Mechanism of Chaperone Function in Small Heat Shock Proteins

To elucidate the mechanism of αA-crystallin chaperone function, a detailed thermodynamic analysis of its binding to destabilized, site-directed mutants of T4 lysozyme was carried out. The selected mutants form a ladder of stabilities spanning the 5–10 kcal/mol range of free energy of unfolding. The crystal structures of the majority of the mutants have been previously determined and found to be similar to that of the wild type with no evidence of static local unfolding. Complex formation between αA-crystallin and T4 lysozyme was observed directly via the changes in the electron paramagnetic resonance lineshape of a nitroxide introduced at a non-destabilizing, solvent exposed site in T4 lysozyme. αA-Crystallin differentially interacts with the mutants, binding the more destabilized ones to a larger extent despite the similar structure of their native states. Our results suggest that the states recognized by αA-crystallin are non-native excited states distinct from the unfolded state. Stable complexes are formed when the free energy of binding to αA-crystallin is on the order of the free energy associated with the transition from the excited state to the native state. Biphasic binding isotherms reveal two modes of interactions with distinct affinities and stoichiometries. Highly destabilized mutants preferentially bind to the high capacity mode, suggesting conformational preference in the use of each mode. Furthermore, binding can be enhanced by increased temperature and pH, which may be reflecting conformational changes in αA-crystallin oligomeric structure.

In the crowded molecular environment of the cell, protein folding, stability, and solubility are critically dependent on a specialized protein machinery consisting of multiple superfamilies of heat shock proteins, the molecular chaperones. By recognizing and binding non-native protein states, chaperones protect the cell from the toxic and pathogenic consequences of protein aggregation (1,2). These non-native states are not only a byproduct of the folding process but are also populated through fluctuations in the native structure, the rate of which increases under extreme physicochemical conditions. The diversity of the molecular chaperones has been associated with distinct roles at different stages of the protein life cycle and in response to different cellular conditions (3).
The small heat shock protein (sHSP) 1 superfamily consists of oligomeric proteins of 4 -40 subunits with a molecular mass of less than 40 kDa/subunit (4 -6). The exact role of the sHSP in the heat shock response is poorly understood, although they have been associated with a spectrum of functions under permissive temperatures (7)(8)(9)(10). In humans, six sHSP have been identified in various tissues (6). The ␣-crystallins are the main protein component in the lens, where they play a critical role in establishing and maintaining its optical properties (11)(12)(13). It has been hypothesized that ␣-crystallin chaperone function prevents the early aggregation and precipitation of lens proteins. In non-lenticular tissues, compelling evidence suggests an important role for sHSP in transduction pathways related to cellular growth and differentiation (10, 14 -17). Inherited mutations in these proteins result in pathogenic conditions that include cataracts and desmin-related myopathy (18 -21).
Two functional characteristics distinguish sHSP from other molecular chaperones such as members of the heat shock protein 60 family. The first is a higher capacity in binding nonnative proteins. Binding stoichiometry by some sHSP reaches one substrate of equal molecular weight per subunit (13,22). This is contrasted to the one chain per 14-subunit oligomer of GroEL (2). The second unique characteristic is the lack of ATP hydrolysis by sHSP, which brings into question the fate of the bound polypeptides. Recent studies indicate that proteins bound to sHSP can be refolded in the presence of ATP-dependent chaperones (23,24). Unlike the chaperonins, the function of which is to supervise the folding process, the high capacity and absence of energy consumption make sHSP proteins an efficient defense mechanism under stressful conditions that promote the indiscriminate unfolding of cellular proteins and where the use of ATP-dependent chaperones might be wasteful of cell resources (23,24).
The detailed mechanism of sHSP chaperone function is not well understood. Most previous studies relied on assays where the native state of the substrate protein is thermally or chemically destabilized causing aggregation and precipitation. The functional characteristics of a sHSP are inferred from its efficiency in suppressing aggregation, as reflected in the reduction of light scattering by the substrate protein or the partial recovery of function upon the return to permissive temperatures.
Despite their convenience and usefulness in establishing some of the basic aspects of the in vitro function of sHSP, aggregation-based assays are inadequate for the thermodynamic and kinetic dissection of the sHSP chaperone function. Because these assays are carried out under non-equilibrium conditions, it is not possible to measure affinities and stoichiometries. Furthermore, the experimental observable, the reduc-tion in light scattering, reflects in part the kinetic competition between self-association of the substrate and the binding to the chaperone (25). This has a confounding effect on the interpretation of the changes in apparent chaperone activity of sHSP mutants and their differential interactions with substrates. In addition, the strongly denaturing conditions used result in a conformationally heterogeneous ensemble of states that is not amenable to structural analysis and may simultaneously destabilize or inactivate the sHSP. The development of a mechanistic model of sHSP chaperone function requires assays that allow the separation of the multiple events that underlie the interaction with non-native substrates. The direct detection of complex formation is a central feature of such assays (25,26).
Fundamentally, sHSP should recognize and bind proteins prior to nucleation of aggregation, i.e. should detect the increased excursion of destabilized proteins toward the aggregation-prone non-native states. This implies that binding can be induced by a progressive reduction in the folding equilibrium constant of a particular substrate until a stable complex is formed. Such incremental changes in protein stability can be achieved by site-directed mutagenesis. The advantage of this approach is that the substrate itself remains predominantly in the native state. Furthermore, the free energy of folding of the mutants can be characterized as a function of physicochemical variables such as pH and temperature and correlated with changes in the binding to the chaperone.
This study uses this approach to gain insight into the mechanism of ␣A-crystallin chaperone function using T4 lysozyme as a substrate. A set of mutants spanning the 5-10 kcal/mol range of free energy of unfolding, ⌬G unf , was constructed. Binding to ␣A-crystallin was detected using a nitroxide spin label introduced on the solvent-exposed surface of the substrate (25,26). ␣A-Crystallin binds T4L mutants using two modes having different affinities and stoichiometries. The two modes have distinct energetic thresholds for binding and may be associated with the recognition of different conformational states of T4L. These results are discussed in the context of the coupling between the dynamics of the oligomeric structure and the chaperone function in sHSP. The knowledge of the structure and folding of the T4L variants provides new insight into the recognition and binding events and the energetics associated with them.

EXPERIMENTAL PROCEDURES
Construction, Expression, and Purification of T4L Mutants-Mutants of T4L were constructed using a single-step polymerase chain reaction strategy as previously described (27). The mutants are designated by specifying the original residue, the residue number, followed by the new residue. All mutants were inserted into the PBR322 vector, except for L99A/F153A and M102K, which were inserted into pET 20bϩ to increase their yield.
Expression of T4L mutants was carried out as described by Mchaourab et al. (27) with modification of the expression temperatures. For M102K, an induction temperature of 24°C was used, whereas the expression of L99A, D70N, and L99A/A130S was induced at 30°C. Because of its lower stability, L99A/F153A formed inclusion bodies regardless of the expression temperature.
All mutants except L99A/F153A were purified using cation exchange chromatography on a Resource S column followed by spin-labeling with a 10-fold excess of reagent I (27), as shown in Scheme 1.
The reaction was allowed to proceed overnight and the mixture further purified by size exclusion chromatography (SEC) on a Superdex 75 column equilibrated with the appropriate buffer in the 6 -8 pH range. The desired pH at different temperatures was achieved by using different combinations of Mes and Tris without further adjustment. We found the SEC step critical for T4L mutants expressed at lower levels. L99A/F153A was refolded following a protocol kindly provided by Dr. B. W. Matthews. Briefly, inclusion bodies consisting of L99A/F153A were solubilized in a pH 3 urea solution. Refolding was initiated by a rapid dilution into a pH 7 buffered solution containing 10% glycerol. Spin labeling was carried out following refolding and the sample processed on the Superdex column. All spin-labeled mutants were concentrated using Filtron Microsep 10K. Protein concentrations were determined using an extinction coefficient of 1.228 cm 2 mg Ϫ1 . Relative protein concentrations were verified using the Bradford assay.
Expression and Purification of ␣A-Crystallin-Recombinant ␣A-crystallin was expressed, purified, and characterized as previously described by Berengian et al. (28).
Thermodynamic Analysis of the T4L Mutants-Unfolding curves for each mutant were constructed by monitoring tryptophan fluorescence at 320 nm as a function of urea concentration at different temperatures and pH on a Spex L-format spectrofluorometer. The curves were then fit to a two-state unfolding model using nonlinear least squares methods as described previously (29). Six parameters were used to fit the data: a slope and an intercept for the pretransition and posttransition regions and ⌬G unf and m for the transition region.
EPR Spectroscopy-EPR analysis of spin-labeled T4L was carried out on a Bruker E500 spectrometer. Samples consisting of 50 M T4L and varying concentrations of ␣A-crystallin were loaded in 25-l glass capillaries. The temperature of the cavity was maintained using a stream of nitrogen gas. Samples were incubated in a water bath at the desired temperature for 100 min and then transferred to the EPR cavity 15 min prior to initiation of data collection.

RESULTS
General Methodology-T4L was chosen as a model system because of the vast amount of structural and thermodynamic data available (30,31). This relatively small protein is devoid of disulfides, and its folding equilibrium is two-state over a wide range of temperatures, pH, and ionic strengths. More importantly, a library of site-directed T4L mutants was constructed in the context of a systematic analysis of the determinants of folding and stability (30). The thermodynamic consequences of these mutations were determined via thermal denaturation circular dichroism spectroscopy, and for many mutants the structural consequences were analyzed via x-ray crystallography. From this library of mutants, a subset was selected for this study such that their ⌬G unf were separated by ϳ1 kcal/mol (see Table I). Although not all the mutants selected have known crystal structures, there is compelling evidence supporting a WT-like structure.
The D70N mutant was chosen primarily because of its pHindependent stability (32). In WT T4L, a partially buried salt bridge between Asp-70 and His-31 results in a pH-dependent stability that peaks at pH 5.7 and decreases at higher pH as the ionization state of both side chains is changed. The D70N substitution reduces the overall ⌬G unf and removes the pH dependence in the 6 -10 range. NMR and CD spectroscopies in SCHEME 1 conjunction with functional analysis of D70N are consistent with a native-like structure (32). The L99A and L99A/F153A mutants have hydrophobic core mutations that reduce the buried surface area, resulting in the enlargement of preexisting cavities (33). The crystal structures of both mutants have been determined. Aside from minor rearrangement in the core, the backbone fold of L99A is virtually identical to the WT. In contrast, more significant changes in structure are associated with the F153A replacement. The structure in the vicinity of Ala-153 relaxes to reduce the resulting cavity. The structural changes seen in the L99A/F153A double mutant are a combination of those seen in the single mutants. Nevertheless, despite the repacking in the core, the crystal structure of this mutant confirms the lack of any local unfolding in the native state.
In WT T4L, Met-102 is located in a buried helix near the center of the C-terminal domain of T4L. The substitution with a lysine residue preserves the overall size of the side chain while introducing a charge. The crystal structure of this mutant is very similar to the WT except for the increased mobility of an ␣-helix consisting of residues 108 -113 (34). It is noted that the increased motion does not significantly change its average position in the static structure. The pK a of the introduced lysine is in the 6 -7 range, resulting in an increase in the stability of this mutant at basic pH in contrast to WT T4L.
The L99A/A130S mutant combines two mutations constructed separately by Matthews and co-workers. A130S was designed to determine the consequences of burying a hydroxyl group at the interface of two helices (35). In the crystal structure of this mutant, the serine side chain is accommodated with little indication of strain or distortion. The A130S substitution results in a 1.3 kcal/mol decrease in ⌬G unf .
In addition to the destabilizing mutations, all T4L variants contained a cysteine residue to allow the attachment of a nitroxide spin label. The cysteine was introduced at residue 151 on the exposed surface of helix J except for M102K, where the spin label was introduced at the exposed site 131 in the middle of helix H. Table I reports the thermodynamic stabilities of the spinlabeled T4L mutants used in this study. ⌬G unf was measured using denaturant unfolding tryptophan fluorescence at various temperatures as detailed under "Experimental Procedures." Measurements of the ⌬G unf of the mutants lacking the cysteine residue confirm that the introduction of the spin label does not result in further changes in stability (data not shown). Although the ⌬G unf of the mutants has not been previously reported, the deviation from that of the WT, ⌬⌬G unf , was obtained via thermal melting studies and is in close agreement with the results of Table II (30). For L99A/A130S, the change in stability of this double mutant reflects the additive effects of each mutation consistent with the conclusion that its structure is similar to that of L99A.
␣A-Crystallin Is a "Sensor" of Protein Stability- Fig. 1 (a and  b) compares the EPR spectrum of spin-labeled T4L L99A in solution to that obtained in the presence of 10-fold molar excess ␣A-crystallin subunit following incubation at 23°C at pH 7.2 in a buffered solution containing 50 mM NaCl. Because the spin label was introduced at the same site in four of the five mutants, the resulting EPR spectra in solution are identical. The relatively sharp lineshape reflects both the tumbling of T4L in solution with a correlation time of ϳ6 ns and the rotational isomerization of the nitroxide around the bonds tethering it to the protein. However, in the presence of ␣A-crystallin, a distinct second component is observed. This component, labeled 1 in Fig. 1, reflects a motionally restricted population of nitroxides and arises from the fraction of T4L associated with ␣Acrystallin. This was confirmed by size exclusion chromatography analysis of the sample to separate the high molecular weight fraction from the T4L fraction followed by SDS-PAGE (data not shown). The complex appears to be relatively stable on the time scale of the SEC experiment, although as discussed below the binding is reversible. Furthermore, upon incubation of the L99A sample at 37°C, a complete conversion to the immobilized spectral component occurs as shown in Fig. 1c. SEC analysis of this sample reveals that almost all the T4L is bound to ␣A-crystallin, confirming the spectral interpretation. The EPR spectrum of M102K labeled at site 131 is similar to that of the 151 labeled mutants (data not shown). Binding to ␣Acrystallin results in a spectral component corresponding to bound T4L with similar EPR parameters to spectrum c in Fig. 1.
In Fig. 1, the spectral component labeled 2 arises from the fraction of T4L that is free in solution. Therefore, aside from a minor correction (see below), the amplitude of this component is a measure of the fraction of the unbound T4L. Comparison of the amplitude of this component between the spectra of the mutants obtained in the presence of a 10-fold molar excess of ␣A-crystallin reveals increased binding for the more destabilized mutants as detailed in Table II. The almost complete binding of the mutant L99A/F153A demonstrates that at the stoichiometry employed ␣A-crystallin can bind all the T4L available. This result can be interpreted to yield two conclusions. First, the binding is reversible at room temperature. Irreversible binding would have resulted in an equal bound fraction for all mutants (i.e. ϳ90% bound). Based on this inter- pretation, the different extent of binding reflects the presence of a kinetic bottleneck that slows down binding in the least bound T4L variant. However, after 2 h of incubation, binding of all mutants leveled off. Second, because all T4L mutants have similar structures, ␣A-crystallin is recognizing either the increased population of the unfolded state (implied by the lower stability) or increased populations of non-native kinetic intermediates. As expected, large scale aggregation is not a requirement for binding. All mutants were soluble for days under the conditions of the binding assay.
Analysis of Binding Isotherms: Two Modes of Binding-The equilibrium nature of the binding process allows measurement of an equilibrium binding constant and determination of the number of available binding sites. For this purpose, a constant amount of the spin-labeled T4L mutant was titrated with increasing amount of ␣A-crystallin. For every ratio the molar fractions of bound and free T4L were determined by measurement of the low field sharp feature highlighted in Fig. 1 (labeled 2). The amplitude was corrected by subtraction of the contribution arising from low levels of unreacted labels (Ͻ5%), the amount of which varied between preparations of T4L. This population, which is spectrally distinct from both component spectra shown in Fig. 1, was easily observed under conditions where all the T4L was in the bound state.
Among the T4L mutants, only the three most destabilized, namely L99A/F153A, M102K, and L99A/A130S, bind to the extent that an isotherm can be constructed at 23°C, pH 7.2. Fig. 2a shows the binding isotherm of L99A/A130S. At pH 7.6, an isotherm for L99A 2 can be constructed, and Fig. 2b compares the binding isotherms of L99A and L99A/A130S at this pH. All three curves are linear and can be fit to Equation 1, where r is the ratio of bound T4L to total ␣-crystallin, L is the fraction of free native state T4L, n is the number of binding sites, and K ba is the binding constant.
Because ␣A-crystallin does not recognize the native state of T4L, K ba is an apparent binding constant that in principle depends on the intrinsic binding constant as well as on the equilibrium constant of the transition between the native state and the state that is recognized.
The superimposed linear fit in Fig. 2 yields a higher K ba (Table III) for the L99A/A130S mutant than for the L99A mutant, consistent with its higher degree of destabilization. Similarly, the apparent affinity for L99A/A130S is higher at pH 7.6 compared with pH 7.2. The origin of this effect will be discussed below. In all isotherms, the number of available binding sites is essentially constant with an average of about 0.25 T4L per ␣A-crystallin subunit.
The binding isotherm of M102K under similar conditions is shown in Fig. 3a. Unlike L99A and L99A/A130S, a distinct change in the slope and intercept is observed in the region of the plot corresponding to high molar ratios of T4L to ␣Acrystallin. This result suggests the presence of two modes of binding with different affinities and/or number of binding sites. Although it is not possible to accurately fit the second component, the data suggest that the second binding mode has a lower affinity but a higher number of binding sites. Because of the larger error associated with the determination of the bound fraction in the low 1/L region of the plot, the isotherm of M102K, as well as other curved isotherms, were determined at least three times using different preparations of ␣A-crystallin and T4L. It is noted that the ␣A-crystallin-T4L complexes obtained at both high and low ratios of T4L to ␣A appear to have similar apparent molecular masses as determined by SEC (data not shown). Fig. 3b shows the binding isotherm of L99A/F153A to ␣Acrystallin obtained under the same conditions of Fig. 2. Although the curve is linear, the number of binding sites is strikingly different. While the average number obtained from analysis of Fig. 2 is 0.25, the fit of Fig. 3b results in 1.03 binding sites/␣A-crystallin subunit, suggesting that the main mode of binding for this mutant is the high capacity mode hinted at in the isotherm of M102K. The change in the two parameters characterizing binding supports the interpretation of the M102K isotherm as reflecting the presence of two modes of binding. The lack of contribution by the high affinity mode suggests preferential binding of L99A/F153A by the high ca-2 ⌬G unf of L99A at pH 7.5 is 7.4 kcal/mol. Effect of Temperature on the Extent of Binding-The comparative analysis of the binding isotherms of L99A and L99A/130S at 30°C, shown in Fig. 4, is consistent with this conclusion. The former is linear, indicating a single mode of binding with a number of binding sites similar to that obtained at 23°C. The latter, however, displays the same behavior as M102K at 23°C, reflecting the two-mode binding. Similar to M102K, the appearance of the curvature in the isotherm is coincident with a ⌬G unf approaching 5 kcal/mol. For L99A/␣A complexes formed at 30°C, reduction of the temperature to 23°C results in a decrease of the bound fraction to that expected from Fig. 2, further confirming the reversible nature of the binding process (data not shown).
A further increase in temperature to 37°C lowers the energetic threshold for the activation of the high capacity binding mode. This is demonstrated by the binding isotherm of L99A (⌬G unf Х 5.3 kcal/mol) at this temperature, shown in Fig. 5a. Although the isotherm is linear, the intercept, reflecting the number of binding sites, is significantly different from that obtained at lower temperatures for L99A and indicates a 1 to 1 stoichiometry (Table IV). In the entire range examined, the binding seems to be primarily through the mode identified in the isotherm of L99A/F153A at 23°C, which has a lower apparent affinity but higher capacity. For mutants with similar stabilities at lower temperatures such as L99A/A130S at 30°C (Fig. 4b) and M102K at 23°C (Fig. 3a), only a minor contribution of the high capacity mode is evident in their isotherms. In fact, L99A at 37°C is more stable than L99A/F153A at 23°C, yet is predominantly bound by the high capacity mode.
Analysis of the slope of the isotherm reveals that the apparent affinity of ␣A-crystallin to L99A at 37 and 23°C are similar despite the significant drop in the mutant stability at 37°C (Tables III and IV). Because the apparent affinity is expected to increase as the stability is reduced, this result must reflect a lower intrinsic affinity of the predominant mode of binding at 37°C, the high capacity mode. This is consistent with the interpretation of the curved isotherms of M102K and L99A/ A130S at 23 and 30°C, respectively.
At 37°C, the binding of D70N is in the range that can be detected by EPR spectroscopy. Fig. 5b reports the binding curve of 70N at pH 7.2. The linear fit yields a number of binding sites in the range of that observed at 23°C for L99A and L99A/ A130S, suggesting exclusive binding at the low capacity mode. Furthermore, the apparent affinity to D70N at 37°C is comparable with the apparent affinity to L99A/A130S at 23°C (Table  I), suggesting the lack of significant temperature activation of the low capacity mode. Taken together, the results from L99A and D70N at 37°C demonstrate that both modes of binding persist at this temperature, with the high capacity mode reserved for lower stability mutants.
It is not possible at 37°C to obtain a binding isotherm for L99A/A130S. The samples tended to have small level of precipitates consisting of both the mutant and ␣A-crystallin except in the presence of ␣A-crystallin in a greater than 1 to 3 ratio. It is noted that L99A/A130S is marginally stable at this temperature.
pH Activation of Binding-The energetic threshold of each binding mode can be modulated by variation in the pH of the binding reaction. Fig. 6 (a and b) shows the pH 8.0 binding isotherms of L99A/A130S at 23°C and L99A at 30°C. Both isotherms are characterized by the curvature associated with the presence of two modes of binding. Similarly, the curvature is observed in the binding isotherm of D70N at 37°C at pH 8 and not at pH 7.2 (Figs. 6c and 5b). In contrast, a similar shift in the pH has an opposite effect on M102K, transforming the pH 7.2 isotherm into a linear one as shown in Fig. 6d. These effects could result from changes in the stability of T4L, the intrinsic binding affinity, and/or the number of binding sites.
These effects can be deconvoluted by measuring the pH dependence of the extent of binding of two mutants, D70N and M102K, where the pH dependence of ⌬G unf has been eliminated or reversed, respectively. Fig. 7a reports the change in the bound fraction of T4L as a function of pH at a 1 to 3 molar ratio of T4L to ␣A-crystallin. The increased binding at higher pH demonstrates a change in the intrinsic affinity between T4L and the chaperone and/or an increase in the number of binding sites available because ⌬G unf of D70N is pH-independent in the 6 -8 range (32). A similar conclusion is obtained for M102K at 23°C at a molar ration of 1:8. The more shallow increase in the bound fraction of M102K, shown in Fig. 7b, compared with L99A/A130S obtained under the same conditions is a result of the increase in stability of the former at higher pH as the pK a of the buried lysine residue is close to 6.5. Nevertheless, even for M102K, higher pH results in increased extent of binding. Taken together, the dependence of D70N and M102K binding on pH strongly indicate that the increased extent of binding is chaperone-specific.
Unfortunately, for most curved isotherms, the range and quality of the data do not permit quantitative analysis by nonlinear least squares fitting. Numerical calculations using two sets of binding sites suggest that the type of curvature in the isotherm of L99A/A130S (Fig. 6a) is observed for K ba1 in the 0.2-1 M Ϫ1 range and K ba2 in the 0.001 to 0.005 M Ϫ1 range, assuming n 1 ϭ 0.25 and n 2 ϭ 1. When compared with the value of K ba1 obtained at pH 7.2 (Tables III and IV), these results suggest an increase in the apparent affinity of both modes at higher pH. A similar conclusion can be obtained from the comparative analysis of the isotherms of L99A and D70N obtained at pH 7.2 and 8.0 at 30 and 37°C, respectively. In both cases, incipient curving in the isotherm is detected at high molar ratios of T4L to ␣A-crystallin. Furthermore, despite the linear nature of the isotherms for M102K and L99A at pH 8, 23°C, the number of binding sites suggests a contribution from the second mode of binding. Numerical calculations were used to verify that, at sufficiently weak affinity of the second mode, the curvature occurs at values of 1/L that are not accessible experimentally. It is manifested, however, by an erroneous linear extrapolation to a smaller intercept, reflect-FIG. 6. pH-induced changes in the binding characteristics of T4L mutants to ␣A-crystallin. All isotherms reflect the two-mode binding. a-c, biphasic isotherms of mutants having ⌬G unf approaching the threshold for activation of the high capacity mode. d and e, linear isotherms for mutants with ⌬G unf Х 6.6 kcal/mol. ing a larger number of binding sites. It is noted that the parameters obtained from the analysis of the M102K and L99A isotherms at pH 8 are similar, which is consistent with their similar ⌬G unf . The curvature reversal of M102K further demonstrates that activation of the high capacity mode is a function of the conformational stability of the substrate. ⌬G unf of M102K increases from 5.3 kcal/mol at pH 7.2 to 6.7 kcal/mol at pH 8. Thus, the contribution of the high capacity mode seems to be correlated with a ⌬G unf below 6 kcal/mol at 23°C and becomes dominant below 5 kcal/mol, as demonstrated by the binding isotherm of L99A/F153A. The threshold seems to be temperature-dependent as L99A stability at 37°C is close to that of M102K at 23°C, yet it is predominantly bound by a different mode.
Taken together, the temperature and pH effects suggest that two mechanisms modulate the use of the two modes of binding. The first is related to the conformational stability of the presented substrate, whereas the second arises from the intrinsic affinity of ␣A-crystallin regardless of the substrate state. DISCUSSION One of the mechanisms of protein aggregation involves the increase in the equilibrium population of non-native states, including the globally unfolded state, characterized by exposed hydrophobic surfaces. These on or off folding pathways intermediate states are populated as a consequence of the finite value of the equilibrium constants that characterize their interconversion with the native state. To be effective, molecular chaperones should detect the increased population of these states prior to the nucleation of aggregation. The recognition and binding either targets these proteins for turnover or allows their refolding.
Similar to other heat shock proteins, sHSP recognize and bind non-native protein states. Although the binding of ATP to sHSP has been reported (36,37), there is no evidence for an energy consumption step in the chaperone mechanism of sHSP. These functional characteristics have been established in extensive studies based on the suppression of aggregation of substrate proteins and enzymatic recovery studies (23,38). For a variety of reasons, these assays were not conducive to defining the mechanistic details of the chaperone function.
The simplest model of ␣-crystallin chaperone function has to include: 1) the identification of the conformational states recognized by the chaperone, 2) the description of the structural changes in the chaperone that are required or accompany recognition and binding, and 3) the measurement of the binding affinity and stoichiometry. Previous studies provided conflicting evidence regarding the nature of the binding process and the identity of the recognized states (39 -44). Changes in the ␣A-crystallin oligomeric structure associated with substrate binding have been deduced from structural analysis, functional studies, and temperature activation of chaperone function (45)(46)(47)(48)(49)(50).
For this purpose a set of T4L mutants forming an energetic ladder of lower stability were selected, expressed, and characterized. When presented with these mutants, ␣A-crystallin was able to sort them according to their stability, binding the mutants with reduced ⌬G unf to a larger extent. Given that the T4L mutants have structures similar to the native state, ␣A-crystallin cannot be recognizing static local unfolding or packing defects. Furthermore, because the aggregation propensity of the variants is negligible on the time scale of binding, aggregation is not a prerequisite for binding.
Conformational States of T4L Recognized by ␣A-Crystallin-The folding equilibrium of all the T4L mutants is two-state. Thus, the simplest interpretation for the differential binding of the T4L mutants is that ␣A-crystallin recognizes the unfolded state, the equilibrium population of which increases as ⌬G unf is reduced. In the range of stability of the mutants reported in this paper, the equilibrium population of the unfolded state is too small to be detected by spectroscopic approaches. Therefore, ␣A-crystallin shifts the folding equilibrium of T4L by thermodynamic coupling, i.e. stable complexes are detected when the free energy of binding to ␣Acrystallin is comparable with the free energy of refolding. A simplified coupled equilibrium that does not include possible structural changes in the chaperone associated with substrate binding is shown by Reactions 1 and 2.

REACTIONS 1 AND 2
These reactions predict that the apparent binding constant of Reaction 2, obtained from linear fits such as those of Fig. 2, is a ratio of the intrinsic binding constant, K bi , to the folding constant of Reaction 1, K f . It follows that the ratio of the apparent binding constants of two mutants should reflect their folding equilibrium constant.
For linear binding isotherms obtained at the same temperature, where a single binding mode predominates, it is possible to test whether this relation holds. An example is the differential apparent affinity at 23°C, pH 7.6, to L99A and L99A/ A130S. For the more than 1 kcal/mol difference in ⌬G unf , at least a 5-fold increase in the apparent binding constant is predicted. The observed change is less than a factor of 2. An alternative interpretation of the results is that ␣A-crystallin is recognizing partially unfolded, excited states that are transiently populated even under native conditions (51,52). Despite the apparent two-state nature of protein unfolding curves, it has been demonstrated that proteins in solution are constantly sampling higher energy conformational states that differ significantly in structure and packing from the native state. Whether these states are on the folding pathway has been the subject of debate (53). Nevertheless, the partial loss of structure in these states implies that they are distributed on the energy scale relative to the unfolded state, making them more accessible. Consequently, the free energy change associated with the transition from the native state to one of these states is smaller than ⌬G unf .
For T4L, native state H/D exchange reveals a continuum of conformations characterized by an average stability in the Cterminal domain that is significantly higher than the N-terminal domain (54). Thus, fluctuations from the native states are likely to occur to states with unfolded N-terminal domain. Furthermore, mutations can alter the energy landscape of these excited states. For instance, recent NMR study of the L99A mutant dynamics reveals interconversion with an excited state that is 2 kcal/mol higher in energy (55). The excited state is characterized by partial unfolding in the vicinity of L99A in the C-terminal domain.
If ␣A-crystallin recognizes one or more of these higher energy states, then the reduced overall stability is expected to enhance the binding to the extent detectable by spectroscopic methods, assuming that the mutations destabilize the native state rather than stabilize the unfolded state. The free energy required for the detection of a stable complex is correspondingly smaller than if the binding occurred to the unfolded state. This model is described by the following coupled equilibrium.
The transition from N to I i is described by an equilibrium constant K ei . This constant replaces K f in Equation 2. From an energetic perspective, the changes in ⌬G unf as a consequence of the decreased stability of N may or may not affect the ⌬G of N to I transition to the same extent, depending on whether the disrupted interactions are present in the excited state. Consequently, the change in K ba is not expected to proportionally track the changes in K f , as observed experimentally. Although our data do not provide definite evidence of ␣A-crystallin trapping of these kinetic intermediates, this interpretation is also consistent with the two modes of binding as described below.
It is possible that I 1 and I i interconvert in the bound state. Furthermore, if these states are on the unfolding pathway, they are likely to be in equilibrium with the bound unfolded state. Whether such interconversion occurs to a significant extent depends on the balance of the equilibrium constants that characterize these transitions relative to those of the corresponding binding equilibria to ␣A-crystallin.
This paper does not report on the kinetic of complex formation. Consequently, the details of the transition between the recognition and binding events including any conformational rearrangements in either substrate or chaperone have yet to be addressed.
Two Modes of Binding-That the binding of the excited states occurs through two modes was demonstrated by the biphasic binding isotherms under different combinations of temperatures and pH. Each mode is characterized by a different affinity and stoichiometry. One of the more significant results of this paper is the selective use of the low affinity mode to bind highly destabilized mutants. This is illustrated in the shift in the number of available sites for L99A/A130S versus those available for L99A/F153A at pH 7.2, 23°C, or the shift in the binding mode observed at 37°C between D70N and L99A. In the context of Reactions 3 and 4, this conformational preference might accompany binding of highly unfolded states such as I 1 that appear more frequently as the global ⌬G unf is reduced. If indeed ␣A-crystallin modes of binding display conformational preferences, it is likely that the energetic thresholds for activation are substrate-specific.
Given the high stoichiometry in the high capacity mode, it is logical to consider the possibility that T4L might be aggregating in the bound state. Such aggregate will be nucleated by the initial population of globally or partially unfolded T4L bound in the low capacity mode. There are two lines of evidence that exclude this possibility. First, T4L aggregation is unlikely to result in a finite and uniform apparent stoichiometry. Second, such T4L-rich aggregates are likely to be detected as separate peaks on SEC. Unpublished studies using bimane-labeled T4L mutants did not detect such complexes and reveal a rather uniform distribution of bound fluorescent T4L across the ␣Acrystallin peak.
Our data do not exclude the possibility that the two modes correspond physically to the same binding sites, thus the use of the term "mode of binding." If this is the case, the change in the available number of binding sites may reflect increased access depending on the presented conformational state of the substrate.
Activation of ␣A-Crystallin Chaperone Function-The binding of ␣A-crystallin to its substrate is enhanced by an increase in temperature. Previous studies probing the temperature enhancement of the chaperone efficiency, i.e. the suppression of substrate aggregation, reported similar conclusions (42)(43)(44)(45)(46). However, this is the first direct demonstration that this occurs through the effective increase in the number of binding sites via the activation of a second mode of binding. The quantitative comparison of affinities and stoichiometries at different temperatures is complicated by the presence of the two modes, particularly as the energetic threshold associated with binding to the second mode is approached. This is at the origin of the seemingly similar affinities obtained for substrates with different stabilities, a striking example being L99A and D70N at 37°C. For the former the main mode of binding occurs at the high capacity sites, whereas for the latter, the binding occurs at the low capacity sites. As a result, despite the reduced stability of L99A relative to D70N, the apparent affinity is similar. The data illustrate the intrinsic problem associated with the interpretation of the aggregation-based assays where it is not possible to distinguish affinity and stoichiometry effects.
Similarly, the pH activation involves an increase in the affinity of both modes. This was demonstrated by the increase in the fraction of bound D70N and M102K at higher pH. The pH-induced contribution of the second mode is manifested either by a curved isotherm or by an increase in the apparent number of binding sites in a linear isotherm. The former is more prevalent when the ⌬G unf approaches 5 kcal/mol at 23°C and 6 kcal/mol at 37°C. The lower apparent affinity at pH 8 for L99A and M102K compared with pH 7.2 reflects a partial contribution from this mode that is reflected in the increase in the effective number of binding sites.
Because the activation process of the high capacity mode is dependent on substrate conformation, a feasible model for activation is through a structural rearrangement in ␣A-crystallin to accommodate the more unfolded excited states. Such a rearrangement can occur via subunit exchange that involves the transient dissociation of a group of subunits (56,57). In the context of this model, an appropriate modification of model 2 would be to add the following equilibrium, where (␣A) a is an activated form of ␣A-crystallin.
␣A º n͑␣A͒ a REACTION 5 Toward a Mechanistic Model of sHSP Chaperone Function-From the results presented in this paper emerges the outline of a mechanistic model of ␣A-crystallin with implications for the entire sHSP superfamily. The chaperone function of ␣A-crystallin involves the recognition of non-native protein states distinct from either the native or the globally unfolded states. These states are populated through fluctuations from the native state as a consequence of the distribution of stabilities across the polypeptide chain (52). The rates of these fluctuations increase as N is destabilized, be it via directed mutagenesis or as a consequence of changes in the physicochemical conditions. The resulting decrease in the ⌬G of interconversion between N and I i makes the association with the chaperone more favorable than refolding to N resulting in a stable complex. Consequently, in a set of mutant of similar native state structures, ␣A-crystallin is able to "sense" the reduction in the stability of the native state.
The transient non-native states are necessarily conformationally heterogeneous as they span a wide range of energies. ␣A-Crystallin seems to classify them in two categories that are correspondingly bound using two different modes. One interpretation of the data presented in this paper is that ␣A-crystallin uses the low capacity mode to bind compact native-like states, whereas the high capacity mode is used to bind states that resemble the globally unfolded states. This model predicts that the threshold for binding to each mode will be dependent on the substrate identity because native state fluctuations are dependent on the stability distribution in the context of a structure.
The activation of the two modes of binding is significant from two perspectives. First, in many lower organisms, sHSPs are involved in thermotolerance; therefore, a temperature-dependent switch in the structure is not unexpected. Second, mammalian sHSPs, particularly the ␣-crystallins, have been shown to have a dynamic oligomeric structure (56,57). The structure is sensitive to pH, ionic strength, and temperature (58). Subunit exchange between oligomers involving a transient dissociation of the oligomeric structure is activated by increased temperature. Therefore, coupling of oligomer dynamics and function has been invoked to explain the increased chaperone efficiency at higher temperatures. The dynamic threshold model of chaperone function proposes that substrate recognition requires the transient dissociation of the oligomeric structure (49,50). In the context of this hypothesis, structural changes in the oligomer induced by pH and temperature allow better steric access for the substrate, thereby increasing both the affinity and the available number of binding sites. Because pH and temperature are maintained within narrow ranges, it is likely that the activation of mammalian sHSP occur in response to a different signal.
The dynamic threshold model has important implications with respect to the molecular design of the sHSP and the structural origin of coupling between oligomer dissociation and substrate recognition. In the non-activated state, the oligomerization of the subunits satisfies their binding potential. Subunit exchange reflects the marginal stability of the native oligomer versus the dissociated state. The appearance of nonnative states of the substrates provides a competing interaction surface that shifts the equilibrium toward the activated smaller oligomers.
The ability to tune up or down the buffering capacity without significant changes in affinity allows sHSP to function in a wide range of conditions while avoiding the role of an "unfoldase" under physiological conditions. In particular, the use of a high capacity/low affinity mode allows sHSP to deal with increased concentration of non-native states that nucleate aggregation. On the other hand, when the refolding energy is more favorable than that associated with chaperone binding, nonnative states constantly associate with the chaperone, promptly dissociate, and then refold. The thermodynamics of this equilibrium is not affected by the number of binding sites.