Small heat-shock protein structures reveal a continuum from symmetric to variable assemblies1

doi:10.1006/jmbi.2000.3657 available online at http://www.idealibrary.com on

J. Mol. Biol. (2000) 298, 261±272

Small Heat-Shock Protein Structures Reveal a Continuum from Symmetric to Variable Assemblies Dana A. Haley1,2, Michael P. Bova2, Qing-Ling Huang2 Hassane S. Mchaourab3 and Phoebe L. Stewart1* 1

Department of Molecular and Medical Pharmacology and Crump Institute for Biological Imaging, UCLA School of Medicine, Los Angeles CA 90095, USA 2

Department of Ophthalmology and Jules Stein Eye Institute UCLA School of Medicine Los Angeles, CA 90095, USA 3

National Biomedical ESR Center, Biophysics Research Institute, Medical College of Wisconsin, Milwaukee, WI 53226, USA

The small heat-shock proteins (sHSPs) form a diverse family of proteins that are produced in all organisms. They function as chaperone-like proteins in that they bind unfolded polypeptides and prevent uncontrolled protein aggregation. Here, we present parallel cryo-electron microscopy studies of ®ve different sHSP assemblies: Methanococcus jannaschii HSP16.5, human aB-crystallin, human HSP27, bovine native a-crystallin, and the complex of aB-crystallin and unfolded a-lactalbumin. Gel-®ltration chromatography indicated that HSP16.5 is the most monodisperse, while HSP27 and the a-crystallin assemblies are more polydisperse. Particle images revealed a similar trend showing mostly regular and symmetric assemblies for HSP16.5 particles and the most irregular assemblies with a wide range of diameters for HSP27. A symmetry test on the particle images indicated stronger octahedral symmetry for HSP16.5 than for HSP27 or the a-crystallin assemblies. A single particle reconstruction of HSP16.5, based on 5772 particle images with imposed octahedral symmetry, resulted in a structure that closely matched the crystal structure. In addition, the cryo-EM reconstruction revealed internal density presumably corresponding to the ¯exible 32 N-terminal residues that were not observed in the crystal structure. The N termini were found to partially ®ll the central cavity making it unlikely that HSP16.5 sequesters denatured proteins in the cavity. A reconstruction calculated without imposed symmetry con®rmed the presence of at least loose octahedral symmetry for HSP16.5 in contrast to the other sHSPs examined, which displayed no clear overall symmetry. Asymmetric reconstructions for the a-crystallin assemblies, with an additional mass selection step during image processing, resulted in lower resolution structures. We interpret the a-crystallin reconstructions to be average representations of variable assemblies and suggest that the resolutions achieved indicate the degree of variability. Quaternary structural information derived from cryo-electron microscopy is related to recent EPR studies of the a-crystallin domain fold and dimer interface of aA-crystallin. # 2000 Academic Press

*Corresponding author

Keywords: alpha-crystallin; cryo-electron microscopy; small heat-shock protein; three-dimensional reconstruction; variable structure

Present address: H. S. Mchaourab, Department of Molecular Physiology and Biophysics, Vanderbilt University Medical School, Nashville, TN 37232, USA. Abbreviations used: cryo-EM, cryo-electron microscopy; CTF, contrast transfer function; DNA-PKcs, DNA-dependent protein kinase catalytic subunit; FRC, Fourier-ring correlation; FSC, Fourier-shell correlation; HSP, heat-shock protein; aB-crys./a-lact., aB-crystallin complexed with unfolded a-lactalbumin. E-mail address of the corresponding author: [email protected] 0022-2836/00/020261±12 $35.00/0

Introduction All organisms produce one or more small heatshock proteins (sHSPs) that function to bind unfolded polypeptides and prevent uncontrolled protein aggregation in the cell. Mammalian sHSPs are expressed constitutively and are upregulated after physiological stresses, such as heat, pH extremes, and osmotic variation (Head et al., 1994; Derham & Harding, 1999). The subunits of sHSPs # 2000 Academic Press

262 have molecular masses in the range of 12 to 43 kDa, and they assemble into large oligomers with a total molecular mass of 140 to over 800 kDa. A characteristic of the sHSPs is that they contain a homologous C-terminal region of 80 to 100 residues referred to as the ``a-crystallin domain'' (de Jong et al., 1998). a-Crystallin is one of the main structural proteins in the vertebrate lens, and it is composed of two polypeptides, aA and aB, that have 57 % sequence similarity. Ignolia & Craig (1982) ®rst discovered the high homology between Drosophila sHSPs and mammalian a-crystallin. Later it was con®rmed that aB-crystallin is a member of the sHSP family (Klemenz et al., 1991) and that a-crystallin possesses in vitro chaperone-like activity (Horwitz, 1992). When native a-crystallin is isolated from lens tissue, it contains a 3:1 molar ratio of aA and aB-crystallin and is found to have numerous post-translational modi®cations (Groenen et al., 1994). aB-Crystallin is expressed in numerous tissues throughout the body, while aA-crystallin is less abundant (Srinivasan et al., 1992; Bhat & Nagineni, 1989; Kato et al., 1991). Elevated levels of aB-crystallin are associated with disorders such as multiple sclerosis (van Noort, 1996), Alzheimer's disease (Renkawek et al., 1994), Creutzfeldt-Jacob disease (Renkawek et al., 1992) and Alexander's disease (Head et al., 1993). A crystal structure has been determined for one member of the sHSP family, HSP16.5 from Methanococcus jannaschii (Kim et al., 1998a). The structure reveals 24 subunits arranged with octahedral symmetry forming a hollow spherical assembly with small ``windows'' or openings. The monomer folds into a b-sandwich with one strand, referred to as b6, contributed by a neighboring subunit at the dimer interfaces. A similar, but not identical, folding pattern has now been shown for the a-crystallin domain in aA-crystallin by EPR and sitedirected spin labeling (Koteiche & Mchaourab, 1999). The major difference between the two folds is that in aA-crystallin there is no b-strand corresponding to b6 in HSP16.5, creating a different dimerization motif. Sequence alignment of the acrystallin domains in sHSPs indicates that aB-crystallin, human HSP27, and mouse HSP25 are also missing the domain-swapped b-strand (Koteiche & Mchaourab, 1999). The homology between the members of the sHSP family is signi®cantly reduced in the N-terminal domain and the C-terminal tail that bracket the conserved a-crystallin domain. In HSP16.5, there are 45 residues before the a-crystallin domain, 32 of which are disordered in the crystal structure. In aA- and aB-crystallin, roughly 60 residues are found before the a-crystallin domain, while in HSP25 and HSP27 the N-terminal domain is over 80 residues. A short ordered C-terminal tail of 12 residues is observed in the HSP16.5 crystal structure. Depending on the exact end point of the a-crystallin domain in aA-, aB-crystallin, and HSP27, there may be up to 20 ordered residues fol-

Symmetric to Variable sHSP Assemblies

lowed by a mobile C-terminal extension (Carver, 1999). Subtle differences in the a-crystallin domain fold as well as presumably major variations in the N-terminal domain and C-terminal tail may be responsible for the dramatic differences observed in the quaternary structure, or oligomeric arrangement, of the sHSPs. A cryo-electron microscopy (cryo-EM) study of aB-crystallin indicates that the quaternary structure is variable (Haley et al., 1998). To compensate for the polydispersity in mass of the sample, a rough selection was performed to select for particle images of assemblies with the average molecular mass, 650 kDa or 32 subunits. The low-resolution reconstruction of the assembly reveals an asymmetric, roughly spherical protein shell surrounding a large central cavity. The ®nding of a variable or open loose quaternary structure is consistent with the observation of dynamic subunit exchange, light scattering and equilibrium sedimentation (Bova et al., 1997; van den Oetelaar et al., 1990; Vanhoudt et al., 1998). Various oligomeric states have been proposed for other sHSPs including a nine subunit trimer of trimers for Mycobacterium tuberculosis HSP16.3 (Chang et al., 1996); a 12 subunit HSP18.1 from Pea (Lee et al., 1997); either a 16 subunit (Ehrnsperger et al., 1999) or a 32 subunit assembly for murine HSP25 (Behlke et al., 1991); and an 24 subunit assembly for Chinese hamster HSP27 (Lambert et al., 1999). Here, we present cryo-EM studies of ®ve different sHSP assemblies, M. jannaschii HSP16.5, human aB-crystallin, human HSP27, bovine a-crystallin, and the complex of human aB-crystallin and unfolded a-lactalbumin. All of these sHSPs are recombinantly produced except for native a-crystallin, which is isolated from bovine lens. Cryo-EM offers the advantage of being able to visualize biological assemblies in a native-like, frozen-hydrated state without the need for crystals. Our results suggest that even the most symmetrical sHSP examined, HSP16.5, may have some degree of structural variability in solution. Since the function of sHSPs is to bind unfolded polypeptides, the plasticity of its quaternary structure may be functionally important for recognizing and binding a diverse population of conformationally ¯exible target proteins.

Results Polydispersity and symmetry of sHSPs The family of sHSPs is generally accepted to be polydisperse in total molecular mass with the exception of M. jannaschii HSP16.5 (Kim et al., 1998b) and possibly M. tuberculosis HSP16.3 (Chang et al., 1996). Comparison of the half bandwidth of the elution pro®les from gel-®ltration chromatography gives an estimate of the range in apparent molecular mass (Figure 1). Gel-®ltration chromatography of HSP16.5 at ambient temperature indicates a monodisperse sample with a half

Symmetric to Variable sHSP Assemblies

Figure 1. Gel-®ltration elution pro®les of a control monodisperse protein assembly, GroEL; M. jannaschii HSP16.5; human aB-crystallin; human HSP27; bovine native a-crystallin; and aB-crystallin with bound unfolded a-lactalbumin. The peak bandwidth, measured at half the height, is shown for each pro®le and indicates the degree of heterogeneity in molecular mass. A shift to the left corresponds to a higher average molecular mass. Peaks labeled a-lact. and Vo correspond to unbound a-lactalbumin and void volume, respectively. Pro®les for GroEL, aB-crystallin, native a-crystallin and aB-crystallin with bound unfolded a-lactalbumin are reprinted with permission from Horwitz et al. (1999).

263 bandwidth similar to that observed for GroEL, the 14-subunit chaperonin from Escherichia coli (Xu et al., 1997). It is interesting, that the HSP16.5 bandwidth broadens when the sample is heated to 85  C to mimic the normal environment of the hyperthermophile M. jannaschii and then returned to ambient temperature before chromatography (data not shown). The effectiveness of HSP16.5 at binding unfolded protein varies considerably for different target proteins. We have found that HSP16.5 is much less effective at binding various target proteins at ambient temperature than aBcrystallin (unpublished data). The elution pro®les of aB-crystallin, HSP27, native a-crystallin, and aB-crystallin complexed with unfolded a-lactalbumin (aB-crys./a-lact.) suggest that these protein assemblies are polydisperse in total molecular mass. For aB-crystallin, the bandwidth is 1.3 ml, indicating an approximate molecular mass range of 650(140)kDa. The bandwidth observed for HSP27 is slightly larger, followed by that of native a-crystallin, and then aBcrys./a-lact. Other studies with different unfolded target proteins complexed with a-crystallin show the formation of high molecular mass aggregates of several million Daltons (Derham & Harding, 1999). We chose a-lactalbumin as the representative target protein since it produced the narrowest gel-®ltration bandwidth as well as the most regularly shaped particles as evaluated by cryo-EM. We have shown that aB-crystallin assemblies Ê in cryo-electron range in diameter from 80 to 180 A micrographs (Haley et al., 1998). In the current study, we collected cryo-EM images of HSP16.5, HSP27, native a-crystallin, and aB-crys./a-lact. It is not surprising that the diameters of native a-crystallin and aB-crys./a-lact. were larger on average Ê ) than that found for aB-crystallin, (120 to 200 A consistent with their higher average molecular masses. The HSP27 particles showed the greatest Ê ) with no single variability in diameter (90 to 220 A diameter being most prevalent. Cryo-electron micrographs of HSP16.5 revealed particles with Ê , in relatively uniform diameters of 110 to 130 A Ê good agreement with the 120 A diameter derived from the X-ray crystal structure (Kim et al., 1998a). In preparation for 3D image processing, particle images of each sHSP were computationally extracted from digitally collected micrographs. The main structural features became clear after ®ltering Ê , which is the position the particle images to 23 A of the ®rst zero in the contrast transfer function (CTF) for the set defocus value (Figure 2). Many of the HSP16.5 particle images appear highly symmetric with clear 2- or 3-fold symmetry, while some looked more expanded or irregular. In contrast, none of the a-crystallin or HSP27 particle images showed clear symmetry. HSP27 assemblies were so variable in diameter that 3D reconstruction was not attempted for this sample. To quantify the degree of octahedral symmetry in the sHSP images, we used the symmetry selfsearch algorithm in the IMAGIC software package

264

Symmetric to Variable sHSP Assemblies

Figure 3. A line histogram plot of the octahedral symmetry residual for projections of the HSP16.5 X-ray crystallographic structure, and cryo-EM particle images of HSP16.5, aB-crystallin and a control asymmetric sample, DNA-PKcs. The projection images of the HSP16.5 crystal structure have added Gaussian noise designed to mimic the signal-to-noise ratio in the cryo-EM images. Note that the curves for aB-crystallin and DNA-PKcs are shifted to the right relative to the curves for HSP16.5, indicating reduced or no overall octahedral symmetry, respectively.

Figure 2. Representative cryo-EM particle images of HSP16.5, aB-crystallin, HSP27, native a-crystallin, and aB-crystallin with bound a-lactalbumin. The particle images are all extracted from micrographs collected with a microscope defocus value of ÿ1.5 mm and are ®lÊ resolution (corresponding to the ®rst zero tered to 23 A of the CTF). The protein appears white in these images. Note that for HSP16.5, some particles reveal clear 2-fold and 3-fold symmetry (top row), while others are more irregular or display expanded diameters (bottom row). Ê. The scale bar represents 100 A

(van Heel et al., 1996). Projection images of the octahedral HSP16.5 crystal structure with added Gaussian noise all gave low residuals in the range of 0.2 to 0.25 indicating near perfect octahedral symmetry (Figure 3). The cryo-EM particle images of HSP16.5 gave a range of residuals from 0.2 to 0.5 with a mean value of 0.3, while aB-crystallin displayed a range of residuals from 0.2 to 0.8 with a mean value of 0.4. As a control, we have also plotted the octahedral symmetry residual for cryoEM images of an asymmetric particle, DNA-dependent protein kinase catalytic subunit (DNA-PKcs),

previously reconstructed in our laboratory (Chiu et al., 1998; Stewart et al., 1999). This monomeric protein is relatively close in molecular mass (470 kDa) to that of the HSP16.5 oligomeric assembly (396 kDa). The residual curve for DNA-PKcs is shifted even further to the right of the HSP16.5 and aB-crystallin curves and represents a sample without any octahedral symmetry. The other sHSPs tested, HSP27, native a-crystallin, and aB-crys./alact., showed mean residual values between that found for aB-crystallin and DNA-PKcs. From these results, we conclude that while HSP16.5 in solution is more octahedrally symmetric than either HSP27 or the a-crystallin assemblies, HSP16.5 may be more variable in solution than suggested by the crystal structure. Reconstruction of HSP16.5 without imposed symmetry Another way to test the symmetry of an assembly is to reconstruct the particle without imposed symmetry and to observe what symmetry emerges. For this purpose, we processed a subset of 1000 particle images of HSP16.5 collected with a microscope defocus value of ÿ1.5 mm. Following seven cycles of re®nement and deconvolution to correct for the CTF of the microscope, the ®nal reconstruc-

Symmetric to Variable sHSP Assemblies

265

tion showed a roughly spherical protein shell with numerous holes suggestive of the packing arrangement in the crystal structure (Figure 4(a)). We were able to identify an axis along which imposing 3fold symmetry produced even clearer octahedral symmetry for the assembly. Figure 4(b) shows the 3-fold symmetrized reconstruction viewed along unimposed, pseudo 2-, 3- and 4-fold axes, which are oriented appropriately for an octahedron. For comparison the same three views are shown of the unsymmetrized reconstruction in Figure 4(a). Note that the pseudo 2- and 3-fold axes show the greatest improvement. This result together with the symmetry self-search plot supports imposing octahedral symmetry on the HSP16.5 cryo-EM data. Reconstruction of HSP16.5 with imposed octahedral symmetry After ®nding that pseudo-octahedral symmetry emerged for HSP16.5, we processed a larger set of cryo-EM data, 5772 particle images collected with three different microscope defocus values (ÿ1.5, ÿ1.0 and ÿ0.8 mm), assuming that the particle has octahedral symmetry. The resulting reconstruction is shown in Figure 5 along with the X-ray crystal Ê resolution. Overall these structure ®ltered to 13 A two structures appear quite similar, with a central cavity and gaps or windows at the 3- and 4-fold axes. The buffer used for crystallization of HSP16.5 is different from the buffer used for cryo-EM, which was designed to mimic physiological conditions, and thus small structural differences might be expected. The resolution of the cryo-EM reconÊ by the Fourier struction was estimated to be 13 A

Figure 4. A single particle reconstruction of HSP16.5 based on 1000 particle images without imposed octahedral symmetry. (a) Three views of the asymmetric reconstruction showing gaps in the protein shell of the assembly. (b) The same three views after imposing 3fold symmetry along a different axis. The emergence of clear non-imposed 2-, 3- and 4-fold symmetry axes (designated by symbols) indicates at least loose octahedral symmetry for HSP16.5 in solution. The scale bar Ê. represents 100 A

Figure 5. Comparison of the X-ray crystal structure of HSP16.5 (yellow) beside the cryo-EM single particle reconstruction with imposed octahedral symmetry Ê resolution (pink). Both structures are ®ltered to 13 A and are viewed along a 4-fold axis along with two density slices at the indicated positions. In the slices, the colors represent the density values, with red corresponding to the strongest density and green to the weakest. The small white circles denote the locations of residue 33 in the crystal structure. Note that the regions of weak density observed in the central cavity of the cryo-EM reconstruction extend from the protein shell at the positions of residue 33. The weak internal density presumably corresponds to the disordered 32 residues missing from the N terminus in the crystal structure. The ®ltered crystal structure is shown contoured as appropriate for 115 ordered residues per subunit. The cryo-EM reconstruction is contoured to account for the full-length monomer (147 residues). The scale bar repÊ. resents 100 A

shell correlation (FSC) method, which involves splitting the data set into two halves and calculating two half reconstructions (BoÈttcher et al., 1997; Conway et al., 1997). When the full cryo-EM reconstruction was compared with the ®ltered crystal structure by the FSC method, the two were found Ê resolution. Soft-masking of to agree to only 30 A the cryo-EM reconstruction to remove the internal Ê , which is density improved the agreement to 28 A roughly the thickness of the HSP16.5 protein shell.

266

Symmetric to Variable sHSP Assemblies

A comparison of slices from the cryo-EM reconÊ struction with the crystal structure ®ltered to 13 A resolution showed that fewer structural details are found within the cryo-EM density (Figure 5). Therefore, the true resolution of the cryo-EM reconstruction is probably somewhere in between Ê. 13 and 28 A One signi®cant difference between the cryo-EM and crystal structures is the presence of internal weak density as shown in two slices through the structures (Figure 5). Kim et al. (1998a) suggested that the disordered 32 N-terminal residues extend into the central cavity of the HSP16.5 assembly, with the ®rst ordered residue on the inner protein surface near the small window at the 4-fold axis. The internal density observed in the cryo-EM reconstruction clearly connects to each subunit near the position of residue 33 and its volume corresponds to roughly 27 residues per subunit. Thus, it is likely that the internal density in the cryo-EM reconstruction represents the most common position for the ¯exible N-terminal region. Reconstructions of sHSPs without imposed symmetry In order to perform a fair comparison between the cryo-EM data on HSP16.5 and the a-crystallin assemblies, we collected and processed data on all of them in the same manner with one additional image-processing step for the a-crystallin assemblies. Since gel-®ltration chromatography suggested that the a-crystallin assemblies are polydisperse in molecular mass, we performed a rough mass selection step on these data sets. Mass-selection was performed on class-sum or average images, which have enhanced signal-to-noise ratios compared to the unclassi®ed particle images. Representative mass-selected class-sum images of aB-crystallin, a-crystallin, and aB-crys./a-lact. are shown in Figure 6 together with class-sum images of HSP16.5 generated for comparison. The majority of the HSP16.5 class-sum images displayed mirrorsymmetry, as is expected for all projections of an octahedral assembly. However, approximately 10 % of the HSP16.5 class-sum images were irregular, supporting the idea that in solution HSP16.5 has some degree of variability. In contrast, almost none of the aB-crystallin, a-crystallin, or aB-crys./ a-lact. class-sum images showed clear mirror symmetry. Also, there is a wider range of diameters in the mass-selected class-sum images of all three acrystallin samples than observed for HSP16.5. Class-sum images of a-crystallin and aB-crys./alact. often show more density in the center than aB-crystallin. The class-sum images indicate some limited variability for HSP16.5 and greater variability for the a-crystallin assemblies, however it is dif®cult to quantify the degree of variability among the samples. Asymmetric reconstructions of the a-crystallin assemblies, based on CTF corrected and massselected data sets, are shown together with the

Figure 6. Class-sum images of HSP16.5, aB-crystallin, native a-crystallin, and aB-crystallin with bound a-lactalbumin. The majority of the HSP16.5 class-sum images display clear mirror symmetry (top row), while a minority are more irregular in shape (bottom row). Massselected class-sum images of all three a-crystallin assemblies display greater variability in size and shape than the HSP16.5 class-sum images. The scale bar represents Ê. 100 A

asymmetric reconstruction of HSP16.5 in Figure 7. All of the reconstructions are irregular and roughly spherical. The cropped view of the aB-crystallin reconstruction reveals a central cavity, which is

Symmetric to Variable sHSP Assemblies

Figure 7. Single particle reconstructions of HSP16.5 (pink), aB-crystallin (blue), native a-crystallin (green), and aB-crystallin with bound a-lactalbumin (lavender). Each reconstruction is based on 1000 particle images and processed without imposed symmetry. The HSP16.5 reconstruction, previously shown in Figure 4(a), is displayed along a pseudo 4-fold symmetry axis. Both a full structure (left) and a cropped view (right) are shown for each reconstruction. In the crop planes, the colors represent the density values as in Figure 5. The resolutions indicated by the Fourier shell correlation method are noted on the far right, and each reconstruction is ®ltered to its respective resolution. The reconstructions are contoured to account for the average molecular mass of the assembly indicated by gel-®ltration chromatography. Ê. The scale bar represents 100 A

similar in size but less regular in shape, than that observed for the asymmetric HSP16.5 reconstruction. The reconstructions of native a-crystallin and aB-crys./a-lact. do not show clear central cavities;

267 however, there are regions of lower density in the center. Note that the aB-crystallin reconstruction presented here is somewhat different than the previously published reconstruction (Haley et al., 1998). The reconstruction in Figure 7 is further re®ned and based on CTF corrected particle images. Although the reconstructions of the a-crystallin assemblies showed no hint of symmetry, we nevertheless tried imposing 2-, 3-, and 4-fold symmetry along a variety of axes, as we had done for the asymmetric HSP16.5 reconstruction. In no case did additional, non-imposed symmetry axes emerge. This result, together with the ®nding of irregular particle and class-sum images supports the idea that none of the a-crystallin assemblies show a strong preference for a particular symmetric quaternary arrangement. It is conceivable that the acrystallin domain forms locally symmetric interfaces, but if another subunit interface is formed by a ¯exible region this could lead to disorder at the quaternary level. The FSC method revealed a range in resolutions Ê for HSP16.5 to for these reconstructions, from 29 A Ê for native a-crystallin (Figure 7). In the case 44 A of HSP16.5 and aB-crystallin, the resolution is approximately the same as the thickness of the protein shell observed in the reconstruction. This indicates that we are not resolving any details of the subunit packing, but only the overall protein distribution within the assembly. The lower resolutions found for native a-crystallin and aB-crys./a-lact. are consistent with our observation that particle images of these samples were the most irregular in shape. We propose that since we have collected and processed these data sets in a similar manner, the resolutions indicate the degree of variability in the assembly. Another test for structural variability is the comparison of particle images with reprojections of the reconstruction in a pair-wise manner. If the assembly has a well-de®ned quaternary structure and if the reconstruction has faithfully reproduced it, then there will be good agreement between the images in the pairs, re¯ected in a good numerical Fourier ring correlation (FRC). The resolutions indicated by the FRC test paralleled those found Ê for HSP16.5, 41 A Ê for aB-crystalby FSC with 31 A Ê for native a-crystallin, and 54 A Ê for aBlin, 56 A crys./a-lact. In the case of native a-crystallin, it is possible that the high degree of post-translational modi®cations contributes to the variability of the assembly. The sHSPs are known to bind unfolded proteins that are in a molten globule state (Carver & Linder, 1998; Das et al., 1996; Rao et al., 1998), and thus it is reasonable that greater variability is found for aB-crys./a-lact. than for the aB-crystallin assembly. We interpret the a-crystallin reconstructions not as de®nitive structures of these complexes, but rather as average representations of variable assemblies.

268

Symmetric to Variable sHSP Assemblies

There are several examples in the literature of cryo-EM single particle reconstructions that closely match X-ray crystal structures of the same particle. The crystal structure of the hepatitis B virus capsid (Wynne et al., 1999) is in excellent agreement with Ê resolution cryo-EM reconstruction the 7.4 A (BoÈttcher et al., 1997). The main difference is in the region of the six C-terminal residues, which are too disordered to observe by crystallography. In the case of the GroEL-GroES complex, a domain from the crystal structure of the GroEL-GroES-(ADP)7 complex (Xu et al., 1997) ®ts well within the cryoEM density envelope of the same complex (Rye et al., 1999). However, subtle differences are observed between the two structures at the interdomain contacts perhaps due to crystal packing Ê resolution reconstrucforces. Comparison of a 20 A Ê restion of the ribosomal 50 S subunit with a 20 A olution density map derived from X-ray data reveals that both maps have the same overall shape (Ban et al., 1998). There are some discrepancies observed between the two structures and these are attributed to the possibility that certain ribosomal domains may adopt different conformations. Overall, these structural comparisons indicate good agreement between the two techniques; however, real differences do arise. Cryo-EM offers the advantage of being able to readily study assemblies in different binding states without the need for crystals. A second advantage is that ¯exible residues not observed by crystallography are sometimes observed in lower resolution cryo-EM reconstructions, including the N-terminal residues of HSP16.5. Finally, cryo-EM may be used to image assemblies that have not been crystallized, such as most of the sHSPs.

without imposed symmetry showed pseudo-octahedral symmetry, we can conclude that in solution at ambient temperature HSP16.5 assemblies have at least loose, if not perfect, octahedral symmetry. We have found that HSP16.5 is far less effective at binding unfolded target proteins than aB-crystallin at ambient temperature. Perhaps the chaperonelike function of HSP16.5 is dependent upon some structural variability although the assembly can clearly adopt an octahedral quaternary arrangement under certain conditions. Our ®ndings indicate that HSP16.5 N termini project into the central cavity of the assembly. Thus it is unlikely that HSP16.5 sequesters unfolded target proteins inside the cavity since it is partially ®lled. We present cryo-EM data for several other members of the sHSP family including aB-crystallin, HSP27, native a-crystallin, and the complex of aB-crystallin with an unfolded target protein, a-lactalbumin. All of these samples displayed some polydispersity by gel-®ltration chromatography. The gel-®ltration bandwidth observed for HSP27 was only slightly larger than that of aB-crystallin; however, the cryo-EM images revealed a wide range in particle diameters. There is disagreement in the literature over the oligomeric structure of HSP27 and the homologous HSP25 (Ehrnsperger et al., 1999; Behlke et al., 1991; Lambert et al., 1999). Since gel-®ltration gives only the apparent molecular mass, it does not necessarily correlate with the true particle size distribution. Conceivably mass selection of the HSP27 cryo-EM particle images would be possible; however, one would need to start with an even larger sample size before selection than we used for the a-crystallin assemblies. The reconstructions of HSP16.5 and the a-crystallin assemblies without imposed symmetry indicate that HSP16.5 is the most symmetric and the resolutions suggest that there is a continuum in the amount of structural variability among these assemblies.

Quaternary symmetry and variability of the sHSPs

Comparison between EPR, cryo-EM, and X-ray crystallographic results

Here, we present two cryo-EM reconstructions of HSP16.5; one without imposed symmetry and the other with octahedral symmetry. It is dif®cult to discern which reconstruction more closely represents the structure of HSP16.5 in solution. The octahedral symmetry residual test indicated that projections of the crystal structure, even with added noise, were more octahedrally symmetric than the cryo-EM images of HSP16.5, but this could be due to the low signal-to-noise ratio of cryo-EM. Most of the cryo-EM particle images of HSP16.5 displayed clear symmetry, while some appeared more irregularly shaped. This could be due in part to the low signal-to-noise ratio in the images, but it does not explain the observation that some HSP16.5 particle images showed larger diameters and not all the class-sum images displayed mirror symmetry. Since the HSP16.5 reconstruction

EPR analysis of aA-crystallin shows a de®ned tertiary fold for the a-crystallin domain and a clear dimer interface, features that are presumably conserved in aB-crystallin and HSP27 (Koteiche & Mchaourab, 1999). How can these results be reconciled with the highly variable quaternary structures of HSP27 and aB-crystallin reported here? A ¯exible linker region between the a-crystallin domain and either the N or C termini would be all that is required to form polydisperse complexes with variable quaternary structures. The degree of ¯exibility may restrict the range of possible oligomers formed. How can sHSPs form such diverse quaternary structures, from octahedral to highly variable? The answer to this question may lie in the structure of the unconserved N- and C-terminal regions surrounding the a-crystallin domain. In the HSP16.5 crystal structure, the region N-terminal to

Discussion Comparison of cryo-EM and X-ray crystal structures

269

Symmetric to Variable sHSP Assemblies

the a-crystallin domain is involved in subunit contacts around the 4-fold axis, while the C-terminal region is involved in contacts at the 3-fold axis (Kim et al., 1998a). If each sHSP has a unique fold for the N- and C-terminal regions, this could lead to quite different overall packing arrangements while still conserving the dimer interface between neighboring a-crystallin domains. In summary, we have shown by parallel cryoEM studies that sHSP assemblies display varying amounts of structural ¯exibility in solution. HSP16.5 assemblies are the most monodisperse and display at least loose, if not perfect, octahedral symmetry. When symmetry is imposed during 3D image processing of HSP16.5, the resulting cryoEM reconstruction is similar to the crystal structure with additional internal density presumably corresponding to the disordered N-terminal 32 residues. At the opposite extreme of the sHSPs studied, HSP27 assemblies were found to have a wide range of particle diameters and oligomeric states. a-Crystallin assemblies, in the middle of the continuum, appeared to have a more restricted range of diameters, however they displayed no overall symmetry. Native a-crystallin and the complex of aBcrystallin with unfolded a-lactalbumin show more structural heterogeneity than aB-crystallin. Perhaps the relaxed nature of the quaternary packing in the a-crystallin assemblies facilitates binding of a heterogeneous population of unfolded target proteins.

Materials and Methods Protein purification Puri®cation of recombinant human aB-crystallin, HSP27 and lens bovine a-crystallin was performed as previously described (Horwitz et al., 1998; Mchaourab et al., 1997). The cDNA of HSP16.5 was cloned from the genomic DNA of M. jannaschii by PCR ampli®cation. The forward primer was designed to contain an NdeI site ¯anking a 15 base sequence identical to that of the HSP16.5 deposited in GenBank under accession number U67483. Similarly, the reverse primer was designed to contain a XhoI site and overlapped the 15 C-terminal bases of the published sequence. The PCR fragment was then digested with NdeI and XhoI and subcloned into the vector pET20b ‡ . The resulting insert was sequenced, and the sequence was veri®ed to code for HSP16.5. The pET20b‡ vector containing the HSP16.5 clone was transferred into E. coli BL21DE3 cells. Cells were grown and harvested as described by Horwitz et al. (1998). Whole cell lysate containing HSP16.5 was applied to a Mono-Q column (Amersham Pharmacia Biotech, Piscataway, NJ) pre-equilibrated with 20 mM Tris-HCl (pH 8.2), 100 mM NaCl. HSP16.5 was eluted from the column with a linear gradient of 0.1 to 1 M NaCl and fractions containing HSP16.5 were pooled and concentrated. After adjusting the pH to 7.0, ammonium sulfate was added to a ®nal concentration of 1 M and this solution was applied to a phenyl Sepharose HP column (Amersham Pharmacia Biotech) pre-equilibrated in 50 mM sodium phosphate, (pH 7.0), 1 M ammonium sulfate. HSP16.5 was eluted from the column with a linear gradient of 1 to 0.1 M ammonium sulfate. Relevant fractions were pooled, concentrated and applied to a FPLC system with a Superose

6HR (10/30) column (Amersham Pharmacia Biotech). SDS-PAGE analysis revealed that HSP16.5 was puri®ed to near homogeneity. For comparison, a test sample of HSP16.5 was also prepared as above and heated to 85  C to mimic the normal environment of the hyperthermophile M. jannaschii. This sample appeared more polydisperse by gel-®ltration with a larger half bandwidth compared to the unheated sample. For the aB-crys./a-lact. complex, a-lactalbumin was obtained from Sigma (St Louis, MO) and unfolded by reducing the disul®de bonds with DTT. Since aB-crystallin does not possess any disul®de bonds, treatment with DTT does not affect its structure. A 2:1 (ww) ratio of aB-crystallin to a-lactalbumin was incubated with 20 mM DTT at 37  C for two hours to allow aB-crys./ a-lact. complex formation. Gel-®ltration chromatography analysis of all samples was performed on a Superose 6HR column. The ¯ow rate was 0.5 ml/minute, and the elution buffer was 50 mM sodium phosphate (pH 7.0), 0.1 M NaCl. The standard proteins used for calibration included: myoglobin (17 kDa), catalase (232 kDa), ferritin (440 kDa), thyroglobulin (670 kDa), GroEL (801 kDa), and ferritin dimer (880 kDa). Gel-®ltration fractions were analyzed by SDSPAGE gels stained with Coomassie blue. Densitometric analysis revealed a molar ratio of 2:1 aB-crystallin to a-lactalbumin in the complex. Cryo-electron microscopy Cryo-EM sample grids were prepared by placing a 4 ml droplet of sample on a glow-discharged holey carbon grid, followed by blotting and plunging into ethane slush (Adrian et al., 1984). Cryo-electron microscopy was performed on a Philips CM120 transmission electron microscope (FEI, Hillsboro, OR) equipped with a LaB6 ®lament, a Gatan (Pleasanton, CA) 626 cryo-transfer holder, and a Gatan slow-scan CCD camera (1024  1024 pixels). Micrographs (384 for aB-crys./a-lact.; 405 for a-crystallin; 72 for HSP27; and 196 for HSP16.5) were taken with a nominal magni®cation of 60,000 and with a set ÿ1.5 mm defocus value. Additional micrographs (203) were collected for HSP16.5 with defocus values of ÿ1.0 mm and ÿ0.8 mm. The particle diameters were measured using the Gatan Digital Micrograph software Ê on the sample scale package, and the pixel size of 3 A was con®rmed by calibration with a catalase crystal. Preliminary image processing The QVIEW software package (Shah & Stewart, 1998) was used to interactively select particles (4973 for aBcrys./a-lact.; 5038 for a-crystallin; 484 for HSP27; and 5772 for HSP16.5) from the cryo-electron micrographs. The selected particles were extracted in 80  80 pixel boxes. Note that out of the 5772 particle images of HSP16.5, 3321 of these were collected with a ÿ1.5 mm defocus value. A subset of 1000 (ÿ1.5 mm defocus) particle images were used for the asymmetric HSP16.5 reconstruction. Further image processing was performed with the IMAGIC software system (van Heel et al., 1996) on Digital Equipment Corporation/Compaq alpha workstations. Self-search test for octahedral symmetry To test how well each sHSP conformed to octahedral symmetry, we performed a self-search for octahedral

270 symmetry in IMAGIC on particle images of each sample. For comparison, we also tested projections of the HSP16.5 crystal structure and cryo-EM images of an asymmetric particle, DNA-PKcs. The atomic coordinates of HSP16.5 (1SHS) were obtained from the Protein Data Bank Research Collaboratory for Structural Bioinformatics web site (www.rcsb.org/pdb/) and converted to a 3D-density volume (803 voxels). This 3D-density Ê resolution using a custovolume was ®ltered to 13 A mized module in the AVS visualization software package (Advanced Visualization Systems, Inc., Waltham, WA). The projections of the ®ltered crystal structure were normalized and Gaussian noise was added in IMAGIC. Charles Chiu provided 1000 cryo-EM particle images of DNA-PKcs which were collected on the same microscope with the same nominal magni®cation of 60,000x and the same defocus value of ÿ1.5 mm (Chiu et al., 1998).

Image classification and mass selection In the case of aB-crystallin, native a-crystallin, and aB-crys./a-lact, a mass selection step was used to select for 1000 particle images corresponding, at least approximately, to assemblies with the average molecular mass. As described earlier (Haley et al., 1998), mass selection was performed on class-sum images, which have enhanced signal-to-noise ratios over the unclassi®ed particle images. The gel-®ltration bandwidths at half peak height were used to estimate the average molecular mass as well as the mass range for each sample (a-crystallin, 650 (140) kDa; native a-crystallin, 800 (200) kDa; and aB-crys./a-lact., 900 (170) kDa). The mass-selected particle images of the Ê resolution) a-crystallin assemblies (®ltered to 23 A were reclassi®ed into 150 class-sum images and a subset is shown in Figure 6. For comparison, 150 classsum images were also generated of HSP16.5 from parÊ resolution and collected ticle images ®ltered to 23 A with a ÿ1.5 mm defocus value.

3D reconstruction For HSP16.5, 3321 unclassi®ed particle images (®lÊ resolution and collected at ÿ1.5 mm tered to 23 A defocus) were used to calculate a preliminary asymmetric reconstruction that went through iterative re®nement cycles. The ®rst preliminary reconstructions of the a-crystallin assemblies were based on the massselected class-sum images described above. Several cycles of re®nement proceeded using the unclassi®ed Ê resolution. The ®nal particle images ®ltered to 23 A asymmetric reconstructions of HSP16.5 and the a-crystallin assemblies presented in Figure 4(a) and Figure 7 were calculated from CTF corrected particle images (1000 for HSP16.5; 919 for aB-crystallin, 858 for native a-crystallin; and 1169 for aB-crys./a-lact.). In addition, asymmetric reconstructions were calculated for each sample using randomly assigned Euler angles covering the asymmetric unit uniformly. In each case, a FSC between two half reconstructions based on random angles indicated worse resolution than found for the non-random asymmetric reconstructions. This result is not surprising for HSP16.5, since the non-random asymmetric reconstruction does reveal loose octahedral symmetry. For the three a-crystallin assemblies,

Symmetric to Variable sHSP Assemblies this result supports the idea that although the non-random reconstructions may represent average structures of variable assemblies, the orientational angles found are not completely random. To test for octahedral symmetry, we imposed 2-, 3-, or 4-fold axes along probable HSP16.5 symmetry axes and along arbitrary axes for the a-crystallin assemblies, which displayed no likely symmetry axes. We then examined the resulting density maps for any additional, non-imposed symmetry axes. The octahedral reconstruction of HSP16.5 was based on 5772 particle images collected with three different defocus values (ÿ1.5 mm, ÿ1.0 mm, and ÿ0.8 mm). A ®rst preliminary reconstruction was calculated from 1000 particle images with a defocus value of ÿ1.5 mm and assuming octahedral symmetry. This reconstruction was used to ®nd orientational angles for all 5772 particle images Ê resolution. During the ®nal re®nement ®ltered to 10 A cycle, orientational angles were determined using a 3  angle search. The ®nal octahedral reconstruction was based on CTF corrected images ®ltered to the Nyquist Ê resolution. frequency of 6 A CTF correction was performed as described previously (Chiu et al., 1998) with the following CTF parameters (Cs ˆ 2 mm, fraction of amplitude contrast ˆ 0.1, kV ˆ 120, decay constant ˆ 10 nm2). Decay constants of 5 and 15 nm2 were also tried for the HSP16.5 octahedral reconstruction, but yielded no improvement for the FSC comparison between the cryo-EM reconstruction and the ®ltered crystal structure. Three methods of resolution assessment were used: the Fourier shell correlation (FSC) between two half reconstructions (each based on one half of the data set) (BoÈttcher et al., 1997; Conway et al., 1997); a modi®ed FSC between a density map based on the Ê resolution crystal structure of HSP16.5 ®ltered to 13 A and the octahedral HSP16.5 reconstruction; and the Fourier ring correlation (FRC) between particle images and their corresponding reconstruction reprojection images (van Heel, 1987b). For each method the resolution was taken to be the point at which the correlation fell to 0.5. The modi®ed FSC between the HSP16.5 crystal structure and the octahedral HSP16.5 reconstruction was performed two ways, both with Ê to remove and without a ``soft'' mask of radius 30 A the internal weak density in the cryo-EM reconstruction. The soft-mask option in IMAGIC uses a Gaussian fall-off to prevent possible correlation at a sharp edge. The AVS visualization package (Advanced Visualization Systems, Inc.) was used to display the 3D reconstructions. Contour levels were chosen assuming a protein density of 1.36 g/cm3.

Acknowledgments We thank Dr Joe Horwitz for his support and helpful discussions throughout the project; Dr Charles Chiu for supplying the DNA-PKcs particle images; and Linlin Ding for help with sample preparation. This work was supported by grants from the National Science Foundation (MCB-9722353) (P.L.S.), Bank of America Giannini Fellowship (M.P.B.), and the National Institutes of Health (EY12018 and EY12683 to H.S.M., EY3897 to Q.L.H., and T32-EY07026 to D.A.H.).

Symmetric to Variable sHSP Assemblies

References Adrian, M., Dubochet, J., Lepault, J. & McDowall, A. W. (1984). Cryo-electron microscopy of viruses. Nature, 308, 32-36. Ban, N., Freeborn, B., Nissen, P., Penczek, P., Grassucci, R. A., Sweet, R., Frank, J., Moore, P. B. & Steitz, Ê resolution X-ray crystallographic T. A. (1998). A 9 A map of the large ribosomal subunit. Cell, 93, 11051115. Behlke, J., Lutsch, G., Gaestel, M. & Bielka, H. (1991). Supramolecular structure of the recombinant murine small heat-shock protein HSP25. FEBS Letters, 288, 119-122. Bhat, S. P. & Nagineni, C. N. (1989). aB Subunit of lensspeci®c protein a-crystallin is present in other ocular and non-ocular tissues. Biochem. Biophys. Res. Commun. 158, 319-325. BoÈttcher, B., Wynne, S. A. & Crowther, R. A. (1997). Determination of the fold of the core protein of hepatitis B virus by electron cryo-microscopy. Nature, 386, 88-91. Bova, M. P., Ding, L. L., Horwitz, J. & Fung, B. K. (1997). Subunit exchange of aA-crystallin. J. Biol. Chem. 272, 29511-29517. Carver, J. A. (1999). Probing the structure and interactions of crystallin proteins by NMR spectroscopy. Prog. Retinal Eye Res. 18, 431-462. Carver, J. A. & Linder, R. A. (1998). NMR spectroscopy of a-crystallin. Insights into the structure, interactions and chaperone action of small heat-shock proteins. Int. J. Biol. Macromol. 22, 197-209. Chang, Z., Primm, T. P., Jakana, J., Lee, I. H., Serysheva, I., Chiu, W., Gilbert, H. F. & Quiocho, F. A. (1996). Mycobacterium tuberculosis 16-kDa antigen (HSP16.3) functions as an oligomeric structure in vitro to suppress thermal aggregation. J. Biol. Chem. 271, 72187223. Chiu, C. Y., Cary, R. B., Chen, D. J., Peterson, S. R. & Stewart, P. L. (1998). Cryo-EM imaging of the catalytic subunit of the DNA-dependent protein kinase. J. Mol. Biol. 284, 1075-1081. Conway, J. F., Cheng, N., Zlotnick, A., Wing®eld, P. T., Stahl, S. J. & Steven, A. C. (1997). Visualization of a 4-helix bundle in the hepatitis B virus capsid by cryo-electron microscopy. Nature, 386, 91-94. Das, K. P., Petrash, J. M. & Surewicz, W. K. (1996). Conformational properties of substrate proteins bound to a molecular chaperone a-crystallin. J. Biol. Chem. 271, 10449-10452. de Jong, W. W., Caspers, G.-J. & Leunissen, J. A. M. (1998). Genealogy of the a-crystallin: small heatshock protein superfamily. Int. J. Biol. Macromol. 22, 151-162. Derham, B. K. & Harding, J. J. (1999). a-Crystallin as a molecular chaperone. Prog. Retinal Eye Res. 18, 463509. Ehrnsperger, M., Lilie, H., Gaestel, M. & Buchner, J. (1999). The dynamics of HSP25 quaternary structure. Structure and function of different oligomeric species. J. Biol. Chem. 274, 14867-14874. Groenen, P. J., Merck, K. B., de Jong, W. W. & Bloemendal, H. (1994). Structure and modi®cations of the junior chaperone a-crystallin. From lens transparency to molecular pathology. Eur. J. Biochem. 225, 1-19. Haley, D. A., Horwitz, J. & Stewart, P. L. (1998). The small heat-shock protein, aB-crystallin, has a variable quaternary structure. J. Mol. Biol. 277, 27-35.

271 Head, M. W., Corbin, E. & Goldman, J. E. (1993). Overexpression and abnormal modi®cation of the stress proteins aB-crystallin and HSP27 in Alexander disease. Am. J. Pathol. 143, 1743-1753. Head, M. W., Corbin, E. & Goldman, J. E. (1994). Coordinate and independent regulation of aB-crystallin and HSP27 expression in response to physiological stress. J. Cell. Physiol. 159, 41-50. Horwitz, J. (1992). a-Crystallin can function as a molecular chaperone. Proc. Natl Acad. Sci. USA, 89, 1044910453. Horwitz, J., Huang, Q. L., Ding, L. & Bova, M. P. (1998). Lens a-crystallin: chaperone-like properties. Methods Enzymol. 290, 365-383. Horwitz, J., Bova, M. P., Ding, L., Haley, D. A. & Stewart, P. L. (1999). Lens a-crystallin: function and structure. Eye, 13, 403-408. Ingolia, T. D. & Craig, E. A. (1982). Four small Drosophila heat-shock proteins are related to each other and to mammalian a-crystallin. Proc. Natl Acad. Sci. USA, 79, 2360-2364. Kato, K., Shinohara, H., Kurobe, N., Goto, S., Inaguma, Y. & Ohshima, K. (1991). Immunoreactive aA-crystallin in rat non-lenticular tissues detected with a sensitive immunoassay method. Biochim. Biophys. Acta, 1080, 173-180. Kim, K. K., Kim, R. & Kim, S. H. (1998a). Crystal structure of a small heat-shock protein. Nature, 394, 595-599. Kim, R., Kim, K. K., Yokota, H. & Kim, S. H. (1998b). Small heat-shock protein of Methanococcus jannaschii, a hyperthermophile. Proc. Natl Acad. Sci. USA, 95, 9129-9133. Klemenz, R., Frohli, E., Steiger, R. H., Schafer, R. & Aoyama, A. (1991). aB-Crystallin is a small heatshock protein. Proc. Natl Acad. Sci. USA, 88, 36523656. Koteiche, H. A. & Mchaourab, H. S. (1999). Folding pattern of the a-crystallin domain in aA-crystallin determined by site-directed spin labeling. J. Mol. Biol. 294, 561-577. Lambert, H., Charette, S. J., Bernier, A. F., Guimond, A. & Landry, J. (1999). HSP27 multimerization mediated by phosphorylation-sensitive intermolecular interactions at the amino terminus. J. Biol. Chem. 274, 9378-9385. Lee, G. J., Roseman, A. M., Saibil, H. R. & Vierling, E. (1997). A small heat-shock protein stably binds heat-denatured model substrates and can maintain a substrate in a folding-competent state. EMBO J. 16, 659-671. Mchaourab, H. S., Berengian, A. R. & Koteiche, H. A. (1997). Site-directed spin-labeling study of the structure and subunit interactions along a conserved sequence in the a-crystallin domain of heat-shock protein 27. Evidence of a conserved subunit interface. Biochemistry, 36, 14627-14634. Rao, C. M., Raman, B., Ramakrishna, T., Rajaraman, K., Ghosh, D., Datta, S., Trivedi, V. D. & Sukhaswami, M. B. (1998). Structural perturbation of a-crystallin and its chaperone-like activity. Int. J. Biol. Macromol. 22, 271-281. Renkawek, K., de Jong, W. W., Merck, K. B., Frenken, C. W., van Workum, F. P. & Bosman, G. J. (1992). aB-Crystallin is present in reactive glia in Creutzfeldt-Jakob disease. Acta Neuropathol. 83, 324-327. Renkawek, K., Voorter, C. E., Bosman, G. J., van Workum, F. P. & de Jong, W. W. (1994). Expression

272

Symmetric to Variable sHSP Assemblies

of aB-crystallin in Alzheimer's disease. Acta Neuropathol. 87, 155-160. Rye, H. S., Roseman, A. M., Chen, S., Furtak, K., Fenton, W. A., Saibil, H. R. & Horwich, A. L. (1999). GroEL-GroES cycling: ATP and nonnative polypeptide direct alternation of folding-active rings. Cell, 97, 325-338. Shah, A. K. & Stewart, P. L. (1998). QVIEW: software for rapid selection of particles from digital electron micrographs. J. Struct. Biol. 123, 17-21. Srinivasan, A. N., Nagineni, C. N. & Bhat, S. P. (1992). aA-Crystallin is expressed in non-ocular tissues. J. Biol. Chem. 267, 23337-23341. Stewart, P. L., Fuller, S. D. & Burnett, R. M. (1993). Difference imaging of adenovirus: bridging the resolution gap between X-ray crystallography and electron microscopy. EMBO J, 12, 2589-2599. Stewart, P. L., Chiu, C. Y., Haley, D. A., Kong, L. B. & Schlessman, J. L. (1999). Resolution issues in single particle reconstruction. J. Struct. Biol. 128, 58-64. van den Oetelaar, P. J., van Someren, P. F., Thomson, J. A., Siezen, R. J. & Hoenders, H. J. (1990). A dynamic quaternary structure of bovine a-crystallin

as indicated from intermolecular exchange of subunits. Biochemistry, 29, 3488-3493. van Heel, M. (1987a). Angular reconstitution: a posteriori assignment of projection directions for 3D reconstruction. Ultramicroscopy, 21, 111-124. van Heel, M. (1987b). Similarity measures between images. Ultramicroscopy, 21, 95-100. van Heel, M., Harauz, G., Orlova, E. V., Schmidt, R. & Schatz, M. (1996). A new generation of the IMAGIC image processing system. J. Struct. Biol. 116, 17-24. van Noort, J. M. (1996). Multiple sclerosis: an altered immune response or an altered stress response? J. Mol. Med. 74, 285-296. Vanhoudt, J., Aerts, T., Abgar, S. & Clauwaert, J. (1998). Quaternary structure of bovine alpha-crystallin: in¯uence of temperature. Int. J. Biol. Macromol. 22, 229-237. Wynne, S. A., Crowther, R. A. & Leslie, A. G. (1999). The crystal structure of the human hepatitis B virus capsid. Mol. Cell, 3, 771-780. Xu, Z., Horwich, A. L. & Sigler, P. B. (1997). The crystal structure of the asymmetric GroEL-GroES-(ADP)7 chaperonin complex. Nature, 388, 741-750.

Edited by W. Baumeister (Received 7 January 2000; received in revised form 22 February 2000; accepted 29 February 2000)