Subunit arrangement in the dodecameric chloroplast small heat shock protein Hsp21

Abstract

Unfolding proteins are prevented from irreversible aggregation by small heat shock proteins (sHsps) through interactions that depend on a dynamic equilibrium between sHsp subunits and sHsp oligomers. A chloroplast-localized sHsp, Hsp21, provides protection to client proteins to increase plant stress resistance. Structural information is lacking concerning the oligomeric conformation of this sHsp. We here present a structure model of Arabidopsis thaliana Hsp21, obtained by homology modeling, single-particle electron microscopy, and lysine-specific chemical crosslinking. The model shows that the Hsp21 subunits are arranged in two hexameric discs, similar to a cytosolic plant sHsp homolog that has been structurally determined after crystallization. However, the two hexameric discs of Hsp21 are rotated by 25° in relation to each other, suggesting a role for global dynamics in dodecamer function.

Introduction

During transient stress, aggregation of unfolding proteins is prevented by the protein–protein interactions with the small heat shock protein (sHsp) chaperones.^1–4 The expression of sHsps is increased by stress, and after upregulation they may become among the most abundant cellular proteins.⁵ The sHsps are oligomeric proteins showing a functionally important rapid subunit exchange,^6–10 in which hydrophobic surfaces hidden in the subunit interfaces of the oligomer become exposed for interactions with the unfolding client proteins.^11–15 Recent data indicate that the sHsp subunits and the protected client proteins form a very polydisperse ensemble of complexes with large client binding site flexibility.^16,17

All sHsps that have been structurally characterized to high resolution show dimeric subunit building blocks, in which each monomer contains a β-sandwich fold^11,12,18,19 in the conserved C-terminal domain. Some sHsps have a homogeneous oligomer population, others show heterogeneous quarternary structures, and many different oligomeric conformations have been detected, including 2-, 4-, 12-, and 24-mers.^1,3,20

The N-terminal domain is especially important for the oligomer disassembly and subunit exchange.^9,21,22 In the chloroplast-localized sHsp, Hsp21, this domain contains a unique motif, with conserved methionines in an amphipathic α-helix.²³ This N-terminal motif is important for regulation, client protein-binding and oxygen radical scavenging,^24–26 and evolved during land plant development.²⁷ Although Hsp21 increases stress resistance in Arabidopsis thaliana plants,²⁸ there is neither structural information concerning the oligomeric configuration of this important plant sHsp nor its subunit arrangement. One plant sHsp has been crystallized and structure-resolved to date, a cytosolic homolog from Triticum aestivum,¹¹ Hsp16.9, showing a dodecamer of two hexameric discs. This Hsp16.9 homolog is considerably smaller than the chloroplast-localized Hsp21 homolog and lacks the methionine-rich motif in the N-terminal domain, which is 41 amino acids longer in Hsp21 than in Hsp16.9.

The N-terminal domain in sHsps is flexible and disordered^29,30 and often obstructing crystallization.¹⁹ We have tried to crystallize Hsp21, sofar without success. To investigate if Hsp21 is also a double-disc dodecamer,¹¹ a tetrahedral dodecamer³¹ or a 24-mer,¹² we have adopted other approaches.

We have previously concluded from nanoelectrospray mass spectrometry that Hsp21 appears dodecameric, and after lysine-specific crosslinking, Hsp21 was detected in denaturing electrophoresis as a 250 kDa band that could correspond to a dodecamer.³² Here, we have used negative stain single particle EM and image reconstruction to generate protein density maps for Hsp21, in samples with and without such lysine-specific crosslinking. Using the dimeric building block of the cytosolic homolog Hsp16.9¹¹ as a template, a structure model for dimeric Hsp21 has been generated and subsequentially fitted into these protein density maps. It emerges that Hsp21, like the cytosolic Hsp16.9, shows D3 symmetry and that a tetrahedral dodecamer arrangement and D6 symmetry can be excluded. Moreover, the two hexameric discs of Hsp21 display a rotation of the upper and lower disc by about 25° in relation to each other, in both native and crosslinked dodecamers.

Results

Homology modeling of Hsp21

The sequence similarity between the chloroplast-localized Hsp21 from A. thaliana and the cytosolic Hsp16.9 homolog from T. aestivum is 38%. There is high sequence similarity in the conserved α-crystallin domain, whereas the N-terminal domain is more variable, and the unique methionine-containing amphipatic α-helix motif is conserved only among chloroplast Hsp21 ortologs (Fig. 1). The crystal structures of sHsp homologs^11,12,19 show the typical β-sandwich fold of the α-crystallin domain. The plant Hsp16.9 from T. aestivum (12-mer) and the archaeal Hsp16.5 from M. jannaschii (24-mer) are both made up of structurally very similar dimeric building blocks.¹¹ The two monomers of each dimer interact via a β-strand exchange where β6 from one monomer binds to β2 in the other monomer [Fig. 2(A)]. Their higher oligomeric organization is also based on essentially the same interaction, namely the C-terminal tail containing a short beta-strand that binds to a hydrophobic groove on a neighboring dimer. Such interactions (12 or 24) on the outside hold the oligomer together. In both Hsp16.9 and Hsp16.5 oligomers, the N-terminal domains are located in the oligomer interior, either resolved or disordered in the crystal structure. As dimers are the least common nominator among sHsps and Hsp16.9 has highest sequence similarity to Hsp21 among the known structures, we selected the dimeric building block of Hsp16.9 as a template for homology modeling of Hsp21. This modeling resulted in a dimeric building block [Fig. 2(B)] that was used to generate a dodecameric Hsp21 structure model after fitting to the protein density maps as described below.

Figure. 1.

Sequence alignment of chloroplast-localized and cytosolic sHsps. The following sHsp sequences were aligned: Hsp21, the chloroplast-localized sHsp from Arabidopsis thaliana (P31170), pea (Pisum sativum, P09886) and wheat (Triticum aestivum, Q00445), here aligned without chloroplast signal sequences, and cytosolic Hsp16.9-homologs (class I) from A. thaliana (P19037), pea (P. sativum, P19243), and wheat (T. aestivum, Q41560), Hsp16.9. The β-strands in the α-crystallin domain as determined by X-ray crystallography for Hsp16.9 (van Montfort et al., 2001) are indicated below the alignment, with the long loop that engages in strand exchange to stabilize the two monomers in italics. The C-terminal binding groove is shown in light grey, the conserved IXI/V-motif in C-terminal extension, which covers the binding groove in black, and the conserved methionine-rich motif in the N-terminal domain of Hsp21 in dark grey. All lysine residues are in bold face. The Hsp21 lysine residues that were detected in lysine-specific crosslinks (Table I) and mapped into the Hsp21 structure model (Fig. 7) are indicated with arrow and number. Multiple alignment was performed with Clustal W (http://www.ebi.ac.uk). [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Figure. 2.

Homology modeling of Hsp21. (A) Schematic view of the dimeric building blocks of the oligomers formed by the cytosolic Hsp16.9 from T. aestivum and Hsp16.5 from M. jannaschii, for which the crystal structures were solved (1GME and 1SHS in PDB). The antiparallel β-sheets made up of β-strands 2, 3, 9, and 8 and 7, 5, and 4 form a β-sandwich in each monomer, and β-strand 6 of each monomer binds to β-strand 2 of the other monomer, which stabilizes the dimer. (B) Homology model of chloroplast-localized Hsp21 dimer, based on the dimeric building block of the cytosolic homolog Hsp16.9. (C) Schematic top and side view of the subunit arrangement in the Hsp16.9 dodecamer. The six dimers formed by monomers A-B, C-D, E-F, G-H, I-J, and K-L (designated as in PDB-file 1GME¹¹) are organized as two hexameric discs. (D) Schematic view of a tetrahedral organization of a dodecameric sHsp, based on Acr1 from M. tuberculosis. Each of the six dimers formed by monomers A-B, C-D, E-F, G-H, I-J, and K-L are located on the six edges of a tetrahedron. The image in Fig. 2(B) was prepared with the molecular visualization program PyMOL (http://www.pymol.org).

The six dimers can be imagined to form a dodecamer that is comprised of two hexameric discs, each of which is built up of three dimers [Fig. 2(C)] as in the structure of the crystallized Hsp16.9 cytosolic homolog,¹¹ or as six dimers located on the six edges of a tetrahedron [Fig. 2(D)], as in the sHsp Acr1 from Mycobacterium tuberculosis.³¹ As the interactions stabilizing either conformation are expected to be the same (C-terminal tails binding to the hydrophobic grooves in neighboring dimers), and as structural studies of sHsps so far have convincingly illustrated the ability of this interaction to connect dimers at quite different angles, Hsp21 could in principle adopt any of the different conformations.

Single particle averaging and image reconstruction

We have previously observed that the Hsp21 dodecamer is stabilized by lysine-specific crosslinking such that it remains dodecameric even after SDS-solulization.³² In this closer structural investigation, the raw negative-stain electron micrographs showed similarly appearing spherical particles of crosslinked Hsp21 [Fig. 3(A)] and native (not crosslinked) Hsp21 [Fig. 3(B)]. These particles were selected for single particle image processing. An interpretable reconstruction was obtained after contrast reversal, and the resulting 3D map was used for identifying domains. Three-dimensional reconstructions with D3, D6, and tetrahedral symmetry were generated.

Figure. 3.

Negative stain EM images of Hsp21 dodecamers. (A) Crosslinked Hsp21. (B) Native Hsp21. Similarly appearing particles were seen in the raw images for both samples. Scale bars, 100 nm. The crosslinked Hsp21 particles were negatively stained by 1% uranyl acetate, whereas the native Hsp21 particles appeared positively stained, that is, the particles were stained more than the background and were used for image reconstruction after inverting the micrograph scans.

Inspection of the reconstruction with tetrahedral symmetry made us discard this alternative, as the only way in which 6 dimeric building blocks can satisfactorily fit into a tetrahedron is at its edges, where no protein density was seen in the reconstruction. Instead, the reconstruction showed protein density at the four planes of a tetrahedron, which is where no density is expected if the 6 dimers are located at the six edges (data not shown). The results of multivariate statistical analysis and two-dimensional classification presented in the Supporting Information also support the exclusion of tetrahedral symmetry.

To decide between D3 and D6-symmetry, we used rotational power-spectra of the respective top view classes in D3 and D6-symmetry. The rotational power spectrum for the raw images classified as top views in the D3-refinement clearly supports the assumed D3-symmetry, because it shows threefold, sixfold, and ninefold peaks. On the other hand, the rotational power spectrum from the top views of the D6-refinement contradicts the assumed D6-symmetry as it shows mainly a ninefold peak, which is compatible with a threefold but not a sixfold symmetry. It also turns out that the ratio between the number of particles in top view, and the typical number in any other view is only 1:5 for the D6-reconstruction, whereas it is 1:2 for the D3-reconstruction. This indicates that particles showing an apparent sixfold symmetry are actually quite rare. Additionally, the agreement between model projections and class averages (particularly for the top view) is better in D3 than in D6. The results of multivariate statistical analysis and two-dimensional classification presented in the Supporting Information also support the choice of D3-symmetry rather than D6. These findings led us to continue all image analysis in the D3 point group.

Effects of crosslinking on the Hsp21 dodecamer

The image reconstructions of the Hsp21 dodecamer particles were obtained at a resolution of about 15 Å, measured at 0.5 Fourier shell correlation. The image reconstructions showed differences between the crosslinked Hsp21 dodecamer, which had a compact appearance [Fig. 4(A)], and the native Hsp21 dodecamer, which had an apparently empty interior cavity [Fig. 4(B)].

Figure. 4.

Surface rendered views of the crosslinked and native Hsp21 dodecamer after image reconstruction. (A) Hsp21 dodecamer crosslinked with DTSSP, top view (left) and side view (right). (B) Hsp21 dodecamer without crosslinker, top view (left) and side view (right). Scale bars, 10 Å.

Comparing the two reconstructions by overlaying the structures in top view [Fig. 5(A)] and side view [Fig. 5(B)] further emphasizes this difference between the native Hsp21 dodecamer (blue) and the more compact crosslinked Hsp21 dodecamer (grey), with the apparently empty interior of the native Hsp21 dodecamer evident especially in clipped views, from top [Fig. 5(C)] and side [Fig. 5(D)]. The 12 subunits could be assigned to the 12 densities [inserts Fig. 5(A,B), using the same subunit letter code as in Fig. 2(C)]. The subunits B, D, F, H, J, and L protrude out from the structure in the native Hsp21 dodecamer (blue). This is not the case in the crosslinked Hsp21 dodecamer (grey).

Figure. 5.

Comparison of the crosslinked and native Hsp21 dodecamers. The top view (A) and side view (B) of the overlaid crosslinked and native image reconstructions show that the crosslinked Hsp21 dodecamer (grey) is more compact compared with the native Hsp21 dodecamer (blue), from which every second subunit is protruding out. Inserts show assignment of the 12 subunits with subunit letter code as in Fig. 2(C). (C, D) Clipped top view and side view reveal an apparently empty interior in the native, but not in the crosslinked, Hsp21 dodecamer.

Docking a structure model of Hsp21 into the density maps

Having established the D3 symmetry of the Hsp21 dodecamer in the image reconstructions, we performed docking of the Hsp21 dimeric building blocks [Fig. 2(B)] into the protein density maps. The image reconstructions, of both the crosslinked and the native Hsp21 dodecamer, suggested that there is a rotation of the upper and lower discs to each other. The six Hsp21 dimers were therefore docked as two separate hexameric rings, resulting in a good fit to the Hsp21 density map [Fig. 6(A)]. The docking of a nonrotated double disc dodecamer gave a less good fit [Fig. 6(B)].

Figure. 6.

Docking of atomic coordinates of the Hsp21 structure model into the density map. The Hsp21 structure model was docked into the density map with two hexameric discs (red and yellow), consisting of three dimers each, docked separately. (A) Good fit with a relative rotation of about 25° between the two Hsp21 hexameric discs, (B) no fit without rotation of the two Hsp21 hexameric discs as in the Hsp16.9 template structure, (C) side view and (D) clipped side view of the rotated Hsp21 structure model, showing how the twisted N-terminal domains of the subunits can account for the density in the interior of the dodecamer. (E) Schematic presentation of the 25° rotation of the two Hsp21 hexameric discs required to fit into the protein density map. Scale bar, 10 Å. All figures show docking onto the density map of crosslinked Hsp21, but the fitting of the subunits required the rotation also for the density map of native Hsp21 (data not shown).

The relative rotation required to obtain this fit was about 25°. In this rotated Hsp21 structure model, the six N-terminal domains have also been twisted around Ile-82 toward the interior of the dodecamer to account for the density in the center of the structure, which is seen more clearly in side view [Fig. 6(C)] and clipped side view [Fig. 6(D)]. The rotation of the two hexameric discs of Hsp21 by 25° in relation to each other is schematically presented in Figure 6(E).

Structural mapping of crosslinks into the Hsp21 structure model

The crosslinked Hsp21 dodecamer, with a more compact appearance [Figs. 4(A) and 5 in grey], has previously been analyzed by protease digestion and mass spectrometry to identify the crosslinked peptides.³² With two complementary proteases, trypsin or endoproteinase Glu-C, six crosslinked peptides were detected (Table I) and mapped into the Hsp21 structure model (Fig. 7). Crosslinked peptides containing half (8 of 16) of the Hsp21 lysine residues (K106, K121, K125, K126, K155, K157, K173, and K177) fitted into the structure model, with the assumption that the distance between two carbon-beta atoms of two lysines is <20 Å (12 + 8 Å = lengths of crosslinker and 2x the lysine side chain).

Table I.

Crosslinked Hsp21-Hsp21 Peptides Mapped into Hsp21 Structure Model

MH+	Crosslinked peptides	Crosslinked lysines		Dist (Å)	Subunits	Protease	Crosslinka
1695.746	M101PGLSKE107 + M101PGLSKE107	K106	K106	20	A-B	Glu-C	I
1466.669	M101PGLSKE107 + Q124KKE127	K106	K125/126	15	A-A	Glu-C	II
2123.947	D115NVLVIKGE123 + Q124KKEDSDD131	K121	K125/126	16	A-A	Glu-C	III
3125.440	I111SVEDNVLVIKGEQK125 + K126EDSDDSWSGR136	K121	K126	19	A-A	Trypsin	III
1506.802	I156KAELK161 + T172KVER176	K157	K173	18	A-A	Trypsin	IV
2252.104	D154KIKAELK161 + K177VIDVQIQ184	K155/157	K177	17	A-D	Trypsin	V

^aCrosslinked distance mapped into structure, see Figure 7.

Crosslinked peptides were detected by mass spectrometry after cleavage with either trypsin or endoproteinase Glu-C as previously described³² and used for mapping into the Hsp21 structure model in Figure 7. Crosslinked peptides involving residues in the N-terminal domain (K18, K27, and the N-terminal amine) are not included in the table, because they cannot be mapped into the Hsp21 structure model in Figure 7, because the N-terminal domain is very flexible, and with low homology with the template Hsp16.9 used for the modeling. When more than one crosslinkable lysine is present in a peptide, the lysine used for the structural mapping is underlined.

Figure. 7.

Structural mapping of crosslinked peptides into the Hsp21 structure model. Detected lysine to lysine crosslinks (Table I) were mapped into the structure model of Hsp21, with the assumption that the distance between two carbon-beta atoms of two lysines is <20 Å (12 + 8 Å = length of DTSSP crosslinker and 2x the lysine side chain). For clarity, only one dimer is presented with subunit A (dark grey) and subunit B (lighter grey), plus one subunit (D) in a neighboring dimer in the same hexameric disc. Subunit letter code is as in Fig. 2(C). The image was prepared using the molecular visualization program PyMOL (http://www.pymol.org). [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Most crosslinks were intramonomeric (K106-K125, K121-K126, and K157-K173). One of the crosslinks, K106-K106, was intermonomeric and would covalently fix the two monomers within a dimer together. The crosslink between K155/157 and K177 is between two different dimers in the same hexameric disc.

Discussion

A rotated double disc conformation of the Hsp21 dodecamer

We have generated protein density maps for dodecamers of the chloroplast-localized sHsp Hsp21, by negative stain EM and single particle image reconstruction. Hsp21 dimers, obtained by homology modeling to the dimeric building block of the smaller cytosolic Hsp16.9 homolog, were used for docking into the density maps. From the fitting of the Hsp21 dimers into the actual protein density maps, we conclude that Hsp21 also has a dodecameric conformation, with two stacked hexameric discs, each containing three dimers, and that there is a rotation of the two discs by about 25° in relation to each other.

The compact appearance of the crosslinked Hsp21 dodecamer

The crosslinked Hsp21 dodecamer had a compact appearance [Fig. 4(A)], compared with the native [Fig. 4(B)]. The crosslinker DTSSP is locking the Hsp21 dodecamer such that it can no longer undergo subunit dissociation.³² The detected crosslinks between lysine residues (Table I, Fig. 7) should indeed reduce the flexibility of the Hsp21 dodecamer. The crosslinking between K106 and K106 should fix the two monomers in a dimer, and crosslinking between K155/157 and K177 should glue together the C-terminal tail of one subunit onto another subunit of a neighboring dimer.

In spite of this crosslinking, the seven amino acids C-terminal of K177 would still be flexible and some substrate binding would be possible, also to the hydrophobic binding groove. Substrate protein aggregation was indeed somewhat suppressed by crosslinked Hsp21 during in vitro measurements (data not shown), yet functionality in vitro is only partly reflecting the full functionality in vivo^29,30 that requires the rapid subunit exchange,^6–10 which of course is not possible with the crosslinked Hsp21.

There is also extensive crosslinking between the N-terminal domains of the subunits (K18, K27 and the N-terminal amine).³² The flexible N-terminal domains that become less flexible when crosslinked is presumably what becomes visible as protein density in the interior of the crosslinked dodecamers [Fig. 6(A,C)].

The flexible tails in the native Hsp21 dodecamer

The protein density maps showed that the native Hsp21 dodecamers are indeed less compact in appearance compared with the crosslinked Hsp21 dodecamers. In the native Hsp21 dodecamer [Fig. 4(B)] six densities that correspond to six halves of the dimers, or even six separate monomers,³³ appear to protrude from the structure. The subunits that protrude (B, D, F, H, J, and L, Fig. 5, blue) correspond to the subunits in the Hsp16.9 crystal structure¹¹ that have C-terminal tails that interact with the other hexameric disc in the dodecamer. A possible interpretation is that the protein density map of the native Hsp21 dodecamer displays C-terminal tails (of the subunits B, D, F, H, J, and L) that are flexible in solution, in contrast to the compact crosslinked Hsp21 dodecamer and in contrast to the Hsp16.9 crystal structure. Such flexibility of the C-terminal tails of sHsps is well known from NMR^34,35 and single point mutations and other alterations that restrict the flexibility reduce the chaperoning activity.^12,22,36

Significance of the rotation of the two Hsp21 hexameric discs

Why would the Hsp21 dodecamer, in contrast to the Hsp16.9 template, be rotated? Is this difference due to that the N-terminal domain is different in the chloroplast-localized Hsp21, having the unique amphipathic α-helix methionine rich motif and being 41 amino acids longer compared with the cytosolic homolog Hsp16.9? Or is the rotated Hsp21 dodecamer configuration just one, out of several possible, dodecameric states and the nonrotated configuration of Hsp16.9 structurally determined after crystallization a special, highly symmetric state selected during the crystallization process? The former possibility is most likely. If several conformations were possible one might have expected these to show up during the classification procedure. On the other hand, the various conformations are not necessarily very abundant in the mixture of conformations and therefore might not influence three-dimensional reconstruction very much. A careful analysis of conformations might reveal them if they are not extremely rare.

Without the rotation, the interdisc interactions are between subunits positioned directly above each other, for example, the hydrophobic binding groove of subunit A is covered by the C-terminal tail of subunit H [Fig. 6(E), upper]. Upon clockwise rotation of the upper and lower disc by ∼25° in relation to each other [Fig. 6(E), lower], the interaction to the other disc might be weaker, and the flexible C-terminal tails that strap across to the other disc (A, G, C, I, E, and K) may or may not still reach across. Perhaps the protein density maps have caught a glimpse of a situation of global dynamics, with the rotated hexameric discs reflecting the dodecamers “in action,” with the functionally important rapid subunit exchange.^6–10

To summarize, by docking onto the protein density maps, obtained by negative stain EM and image reconstruction, we have obtained an improved structure model for Hsp21 dodecamer in which two hexameric rings are rotated in relation to each other. The structure model is validated by detected lysine-specific crosslinks and will be useful in further structural mapping and crosslinking, within Hsp21 and between Hsp21 and client proteins.

Materials and Methods

Sample preparation

Hsp21 from Arabidopsis thaliana was recombinantly expressed in E. coli and purified as described previously.²⁴ Samples (12 μM Hsp21, 50 mM HEPES pH 8.0, and 5 mM MgCl₂) were treated with the crosslinker DTSSP (3,3′-dithiobis[sulfosuccinimidylpropionate]) from Pierce (Nordic BioLabs, Täby, Sweden) as described in Ref.32. Final concentration of crosslinker was 5 mM (1 μL from a freshly prepared stock solution in 50 mM HEPES pH 8 and 5 mM MgCl₂). The crosslinking reaction was quenched after 20 min by tris-(hydroxyl-methyl)-aminomethane, added to a final concentration of 20 mM. Samples were frozen in liquid nitrogen and stored at −80°.

For electron microscopy, samples were thawed and kept in room temperature for 30 min and then applied to glow discharged carbon-coated grids and stained with 1% uranyl acetate.

Crosslinked peptides were detected by mass spectrometry after overnight digestion at 37°C with trypsin (Promega, Madison) or Endoproteinase Glu-C (Roche Applied Science, Bromma, Sweden), according to Ref.32.

Homology modeling and docking of atomic coordinates into map

The Hsp21 structure model was obtained by modeling the Arabidopsis thaliana Hsp21 sequence (P31170, after replacing the chloroplast transit peptide with a start methionine used in recombinant expression), onto chains A and B of the structure of Hsp16.9,¹¹ retrieved from the Protein Data Bank (http://www.pdb.org, 1GME), using the program 3D-JIGSAW.³⁷ Docking of the structure model onto density maps for further refinement, as described in the Results section, was performed using the “fit-to-map” utility in Chimera³⁸ and the model building in O.³⁹

Electron microscopy and image processing

Images were recorded, using a Philips CM 120 microscope at 120 kV and a magnification of 60,000 times, onto Kodak SO-163 films and digitized with a Zeiss SCAI scanner, with 28-μm scan steps, corresponding to a pixel size of about 4.7 Å at the specimen level. A total of 14,829 particles was selected from 18 micrographs and 9546 particles from six micrographs for the crosslinked and noncosslinked Hsp21 samples, respectively, and cut out into 64 × 64 pixel boxes using the EMAN-tool Boxer.⁴⁰ The effects of the PCTF were corrected in the stack of raw images by inverting every second resolution band (phase-flipping). The defocus of the micrographs, ranging from 0.3 to 2.6 μm under focus, was determined by fitting the phase contrast transfer function using the EMAN-tool Ctfit.⁴⁰

For 3D reconstruction D3- and D6- or quasi-D6-symmetries were considered based on the D3-symmetry of the dodecameric Hsp16.9 from T. aestivum,¹¹ as well as tetrahedral symmetry, which has been observed for the dodecameric sHsp Acr1 from M. tuberculosis by negative stain EM.³¹ We initially calculated 3D-maps of the crosslinked sample in the D3- and D6-point groups from 4328 images taken at defocus-values between 1.1 and 1.6 micrometers and, therefore, exhibiting somewhat higher contrast. To aid in choosing between D3 and D6 symmetry, the particles classified as top views (threefold or sixfold view, respectively) were used to determine rotational power spectra according to the following method in the image-processing package Khoros:⁴¹ (1) Particle images were centred by translationally aligning them with a smooth ring of comparable dimensions, (2) centred particle images were transformed to polar coordinates, (3) images in polar coordinates were Fourier transformed with respect to the angular coordinate, and (4) the magnitudes of these Fourier transforms were averaged over a radial range of 4–6.5 nm (8–14 pixels) for each particle and the results were averaged over all images to give an average rotational power spectrum for all top views.

Starting models in C6 and C3 symmetry for initiating Eman's “refine”-procedure were generated using the Eman command “startcsym” on the CTF-corrected (phase-flipped) particle images. The multivariate statistical analysis and 2D-classification presented in the Supporting Information were only used for choosing the appropriate symmetry but otherwise had no part in calculating the 3D-reconstructions. To test tetrahedral (2,3) symmetry, the reconstructions were also refined in tetrahedral symmetry using the Eman procedure “refine” with reconstructions in D3 symmetry as starting models. The resolution was determined by Fourier shell correlation between two 3D-maps calculated from half the set of particle images each. The two half sets were generated by splitting the images within each angular class into a set of odd-numbered, and a set of even-numbered images without changing the angular assignment.

The programs used for image processing of the micrographs and surface rendering of the 3D-models were EMAN⁴⁰ and Chimera,³⁸ respectively. More information concerning the 3D reconstructions can be found in the Supporting Information.