Visualizing chaperonin function in situ by cryo-electron tomography

Plasmids and strains

Escherichia coli BL21(DE3) Gold cells (Stratagene) were used for growth analysis, electron tomography and protein expression. For tomography and biochemical experiments, GroEL was expressed from a pBAD33 plasmid containing the groEL gene under the control of an araBAD promotor (EL⁺ cells)²⁶. For overexpression of GroEL and GroES, a pBAD33 plasmid containing both groEL and groES genes under the control of an araBAD promotor was used⁴⁸. MetK was expressed from a pET22b plasmid previously described²⁶.

Antibodies

Polyclonal antisera used against GroEL, GroES, MetK and GAPDH were previously described²⁶, and the rabbit antiserum against α-lactalbumin was a product of East Acres Biologicals immunization service.

E. coli growth

E. coli cells were grown in lysogeny broth (LB) medium that contained, depending on the plasmids used, the antibiotics ampicillin (200 μg ml^−1, pET22b-MetK) and chloramphenicol (32 μg ml^−1, pBAD33 variants). For overexpression of GroEL, GroES and MetK (MetK cells), transformed E. coli Bl21 (DE3) pBAD33-GroEL:ES pET22b-MetK cells were grown to early exponential phase at 37 °C, and GroEL–GroES expression using the pBAD33 promoter was induced for 90 min by supplementation of LB medium with arabinose to a final concentration of 0.2% (w/v). Cells were subsequently harvested by centrifugation at 8,000g (4 °C for 10 min) and resuspended to an optical density (600 nm, OD₆₀₀) of 0.1–0.2 in fresh LB medium containing both antibiotics and 1 mM isopropyl β-d-thiogalactopyranoside (IPTG), to induce MetK expression under control of the T7 promoter for 40 min. GroEL expression (EL⁺) was induced in E. coli Bl21 (DE3) pBAD33-GroEL by supplementation of LB medium with arabinose to a final concentration of 0.1% (w/v) and growth of the culture at 37 °C. To expose E. coli Bl21 (DE3) cells to HS, cells were first cultured to early exponential phase at 37 °C and then incubated in a shaking water bath at 46 °C for 2 h.

E. coli growth curves

Cells were cultured as described above. Aliquots were removed at the time points indicated for optical density measurement at OD₆₀₀. To ensure exponential growth conditions, growing cultures were diluted to an OD₆₀₀ of 0.1 with prewarmed LB medium containing the necessary antibiotics and arabinose when OD₆₀₀ just exceeded 0.4. Growth curves for MetK and EL⁺/MetK cells were measured following termination of GroEL induction by transfer of cells into arabinose-free medium containing 1 mM IPTG for MetK overexpression. The first sample was taken 5 min after changing the medium. Data were processed for fitting in R.

Protein expression and purification

GroEL, GroES and MetK proteins were expressed and purified as previously described^26,49.

Measurement of protein concentration

Concentrations of purified proteins were determined by measurement of absorbance at 280 nm using absorbance coefficients calculated from the protein sequence with the program ProtParam⁵⁰. Protein concentrations of cell lysates were determined with the Pierce Coomassie Plus (Bradford) Assay Kit (Thermo Fisher Scientific) as described by the manufacturer.

Preparation of cell lysates

Cultures were prepared as described above, harvested by centrifugation and the cell pellet flash-frozen in liquid nitrogen before further processing. Spheroplasts were prepared at 4 °C as previously described⁵¹. In brief, cells were resuspended in 100 mM Tris-HCl pH 8.0 and washed twice with 2 ml of buffer. The pellet was then resuspended in HMK buffer (50 mM HEPES-KOH pH 7.2, 20 mM Mg acetate, 50 mM K acetate) supplemented with 20% (w/v) sucrose and 0.25 mg ml^−1 lysozyme. Cells were then incubated on ice for 7 min and transferred to 37 °C for 10 min. The resulting suspension was supplemented with Complete EDTA-free protease inhibitor cocktail (Roche), and spheroplasts were lysed by the addition of 0.1% (v/v) Triton X-100 and subsequent sonication.

Mass spectrometry

Cell lysates were reduced by the addition of dithiothreitol (DTT) to a final concentration of 10 mM and heated to 56 °C for 45 min. Acylation of thiol groups was performed by the addition of chloroacetamide to a final concentration of 55 mM and incubation for 45 min in the dark, followed by a first digestion step with Lys-C (Wako) at a w/w ratio of 1:20 for 2 h at 37 °C. This was followed by a second digestion step overnight with trypsin (Roche) at a 1:20 (w/w) ratio at 37 °C. The reaction was stopped by the addition of trifluoroacetic acid to a final volume of 1%. Peptides were desalted using OMIX C18 (100 μl) tips (Agilent Technologies, no. A57003100) according to the manufacturer’s instructions.

Desalted peptides were dissolved in 12 µl of 5% formic acid, sonicated in an ultrasonic bath, centrifuged and transferred to autosampler vials (Waters). Samples were analysed on an Easy nLC-1200 nanoHPLC system (Thermo) coupled to a Q-Exactive Orbitrap HF mass spectrometer (Thermo). Peptides were separated on pulled-spray columns (ID 75 μm, length 30 cm, tip opening 8 μm, NewObjective) packed with 1.9 μm C18 particles (Reprosil-Pur C18-AQ, Dr Maisch) using either a stepwise 196 min gradient (comparison of 37 °C, HS and MetK) or a stepwise 67 min gradient (all other samples) between buffer A (0.2% formic acid in water) and buffer B (0.2% formic acid in 80% acetonitrile). Samples were loaded on the column by the nanoHPLC autosampler at a pressure of 900 bar. The high-performance liquid chromatography flow rate was set to 0.25 μl min^−1 during analysis. No trap column was used. The following parameters were used for comparison of growth conditions 37 °C, HS and MetK: MS, resolution 60,000 (full-width at half-maximum (FWHM) setting); MS mass range 300–1,650 m/z; MS-AGC-setting 3 × 10⁶; MS-MaxIT 50 ms; MS/MS fragmentation of the 15 most intense ions (charge state 2 or higher) from the MS scan; MS/MS resolution 15,000 (FWHM setting); MS/MS-AGC-setting 10⁵; MS/MS-MaxIT 50 ms; MS/MS isolation width 1.8 m/z; collision-energy setting 29 (NCE). All other samples were analysed with the following parameters: MS resolution 120,000 (FWHM setting); MS mass range 300–1,650 m/z; MS-AGC-setting 3 × 10⁶; MS-MaxIT 100 ms; MS/MS fragmentation of the ten most intense ions (charge state 2 or higher) from the MS scan; MS/MS resolution 15,000 (FWHM setting); MS/MS-AGC-setting 10⁵; MS/MS-MaxIT 50 ms; MS/MS isolation width 1.2 m/z; collision-energy setting 29 (NCE).

MS data analysis

Protein identification was performed using MaxQuant with default settings. The E. coli K12 strain sequences of UNIPROT (v.2023-03-01) were used as the database for protein identification (Supplementary Information). MaxQuant uses a decoy version of the specified UNIPROT database to adjust false discovery rates for proteins and peptides below 1%.

Quantification of MetK binding to GroEL

To quantify the fraction of GroEL with bound MetK in MetK-overexpressing cells, we immunoprecipitated GroEL with GroEL antibody followed by GroEL and MetK immunoblotting and liquid chromatography–tandem mass spectrometry. Cells were prepared and lysed as described above, but with the addition of apyrase (25 U ml^−1 final concentration) to rapidly deplete the ATP pool in the lysate and arrest the GroEL reaction cycle²⁶. The lysate was clarified by centrifugation at 16,000g (4 °C for 10 min). Either 20 μl of a non-specific antibody (against α-lactalbumin) or a GroEL-specific antibody was coupled to 100 μl of recombinant protein A Sepharose 4B beads (Thermo Fisher Scientific) as described by the manufacturer. The beads were loaded with sample (180 μg of protein) and incubated in 650 μl of HMK buffer for 1 h. The beads were washed twice with 600 μl of HMK buffer and then twice more with HMK containing 0.1% Triton X-100. For immunoblotting, elution was performed with 50 μl of 2× lithium dodecyl sulfate (Pierce) containing β-mercaptoethanol 5% (v/v) as prescribed by the manufacturer. For liquid chromatography–tandem mass spectrometry analysis, elution and digestion were performed with the IST MS sample preparation kit (Preomics) using the manufacturer’s on-bead digestion protocol. Mass spectrometry was performed as described above.

SDS–PAGE and immunoblotting

Before SDS–polyacrylamide gel electrophoresis (SDS–PAGE) analysis, cells were resuspended in HMK buffer supplemented with 2 mM DTT, 1 mM EDTA and 5% glycerol and subsequently sonicated, followed by centrifugation (20 min, 16,000g at 4 °C). Protein samples were separated by electrophoresis on NuPAGE 10% Bis-Tris SDS gels (Invitrogen) using NuPAGE MES SDS running buffer (Invitrogen) at 150 V. Proteins were transferred to polyvinylidene difluoride membranes in blotting buffer (25 mM Tris, 192 mM glycine, 20% methanol) at 150 mA. Membranes were first incubated with primary antibodies in TBST buffer overnight at 4 °C and subsequently with horseradish peroxidase-conjugated secondary antibody for chemiluminescence detection. Uncropped immunoblots are provided in the Source Data file to Extended Data Fig. 3.

In situ cryo-ET analysis

Cell cultures were grown as described above. For cryo-ET analysis, cells in exponential growth (approximate OD₆₀₀ 0.4) were rapidly (for about 2 min) concentrated to an approximate OD₆₀₀ of 10 by centrifugation at 8,000g and subsequently applied to R 2/1 100 Holey carbon film Cu 200 mesh grids (Quantifoil) that were previously plasma cleaned for 30 s. The sample was blotted for 9 s at force 10 and then plunge-frozen in a mixture of liquid ethane and propane cooled by liquid nitrogen using a Vitrobot Mark IV (Thermo Fisher Scientific) at 70% humidity and 22 °C. Frozen grids were transferred to a dual-beam, cryo-focused ion beam (FIB)/scanning electron microscope (Thermo Fisher Scientific; either Scios, Quanta, Aquilos or Aquilos 2). Cells were coated with a layer of inorganic platinum, if available in the system used, followed by the deposition of organometallic platinum using an in situ gas injection system (working distance, 10 mm; heating, 27 °C; time, 8 s). Removal of bulk material was done at a stage angle of 20–25° using gallium ions at 30 kV, 0.5 nA. Fine milling of lamellae was done at 11–13° stage tilt with successively lower currents between 0.3 nA and 30 pA, aiming for a final thickness of 100–200 nm (ref. ⁵²). Lamellae for the selective GroEL overexpression dataset were prepared using Serial FIB⁵³, and an additional layer of inorganic platinum was added following fine milling to avoid charging during image acquisition⁵⁴. The resulting lamellae were transferred to a TEM (Titan Krios, field emission gun 300 kV, Thermo Fisher Scientific) equipped with an energy filter (Quantum K2, Gatan), a direct detection camera (K2 Summit, Gatan), and tomograms were acquired at a magnification of ×42,000 (pixel size 3.52 Å), defocus ranging from −5.0 to −3.0 μm and the energy filter slit set to 20 eV using SerialEM 3.9.0 (ref. ⁵⁵). Tomograms were recorded in dose-fractionated super-resolution mode, with a total dose of roughly 120 e^−/Ų per tilt series. A dose-symmetric tilt scheme was used with an increment of 2–3° in a total range of ±60° from a starting angle of approximately 10° to compensate for lamellar pretilt (mostly around 11°)⁵⁶. Frames were aligned using MotionCor2 (v.1.4.0, https://emcore.ucsf.edu/ucsf-software)⁵⁷. The reconstruction was performed in IMOD using patch tracking (v.4.11.1, RRID:SCR_003297, https://bio3d.colorado.edu/imod/)⁵⁸ using the TOMOgram MANager (TOMOMAN) wrapper scripts⁵⁹. Tilt-series images were dose filtered using TomoMAN’s implementation of the Grant and Grigorieff exposure filter⁶⁰. Defocus was estimated using CTFFIND4 (ref. ⁶¹).

Tomograms of the EL⁺ dataset were acquired on a Krios G4 equipped with a Selectris X energy filter and Falcon 4 direct electron detector (Thermo Fisher Scientific). Tilt series were collected with a dose-symmetric tilt scheme using TEM Tomography 5 software (Thermo Fisher Scientific). A tilt span of ±60° was used with 2° steps, starting at ±10°, to compensate for lamellar pretilt. Target focus was changed for each tilt series in steps of 0.5 µm over a range of −2.5 µm to +5 µm. Data were acquired in EER mode of Falcon 4 with a calibrated physical pixel size of 3.02 Å and a total dose of 3e^−/Ų per tilt over ten frames. A 10 eV slit was used for the entire data collection. Data were preprocessed using TOMOMAN⁵⁹. EER images were motion corrected using RELION’s implementation of MotionCor2 (ref. ⁶²). Defocus was estimated using CTFFIND4 (ref. ⁶¹). Reconstruction was performed with IMOD using local deposits of the inorganic platinum that was applied by sputtering following milling as fiducials. All tomograms were reconstructed using NovaCTF⁶³.

E. coli membranes were segmented for visualization using TomoSegMemTV 1.0.

Cryo-ET analysis of in vitro reconstituted GroEL–GroES complexes

For generation of a GroEL–GroES reference for in situ tomographic analysis containing a defined substrate protein in a folded state and in a known topology, we imaged in vitro reconstituted GroEL–GroES–MetK complexes using the same data collection strategy and parameters as above for WT cells.

Subtomogram averaging

For subtomogram averaging, all datasets acquired on the same microscope (37 °C, HS, MetK) were combined and processed together; the EL⁺ dataset was processed separately. The overall processing workflow is depicted in Extended Data Fig. 1b.

For template matching, PDB entry 1AON was used for EL–ES₁, 4PKO for EL–ES₂ and 5MDZ for 70S ribosomes to generate templates at a resolution of 40 Å using the molmap⁶⁴ command in Chimera⁶⁵. Initial positions for a subset of EL–ES₁ and EL–ES₂ complexes and ribosomes were determined using the noise correlation template-matching approach implemented in STOPGAP, by fourfold binning to a pixel size of 14.08 Å (ref. ⁶⁶). This subset of the data was subsequently aligned and classified in STOPGAP to generate a reference from the tomographic data with a Fourier shell correlation (FSC) value close to 1 at 40 Å template-matching resolution. Template matching with various GroEL₁₄ species was attempted, but never yielded an average of GroEL₁₄ with a resolution better than the template resolution. The data-derived references of all three different structures were used for an additional round of template matching on the complete dataset. Cross-correlation cut-off was chosen separately for every tomogram by visual inspection of the generated hits and comparison with the tomogram. To reduce the level of false-positive detection, a mask for the cytosol of the cell was first created using AMIRA (Thermo Fisher Scientific) and subsequently used to filter out hits outside of the cytosol. Putative particles were deliberately overpicked with low-resolution templates in the initial stage to avoid false-negative assignments.

This procedure yielded 176,408 initial subtomograms for the EL–ES₁ reference and 125,860 for the EL–ES₂ reference. These were then further aligned and classified separately in STOPGAP, each yielding classes containing both EL–ES₁ and EL–ES₂ particles. The combined number of particles contained in classes with emergent high-resolution features (Supplementary Fig. 1a) for the EL–ES₁ reference was 19,239, and 17,614 for the EL–ES₂ reference (Extended Data Fig. 1 and Supplementary Fig. 1b). Because both references pick up a subset of the other particles, the particles were then combined and duplicates removed. The resulting combined dataset was split by reference-free, three-dimensional classification in STOPGAP, resulting in a set of 17,598 EL–ES₁ and 11,213 EL–ES₂ complexes that were then independently refined. This resulted in a resolution at the FSC cut-off of 0.143 following the application of symmetry at 11.6 Å for the EL–ES₁ complex (C7 symmetry) and 11.9 Å for the EL–ES₂ complex (D7 symmetry). Classification was performed using simulated annealing stochastic hill-climbing multireference alignment as previously described⁶⁷. All classifications were done repeatedly with different, random initial starting sets of 250–500 subtomograms to generate the initial references. Only particles that ended up in the same class for all independent rounds of classifications were retained⁶⁷. Further refinements with the established WARP, RELION, M pipeline were attempted but did not yield any further improvements. EL–ES₁ wide and narrow complexes were separated by classification with a focused, disk-shaped mask on the apical domains of the EL–ES₁ trans-ring. This resulted in 6,681 narrow complexes that were refined to a resolution of 13.5 Å, and 10,130 wide EL–ES₁ complexes refined to a resolution of 12.0 Å.

The EL⁺ dataset was processed in the same way, but starting with the structures from the other datasets, low-pass filtered to 40 Å, as initial references for template matching. Template matching was then repeated once with structures generated by averaging a subset of particles from this dataset. To improve the resolution for model building, the dataset was exported to WARP⁶⁸ and angles and positions refined using RELION v.3.0.8 (ref. ⁶⁹). This yielded a GroEL₁₄ structure at a global resolution of 13 Å. GroEL 14-mer particles were corefined for geometric distortions with ribosomes in M. The resulting GroEL 14-mer particles were exported for further alignment and classification in RELION. Classification was performed with a regularization parameter T of four and six classes for 25 iterations without angular search, resulting in a more homogeneous subset of 12,421 particles. These particles were again corefined in M for geometric distortions and per-particle defocus for contrast transfer function (CTF) estimation, resulting in a final structure with nominal resolution of 9.8 Å at 0.143 FSC cut-off.

Owing to their high molecular weight and density, ribosome template matching achieves a higher precision and recall. During initial rounds of classification in STOPGAP, because no false-positive particles were detected, all ribosomal hits from template matching were aligned first in STOPGAP at progressively lower binnings (bin4, bin2, bin1). The resulting particles were then exported to WARP using TOMOMAN. Subtomograms were reconstructed for RELION v.3.0.8 using WARP at a pixel size of 3.52 Å per pixel. An iterative approach with subtomogram alignment in RELION and tilt-series refinement in M⁷⁰ were performed until no further improvement in gold-standard FSC was obtained. This resulted in a final structure of the ribosome at a resolution of 8.6 Å for the combined 37 °C, HS and MetK datasets, and 6.3 Å for the EL⁺ dataset, which was processed separately.

In vitro cryo-ET data for GroEL–GroES complexes were processed analogous to the in situ data, resulting in 39,518 initial hits for the EL–ES₂ template and 46,093 for the EL–ES₁ template, with both sets having a significant overlap. These were then further aligned and classified separately in STOPGAP, yielding 5,832 and 13,688 particles, respectively, following duplicate removal.

Classification of SP occupancy of GroEL–GroES complexes in situ

For the resolution of densities corresponding to substrate proteins in the GroEL–GroES chamber we first performed symmetry expansion around the C2 axis of the EL–ES₂ complexes and aligned the new set of GroEL–GroES chambers with the cis-ring of the EL–ES₁ complexes. The resulting subtomograms of the chambers were then denoised using TOPAZ’s three-dimensional pretrained denoising function⁷¹. Because initial attempts to classify the interior of the chamber using STOPGAP multireference-based alignment showed only separation by missing wedge, the subtomograms were combined into 5,000 random bootstraps containing 250 random subtomograms each. These averages were then used to perform k-means clustering with two classes. Bootstraps from the resulting clusters were averaged and used as initial start structures for multireference alignment in STOPGAP. For this, stochastic hill climbing was performed with a temperature factor of 10 for simulated annealing, followed by 40 iterations of multireference alignment with two classes and a mask around the interior of the chamber. This process was repeated five times. Only particles consistently assigned to the same classes were used for a final round of subtomogram averaging, resulting in one class showing weak diffuse density inside the chamber and a second showing strong density near the bottom. Attempts to further subdivide these two classes resulted only in separation based on missing wedge. Because it was not possible to resolve the C7 symmetry mismatch of the substrate and enclosing chamber, final averages were produced for all different biological conditions with C7 symmetry applied to increase the signal-to-noise ratio. The class showing a strong density near the bottom contained 12,255 subtomograms, the one showing only a weak diffuse density with 24,435 subtomograms for the combined 37 °C, HS and MetK datasets.

In vitro data were processed analogously. The resulting classes were then again split into EL–ES₁ and EL–ES₂ complexes corresponding to their substrate state and exported to WARP⁶⁸. An additional round of alignments was performed in RELION for all different classes and complexes. A prior was set for all angles. Local search was performed with a sigma of 0.5 and search angle of 0.9°. The resulting particles were separately refined in M, correcting for geometrical distortions. Particles were again exported from M⁷⁰ and signal subtraction preformed in RELION of the trans-ring for EL–ES₁ and the opposing chambers for EL–ES₂. Based on their previous classification results in STOPGAP, the refined signal-subtracted, single-chamber complexes were combined in two groups resulting in 7,087 GroEL–GroES chambers containing an ordered SP and 14,371 that either contained a disordered SP or were empty. The resulting chambers were again locally refined in RELION using priors and a sigma on all angles, yielding a resolution of 9.4 Å for GroEL–GroES chambers containing ordered SP and 8.8 Å for the remaining chambers.

Cryo-EM single-particle analysis of GroEL–GroES–MetK complexes

For generation of substrate-bound GroEL–GroES complexes, 4 μM MetK was denatured in the presence of 1 μM GroEL (14-mer) in buffer A (20 mM MOPS-NaOH pH 7.4, 200 mM KCl, 10 mM MgCl₂, 5 mM DTT) containing 30 mM NaF and 5 mM BeSO₄ by first incubation of the mixture at 60 °C for 15 min and then cooling to 25 °C in a thermomixer (Eppendorf). The addition of 2 μM GroES (7-mer) and 1 mM ATP (pH 7.0) resulted in stable chaperonin complexes with encapsulated MetK⁴⁰. Biochemical analysis of this preparation was performed by size exclusion chromatography on a Superdex 200 3.2/300 GL column. Fractions were analysed by SDS–PAGE electrophoresis (NuPAGE, Bis-Tris 4–12% gels), and MetK loading of GroEL–GroES complexes was estimated by mass spectrometry using intensity-based absolute quantification values⁷². For analysis by mass spectrometry, fractions F1 and F2 (Extended Data Fig. 5a) were analysed separately but intensities pooled for the determination of intensity-based absolute quantification ratios.

GroEL–GroES–MetK samples were concentrated tenfold by ultrafiltration using a 100 kDa Amicon centrifugal concentrator (Millipore) at room temperature. As a control, GroEL and GroES were treated identically in the absence of MetK. Before freezing, 1 μl of a n-octyl-β-d-glucopyranoside stock solution (87.5 mg ml^−1 in buffer A) was added per 50 μl of sample. For single-particle analysis and in vitro cryo-ET experiments, 4 μl of the sample was applied onto R 2/1 100 Holey carbon film Cu 200 mesh grids (Quantifoil) previously plasma cleaned for 30 s. This grid was blotted for 3.5 s at force 4 and plunge-frozen in a mixture of liquid ethane and propane cooled by liquid nitrogen using a Vitrobot Mark IV (Thermo Fisher Scientific) at 100% humidity and 4 °C.

Cryo-EM data for the EL–ES–MetK dataset were acquired using a FEI Titan Krios transmission electron microscope and SerialEM software⁵⁵. Video frames were recorded at a nominal magnification of ×22,500 using a K3 direct electron detector (Gatan), with a total electron dose of around 55 electrons per Ų distributed over 30 frames at a calibrated physical pixel size of 1.09 Å. Micrographs were recorded within a defocus range of −0.5 to −3.0 μm.

On-the-fly image processing and CTF refinement of cryo-EM micrographs were carried out using the Focus software package⁷³. Only micrographs that met the selection criteria (ice thickness under 1.05, drift 0.4 Å < x < 70 Å, refined defocus 0.5 μm < x < 5.5 μm, estimated CTF resolution under 6 Å) were retained. Micrograph frames were aligned using MotionCor2 (ref. ⁵⁷), and the CTF for aligned frames was determined using GCTF⁷⁴.

The control dataset of GroEL–GroES complexes without MetK was acquired similarly but with a nominal magnification of ×29,000, resulting in a calibrated pixel size of 0.84 Å.

Image processing, classification and refinement for single-particle analysis

From the resulting 8,945 micrographs of the GroEL–GroES–MetK dataset, 1,561,482 particles were picked using a trained crYOLO network⁷⁵ and extracted with RELION v.3.1.3 (ref. ⁶⁹). An initial round of two-dimensional classification was performed and the remaining particles were passed into CryoSPARC⁷⁶ for further two-dimensional classification, ab initio model building, alignment and initial three-dimensional classification to separate EL–ES₁ from EL–ES₂ complexes. The remaining EL–ES₂ (659,866 particles) and EL–ES₁ (294,250 particles) complexes were then exported separately to RELION for additional alignment with imposed symmetry, CTF refinement and Bayesian polishing. For the EL–ES₂ complexes, symmetry expansion around the C2 axis was performed and the opposing half removed using RELION’s signal subtraction.

The resulting asymmetric EL–ES₁ complexes were then classified further with CryoDRGN⁷⁷, resulting in a clean subset of 242,276 particles. The trans-rings of the EL–ES₁ complexes were classified in CryoSPARC using a focused mask on the apical domains of the trans-ring, resulting in 169,454 particles in the narrow conformation and 34,755 in the wide conformation. The resulting structures were refined in CryoSPARC under the application of C7 symmetry to a nominal resolution of 2.9 and 3.1 Å, respectively. For the analysis of the cis-chamber, all EL–ES₁ particles were pooled and the trans-ring was removed by signal subtraction in RELION.

The resulting GroES-bound, single-ring particles (1,562,002 particles) were then aligned to a common reference in RELION and exported to CryoSPARC for further alignment without imposed symmetry. The resulting mask and reference were reimported into RELION and used for an additional alignment step with the goal of aligning the asymmetric MetK substrate contained inside the chamber (Extended Data Fig. 5d). Subsequently a second round of signal subtraction was performed and the resulting particles, comprising only MetK density, were further subjected to three-dimensional classification without angular search in RELION. A subset of the resulting classes showed visible secondary structure elements in different orientations (Extended Data Fig. 5d). These classes were then combined and aligned into a single frame of reference in Matlab 2015b by manual rotation with the respective multiple of 360°/7 around the sevenfold symmetry axis. This was done by adding the corresponding increment to particle rotation angles in the particle table (.star file).

These folded MetK (fMetK) particles were then further locally aligned in CryoSPARC. An additional round of three-dimensional classification was performed followed by a final round of local alignment (322,800 particles), resulting in density for MetK at a resolution of 3.7 Å.

For the study of MetK contacts with the inner wall of GroEL–GroES chamber, we reverted the signal subtraction in RELION to generate single-ring GroEL–GroES–MetK particles for both the folded MetK and mixed population of chambers either containing disordered MetK or empty; both were refined and aligned in CryoSPARC. The subset containing a mixed population was additionally classified in CryoDRGN between the final alignment steps, resulting in a global resolution of 3.04 Å for the GroEL–GroES–MetK complex containing folded MetK and of 2.94 Å for the complex containing a mixed population of disordered MetK or empty chambers.

GroEL–GroES complexes without MetK were processed analogously but without Bayesian polishing and CTF refinement in RELION. Signal subtraction was performed in CryoSPARC; using 293,974 particles, this resulted in a map with a global resolution of 2.5 Å following the application of C7 symmetry.

Densities were visualized and rendered using ChimeraX^78,79.

Model building and refinement

Model building was initiated by rigid-body fitting the GroEL subdomains, GroES and MetK from the crystal structures PDB 1SX3 (ref. ⁸⁰), 5OPW¹² and 7LOO⁴², respectively, into cryo-EM density, followed by manual editing using Coot⁸¹. The models were subsequently refined in real space with Phenix⁸². For the refinement of models against low-resolution data from STA, automatically generated restraints from reference structures such as PDB 8P4M (this study) were used. Residues with disordered sidechains were truncated at C-beta.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.