Sci 71, 3507-3521. Bolognesi B, Kumita JR, Barros TP, Esbjorner EK, Luheshi LM, Crowther DC, Wilson MR, Dobson CM, Favrin G, Yerbury JJ (2010) ANS binding reveals common features of cytotoxic amyloid species. ACS Chem Biol 5, 735-740. Fink AL (1998) Protein aggregation: Folding aggregates, inclusion bodies and amyloid. Fold Des 3, R9-23. Kepp KP (2014) Computing stability effects of mutations in human superoxide dismutase 1. J Phys Chem B 118, 17991812. Castellani RJ, Perry G (2014) The complexities of the pathology-pathogenesis relationship in Alzheimer disease. Biochem Pharmacol 88, 671-676. Pena F, Gutierrez-Lerma A, Quiroz-Baez R, Arias C (2006) The role of ␤-amyloid protein in synaptic function: Implications for Alzheimer’s disease therapy. Curr Neuropharmacol 4, 149-163.