Mass Spectrometry : B-crystallin by NMR and α Chaperone-Microglobulin and the Molecular 2 β Monitoring the Interaction between Molecular Biophysics :

Stoppini, Agata Rekas, Vittorio Bellotti, Patrizia Mangione, Sofia Giorgetti, Monica Federico Fogolari, Maurizio Polano, P. Corazza, Alverdi, Fabio Pettirossi, Alessandra Gennaro Esposito, Megan Garvey, Vera OLIGOMERS 2-MICROGLOBULIN β DISSOCIATES B-CRYSTALLIN α Mass Spectrometry: B-crystallin by NMR and α Chaperone -Microglobulin and the Molecular 2 β Monitoring the Interaction between Molecular Biophysics:


aggregation at the various stages of its aggregation pathway. Our findings are highly relevant to the emerging view that molecular chaperone action is intimately involved in the prevention of in vivo amyloid fibril formation.
It is well recognized that there is a close association between a wide range of diseases and protein aggregation. For example, Alzheimer, Parkinson, Creutzfeldt-Jakob, and Huntington diseases are characterized by the accumulation of protein deposits or plaques in which the aggregated protein is arranged in highly ordered amyloid fibrils (1). Understanding the relationship between the formation of amyloid fibrils and protein aggregation diseases is the focus of much research activity. It is clear that amyloid fibril formation arises from the repertoire of partially folded intermediate states of proteins, and there is significant evidence that the process of aggregation of these states causes cellular toxicity which, in turn, contributes to the particular disease (1). Intriguingly, various reports have suggested that the cellular toxicity of the putative causative agent in Alzheimer disease, the amyloid ␤ (A␤) 5 peptide, and other amyloidogenic proteins, does not arise from the insoluble amyloid deposits but rather from the soluble prefibrillar oligomers (2)(3)(4)(5)(6).
Molecular chaperone proteins interact with the intermediate states of proteins. Most likely, they are important in preventing the accumulation of prefibrillar species in vivo, and accordingly, they have been proposed as therapeutics in the treatment of protein conformational diseases (7,8). In particular, small heatshock molecular chaperone proteins (sHsps) are attractive in this aspect because of their ability to interact with and stabilize long-lived intermediate states of proteins early on their offfolding pathway, prior to large scale aggregation (8,9). ␤ 2 -Microglobulin (␤2m) is the nonpolymorphic light chain component of class I major histocompatibility complex (MHC-I) (10). In vivo, ␤2m causes amyloidosis via accumulation in the tissues of patients with renal failure as a result of long term hemodialysis (11). The structure of ␤2m and the molecular aspects of its amyloid transition have been extensively investigated in recent years, not only because of the clinical relevance of dialysis-related amyloidosis but also as a model for other amyloidogenic proteins (12)(13)(14)(15)(16)(17)(18)(19).
Among the amyloidogenic proteins, ␤2m has the unusual property of not requiring mutation or proteolysis to form fibrils. In vitro, however, the transition does not occur spontaneously at neutral pH but requires moderately acidic conditions and proper ionic strength or the presence of chaotropes (20 -22). The formation of ␤2m fibrils at neutral pH occurs upon the addition of seeds and 2,2,2-trifluoroethanol, surfactants, or glycosaminoglycans (23,24). Furthermore, temperature and pH conditions similar to those occurring upon inflammation in periarticular compartments enhance aggregation (25) and trigger fibril formation in the presence of collagen fibers (26) and heparin (27). The availability of several variants of the natural protein with different stabilities and amyloidogenic properties and the recent finding of a naturally occurring mutation (D76N) leading to a highly amyloidogenic species responsible for a systemic pathology (28) make ␤2m a convenient system to investigate the unfolding and aggregation processes. For instance, the fragment of ␤2m devoid of the N-terminal hexapeptide (⌬N6␤2m) that occurs to a significant extent (ϳ30%) in ex vivo deposits of the protein (13) has an enhanced tendency to aggregate and form fibrils, even at neutral pH (14). Furthermore, the Arg-3-to-Ala mutant of ␤2m (R3A␤2m) forms a folded structure very close to that of the wild-type protein, but it exhibits a spontaneous propensity to unfold, aggregate, and precipitate over a period of days to weeks (19).
sHsps are ubiquitous intracellular chaperones. ␣-Crystallin is an sHsp that is composed of two closely related subunits, A and B, each of about 20 kDa in mass. Like most other sHsps, the ␣-crystallin subunits exist as large heterogeneous aggregates of a mass range ϳ300 -1000 kDa (29,30). ␣-Crystallin has been studied for many years because of its presence as the major eye lens protein and its crucial role, via its chaperone ability (31), in preventing crystallin protein aggregation and consequent cataract formation. Interest in ␣-crystallin has increased significantly since the discovery that ␣B-crystallin also occurs extensively outside the lens (32). Its expression levels are enhanced significantly under stress conditions, e.g. elevated temperature. The ␣B-crystallin oligomer has an average size of ϳ32 subunits with the dimer as the basic building block (31,33). The protein is composed of three domains as follows: a central ␤-sheetcontaining "␣-crystallin" domain that is flanked by relatively disordered N-and C-terminal domains, the latter of which also contains a short, highly mobile, and unstructured C-terminal extension (34 -36). Crystal and NMR structures have recently become available for the ␣-crystallin domain (37)(38)(39).
This study exploited the complementary insight obtainable from NMR spectroscopy and mass spectrometry (MS) into the early stages of ␤2m amyloid fibril formation and the nature of the interaction between ␣B-crystallin and ␤2m. For this purpose, we used wild-type ␤2m and R3A␤2m under conditions of a convenient time scale for spectroscopic monitoring of processes that precede and lead to ␤2m aggregation, along with the effects of ␣B-crystallin on these processes. By use of 1 H NMR spectroscopy, we determined that a nonspecific interaction occurs between the exposed surface of R3A␤2m and ␣B-crystallin during chaperone action. Furthermore, MS and NMR diffusion coefficient measurements enabled the ␤2m oligomerization distribution to be quantified in the absence and presence of ␣B-crystallin. Importantly, ␣B-crystallin induced dis-aggregation of ␤2m oligomers. It is concluded that ␣B-crystallin is multifaceted in its ability to prevent fibril formation by interacting with ␤2m at the various stages of its aggregation pathway.
Thioflavin T Fluorescence Assay-Thioflavin T (ThT) fluorescence assays (440:490 nm excitation/emission) to monitor amyloid fibril formation of R3A␤2m (500 g/ml) were undertaken for up to 17 days in 100 mM sodium phosphate buffer at pH 7.4 and 310 K with orbital shaking at 700 rpm in the presence of 10 M ThT (51).
NMR Spectroscopy-NMR spectra were obtained at 500.13 MHz with a Bruker Avance spectrometer on 0.1-0.9 mM protein samples dissolved in H 2 O/D 2 O (90:10 and 95:5 v/v) with 50 -70 mM phosphate buffer, 50 -100 mM NaCl, and pH* in the range 6.5-6.7 (where pH* is the pH meter reading without isotope effect correction). When necessary, samples were centrifuged and/or filtered using filters with a threshold of 0.22 m (Millipore, Bedford, MA) or 0.02 m (ANOTOP type, Whatman, Maidstone, UK). Studies were carried out at 310 K apart from a few experiments acquired at 313 K, which did not show any NMR-detectable differences. The samples containing two proteins were prepared using ␣B-crystallin/R3A␤2m at molar subunit ratios of 1:60, 1:15, 1:5, 1:2, and 1:1, and ␣-crystallin/ ␤2m molar subunit ratios of 1:4.9 ϫ 10 2 , 1:95, 1:60, and 1:35. The ␤2m mutant and the stock ␣B-crystallin solutions employed for measuring the signal attenuations were filtered using 0.02-m microfilters with either lyophilized or un-lyophilized purified ␤2m proteins being used. The 1 H two-dimensional total correlation spectroscopy (TOCSY) (52) NMR spectra were collected and processed as reported previously (19,53). Based on the signal-to-noise ratio, amplitude errors typically ranged within Ϯ5%, although much larger errors are to be expected for very weak resonances. Values of average accessible surface area of backbone amide and C ␣ hydrogen atoms were computed with the software available at the public domain biocomputation suite of EMBL, using a probe radius of 0.14 nm and the 20 deposited NMR structure coordinates of R3A␤2m (19). Molecular structures were generated with InsightII (Accelrys Inc., San Diego), MOLMOL (54), and Swiss-Pdb Viewer software packages.
The diffusion coefficients of a 0.44 mM solution of wild-type ␤2m at 310 K (93:7 v/v H 2 O/D 2 O, 100 mM NaCl, 70 mM phosphate buffer, pH 6.6), as a function of the ␣-crystallin content (at the ratios reported above), were measured using the convection-compensated two-dimensional double-stimulated echo diffusion ordered spectroscopy (DSTE DOSY) (55) pulse sequence to record matrices of 2048 (t2) by 80 points (t1). The gradient strength was varied linearly from 2 to 95% of its maximum value (61.1 G/cm). The lengths of the diffusion interval (200 ms) and diffusion gradient (2.5 ms) were optimized to obtain a substantial decay of the signal (95% loss, at least). The data analysis along t1 was performed using the Bruker DOSY software by imposing a single exponential fitting. Water suppression was achieved with the addition of an excitation sculpting WATERGATE module (56,57).
Mass Spectrometry-MS measurements were performed on lyophilized wild-type ␤2m solutions dissolved in 100 mM ammonium acetate buffer, pH 6.82, at a final concentration of 120 M, in the absence or in the presence of ␣B-crystallin. All sets of measurements were carried out using ␤2m solutions divided into two equal portions that were or were not filtered through 0.02-m filters.
For the first set of experiments, ␤2m was dissolved in a final volume of 500 l, at the concentrations reported above. The sample was divided into 2 aliquots. The 1st aliquot was the ␤2m control solution, and the second one was mixed with ␣B-crystallin in 100 mM ammonium acetate buffer, pH 6.82, at a final ratio of 30:1 ␤2m/␣B-crystallin on a molar subunit basis.
In the second set of experiments, concentrated solutions of ␣B-crystallin in 100 mM ammonium acetate were added to 50 l of an aged (20 days old) 120 M ␤2m solution at a ratio of 30:1 ␤2m/␣B-crystallin on a molar subunit basis. Before electrospray ionization (ESI), the samples were diluted to a final concentration of 20 M in 100 mM ammonium acetate buffer, pH 6.82. This concentration was chosen because the physiological concentration of circulating ␤2m is 0.3 M in healthy individuals but is elevated 20 -60-fold in long term hemodialysis patients (58,59).
The native MS experiments were performed using an ESI time-of-flight (TOF) LCT mass spectrometer equipped with a Z-spray nanoflow electrospray source (Micromass UK Ltd., Manchester, UK). Aliquots of 2 l of the 20 M protein solutions were infused into the mass spectrometer using an in-house pulled and gold-coated borosilicate glass needle.

Amyloid Fibril Formation of R3A␤2m Is Prevented by
␣B-crystallin-At the global level, amyloid fibril formation of R3A␤2m was monitored via ThT binding. ThT fluorescence is markedly enhanced upon binding to the cross-␤-sheet structure associated with the amyloid fibril structure. The ThT binding curves for an agitated R3A␤2m solution, in the absence and presence of ␣B-crystallin, are shown in Fig. 1A. Interestingly, the ThT curves reproducibly displayed a biphasic profile, with an initial increase in fluorescence occurring after a lag phase of some 100 h and a plateau phase extending up to about 250 h, followed by a second much larger increase in fluorescence. Transmission electron microscopy images of R3A␤2m, taken 400 h after the commencement of incubation, confirmed the extensive presence of fibrillar species (data not shown). The presence of increasing amounts of ␣B-crystallin led to the partial, and at higher concentrations, complete suppression of ThT fluorescence, due to the inhibition of fibril formation. When only partial suppression of ThT fluorescence increase was observed, i.e. at ␣B-crystallin/R3A␤2m molar subunit ratios of 1:1000, 1:100, and 1:10, little if any inhibition of the first ThT fluorescence increment occurred, whereas the effect on the second ThT fluorescence increment was significant, implying that ␣B-crystallin preferentially interacts with the second phase of R3A␤2m aggregation. Assessment of the Stability of R3A␤2m Solutions in the Presence and Absence of ␣B-crystallin as Monitored by 1 H NMR Spectroscopy-The 1 H NMR spectrum of R3A␤2m has been assigned (19) thereby enabling investigation of alterations in chemical shifts and intensities of its resonances upon interaction with ␣B-crystallin. Because of its relatively rapid rate of aggregation and precipitation, the mutant exhibits a more convenient aggregation time scale for monitoring by NMR than the wild-type protein.
Depending on the procedures undertaken after expression, purification, and sample preparation, the time scale for R3A␤2m aggregation was variable. We found that the timing of the onset of precipitation and its extent depended on several factors. Besides the expected effects of protein concentration, pH, ionic strength, and temperature, the onset of precipitation was also affected by filtering the freshly prepared solutions and, specifically, by the exclusion size threshold of the filter (0.22 or 0.02 m). Different precipitation patterns were also observed when the freshly prepared solutions were centrifuged rather than filtered. Other sources of variability were whether the protein had been subjected to lyophilization after final purification by dialysis, as well as omitting the lyophilization step so as to utilize only frozen aliquots of purified protein solutions. Solutions of lyophilized protein that were used unfiltered or subjected to filtration through a 0.22-m filter prior to use are referred to as ordinary samples. Solutions of protein not subjected to lyophilization that had been filtered through the 0.02-m filter are referred to as nonordinary samples.
As already reported (19), the 1 H NMR spectrum of an ordinary solution of R3A␤2m, acquired immediately after dissolution, shows the same pattern as observed with wild-type protein (12), which is consistent with the strong conformational similarities between the two species (19). Data were collected using clear solutions prepared without filtering and, sometimes, with mild centrifugation. After a lag phase that could last one or several days depending on the concentration and previous history of the sample, the NMR spectrum underwent changes caused by the concomitant processes of unfolding and aggregation followed by precipitation, as reported previously (19). The progress of unfolding is highlighted by gradual loss of the isolated hallmark resonances of the folded species, coupled to a simultaneous growth of resonances at random coil chemical shift values (19) (Fig. 2). Precipitation was readily ascertained by visual inspection of the sample and by a decrease in the total NMR spectral integral. Fig. 2A shows the one-dimensional 1 H NMR spectrum of a 1:1 mixture (on a molar subunit basis) of an ordinary sample of R3A␤2m and ␣B-crystallin. ␣B-crystallin forms a large oligomer of average mass of ϳ650 kDa (31). The ␣B-crystallin oligomers tumble very slowly, and therefore the majority of their resonances is not observed in the NMR spectrum. The exception is the flexible and unstructured C-terminal extension of 12 amino acids (34 -36) that gives rise to additional, readily detectable cross-peaks at random coil chemical shift values in the fingerprint region of two-dimensional NMR spectra (e.g. TOCSY) ( Fig. 2A, insets). The same pattern was also observed in NMR spectra at half of this ratio, i.e. 0.5:1.0 ␣B-crystallin/ R3A␤2m (data not shown). The spectra shown in Fig. 2A are interpreted as evidence for a weak interaction between R3A␤2m and ␣B-crystallin that introduces a limited broadening of the R3A␤2m resonances due to transient mutual association.
Quantification of the effects of ␣B-crystallin on R3A␤2m spectra was undertaken by integration of selected resonances and cross-peaks in the one-and two-dimensional TOCSY spectra. Fig. 2B documents the time-dependent intensity change of an upfield-shifted methyl peak of R3A␤2m (L23 ␦ 1 -CH 3 ). Similar results are shown also for amide cross-peaks from the twodimensional TOCSY fingerprint (Fig. 2, C and D). In the absence of ␣B-crystallin, after a lag phase of ϳ50 -60 h, the resonance intensity decreased in a linear time-dependent manner and achieved about 80% loss over the time course of the experiment (ϳ195 h). In the presence of ␣B-crystallin, for the same resonance, the loss of intensity was slowed down 3.4-fold, on average (Fig. 2B). No detectable lag phase also occurred in the presence of ␣B-crystallin. Overall, in the presence of ␣B-crystallin, a 6-fold reduction in the increase in intensity of the aliphatic resonances that are characteristic of unstructured peptides was measured, compared with the corresponding measurements in the absence of aB-crystallin under the same conditions (data not shown). Moreover, the samples with ␣B-crystallin present exhibited hardly detectable precipitation over the same time period compared with samples without ␣B-crystallin, where significant precipitation occurred.
In the absence of ␣B-crystallin, the timing for the lag phase and subsequent loss of NMR resonance intensity (Fig. 2, B and D) matches the lag phase and first rise of ThT fluorescence (Fig.  1). Following a latency period of 60 Ϯ 20 h, the NMR signal loss is due to concomitant precipitation and formation of large soluble aggregates that are poorly observable or totally unobservable by NMR due to their very large linewidth. The total loss of the NMR signal occurs around 200 -250 h, i.e. at the time as the first plateau in the ThT fluorescence data (Fig. 1). So, the initial increase of ThT fluorescence could coincide with the loss of the NMR intensity due to formation of large prefibrillar aggregates.
We hypothesize that these prefibrillar aggregates, possibly containing some extent of intermolecular cross-␤-structure, occur along nonamyloid and/or slow amyloid-forming pathways that are present at high ␤2m concentration and are enhanced by solution agitation (60).
Mapping R3A␤2m-␣B-crystallin Interaction by NMR Spectroscopy-By contrast to the situation in the previous section with ordinary samples of R3A␤2m, neither unfolding nor precipitation of R3A␤2m was detected over an extended time FIGURE 2. A, 500 MHz 1 H NMR spectra of an ordinary sample of R3A␤2m (ϳ0.9 mM, 310 K, pH 6.6) in the absence (lower trace) and presence (upper trace) of an equimolar amount of ␣B-crystallin subunit. The only apparent consequence of the sHsp addition is a general broadening, with conservation of the spectral pattern of R3A␤2m. The insets show a portion of the H N -H ␣ connectivity (fingerprint) region of the corresponding two-dimensional TOCSY spectra. The boxed cross-peaks arise from the unstructured and flexible C-terminal extension of ␣B-crystallin. Only this region of 12 amino acids is present in the spectrum by virtue of its fast local tumbling as opposed to the extremely slow overall motion of ␣B-crystallin oligomer (ϳ650 kDa) that causes the extreme broadening of other resonances (34 -36, 60). Assignments are indicated for R3A␤2m NH to ␣-CH cross-peaks (12,19). Arrows mark visibly attenuated R3A␤2m connectivities for Q69 and T86. B, time evolution of the resolved upfield-shifted methyl resonance (L23 ␦ 1 CH 3 ), which is diagnostic for the correctly folded species, of an ordinary sample of R3A␤2m obtained from 500 MHz 1 H one-dimensional NMR spectra, in the absence (circles) and in the presence of ␣B-crystallin at equimolar (squares) or halved concentration (triangles) with respect to R3A␤2m. The average rate of loss of signal intensity was (5.6 Ϯ 0.5) ϫ 10 Ϫ3 h Ϫ1 in the absence of ␣B-crystallin and (1.5 Ϯ 0.5) ϫ 10 Ϫ3 or (1.8 Ϯ 0.5) ϫ 10 Ϫ3 h Ϫ1 for 1:2 or 1:1 ␣B-crystallin/R3A␤2m mixtures, respectively. The errors on the rates were estimated by considering the deviations obtained when assessments were made using absolute resonance amplitudes rather than self-normalized values (53) to account for the uncertainties introduced by the scaling procedure. Generally speaking, such uncertainties arise when resonances with smaller linewidths overlap preexisting signals, as well as from the general linewidth increase upon addition of ␣B-crystallin. C, two-dimensional TOCSY fingerprint details obtained from an 82-h-old solution of R3A␤2m (0.8 mM, 310 K, pH 6.6). The boxed cross-peaks represent the NH-␣CH connectivities whose time evolution is displayed in D, respectively, from Phe-30, in the folded protein, and from an unknown residue, in the unfolded soluble species. D, time dependence of the two-dimensional TOCSY cross-peak amplitudes of C (Phe-30 ϭ squares, unassigned ϭ diamonds). In the absence of ␣B-crystallin, the evolution of the resolved fingerprint cross-peaks showed typically a similar trend as that obtained by considering the evolution of the resolved methyl resonances in one-dimensional 1 H spectra (B). Indeed, the slope of the decreasing Phe-30 cross-peak amplitude is Ϫ5.9 ϫ 10 Ϫ3 h Ϫ1 , i.e. in absolute value the same, within the experimental error, as the slope of the unknown cross-peak intensity.
period in nonordinary solutions prepared from un-lyophilized protein that had been filtered through 0.02-m filters. Thus, analysis of NMR spectra of these samples in the presence of ␣B-crystallin was not affected (over typical measurement time intervals) by systematic uncertainties due to the loss of soluble protein or the onset of resonances with unknown linewidth due to R3A␤2m aggregation. Therefore, measurements of the attenuation of resonances induced by ␣B-crystallin were undertaken using ␣B-crystallin/R3A␤2m at molar ratios suitable for NMR quantification. Attenuation of two-dimensional TOCSY fingerprint (H N -H ␣ ) cross-peaks was measured at 1:15 and 1:5 molar subunit ratios (data not shown). Qualitatively, identical patterns were observed at both ratios, and therefore, only the more precise results at the higher ratio (1:5) are reported (Fig. 3A).
To avoid a complex range of local mobilities, the quantitative attenuation analysis was restricted to connectivities from backbone hydrogens, i.e. a class of nuclei with relatively homogeneous motional properties (61). The fingerprint resolution of the TOCSY spectrum was not always sufficient to estimate individual cross-peak amplitudes, so only 67 out of 94 H N -H ␣ correlations could be integrated, which introduced discontinuity in the data of Fig. 3A. The interaction of R3A␤2m with ␣B-crystallin is very dynamic due to a weak, nonspecific association between the two proteins. No chemical shift changes occurred, so the observed broadening of cross-peaks is ascribed . The attenuations exhibited a qualitatively identical pattern at a 1:15 molar ratio. However, greater attenuation (whose maximum was around 50%), which was more convenient for quantification, was observed at the higher ␣B-crystallin ratio. The location of the ␤-strand segments according to the NMR solution structure (19) is given at the bottom of the figure. The attenuation trend along the R3A␤2m sequence exhibited a clear analogy with the results of the experiments performed at an equimolar concentration on an ordinary sample ( Fig. 2A). B, schematic representation of ␤2m secondary structure showing A data, i.e. the attenuation levels of R3A␤2m H N -H ␣ TOCSY connectivities due to ␣B-crystallin interaction observed in nonordinary samples. Class I MHC heavy chain (backbone schematic and space filling) is shown in transparency. The crystal structure coordinates of the complex were used (Protein Data Bank code 3HLA (10)). A three-color code was employed to report R3A␤2m experimental attenuation, as indicated explicitly. The residue positions for which no attenuation data could be measured is colored gray. The small schematic aside shows the ␤-strand naming scheme of ␤2m. The drawing was prepared with PyMOL (DeLano Scientific LLC).
to the high effective mass that transiently is experienced by R3A␤2m upon interaction with the chaperone.
At first glance, the profile of the fingerprint cross-peak attenuations appears to be spread over the whole R3A␤2m sequence, except for the segment of residues 50 -70 where the data set is sparse (Fig. 3B). The attenuations did not show any specific dependence on amino acid type, or any correlation with the secondary and tertiary structure, or a high surface exposure of the NH-C ␣ H moieties. In detail, the largest attenuation values of H N -H ␣ correlations were mainly located in the inter-strand loop regions, namely the A-B and E-F loops and the CЈ strand (five out of the nine most attenuated H N -H ␣ connectivities with an attenuation range of 40 -50%). Therefore, the upper apical region of R3A␤2m (Fig. 3B) may be preferentially involved in the interaction with ␣B-crystallin. Other backbone nuclei that experience comparable signal amplitude reductions are spread over the molecule. Fig. 3B provides a color-coded structural overview of the involvement of R3A␤2m regions in interaction with ␣B-crystallin, as inferred from the NMR data, and shows that the contact patches are disseminated over the whole surface of R3A␤2m. Among them, the regions of ␤2m in contact with the heavy chain of MHC-I (in transparent format) are readily apparent (10). Therefore, the weak, nonspecific interaction between R3A␤2m and ␣B-crystallin also involves ␤2m regions with a propensity toward protection from the aqueous environment to limit their hydrophobic exposure.
Assessing the Effect of ␣-Crystallin on Wild-type ␤2m Aggregation by NMR Diffusion Measurements-We utilized NMR DOSY experiments (55) to measure the translational diffusion coefficient (D T ) of wild-type ␤2m as a function of ␣-crystallin concentration (Fig. 4). At 310 K, a 0.44 mM solution of wild-type ␤2m reproducibly exhibited a D T value of 1.83 ϫ 10 Ϫ10 m 2 /s that did not change appreciably over 5 days. Assuming a spherical model and using a mass-dependent protein density estimate (62), this value is consistent with an apparent mass of 19.9 kDa. The molecular mass of ␤2m is 11.9 kDa, implying that the protein undergoes aggregation processes of variable stoichiometry, i.e. as per the association equilibria previously proposed (19) with a dimer being the most populated species. In this respect, the conclusion generated by NMR is different from that obtained by MS (see below), which we ascribe to the 4-fold ␤2m concentration difference between the two experiments. Consistent with this, NMR diffusion measurements with freshly prepared ␤2m solutions at a much lower concentration (ϳ0.1 mM) generated a D T value consistent with a mass of 12 kDa in agreement with the MS data on similar samples that showed a predominance of monomeric species.
The diffusion coefficient of a 0.44 mM solution of ␤2m was also determined by NMR following sequential additions of ␣-crystallin. The D T value did not alter for an ␣-crystallin/␤2m ratio of 1:4.9 ϫ 10 2 , but at ratios of 1:95, 1:60, and 1:35, a D T value of 1.89 ϫ 10 Ϫ10 m 2 /s was obtained (Fig. 4), with a slight increase observed over 7 months after the last addition of ␣-crystallin to the solution. For the D T value measured after this last addition, the corresponding apparent mass of ␤2m was determined to be 17.9 kDa, a value that suggests a small but significant shift of the oligomerization equilibria toward lower mass species in the presence of ␣-crystallin, a shift that occurred to a greater extent over the long term (Fig. 4). A control TOCSY spectrum confirmed the stability of the ␤2m sample whereby ␣-crystallin preserved the native conformation of ␤2m as was also observed by NMR spectroscopy for R3A␤2m in the presence of ␣B-crystallin (Fig. 3A).
Time Course Oligomerization of Wild-type ␤2m as Monitored by Mass Spectrometry-The alteration over 20 days in the mass spectrum of wild-type ␤2m after dissolving the protein in solution at pH 6.8 is reported in Fig. 5. The spectra show timedependent changes in peak identities and intensities, in particular as new (oligomeric) species are formed in solution. Immediately after dissolution, mainly monomeric and some dimeric ␤2m species were detected, and after 4 h, oligomers up to hexamers were readily recognizable (data not shown). Later, species with higher mass appeared with increasing intensities, but their unambiguous identification could only be determined up to the octameric species. In Fig. 5, the development with time of increasing peak intensities, corresponding to an increase in concentration of dimer, trimer, tetramer, and the onset of larger aggregated species, is consistent with the formation of oligomers in solution that may be the forerunners of nucleation and fibrillogenesis (19).
To characterize quantitatively the time course of ␤2m oligomerization, peak areas in the mass spectra were evaluated. Because of the absence of an internal time-independent reference, raw data of areas corresponding to dimers or the sum of higher oligomers, i.e. from trimers to octamers, were normalized in each spectrum to the area corresponding to the overall sum of the ␤2m peaks. This normalization procedure can be considered an internal scaling of the experimental data sets, and thus the intensity data obtained thereof are referred to as autoscaled areas. It should be noted that, despite mass conservation, comparisons among autoscaled areas may be affected by systematic errors due to the failure to detect very large oligomeric species (i.e. larger than octamers) in the mass spectra, especially for samples that were aged for a long period.
The time course of signal intensity for mass spectra of the ␤2m dimer species and the higher oligomer pool, from trimers up to octamers, normalized over the total area sum, was quantified (Fig. 6). The percentage of dimer species present in solu-tion increased steadily with time reaching 32% of the total 8 days after preparation. A similar trend was seen also for the sum of the other oligomers larger than dimers. In all cases, oligomer autoscaled areas were more intense in the unfiltered than the filtered samples with a 0.02-m threshold (20-nm-filtered samples). The difference between 20-nm-filtered and unfiltered samples was greater for the oligomer pool compared with the dimeric species (Fig. 6).
Effect of ␣B-crystallin on Wild-type ␤2m Oligomerization as Monitored by Mass Spectrometry-To examine the effect of ␣B-crystallin on ␤2m aggregation, mixtures of wild-type ␤2m and ␣B-crystallin at molar subunit ratios of 30:1, 60:1, and 90:1 ␤2m/␣B-crystallin were prepared and examined by MS. Two sets of experiments were performed. The first set involved the preparation of a fresh mixture of ␤2m and ␣B-crystallin in solution, followed by its monitoring over time by MS in tandem with a control solution that contained ␤2m only. In the second set of experiments, ␣B-crystallin was added to an aged (20-dayold) solution of ␤2m, at the same molar ratios listed above. Fig. 7 shows the mass spectrum with time of the solution containing simultaneously dissolved ␤2m and ␣B-crystallin (at a subunit molar ratio of 60:1) as monitored over 20 days (A-E) compared with the corresponding control spectrum of ␤2m after 20 days in the absence of ␣B-crystallin (F). As is readily evident from inspection of Fig. 7, the presence of ␣B-crystallin potently inhibits the development of oligomeric ␤2m species. After 1 day (Fig. 7B), the mass spectrum of the mixture was essentially dominated by peaks from the monomer and dimer species and was highly comparable with the corresponding "time 0" spectrum of ␤2m on its own. In the absence of ␣B-crystallin, the corresponding spectrum of a ␤2m solution contained peaks arising from oligomeric species up to hexamers (data not shown). Furthermore, contrary to the spectrum of a 20-day-old solution of ␤2m, the spectrum of a 20-day-old mixture of ␤2m and ␣B-crystallin did not exhibit any indication of conversion to large oligomeric ␤2m species.
Similar results were obtained using ␤2m/␣B-crystallin molar subunit ratios of 30:1 and 90:1, with either unfiltered and filtered solutions. At a 90:1 ␤2m/␣B-crystallin ratio, the sHsp was similarly effective at inhibiting the formation of ␤2m oligomers. Thus, MS demonstrates the impressive capacity of ␣B-crystallin, even at very low stoichiometric ratios, to completely inhibit the aggregation of wild-type ␤2m. Fig. 8, A and B, shows the mass spectra of an aged (20-dayold) ␤2m solution at 1 and 20 days, respectively, after addition of ␣B-crystallin (at a 60:1 ␤2m/␣B-crystallin ratio). The peaks corresponding to the oligomeric species that occur in aged ␤2m samples in the absence of ␣B-crystallin (Fig. 8C) are significantly decreased in its presence so that the mass spectrum of the aged ␤2m solution after 20 days of equilibration with ␣B-crystallin displays no oligomeric species and even a net decrease of the residual dimer peaks that were detected after a single day of equilibration with ␣B-crystallin. Importantly, the data in Fig. 8 provide compelling evidence for the capability of ␣B-crystallin to break down ␤2m oligomers in solution. In doing so, ␣B-crystallin converts ␤2m oligomers to monomers.
Similar MS experiments were conducted with two unrelated control proteins, GroEL, the well characterized molecular chaperone involved intimately with the folding of target proteins (for MS see ref. 63), and lysozyme, as representative of a typical medium-sized protein that has no chaperone ability. Independently of the molar subunit ratio (90 or 60 or 30:1 ␤2m/ control) and preliminary solution filtering, none of the effects observed for ␣B-crystallin, i.e. prevention of ␤2m oligomerization and dissociation of ␤2m oligomers, was observed with GroEL ( Fig. 9) or lysozyme (data not shown).

DISCUSSION
In this study, we have performed complementary NMR and MS experiments, addressing the very first stages of the ␤2m aggregation pathway and its inhibition due to interaction with the molecular chaperone ␣B-crystallin. We have previously shown that R3A␤2m accumulates into fibrillar deposits in dialysis-related amyloidosis and exhibits a convenient aggregation time scale, while preserving all the features of the parent protein (19). Here, we have shown that R3A␤2m forms fibrils at neutral pH upon agitation, a process that is inhibited by substoichiometric amounts of ␣B-crystallin.
We monitored the effect of ␣B-crystallin on R3A␤2m in detail by NMR spectroscopy. First, we used ordinary solutions prepared from lyophilized R3A␤2m, either unfiltered, centrifuged, or filtered through 0.22-m filters. In the absence of ␣-crystallin, these samples exhibited a wide range of R3A␤2m aggregation, precipitation, and unfolding behavior. Second, we used nonordinary R3A␤2m solutions, i.e. solutions that were prepared under carefully controlled conditions, with un-lyophilized protein and filtered through 0.02-m filters.
In ordinary samples of R3A␤2m, aggregation was preceded by a lag phase that is consistent with the initial steps of the nucleated conformational conversion mechanism that R3A␤2m undergoes during its aggregation (19). In the presence of ␣B-crystallin, the unfolding and aggregation of R3A␤2m . A similar trend is seen also for the sum of the other oligomers larger than dimers (A). In all cases, oligomer autoscaled areas are more intense in the unfiltered than in the 20-nm-filtered samples. The difference between 20-nm-filtered and unfiltered samples is greater for the oligomer pool compared with the dimeric species. Because of the lack of an internal reference species with time-independent area, the raw area data corresponding to oligomers were normalized in each spectrum using the overall ␤2m peak area sum. This normalization procedure can be considered an internal scaling of the experimental data sets, and thus the intensity values obtained thereof will be referred to as autoscaled intensities. It should be mentioned that, despite the mass conservation, comparisons among autoscaled intensities may be affected by systematic errors due to failure to observe in the mass spectrum large oligomeric species, especially for long aged samples.
In ordinary samples of R3A␤2m, the lag phase prior to precipitation was 1 to several days, depending on the sample's concentration and previous history. Removing solute heterogeneity required filtered, un-lyophilized samples. These nonordinary samples enabled reproducible data to be obtained on the very early stages of aggregation.
NMR studies did not detect any protein aggregation or unfolding in the nonordinary samples of R3A␤2m, even over extended time intervals. There was no specific interaction site of R3A␤2m with ␣B-crystallin because a general dipolar broadening, associated with dipole-dipole interactions between closely spaced 1 H nuclei, was observed that affected the relaxation and increased the linewidths of R3A␤2m resonances. It is therefore concluded that a large region of R3A␤2m interacts with ␣B-crystallin. By evaluation of the backbone R3A␤2m resonance attenuations, broadening was quantified, and the contacts between the two proteins were mapped (Fig. 3). ␤2m is a protein that in vivo, apart from the fibrillar deposits in pathological conditions, normally occurs in complex with the polymorphic moiety of MHC-I. The finding that ␣B-crystallin is in contact with the regions that are involved in MHC-I assembly suggests the involvement of hydrophobic interactions in driving the association between R3A␤2m and ␣B-crystallin. In our NMR examination of the interaction between ␣-synuclein and ␣B-crystallin, -casein and ␣-crystallin, and ataxin-3 and ␣Bcrystallin, we also observed nonspecific association between the two proteins and a large region of interaction on the target protein (43,47,50). Thus, it would seem that ␣B-crystallin has a general, essentially entropy-driven mechanism of interaction with fibril-forming target proteins that involves transient, low affinity interactions with a large solvent-exposed region of the latter proteins. Another important feature of this interaction is that as a consequence of the weak transient contacts, the turnover efficiency leads to an extensive sampling of the target protein surface by the sHsp. This prerogative of sHsps is crucial for preventing the aggregation of the target proteins and appears to be a specific feature of sHsps when compared with the interaction of other molecular chaperones with target proteins. Thus, the failure of GroEL to prevent ␤2m oligomerization suggests that it is not as versatile as sHsps in interacting with target proteins under stress conditions, i.e. it only interacts with target proteins along their folding pathway (9). The same inability of lysozyme ensures that the efficient, albeit nonspecific, chaperone action of sHsps cannot be attributed to a generic protein interaction.
The low affinity of ␣B-crystallin for ␤2m is an advantage in terms of preventing ␤2m aggregation. Most likely, the very potent chaperone ability of ␣B-crystallin, i.e. its ability to inhibit ␤2m aggregation at very low ratios of ␣B-crystallin to ␤2m, arises from the combination of the fast rate of transient binding and dissociation of ␣B-crystallin to ␤2m coupled with the slow process of unfolding of ␤2m toward an amyloidogenic conformation. Furthermore, at any one time, there are very few ␤2m molecules undergoing this unfolding process. As a result of these factors, a single ␣B-crystallin molecule can potentially contact multiple ␤2m molecules to stabilize the native ␤2m conformation. Contributing to ␣B-crystallin's high chaperone activity is its highly dynamic nature that arises from its inherent flexibility due to a large amount of structural disorder and its property of undergoing continuous subunit exchange. The ability of ␣-crystallin subunits to prevent amyloid fibril formation by wild-type ␤2m has been investigated at pH 2.5 under conditions in which ␤2m aggregates spontaneously (22). At this pH, ␣-crystallin subunits partially unfold and lose or have a significantly reduced oligomeric state (64,65). No stable complex was formed by either ␣-crystallin subunits with monomeric ␤2m (22), a conclusion in agreement with the results reported herein. Likewise, no complex is formed between ␣-crystallin and the fibril-forming protein, apolipoprotein C-II, although ␣-crystallin is a potent inhibitor of the latter's fibril formation (42). At acidic pH, the required concentration of chaperone for suppression of aggregation was nearly equal to that of ␤2m on a molar subunit basis (22). Under the conditions employed in this study (i.e. at neutral pH), the suppression of ␤2m fibril formation by ␣B-crystallin is much more efficient. The efficiency in our case can be explained in terms of the extremely low concentration of the dynamic ␤2m oligomers that are present at neutral pH, although the dissociation at acidic pH of the ␣B-crystallin oligomer and the concomitant unfolding of the monomer, which decrease the chaperone ability of ␣B-crystallin, 6 would have a significant effect on its ability to suppress ␤2m fibril formation.
An important observation from the MS studies presented herein is that ␣B-crystallin is capable of dissociating soluble wild-type ␤2m oligomers into monomers in aged ␤2m solutions incubated in the absence of ␣B-crystallin. Thus, ␣B-crystallin can dissociate soluble ␤2m oligomers into smaller species and thereby remove them from the fibril-forming, off-folding 6 C. Brockwell, H. Ecroyd, and J. A. Carver, unpublished results.  . Mass spectra of a ␤2m and GroEL mixture (molar ratio of 60:1 ␤2m/GroEL), 20 min (A), 1 day (B), and 20 days (C) after mixing the proteins together are shown. The mass spectrum of the 20-day-old ␤2m control solution is also reported (D). The data show clearly that GroEL has no effect on the ␤2m oligomerization process, i.e. the spectra at 20 days from solution preparation in the presence (C) and in the absence (D) of GroEL are highly comparable. Similar results were observed also at molar ratios 30:1 and 90:1 ␤2m/GroEL (data not shown).
FIGURE 10. Schematic mechanism for the inhibition of amyloid fibril formation by ␣B-crystallin. During the formation of amyloid fibrils, target proteins (e.g. ␤2m), either folded or unfolded in their native state, can adopt a partially folded intermediate that is prone to aggregation to form prefibrillar oligomers. These can then sequester other intermediates to form the fibrillar species. ␣B-crystallin can interact with any of these species. Early along the aggregation pathway, i.e. at the intermediate or early oligomerization stage, ␣B-crystallin interacts transiently with the target protein and pushes it back into its monomeric, native conformation. It can also bind to the amyloid fibril to stabilize it and prevent its fragmentation and elongation. pathway (9,66,67). As these oligomeric species are proposed to be cytotoxic, the ability of ␣B-crystallin to remove them is potentially of great importance in mitigating against their toxicity in diseases such as Alzheimer and Parkinson. Our results showing that ␣B-crystallin prevents the cellular toxicity of fibril-forming species, e.g. the A␤ peptide (41), is consistent with the ability of ␣B-crystallin to dissociate ␤2m oligomers.
Furthermore, our recent work has shown that ␣B-crystallin binds along the length and to the ends of fully formed amyloid fibrils of ␣-synuclein (50,68), A␤ (69), and apolipoprotein C-II (70) to prevent fibril elongation and fragmentation. Fragmentation is recognized to be a major factor in fibril propagation due to it leading to secondary nucleation events and hence enhanced fibril formation (71).
The ability of ␣B-crystallin to interact with species at the various stages of the amyloid fibril-forming pathway is schematically represented in Fig. 10. The interaction at the early stages (with monomers and oligomers) is transient while binding of ␣B-crystallin to amyloid fibrils occurs (68 -70). Thus, ␣B-crystallin is multifaceted in its ability to prevent fibril formation. Clearly, this multiplicity and plasticity in functionality provide ␣B-crystallin with enhanced capabilities in vivo in preventing the generation of toxic species arising from amyloid fibril formation and is consistent with the important role that ␣B-crystallin (and other sHsps) plays in minimizing protein aggregation.