Preparation of cotranslational folding intermediates

As a model for studying protein biogenesis, we used E. coli DHFR, an essential oxidoreductase enzyme with close homologs in all domains of life35. DHFR is a single-domain monomeric protein comprising 159 amino acids (aa), with a central discontinuous eight-stranded β-sheet and four flanking α-helices (Fig. 1a,c). During refolding from denaturant in vitro, DHFR undergoes global collapse on the microsecond timescale, and the order of strands in the β-sheet motif is established within 6 ms36. Importantly, folding of DHFR is substantially faster than its synthesis on the ribosome, which would require ~8–16 s (~10–20 codons translated per second37). Stalling the ribosome is therefore not expected to distort NC conformational sampling that occurs during normal translation.

Fig. 1: Design of DHFR RNCs.

a, Structure of E. coli DHFR (PDB 5CCC). b, Schematic illustrations of stall-inducing constructs. The stall site is indicated in red, and subdomains are colored gold (discontinuous loop subdomain, DLD) or bronze (adenosine binding subdomain, ABD), with the artificial linker in FL + 58RNC in black. The region of the NC expected to span the exit tunnel is indicated. PTC, peptidyl transfer center. c, Secondary structure elements in DHFR, colored according to subdomains as in b. Sites where the stall sequence was inserted are marked with arrows.

To sample DHFR cotranslational folding intermediates along the pathway of vectorial synthesis, we prepared stalled ribosome–NC complexes (RNCs) representing snapshots of folding in vivo (Fig. 1b)38,39,40. The NC sequence was truncated at the boundaries between discrete structural motifs, also considering the annotation of DHFR 'subdomains' (adenosine binding subdomain, ABD (residues 38–106); discontinuous loop subdomain, DLD (residues 1–37 and 107–159)). Translation was stalled by encoding, C-terminal to each DHFR fragment, an 8 aa ribosome stall-inducing sequence that is resistant to folding-induced release41. As a control, we prepared a construct consisting of the complete DHFR sequence separated from the stall site by an unstructured C-terminal linker of 50 aa, such that the full-length DHFR is a total of 58 residues from the peptidyl transfer center (denoted FL + 58RNC). Prior force profile analyses suggested that DHFR stalled at this linker length can fold into a conformation capable of binding the inhibitor methotrexate17. Note that ~30 residues in an extended conformation are required to span the exit tunnel42.

RNCs purified from E. coli were homogeneous, stable and insensitive to puromycin, indicating that the ribosomes were accurately stalled (Extended Data Fig. 1a–f). MS confirmed the presence of the NC and all ribosomal proteins at close to the expected stoichiometries, as well as the absence of significant contamination (Extended Data Fig. 2a,b). The exception was the chaperone TF, which copurified predominantly with RNCs containing the first 106 or 126 residues of DHFR (1–106RNC and 1–126RNC; Extended Data Figs. 1c and 2c).

Purified RNCs were exposed to deuterated buffer for 10 s, 100 s or 1,000 s and the entire system was digested into peptides. The peptide mixture was separated by liquid chromatography and ion mobility, and the relative deuterium uptake of peptides from all proteins in the system was measured by mass spectrometry (Fig. 2a and Supplementary Data 1). The degree of deuterium incorporation indicates protein conformation, as backbone hydrogens are protected from exchange when involved in stable secondary structure, buried in the core of a folded protein or at a protein–protein interface43. Comparative HDX measurements are therefore a sensitive probe of the local environment of amide hydrogens. Although different factors can influence HDX rates, quantitative comparison with appropriate reference samples can readily distinguish folded versus unfolded and native versus non-native states at the peptide level44,45,46,47.

Fig. 2: HDX MS of RNCs.

a, Schematic of HDX MS experiment. Purified RNCs are deuterated for 10 s, 100 s or 1,000 s before the reaction is quenched and the entire system is digested into peptides. Peptides are resolved in solution (LC, liquid chromatography) and in the gas phase (IMS, ion-mobility spectrometry), followed by analysis of deuterium incorporation by MS. b, Relative deuterium uptake of DHFR peptides after 10 s exposure to deuterium as a percentage of the maximum possible exchange, for isolated DHFR (FL DHFR) and FL + 58RNC. A maximally deuterated control sample (FL DHFR maxD) is shown as a reference. Lines are a guide for the eye only. Data are presented as mean values of two to four replicates. Error bars, s.d.; n = 3 or n = 4 independent experiments. See also Extended Data Fig. 3 and Supplementary Data 1.

We first analyzed FL + 58RNC, in which DHFR was expected to be completely emerged from the ribosome and folded. A total of 34 peptides from the NC covering 94% of the sequence could be followed, despite the analytical complexity of the system (Extended Data Fig. 3a). For comparison, we analyzed isolated DHFR (here called FL DHFR). Deuterium uptake for DHFR peptides was almost identical between FL + 58RNC and FL DHFR (Fig. 2b, Extended Data Fig. 3b and Supplementary Data 1), indicating that DHFR, fully translated yet still coupled to the ribosome, is essentially natively folded. Two exceptions were subtle protection from exchange of peptide 63–92 in the NC, and deprotection of peptide 94–117, both apparent at longer deuterium exposure times. These differences may arise from weak interactions between the ribosome surface and folded DHFR, discussed below. Overall, these data demonstrate that our approach allows for the analysis of HDX in an extremely complex mixture and can accurately report on local NC conformation even in the background of the entire ribosomal protein complement.

Cotranslational folding pathway of DHFR

To define the folding dynamics of DHFR on the ribosome, we compared peptide-resolved HDX at different NC lengths. FL DHFR and FL + 58RNC served as fully folded references, and deuterium uptake at three exchange times provided a fingerprint for the native conformation. We initially focused on the N-terminal region including part of strand β1 in the DLD (residues 5–30), which presented a set of overlapping peptides common to all RNCs. As a representative example, deuteration of peptide 9–28 as a function of NC length and deuterium exposure time is shown in Fig. 3a. Note that the same behavior was observed in multiple peptides covering this region of DHFR (Fig. 3b). At short chain lengths (1–37RNC), the N-terminal region is protected relative to the maximally deuterated control but is readily distinguishable from the same sequence in native DHFR, which was much more protected at 10 s deuteration. Non-native protection at short chain lengths may be a result of interactions with the ribosome or reflect transient non-native structure stabilized by the exit tunnel4. Although substantially deprotected relative to native DHFR, the N-terminal region was not completely deuterated in the RNCs, as judged by comparison to deuteration of a synthetic peptide comprising residues 1–37 (1–37peptide), which was maximally exchanged at the earliest labeling time. Elongation of the NC to extend the N terminus fully beyond the exit tunnel (1–64RNC) resulted in further deuteration. The N-terminal region remained highly deuterated with little variation in uptake, even as the NC extended to 106 and 126 residues during synthesis of the ABD. Folding of the DLD is therefore delayed until the complete subdomain is available and the C terminus is released from the ribosome, or artificially extended beyond the tunnel through a linker as in FL + 58RNC. Notably, this folding pathway is different than the in vitro pathway of refolding of DHFR from denaturant, in which the central eight-stranded β-sheet that spans the two subdomains is established early in the folding pathway48.

Fig. 3: Conformational dynamics of nascent DHFR on the ribosome.

a, Deuterium incorporation into peptide 9–28 of DHFR as a function of deuterium exposure time and NC length. Data are presented as mean values of two to four replicates. Error bars, s.d.; n = 3 or n = 4 independent experiments. b, Relative deuterium uptake of DHFR peptides after 10 s exposure to deuterium. Peptides belonging to the DLD or ABD subdomains are indicated. Data are presented as mean values of two to four replicates. Error bars, s.d.; n = 3 or n = 4 independent experiments. c, Peptide-resolved folding of nascent DHFR, illustrated on the structure of native DHFR (PDB 5CCC). Peptides with the same deuterium uptake (within 0.5 Da) as the native controls (FL DHFR or FL + 58RNC) are colored blue, while peptides that are deprotected relative to the controls are colored red. Regions for which we did not measure HDX are colored gold. The C-terminal 22 aa in each construct, expected to be occluded in the exit tunnel at that chain length, is colored green. The part of DHFR that is not yet synthesized at each chain length is colored white. d, Difference in deuterium uptake between FL DHFR and 1–126stop after 10 s deuteration. Darker red indicates more deuterium in 1–126stop relative to FL DHFR (increasing deprotection). Residues 127–159, not present in 1–126stop, are colored white. e, Difference in deuterium uptake between FL DHFR and FL + 50stop after 100 s deuteration. Darker red indicates increasing deprotection of FL + 50stop relative to FL DHFR. Peptides that do not differ in exchange are colored white. See also Extended Data Fig. 4 and Supplementary Data 1.

Extending this analysis to additional peptides across the RNCs, each reporting local folding information, allowed us to reconstruct the complete cotranslational folding pathway of DHFR. Peptide-resolved HDX for each different length RNC is shown in Fig. 3b, Extended Data Fig. 4a,b and Supplementary Data 1, and folding information is mapped onto the structure of native DHFR in Fig. 3c. Although not all peptides were uniformly detected across stalled RNCs of different lengths, clear patterns of HDX protection corresponding to NC folding were apparent. As described above, the N-terminal region including strand β1 is not folded when initially exposed outside the exit tunnel (1–37RNC), with deuteration levels close to the maximally deuterated control. In 1–64RNC, residues 25–36 emerge from the tunnel and fold into a native helix α1 belonging to the DLD, while the N-terminal region remains unstructured. Synthesis of the remainder of the ABD (1–106RNC) allows β2–β4 to exit the ribosome and acquire native-like protection from HDX. This protection was not a result of association with TF, as discussed in subsequent sections. Folding of the ABD completes when 126 residues of DHFR have been synthesized (1–126RNC), allowing β5 and α4 to coalesce with the remainder of the ABD outside the exit tunnel. At this point, β6 is still occluded in the tunnel, precluding folding of the adjacent β1 in the DLD. Synthesis and release of the C-terminal strand β8 from the ribosome triggers the final folding of the DLD, including the N-terminal segments (FL DHFR and FL + 58RNC).

The ribosome stabilizes a cotranslational folding intermediate

Unlike the DLD, peptides corresponding to the β-sheet of the ABD became progressively protected from HDX during translation, requiring neither the sequence context of FL DHFR nor even the complete subdomain (Fig. 3c). By contrast, previous work has shown that fragments of DHFR produced by chemical cleavage are disordered in isolation49. To test whether the ribosome modulates the conformation of the nascent ABD, we expressed isolated fragments corresponding to the ABD in E. coli. Only the longest fragment, consisting of residues 1–126 (herein 1–126stop), was soluble, and this was contingent on maintaining it as a fusion protein, with monomeric ultrastable GFP (muGFP50) at the N terminus. HDX MS showed that 1–126stop was substantially deprotected relative to native DHFR (Fig. 3d and Supplementary Data 1). High levels of exchange were observed for the N-terminal region, as expected in the absence of C-terminal strands β7 and β8 that complete the DLD. The β-strands (β2–β4) and peripheral helices comprising the core of the ABD were also strongly deprotected relative to FL DHFR. The ABD, therefore, fails to fold into a stable structure in isolation, although native-like folding of this subdomain is supported on the ribosome in the context of an RNC.

We asked why incomplete chains of DHFR can fold on the ribosome but not in free solution. We noticed that residues 93–118 near the C terminus of 1–126stop were maximally deuterated even at short deuteration times, indicative of structural disorder (Fig. 3d and Supplementary Data 1). In the corresponding RNC, however, this sequence is sterically confined in the ribosome exit tunnel (Figs. 1b and 3c). Therefore, we considered whether, in the absence of the ribosome, a region of C-terminal disorder may destabilize folded DHFR. Previous work has shown that unstructured termini can generate an entropic force that modulates protein conformation51. To explore this idea, we used FL + 58RNC as a model. FL + 58RNC consists of FL DHFR tethered to the peptidyl transfer center (PTC) by a flexible 50 aa linker that is largely occluded in the exit tunnel (Fig. 1b). Our HDX analysis showed that DHFR is natively folded in this context and almost indistinguishable from the isolated domain (Fig. 2b and Extended Data Fig. 3b,c). We reasoned that releasing the protein from the ribosome would remove the protective effect of the exit tunnel and allow an unstructured terminus to disrupt the conformation of DHFR. To test this idea, we replaced the stalling sequence with a stop codon, resulting in FL + 50stop. HDX MS analysis of FL + 50stop confirmed that the linker was highly dynamic, as expected (Extended Data Fig. 4c–e and Supplementary Data 1). Furthermore, we observed extensive deprotection in the DLD and ABD relative to native FL DHFR, consistent with increased structural dynamics (Fig. 3e and Extended Data Fig. 4c–e). Combined, our observations suggest a model whereby domain instability induced by unstructured C termini is rescued by confinement in the ribosome exit tunnel, potentially facilitating the folding of structured intermediates during translation.

DHFR can fold close to the ribosome

Proximity to the ribosome surface thermodynamically destabilizes some full-length nascent proteins relative to the native state of the ribosome11,12,13,15, prompting us to consider whether tethering to the ribosome might alter the conformation of FL DHFR. HDX analysis of FL + 58RNC did not reveal substantial differences in conformation compared to isolated DHFR (Fig. 2b). We therefore prepared an RNC with a 38-residue linker (FL + 38RNC), designed to bring DHFR closer to the ribosome surface. The folded core of DHFR in FL + 38RNC was similar in protection to isolated DHFR, although peripheral loops and part of the DLD were affected (Fig. 4a, Extended Data Fig. 5 and Supplementary Data 1). Furthermore, both FL + 58RNC and FL + 38RNC were enzymatically active (Fig. 4b). Therefore, close proximity to the ribosome exit tunnel does not substantially disrupt the conformation of folded wild-type DHFR.

Fig. 4: Ribosome contacts modulate the structure of nascent DHFR without inducing unfolding.

a, Relative deuterium uptake of FL + 38RNC compared to FL DHFR. DHFR peptides that are protected in the RNC relative to FL DHFR, at any deuteration time point, are colored blue; deprotected peptides are colored red. The Met20 loop is indicated with a green dashed ellipse. See also Extended Data Fig. 6a and Supplementary Data 1. b, Oxidoreductase activity of 25 nM FL DHFR, empty 70S ribosomes without an NC or RNCs. Where indicated, reactions are supplemented with 50 µg ml−1 RNaseA and 50 mM EDTA, a peptide corresponding to the C terminus of DHFR (C10, SYCFEILERR) or a scrambled-sequence control peptide (Scr, RFIERCELYS). Data are presented as mean values; error bars, s.d.; n = 3 independent experiments.

Source data

Remarkably, FL + 38RNC and FL + 58RNC were at least twofold to fourfold more active than isolated DHFR (Fig. 4b and Extended Data Fig. 6a,b). Vmax and, importantly, KMDHF were higher for the RNC, indicating that this effect is not explained by errors in measurement of enzyme concentration (Fig. 4b and Extended Data Fig. 6d). The results were also not explained by tube adsorption (Extended Data Fig. 6a). The extent of activity stimulation at fixed substrate concentration may have been underestimated because RNC concentration was calculated by assuming 100% occupancy of NCs on ribosomes (See Methods). This was bona fide DHFR activity, as it was fully inhibited by methotrexate, and neither empty ribosomes nor intermediate-length RNCs were active (Fig. 4b and Extended Data Fig. 6e). We observed the same effect for an RNC with a different linker sequence, indicating that hyperactivity is not an artefact of linker chemistry (Extended Data Fig. 6c). Notably, the region near the N terminus of DHFR that is protected in FL + 38RNC includes the 'Met20 loop' at the folate binding site of DHFR, the dynamics of which are known to strongly influence catalysis35 (Fig. 4a). Thus, physical interactions with the ribosome may alter the active site of DHFR without inducing global unfolding.

To test the threshold linker length for folding of DHFR on the ribosome, we prepared FL + 28RNC. This RNC was devoid of oxidoreductase activity (Fig. 4b) and copurified with TF (Extended Data Fig. 2c). TF was not responsible for the lack of activity, as FL + 28RNC purified from a TF-free background was also inactive (Extended Data Fig. 6c). We reasoned that a 28 aa linker is insufficient to expose the C terminus of DHFR outside the exit tunnel, precluding folding of the DLD. Indeed, DHFR activity could be restored by releasing the NC from FL + 28RNC with EDTA and RNase (Fig. 4b). This observation is consistent with our interpretation, based on HDX MS, that DHFR folding completes post-translationally when the C terminus emerges from the tunnel (Fig. 3a–c). To test whether DHFR in FL + 28RNC is folding-competent before release from the ribosome, we added a peptide comprising the C-terminal 10 aa of DHFR (peptide C10) in trans. Peptide C10 activated FL + 28RNC in a concentration-dependent manner, whereas a scrambled-sequence control had no effect (Fig. 4b and Extended Data Fig. 6c). None of the peptides influenced the activity of isolated FL DHFR. TF was not required for reactivation, as TF-free FL + 28RNC was similarly reactivated (Extended Data Fig. 6c). Peptide C10 did not release the NC from ribosomes, nor did it displace TF (Extended Data Fig. 6f). DHFR is therefore poised to complete folding upon emergence of the C terminus from the ribosome exit tunnel, and neither close proximity to the ribosome surface nor association with TF prevent acquisition of the native state.

Interaction of TF with DHFR on the ribosome

We next sought to determine the role of TF in the cotranslational folding of DHFR. TF is a ribosome-associated chaperone with a three-domain architecture (Fig. 5a). The ribosome binding domain (RBD) contains a conserved ribosome-interaction motif52, and the substrate binding domain (SBD) is required for chaperone function off the ribosome53. The role of the peptidyl-prolyl isomerase domain (PPD) in de novo folding is enigmatic54. Given that chaperone–RNC complexes are highly dynamic, details of the interaction between TF and NCs have eluded structural characterization55,56, although site-specific photocrosslinking showed that NCs contact all three domains of TF56,57.

Fig. 5: NC interactions with TF.

a, Structure of E. coli TF (PDB 1W26). RBD, ribosome binding domain; SBD, substrate binding domain; PPD, peptidyl-prolyl isomerase domain. The ribosome binding motif in the RBD is indicated with a red dashed rectangle. b, Difference in deuterium uptake between isolated TF and TF bound to DHFR RNCs after 10 s deuteration. Darker blue indicates less deuteration of RNC-bound TF relative to isolated TF (increasing protection). The sites on arm 2 that preferentially engage 1–126RNC are marked by ellipses. See also Supplementary Data 1. c, TF surface colored according to hydrophobicity75.

TF copurified with 1–106RNC and 1–126RNC, coinciding with the synthesis of the complete ABD (Extended Data Figs. 1c and 2c). To characterize the interaction between TF and nascent DHFR, we analyzed deuterium uptake for endogenous TF in the RNC samples and compared these data to HDX MS measurements of isolated TF. We followed 181 peptides for TF in each condition, with 99.5% sequence coverage (Extended Data Fig. 7a and Supplementary Data 1). A potential confounding factor is that isolated TF is weakly dimeric (dimerization Kd ~1 µM) but binds ribosomes as a monomer57,58,59. To account for this effect, we used both wild-type TF and a constitutively monomeric variant60 as reference controls for isolated TF. HDX MS confirmed that the monomeric variant was deprotected relative to wild-type TF at sites that are normally buried by the dimer interface (Extended Data Fig. 7b and Supplementary Data 1).

Compared to isolated TF, TF that was bound to 1–106RNC and 1–126RNC was very strongly protected from HDX in the ribosome-interacting motif of the RBD, as expected52 (Fig. 5b). We also observed protection of additional regions in all three domains of TF, which we attribute to interaction with nascent DHFR (Fig. 5b). Strong protection was observed in the two 'arms' of the SBD, with intermediate protection in the RBD, 'neck' region and catalytic site in the PPD. Analysis of the binding interface revealed a mix of hydrophobic and hydrophilic surfaces (Fig. 5c). The interaction sites in the RBD and PPD are predominantly hydrophobic. In the SBD, the interface includes the hydrophobic pocket in the crook of arm 2 as well as hydrophilic surfaces in arm 1 and the neck. The hydrophilic part of the neck situated between the arms, previously implicated in substrate engagement off the ribosome61, was not protected from exchange. The protected sites were the same in TF bound to 1–106RNC and 1–126RNC, indicating a common interaction surface. We did, however, detect increased protection of hydrophilic regions in arm 2 (residues 351–374 and 388–409) when TF was bound to the longer NC (Fig. 5B (red ellipses) and Supplementary Data 1).

Nascent DHFR can fold while associated with TF

To determine how TF influences the folding of DHFR on the ribosome, we purified 1–106RNC and 1–126RNC from E. coli lacking TF (ref. 22), resulting in 1–106RNCΔTF and 1–126RNCΔTF. MS confirmed that the absence of TF did not result in other chaperones (for example DnaK, DnaJ or GroEL) copurifing with the RNCs (Extended Data Fig. 2c). Pelleting assays showed that 1–126RNCΔTF was still competent to bind purified TF in vitro, and binding was sensitive to mutation of the ribosome-interacting motif in TF (Extended Data Fig. 8a).

To probe the conformation of the NC without TF, we analyzed the HDX behavior of nascent DHFR in 1–106RNCΔTF and 1–126RNCΔTF (Extended Data Fig. 7c and Supplementary Data 1). A set of peptides reported on the N terminus and β-strands in the ABD enabled us to compare wild-type and ΔTF RNCs. Although some marginal differences in deuterium uptake could be detected in the absence of TF, these were not consistently observed across overlapping peptides. Cotranslational folding of the ABD therefore occurs irrespective of the presence of TF, and TF binding does not explain our observation that the N-terminal region remains unfolded until release of the C terminus from the ribosome (Fig. 3c).

Considering that DHFR folding can occur in the presence of TF, we questioned what features of the NC are recognized by the chaperone. To probe the contribution of electrostatic versus hydrophobic interactions to binding, we tested the salt sensitivity of the TF–RNC interaction. TF preferred 1–106RNC over 1–126RNC when the complexes were purified under high-salt conditions (Extended Data Fig. 2c). By contrast, RNCs purified under low-salt conditions bound similar amounts of TF (Extended Data Fig. 8b). TF binding to 1–126RNC is therefore partially stabilized by electrostatic interactions, unlike binding to 1–106RNC, which is predominantly mediated by hydrophobic contacts and therefore stabilized by high salt. This observation is consistent with the folding-induced burial of the ABD hydrophobic core in 1–126RNC (Fig. 3c) as well as the preference of this NC for binding hydrophilic surfaces on TF (Fig. 5b).

To directly test the contribution of NC folding to TF binding, we introduced destabilizing mutations62 into 1–126RNC (Extended Data Fig. 8c). The mutated RNC bound more TF than wild-type 1–126RNC when purified under high-salt conditions, but the difference was much less pronounced when binding was reconstituted in vitro under low-salt conditions (Extended Data Fig. 8d,e). Thus, TF engages poorly folded intermediates by hydrophobic surfaces. Taken together, these observations indicate that TF uses a composite hydrophobic–hydrophilic interface to accommodate both folded and unfolded NCs, and provide indirect evidence supporting our conclusion that the ABD is natively folded in wild-type 1–126RNC. Our low-salt conditions are in a similar range of ionic strength to the E. coli cytosol (~100–200 mM (ref. 63)). In vivo, TF would thus be expected to bind equally well to NCs exposing different amounts of hydrophobic surface, exploiting different types of interaction in each case.

NC interactions with ribosomal proteins

To identify possible NC–ribosome interaction sites, we compared the HDX of ribosomal proteins in the RNCs to the same proteins in empty 70S ribosomes. We focused on a set of five ribosomal proteins (L4, L22, L23, L24 and L29) that are near or in the exit tunnel and therefore likely to contact the emerging NC (Fig. 6a). Sequence coverage of these proteins was close to 100% (Extended Data Fig. 9a). We identified several sites of HDX protection in ribosomal proteins when NC was present, often dependent on NC length (Fig. 6b–g and Supplementary Data 1).