The shape of the bacterial ribosome exit tunnel affects cotranslational protein folding

The E. coli ribosome exit tunnel can accommodate small folded proteins, while larger ones fold outside. It remains unclear, however, to what extent the geometry of the tunnel influences protein folding. Here, using E. coli ribosomes with deletions in loops in proteins uL23 and uL24 that protrude into the tunnel, we investigate how tunnel geometry determines where proteins of different sizes fold. We find that a 29-residue zinc-finger domain normally folding close to the uL23 loop folds deeper in the tunnel in uL23 Δloop ribosomes, while two ~ 100 residue proteins normally folding close to the uL24 loop near the tunnel exit port fold at deeper locations in uL24 Δloop ribosomes, in good agreement with results obtained by coarse-grained molecular dynamics simulations. This supports the idea that cotranslational folding commences once a protein domain reaches a location in the exit tunnel where there is sufficient space to house the folded structure.


Introduction
A large fraction of cellular proteins likely start to fold cotranslationally in the~100 Å long exit tunnel in the ribosomal large subunit (Milligan and Unwin, 1986;Ban et al., 2000;Nissen et al., 2000), before they emerge into the cytosolic environment. In E. coli ribosomes, portions of the 23S rRNA and a few universally conserved proteins line the exit tunnel, Figure 1A. The tunnel proteins uL4, uL22, and uL23 consist of globular domains that are buried within the rRNA, and b-hairpin loops that protrude into the tunnel (Klein et al., 2004). These loops help stabilize the tertiary structure of 23S rRNA (Lawrence et al., 2016) and contribute towards the unique geometry of the tunnel Nissen et al., 2000;Harms et al., 2001). uL24 and uL29 are located near the end of the tunnel, and a hairpin loop in uL24 forms a finger-like structure that partially obstructs the tunnel exit port.
Inspired by observation that protein domains fold in different parts of the exit tunnel depending on their molecular weight (O'Brien et al., 2010;O'Brien et al., 2011;Trovato and O'Brien, 2016;Samelson et al., 2018;Farías-Rico et al., 2018), we now ask what role the geometry of the exit tunnel plays in determining where these domains fold. To explore this question, we employ the same arrest peptide-based approach (and coarse-grained MD simulations) used in our previous studies of cotranslational protein folding (Nilsson et al., 2015;Nilsson et al., 2017), but with ribosomes that carry deletions in either the uL23 or the uL24 hairpin loop. Our findings provide strong evidence that the tunnel geometry determines where in the tunnel a protein starts to fold.

Results and discussion
The folding assay Our experimental set-up, Figure 1B, exploits the ability of the SecM translational arrest peptide (AP) (Nakatogawa and Ito, 2001) to act as a force sensor (Ismail et al., 2015;Ismail et al., 2012;Goldman et al., 2015), making it possible to detect the folding of protein domains in the exit tunnel (Nilsson et al., 2015;Goldman et al., 2015). In brief, the domain to be studied is cloned, via a linker, to the AP, L residues away from its C-terminal proline. The AP is followed by a C-terminal tail, to ensure that arrested (A) nascent chains can be cleanly separated from full-length (FL) chains by SDS-PAGE. Constructs with different L are translated in the PURE in vitro translation system (Shimizu et al., 2005), and the fraction full-length protein (f FL ) is determined for each L. For linkers that, when stretched, are long enough to allow the protein to reach a part of the exit tunnel where it can fold, force will be exerted on the AP by the folding protein, reducing stalling and increasing f FL . (Tian et al., 2018), Figure 1C. A plot of f FL vs. L thus shows where in the exit tunnel a protein starts to fold and at which linker length folding no longer causes increased tension in the nascent chain.
A number of earlier studies have provided strong support for the notion that the dominant peak in a f FL profile corresponds to folding into the native state (as opposed to, e.g., non-specific compaction of the nascent chain): (i) folded proteins have been visualized in the exit tunnel by cryo-EM of ribosome-nascent chain complexes at L-values corresponding to the dominant f FL peak (Nilsson et al., 2015;Nilsson et al., 2017;Tian et al., 2018), (ii) the dominant f FL peak disappears when proteins that depend on metals or other ligands for folding are translated in the absence of the ligand (Farías-Rico et al., 2018;Nilsson et al., 2015), (iii) the dominant f FL peak corresponds closely to the tether length at which protein domains become resistant to on-ribosome pulse-proteolysis by thermolysin (Farías-Rico et al., 2018) or at which folding can be detected by other techniques such as NMR or FRET (Kemp et al., 2018), (iv) the amplitude of the f FL peak correlates with the folding free energy of a domain (Farías-Rico et al., 2018). uL23 Dloop and uL24 Dloop ribosomes The E. coli strains HDB143 (uL23 Dloop; uL23 residues 65-74 deleted) and HDB144 (uL24 Dloop; uL24 residues 43-57 deleted) have previously been shown to be viable (Peterson et al., 2010), as is a strain where uL23 has been replaced by a homologue from spinach chloroplast ribosomes that also lacks the b-hairpin loop (Bubunenko et al., 1994;Bieri et al., 2017). These strains were used to purify high-salt-washed ribosomes that were used to translate proteins in the commercially available PURExpress D-Ribosome kit. Analysis of the purified ribosomes by SDS-PAGE and western blotting demonstrated the expected size differences compared to wildtype for the uL23 Dloop and uL24 Dloop proteins, Figure 1-figure supplement 1.

Cryo-EM structure of uL23 Dloop ribosomes
The loop deleted in the uL24 Dloop ribosomes does not interact with neighboring parts of the ribosome, Figure 2A, and hence its removal would not be expected to alter the structure of other parts of the exit tunnel. In contrast, the loop deleted in uL23 Dloop ribosomes is located deep in the exit tunnel, Figure 2B,~40-50 Å from the exit and it is not clear a priori whether its removal may cause rearrangements in other tunnel components. For this reason, we determined a cryoEM structure of the uL23 Dloop 70S ribosome at an average resolution of 3.3 Å , Figure 2C, Figure 2-figure supplement 1, and found that the shape of the tunnel remains unchanged in the uL23 Dloop ribosome when compared with wildtype (WT) E. coli ribosomes, except for an increase in volume resulting from the absence of the uL23 loop Figure 2C-D. We estimated this increase using the POVME algorithm (Durrant et al., 2011;Durrant et al., 2014). Compared to WT E. coli ribosomes, the tunnel volume increases by 2,064 Å 3 in uL23 Dloop ribosomes, see Video 1, about 1/3 of the size of ADR1a (5,880 Å 3 ) calculated by the same method.  (Zhang et al., 2015), with tunnel proteins uL4 and uL22 indicated in gray. The globular domain of uL23 is indicated in orange with the b-hairpin loop depicted in yellow. uL24 is shown in dark blue, with the loop at the tunnel exit shown in light blue. The exit tunnel, outlined by a stalled SecM nascent chain (purple), is~100 Å in length. (B) The arrest-peptide assay (Nilsson et al., 2015). The domain to be studied is placed L residues upstream of the critical proline at the C-terminal end of the 17-residue long arrest peptide (AP) from the E. coli SecM protein. A 23-residue long stretch of the E. coli LepB protein is attached downstream of the AP, allowing us to separate the arrested (A) and full-length (FL) products by SDS-PAGE after translation. Constructs are translated in the PURExpress in vitro translation system supplemented with WT, uL23 Dloop, or uL24 Dloop high-salt washed ribosomes for 20 min. The relative amounts of arrested and full-length protein were estimated by quantification of SDS-PAGE gels, and the fraction of full-length protein was calculated as f FL = I FL /(I A +I FL ) where I A and I FL are the intensities of the bands corresponding to the A and FL products. (c) f FL is a proxy for the force F that cotranslational folding of a protein domain exerts on the AP. At short linker lengths, both F and f FL » 0 because the domain is unable to fold due to lack of space in the exit tunnel. At intermediate linker lengths, F and f FL > 0 because the domain pulls on the nascent chain as it folds. At longer linker lengths, F and f FL » 0 because the domain is already folded when the ribosome reaches the end of the AP. mM Zn 2+ (to promote folding of ADR1a) or in the presence of 50 mM of the zinc-specific chelating agent TPEN (to prevent folding of ADR1a; TPEN is required to remove residual amounts of Zn 2+ from the PURE lysate) ( Figure 3A, Figure 3-figure supplements 1-4). Translation rates in PURE are~10 fold slower than in vivo (Capece et al., 2015), but since the proteins studied here fold on micro-to-millisecond time scales, that is considerably faster than the in vivo translation rate, it is safe to assume that the folding reaction has time to equilibrate between each translation step both in vivo and in the PURE system. Similar to previous results (Nilsson et al., 2015), we saw efficient stalling when ADR1a was translated in the presence of TPEN at L ! 19 residues, Figure 3-figure supplement 8. Further, there is a slight but significant increase in f FL at L = 17 residues in the presence of TPEN (hence not related to folding); this has been observed before (Nilsson et al., 2015) and we hypothesize that it is due to a weakening in the arrest potency of SecM by the ADR1a residues that abut the AP in this construct (see Figure 3-figure supplement 9 for sequences). To correct for this effect, we calculated Figure 3A. In the presence of Zn 2+ , the Df FL profiles for WT and uL24 Dloop ribosomes are very similar: Df FL starts to increase around L = 20-21 residues and peaks at L = 25 residues (gray and blue curves). In contrast, for the uL23 Dloop ribosomes, Df FL starts to increase already at L = 17 residues and peaks at L = 21-24 residues (red curve). To quantify these differences, for each f FL curve we calculated the linker lengths characterizing the onset and end of the peak (L onset and L end ; defined as the L-values for which the curve has half-maximal height, as indicated in Figure 3A), as well as the L-value corresponding to the peak of the curve (L max ), Table 1.
A previous cryo-EM study demonstrated that the 29-residue ADR1a domain folds deep inside the ribosome exit tunnel in a location where it is in contact with the uL23 loop (Nilsson et al., 2015), Figure 2B. The additional space available in uL23 Dloop ribosomes makes it possible for ADR1a to start to fold at 3-4 residues shorter linker lengths (L onset ). Assuming an extended conformation of the linker segment (~3 Å per residue), ADR1a folds~9-12 Å deeper in the exit tunnel in uL23 Dloop ribosomes than in WT ribosomes.

Spectrin and titin domains fold deeper in the exit tunnel in uL24 Dloop ribosomes
The 109-residue a-spectrin R16 domain has been shown to fold cotranslationally at L » 35 residues, in close proximity to uL24 in the exit port region (Nilsson et al., 2017). As seen in Figure 3B and Table 1, with both WT and uL23 Dloop ribosomes, R16 has L onset = 31 residues and L max = 35 residues (gray and red curves). For the uL24 Dloop ribosomes however, L onset = 29 residues and L max = 33 residues (Table 1), suggesting that that spectrin R16 folds~6-7 Å deeper in the exit tunnel when the uL24 loop does not obstruct the tunnel exit port.
Similar results were obtained for the 89-residue titin I27 domain, Figure 3C. Previous studies have shown that the I27 domain folds at linker lengths L = 35-39 residues and that it folds in about the same location as does spectrin R16, in close proximity to the uL24 loop (Tian et al., 2018). The f FL profile is not affected by the uL23 loop deletion, but folding commences at~4 residues shorter linker lengths in uL24 Dloop ribosomes, similar to R16 (Table 1).  (Nilsson et al., 2015). The ADR1a structure is from PDB 5A7U. (C) Cryo-EM structure of the uL23 Dloop 70 S ribosome (EMD-4319), fitted to PDB 3JBU (that includes a Gly-tRNA and a 26-residue long arrested SecM AP) to locate uL23 (orange) and the exit tunnel. The enlarged region shows a difference map (in mesh) obtained by subtracting the cryo-EM map of the uL23 Dloop 70 S ribosome from a map generated from 3JBU in Chimera. The difference map shows that the only difference in volume between the two maps is the tRNA (in magenta), the SecM AP (in pink), and the loop deleted from uL23.

Coarse-grained molecular dynamics simulations
In order to provide a more detailed structural framework for interpreting the f FL profile results, we performed coarse-grained molecular dynamics simulations of the cotranslational folding of ADR1a, spectrin R16, and titin I27 in WT, uL23 Dloop, and uL24 Dloop ribosomes, using a recently described model that allows us to calculate f FL profiles from the simulations (Tian et al., 2018). The essence of the method is that the simulations are used to determine folded and unfolded populations at each linker length, and the forces associated with them. Combining this information with the experimentally determined force-dependent escape rate of the AP from the ribosome (Goldman et al., 2015) in a kinetic model allows f FL to be calculated. Simulated (full lines) and experimental (dashed lines) f FL profiles are shown in Figure 3D-F, and detailed simulation results, together with representative snapshots from the simulations of the folded domains at L » L onset , are shown in Figure 3 The simulated ADR1a f FL profile for uL23 Dloop ribosomes, while showing an early onset of folding in agreement with the experimental profile, has a much smaller L max value. We also performed a simulation using a ribosome model with a smaller deletion in the uL23 loop (residues 70-72; red curve marked by X's); in this case, the peak in the simulated profile extends between L onset and L end values that are more similar to the experimental profile for uL23 Dloop ribosomes. The shape of the f FL profile for ADR1a is clearly highly sensitive to fine structural details of the exit tunnel and therefore somewhat difficult to reproduce by coarse-grained simulations.
In summary, both the experimental and simulation results are consistent with the idea that proteins start to fold as soon as they reach a part of the exit tunnel that is large enough to hold the folded protein.
Judging from the f FL profiles, the 29-residue ADR1a domain folds approximately~9-12 Å deeper in the exit tunnel in uL23 Dloop ribosomes than in WT and uL24 Dloop ribosomes, while the 89-and 109-residue titin and spectrin domains fold~6-10 Å deeper inside the tunnel in uL24 Dloop ribosomes than in WT and uL23 Dloop ribosomes; the corresponding values estimated from the simulations are~6 Å for ADR1a and~13-15 Å for I27 and R16 (Figure 3-figure supplement 10 panel B).
Both the uL23 and uL24 loops thus serve to reduce the space available for folding, but in different parts of the exit tunnel. The uL24 loop is particularly interesting in this regard. In bacterial ribosomes, it partially blocks the tunnel exit port, closing off what would otherwise be a wide, funnel-like opening, Figure 2A, and thereby prevents domains of M w !10 kDa from folding inside the exit tunnel. It is conserved (in length, if not in sequence) in bacterial ribosomes, Figure 1-figure supplement 2, suggesting that this will be the case not only for E. coli ribosomes but for bacterial ribosomes in general. Eukaryotic ribosome tunnels have different geometries owing to expansion segments in their rRNA as well as an increased number of proteins and a wider exit port (Wilson and Doudna Cate, 2012;Filipovska and Rackham, 2013); uL24 is among the most divergent proteins compared to bacteria (Melnikov et al., 2015). We therefore expect the precise relation between the onset of folding Video 1. The ribosome exit tunnel (mesh), as calculated for PDB 3JBU, uL23 Dloop and uL24 Dloop ribosomes by POVME. See Figure 1A for coloring scheme. The b-hairpin loops deleted in uL23 Dloop and uL24 Dloop ribosomes are shown in yellow and light blue, respectively. To facilitate the visualization of the exit tunnel, spheres left outside the exit tunnel after POVME processing were manually removed. DOI: https://doi.org/10.7554/eLife.36326.007 and protein M w to be somewhat different in eukaryotic ribosomes, as also suggested by a recent study (Schiller et al., 2017). At present, we do not know to what extent the shape of the ribosome exit tunnel has evolved to optimize the conditions for cotranslational protein folding in different organisms and organelles, but it is not unlikely that such a connection exists.

Enzymes and chemicals
The PURExpress D Ribosome kit was purchased from New England Biolabs (Cat no. E3313S). The components used to prepare Lysogeny Broth (LB Medium) for ribosome isolation were obtained Table 1. L onset , L max , and L end values calculated from the f FL profiles in Figure 3. from BD Biosciences and all other chemicals used were sourced from Merck Sigma Aldrich. ( 35 S) Methionine was purchased from Perkin Elmer. Bis-Tris gels and plasmid isolation kits were obtained from Thermo Scientific.

Plasmids
All ADR1, spectrin and titin constructs fused to the E. coli SecM AP via a variable linker were expressed from the pET19b vector, as described previously (Nilsson et al., 2015;Nilsson et al., 2017;Tian et al., 2018). The spectrin constructs used in this study lacked the soluble domain of LepB at the N-terminus.

Strains and antisera
Strains HDB140 (referred to as WT), HDB143 (referred to as uL23 Dloop) and HDB144 (referred to as uL24 Dloop), as well as rabbit polyclonal antisera against uL23 and uL24, are described in (Peterson et al., 2010).

Isolation of ribosomes
Ribosomes were purified from the strains HDB140, HDB143, and HDB144. The strains were cultured in Lysogeny broth (LB) to an A 600 of 1.0 at 37˚C and chilled on ice for 15 min before they were harvested by centrifugation at 4000 g for 10 min. The cell pellet was washed twice with Buffer A at pH 7.5 (10 mM Tris-OAc, 14 mM Mg(OAc) 2 , 60 mM KOAc, 1 mM DTT, 0.1% Complete Protease Inhibitor) and lysed using the Emulsifex (Avestin) at a pressure of 8000 psi. The cell lysate was loaded on a sucrose cushion at pH 7.5 (50 mM Tris-OAc, 1 M KOAc, 15 mM Mg(OAc) 2, 1.44 M sucrose, 1 mM DTT, 0.1% Complete Protease Inhibitor) and centrifuged at 80,000xg in a Ti70 rotor (Beckman-Coulter) for 17 hr. The obtained ribosomal pellet was resuspended in Buffer B at pH 7.5 (50 mM Tris-OAc, 50 mM KOAc, 5 mM Mg(OAc) 2 , 1 mM DTT), flask frozen in liquid nitrogen and stored at À80˚C. This suspension of ribosomes is presumed to consist of a pool of non-translating 30S, 50S and 70S particles due to the concentration of Mg 2+ in the buffer they are in. Each batch of ribosomes that was prepared was tested for optimal translation by titrating different volumes in the PURExpress D-Ribosome kit.

In vitro transcription and translation
The generated constructs were translated for 20 min. in the PURExpress D-Ribosome kit supplemented with high-salt-washed ribosomes isolated from HDB140, HDB143 (uL23 Dloop), or HDB144 (uL24 Dloop). Plasmid DNA of each construct (300 ng) was used as a template for polypeptide synthesis, and translation was carried out in the presence of ( 35 S) Methionine at 37˚C for 20 min and shaking at 500 r.p.m. For ADR1a constructs, the translation reactions also included either 50 mM zinc acetate or 50 mM of the Zn 2+ chelator TPEN. Translation was stopped by treating the sample with a final concentration of 5% trichloroacetic acid (TCA) and incubated on ice for 30 min. The TCA precipitated samples were subsequently centrifuged at 20,000 g for 10 min in a tabletop centrifuge (Eppendorf) and the pellet obtained was solubilized in sample buffer, supplemented with RNaseA (400 mg/ml), and incubated at 37˚C for 15 min. The samples were resolved on 12% Bis-Tris gels (Thermo Scientific) in MOPS buffer for ADR1 and MES buffer for Spectrin and Titin. Gels were dried and subjected to autoradiography and scanned using the Fujifilm FLA-9000 phosphorimager for visualization of radioactively labeled translated proteins.

Quantification of radioactively labelled proteins
The protein bands on the gel were quantified using MultiGauge (Fujifilm) from which one-dimensional intensity profiles of each gel lane was extracted. This information was subsequently fit to a Gaussian distribution using EasyQuant (Rickard Hedman, Stockholm University). The sum of the arrested and full-length bands was calculated, and this was used to estimate the fraction full-length protein for each construct.

Cryo-EM sample preparation and data processing
The uL23 Dloop ribosomes (4 A 260 /ml) diluted in grid buffer (20 mM HEPES-KOH, 50 mM KOAc, 10 mM Mg(OAc) 2 , 125 mM sucrose, 2 mM Trp, 0.03% DDM) were loaded on Pelco TEM 400 mesh Cu grids pre-coated with 2 nm thick carbon and frozen using the Vitrobot Mark IV (FEI). Data were collected on the Titan Krios (FEI) microscope operated at 300 keV and equipped with a Falcon II direct electron detector. The camera was set to a nominal magnification of 75,000X, which resulted in a pixel size of 1.09 Å at the sample level and a defocus range of À1 to À3 mm. The frame dose used was 1.17 e/Å 2 , and 20 frames were aligned using MotionCor2 (Li et al., 2013) within the Scipion software suite (de la Rosa-Trevín et al., 2016). The micrographs were visually inspected and those within a resolution threshold of 5 Å were selected, yielding 3522 micrographs. 471,272 particles were picked using Xmipp manual-pick followed by particle extraction within Scipion and further processing in CryoSPARC (Punjani et al., 2017). Two rounds of 2D classification were done, and particles resembling 30S and 50S subunits alone were discarded after visual inspection of the classes. The remaining 297,363 particles of the 70S ribosome were subjected to ab initio reconstruction into three classes to further sort out heterogeneity. A single homogeneous class consisting of 132,029 particles was used for final homogeneous refinement that resulted in a final map with an average FSC resolution at 0.143 of 3.28 Å . The obtained map of the 70S ribosome was sharpened and corrected for handedness in CryoSPARC fitted with PDBs 3JBU and 4YBB in Chimera (Pettersen et al., 2004). Local resolution and FSC at 0.143 was estimated in cryoSPARC. The electron microscopy map was deposited in the Electron Microscopy Data Bank.
The initial model for uL23 Dloop was built with Coot, and improved by energy minimization in a solvated dodecahedron box of explicit TIP3P waters, neutralized with chloride ions and using the Amber 99SB-ILDN force field (Lindorff-Larsen et al., 2010). The steepest descent minimization method implemented in GROMACS 2016.1 was used (Abraham et al., 2015;Pal et al., 2014). Even after minimization, the backbone of the new loop formed after the deletion of residues 65-74 still showed improper geometry and Ramachandran outliers, so we used kinematic sampling (Bhardwaj et al., 2016) to model alternative loop conformations, and then we selected the loop that could fit the electron density and had the best Ramachandran score.

Calculation of tunnel volume
The volume calculations were performed with POVME 2.0 (1). We used the E. coli SecM structure PDB 3JBU as a reference. To determine the inclusion region, we generated a series of overlapping spheres-eight with a 20 Å radius, and one with a 40 Å radius. In order to have a complete coverage of the exit tunnel, the centers of the spheres were chosen to match the coordinates corresponding to alternating Ca atoms of the amino acids of the SecM arrest peptide located within the exit tunnel (for the 20 Å radius spheres the residues use as centers were D11, F13, T15, V17, I19, Q21, Q23, I25, A27, G28 and for the 40 Å radius sphere the residue was E3). Grid Spacing was set to 2.0 Å , and the distance cut-off to 1.09 Å . For all three cases (WT, uL24 D loop, uL23 D loop), we used the same inclusion region. We also removed the SecM arrest peptide located within the exit tunnel. For uL23 Dloop ribosomes residues 65-75 were removed from uL23, and for uL24 Dloop ribosomes residues 42-57 were removed from uL24 (numbering based on PDB 3JBU) prior to the calculation.

Kinetic model to calculate fraction full length protein f FL (t)
The theoretical force profiles ( Figure 3D-F) for ADR1a, I27, and R16 were calculated based on a kinetic model introduced in our previous study (Tian et al., 2018). Briefly, the rate, k e , of the arrest peptide sequence escape from the peptidyl transfer center with a force (F) exerted by the folding protein can be calculated using the Bell model: where Dx z is the distance from the free energy minimum to the transition state, k 0 the rupture rate when force equals to zero, k B is Boltzmann's constant, and T the absolute temperature. In this study, k 0 and Dx z are set to be 3.4 Â10 À4 s À1 and 4.5 Å , respectively, based on a previous experimental study (Goldman et al., 2015) in which k 0 and Dx z were estimated to be in the range of 0.5 Â10 À4 to 20 Â10 À4 s À1 and 1-8 Å , respectively.