Structure and conformational cycle of a bacteriophage-encoded chaperonin

Chaperonins are ubiquitous molecular chaperones found in all domains of life. They form ring-shaped complexes that assist in the folding of substrate proteins in an ATP-dependent reaction cycle. Key to the folding cycle is the transient encapsulation of substrate proteins by the chaperonin. Here we present a structural and functional characterization of the chaperonin gp146 (ɸEL) from the phage EL of Pseudomonas aeruginosa. ɸEL, an evolutionarily distant homolog of bacterial GroEL, is active in ATP hydrolysis and prevents the aggregation of denatured protein in a nucleotide-dependent manner. However, ɸEL failed to refold the encapsulation-dependent model substrate rhodanese and did not interact with E. coli GroES, the lid-shaped co-chaperone of GroEL. ɸEL forms tetradecameric double-ring complexes, which dissociate into single rings in the presence of ATP. Crystal structures of ɸEL (at 3.54 and 4.03 Å) in presence of ATP•BeFx revealed two distinct single-ring conformational states, both with open access to the ring cavity. One state showed uniform ATP-bound subunit conformations (symmetric state), whereas the second combined distinct ATP- and ADP-bound subunit conformations (asymmetric state). Cryo-electron microscopy of apo-ɸEL revealed a double-ring structure composed of rings in the asymmetric state (3.45 Å resolution). We propose that the phage chaperonin undergoes nucleotide-dependent conformational switching between double- and single rings and functions in aggregation prevention without substrate protein encapsulation. Thus, ɸEL may represent an evolutionarily more ancient chaperonin prior to acquisition of the encapsulation mechanism.


Introduction
Chaperonins are large double-ring complexes that mediate protein folding in an ATP-dependent mechanism in all domains of life [1]. Two major groups of chaperonins exist, which transiently encapsulate non-native substrate protein (SP) for folding to proceed in an aggregationfree environment: Group I chaperonins occur in eubacteria and organelles of prokaryotic origin, mitochondria and chloroplasts (GroEL, Hsp60 and Cpn60, respectively). These chaperonins have 7-membered rings. The group II chaperonins in the cytosol of archaea and eukaryotes (thermosome and TRiC/CCT, respectively) have 8-mer rings. The subunits of both Gene product 146 (gp146 also known as ɸEL) from Pseudomonas bacteriophage EL encoded by the plasmid pET22b-phi-GroEL was expressed in E. coli BL21(DE3) cells. Cells were grown in LB medium containing 100 mg L -1 ampicillin at 37˚C to an optical density of 0.5 at 595 nm. Expression of ɸEL was induced with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) for 3 h at 30˚C. Cells were harvested and re-suspended in ice-cold buffer PA (50 mM Tris-HCl pH 7.5). The cells were lysed by the French press method on ice using an Emulsiflex C5 apparatus (Avestin, Ottawa, Canada). After removal of cell debris by centrifugation at 120,000 g for 45 min, the supernatant was applied to a 50 mL Source 30Q column (GE Healthcare) equilibrated in buffer PA. The column was washed with 3 column volumes (CV) buffer PA containing 50 mM NaCl, followed by a linear salt gradient in buffer PA (50-500 mM NaCl, 10 CV). The eluate fractions were analyzed by SDS-PAGE. Fractions containing ɸEL (eluting at 17-21.5 mS cm -1 ) were merged and transferred into buffer PB (50 mM Tris-HCl pH 7.5, 1 mM EDTA, 1 mM DTT). Next, the material was passed over a 20 mL Heparin Sepharose Fast-Flow column (GE Healthcare), which was eluted with a gradient from 0 to 500 mM NaCl in buffer PB (10 CV). Finally, the concentrated fractions containing ɸEL (eluting at 8-15 mS cm -1 ) were subjected to size exclusion chromatography (SEC) on Sephacryl S-400 (GE Healthcare) in buffer, 20 mM MOPS-NaOH pH 7.2, 100 mM NaCl and 10% (v/v) glycerol.

Size-exclusion chromatography coupled to multi-angle static light scattering (SEC-MALS)
Purified ɸEL at 2 g L -1 was analyzed using static and dynamic light scattering by auto-injection of the sample onto a SEC column (5 μm, 4.6x300 mm column, Wyatt Technology, product # WTC-030N5) at a flow rate of 0.35 mL min -1 in buffers EM50 (20 mM HEPES-NaOH pH 7.5, 50 mM KCl, 4 mM Mg acetate) or EM100 (20 mM HEPES-NaOH pH 7.5, 100 mM KCl, 2 mM Mg acetate) at 25˚C in the presence or absence of nucleotide (2 mM). The column was in line with the following detectors: a variable UV absorbance detector set at 280 nm (Agilent 1100 series), the DAWN EOS MALS detector (Wyatt Technology, 690 nm laser) and the Optilab rEXTM refractive index detector (Wyatt Technology, 690 nm laser) [12]. Molecular masses were calculated using the ASTRA software (Wyatt Technology) with the dn/dc value set to 0.185 mL g -1 . Bovine serum albumin (Thermo) was used as the calibration standard.
Rho refolding assays were performed as described previously with minor modifications [14]. GuHCl-denatured Rho was diluted 200-fold to a final concentration of 0.5 μM into buffer, 20 mM Tris-HCl pH 7.5, 100 mM KCl and 5 mM MgCl 2 , either lacking chaperone or containing ɸEL, ɸEL/GroES, GroEL or GroEL/GroES. The concentrations of GroEL and ɸEL were 1 μM (tetradecamer) and GroES 2 μM (heptamer). Refolding was initiated upon addition of ATP (5 mM). When indicated, chaperonin action was stopped by 1,2-cyclohexylenedinitrilotetraacetic acid (CDTA) (10 mM). Enzymatic assays of Rho were performed as previously described [14]. Spontaneous refolding of Rho was inefficient (< 10% yield) due to aggregation. Crystallization ɸEL at 19.1 g L -1 in buffer, 20 mM MOPS-NaOH pH 7.3, 100 mM KCl and 2 mM Mg-acetate, was crystallized by the sitting-drop vapor diffusion method by mixing 100 nl sample with 100 nl reservoir solution using the robotics setup at the Crystallization Facility of the Max Planck Institute of Biochemistry. The drops were equilibrated against 150 μl reservoir solution at 16˚C. Crystals of ɸEL were obtained with the Complex crystallization screen [15,16].
Crystal form I was obtained in presence of 2 mM ATP, 5 mM BeSO 4 and 20 mM NaF with a reservoir solution containing 5% (w/v) PEG-4000, 0.2 M Na-acetate and 0.1 M Na 3 -citrate pH 5.5.
Crystal form II was obtained in presence of 2 mM ATP, 5 mM BeSO 4 and 20 mM NaF with a reservoir solution containing 8% (w/v) PEG-6000, 0.15 M NaCl and 0.1 M Tris-HCl pH 8.0.
For vitrification, the crystals were sequentially incubated in reservoir solution containing additionally 12.5 and 25% (v/v) glycerol for 15 min each and were then rapidly cooled in liquid nitrogen.

Diffraction data collection and processing
The X-ray diffraction data were collected by the oscillation method at beamline ID30B of the European Synchrotron Radiation Facility (ESRF) in Grenoble, France. All data were integrated and scaled with XDS [17]. Pointless [18], Aimless [19] and Ctruncate [20], as implemented in the CCP4i graphical user interface (GUI) [21], were used for data reduction.

Crystal structure solution and refinement
The space group symmetry and size of the asymmetric unit suggested that crystal forms I and II contained single-ring heptamers at 65% solvent content, which is within the expected range for protein crystals (~75-40% solvent) [22]. Analysis of the self-rotation function calculated with Molrep [23] indicated the presence of seven-fold non-crystallographic symmetry (NCS) consistent with the presence of single-ring heptamers in the crystal lattice. For solving the structure of crystal form II by molecular replacement, the cryo-EM density for the heptadecameric ɸEL•ATP at 6.8 Å resolution (EMDB entry EMD-6492, [10]) was segmented with Chimera [24] and the density for a single-ring heptamer extracted. With this density as a search model, a plausible molecular replacement solution was obtained. The density modification program Resolve [25] was used to extend the phases beyond 6.8 Å employing seven-fold NCS averaging and refinement. After B-factor sharpening, the resulting experimental electron density revealed features of secondary structure sufficient for manual model building with Coot [26]. ɸEL crystal form I was solved by molecular replacement using Molrep [23] with these coordinates as a search model. To identify the nucleotides bound to subunits of ɸEL, omit maps were calculated after applying random coordinate shifts to the preliminary model with Pdbset (to suppress model bias) and refinement with Refmac5 [27], which revealed density consistent with ADP-bound to chains B, D and F in crystal form II, and with ATP/ADP•BeF x in all other chains. Since ATP and ADP•BeF x cannot be distinguished at the resolution, and since ATP was added to the crystallization mix, ATP was included into the model. The models were refined with Refmac5 using local NCS restraints and translation-libration-screw (TLS) parametrization of B-factors [27]. Residues 1 and 553-558 were disordered in all chains of the final models. Furthermore, no interpretable density was observed for loop residues 290-294 in crystal form I and in chains A, C, E and G of crystal form II. Residues facing solvent channels with disordered sidechains were truncated after C-β.
Crystallographic structure factors and model coordinates for crystal forms I and II of ɸEL•ATP-BeFx have been deposited to wwPDB under accession numbers 6TMT and 6TMU, respectively.

Cryo-electron microscopy and single particle analysis
The samples were prepared by mixing equal volumes of ɸEL stock solution (2.25 g L -1 ) in buffer, 20 mM HEPES-NaOH pH 7.5, 100 mM NaCl and 1 mM EDTA, and dilution buffer (20 mM HEPES-NaOH pH 7.5 and 8 mM MgCl 2 ) containing either no nucleotide or 4 mM ADP or ATPγS. The dilution buffer with ADP also contained 0.08% (w/v) n-octyl-β-D-glucopyranoside. Subsequently the mixture was incubated at room temperature for 5 min. Holey carbon supported copper-grids (Quantifoil R2/1 300 mesh) were plasma-cleaned for 30 s (Harrick Plasma) immediately before use. All cryo-grids were prepared using a Vitrobot Mark 4 (FEI) by applying sample (5 μL) to a plasma-cleaned grid at 25˚C and 100% humidity, then semi-automatically blotted for 2 s and plunge-frozen in liquid ethane.
The grids with ɸEL•ADP or ɸEL•ATPγS were analyzed on a Talos Arctica (FEI) transmission electron microscope (TEM) at 200 kV. Frames were recorded with a Falcon 3EC direct detector (FEI) operated in movie mode at 0.05 s per frame, at a pixel size of 1.997 Å and 2.019 s total exposure, with an estimated cumulative dose of 42-43 e − Å -2 . EPU (FEI) software was used for automated data collection. MotionCor2 [28] was employed to correct the movie stacks for beam-induced motion and dose weighting. Ctffind4.1 [29] was used to estimate the defocus. For generating templates used in auto-picking, 6,668 single particles of ɸEL•ADP were picked manually and subjected to 2D classification with RELION 3.0 [30,31]. The eight largest 2D classes were selected and used in Gautomatch (http://www.mrc-lmb.cam.ac.uk/ kzhang/Gautomatch) as templates for automated particle picking.
The grid with apo-ɸEL was analyzed on a Titan Krios (FEI) TEM at 300 kV with a pixel size of 1.09 Å. Data were collected with a K3 direct detector (Gatan) recording 50 frames per movie during 5.992 s total exposure with an estimated cumulative dose of 77.6 e − Å -2 . SerialEM software was used for automated data collection [32,33]. Motion correction, dose weighting, defocus analysis and particle picking were carried out automatically during data collection using the Focus software package [34]. Movies with large drift, exhibiting ice diffraction or poor resolution (> 5 Å) in the power spectra were immediately discarded.
In absence of nucleotide and in presence of ADP, ɸEL had a tendency to associate into large aggregates. Images with thick aggregates were removed after visual inspection. RELION 3.0 was used for further data processing [31]. The complete data sets went through two rounds of 2D classification to remove contaminations or false positive particles. Using RELION 3.0, an initial model was generated from the remaining particles and used as a reference map for symmetry-free 3D refinement. The aligned particles were subjected to 3D classification. The particles from the largest class were used for further 3D refinement and post-processing including mask application and B-factor sharpening. The resulting EM density maps were inspected with Chimera [24], and C2 (apo-ɸEL) and D7 (ɸEL•ADP and ɸEL•ATPγS) particle symmetry were detected. Re-processing of the particles using the same protocol as above, but with application of symmetry restraints, yielded improved maps for manual model building with Coot [26]. First, the apical, intermediate and equatorial domains of ɸEL from the form-I crystal structure were separately placed into density by rigid-body real-space fitting. After manual adjustment of the coordinates to the density, the models were refined with Refmac5 in reciprocal space, using jelly-body and NCS restraints [27].
Cryo-EM density maps for apo-ɸEL, ɸEL•ADP and ɸEL•ATPγS have been deposited to EMDB under accession numbers 10528, 10529 and 10530, respectively. The corresponding model coordinates have been deposited to wwPDB under accession codes 6TMV, 6TMW and 6TMX.

Results
Oligomeric state of ɸEL ɸEL was expressed as a soluble protein in E. coli and purified (S1A Fig). To determine the oligomeric state of ɸEL at different ionic strength and in the absence or presence of nucleotide (at 25˚C), we subjected ɸEL to size exclusion chromatography combined with multi-angle light scattering (SEC-MALS). At 100 mM KCl, ɸEL fractionated at a molecular weight (MW) of 770 kDa (Fig 1A), indicating a high population of double-ring complexes (theoretical MW 863 kDa). A similar MW (~725 kDa) was observed in the presence of ADP ( Fig 1B). In contrast, ɸEL in the presence of ATP shifted to a homogeneous population of complexes with a MW of~409 kDa, close to the MW of the single-ring heptamer (theoretical MW~431 kDa) ( Fig 1C). This suggests that binding of ATP to ɸEL in the presence of close to physiological salt concentration destabilizes the inter-ring contacts in the double ring structure. However, at a lower salt concentration of 50 mM KCl, this destabilization was not observed and mainly double-ring complexes were populated, as for apo and ADP-bound ɸEL (Fig 1A-1C). These results suggests that under conditions of ongoing ATP-binding and hydrolysis (at 100 mM KCl), ɸEL may cycle between double-ring and single-ring complexes.

ATPase activity of ɸEL
ATP hydrolysis by group I and II chaperonins exhibits positive cooperativity within rings, with higher ATP occupancy triggering ATP hydrolysis (Hill coefficient for GroEL~2.8) [38]. In addition, negative cooperativity between the rings results in a reduced hydrolysis activity at still higher ATP concentration [38]. At 25˚C and 10 mM KCl, ɸEL hydrolyzed ATP with near Michaelis-Menten kinetics, reaching a maximal turnover number of 558 ± 28 ATP min -1 per tetradecamer, which is~8-fold higher than that of E. coli GroEL [4,38]. A Hill coefficient of 1.21 ± 0.05 and an apparent K M of 0.68 ± 0.07 mM ATP were determined (Fig 2A). At 100 mM KCl, we measured a similar maximal turnover number (576 ± 7 ATP min -1 per tetradecamer), a Hill coefficient of 1.61 ± 0.18 and a~10-fold lower K M of 0.072 ± 0.008 mM ATP ( Fig  2B). Thus, at the near physiological salt concentration the affinity of ɸEL for ATP is increased with weak positive cooperativity for ATP hydrolysis. No evidence for negative cooperativity in ATP hydrolysis was detected. The higher ATPase activity and lower Hill coefficient of ɸEL compared to GroEL are consistent with a reduced level of allosteric coordination within and between rings of ɸEL.

No interaction of ɸEL with E. coli GroES
The co-chaperone GroES (2-fold molar excess) inhibited the GroEL ATPase activity by~50% both at 10 and 100 mM KCl (Fig 3A and 3B) [39]. In contrast, we observed no effect of E. coli GroES on the ATPase activity of ɸEL (at 1 mM ATP) independent of salt concentration (

PLOS ONE
and 3B). Note that at a concentration of 1 mM ATP, the ATPase rate of ɸEL measured at 10 mM KCl is~50% lower than in the presence of 100 mM KCl. Increasing GroES up to 8-fold excess over ɸEL also showed no effect (S1B Fig), suggesting the absence of a functional interaction. E. coli GroES shares 61% amino acid identity (90% similarity) with the GroES of the host bacterium of phage EL. Specifically, the mobile loop residues of GroES that mediate binding to the apical domains of GroEL are completely conserved between E. coli GroES and P. aeruginosa GroES (S1C Fig). To further analyze a possible interaction of GroES with ɸEL, we took advantage of the fact that the mobile loops of GroES become protected against degradation by proteinase K (PK) upon complex formation with GroEL in the presence of ADP (  We next analyzed the effect of SP on the ATPase of ɸEL. We used as model SP the double mutant of maltose binding protein (DM-MBP), which folds slowly (t 1/2~3 5 min at 25˚C) [43] in the absence of chaperonin and has a low aggregation propensity, thus allowing us to

PLOS ONE
perform ATPase measurements under SP saturation of chaperonin [44]. Binding of non-native DM-MBP stimulated the GroEL ATPase by~2-fold independent of salt concentration (Fig 3A  and 3B). Interestingly at 10 mM KCl, DM-MBP inhibited the ɸEL ATPase by~60% (Fig 3A). In contrast, no inhibition of ɸEL ATPase was observed at 100 mM KCl (Fig 3B). Thus, the effect of SP on the ATPase appears to depend on whether ɸEL populates double-ring complexes (low salt) or single-rings (high salt) (Fig 1C).
In summary, GroEL and ɸEL differ substantially with regard to their interaction with cochaperone (E. coli GroES) and SP.

Chaperone activity of ɸEL
Next, we investigated the molecular chaperone activity of ɸEL. We first tested the ability of ɸEL to prevent aggregation of the model chaperonin SP rhodanese (Rho;~30 kDa). This protein has a high propensity to aggregate upon dilution from denaturant into buffer, but folds efficiently upon transient encapsulation in the GroEL-GroES cage [42,[45][46][47][48]. Aggregation was monitored spectrophotometrically by measuring turbidity at 320 nm. A time-dependent aggregation of Rho was observed upon dilution from denaturant to a final concentration of 0.5 μM (Fig 4A and 4B). Note that Rho aggregation was independent of the presence of nucleotide (Fig 4A and 4B). GroEL at a 1:1 molar ratio, in the absence of nucleotide, completely prevented the aggregation of Rho in the turbidity assay. In contrast, ɸEL failed to prevent Rho aggregation ( Fig 4A). However, ɸEL inhibited Rho aggregation with high efficiency in the presence of ATP or ADP (Fig 4A and 4B). In the case of GroEL, aggregation was efficiently prevented in the presence of ADP (Fig 4B), whereas addition of ATP reduced the ability of GroEL to bind non-native Rho [45] (Fig 4A). Thus, in contrast to GroEL, SP binding to ɸEL requires presence of nucleotide.
Rho activity assays were performed to determine whether aggregation prevention by ɸEL was coupled to productive folding. Control reactions showed only~5% recovery of Rho enzyme activity during spontaneous folding (Fig 4C), due to Rho aggregation. Efficient refolding was obtained by GroEL in a manner dependent on GroES and ATP [42,45]. In contrast, only~8% of Rho enzyme activity was recovered in the presence of ɸEL and ATP (Fig 4C). Addition of GroES was without effect (Fig 4C), consistent with the absence of a functional interaction of ɸEL with E. coli GroES (Figs 3 and S1B).
To investigate the fate of Rho during attempted refolding with ɸEL and ATP, we analyzed the reaction after 30 min by SEC in the presence of ATP, followed by SDS-PAGE and immunoblotting with anti-Rho. ɸEL fractionated in an asymmetric distribution (with the peak in fraction 8), consistent with cycling between double-and single-ring complexes (S2A Fig). Rho fractionated with a higher mass than ɸEL (with the peak in fraction 7), indicating that it formed a high molecular weight aggregate as it dissociated from ɸEL during gel filtration (S2A In conclusion, ɸEL can bind non-native Rho in the presence of nucleotide thereby preventing aggregation. However, ATP-dependent cycling of Rho fails to promote productive folding, independent of the presence of GroES. Apparently, during cycling Rho is transiently released in an unfolded state, explaining its aggregation during gel filtration when rebinding to ɸEL is precluded.

Crystal structures of ɸEL in presence of ATP•BeF x
To obtain insight into the structural features underlying the ATP-dependent double-to singlering transition in ɸEL, we tried to determine the crystal structures of apo-ɸEL and of ɸEL in presence of ADP or ATP•BeF X (ADP•BeF X is a mimic of bound ATP prior to hydrolysis). However, only crystals of ɸEL with ATP•BeF X diffracted below 7 Å resolution. Two crystal forms were identified that diffracted to 4.03 and 3.54 Å resolution ( Table 1). The structure was solved by molecular replacement at 6.8 Å resolution using the cryo-EM map of the ɸEL-ATP tetradecamer as a search model (EMD-6492) [10]. The seven-fold non-crystallographic symmetry was then employed to extend the phases to 3.54 Å, resulting in an electron density map of sufficient quality to build an atomic model. The other crystal form was solved by molecular replacement using this model. Both structures consist of heptameric single-ring complexes of ɸEL (Fig 5A and 5B).
As anticipated, the ɸEL subunits displayed the typical three-domain architecture of chaperonins, composed of an equatorial domain (residues 2-130 and 425-552), an intermediate domain (residues 131-188 and 388-424), and an apical domain (residues 189-387) (Fig 6A). These domains formed rigid-body units in the heptamer rings, which tend to move en bloc when comparing subunits (S3A-S3D Fig). The apical domains appeared to be the most mobile units, as judged from their comparatively poor electron density when not stabilized by crystal contacts. The equatorial domains were well-defined, as they form the majority of the contacts between adjacent subunits, similar to known chaperonin structures. In conformation I, the tip of the protruding αK-αL helical hairpin of the apical domain contacts the equatorial domain ( Fig  6A). The αK-αL helical hairpin makes additional contacts to the intermediate domain ( Fig  6A). These contacts appear to stabilize the orientation of the apical domain. No such contact is found in group I chaperonin structures (group II chaperonins do not contain this helical hairpin). There were no detectable contacts between apical domains unless forced by crystal packing. All subunits as judged from the electron density had either ATP or ADP•BeF X bound in the equatorial domain. For simplicity, we modelled the bound nucleotide as ATP. ATP was cradled by the αA-β2 loop (residues [30][31][32][33], the N-terminal ends of helix αD (residues 86-90) and helix αN1 (residues 428-430), helix αO (residues 474 and 478) and residues 504-506 and 519-521 ( Fig  6B). Thr145 and Gln149 at the C-terminal end of αF in the intermediate domain approached the nucleobase, but did not appear to make full contact (Fig 6B). Intermediate domain helix αM with the catalytic residue Asp412 (equivalent to Asp398 in E. coli GroEL) was at~6.5 Å distance to the γ-phosphate of ATP, consistent with a conformation poised for ATP hydrolysis (Fig 6B).  (Fig 6D and 6E). This re-orientation of the apical domains of chains A, C and E also results in the αK-αL helical hairpin to point outwards into the solvent (Figs 6D and S3C). The apical-intermediate domain intra-ring contact also requires a 19-21˚outward rotation of the intermediate domains in chains B, D and F (Figs 6C and S3B). All other inter-subunit contacts are limited to the equatorial domains ( Fig 6F).
All subunits in crystal form II contained electron density for a bound nucleotide (Fig 6G  and 6H). The orientation of the intermediate domain correlated with the identity of the bound nucleotide: The subunits with outward-oriented intermediate domain (chains B, D and F) contained weak nucleotide density, consistent with partial occupancy by ADP (Fig 6G). The βphosphate of ADP was coordinated by the amides at the N-terminal end of helix αD; additional density for BeF x was not detectable. The nucleotide electron density for the subunits with inward-oriented intermediate domain (chains A, C, E and G) was consistent with nucleoside triphosphate, modelled as ATP and Mg 2+ (Fig 6H). Compared to crystal form I, the catalytic residue Asp412 was more distant from the γ-phosphate (~8.5 Å) in these subunits. The nucleotide binding pattern is thus alternating in the ring, with the exception of the adjacent subunits G and A, which both harbor ATP and Mg 2+ . Interestingly, in two of the ADP-bound subunits (chains B and D), the tip of the αA-β2 loop, which cradles the nucleotide in the ATPbound subunits, was remodeled and flipped away from ADP (compare Fig 6B and 6G). This conformational change might facilitate dissociation of ADP from the chaperonin.

Cryo-EM structure of ɸEL in absence of nucleotide
To obtain the solution structure of apo-ɸEL, we next performed cryo-EM and single particle analysis. The raw micrographs and the particle 2D class averages suggested the presence of contact is highlighted by a box. (F) Contact between subunits in conformations III and II, as observed between chains B and A in crystal form II. The contacts are almost exclusively formed between the equatorial domains (boxed area). (G) Zoom-in on the bound ADP in chain D of crystal form II (conformation II). Unbiased Fo-Fc omit density at 1.7σ for the nucleotide is shown in green as meshwork. Movement of the intermediate domain and remodeling of the αA-β1 loop, respectively, are indicated by arrows. (H) Zoom-in on the bound ATP and Mg 2+ in chain A of crystal form II (conformation III). Unbiased Fo-Fc omit density at 2.85σ for the nucleotide is shown in green as meshwork.
https://doi.org/10.1371/journal.pone.0230090.g006 double-ring structures with approximate seven-fold rotational symmetry (S4A and S4B Fig). The particles appeared to have a strong tendency to associate via their apical domains (S4A Fig). Symmetry-free 3D class averaging indicated a tetradecameric double-ring structure with C2 symmetry (S4C Fig). Refinement of the particles resulted in a density map at 3.45 Å resolution (Figs 7A and S4D), which allowed construction of an atomic model (Fig 7B and Table 2). The contacts between subunits were well defined in the density (Fig 7A-7C). Similar to the crystal structures, the apical domains were the most mobile elements of the structure (S4E Fig). Apo-ɸEL has an open structure with a ring opening diameter at the apical domains of 65 Å (Fig 7A). This is in contrast to the previously reported 9 Å cryo-EM map of apo-ɸEL (EMD-6494) in which the ring opening is reduced to~27 Å due to rearrangement of the apical domains [10]. Overall each ring of the apo-ɸEL structure reported here is similar to the ɸEL heptamer of crystal form II (r.m.s.d. Cα positions 2.4 Å, versus 4.2 Å with crystal form I) in that the subunits alternate between conformations II (chains B, D, F) and III (chains A., C, E) and one subunit (chain G) is in conformation I (S5A-S5D Fig). Similar conformations of alternating subunits were recently observed by cryo-EM for the apo-state of the chaperonin of Pseudomonas phage OBP [50]. In all subunits of ɸEL the tip of the αA-β2 loop in the equatorial domain is moved away from the nucleotide binding site as in chains B and D of crystal form II (Fig 6G). Other nucleotide binding elements such as the P-loop (residues 79-86) and the β17-β18 hairpin (residues 503-512) are also moved outward (~1.5 Å) in the apo structure. These local conformational changes might facilitate ATP-binding.
Similar to group II chaperonins and unlike the staggered subunit contacts in group I, the ring-ring interface of apo-ɸEL is formed by two-fold symmetrical contacts between subunits. This results in the burial of 489-537 Å 2 of accessible surface area (Fig 7C and 7D). For comparison, the buried surface area between subunits within heptamer rings is 1218-1401 Å 2 . The inter-ring contacts involve residues 492-494 in one subunit forming van-der-Waals contacts with residues 460 � and 461 � of the subunit in the opposite ring (Fig 7D). The negative dipole at the C-terminal end of helix αN2 is positioned close to His492 � . Residues Asn465 and Asn465 � form a symmetrical contact; Glu466 makes polar contacts with Arg469 � and the backbone at Gly462 � . Moreover, there is a symmetrical van-der-Waals contact with the neighboring subunit in the opposite ring, involving residues Lys486 (Fig 7D). None of these contacts are observed in group I and II chaperonins due to the high divergence of sequences.

Cryo-EM structures of ɸEL in presence of ADP or ATPγS
To further elucidate the allosteric cycle of ɸEL, we determined the solution structure of ɸEL in the presence of ADP by single-particle cryo-EM. ɸEL formed double-ring structures with seven-fold rotational symmetry (S6A and S6B Fig). Symmetry-free 3D class averaging indicated an open double-ring structure with nearly uniform rings, and thus D7 symmetry was applied, resulting in a refined cryo-EM density map at 5.9 Å resolution (Figs 8A and S6C). Note that the resolution is simply limited by the experimental setup. No particle class with closed single rings was found, in contrast to a previous report showing that ɸEL•ADP (EMD-6493) forms single rings resembling a hollow sphere [10] (S6B Fig). We used the domain structures from the 3.54 Å crystal structure to rigid-body fit and refine a pseudo-atomic molecular model of the ADP-bound ɸEL double-ring (Fig 8A and Table 2). In all the subunits, the αK-αL helical hairpin in the apical domain contacts the equatorial domain and the intermediate domain is in the inward-rotated conformation (subunit conformation I). The ring-ring interface was similar to that of apo-ɸEL (r.m.s.d. 1.4 Å for Cα atoms of all 14 equatorial domains) (Fig 8B).
To mimic the conformational changes upon ATP-binding in solution, we also determined the cryo-EM structure of ɸEL in the presence of the slowly hydrolyzing ATP analog, ATPγS.

PLOS ONE
Double-ring structures were again observed (S6D and S6E Fig) and symmetry-free 3D class averaging indicated nearly uniform rings. Applying D7 symmetry resulted in a refined cryo-EM density map at 5.8 Å resolution (Figs 8C and S6F). The subunits assumed conformation I and were slightly tilted outward compared to ɸEL•ADP (Fig 8D). Interestingly, we find a rotation at the ring-ring interface of 2.9˚with a vertical displacement of 2.9 Å (Fig 8D). This re-orientation would break the contacts between Asn465 and Glu466 with Asn465 � and Arg469 � / Gly462 � , respectively (Fig 7D). Moreover, the symmetrical van-der-Waals contact at the Lys486 residues can no longer form. Thus, binding of ATPγS weakens the ring-ring interface and this effect may be more pronounced in the presence of ATP, consistent with the finding of mainly single-rings in solution (Fig 1C).

PLOS ONE
9A-9C), but share the same secondary structure topology, i.e. the β-sandwich core structure, α-helices αH, αI and αJ at the surface and the helical hairpin formed by helices αK and αL ( Fig  9D and 9E). However, the length of helices αH, αJ and αK and of the connecting surface loops differ substantially, resulting in a re-orientation of helices and re-modelling of long surface loops, such as residues 191-215 and 306-331, respectively (Fig 9D, boxed areas). The groove between helices αH and αI, which forms the binding site for substrate and GroES in GroEL, is more narrow and less deep in ɸEL. Accommodation of a helix or a β-hairpin element from SP would require structural remodeling of this site in ɸEL. In crystal form I, the putative SP binding site is solvent exposed, consistent with the ability of ɸEL to bind SP in the presence of ATP (Fig 4A). In contrast, in the apo-ɸEL (and crystal form II), helices αH and αI of chains A, C and E are partially occluded by the respective adjacent apical domain. The contact between the tip of the αK-αL helical hairpin to the equatorial domain in ɸEL is absent in GroEL, and the αK-αL connection is elongated and re-modelled in ɸEL (Fig 9A-9C). The intermediate domains of GroEL and ɸEL are quite similar (Fig 9F and 9G), consistent with a conserved function in coupling domain movements with changes of the nucleotide status of the equatorial domain.
The equatorial domains of ɸEL and GroEL differ mainly at the ring-ring interface contacts, with ɸEL making 1:1 subunit interactions and GroEL a 1:2 staggered interaction (Fig 9H and  9I). Helix αD is shortened in ɸEL by one turn compared to GroEL and helix αN is elongated by two turns (αN1), followed by insertion of a short helix, αN2. Helix αQ in GroEL (residues 462-471) is replaced by a loop connection in ɸEL (residues 479-503). This loop exhibits structural plasticity in the crystal structures of ɸEL single-ring.
Inter-domain salt bridges within subunits (equatorial domain D83 to apical domain K327) and between subunits (intermediate domain R197 to apical domain E386), which are important in allosteric regulation of GroEL [38], are not conserved in ɸEL.

Discussion
Our structural and functional analysis of the chaperonin ɸEL from the bacteriophage EL of P. aeruginosa revealed that the protein is ATPase active and functions in aggregation prevention of denatured proteins. ɸEL forms tetradecameric double ring complexes, which dissociate into a population of single rings in the presence of ATP and at physiological salt concentration. In contrast, the recently observed dissociation of GroEL into single rings occurs only transiently during the reaction cycle [4]. The nucleotide bound single-ring complexes in the crystal structures of ɸEL closely resembled the individual rings of the double-ring complexes analyzed by cryo-EM. We could not confirm the existence of a sphere-like single-ring structure proposed to function in SP encapsulation [10]. Our functional data rather suggests that the chaperone mechanism of ɸEL is encapsulation independent, and represents an evolutionary precursor of the more complex encapsulation mechanisms used by the group I and II chaperonins [2,5].
Based on our structural and functional analysis, we propose the following hypothetical model for the chaperonin cycle of ɸEL in SP binding and release, coupled to transitions between single-and double-ring complexes. Unlike group I and II chaperonins, the ability of ɸEL to prevent protein aggregation, as tested with Rho as a model SP, was strictly nucleotidedependent. In the apo-state, ɸEL is a double ring with six of the seven subunits per ring assuming alternating states (conformations II and III) and one subunit in conformation I (Fig 10). The putative SP binding sites in the apical domains of subunits A, C and E are partially occluded. This form acquires competence in SP-binding upon binding of ATP, whereupon the apical domains of all seven subunits per ring are shifted to the same state (conformation I) and the equatorial domains are poised for ATP hydrolysis (Fig 10). ATP-binding weakens the

PLOS ONE
ring-ring interface, resulting in single-ring formation at physiological salt concentration. SPs with a lower requirement for the availability of binding sites on adjacent subunits may also bind to the apo-state of ɸEL. ATP hydrolysis appears to occur in two stages: in the first stage, ATP is hydrolyzed in three alternating subunits, resulting in a ring conformation closely similar to the apo-state. This step may weaken the interaction with bound SP, perhaps allowing partial folding of some SPs. In the second stage, the remaining four subunits hydrolyze their ATP, followed by ADP dissociation generating the double-ring apo-state. This step presumably results in complete SP release for folding, as the apo-state is not SP-binding competent. Folding failed in the case of Rho, presumably because rebinding to ɸEL is faster than folding, consistent with the dependence of this slow folding protein on the encapsulation mechanism provided by GroEL/ES. We note that Rho is not a natural substrate of ɸEL and it would be important in future studies to investigate the mechanism of ɸEL in folding its cognate substrate proteins. Our structural analysis did not reveal an encapsulating state for ɸEL. Moreover, ɸEL did not functionally interact with GroES and does not contain a built-in-lid extension in its apical domains. However, the existence of a distant GroES homolog in the phage EL genome or an unrelated co-factor cannot be ruled out at this point. Distantly related GroES homologs have been identified in other phage genomes [53].
Only some bacteriophage genomes encode their own chaperonin, presumably to assist in the folding of an essential phage protein(s) that cannot utilize (or does so only partially) the host chaperonin system for folding [10,54]. Alternatively, ɸEL may prevent the folding of a phage protein up to a point when its function is required. For example, the putative ɸEL substrate gp188 is a cell wall endolysin [9] that should only function once a large number of phage particles have been produced by the host cell. ɸEL may stabilize gp188 in a non-native, inactive state until ATP levels may become depleted at the peak of phage production, resulting in concerted release from ɸEL and activation. It is also possible that additional regulatory factors play a role in triggering gp188 release. Future investigations will have to test the feasibility of such scenarios. showing the distribution of hydrophobic and charged residues. In the GroEL apical domain, helices αI and αH are separated by a wide, deep cleft, revealing the hydrophobic underlying segment. In the ɸEL apical domain, the respective helices contact each other, resulting in a shallow furrow. The underlying hydrophobic segment is occluded. Positively and negatively charged functional groups are indicated in blue and red, respectively, hydrophobic side chains in yellow and the rest of the surface in white. (EPS) staff at MPIB crystallization facility for performing automated crystallization screening. Expert assistance by Tillman Schäfer and Daniel Bollschweiler from the MPIB cryo-EM facility is gratefully acknowledged.