Dynamic disulfide exchange in a crystallin protein in the human eye lens promotes cataract-associated aggregation

Increased light scattering in the eye lens due to aggregation of the long-lived lens proteins, crystallins, is the cause of cataract disease. Several mutations in the gene encoding human γD-crystallin (HγD) cause misfolding and aggregation. Cataract-associated substitutions at Trp42 cause the protein to aggregate in vitro from a partially unfolded intermediate locked by an internal disulfide bridge, and proteomic evidence suggests a similar aggregation precursor is involved in age-onset cataract. Surprisingly, WT HγD can promote aggregation of the W42Q variant while itself remaining soluble. Here, a search for a biochemical mechanism for this interaction has revealed a previously unknown oxidoreductase activity in HγD. Using in vitro oxidation, mutational analysis, cysteine labeling, and MS, we have assigned this activity to a redox-active internal disulfide bond that is dynamically exchanged among HγD molecules. The W42Q variant acts as a disulfide sink, reducing oxidized WT and forming a distinct internal disulfide that kinetically traps the aggregation-prone intermediate. Our findings suggest a redox “hot potato” competition among WT and mutant or modified polypeptides wherein variants with the lowest kinetic stability are trapped in aggregation-prone intermediate states upon accepting disulfides from more stable variants. Such reactions may occur in other long-lived proteins that function in oxidizing environments. In these cases, aggregation may be forestalled by inhibiting disulfide flow toward mutant or damaged polypeptides.

In vivo populations of many proteins exhibit conformational and chemical heterogeneity (1)(2)(3)(4)(5)(6)(7)(8). This heterogeneity may arise from post-translational modifications, somatic mutations, or roughness of the conformational landscape or simply as a natural effect of heterozygosity. Biochemical heterogeneity is particularly important for highly abundant, long-lived (low-turnover) proteins, such as the collagens and aggrecans of the joints; elastins of the skin; and crystallins of the eye, which accumulate damage throughout the course of life (9 -11). For well structured proteins, partially populated intermediate and misfolded conformational states give rise to heterogeneous conformational ensembles in living cells, which sometimes leads to aggregation (12-15). The emergent properties of subtly heterogeneous macromolecular populations are largely unexplored, yet there is ample evidence that they can be significant (1,4,16). Perhaps the most spectacular case is the prion effect wherein polypeptides in an aggregation-prone conformational state can catalyze conformational conversion and aggregation in the natively folded polypeptides (17). Specific mutations or posttranslational modifications can enhance or suppress these aggregation-promoting interactions, e.g. in the ALS-associated enzyme superoxide dismutase 1 (18); in ␤ 2 -microglobulin, associated with dialysis amyloidosis (19); and in the yeast prion Sup35 (20).
Intramolecular disulfides are crucial for the correct folding of many proteins that function in relatively oxidizing environments, such as antibodies and hormones in the blood, transmembrane proteins, and digestive enzymes as well as a wide variety of protein toxins (21)(22)(23)(24)(25)(26)(27). At the same time, internal disulfides are well known conformational traps on the protein folding landscape, capable of stabilizing both productive folding intermediates and aggregation-prone misfolded states (28 -32). Redox chemistry is known to be crucial for many proteins' aggregation pathways. These include superoxide dismutase 1, associated with ALS (31), as well as ␤ 2 -microglobulin, the amyloid-forming protein in dialysis amyloidosis (33,34). Disulfide scrambling is an important failure mode of therapeutic antibodies (21) and may be involved in light-chain amyloidosis (35). Domain swapping is a particularly sensible mechanism for aggregation via disulfide-stabilized intermediate states, and indeed this may play a role in serpinopathies (36). Dynamic disulfide exchange among identical or nearly identical polypeptides has only been recently recognized, in the case of proteindisulfide isomerase (37), and is likely to be found in other systems. Most known proteins are capable of undergoing some reversible post-translational modifications, including a wide variety of disulfide bonds (1,4,38). Redox potential in various tissues changes over time in the course of development or across the cell cycle, due to episodes of oxidative stress, or in the course of aging (39 -41).
The eye lens is a remarkable study in proteome aging because the proteins in its core are never replaced. Their eventual misfolding and aggregation leads to cataract disease, lens opacification due to light scattering by the protein aggregates (42). Crystallins in the lens core are highly stable and soluble, resisting any long-range packing interactions, including formation of both native (crystals) and nonnative (amyloid or amorphous) aggregates (43). However, they accumulate damage over time, such as Cys and Trp oxidation as well as deamidation, truncation, and other changes (44 -47). One of the most readily formed yet most consequential post-translational modifications that accumulate with aging is formation of disulfide bonds. Cells of the human eye lens core lack nuclei or organelles and are metabolically quiescent (48,49), so their cytoplasm gets more oxidizing over time (50). Due to this lack of active metabolism, neither the redox enzymes nor the crystallins, which make up the bulk of the lens proteome, turn over (51,52). The result is progressively higher disulfide content in lens crystallins during aging (51,(53)(54)(55)(56). Oxidation of specific Cys residues to disulfides in lens crystallins correlates with the onset and progression of cataract disease (57)(58)(59)(60).
GSH, which acts as the redox buffer in young lens, becomes depleted from the lens with age (50,61,62). Reduced and oxidized forms of GSH are both depleted, whereas protein and mixed disulfides increase (58,63,64). In fact, the concentration of GSH in aged and especially cataractous lenses is 0.2-2 mM (50,63); crystallins are present in lens nuclei at up to 400 mg/ml (48), and ␥-crystallins account for 20 -25% of this total (65,66), or ϳ4 -5 mM, with 5-7 Cys residues each (depending on the type). This raises the possibility that even the redox buffering functions are transferred to the lens proteome during aging and cataractogenesis. The W42Q mutation mimics oxidative dam-age to Trp side chains (conversion to kynurenine, which is more hydrophilic) that is known to arise during aging (44,47,67); the W42R variant, whose biophysical properties are highly similar, causes hereditary cataract in humans (32, 68 -70). We report that WT H␥D 2 is able to act as an oxidoreductase, oxidizing the W42Q variant in a way that triggers its aggregation. Specifically, a pair of Cys residues in the C-terminal domain of H␥D is capable of forming a labile disulfide bond. These disulfides are dynamically exchanged among crystallin molecules under physiological conditions in solution, creating a "redox hot potato" competition: the disulfide is passed around until it is transferred to a destabilized variant where it can lock the aggregation-prone intermediate; the ensuing aggregation removes both the destabilized variant and the disulfide from solution.

WT H␥D crystallin must be oxidized to promote aggregation of the W42Q mutant
We have reported previously (71) the surprising interaction between WT H␥D crystallin and its cataract-associated W42Q mutant whereby the WT protein promoted the mutant's aggregation in a temperature-and concentration-dependent manner without aggregating itself. Formation of an internal disulfide bond was crucial for the mutant protein's ability to aggregate (32). Therefore, we explored the possibility that redox chemistry is involved in this aggregation-promoting mechanism.
We found that fully reduced purified WT protein had no activity with respect to W42Q aggregation. However, it could  42 . B, turbidity traces from mixtures of reduced W42Q with reduced (black), oxidized (red), and oxidized and re-reduced WT (yellow). The concentration of each protein was 50 M. C, whole-protein isotopically resolved mass spectra showing WT(ox.) to be 2 Da lighter than the reduced version, indicating formation of one internal disulfide bond per molecule. Theoretical isotopically resolved mass spectra based solely on the atomic composition (green and red) are overlaid on the respective experimentally determined deconvoluted isotopically resolved mass spectra (orange and blue). D, WT H␥D and variants at every Cys position were oxidized in parallel, and their ability to catalyze W42Q aggregation was tested (each protein at 50 M). Resulting turbidity traces confirmed Cys 108 and Cys 110 as the residues predominantly responsible for the aggregation-promoting activity.

Crystallin aggregation via disulfide exchange
be converted to an active form in vitro (Fig. 1B) by a commonly used oxidizing treatment with a 1:3 mixture of Cu(II) and phenanthroline (72,73). After removing the oxidant by extensive dialysis, the WT(ox.) protein efficiently promoted aggregation of W42Q; reduction of WT(ox.) with tris(2-carboxyethyl)phosphine inactivated it again (Fig. 1B). Isotopically resolved wholeprotein electrospray MS confirmed that the peak isotope mass of the WT sample prior to oxidation was 20,605.9 Da, which precisely matches the peak isotope mass predicted from the amino acid sequence (20,605.8 Da), and the overlap of isotopic distributions was also nearly perfect between the theoretical and experimentally determined mass spectra (Fig. 1C). The same experiment was carried out on the WT(ox.) sample whose activity was assayed in Fig. 1B, and the resulting molecular weight and isotope distribution both precisely matched those predicted for the protein in the case of one intramolecular disulfide bond per polypeptide (-SH ϩ -SH 3 -S-S-; hence the loss of 2 Da) following the oxidizing treatment. Thus, the oxidation was highly specific, leaving the remaining 4 Cys residues unaffected, and essentially quantitative under these conditions.

A specific disulfide bond in the WT protein enables W42Q aggregation
The native-state crystal structure of H␥D (Fig. 1A) has no disulfide bonds despite the statistically high proportion of Cys residues in this protein (6 of 173 residues). Tissue proteomic studies have revealed disulfide formation in this protein in the oxidizing cytosol of aged and cataractous lenses (58,74). At least two pairs of Cys residues (18-78 and 108-110) are sufficiently close in the native structure that a disulfide bond might form with only a short-range structural perturbation. Other disulfides have been observed following mutation or denaturation of the protein (32,74).
We therefore used site-directed mutagenesis to eliminate, one at a time, each of the 6 Cys residues on the WT background. Simultaneous oxidation of WT and Cys-mutant proteins under identical conditions showed no declines in aggregation-promoting activity, relative to WT, for any mutants except C108S and C110S, which lost all or most of this activity (Fig. 1D). Thus, the 108 -SS-110 disulfide appeared to be crucial for the aggregation-promoting activity. We were able to directly detect this disulfide in LC-MS/MS spectra of chymotryptic digests of the oxidized protein (Table S1 and Fig. S2). This was additionally confirmed by top-down proteomics with incomplete chymotrypsin/Glu-C digestion, which yielded a peak at 10,382.4101 Da (consistent with residues 65-150 with one disulfide), which fragmented to yield a peak at 6921.9584 Da (consistent with residues 93-145 with 108 -SS-110 as no other Cys residues are present between these positions). By contrast, combined tryptic/chymotryptic digests revealed that the N-terminal Cys residues (particularly Cys 18 and Cys 78 ) were mostly in the reduced state (Table S1). A smaller amount of Cys 18 -Cys 32 disulfides was observed, although detection of Cys 32 -and Cys 41 -containing peptides proved challenging, in line with previous proteomic data (45).
Notably, isotopically resolved MS of the C108S and C110S variants following oxidative treatment revealed significant frac-tions of each variant with a rightward mass shift (Fig. S3), possibly due to oxidation of a His side chain. Thus, in our experiments the 108 -SS-110 bond in H␥D appeared to act as a preferred oxidation scavenger site to avoid irreversible oxidation of other residues. Furthermore, the C108S/C110S double mutant exhibited much less of the heavier species (Fig. S3), suggesting that this form of covalent damage was cysteine-mediated in some way.

Disulfide transfer from WT to W42Q promotes W42Q aggregation
We next tested whether this newly discovered oxidoreductase activity was related to the aggregation-promoting activity of the WT protein. This was accomplished by incubating a 1:1 mixture of oxidized WT and reduced W42Q until it turned turbid, indicating insolubilization of some fraction of the latter. Then, fresh W42Q was added, and the reaction was repeated for a total of four cycles, so that by the end the total amount of W42Q added was roughly 4ϫ WT. By the fourth cycle, turbidity was visibly lower than initially. After each step, the turbid solution was centrifuged, and the pelleted fractions were washed, resuspended in sample buffer, and combined.
Whole-protein isotopically resolved ESI-TOF ( Fig. 2) revealed that the supernatant at the end of the fourth aggregation cycle contained predominantly reduced WT protein and reduced W42Q. Thus, most of the disulfide-containing WT protein had been converted to the reduced form. Analysis of the pelleted fraction proved more difficult because the proteins were in an aggregated state. The aggregates were partially resolubilized in low-pH buffer (pH 5 ammonium acetate). ESI-TOF of this resolubilized protein revealed that it consisted predominantly of W42Q with one disulfide bond (Ϫ2-Da mass shift), consistent with our initial observations (71). It follows that oxidized WT H␥D promoted W42Q aggregation by transferring a disulfide to it. The result can be summarized as a chemical reaction where the reactants WT(ox.) and W42Q are both soluble, whereas the products are soluble WT and insoluble W42Q(ox.). Aggregation of the oxidized mutant may shift the equilibrium in favor of disulfide transfer to the mutant.
The ability of W42Q to accept a disulfide bond and aggregate depended on the nature of the oxidant disulfide. Thus, GSH disulfide (GSSG) caused strong aggregation of W42Q with kinetics comparable with those observed upon addition of the oxidized WT protein. By contrast, the same amount of hen egg lysozyme had no effect on W42Q (Fig. 3A) despite the fact that lysozyme contains four disulfide bonds per molecule. We interpret the difference as arising from the much higher accessibility and/or redox potential of both the reversible Cys 108 -Cys 110 disulfide in WT H␥D and the GSSG disulfide as compared with the structural disulfides in lysozyme. It is worth noting that disulfide geometry has a strong effect on whether it is a catalytic or a structural bond (38) and that the close sequence proximity of Cys 108 and Cys 110 is likely to give it a strained, catalytic-like configuration (24,38,75).
Because Trp oxidation is a relatively rare event in the lens, the concentration of species mimicked by the W42Q substitution is expected to be relatively low compared with WT. Therefore, we also carried out aggregation experiments in a different concen-tration regime: 20 M W42Q and 100 M WT(ox.) or GSSG. Surprisingly, the relative aggregation-promoting abilities of GSSG and WT(ox.) were reversed in this concentration range, as shown in the representative traces in Fig. 3B. The reason WT(ox.) appears to be a weaker aggregation inducer than GSSG when [W42Q] is high but a stronger aggregation inducer when [W42Q] is low is the subject of ongoing investigation. As the pool of GSH becomes depleted in the lens with age, [WT(ox.)] is expected to increase, and the protein disulfides are likely to play a bigger role in promoting aggregation of damaged H␥D polypeptides.
The disulfide required for W42Q aggregation was distinct from the dynamic 108 -SS-110 disulfide in WT. LC-MS/MS (Table S4) revealed the internal disulfide in the W42Q aggregates to be predominantly 32-SS-41, consistent with previous results (32) as well as tissue proteomic analysis (58). However, given the importance of the 108 -SS-110 bond in the WT(ox.) protein, we examined whether the same bond in the mutant affected its aggregation. The W42Q/C110S double mutant aggregated comparably with W42Q at neutral pH when mixed with the WT(ox.) protein but less so at lower pH (Fig. S5A). The ability to form the 108 -SS-110 disulfide, however, appeared to confer modest aggregation resistance at lower concentrations of the oxidizing agent GSSG, perhaps because this site acted as an oxidation decoy, reducing or delaying oxidation of the N-terminal domain (Fig. S5B). Despite similarity at neutral pH, and theoretically predicted (purple) isotope distributions, and the WT(ox.) sample had been almost completely reduced back to WT because the observed peak (red; 20,605.8 Da) skewed only slightly to the left compared with the theoretically predicted isotope distribution for the fully reduced WT (black). The pooled pelleted fractions contained exclusively Ϫ2-Da shifted W42Q as evidenced by the exact overlap between the experimental data (blue; 20,545.8 Da) and theoretically predicted isotope distribution for W42Q with one internal disulfide per molecule (green). Hence, the internal disulfide was transferred from soluble, oxidized WT to the soluble, reduced W42Q during the reaction, leading to formation of insoluble, oxidized W42Q.

Crystallin aggregation via disulfide exchange
lower pH (pH 5) led to dramatically different behavior for the W42Q and W42Q/C110S variants. When mixed with the oxidizing agent Cu(II)/phenanthroline, only the single mutant aggregated at pH 5, whereas both did so at pH 7 ( Fig. S5C). This difference suggests that Cys 110 may retain its reactivity even at reduced pH; however, the mechanism of aggregation, including any potential intramolecular disulfide transfer, under these conditions remains to be investigated. Formation of 108 -SS-110 in the mutant may promote formation of the N-terminal disulfide there under low pH conditions.
To gain quantitative understanding of the distribution of the number of free thiol groups per molecule, we also resolubilized aggregates in 5% SDS at 95°C, pH 5, and used a PEGylation gel-shift assay similar to those reported previously (76 -78). Free thiol groups reacting with maleimide-conjugated polyethylene glycol (PEG) generated predictable upshifts in migration through an SDS-polyacrylamide gel. PEGylation of free thiols in the aggregated state (Fig. 4) revealed that W42Q/C110S aggregates lost two free thiols per molecule, consistent with formation of one internal disulfide bond. The W42Q aggregates contained a major one-disulfide population and a minor, but still significant, population of molecules with two internal disulfide bonds. We infer that one of those bonds is 108 -SS-110 and that its formation in the mutant is not required for aggregation.
The precise mechanism of W42Q misfolding and the precise extent of disordering of its N-terminal domain prior to 32-SS-41 formation remains to be investigated. The T m for the first unfolding transition, corresponding to unfolding of the mutated N-terminal domain, is 55°C (32). The domain contains 83 residues. Robertson  Using these values and assuming two-state unfolding of the W42Q N-terminal domain, we arrive at a predicted ⌬G unfolding of ϳ4 kcal/mol at 37°C, or unfolding probability of ϳ2 ϫ 10 Ϫ3 . Conversely, the W42Q variant contains the full complement of 6 Cys residues and hence could theoretically form any of 14 internal disulfides besides 108 -SS-110. However, only one of those disulfides, 32-SS-41, has been found in W42Q aggregates in this study, consistent with the prior report that used a different experimental procedure (32). No detectable amount of disulfide was found in the soluble fraction of W42Q ( Fig. 2A). This suggests the misfolding reaction leading to 32-SS-41 has a degree of specificity; it is likely to proceed through an intermediate with only a partially unfolded N-terminal domain.

Oxidoreductase activity of H␥D enables dynamic disulfide exchange among soluble H␥D variants
The reversible Cys 108 -Cys 110 disulfide of H␥D bears similarity to the "CXC" motif of some thioredoxins and disulfide isomerases as well as the redox-switchable chaperone Hsp33 (75,80,81), so this ␥-crystallin may have a previously unrecognized oxidoreductase function. The Cys 110 thiol group is solvent-accessible in the native structure (82), suggesting that disulfide interchange among H␥D molecules may occur in solution. It was recently reported (37) that some protein-disulfide isomerases are capable of dynamically exchanging disulfide bonds in solution, essentially forming a proteinaceous redox buffer in the endoplasmic reticulum. Given the high abundance of H␥D in the lens core (up to ϳ10 mg/ml), it could potentially act as a supplementary redox buffer as GSH levels are depleted with age.
To explore this possibility, we incubated mixtures of H␥D variants where one variant (or WT) started out as fully oxidized at the 108-110 site and the other started with those two residues fully reduced. The typical pK a of a Cys residue is ϳ8. Therefore, incubations were carried out either at pH 8 (permissive for thiol-disulfide interchange) or at pH 5 (inhibiting interchange). Disulfide transfer was determined by quantifying the distributions of the number of free thiols per molecule for each of the variants at the end of the incubation via the PEGylation/gelshift assay. As shown in Fig. 5, when reduced WT protein was incubated with the oxidized triple mutant C18T/C41A/C78A (abbreviated CCC) at pH 5, the vast majority of each respective polypeptide population was found at the migration positions expected for no disulfide exchange: the WT protein retained six free thiols per molecule (hence the prominent 6xPEG band), whereas the oxidized triple mutant had only one free thiol per molecule (hence the 1xPEG band). The band at 2xPEG was likely due to incomplete reaction between the CCC construct and the labeling reagent. The corresponding incubation mixture at pH 8 showed a significant decrease in the amount of The triple-Cys mutant was used to ensure that any contamination from the non-W42Q proteins in the aggregates could be easily seen by SDS-PAGE upon PEGylation (as ϩ1 or ϩ3 bands in the W42Q lanes and ϩ4 or ϩ6 bands in the W42Q/C110S lane, respectively). The predominant band for resolubilized monomers was at ϩ3 PEG (of 5 Cys/molecule), indicating one internal bond per molecule. Lack of a prominent minor band at ϩ1 PEG indicated that monomers with two internal disulfides were extremely rare. Right lane, for comparison, the same was done for W42Q mixed with oxidized WT (40 M each protein) at pH 8 with results very similar to those obtained at pH 5.5. Because substantial numbers of monomers with two internal disulfides were found in W42Q aggregates but not W42Q/C110S aggregates, we attribute the second disulfide to 108 -SS-110.

Crystallin aggregation via disulfide exchange
6xPEG (WT) and 1xPEG (CCC(ox.)) bands and a concomitant increase in the 3xPEG and 4xPEG bands, attributable to formation of reduced CCC mutant and oxidized WT, respectively.

Formation of the 108 -SS-110 disulfide conformationally strains the C-terminal domain
As we have demonstrated previously (32), oxidative aggregation of W42Q/R mutants proceeds at temperatures ϳ10°C below the onset of any detectable unfolding of the N-terminal domain. However, it is clear by inspection that formation of the 108 -SS-110 disulfide should introduce strain into the native structure. Most thioredoxin-type motifs are "CXXC" where XX are often Pro and Gly to enable the tight turn required to form the disulfide. In cases where CXC motifs are found, X is typically Gly (75, 80) because its range of backbone angles is the broadest. We therefore hypothesized that the CSC motif found here was likely to be strained, which accounts for the ease of disulfide exchange but may also contribute to conformational strain in the protein, potentially leading to partial unfolding.
To test this prediction experimentally, we studied both WT and the library of single-and multiple-Cys mutants in the WT background to determine whether 108 -SS-110 formation indeed strained the native state. (The W42Q variant could not be oxidized efficiently due to extensive aggregation during the oxidizing treatment as shown in Fig. S5C.) Differential scanning fluorometry experiments revealed that all variants capable of forming the 108 -SS-110 bond had decreased apparent T m values upon oxidation relative to their reduced state (Fig. 6A). These decreases were not observed in constructs lacking either Cys 108 or Cys 110 but were present in all other cases. The conformational strain disappeared upon mild reduction, indicating that it was not due to any irreversible oxidative damage, such as Met or Trp oxidation (Fig. 6B). The average downshift in DSF T m attributable to the 108 -SS-110 bond was 5.4°C with a standard deviation of 1.8°C.
We next examined whether the conformational strain is global or confined to the C-terminal domain, which contains the disulfide. Differential scanning fluorometry assumes that the reporter fluorophore, in this case SYPRO Orange, binds to all relevant molten-globule states. However, DSF measurements conducted on the W42Q mutant (Fig. 6C) revealed only one melting transition, rather than the two transitions previously reported for this mutant by calorimetry, guanidinium denaturation, and intrinsic fluorescence (32,68,70,71). The temperature of the transition was above the previously measured T m of the N-terminal domain of the W42Q protein and close to its C-terminal T m . Thus, SYPRO Orange likely bound preferentially to the molten-globule state of the C-terminal domain rather than the N-terminal domain. This surprising observation may help explain why chaperones fail to recognize the partial unfolding of the N-terminal domain in aggregation-prone, cataract-associated variants of this protein (70,83). Apparently, unfolding of the N-terminal domain did not form a sufficiently hydrophobic binding pocket. This may be due in part to some of the N-terminal domain's residues being sequestered upon partial unfolding by interaction with the C-terminal domain, for example as proposed in Serebryany et al. (32).
To confirm the effect of the disulfide strain by a complementary label-free measurement, we carried out differential scanning calorimetry experiments at pH 5 to inhibit any disulfide exchange, comparing stock, oxidized, and oxidized-and-thenreduced WT H␥D as well as identically treated C18T/C41A/ C78A triple mutant. The results are summarized in Table 1. The stock WT protein was destabilized only modestly (3-4°C) at lower pH compared with the previous neutral-pH measurements (32). However, oxidation led to a downshift of ϳ10°C in the melting transition of one domain in the WT and ϳ7°C in the triple mutant, whereas the other melting transition remained unchanged in both cases (Fig. S6). We attribute the redox-dependent melting transition to the C-terminal domain because the redox-active Cys pair is located there. The fitted calorimetric enthalpies for the two domains (shown in Table 1) were consistent with this assignment. Although these calculated enthalpies may be subject to variations in protein concentration or to aggregation during the experiment, the overall pattern was clear: enthalpy of the C-terminal domain was significantly higher than that of the N-terminal domain in reduced samples but dropped to become nearly identical to it following 108 -SS-110 formation. The oxidized-and-then-reduced samples yielded traces and melting transitions very similar to those of the never-oxidized samples, indicating that destabilization

Crystallin aggregation via disulfide exchange
due to oxidation was almost entirely redox-reversible and thus attributable to conformational strain induced by the 108 -SS-110 disulfide.
The W130E oxidation-mimicking mutation in the C-terminal domain of H␥D opens up a distinct nonnative, nonamyloid aggregation pathway, which is chaperone-suppressible but also partially redox-dependent (70). It is likely that the destabilization produced by the C-terminal mutation is in synergy with the conformational strain produced by the 108 -SS-110 bond, and thus the disulfide indirectly promotes aggregation of this cataract-associated protein. Notably, UV irradiation, one of the best characterized causes of cataract, is known to damage Trp 130 in H␥D (67), and the UV-induced aggregation pathway requires oxidation (84). The N-terminal domain of H␥D derives part of its stability from its interface with the C-terminal domain; disrupting this interface leads to N-terminal destabilization (85)(86)(87)(88). The earliest detectable unfolding of the W42Q protein occurs at ϳ43°C (32,71), so if the 108 -SS-110 bond forms in that variant and the resulting destabilization propagates to the domain interface, it may be sufficient to increase the population of aggregation-prone intermediates near 37°C and in this way allosterically promote formation of the nonnative 32-SS-41 disulfide bond in the N-terminal domain that is required for aggregation. We have not found evidence of such an allosteric effect; however, a subtle allosteric effect on the N-terminal  (32)). Right, DSF traces of W42Q showed only a single melting transition at the same temperature as seen in the C41A/C78A double mutant as well as in WT(ox.) and only ϳ5°C below that of WT. Thus, SYPRO Orange is a reporter of the melting of the C-terminal domain. A shoulder at ϳ50°C in the oxidized C41A/C78A double mutant may be due to some misfolding or aggregation during the measurement. RFU, relative fluorescence units.

Discussion
The GSH redox buffer becomes depleted in the lens nucleus with age, whereas protein disulfide content increases (50,58,(61)(62)(63)(64). Other forms of protein oxidation, such as conversion of tryptophan to kynurenine, increase at the same time (64, 67, 89 -91). We have described an unexpected synergy between these two modes of oxidative damage that results in rapid lightscattering aggregation due to the existence of a previously unrecognized oxidoreductase activity in human ␥D-crystallin. The oxidation-mimicking W42Q mutation, by partially destabilizing the N-terminal domain, enables 2 Cys residues that are distant and buried in the natively folded N-terminal domain to become solvent-exposed and transiently approach each other (32). We have now found that a redox-active disulfide in the C-terminal domain of the WT protein can then be transferred to these newly exposed N-terminal Cys residues in the mutant. The products of this bimolecular redox reaction are a soluble, fully reduced WT molecule and a W42Q molecule bearing a nonnative internal disulfide that locks its N-terminal domain in an aggregation-prone intermediate state (Fig. 7). The structure of this intermediate, and of the resulting aggregated state, has been proposed (32). The intermediate structure, detachment of the N-terminal ␤-hairpin, is consistent with that derived from single-molecule force spectroscopy (92), and evidence of the specific disulfide bond that traps this intermediate has been found in patient lenses and correlates strongly with age-onset cataract (58). It is worth noting that the latter two studies were carried out on unmutated proteins and that a large variety of congenital cataract-linked point mutations in the ␥-crystallins (including the W42R mutation) cluster near this N-terminal hairpin (43). These observations raise the possibility that the W42Q mutation merely increases the population of a conformational intermediate that is already accessible, albeit extremely rare, in the WT protein and that other mutations and age-related post-translational modifications converge on the same aggregation-prone intermediate. In this model, the fact that the oxidized WT H␥D does not aggregate on experimentally measurable timescales is simply due to the high kinetic and thermodynamic stability (93,94) of its N-terminal domain.
The current study does not address the detailed mechanism by which H␥D or its W42Q variant misfolds to form an aggregation-prone 32-SS-41 conformer. However, data in Fig. 3 suggest that the WT(ox.)/W42Q interaction is not limited to disulfide transfer. Specifically, W42Q at 50 M aggregates much more rapidly with GSSG than with WT(ox.) (Fig. 3A), but W42Q at 20 M aggregates much more rapidly with WT(ox.) than with GSSG (Fig. 3B). This surprising switch in relative efficacy of aggregation inducers suggests that WT(ox.) serves not only as an oxidant but also as an aggregation catalyst at low [W42Q]. We have previously hypothesized that an aggregation-promoting binding interaction between WT and W42Q may exist (71). Whether and how such an interaction affects misfolding and disulfide transfer to W42Q and other cataractassociated variants will be the subject of future research.
The specific model proposed here postulates transfer of the Cys 108 -Cys 110 disulfide from WT H␥D to a nonnative N-terminal disulfide (Cys 32 -Cys 41 ) of the W42Q mutant. Of the 6 Cys residues in H␥D, five are well conserved among the ␥-crystallins; Cys 110 is the exception, being present in humans and chimpanzees but not in chicken, mouse, rat, or bovine ␥D sequences (95). Remarkably, however, each of those species has a Cys residue in position 110 in the closely related ␥C-crystallin, whereas human and chimpanzee versions of ␥C have Ser 110 instead. Thermodynamically, H␥D is more stable than H␥C (43), and human lens proteomics has revealed that H␥D persists longer than H␥C, at least in its soluble form, becoming the most abundant soluble ␥-crystallin by age ϳ55 (11). Thus, the apparent evolutionary switch of the Cys 108 -Cys 110 pair from ␥C to ␥D The WT, undamaged protein (left) may become oxidized, forming a conformationally strained and hence redox-active disulfide. This disulfide can then migrate to either a reduced WT or WT-like protein in a fully reversible reaction (blue arrow) or a damaged or mutated protein (here indicated by a perturbed native interaction in the N-terminal domain) (green arrow). In the latter case, it may migrate to a different location in the molecule, forming a conformationally relaxed and hence redox-inactive disulfide between Cys residues that are normally buried in the native state. This nonnative disulfide then traps a partially unfolded intermediate in a reaction that is essentially irreversible under mildly oxidizing conditions. The intermediate cannot revert to the native state and therefore partitions to aggregated structures. We have previously proposed a structural model for this aggregated state (32). C-td, C-terminal domain.

Crystallin aggregation via disulfide exchange
might be a consequence of increased lifespan of the organism. At the same time, it is worth noting that at least one study identified other sequence-proximal disulfides in the ␥-crystallins: the Cys 78 -Cys 79 bond in H␥C and the Cys 22 -Cys 24 bond in H␥S (74). It is not yet known whether these bonds are dynamic. Conversely, previous research also identified other nonnative internal disulfides in cataractous lenses, particularly in ␤-crystallins (57,59), as well as one such bond in the lens-specific chaperone ␣A-crystallin, which appeared to diminish its chaperone activity as well as contribute to aggregation (60,96). More recently, many disulfide bonds were discovered in noncrystallin proteins in the lens (97). It is possible that many of these longlived lens proteins fall within the redox "hot potato" model we have proposed.
In redox-active proteins with a CXC or CXXC motif, the residues in the X positions are typically Gly, whose flexibility can best accommodate the conformational strain from the disulfide bond. In H␥D, the X residue is Ser instead. Although this is the second most flexible amino acid after Gly (98), its presence is likely to produce a more strained disulfide with a higher redox potential than in the Cys-Gly-Cys disulfide isomerases. Nevertheless, a very recent reanalysis of proteomic data by Ramkumar et al. (99) revealed that 108 -SS-110 disulfides in H␥D are highly abundant in the human lens even prior to the onset of cataract.
What, then, is the possible fitness benefit of these sequenceproximal and likely redox-active disulfides? A possible explanation is that GSH levels become depleted in lens tissue during aging (50), and the lens core gradually becomes impermeable even to GSH generated in the lens cortex or present in vitreous humor (62), resulting in increased disulfide formation in lens cytoplasmic proteins (97). The ␥-crystallins have always been thought of as purely structural proteins with no known biochemical function aside from their structural stability and optical properties (42,43,48). The search for the aggregationpromoting mechanism led us to a previously unrecognized oxidoreductase activity in WT H␥D. Dynamic exchange of disulfide bonds among the protein molecules makes reduced and oxidized forms of H␥D a redox couple. Given the high abundance of ␥-crystallins, we speculate that they may constitute a protein-based redox buffer in the aging, GSH-depleted lens. If so, there is some evidence that this buffering capacity is enzymatically regulated in vivo. Early proteomic studies of the human eye lens indicated that Cys 110 of H␥D is partially (ϳ30 -70%) methylated (89). The significance of this observation was not known at the time but can now be interpreted as a likely regulation mechanism for the protein's oxidoreductase activity. Like Cys 110 of H␥D, Cys 79 of H␥C and Cys 24 of H␥S have also been found to be partially methylated in vivo (89,100), suggesting a common physiological regulation mechanism. Cysteine methylation is an enzyme-catalyzed reaction in vivo (101). Our biochemical observations thus open up the possibility that an enzymatic pathway in the lens core actively regulates the redox buffering capacity of the crystallin proteome and may influence the age of onset of cataract.
Moreover, we have revealed a likely failure mode of such a protein-based redox homeostasis system: although sequenceproximal disulfides are kinetically favorable, nonnative disul-fides are favored thermodynamically due to their propensity to aggregate. Destabilizing mutations or modifications of the H␥D core, such as oxidation of Trp residues, may generate disulfide sinks analogous to the W42Q mutant. In general, we propose a redox hot potato mechanism in which polypeptides with lower kinetic stability (due to mutations or post-translational modifications) become trapped in aggregation-prone intermediate states and insolubilized as a consequence of accepting disulfides from more stable variants. The result is, essentially, a kinetic stability competition among protein variants competent for disulfide exchange with the conformational "weakest link" variant driven into an aggregated state upon receiving a disulfide. To participate in such a competition, the main features of the protein should be 1) the ability to exchange disulfides, or perhaps other reversible modifications, under physiologically relevant conditions and 2) a folding landscape that contains aggregation-prone intermediates that become more populated upon receiving the modification, in this case a disulfide bond, effectively serving as a sink for modified proteins. This process may have played a role in the evolution of the exceptional stability of the ␥-crystallins or other long-lived proteins.

Site-directed mutagenesis
Site-directed mutagenesis was carried out using either the QuikChange II kit (Agilent, Santa Clara, CA) or the Q5 kit (New England Biolabs, Ipswich, MA) according to the manufacturers' instructions. Resulting plasmids, amplified in the XL-1 Escherichia coli strain (Agilent), were confirmed by sequencing and transformed into the BL21-RIL strain of E. coli for expression.

Protein expression and purification
Protein expression and purification were carried out largely as described (32,71) with some modifications. Briefly, overnight starter cultures of BL21-RIL E. coli were inoculated in 10 -40 ml of reconstituted SuperBroth medium (Teknova, Hollister, CA) supplemented with ampicillin and chloramphenicol. Expression cultures using the same medium in standard 2-liter flasks were inoculated at 1:100 from the starters and grown at 37°C with shaking at 250 rpm for 6 -8 h until reaching stationary phase (A 600 ϳ 2). Cultures were then chilled to 18°C for 30 min and induced overnight at 18°C by adding isopropyl 1-thio-␤-D-galactopyranoside (Promega, Madison, WI) to 1 mM final concentration. Harvested cells were resuspended in the presence of protease inhibitors (Complete-mini, EDTA-free, Roche Applied Science) and stored at Ϫ70°C until used.
Following lysis by sonication in the presence of DNase and lysozyme, purification was carried out by salting out with ammonium sulfate as described (71). The 50% ammonium sulfate supernatant was dialyzed at 4°C overnight against 4 liters of sample buffer (10 mM ammonium acetate, 50 mM NaCl) to remove the ammonium sulfate. The dialysis step typically caused significant precipitation, attributable to peptidoglycan contaminants. Following centrifugation, these samples were then further scrubbed of lipids and peptidoglycans by passage through an anion-exchange column comprising two 5-ml Q-Sepharose columns (GE Healthcare) in tandem equilibrated in sample buffer. Crystallins eluted in the flow-through. The

Crystallin aggregation via disulfide exchange
ion-exchanged samples were concentrated using VivaSpin Turbo 5000 molecular weight cutoff centrifugal filters (Sartorius, Göttingen, Germany), loaded onto a Superdex 75 26 ϫ 600-mm column (GE Healthcare), and separated at 2-3 ml/min flow rate at room temperature. Resulting fractions were collected and stored at 4°C. SDS-PAGE (Criterion, Bio-Rad) was used to determine Ͼ95% purity of the samples. Proteins with the W42Q mutation were often found to have a minor degradation product at ϳ10 kDa in the gel, but its presence did not noticeably affect the overall aggregation behavior. Proteins were concentrated for storage at 4°C to 100 -500 M; concentrations were determined by A 280 using a NanoDrop 2000 instrument (Thermo Fisher, Waltham, MA).

In vitro oxidation
In vitro oxidation to form the 108 -SS-110 disulfide bond was carried out by mixing 20 M CuSO 4 and 60 M phenanthroline (MilliporeSigma, Burlington, MA) in sample buffer at room temperature to chelate the copper and then adding the protein to 20 M final concentration. This mixture, typically in 5-10-ml volume, was incubated at room temperature for 1-2 h in closed 15-ml conical tubes followed by addition of 1 mM EDTA and three rounds of dialysis through metal-free Spec-traPor 7 dialysis membrane (Spectrum Labs, Breda, The Netherlands) at 4°C against sample buffer for at least 3 h at a time (at least once overnight). The first round of dialysis included 1 mM EDTA in the sample buffer to fully chelate and remove the Cu 2ϩ ions from the sample and avoid any copper-induced aggregation or oxidation during the subsequent experiments (102). To ensure that biochemical differences between the oxidized and reduced samples were due to reversible oxidation, i.e. disulfide formation, oxidized samples were subjected to mild reducing treatment, 1 mM DTT for 2 h at 37°C, generating the oxidized 3 reduced controls.

Aggregation assay
Aggregation assays were carried out at 37°C unless otherwise indicated using half-area clear polypropylene plates (Greiner Bio-One North America, Monroe, NC) in a Power-Wave HT plate reader (BioTek, Winooski, VT) in a 100-l volume without shaking. ϳ20% of the sample typically evaporated during the 3-h aggregation experiment. The resulting path length was ϳ0.5 cm, so the reported turbidity values are expected to be ϳ50% lower than in the typical 1-cm-pathlength spectrophotometer cuvette. Aggregation occurred on a comparable time scale in other reaction vessels, such as capped microcentrifuge or thermocycler tubes.

Differential scanning fluorometry
Differential scanning fluorometry was carried out using a 1:1 mixture of sample buffer and sodium phosphate (pH 7, 100 mM) buffer to maintain neutral pH during heating in 96-well format in a CFX96 RT-PCR instrument (Bio-Rad). 1ϫ SYPRO Orange dye (Life Technologies) was added as the hydrophobicity probe; control (no-protein) samples were subtracted. The temperature ramp was 1°C/min between 25 and 95°C. Melting temperatures were defined as the minima (rounded to the nearest°C) of the derivative of the empirically fit sigmoid functions in the CFX Manager software (Bio-Rad).

Differential scanning calorimetry
Differential scanning calorimetry was carried out as reported previously (32) except in pH 5 buffer (20 mM ammonium acetate, 50 mM NaCl) to inhibit disulfide exchange during the measurement. Proteins were at 25 M concentration. Samples were kept at 4°C until immediately prior to analysis, which ran from 25 to 95°C. Each melting trace was fitted to a sum of two twostate scaled models using NanoAnalyze software (TA Instruments, New Castle, DE).

PEGylation assays
PEGylation assays were carried out as follows. Samples were denatured for 5 min at 95°C in pH 5 buffer (50 mM ammonium acetate) with 5% (w/v) SDS at [protein] of ϳ30 M in a total volume of 10 l per sample. Once cooled to room temperature, 10 l of pH 8 buffer (100 mM sodium phosphate) was added to neutralize followed by 6 l of 1 mM maleimide-conjugated PEG, M r ϳ5000 (MilliporeSigma), and 4 l of 4ϫ NuPAGE SDS-PAGE gel-loading buffer (Thermo Fisher). The reaction mixtures were incubated at 50°C for 2 h and then analyzed directly by SDS-PAGE with Coomassie stain (Thermo Fisher).

Isotopically resolved intact mass determination
Isotopically resolved intact mass determination was accomplished by electrospray ionization MS. The protein samples were analyzed on a Bruker Impact II q-TOF mass spectrometer equipped with an Agilent 1290 HPLC. The separation and desalting were performed on an Agilent PLRP-S column (1000 Å, 4.6 ϫ 50 mm, 5 m). Mobile phase A was 0.1% formic acid in water, and mobile phase B was acetonitrile with 0.1% formic acid. A constant flow rate of 0.300 ml/min was used. Ten microliters of the protein solution was injected and washed on the column for the first 2 min at 0% B, diverting nonretained materials to waste. The protein was then eluted using a linear gradient from 0% B to 100% B over 8 min. The mobile phase composition was maintained at 100% B for 1 min and then returned to 0% B over 0.1 min. The column was re-equilibrated to 0% B for the next 5.9 min. A plug of sodium formate was introduced at the end of the run to perform internal m/z calibration to obtain accurate m/z values. The data were analyzed using Bruker Compass DataAnalysis TM software (Version 4.3, Build 110.102.1532, 64 bit). The charge-state distribution for the protein produced by electrospray ionization was deconvoluted to neutral charge state using DataAnalysis implementation of the maximum entropy algorithm. Predicted isotope patterns were calculated at a resolving power of 50,000 and compared with isotopically resolved neutral mass spectra calculated using maximum entropy from the experimental charge-state distribution.

Disulfide mapping
Disulfide mapping was carried out using LC-MS/MS on an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher) equipped with an EASY NanoLC 1000 pump (Thermo Fisher). Peptides were separated onto a 100-m-inner diameter Crystallin aggregation via disulfide exchange microcapillary Kasil frit trapping column packed first with ϳ5 cm of C 18 Reprosil resin (5 m, 100 Å; Dr. Maisch GmbH, Ammerbuch-Entringen, Germany). Separation was achieved through applying a gradient from 5 to 27% acetonitrile in 0.1% formic acid over 90 min at 200 nl/min. Electrospray ionization was enabled through applying a voltage of 2 kV using a homemade electrode junction at the end of a Thermo Nano Viper 1200 75-m ϫ 550-mm column. The Fusion Lumos was operated in data-dependent mode for the MS methods. The MS survey scan was performed in the range of 395-1,800 m/z at a resolution of 6 ϫ 10 4 with HCD fragmentation in the Fusion trap using a precursor isolation width window of 2 m/z. HCD collision energy was set at 32 V, isolation window was 3 Da with 30,000 Orbitrap, resolution for MS2 electron transfer and higher-energy collision dissociation scan was 200 ms, activation time was 10 ms, and automatic gain control was set to 50,000. Ions in a 10-ppm m/z window around ions selected for MS2 were excluded from further selection for fragmentation for 60 s.
WT(ox.) samples were treated with either chymotrypsin alone or the combination of trypsin and chymotrypsin. Samples were not carbamidomethylated or reduced; rather the pH of the digestion mixture was kept neutral or below at all times to minimize disulfide scrambling as in our previous report (32). W42Q aggregates were pelleted by centrifugation at 12,000 ϫ g for 5-10 min, rinsed with 3 ml of sample buffer, and then resuspended in pH 5 ammonium acetate buffer. The suspension was centrifuged again, the supernatant was collected as the "pellet 1" fraction, and the remaining solid material was designated "pellet 2." Both were then denatured in SDS (the pellet 2 fraction could not be fully solubilized even in 5% SDS at 80°C) and trypsinized using the suspension trapping (STrap) method (103) in spin column format (Protifi, Huntington, NY) with samples collected at 15, 45, and 90 min of incubation. Results were analyzed using Comet software and additionally verified by probability scores using PeptideProphet software with searches against the relevant subset of the CRAPome database of common contaminants (104), such as human keratins, within the Trans-Proteomic Pipeline. Only peptides with probability scores Ͼ0.95 were included in the analysis.
Top-down proteomics were used to confirm disulfide assignments using the same equipment as described above. WT(ox.) sample was codigested with chymotrypsin and Glu-C under incomplete-digestion conditions. Fragment MS1 spectra were deconvoluted by maximum entropy and assigned on the basis of their molecular weight as well as the weights of subfragments arising from in-source decay.