Cryo-EM sample screening at the EMBL Imaging Centre

To produce and identify a cryo-EM grid suitable for high-end data collection on a Titan Krios microscope, a total of 19 grids were vitrified over three freezing sessions and screened in two sessions using a Glacios (ThermoScientific) transmission electron microscope. At first, the screening was aimed at optimising sample spreading on the grid and ice thickness. Here, two different vitrification devices (ThermoScientific Vitrobot Mark IV and Leica EM GP2) and multiple different grid types (holey carbon grids with and without an additional layer of continuous carbon) were employed. Grid quality assessment was based on the low magnification grid overview which was acquired for each grid (commonly termed “grid map” or “grid atlas”) and high magnification screening images from those grids which showed promising ice thickness and had sufficient imageable areas for potential high-resolution data collection (Fig. 1 and Supplementary Fig. ^{1a, b}).

Open in a separate window

Fig. 1

Initial Cryo-EM sample screening to select the grid for high-resolution data collection.

a Cryo-EM micrographs of the ribosome particles applied to holey carbon grids with (plus carbon, green) and without (no carbon, blue) an additional thin layer of continuous carbon (2nm). One micrograph out of 919 (plus carbon) or 1082 (no carbon) is shown. b Sample screening processing scheme using cryo-EM data collected overnight (o/n) on a ThermoScientific Glacios electron microscope. Colours as in “a”. c Fourier Shell Correlation (FSC) plots for the data collected on holey grids with and without an additional layer of continuous carbon. d Orientation distribution of the particles from data collected on plus and no carbon grids.

During the second stage of screening, the selected grids were assessed for ribosome particle density and distribution. A small overnight data set (~1000 movies) was then collected from the most promising grid at the end of each Glacios screening session. The two selected grids were a Quantifoil R2/1 grid with an additional 2nm layer of continuous carbon vitrified with the Vitrobot (“plus carbon” grid) and a Quantifoil R2/1 grid without the additional layer of continuous carbon vitrified with the EM GP2 using back side blotting to increase particle density in the foil holes (“no carbon” grid) (Fig. 1a and Supplementary Fig. ^{1b, c}). Each dataset was analysed by employing the same processing workflow yielding 3D reconstructions of the 70S ribosome at ~3Å resolution from approximately 100,000 particles (Fig. 1b, c and Supplementary Fig. ^1d). Despite the similar nominal resolutions obtained, the ‘plus carbon’ dataset had a substantially better particle orientation distribution compared to the “no carbon” dataset which suffered from severe preferred orientation (Fig. 1d). Therefore, we selected the ‘plus carbon’ grid for further high-end data collection on a Titan Krios G4 (ThermoScientific) equipped with a cold-FEG electron source, Selectris X energy filter and Falcon4 direct electron detector; over a 3-day session, a total of ~20,000 movies were collected.

High-resolution structure of P-site bound translating ribosome

We chose to process particles using a combination of both cryoSPARC¹⁸ and RELION¹⁹ software. Particle picking was done semi-automatically with crYOLO²⁰ using the pre-trained general model, then followed by standard 2D classification and ab initio model generation which suggested the sample contained both translating ribosomes and free 50S subunit particles. Heterogeneous refinements in cryoSPARC were then employed to remove particles that belong to the free large ribosomal subunit and resulted in ~750,000 70S ribosome particles. Further 3D classification resolved a class containing ~650,000 particles of a translating ribosome containing P-site tRNA and a weaker density for the E-site tRNA with the small 30S subunit in the unrotated state (Supplementary Fig. ²–⁴). This is a widely observed state for translating ribosomes isolated using in vitro translation systems^21–23. In this map, the P-site tRNA is tethered to a nascent chain (NC) density with weak density traceable to the backbone of the alpha carbons (Supplementary Fig. ⁵). However, clear densities of side chains were not observed in the polypeptide exit tunnel due to averaging of several states during translation and thus we did not fit an atomic model in this area.

A 3D variability analysis focused on the tRNA sites identified two subclasses with P- and E-site (~500,000 particles) and A-, P- and E-site (~150,000 particles) tRNA densities, respectively. A combination of 3D refinement schemes, contrast transfer function (CTF) refinement schemes as well as particle polishing¹⁹ were then used to obtain a final reconstruction to 1.55Å overall resolution of the larger class containing EM-densities corresponding to P- and E-site tRNAs (Supplementary Fig. ² and ⁴, Supplementary Table ¹). The local resolution of the small subunit was slightly lower (~1.9Å) and was further improved to ~1.7Å after performing focused 3D refinements and signal subtraction of the large subunit (Supplementary Fig. ²).

In the obtained structure, we can visualise rRNA modifications and protein side-chain conformers (Fig. 2) similar to those observed in a previously reported structure of the bacterial ribosome¹⁵ bound to mRNA and tRNA at 2.0Å. At the current resolution, we can observe separate EM-densities of water molecules coordinating magnesium ions bound to the RNA backbone (Fig. 2b). In addition, we modelled EM-densities for potassium ions (Fig. 2b), based on their identified positions in a recent X-ray structure²⁴. Although the state of the ribosome is paused at different states during translation, we visualised the mRNA with defined anticodon-codon interactions being visible (Supplementary Fig. ⁶). It is thus likely that the ribosome is paused at preferred sites due to visible densities of purine and pyrimidine rings. However, a reliable assignment of the pausing site could not be assigned and is not further discussed here.

Open in a separate window

Fig. 2

1.55Å resolution reconstruction of the 70S ribosome.

a Cryo-EM map of the bacterial translating ribosome at 1.55Å coloured in dark yellow (16S rRNA), light yellow (ribosomal proteins in the small subunit), light blue (large subunit ribosomal proteins), dark blue (23S and 5S rRNA). b, c Closeup densities of protein side chain conformation, water molecules, magnesium, and rRNA modifications are shown as sticks with overlaid EM-densities shown as mesh. Density thresholds: 0.9 (Mg²⁺, b) and (K⁺, b), 0.5 (Protein S5, c), 0.65 (Protein L3, c) and 0.8 (23S rRNA modification).

A C-G base-pair swap in the 23S rRNA observed in the high-resolution map

Interestingly, we observed a swap between the E. coli MRE600 strain sequence of the 23S rRNA docked into our cryo-EM map. The initially identified swap was a G-C base-pair swap in the position G2210 and C2216 (MRE600 numbering) in helix 79 of the 23S rRNA. After closer inspection, we realised that the PURE in vitro translation system kit was developed based on the E. coli B-strain instead^25,26, of which the sequence fits well into the EM-density of our maps. Through sequence alignments of the seven different 23S rRNA operons of the E. coli B-strain and comparing their fit to our EM density we could identify loci C2566_01480 and C2566_04790 as the predominant ones in our 70S ribosome sample. There are two additional mutations within the same rRNA helix which fit to our density (Fig. 3a–c). All other differences between the MRE600 and B-strain rRNA sequences (mutations and insertions), are also reflected by our EM map. Sequence polymorphisms within the 23S and 16S rRNA is a common property of bacterial strains and environmental isolates of microbes which could be used as a possible indicator of strain virulence^27,28. The C-G swap identified here is an intriguing example of co-evolution where two separate mutations in the 23S rRNA sequence were attained to maintain the base-pairing and the integrity in the 23S rRNA helix.

Open in a separate window

Fig. 3

Direct sequencing of rRNA polymorphism resolved by the high-resolution 70S structure.

a Partial sequence alignment of H79 of the 23S rRNA between the E. coli B-strain and MRE600. Base-swap positioning is indicated. Other observed mutations are indicated by an asterisk. b Snapshot of H79 shown as cartoon illustration with base colouring as in “a”. Overlaid cryo-EM density is shown as transparent isosurface. c Close-up of the base pair swap using B-strain sequence and MRE600 sequence (PDB 7K00) with overlaid EM density shown as mesh.

Density corresponding to the ribosome protein bL9

In the current map, the spread of local resolution at the periphery is around 2–3Å (Supplementary Fig. ^7a), which underscores the stability of our sample preparation and data processing schemes. We observe a well-resolved density at side-chain resolution of the ribosome proteins bL9 and uS6 (Fig. 4a–c, and Supplementary Fig. ^{7b, c}). bL9 is located at a peripheral position of the ribosome forming a strut-like shape below the L1 stalk and bridging the large and the small subunits of the ribosome. The C-terminal domain of bL9 is highly mobile and was observed in an open conformation in previous X-ray structures²⁹ and later observed in the closed conformation in high-resolution cryo-EM structures³⁰ contacting the small ribosomal subunit. In the current structure, bL9 is observed in its entirety with the C-terminal domain establishing contacts with uS6 where it locks the L1 stalk in the closed conformation contacting the E-site tRNA (Fig. 4a). The closed conformation of this protein is stabilised via a salt bridge between Arg24 (uS6) and Glu87 (bL9) (Fig. 4c). Notably, the strut-like shape of bL9 is rich in negatively charged residues through its extended structure (Supplementary Fig. ^7d), which would otherwise repel away from the ribosome. Furthermore, a mutant in the C-terminal domain (S93F) affects forward slippage on mRNA, suggesting a role for L9 in fidelity of translation³¹. We speculate that this unusual feature of a ribosomal protein carries an important function during protein synthesis that could also contribute to its detachment and flexibility in previous X-ray and cryo-EM structures that were done under variable salt conditions.

Open in a separate window

Fig. 4

Full-length bL9 in the closed conformation resolved by the high-resolution 70S structure.

a bL9 (cyan) and uS6 (yellow) highlighted in the context of the cryo-EM map of the 70S ribosome. Large subunit is in light grey, small subunit in dark grey. L1 stalk of the ribosome is indicated. b, c Closeups of bL9 density with surrounding interactions with the small subunit. Ribosomal protein uS6 is indicated. Overlaid EM density is shown as a transparent surface in (b). Colouring as in (a).

The ribosomal protein bL9 plays an important role in influencing reading frame maintenance and restraining forward slippage on mRNA. This restraint can be circumvented by recoding signals that cause programmed frameshifting that are availed of by many viruses and mobile elements^31–33. Recently, L9 interactions across neighbouring ribosomes translating the same mRNA were also shown to play a role in polysome formation³⁴ and ribosome quality control pathways bridging interaction between two colliding ribosomes^35,36. In addition, L9 interacts with elongation factor P (EF-P)³⁷ as well as the ribosome biogenesis GTPases EngA³⁸, extending the current known function of this protein. High-resolution of bL9 is thus highly pertinent to determining the outstanding functional aspects of this enigmatic dual-mode protein.

Cryo-EM sample preparation

Cryo-EM grids have been prepared with a Vitrobot Mark IV (plus carbon sample, ThermoScientific) or an EM GP2 (no carbon sample, Leica) plunger. Environmental chambers were set to 6°C and 100% humidity. Grids were rendered hydrophilic by plasma cleaning for 30s in a 90/10% Ar/O mixture with a Fischione 1070 plasma cleaner right before plunge freezing. 2.5µL of the undiluted ribosome sample were applied to the foil side of the grid. The plus carbon sample (Quantifoil Cu300 R2/1+2nmC, as commercially available) was blotted for 2s with blot force +3 after a 30s wait time after sample application and immediately plunge frozen in liquid ethane. The no carbon sample (Quantifoil Cu200 R2/1) was blotted for 1.5s from the back side right after sample application and immediately plunge frozen in liquid ethane. In all cases Whatman 597 blotting paper was used.

Cryo-EM data collection

Grid screening and screening data acquisition was performed on a Glacios TEM operated at 200kV equipped with a Selectris X energy filter and Falcon 4 direct electron detector (ThermoScientific). Data was acquired at a nominal magnification of 100 kx resulting in a calibrated pixel size of 1.154Å on the camera. A 50µm C2 and a 100µm objective aperture were inserted and the width of the energy filter slit was set to 10eV. The TEM was operated in nanoprobe mode at spot size 6. Movies were acquired over 8.3s exposure time with a total accumulated dose of ~40 e^-/Ų (~6.4 e^-/px/s over an empty area on the camera level) in counting mode and saved in the EER file format. For the plus carbon data set, 919 movies were acquired in one overnight data collection (1 exposure per hole/stage movement) and for the no carbon data set 1,082 movies were acquired in the same way.

After evaluation of the two screening data sets (see below), the plus carbon grid was selected for high-resolution data acquisition on a Titan Krios TEM operated at 300kV equipped with a C-FEG, Selectris X energy filter and Falcon 4 direct electron detector (ThermoScientific). A total of 19,449 movies were collected over three days at a nominal magnification of 165 kx resulting in a calibrated pixel size of 0.731Å on the camera. A 50µm C2 and a 100µm objective aperture were inserted and the slit width of the energy filter was set to 10eV. The TEM was operated in nanoprobe mode at spot size 5 and a beam diameter of ~460nm. Movies were acquired over a 5.2s exposure time with a total accumulated dose of ~40 e^-/Ų (~4.1 e^-/px/s over an empty area on the camera level) in counting mode and saved in the EER file format (1293 fractions per movie). The target defocus was set to −0.5 to −1.5µm with 0.1µm steps between holes. In total, 16 movies were acquired per hole using beam image shift (11 in the outer ring, 5 in the inner ring) before moving to the next hole by stage shift. All TEM screening and data acquisition was performed using SerialEM⁴⁶.

Data processing and model building

Cryo-EM data processing workflows for the Glacios screening data sets and the Krios high-resolution data set are shown in Fig. 1 and Supplementary Fig. ¹, ² and ³, respectively. Motion correction of EER movies was performed with the CPU implementation of MotionCor2 from Relion4⁴⁷. Initial CTF estimation was performed using CTFFIND4⁴⁸. Gautomatch (https://www2.mrc-lmb.cam.ac.uk/download/gautomatch-056/) was used for particle picking of the Glacios screening data, crYOLO²⁰ with the pre-trained general model was used for picking of the high-resolution Krios data set. Particle extraction, 3D classification with alignment as well as Bayesian polishing were performed with Relion4. All other cryoEM data processing steps were performed in cryoSPARC¹⁸. Separation of 50S and 70S particles was achieved via Heterogeneous Refinement, different tRNA states were identified using a combination of 3D classification and 3D Variability analysis. Maps were post-processed by local filtering after local resolution estimation as implemented in cryoSPARC. Conversion of cryoSPARC ‘.cs’ files to Relion ‘.star’ files was performed using UCSF pyem⁴⁹ in combination with in-house written bash scripts⁵⁰ (https://github.com/simonfromm/miscEM). All final Homogeneous Refinements were done with per-particle defocus and per-group CTF parameter optimization as well as Ewald Sphere Correction switched on. Exposures from all beam image shift positions of the acquisition pattern for the dataset collected on the Titan Krios were put into a separate group resulting in a total of 16 exposure/optics groups.

Model building was started with individually rigid-body fitting all chains from the published 2Å E. coli 70S ribosome structure (PDB 7K00) in COOT⁵¹. After manual inspection of the rRNA of the large and small ribosomal subunits for mutations, the residue numbering of the E. coli B-strain was used. All chains were inspected and adjusted manually in COOT and subsequently refined with real-space refinement in PHENIX⁵² against the unsharpened map with secondary structure and Ramachandran restraints switched off (Supplementary Table ¹). The C-terminal part of bL9 absent from PDB 7K00 was built de novo based on the AlphaFold2 model. Water molecules were added to the 30S and 50S part of the structure separately using “Find Waters” in COOT. Mg²⁺ and K⁺ positions were inferred from PDB 7K00 and 6QNR, respectively. All Mg²⁺ and K⁺ ion positions were checked and modified manually when necessary. The model geometry was validated using MolProbity⁵³. The model vs. map FSC at FSC=0.5 coincides well with the one determined between the map half-sets at FSC=0.143 (Supplementary Fig. ⁴). All cryoEM density and model renderings were generated with ChimeraX⁵⁴.

We would like to thank members of the Mattei, Jomaa and Atkins labs for discussions. We are grateful to Dr. Jochen Zimmer for his comments on the manuscript. We also thank Maciej Gluc for discussions. This work was funded by the European Molecular Biology Laboratory to S.M., by the Department of Molecular Physiology and Biological Physics and the School of Medicine at the University of Virginia through a start-up grant to A.J., and by the Irish Research Council Advanced Laureate (IRCLA/2019/74) to J.F.A. This work benefited from access to EMBL Imaging Centre and has been supported by iNEXT-Discovery, project number 871037, funded by the Horizon 2020 program of the European Commission. We acknowledge the access and services provided by the Imaging Centre at the European Molecular Biology Laboratory (EMBL IC), generously supported by the Boehringer Ingelheim Foundation.