Bacterial ribosome collision sensing by a MutS DNA repair ATPase paralog

Abstract

Ribosome stalling during translation is detrimental to cellular fitness, but how this is sensed and elicits recycling of ribosomal subunits and quality control of associated mRNA and incomplete nascent-chains is poorly understood^1,2. Here we uncover Bacillus subtilis MutS2, a member of the conserved MutS-family of ATPases that function in DNA-mismatch repair³ as an unexpected ribosome-binding protein with an essential function in translational quality control. Cryo-EM analysis of affinity-purified native complexes reveals that MutS2 functions in sensing collisions between stalled and translating ribosomes and suggests how ribosome collisions can serve as platforms to deploy downstream processes: MutS2 has an RNA endonuclease Small MutS-Related (SMR) domain, as well as an ATPase/clamp domain that is properly positioned to promote ribosomal subunit dissociation, which is a requirement both for ribosome recycling and for initiating ribosome-associated protein quality control (RQC). Accordingly, Muts2 promotes nascent chain modification with Alanine-tail degrons—an early step in RQC—in an ATPase domain-dependent manner. The relevance of these observations is underscored by evidence of strong co-occurrence of MutS2 and RQC genes across bacterial phyla. Overall, the findings reveal a deeply-conserved role of ribosome collisions in mounting a complex response to the interruption of translation within open reading frames.

Ribosome stalling is prevalent during protein synthesis, estimated to take place at least 0.4% of the translation events in exponentially growing E. coli cells⁴. Ribosome stalling can be detrimental to cellular fitness, among other reasons, for producing aberrant and potentially toxic nascent polypeptides^4,5. Bacteria and eukaryotes were initially thought to have evolutionarily unrelated processes for targeting such products of ribosome stalling for degradation—transfer-messenger RNA (tmRNA or SsrA) and ribosome-associated quality control (RQC), respectively^4–7. More recently, however, it has been discovered that the conserved protein, RqcH, mediates a process in bacteria that is homologous to eukaryotic RQC⁸.

There has been great recent interest in understanding how stalled ribosomes are sensed and elicit RQC and other processes (see e.g.,^1,2,5,9,10). An emerging notion is that the mRNA location of a stalling site (i.e., at the 3’ end versus at an internal position) confers unique structural features to stalled ribosomal complexes, which in turn dictate alternative sensing and ‘rescue’ strategies.

Biochemical evidence from eukaryotic systems suggest that ribosomes stalled at internal mRNA sites are sensed after trailing elongating ribosomes collide with them^2,10–13. The resulting collided ‘disomes’ are recognized by factors such as the E3 ligase Hel2 (ZNF598 in mammals), which in turn recruits downstream effectors in a ubiquitylation-dependent manner^11,13–18. One such effector is the mRNA endonuclease Cue2 (NONU-1 in worms, N4BP2 in mammals)^19,20. Cue2-mediated mRNA cleavage can facilitate exonuclease-mediated mRNA degradation^19,20 and would also cause trailing ribosomes to stall in the newly-generated 3’end, which are rescued by the Dom34-Hbs1-ABCE1 pathway^5,9. Another Hel2 effector is the Slh1 helicase (mammalian ASCC3). Both Slh1/ASCC3 and the AAA ATPase ABCE1 dissociate stalled ribosomes^16,17,21, converging in the generation of large ribosomal subunits still obstructed with nascent chain-tRNA conjugates. The obstruction is recognized by RqcH and its orthologs—yeast Rqc2 or mammalian NEMF—to initiate RQC, ultimately leading to the nascent chain becoming marked for degradation, and to recycling of the large subunit^5,6. Whether bacteria also rely on ribosome collisions for sensing and responding to translational stalling within open reading frames (ORFs), the factors involved, as well as the structural basis for how collisions are sensed and elicit ribosome recycling and degradation of associated mRNA and nascent-chains in any organism, have remained unknown.

MutS2 protects from translational stress

In eukaryotes, Cue2 orthologs are recruited to Hel2-ubiquitylated, collided disomes via ubiquitin-binding CUE domains²² and additionally have endonuclease SMR (Small MutS-Related) domains that cleaves the associated mRNA^19,20 (Supplementary Fig. 2a). The role of this SMR domain-containing protein upstream of eukaryotic RQC turned our attention to the only B. subtilis SMR domain protein, MutS2^23,24. MutS2 is a paralog of the conserved DNA mismatch repair factor, MutS^3,25,26. Notably, the MutS domains I and II, which bind DNA and recruit the MutL DNA repair cofactor, respectively, are absent in MutS2 (Supplementary Fig. 2a,b). On the other hand, MutS and MutS2 share domains III/IV (levers/clamp) and V (ATPase), which in MutS2 are followed by a C-terminal extension containing a coiled-coil, a KOW domain found in proteins implicated in RNA binding and translation²⁷, and a C-terminal endonuclease-active SMR domain²⁸ with a conserved catalytic Aspartate^19,20 (Supplementary Fig. 2a).

The B. subtilis MutS2-encoding gene, mutSB, is not essential for growth²³. To seek genetic evidence for its role in the cellular response to ribosome stalling, we first examined whether mutSB deletion (ΔmutSB) caused increased sensitivity to translation elongation inhibitors and found that the growth of ΔmutSB cells was indeed impaired under erythromycin concentrations which allowed wild type (WT) cells to grow normally (Fig. 1a and below). Moreover, growth assays show that mutSB interacted with genes of both trans-translation (ssrA) and RQC (rqcH) pathways (Fig. 1a).

Fig. 1. MutS2 binds to ribosomes and protects cells from translational stress.

a, mutSB protects cells against erythromycin and interacts genetically with the translational quality control pathways, RQC and SsrA. b, MutS2 binds to 70S ribosomes. Coomassie staining of FLAG-MutS2 and RqcH-FLAG co-IP’ed proteins. E.V., empty vector (n>3). c, Proteins co-IP’ed with FLAG-MutS2 identified by label-free LC-MS/MS. Individual proteins are represented as dots, with the Rel stringent response factor and RNA degradosome factors in green, 30S proteins in blue, and 50S proteins in red. d, Erythromycin-stimulated association of MutS2 with micrococcal nuclease (MNase)-resistant disome fractions. 3xFLAG-MutS2 expressing cells were exposed (right) or not (left) to erythromycin and lysates were treated with MNase before fractionation in sucrose gradient and analysis by anti-FLAG blot (top) or Coomassie staining (bottom). For the top fraction, only 1/20^th of the volume was loaded. ‘30S + top’ indicates contamination of 30S with top fractions (n=2). e, Erythromycin-stimulated association of MutS2 with heavy polysome fractions (n=2). As in ‘d’, but without MNase treatment. For gel source data, see Supplementary Fig. 1.

To begin investigating whether the effects of MutS2 on the response to ribosome stalling are direct, we immunoprecipitated (IP’ed) FLAG-tagged MutS2 and examined associated proteins by SDS-PAGE (Fig. 1b). MutS2 selectively co-IP’ed with proteins with a characteristic pattern of 70S ribosomes, whose enrichment was confirmed by both mass spectrometry (Fig. 1c) and cryo-electron microscopy (cryo-EM) (below).

Collided disomes are MutS2 substrates

To obtain insights into MutS2 function, we subjected affinity-purified FLAG-tagged MutS2-ribosome complexes to cryo-EM single particle analysis (Extended Data Fig. 1). The most abundant ribosomal species were 70S ribosomes in various states of the translational elongation cycle (~ 85 %) followed by 50S large ribosomal subunits obstructed with a nascent chain-linked P-site tRNA (~ 11 %) and structurally defined disomes (~ 4 %), in which a trailing ribosome was collided into a leading (or “stalled”) ribosome. Computational sorting further produced a subpopulation of disomes associated with a well-defined tripodal ligand (~ 44 %; see below). Docking an AlphaFold-Multimer-based model for B. subtilis MutS2 (Methods) led to unambiguous identification of the density as a MutS2 homodimer (below), with the model further validated by high similarity to bacterial MutS homodimer structures^26,28,29 (Supplementary Fig. 1a, Extended Data Fig. 2). Despite extensive computational particle sorting, MutS2 could not be observed associated with the excess, non-collided 70S ribosomes.

Substantiating the structural interpretation that MutS2 senses ribosome collisions, sucrose gradient centrifugation analyses revealed that MutS2 remained associated with the disome fraction following micrococcal nuclease (MNase) digestion (Fig. 1d, Supplementary Fig. 3). The enrichment of MutS2 in MNase-resistant disomes is underscored by the larger ratio of MutS2 to total ribosomes in disome compared to 70S fractions. Without MNase digestion, MutS2 could be found in heavy polysomes (Fig. 1e, Supplementary Fig. 3), consistent with the notion that ribosomes are more likely to collide on mRNAs that have higher translation initiation rates¹³. Finally, further supporting our model, MutS2 association with heavy polysomes and MNase-resistant disomes was greatly increased by erythromycin-induced translational stalling (Fig. 1d,e). Together, these observations suggest that collided disomes are direct MutS2 substrates.

Structural remodeling of the disome

We first structurally characterized the MutS2-co-IP’ed collided disomes (Fig. 2a, Extended Data Fig. 3a–d, Extended Data Table 1, Methods). The leading and collided ribosomes interact via an extended interface between the 30S subunits that mainly clusters around the inter-ribosomal mRNA (Fig. 2b, Extended Data Fig. 4). The architecture of the bacterial disome is confined by the trajectory of the inter-ribosomal mRNA and the shape of the ribosome, and is similar to eukaryotic collided disomes^11,13,30. However, universally-conserved ribosomal proteins that are central to collided disome formation in bacteria contribute less to the eukaryotic disome interface.

Fig. 2. Cryo-EM structure of the bacterial collided disome.

a, Structure of the collided bacterial disome at ~ 3.5 Å resolution. The canonical position of uS2 on the stalled ribosome is indicated. b, The disome interface is formed by multiple contacts between leading and collided ribosomes. See Extended Data Fig. 4 for details. c, In the collided disome, ribosomes adopt distinct rotational states associated with specific tRNA occupancy. In all panels, local resolution filtered cryo-EM densities are shown. Density for 30S subunits and mRNA was obtained by focused refinement on the 30S subunits. Density for the 50S subunits and tRNAs was obtained by focused refinement on the full 70S particles (Extended Data Figs. 1 and 3).

Also as in eukaryotes, ribosomes in the bacterial collided disome adopt specific rotational states. The leading ribosome is unrotated and associated with a nascent-chain linked tRNA in a canonical P-site conformation (Fig. 2c). The mRNA can be continuously traced throughout the ribosomal mRNA channel including the ribosomal A-site, indicating that the leading ribosomes have stalled in internal sites of endogenous mRNAs. In contrast, the collided ribosome is in a rotated, pre-translocation state, associated with a nascent chain-linked A/P-site tRNA and a P/E-site tRNA (Fig. 2c). The mRNA can be traced throughout the mRNA channel and projects out from the mRNA exit site (Fig. 2c) interacting with the anti-Shine-Dalgarno rRNA (Extended Data Fig. 5a). For ~5 % of disomes we observed density corresponding to a third ribosome collided into the disome unit (Extended Data Fig. 1e). Even though only moderately resolved, the third ribosome is similar to the collided ribosome of the disome unit regarding overall structure.

Strikingly, and differently from eukaryotic collided disomes^11,13,30, the leading ribosome has undergone specific compositional remodeling when compared to the collided ribosome or to actively translating B. subtilis ribosomes³¹. Ribosomal protein uS2, which is normally situated close to the 30S subunit mRNA exit site, is entirely missing (Fig. 2a, Extended Data Fig. 5b) and conversely, bS21 and bL9 proteins that are absent from the canonical B. subtilis ribosome, are specifically recruited to the leading ribosome (Fig. 2c) and assume similar positions as in the E. coli ribosome, of which these two proteins are constitutive components.

bS21 appears to stabilize the inter-ribosomal mRNA segment, which projects in a straight trajectory from the E-site of the leading ribosome towards the A-site of the collided ribosome (Fig. 2c). This strain-induced altered trajectory (Extended Data Fig. 5a) likely underlies structural remodeling of the mRNA exit region in the leading ribosome, including recruitment of bS21 and dissociation of uS2.

bL9 is specifically recruited to the leading ribosome 50S subunit near the L1-stalk via its N-terminal half (Fig. 2b, Extended Data Fig. 4d,e), while the C-terminal half bridges over to the collided ribosome and contacts the 30S subunit 16S rRNA, where it would clash with the mRNA-translocating elongation factor G (EF-G) (Extended Data Fig. 6), thereby locking the collided ribosome in a rotated state. Notably, bL9 suppresses translational frameshifting caused by ribosome collisions³². Thus, in locking the collided ribosome in the rotated state, bL9 is likely important to prevent frameshifting caused by the strained inter-ribosomal mRNA.

MutS2 selectively binds collided disomes

We next examined the structural basis for MutS2 binding to collided disomes. Like MutS, whose clamp domains encompass the DNA duplex in DNA mismatch repair²⁶ (see Supplementary Fig. 2a), MutS2 utilizes its clamp domains to bind to the leading ribosome in translational quality control (Fig. 3a,b). MutS2 monomer 1 (MutS2-A) contacts the A-site finger, the central protuberance rRNA and ribosomal protein uL5 (Fig. 3a). Higher local resolution for this interaction site (Extended Data Fig. 3g) indicates that it is the dominant and most stable contact with the leading ribosome. MutS2 monomer 2 (MutS2-B) binds to the central protuberance region from the E-site facing side of the ribosome in two different conformations (Fig. 3b, Extended Data Fig. 7).

Fig. 3. Structural basis for ribosome collision sensing by MutS2.

a, Frontal view on the structure of the MutS2-disome complex. The MutS2 homodimer density is shown with transparency and the atomic model superposed. Right: zoom onto the MutS2-A clamp domain binding to the A-site finger and the central protuberance of the leading ribosome. b, View rotated by 180°. The MutS2 homodimer encompasses the central protuberance from the A-site and E-site facing sides. Right: zoom on MutS2-B clamp domain insertion into the gap between ribosomal subunits and the L1-stalk. The additional contact between the MutS2-B lever domain and the collided ribosome 30S subunit is indicated. c, Top view onto the MutS2-disome complex showing how the MutS2 coiled-coil aligns with a composite binding site formed by the two 30S subunits. The interaction sites between the coiled-coil and helices 39 of the 30S subunit rRNAs are indicated (asterisks). d, Model for ribosomal subunit splitting mediated by conformational changes of MutS2. The two observed MutS2 conformations (transparent blue) and one additional extrapolated conformation (solid blue) are shown. Local resolution-filtered ribosome densities are shown as obtained after ribosome-wise particle refinement of all MutS2-containing particles. MutS2 density is shown as obtained after particle refinement centered on the stalled ribosome, for either all MutS2-containing particles (panel ‘a’, zoom) or only particles representing MutS2 conformation 2 (all other panels) (see Extended Data Figs. 1 and 3).

The mode of interaction between MutS2 and the leading ribosome provides insights into how the two MutS2 clamp regions selectively bind to the stalled ribosome by reading the activity status of the ribosome in concert.

The clamp region of MutS2-A binds into a narrow groove in between ribosomal protein L5 and the A-site finger of the leading ribosome (Fig. 3a, ‘finger contact’). The width of this groove changes dynamically during translation elongation (Extended Data Fig. 8), allowing MutS2-A binding only in an unrotated state as adopted by the stalled ribosome. Consistently, the loss of mutSB did not confer sensitivity to the translation elongation inhibitors spectinomycin and hygromycin B (Supplementary Fig. 4), which stall ribosomes in a rotated conformation^33,34, while causing increased sensitivity to erythromycin and chloramphenicol (Fig. 1a and Supplementary Fig. 4), which stall ribosomes in an unrotated state³⁵. Importantly, the binding mode of MutS2-A allows for a negative selection mechanism—MutS2-A binding to a ribosome in an unrotated state during active translation or translational pausing would be followed by dissociation with the next intersubunit rotation.

The clamp region of MutS2-B inserts far into a gap between the central protuberance and the L1-stalk (Fig. 3b). Access to this gap is spatially constrained for ribosomes engaged in active translation, but it is widened and well-accessible in the leading ribosome, where all major obstacles for MutS2-B binding present on actively translating ribosomes have been moved out of the way: the E-site tRNA is absent, the 30S subunit does not pose an obstacle in the unrotated state, and the L1-stalk is locked in an extreme ‘out conformation’. Critically, the locking of the L1 stalk is mediated by interactions occurring exclusively in collided disomes, namely with both bL9 and the collided ribosome 30S subunit (Fig. 2a, Extended Data Fig. 4c,d). bL9 was not essential for MutS2-ribosome binding or erythromycin resistance (Supplementary Fig. 5), suggesting that the 30S subunit contact is likely sufficient for locking the L1-stalk in the ‘out conformation’.

Cumulatively, the above observations indicate that ribosome stalling and collision are strictly required for stable binding of the MutS2 clamp domains to the stalled ribosome, thereby providing a structure-based rationale for the first step in ribosome collision sensing.

MutS2 proteins have a predicted coiled-coil (513-586) extending from the ATPase domain (Supplementary Fig. 2a). This coiled-coil projects along the disome interface, coordinated by interactions with helix 39 of the 30S subunit rRNA of both ribosomes (Fig. 3c), which together form a composite binding site for the coiled-coil and thus provide another layer of selectivity in collision sensing by MutS2.

MutS2-mediated downstream processes

The MutS2-disome structure provides insights into the roles of the ATPase/clamp and SMR domains in the ribosome collision response.

In the MutS2 cryo-EM density, the C-terminal ~200 amino acid-long segment including KOW and SMR domains and linker sequences is not visible (Fig. 3a,c; Methods), suggesting that the linkers likely confer versatility to the SMR domains for endonucleolytic cleavage of the mRNA near the collision site.

In the MutS2 homodimer, the ATPase and clamp domains are shaped like a tweezer that holds to the stalled ribosome (Fig. 3). In ABC-type ATPases, like the Rli1/ABCE1 ribosome splitting factor and MutS DNA repair factors, ATP-binding-induced conformational changes generate power strokes that drive structural changes in associated domains or macromolecules^21,26. Based on the positioning of the MutS2 clamp domains directly by the ribosomal inter-subunit interface, we hypothesize that ATP-driven rearrangement of one monomer relative to the other, similar in extent and direction to what has been observed for paralogous MutS proteins²⁶, separates the subunits of the stalled ribosome (Fig. 3d). Accordingly, we could disentangle two MutS2 conformational states (Extended Data Fig. 7) that are consistent with initial steps of such a splitting reaction and we observe obstructed 50S subunits in the MutS2 co-IP’ed sample (Extended Data Fig. 9) likely arisen from ribosome splitting during purification.

ATPase domain-dependent RQC initiation

To test the hypothesis that ATP-dependent conformational changes in MutS2 promote ribosomal splitting, we first mutated MutS2 residues required for ATP binding (K341A mutation in the Walker A motif) or hydrolysis (E416A) and investigated their effect on MutS2 binding to collided ribosomes. Consistent with the hypothesis, both mutants were strikingly enriched in the heavy polysome fraction of sucrose gradients (Fig. 4a and Supplementary Fig. 6a) as well as in MNase-resistant disomes (Supplementary Fig. 6a).

Fig. 4. MutS2 promotes ribosome splitting upstream of RQC.

a, ATPase domain mutation-stimulated association of MutS2 with heavy polysome fractions. As in Fig. 1d, but with no erythromycin (n= 2). b, A novel assay for Ala-tailing reveals a role for MutS2 upstream of RqcH. Lysates of strains episomally expressing a ribosome stalling reporter (GFP-ns) were utilized for GST-Pirh2 pulldown and analyzed by anti-GFP blot (n= 2). c, MutS2 wt, but not ATPase domain mutants, rescues the Ala-tailing defect of ΔssrA ΔmutSB cells. Left, as in ‘b’, except that a genome-integrated GFP-ns reporter was used. Right, effect of mutSB deletion on the ssrA background. The ratio of modified to input GFP-ns is shown. Data are presented as mean (n= 2). d, mutSB and rqcH strongly co-occur among bacterial genomes. Bar graphs indicate the number of species having both mutSB and rqcH or only one of the genes, among completely sequenced genomes. e, MutS2 protects cells against erythromycin in an ATPase domain-dependent manner. Strains expressing FLAG-MutS2 (wt and mutants) or empty vector (E.V.), analyzed as in Fig. 1a. f, MutS2 ATPase and SMR domains are both required for cellular fitness. Mutant strains were mixed with the WT strain at 1:1 OD₆₀₀ ratio, and subjected to competitive growth in two separate experiments for either 20 (orange bars) or ~40 (blue bars) doubling times (DT), before determining the frequency of each strain in the final population by colony PCR. For gel source data, see Supplementary Fig. 1.

The splitting of stalled ribosomes produces obstructed large subunits, which are the substrates of RQC. To test whether MutS2 initiates RQC, we monitored the modification of ribosomal stalling products with C-terminal Alanine (Ala) tails, which are synthesized by RqcH with the assistance of RqcP^{8 36–38} and function as degrons⁸. To monitor Ala-tailing, we have developed ‘Ala-PIRHification,’ a novel affinity purification method using the human Pirh2 Ala tail-binding protein³⁹. Ala-PIRHification was first validated by analyzing the RqcH-dependence of the pulldown of a ribosome stalling reporter lacking stop codons (GFP-nonstop, or GFP-ns). Since ribosomes stalled at the 3’end of the GFP-ns mRNA are preferred substrates for SsrA^4,7,8, ssrA-deleted strains were also included in the analysis. In total lysates, two GFP-ns bands were observed: a faster-migrating band, corresponding to either unmodified or Ala-tailed reporter; and a slower-migrating band, corresponding to SsrA-tagged reporter (Fig. 4b). As expected, Pirh2 pulled down Ala-tailed GFP-ns, which required ssrA deletion to be observed (lane 4). Pirh2 also pulled down SsrA-tagged GFP-ns (e.g., lane 3)—not surprisingly, since this tag is also C-terminal and ends in Ala-Leu-Ala-Ala in B. subtilis. Importantly, Pirh2 did not pull down GFP-ns from the ΔrqcH ΔssrA strain (lane 7), indicating that it strictly required at least one of the modifications.

While mutSB deletion did not affect SsrA tagging of GFP-ns (Figs. 4b,c), it specifically decreased Ala-tailing by ~80% (Fig. 4c; compare ΔssrA to ΔmutSB ΔssrA) and this effect was rescued by re-expressing MutS2 wt. Importantly, the Ala tailing defect in ΔmutSB ΔssrA cells could not be rescued by expressing MutS2 ATPase domain mutants (Fig. 4c) even though expression of wt and mutants was comparable (Supplementary Fig. 7). These results are consistent with MutS2 promoting ribosomal splitting upstream of RqcH; together with genetic interaction and structural data above, the residual Ala-tailing in ΔmutSB ΔssrA cells suggests that yet unidentified stalled ribosome-sensing factors can act redundantly with MutS2.

Strikingly, bioinformatic analyses revealed a strong and significant genetic association of mutSB with rqcH. First, while mutSB and rqcH were found, respectively, in 33 % and 29 % of 10,982 completely sequenced bacterial genomes analyzed, these genes mostly co-occur, with only 6 % and 1 % of species appearing to have mutSB or rqcH alone, respectively (Fig. 4d and Extended Data Fig. 10a). This corresponds to a 308-fold enrichment over the expected distribution (Fisher test p-value close to 0). The association was not due to database bias, as it could still be observed when quantitated separately for each of several bacterial phyla (Extended Data Fig. 10a). Finally, in several species mutSB and rqcH map to the same gene neighborhood (Extended Data Fig. 10b) raising the possibility that they may also be co-regulated.

We next analyzed the requirement for the ATPase domain in MutS2 function under translational stress. Strikingly, both MutS2 ATPase domain mutants were completely unable to rescue the erythromycin sensitivity of ΔmutSB cells (Fig. 4e), suggesting an essential role for the domain. In contrast, mutation of a predicted catalytic residue of the SMR domain (D711A; Fig. 4e and Supplementary Fig. 2c) and even its complete deletion (MutS2 1-701; Supplementary Fig. 8) led to comparably small defects in rescuing the phenotype. To further probe the physiological relevance of the SMR domain, we utilized a more sensitive cellular fitness assay. The reference parental (WT) strain was mixed at 1:1 ratio pairwise with ΔmutSB strains expressing or not MutS2 wt or mutants. With growth time, we observed the progressive outcompetition of strains with MutS2 mutations (including SMR domain deletion) from the final population (Fig. 4f).

Discussion

This work presents the cryo-EM structure of a native bacterial collided disome; overall, the ribosome arrangement is reminiscent of inter-ribosomal contacts often observed in X-ray crystallography of bacterial ribosomes⁴⁰, which may reflect an intrinsic tendency of ribosomes to aggregate in this particular way. Importantly, the work establishes a role for ribosome collisions in bacterial translational quality control. We identify MutS2 as a factor linked genetically and functionally to translational stress and that acts as a ribosome collision sensor in bacteria. Furthermore, our cryo-EM structure of a disome in complex with MutS2 supports a model in which MutS2 acts in ribosomal rescue by mediating ATP-driven ribosomal subunit splitting itself, via conformational rearrangement similar to the MutS paralog (Fig. 3d). Finally, we present evidence linking MutS2 to functional consequences of ribosomal rescue: i) obstructed 50S subunits were found in the MutS2 co-IP’ed sample (Extended Data Fig. 9); ii) MutS2 acted upstream of Ala-tailing (Fig. 4b,c); iii) ATPase domain mutations led to the predicted effects associated with defective recycling, namely mutants were enriched in heavy polysome fractions (Fig. 4a and Supplementary Fig. 6) and failed to promote Ala-tailing and proteolysis of ribosomal stalling products (Fig. 4c). The notion that a major function of MutS2 is to elicit RQC provides a plausible explanation for the strong co-occurrence of mutSB and rqcH in bacteria (Fig. 4e and Extended Data Fig. 10). Hence, we propose as an alternative name for MutS2, RqcU (RQC-upstream factor). In conclusion, MutS2 is likely to function both as a collision sensor and ribosome splitting factor, although our data does not rule out the possibility that MutS2 promotes splitting indirectly.

Our findings also shed light onto the long-standing problem of how bacteria resolve ribosome stalling in mRNA 3’ ends versus within ORFs (Extended Data Fig. 11). While ribosomes stalled in mRNA 3’ ends (as in GFP-ns) can be directly targeted by SsrA^4,7, this factor cannot act directly on ribosomes stalled within ORFs, with mRNA occupying the A-site and mRNA entry channel. Rather, the latter are primary targets of MutS2 after collision occurs. Analogously to eukaryotes, collided disomes then serve as a platform for launching RQC from stalling within ORFs in bacteria. The pathways for ribosomal rescue in mRNA 3’ ends versus within ORFs are likely to cross-talk, however. SsrA may act downstream of ribosome stalling within ORFs if the mRNA is cleaved to generate free 3’ ends (Extended Data Fig. 11), such as by virtue of MutS2 SMR domain action. Conversely, if stalled ribosomes remain associated long enough with the 3’ end of non-stop mRNA—such as when SsrA activity is limiting (Fig. 4b,c)—collisions can take place, leading to MutS2-mediated ribosomal rescue.

The MutS2 and bL9 co-IPs uncovered all catalytic subunits of the RNA degradosome⁴¹ among the top hits (Fig. 1c), suggesting that the latter associate with stalled or collided ribosomes. The MutS2 and bL9 co-IPs additionally uncovered the Rel translational stress signaling factor as a main interactor (Fig. 1c). Together, our results suggest that a variety of damage-control responses are orchestrated already at the level of collision sensing in bacteria, namely nascent chain proteolytic tagging, mRNA decay, stress signaling, ribosomal recycling, and inhibition of frameshifting. The relevance of the collision response is underscored both by this complexity and the apparent functional redundancy at different steps of the pathway, including in collision sensing and at the level of mRNA endonucleolytic cleavage.

Methods

Bacillus strains and cultures

All strains were derived from B. subtilis 168 1A700 and were grown in LB media at 37 °C. To generate strains deleted for mutSB and rplI we followed the procedure described by⁴² and as reported in^8,36. Plasmid transformation was performed as reported in³⁶. Overnight cultures were diluted to OD₆₀₀ = 0.1 and grown until OD600 = 1.5 in BMK media containing 60 mM K₂HPO₄, 40 mM KH₂PO₄, 2% dextrose, 3 mM trisodium citrate dihydrate, 0.08 mM ferric ammonium citrate, 15 mM potassium aspartate, 10 mM MgSO₄, 150 nM MnCl₂, 0.2 mM L-tryptophan, 0.05% yeast extract⁴³. 350 μL of culture were incubated with 1 μg plasmid DNA at 37°C, 160 rpm for 1 h followed by plating on agar in presence of chloramphenicol (5 μg/mL) or a combination of erythromycin (1 μg/mL) and lincomycin (25 μg/mL).

Bioinformatics

Sequences from the ‘representative genomes’ section of the ProGenomes database v 2.1 (https://progenomes.embl.de/data/repGenomes/freeze12.genes.representatives.fasta.gz)⁴⁴ were split into bacterial and archaeal groups and analyzed separately by generalized profiles⁴⁵ for the RqcH family and the MutS2-specific C-terminal domain. Significant profile hits (p<0.01) were mapped to the underlying species ID and the overlap set for species containing both protein classes was determined and subjected to detailed gene-order analysis⁴⁵. For the association analysis, bacterial species were grouped by phylum, using taxonomic information included in the ProGenomes database. A list of all species analyzed can be downloaded from http://progenomes.embl.de/data/proGenomes2.1_specI_lineageNCBI.tab.

Constructs

DNA cloning was performed in E. coli DH5α by using standard cloning procedure as described^8,36. To generate constructs for deleting mutSB and rplI (pMAD ΔmutSB and pMAD ΔrplI, respectively), 1-Kb fragments flanking the corresponding gene were amplified from bacterial chromosomal DNA. Gene-upstream and -downstream fragments were cloned with the BamHI/EcoRI and the EcoRI/NcoI restriction sites, respectively. Double-digested fragments were ligated to each other through the EcoRI ends and cloned into the BamHI/NcoI sites of the linearized pMAD vector.

FLAG-MutS2 and bL9-FLAG were cloned into pHT01 P₄₃ wRBS RsfS-FLAG T_gyrA, which contains the 43 promoter (P₄₃) and a weak ribosome binding site (wRBS) inserted between the KpnI and the XbaI restriction sites, and RsfS-FLAG inserted between the XbaI and the NotI restriction sites followed by the gyrA terminator (TgyrA). P₄₃ together with a strong ribosome binding site (RBS) were inserted by primers annealing in place of P₄₃ wRBS. MutS2 (wt and 1-701 mutant) and bL9 were amplified by PCR with primers encoding a FLAG tag followed by a GSX4 linker, or a 3xFLAG tag at the N-terminus or at the C-terminus, respectively, and then cloned between the XbaI and the NotI restriction sites. The mutSB endogenous promoter and ribosome binding site (referred as P_mutSB) were inserted in place of P₄₃ RBS by inverse PCR and primers phosphorylation. MutS2 D711A, K341A and E416A mutants were generated by inverse PCR.

All constructs were verified by sequencing.

Purification of MutS2-ribosome complexes for cryo-EM analysis

The procedures were performed as described^8,36 with minor modifications. 2x2-L cultures were grown in LB medium supplemented with 5 μg/mL chloramphenicol at 37°C, 120 rpm to a final OD₆₀₀ of ~1.5, then cells were harvested, frozen in liquid nitrogen and stored at −80°C. Cells were disrupted on ice by sonication (10 cycles of 30 s each; 100% amplitude) in sterile-filtered buffer containing 50 mM Tris-HCl pH 7.4 (at 4°C), 100 mM NaCl, 10 mM MgCl₂, 5% glycerol (v/v), 0.1% NP-40 (v/v), 1 mM DTT, 1x protease inhibitor (Roche #04693159001) and ribonuclease inhibitor (Thermo Fisher #10777-019).

Lysates were clarified by centrifugation at 17,000 rpm in an SS-34 fixed-angle rotor for 22 min at 4°C, and the supernatant was incubated for 3 h at 4°C on a turning wheel (Labinco) with 2x200 μL pre-washed anti-FLAG M2 Affinity Gel (Sigma-Aldrich #A2220). Beads were collected by centrifugation, and after 3 washes with 20 mL lysis buffer were transferred to a Mobicol ‘F’ filter with a 35-mm pore size (MoBiTec) and washed by gravity flow with lysis buffer without glycerol and NP-40. Proteins were eluted by addition of lysis buffer without glycerol and NP-40 added with 0.3 mg/ml 3X FLAG-peptide (Sigma-Aldrich #F4799) for 45 min at 4°C on a turning wheel and then a 15 min standing on ice after a brief centrifugation. Afterward the eluate was collected and a portion was used for SDS-PAGE analysis and the remaining samples was utilized for cryo-EM grids preparation.

FLAG-MutS2 and bL9-FLAG co-immunoprecipitation

The FLAG immunoprecipitation was performed as described for RqcH-FLAG^8,36 with minor modifications. 1-L cultures were grown in LB medium supplemented with 5 μg/mL chloramphenicol at 37°C, 120 rpm to OD600 ~ 1.5, then cells were harvested, frozen in liquid nitrogen and stored at −80°C. Cells were disrupted on ice by sonication (10 cycles 30 s each; 100% amplitude) in lysis buffer containing 50 mM Tris-HCl pH 7.4 (4°C), 100 mM NaCl, 10 mM MgCl₂, 5% glycerol, 0.1% NP-40, 1 mM DTT and 1x protease inhibitor (Roche #04693159001). Lysates were clarified by centrifugation at 17,000 rpm in an SS-34 fixed-angle rotor for 22 min at 4°C, and the supernatant was incubated for 1.5 h at 4°C on a turning wheel (Labinco) with 100 μL pre-washed anti-FLAG M2 Affinity Gel (Sigma-Aldrich #A2220). Beads were collected by centrifugation, and after 3 washes with 20 mL lysis buffer were transferred to a Mobicol ‘F’ filter with a 35-mm pore size (MoBiTec) and washed by gravity flow. Proteins were eluted with 200 μL lysis buffer containing 0.2 mg/mL 3X FLAG-peptide (Sigma-Aldrich #F4799) for 45 min at 4°C on a turning wheel. For gel analyses, eluates were precipitated with 10% TCA, washed twice with acetone and separated on NuPAGE™ 4%–12% Bis-Tris gels (Invitrogen, USA). Gels were stained with Coomassie InstantBlue™ (Expedeon).

Sucrose gradient analysis

200 mL cultures were grown until O.D₆₀₀=0.5, then harvested by filtration on nitrocellulose membrane (0.2 μM pore size, Bio-Rad) and snap-frozen in liquid nitrogen. When stated, erythromycin was used at 0.04 μg/mL for 1h. Cells were pulverized using a mixer mill and consequently lysed with lysis buffer containing 100 mM KCl₂, 12 mM MgCl₂, 5 mM CaCl₂ 50 mM Tris-HCl pH 7.4 (4°C), 0.1% NP-40, 0.4% TritonX-100, 20 U/ml DNase I (Thermofisher #89836), 1 mM chloramphenicol, 1 mM DTT. Lysates were clarified through centrifugation (20000 x g, 10 min at 4°C). When specified, lysates were treated with MNase (150U/OD) for 5 min at 25°C, then the reaction was quenched with 6 mM EGTA. 7-15 OD equivalents were loaded onto 5-45% sucrose gradients (100 mM KCl₂, 12 mM MgCl₂, 50 mM Tris-HCl pH 7.4 at 4°C, 1 mM chloramphenicol, 1 mM DTT), and centrifuged for 2.5 h at 35000 RPM in a SW40 rotor (Beckman-Coulter).

Gradients were fractionated into 40 fractions (300 μL each) using a Gradient Station (Biocomp). Fractions were collected, combined, precipitated with 10% TCA, washed twice with acetone and the dried pellet was resuspended in Laemmli sample buffer 1X for running in SDS-PAGE and analyzing by western blot or Coomassie staining.

Recombinant Pirh2 Purification

GST-Pirh2 (1-195) was purified as previously described³⁹, with modifications. For protein expression, E. coli BL21 (DE3) cells transformed with pET15b-GST- Pirh2 (1-195) were grown in LB media supplemented with ampicillin (100 μg/mL) until OD ~0.6 at 37°C, at which point IPTG (0.75 mM) was added and cells were further grown overnight at 18°C. For protein purification, cells were lysed in chilled buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM DTT, 10 μM ZnCl₂, 1% Triton X-100 and EDTA-free protease inhibitor (Roche #04693159001) using sonication. Clarified lysate was incubated on a rotator at 4°C for 2 hours with 500 μL of glutathione agarose bead slurry pre-washed twice with wash buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM DTT, 10 μM ZnCl₂). After incubation, beads were washed 3 times using wash buffer. GST-Pirh2 was eluted with elution buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM DTT, 10 μM ZnCl₂, 10 mM reduced glutathione). In the end, glutathione was removed from purified protein by buffer exchange with wash buffer using Amicon Ultra-4, 10 kDa filters.

Ala-PIRHification (Pirh2 pulldown)

20 mL B. subtilis cultures expressing episomal or genome-integrated GFP-ns were grown in LB medium supplemented with 5 μg/mL chloramphenicol at 37°C (160 rpm), treated with 40 μM bortezomib (Alfa Aesar #J60378) for 3.5 h and harvested as above. Modified GFP reporters were pulled down using purified recombinant GST-Pirh2 (1-195). Pre-washed Gluathione agarose beads (10 μL per sample) were blocked with assay buffer (20 mM Tris-HCl pH 7.5, 100 mM NaCl, 0.1% Tween-20) containing 1% BSA for 1 h at 4°C. Beads were washed, incubated with purified GST-Pirh2 (1-195) (10 μg protein per 10 μL beads) for 2 h at 4°C, then washed to remove excess unbound GST-Pirh2 (1-195). In the meantime, cells treated with bortezomib (40 μM, 3.5 h) were disrupted with a FastPrep-24™ 5G instrument (MP Biomedicals) in lysis buffer containing 50 mM Tris-HCl pH 7.4 (4°C), 100 mM NaCl, 5 mM MgCl₂, 5% glycerol, 0.1% NP-40, 1 mM DTT and 1x protease inhibitors (Roche #04693159001). Afterwards, 10 μL beads bound with GST-Pirh2 (1-195) were added to 1.5 mL tubes containing 200 μL of each pre-cleared lysate and incubated at 4°C for 2 h. Finally, beads were washed four times in 500 μL assay buffer (1000 x g, 2 min at 4°C). For analysis of pulled down products, beads were boiled in Laemmli buffer for 10 min at 90°C and used for immunoblot.

Mass spectrometry: LC-MS/MS and protein identification

For label-free analysis of MutS2 interactors, three replicate protein mixture samples of the empty vector control and FLAG-MutS2 IP were brought to pH 8 via the addition of 50 mM ammonium bicarbonate, and digested with 1:50 (w:w) trypsin (Pierce Biotechnology, Rockford, IL) for 1 h at 50 °C using ProteaseMax™ Surfactant trypsin enhancer following reduction and alkylation with dithiothreitol and iodoacetamide, respectively, according to the manufacturer’s instructions (Promega Corporation, Madison, WI). LC-MS/MS analysis of extracted peptides was carried out using an Orbitrap Fusion Tribrid mass spectrometer as previously described⁸, following sample clean-up with a 2 μg capacity ZipTip C18 (Millipore, Billeric, MA) according to the manufacturer’s instructions, with the exception that separation was performed on as EASY PepMap™ RSLC C18 column (2 μm, 100Å, 75 μm x 50 cm, Thermo Scientific, San Jose, CA), and ions were created with an EASY Spray source (Thermo Scientific, San Jose, CA) held at 50 °C using a voltage of 1.8 kV. The gradient was also modified to deliver solvent B (80/20 acetonitrile/water, 0.1% formic acid) from 5-25% in 45 min, followed by 25-44% solvent B in 15 min, 44-80% solvent B in 0.10 min, a 10 min hold of 80% solvent B, a return to 5% solvent B in 5 min, and finally with another 5-min hold of 5% solvent B. The gradient was then extended for the purpose of cleaning the column by increasing solvent B to 98% in 3 min, a 98% solvent B hold for 10 min, a return to 5% solvent B in 3 min, a 5% solvent B fold for 3 min, an increase of solvent B to 98% in 3 min, a 98% solvent B hold for 10 min, a return to 5% solvent B in 3 min and a 5% solvent B hold for 3 min and finally, another increase to 98% solvent B in 3 min and a hold of 98% solvent B for 10 min. All flow rates were 250 nL/min delivered using an nEasy-LC1000 nano liquid chromatography system (Thermo Fisher Scientific, San Jose, CA). Solvent A was 0.1% formic acid.

For the bL9-FLAG IP, TMT labeling was performed. Samples were precipitated overnight using 4x ice-cold acetone (v:v). Dried pellets were resolubilized in 92 μL of 5% SDS and processed for trypsin digestion using micro S-Traps™ (Protifi, Huntington, NY) according to the manufacturer’s instructions at a ratio of 1:10 trypsin:protein (w:w). Following 1 h digestion at 47 °C, peptides were eluted from the S-Traps™ and dried under vacuum. Peptides were then resolubilized in 30 μL of 50 mM triethyl ammonium bicarbonate (TEAB), reacted with TMT labels (16-plex) according to the manufacturer’s instructions (Thermo Fisher Scientific, Waltham, MA), and pooled. The 16-plex labeling matrix is available upon request. The pooled plexed samples were dried under vacuum, resolubilized in 200 μL of 1% trifluoroacetic acid (TFA), and desalted using a C18 ZipTip according to the manufacturer’s instructions (Millipore, Billerica, MA). Dried TMT-labelled peptides were reconstituted in 5 μL of 0.1% TFA and mass spectrometry was performed as described previously with the exception that ions were created with an EASY Spray source (Thermo Scientific, San Jose, CA ) held at 50 °C using a voltage of 2.3 kV⁴⁶.

Proteomic Data Processing and Statistical Analysis

For both label-free and TMTpro 16-plex experiments, quantitative analyses were performed simultaneously with protein identification using Proteome Discoverer 2.5 (PD) software. For TMTpro, Precursor and fragment ion mass tolerances were set to 10 ppm and 0.6 Da, respectively, whereas mass tolerances were 10 ppm and 0.02 Da for label-free searches. For both experiments, SEQUEST searches used a maximum of 2 missed tryptic cleavages, the Uniprot Bacillus subtilis proteome (UP000001570 downloaded in 2020) and FASTA files of common contaminants. Impurity correction factors obtained from Thermo Fisher Scientific for the TMTpro 16-plex kit were included in the search and quantification. The following settings were used to search the label-free data: dynamic modifications; oxidation / +15.995Da (M), deamidated / +0.984 Da (N, Q), and static modifications of MMTS / +45.988 Da (C). For TMTpro searches same dynamic and statistic modifications were used with addition of static modification of TMTpro / +304.207 Da (N-Terminus, K). Only unique+ Razor peptides were considered for quantification purposes. The Percolator feature of PD was used to set a false discovery rate (FDR) of 0.01. Samples were normalized by volume due to expected global changes in total amount of protein per sample.

For label-free experiments, proteins that were detected in at least one treatment group in all replicates were included. The mean of summed abundances of intensity values for each group was used to calculate ratios. For the TMTpro experiment, the Protein Abundance Based method was used to calculate protein level values. For both label-free and TMTpro experiments, contaminant proteins were excluded from the data set and treatment groups were compared to control using pooled variances t-test method in JMP Pro (SAS) 15.2.1 to identify proteins that were differentially present. Adjusted p-values were used to set FDR to 0.05 using the Benjamini Hochberg method. For volcano plots, since a few proteins were not identified or quantified in the empty vector groups, 1 was added to the denominator for each data point.

Immunoblotting

Immunoblotting procedures were performed as previously described³⁶. 20 mL cultures were grown in LB medium supplemented with 5 μg/mL chloramphenicol at 37°C, 160 rpm to a final OD₆₀₀ of ~1, then cells were harvested as above. Cells were disrupted with a FastPrep-24™ 5G instrument (MP Biomedicals) in lysis buffer containing 50 mM Tris-HCl pH 7.4 (4°C), 100 mM NaCl, 5 mM MgCl₂, 5% glycerol, 0.1% NP-40, 1 mM DTT and 1x protease inhibitors (Roche #04693159001). Proteins were separated on a NuPAGE™ 4%–12% Bis-Tris gel (FLAG immunoblot) or GenScript 4%-20% Bis-Tris (GFP immunoblot) and transferred to a 0.45 mm PDVF membrane using the Invitrogen Mini Blot Module (Invitrogen, USA). Membranes were sequentially incubated with monoclonal anti-FLAG M2 antibody (Sigma-Aldrich #F1804), mouse monoclonal anti-GFP antibody (Roche #11814460001), goat polyclonal anti-GST (Cytiva #27-4577) or rabbit polyclonal anti-FtsZ antibody (Sigma-Aldrich #ABS2200) and HRP-conjugated secondary antibodies. To develop the signal, membranes were treated with enhanced chemiluminescence substrate kit (ECL, GE Healtcare). Images were acquired with a Amersham ImageQuant 600 (Cytiva) or a GE ImageQuant LAS 4000 instrument and analyzed using Fiji⁴⁷.

Growth assays

Strains were grown at 37°C in LB medium supplemented or not with 5 μg/mL chloramphenicol to OD₆₀₀ = 1.0-1.5, then diluted to OD₆₀₀ = 0.1 and 10-fold serial dilutions spotted onto LB agar plates containing or not the indicated drugs and grown for 16 h at 37°C. For complementation experiments, LB agar plates were supplemented with 5 μg/ml chloramphenicol to allow plasmids selective pressure.

For competitive growth assays, cultures were grown until OD₆₀₀ ~ 0.4, diluted and mixed 1:1 to a final OD₆₀₀ of 0.1 and then grown for 20 or 40 doublings in 2 separate, independent experiments. To maintain cultures in logarithmic growth, cells were diluted to OD_{600 =} 0.1 every 5 doublings. For the 40-doubling experiment, after 15 doublings cells were diluted to OD₆₀₀ =0.005, grown overnight (10 doublings estimated), diluted again and kept in logarithmic growth for additional 15 doublings. At the end, cultures were plated and 46 individual colonies were genotyped by colony PCR for each condition.

Cryo-EM grid preparation and data acquisition

Glow-discharging of holey carbon grids (Cu R2/1; 200 mesh; Quantifoil) was conducted for 20 s under oxygen atmosphere using a Solarus 950 plasma cleaner (Gatan). Using a Vitrobot Mark IV (Thermo Fisher), EM-grids were prepared at ambient temperature and humidity by applying 4 μl of sample onto the grids, blotting with Whatman filter paper Nr.1 for 3 s (blot force = 0), and plunging them into liquid ethane cooled by liquid nitrogen.

Two datasets were acquired using the EPU software package (Thermo Fisher) on a Titan Krios TEM (Thermo Fisher/FEI). Data was acquired in dose fractionation mode at 300 kV using an energy-filtered K3 camera (Gatan) operated by the Gatan Microscopy Suite (version 3.32). The defocus range was set to −0.5 μm to −1.5 μm at an object pixel size of 1.07 Å/px. Micrograph movie stacks contained 30 frames with a cumulative dose of 48 e/A² and 45.4 e/A² for the first and second dataset, respectively. In each selected hole, micrograph movies stacks were acquired at four pre-defined positions after the defocus was automatically adjusted.

Cryo-EM image processing

Image processing was performed in Relion 3.1, if not stated otherwise⁴⁸. The image processing workflow, including detailed information on the number of micrographs, the number of particles and estimated resolution of cryo-EM densities is summarized in figs. S5 and S6. Pre-processing of micrograph movie stacks was conducted separately on both datasets as follows: Micrograph movie stacks were motion-corrected based on a 5x5 grid using MotionCor2 (implemented in Relion 3.1). The contrast transfer function (CTF) was estimated using gCTF (implemented in Relion 3.1). To generate a template for autopicking in Relion, approximately 1,000 particles were manually located, extracted at a scaled pixelsize of 4.28 Å/px and box size of 128x128 px, and subjected to 2D classification. 2D classes depicting 70S ribosomes were used as input for autopicking. Autopicked particles were extracted as described above and split into multiple randomized subsets to reduce processing time. The individual subsets were subjected to an initial tier of 3D classification. 3D classes representing 70S ribosomes with high-resolution features from all subsets of the respective datasets were merged and subjected to 3D auto-refinement. Subsequent 3D classification without translational and orientational sampling yielded classes depicting 70S ribosomes with a 30S-shaped additional density at the mRNA exit site. Particles from both datasets included in these classes were merged and subjected to another round of 3D auto-refinement focused on the central 70S ribosome, followed by 3D classification without sampling using a larger spherical mask (425 Å diameter). Only 3D classes depicting a second 70S ribosome with high-resolution features were retained. This set of particles was subjected to another round of auto-refinement focused on the central 70S ribosome and is in the following referred to as “Collided Disomes”.

For higher resolution analyses, the particles were extracted at a scaled pixelsize of 1.604 Å/px and with a box size of 368x368 px. The large size of the disome complex in conjunction with slight conformational plasticity in between the two individual ribosomes posed a challenge for structure refinement, which we overcame by extracting particles either centered on the leading or the collided ribosome. After extraction, the particles were subjected to 3D auto-refinement, CTF refinement (including beam tilt estimation, per-particle defocus estimation and anisotropic magnification estimation) and Bayeshian Polishing (performed in Relion 3.0). The resulting particles were subjected to four separate runs of 3D auto-refinement, focused on the 30S subunit or the full 70S of the two individual ribosomes, respectively. For interpretation of the disome interface, the four cryo-EM densities were mapped back into a common coordinate system.

For analysis of the MutS2 structure, particles contained in the ‘Collided disomes’ set were extracted centered on the leading ribosome at a scaled pixelsize of 4.28 Å/px and at a box size of 128×128 px. These particles were subjected to an initial round of 3D auto-refinement, followed by a focused 3D classification without sampling using a mask encompassing the Clamp-, Lever and ATPase-domain of both MutS2 dimers. This allowed initial separation of MutS2-containing and -lacking particles and both subsets of particles were extracted at a scaled pixelsize of 1.604 Å/px and with a box size of 368×368 px and subjected to 3D auto-refinement, CTF refinement, Bayesian Polishing and another round of 3D auto-refinement centered on the leading ribosome, as described above. Within the resolution limit, the resulting cryo-EM densities of MutS2-containing and -lacking disome complexes were indistinguishable except for MutS2 density, as judged by density subtraction. Subjecting the MutS2-containing set of particles to an additional tier of 3D classification without sampling focused on MutS2 allowed to reject remaining MutS2-lacking particles and revealed two different conformations of the MutS2 dimer (referred to as “Conf1” and “Conf2”) that were mostly distinct in the orientation of the MutS2-B clamp domain. Particles corresponding to the two individual conformations were subjected separately to 3D auto-refinement centered on the leading ribosome, obtaining the final cryo-EM densities for MutS2 conformations. At the same time, particles corresponding to both conformations were merged and subjected to 3D auto-refinement centered on both the leading and collided (after re-extraction) ribosomes to obtain the final cryo-EM density of the full set of MutS2-containing particles that was best suited for analysis of the interaction between the leading ribosome and the MutS2-A clamp domain.

To obtain a cryo-EM structure of the 50S subunit obstructed with a nascent chain-linked P-site tRNA, the 50S-depicting particles from dataset 1 were re-extracted at a scaled pixelsize of 1.604 Å/px and with a box size of 368×368 px and subsequently subjected to 3D auto-refinement.

To enrich for particles that depict collided trisomes, particles from collided disomes centered on the collided ribosome were subjected to 3D classification focused on the third ribosome. Retained particles were re-extracted centered on the third ribosome (Collided 2) at a scaled pixelsize of 4.28 Å/px and at a box size of 128×128 px and subjected to 3D auto-refinement.

Atomic model of the collided disome

The atomic model of the collided disome was assembled in a stepwise manner: Using USCF Chimera ⁴⁹, the atomic model of a translating B. subtilis 70S ribosome (PDB: 6HA1) ⁵⁰ and the SD-helix of a stalled B. subtilis 70S ribosome (PDB: 3J9W) ³¹ were docked as rigid bodies into the respective segments of the cryo-EM densities, which were refined using a mask encompassing the complete 70S ribosome. Extra ribosomal densities could be identified as bS21 and bL9 by comparison to atomic models of E. coli 70S ribosomes. Homology models for B. subtilis bS21 and bL9 were generated based on a cryo-EM structure of the S. aureus 70S ribosome (PDB 5NGM, bS21) ⁵¹ and an X-ray structure of G. stearothermophilus bL9 (PDB 1DIV) ⁵² using HHpred ⁵³. The derived homology models for bS21 and bL9 were docked as rigid bodies using USCF Chimera.

For atomic modeling and model refinement, we generated a composite density from the four individual cryo-EM reconstructions using the Phenix functionality “Combine Focused Maps”⁵⁴ in conjunction with the atomic model of the collided disome after rigid body docking. In some regions, homology models of the identified proteins were adjusted using Coot ⁵⁵. These adjustments included, the N-terminal and C-terminal interaction sites of bL9 with the respective rRNA of the leading and collided ribosome, the curvature of the alpha-helix of bL9, the L1 stalk of the leading ribosome and the extension of ribosomal proteins uS7 (C-terminus, 2 residues) and uS11 (N-terminus, 4 residues). The model was refined iteratively in ISOLDE ⁵⁶ and Phenix and validated in Phenix ⁵⁴ using the composite density.

Atomic model of the MutS2 homodimer

First, a homology model for the B. subtilis MutS2 monomer was created based on the AlphaFold model ⁵⁷ for S. aureus MutS2 (UniProt Q2FZD3) using ‘one to one threading’ in Phyre2 ⁵⁸. The SMR and KOW domains that were not resolved in the cryo-EM reconstructions, as well as the coiled coil helix, were deleted from the monomer. Using USCF Chimera, two copies of the resulting atomic model were docked as rigid bodies into the local resolution-filtered cryo-EM density of the MutS2 homodimer in conformation 2. To remove structural clashes, the atomic model of the MutS2 homodimer was refined in real space using molecular dynamics flexible fitting (MDFF) performed in VMD v1.9.3 ⁵⁹. For preparation of the model, the plugin QwikMD was used ⁶⁰. To generate the namd configuration file the plugin MDFF v0.4 was used ⁶¹. MDFF was run using NAMD v2.14 ⁶² with 2000 minimization steps, 20000 simulation steps in vacuum and the application of a grid force of 0.3. The output atomic model was subjected to a real space refinement with 1 macro cycle using Phenix, generating an initial MutS2 homodimer model.

To improve the MutS2 dimerization interface and to accurately position the MutS2 coiled-coil helices, we generated a model for dimeric B. subtilis MutS2 ATPase and coiled-coil domains using AlphaFold-Multimer (https://github.com/deepmind/alphafold) and superposed it to the initial MutS2 homodimer model by the ‘matchmake’ command in UCSF Chimera. The two models were fused at residues Lys293 and Val289 for MutS2-A and MutS2-B, respectively, and the fusion sites were adjusted in Coot⁵⁵. The dimeric coiled-coil segment (Ala514-Glu600) was already overlapping well with the cryo-EM density after initial placement and had to be only slightly adjusted as a rigid body to optimize the fit. To remove structural clashes, the atomic model was refined in ISOLDE⁵⁶ around regions with increased clashscore. Due to low local resolution, regions Asp215-Leu223 of MutS2-A and Arg177-Ile230 of MutS2-B were removed from the fused atomic model.

Analysis of cryo-EM densities and atomic models

Localization of the SMR and KOW domains: In an attempt to locate a specific binding site for the SMR domain on the collided disome, we masked out cryo-EM density explained by the modelled segments of MutS2 and the collided disome (including tRNAs and mRNA), but could not observe any obvious additional density likely to correspond to the SMR domain. Similarly, we subtracted cryo-EM densities of the MutS2-containing and -lacking collided disomes from each other, again not observing any significant difference density except for the modelled MutS2 segments.

Steric exclusion of elongation factor - G (EF-G) by the bL9 C-terminus: The atomic model of an elongating bacterial 70S ribosome with EF-G bound in the GTP conformation (PDB 7N2V)⁶³ was docked into the cryo-EM density of the collided ribosome’s 30S subunit (‘collided disomes’: 30S collided ribosome) using UCSF Chimera⁴⁹. The EF-G density was simulated using Chimera’s “Molmap” functionality at a resolution of 5 Å.

Conformational plasticity of the A-site finger: Two atomic models for different states of the bacterial translational elongation cycle (PDB 6WDB and PDB 6WDG)⁶⁴ as well as our atomic model for the leading ribosome were aligned with respect to ribosomal protein uL5 using UCSF Chimera’s “Matchmaker” functionality⁴⁹.

Presentation of cryo-EM densities and atomic models

All cryo-EM densities and atomic models were visualized using Chimera⁴⁹ or ChimeraX⁶⁵. Data was plotted using Excel. Figures were compiled using Inkscape.

Extended Data

Extended Data Fig. 1. Cryo-EM data processing scheme.

a, Two datasets were acquired and processed separately using Relion. To reject false-positive particles, auto-picked particles were subjected to a 3D classification. A subset of 50S depicting particles from dataset 1 was subjected to 3D auto-refinement. b, To enrich for collided disomes, 70S depicting particles were subsequently subjected to multiple rounds of 3D classification using a spherical mask with a diameter of 435 Å (dataset 1) or focused on the 30S subunit of the collided disome (dataset 2). After the first round of classification particles were merged. c, Particles of collided disomes were re-extracted centered on the leading and the colliding ribosome. 3D auto-refinements were focused on the 30S subunit or the full 70S of the respective ribosome. d, MutS2 conformations were sorted by two rounds of 3D classification using a mask encompassing the MutS2-homodimer. To increase local resolution for the MustS2-A clamp domain, the two separate conformations were merged and subjected to 3D auto-refinements based on either the leading or the colliding ribosome. e, To enrich for particles that depict collided trisomes, particles from collided disomes centered on the collided ribosome were subjected to a 3D classification focused on the third ribosome. Retained particles were re-extracted centered on the third ribosome (Collided 2) and subjected to 3D auto-refinement.

Extended Data Fig. 2. Structural similarity between the predicted MutS2 structure and the X-ray structure of a bacterial MutS protein.

a, Overall structures of B. subtilis MutS2 (this study) and Neisseria gonorrhoeae MutS (PDB 5X9W). Segments of MutS2 that were not resolved by cryo-EM, including the KOW and SMR domains, are shown in grey for one monomer as predicted by AlphaFold. b-c, Superposition of the two models with respect to the ATPase domain (b), the individual MutS clamp and lever domains (c,d) and all conserved segments of the individual monomer structures (e,f). Dashed boxes indicate the zoomed area for panels b-f.

Extended Data Fig. 3. Global and local resolution estimation.

For each 3D auto-refinement run, densities have been colored according to local resolution. Mask-corrected Fourier shell correlation (FSC) has been calculated between two independently refined half-sets of the data (‘gold standard’; FSC threshold at 0.143). Shown particle sets are “Collided Disomes” (a, b, c, d,), “+MutS2” (g, h), MutS2 conformation 1 and 2 (respectively e, f) and 50S subunits obstructed with a nascent chain-linked P-site tRNA (i). Particles are either centered on the leading (a, b, e, f, g) or on the colliding ribosome (c, d, h) and refined using either a 30S (a, c) or a 70S mask of the respective ribosome (b, d, e, f, g, h).

Extended Data Fig. 4. Molecular details of contacts stabilizing the disome interface.

a, The 30S subunit heads interact via complementary charged patches on uS9 of the leading ribosome and uS10 of the collided ribosome (‘head contact’). Surface representations for uS9 and uS10 are shown and colored according to electrostatic potential (blue: positive, red: negative, white: neutral). The interacting patches are indicated. b, On the opposing side of the inter-ribosomal mRNA trajectory, helix 25 of the leading ribosome 30S subunit rRNA is accommodated in a groove on the 30S subunit of the collided ribosome formed by uS2 and uS8 (‘body contact’). In particular, helix 25 of the leading ribosome directly interacts with a surface exposed α-helix (Leu43 - Glu63) and the partially unordered C-terminus of uS2 of the collided ribosome. c, In close proximity, two additional contacts between the two collided ribosomes (‘platform contacts’) complete the network of interactions clustered around the mRNA entry and exit sites. First, the L1-stalk adopting an extreme out-conformation on the leading ribosome directly contacts the 30S subunit rRNA of the collided ribosome. Second, uS11 of the leading ribosome contacts uS4 of the collided ribosome, mainly via hydrophobic interactions (Val14 and Ile18 of S11; Val157 and Gly23 of S4) and aromatic stacking (His40 of S11; Phe160 of S4). d, A more peripheral interaction is mediated by ribosomal protein bL9 of the leading ribosome (‘bL9 contact’). e, The binding site of bL9 on the B. subtilis 50S subunit has been significantly remodeled compared to the E. coli ribosome⁶⁶ (PDB 6BY1). While the interaction area between bL28 and the N-terminal half of bL9 is reduced in B. subtilis, this is compensated by an expansion of the 50S subunit rRNA (helix 15), which together with the L1-stalk forms an extended binding groove for bL9.

Extended Data Fig. 5. Structural and compositional remodeling of the mRNA exit site on the leading ribosome.

a, Upper panel: Trajectory of unstrained mRNA exiting the mRNA channel in a defined direction to interact with the anti-Shine-Dalgarno (SD) rRNA sequence of the 16S rRNA 3’end, thereby forming an RNA duplex reminiscent of the SD helix during translation initiation³¹ (PDB 3J9W, EMDB 6306). Middle panel: The unstrained mRNA exiting the collided ribosome interacts with the anti-SD rRNA to form an RNA duplex. Bottom panel: In the collided disome, the mRNA under strain follows a vastly different trajectory, which is accompanied by remarkable structural remodeling of the mRNA exit site of the leading ribosome. In particular, the rRNA anti-SD sequence of the leading ribosome can no longer interact with the mRNA and becomes disordered, which renders the binding site for the bS21 C-terminus on the leading ribosome accessible and at the same time reduces the interaction surface for uS2 on the 30S subunit. Superposition: Superposition of mRNAs exiting the ribosomes and the respective SD-helices in the translating and collided ribosomes, as well as bS21 in the leading ribosome. b, Comparison of uS2 density in the leading and collided ribosomes at the same density threshold level. In all panels, local resolution filtered densities based on 3D auto-refinements focused on either the 30S or the 70S of the respective ribosome are shown (See Extended Data Fig. 1).

Extended Data Fig. 6. The C-terminal half of bL9 sterically excludes binding of EF-G on the collided ribosome.

a, Binding site of the bL9 on the 30S subunit of the collided ribosome. b, Atomic model and simulated density of EF-G⁶³ (PDB 7N2V) mapped onto the 30S subunit of the collided ribosome by fitting the 30S-EF-G complex as a rigid body. Overlapping segments of bL9 and EF-G are shown in transparent grey. c, As in ‘b’, but not showing the EF-G atomic model and with overlapping segments of bL9 and EF-G colored in purple.

Extended Data Fig. 7. Conformational plasticity of the MutS2-B clamp region.

a, Two views on the local resolution-filtered density of the MutS2 dimer after 3D auto-refinement of all MutS2-containing particles. Highly fragmented density for the MutS2-B clamp and lever domains (left panel) indicated conformational heterogeneity. b, c, Computational particle sorting focused on MutS2-B produced two structurally distinct subpopulations slightly differing in the positioning of the MutS-domains III and IV, in which the clamp region either binds to ribosomal protein L5 (b) or the nascent chain-associated P-site tRNA of the leading ribosome (c). d, Overlay of the two conformations from ‘b’ and ‘c’. In all panels, local resolution-filtered cryo-EM densities are shown as obtained after 3D autorefinement centered on the leading ribosome using either all MutS2-containing particles (ribosome densities in all panels, MutS2 density in panel A) or using only particles representing one of the two different MutS2 conformations (MutS2 density in panels b, c).

Extended Data Fig. 8. Conformational plasticity of the A-site finger during the translational elongation cycle.

a-c, Atomic models for rRNA, tRNAs and ribosomal protein uL5. a, The leading ribosome of the B. subtilis collided disome. b, The E. coli ribosome in accommodation state IV-A⁶⁴ (PDB 6WDB). c, The pre-translocation state VI-B⁶⁴ (PDB 6WDG). d, Superposition of the structures shown in (a-c) according to ribosomal protein uL5, demonstrating remodeling of the MutS2-A binding site during the translation elongation cycle.

Extended Data Fig. 9. Structure of the 50S ribosomal subunit obstructed with a nascent chain-linked P-site tRNA.

a, Local resolution-filtered cryo-EM density of the 50S ribosomal subunit obstructed with the nascent chain-linked P-site tRNA. Due to conformational flexibility, cryo-EM density for peripheral segments of the P-site tRNA is fragmented. b, Density was sliced open to allow for an unobstructed view on the P-site tRNA and the associated incomplete nascent chain. c, Model of a nascent chain-linked P-site tRNA³⁶ (PDB 7AQC) superposed to the cryo-EM density. Zoom on the CCA-tail of the P-site-tRNA with linked nascent chain.

Extended Data Fig. 10. Evidence from genomic analyses link mutSB and rqcH.

a, mutSB and rqcH strongly co-occur. Distribution of rqcH and mutSB quantitated separately for the indicated bacterial phyla. Both absolute numbers and frequencies are presented. For the latter, the higher the frequency the darker the background red color. b, mutSB localizes in the vicinity of rqcH in diverse bacteria. Genes are represented by arrows reflecting the direction of transcription, with mutSB and rqcH indicated in blue and red, respectively. In the instances where mutSB and rqcH are separated by genes represented as grey arrows, those genes are highly diverse, have no obvious relationship to translational quality control, and are generally unrelated between different species.

Extended Data Fig. 11. Model for MutS2 function in sensing ribosome collisions and eliciting downstream responses.

The model depicts, from top to bottom: ribosomes translating an mRNA with a stalling site within the open reading frame (“Translation”); the leading ribosome becoming stalled (“Internal stalling”); a trailing ribosome colliding with the stalled ribosome (“Ribosomal collision”); MutS2 sensing the collision and promoting both separation of the ribosomal subunits (“Ribosome splitting”) and endonucleolytic cleavage (“mRNA cleavage”). Left side: Ribosomal splitting generates a 50S subunit still obstructed with a nascent chain-tRNA conjugate, which is sensed by RqcH and RqcP, resulting in elongation of the nascent chain with a C-terminal Ala tail (“Ala tailing”). Nascent-chain release is accompanied by 50S recycling (“Ribosome recycling”, dotted line). Right side: mRNA cleavage can result in mRNA decay (dotted line) or in trailing ribosomes becoming stalled at the mRNA 3’end, which are sensed by SsrA/tmRNA. The SsrA reaction leads to ribosome recycling (“Ribosome recycling”, dotted line) and to nascent-chain modification with a C-terminal SsrA tag (“SsrA tagging”). Both Ala-tails and the SsrA tag act as degrons, recognized by ClpXP and other proteases (“Proteolysis”). See the main text for additional details. Objects: the mRNA is shown in red, with a stalling site within the open reading frame represented by ‘!’ within a triangle and the mRNA stop codon shown as a ‘Stop’ traffic sign; the direction of translation is indicated by arrows; the stalled ribosome is shown in orange (light, 50S subunit; dark, 30S subunit); trailing and collided ribosomes are shown in green (dark, 50S subunit; light, 30S subunit). Quality control factors are indicated by their names.

Extended Data Table 1.

Cryo-EM data collection, refinement and validation statistics

	Leading Ribosome (EMDB-14157) (PDB 7QV1)	Collided Ribosome (EMDB-14157) (PDB 7QV2)	MutS2+ (Conf. 1 / 2) (EMDB-14156, EMDB-14159) (PDB 7QV3)	MutS2+ (Conf. 1+2) (EMDB-14160, EMDB-14161)
Data collection and processing	70S/30S	70S/30S	MutS2 Conf.1/MutS2 Conf.2	Leading 70S / Collided 70S
Magnification	81,000	81,000	81,000	81,000
Voltage (kV)	300	300	300	300
Electron exposure (e−/Å2)	Dataset1=48	Dataset1=48	Dataset1=48	Dataset1=48
	Dataset2=45	Dataset2=45	Dataset2=45	Dataset2=45
Defocus range (μm)	−0.75 to −2.0	−0.75 to −2.0	−0.75 to −2.0	−0.75 to −2.0
Pixel size (Å)	1.07	1.07	1.07	1.07
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	8,297,591	8,297,591	8,297,591	8,297,591
Final particle images (no.)	27,833	27,833	3,645 / 5,078	8,723
Map resolution (Å)	3.38 / 3.56	3.43 / 3.47	5.68 / 5.14	4.22 / 4.92
FSC threshold	0.143	0.143	0.143	0.143
Map resolution range (Å)	3.2–6.0 / 3.2–6.5	3.2–6.0 / 3.2–6.5	4.0–13.0 / 4.0–13.0	4.0–10.0 / 4.0–10.0
Refinement	Composite map	Composite map	MutS2 Conf. 2: (Leading 70S)	Not modeled
Initial model used (PDB code)	6HA1, 5NGM, 1DIV	6HA1, 3J9W	6HA1, 5NGM, 1DIV, AlphaFold Q2FZD3
Model resolution (Å)	3.2	3.2	3.6
FSC threshold	0.143	0.143	0.143
Model resolution range (Å)	3.2-3.5	3.2-3.5	3.7-5.3
Map sharpening B factor (Å2)	−23 / −30	−25 / −25	24 / −98
Model composition
Non-hydrogen atoms	140,774	142,887	149,709
Protein residues	5,278	5,292	6,419
Nucleotides	4,635	4,726	4,635
B factors (Å2)
Protein	111.67	114.08	161.28
Nucleotide	109.27	121.38	109.27
R.ms. deviations
Bond lengths (Å)	0.003	0.003	0.004
Bond angles (°)	0.596	0.601	0.709
Validation
MolProbity score	1.70	1.66	1.66
Clashscore	5.90	5.95	5.59
Poor rotamers (%)	0.41	0.25	0.47
Ramachandran plot
Favored (%)	94.56	95.13	94.89
Allowed (%)	5.35	4.75	4.94
Disallowed (%)	0.10	0.12	0.17

Data availability

Cryo-EM densities have been deposited in the Electron Microscopy Data Bank under accession codes EMD-14163 [https://www.ebi.ac.uk/emdb/EMD-14163] (‘collided disomes’: 30S leading ribosome), EMD-14162 [https://www.ebi.ac.uk/emdb/EMD-14162] (‘collided disomes’: 70S leading ribosome), EMD-14165 [https://www.ebi.ac.uk/emdb/EMD-14165] (‘collided disomes’: 30S collided ribosome), EMD-14164 [https://www.ebi.ac.uk/emdb/EMD-14164] (‘collided disomes’: 70S collided ribosome), EMD-14157 [https://www.ebi.ac.uk/emdb/EMD-14157] (‘collided disomes’: composite density), EMD-14166 [https://www.ebi.ac.uk/emdb/EMD-14166] (Dataset 1: 50S ribosome with nascent chain-linked P-site tRNA), EMD-14160 [https://www.ebi.ac.uk/emdb/EMD-14160] (‘+MutS2’: 70S leading ribosome), EMD-14161 [https://www.ebi.ac.uk/emdb/EMD-14161] (‘+MutS2’: 70S collided ribosome), EMD-14156 [https://www.ebi.ac.uk/emdb/EMD-14156] (‘+MutS2 Conformation 1’: 70S leading ribosome), EMD-14159 [https://www.ebi.ac.uk/emdb/EMD-14159] (‘+MutS2 Conformation 2’: 70S leading ribosome). Atomic coordinates have been deposited in the Protein Data Bank under accession codes PDB-7QV1 [https://www.rcsb.org/structure/7QV1] (‘collided disomes’: 70S leading ribosome), PDB-7QV2 [https://www.rcsb.org/structure/7QV2] (‘collided disomes’: 70S collided ribosome) and PDB-7QV3 [https://www.rcsb.org/structure/7QV3] (‘+MutS2 Conformation 2’: 70S leading ribosome bound to MutS2 in conformation 2). Published structural data used in this article: From Protein Data Bank: PDB-6HA1 [https://www.rcsb.org/structure/6HA1], PDB-5NGM [https://www.rcsb.org/structure/5NGM], PDB-1DIV [https://www.rcsb.org/structure/1DIV], PDB-3J9W [https://www.rcsb.org/structure/3J9W], PDB-7N2V [https://www.rcsb.org/structure/7N2V], PDB-6WDB [https://www.rcsb.org/structure/6WDB], PDB-6WDG [https://www.rcsb.org/structure/6WDG], PDB-5X9W [https://www.rcsb.org/structure/5X9W], PDB-1EWQ [https://www.rcsb.org/structure/1EWQ]. From AlphaFold Protein Structure Database: Q2FZD3 [https://alphafold.ebi.ac.uk/entry/Q2FZD3].