## Main

In the cell, most nascent polypeptide chains begin to fold during biosynthesis1,2,3. In many cases, co-translational folding increases the ability of a protein to efficiently attain its native structure4,5,6,7. This may in part be due to the ribosome modulating the conformational ensembles sampled by nascent chains8,9,10,11. The ribosome constrains disordered chains close to its charged surface12, and can promote the early formation of compact states during co-translational folding13,14,15,16. In general, nascent chains have been found to be destabilized when they are bound to the ribosome5,17,18,19. This could be linked to interactions between the disordered nascent chain and the ribosome surface20, some of which bear a partial electrostatic character21,22. Ribosome–nascent chain interactions have also been suggested to attenuate co-translational folding rates23, and to compete with co-translational assembly between a nascent chain and its binding partner24. Collectively, these studies point clearly towards a role for the ribosome in shaping the onset of co-translational folding3. However, due in large part to the technical difficulty of measuring the intramolecular equilibria associated with ribosome–nascent chain interactions, a link between the energetics of ribosome interactions and co-translational folding outcomes has not yet been established in quantitative terms.

In this article we study the co-translational folding of FLN5, a 105 residue immunoglobulin-like domain from the tandem repeat protein filamin (Fig. 1a), using SecM-arrested ribosome–nascent chain complexes (RNCs) in which FLN5 is tethered to the ribosome via variable lengths of the following domain, FLN6 (Fig. 1b)20,25. Measurements of the accessibility of a C-terminal cysteine to covalent modification by PEG-maleimide (PEGylation) showed that the entire FLN5 domain emerges beyond the ribosome exit tunnel for linkers comprising at least 31 residues20 (FLN5+31). However, NMR observations demonstrated that FLN5 remains partially unfolded until the linker extends beyond 42 residues. This offset between the emergence of FLN5 and its folding suggests that the ribosome has a destabilizing effect on co-translational folding, which we speculated could be due to interactions between the unfolded nascent chain and the ribosome surface20. We use NMR spectroscopy together with protein engineering, molecular dynamics simulations and PEGylation measurements of nascent chain stability to identify a series of interaction sites on the nascent chain of varying affinities, and their impact on co-translational folding.

## Results

### Identification of regions of FLN5 RNCs interacting with the ribosome surface

Since the FLN5+31 RNC represents the first point during protein biosynthesis when the entire sequence is available for folding, we have used this biosynthetic snapshot as a starting point to examine how the ribosome might modulate the dynamic properties of a nascent chain. Previously, a comparison of the 2D 1H,15N NMR correlation spectrum of a FLN5+31 RNC against an isolated unfolded variant, Y719E, revealed site-selective line broadening that we interpreted as evidence of such interactions (Fig. 1c,d). From this, we have identified three main regions for investigation here. First, two clusters of aromatic residues, located within the core of the folded domain (Fig. 1a), were identified as displaying substantial broadening; these are referred to here as ‘F3’ (residues F665–F675) and ‘Y3’ (residues Y715–Y727) sites (Fig. 1d, green). Second, strong line broadenings were observed in the mildly basic C-terminal region of FLN5 (residues N728–C747, the ‘C-terminal segment’ (Fig. 1d, red). These observations correlate well with measurements of the interaction between individual amino acids and purified 70S ribosomes, which are strongest for aromatic and basic side chains (Fig. 1e and Extended Data Fig. 1). Third, resonances of the FLN6 tether are broadened to varying extents across FLN5 RNCs of increasing lengths, which, in part, relates to its occlusion within the ribosomal exit tunnel20 (Fig. 1d, cyan).

#### Interactions of F3 and Y3 aromatic clusters with the ribosome surface

We prepared a series of constructs, termed F3A3, A3Y3 and A3A3, in which the clusters of aromatic residues were mutated to alanine (Fig. 2a and Extended Data Fig. 2). The close overlay between 1H,15N NMR spectra of the isolated variants and disordered FLN5 indicates they are natively unfolded, as expected given large changes to the hydrophobic core (Extended Data Fig. 2c–f). On the ribosome, the 1H,15N correlation spectra of the corresponding FLN5+31 RNC variants showed substantial increases in cross-peak intensities relative to wild-type (wt) FLN5+31 RNC, broadly localized to the alanine mutation sites (Fig. 2b and Extended Data Fig. 3). Residues between A665 and G700 within the FLN5+31 A3A3 RNC variant reached a relative intensity close to 1, indicating that the mobility of this segment of the nascent chain is comparable to that of the isolated unfolded protein. These data demonstrate that the aromatic residues mediate a large proportion of the interactions at these sites. By contrast, in all variants the C-terminal segment of FLN5 (N728–C747) remained broadened beyond detection. This suggests additional interactions within this nascent chain segment (Fig. 2b).

To measure the strength of these aromatic interactions, we developed and acquired sensitivity-optimized measurements of the transverse cross-correlated relaxation (CCR) rate (Extended Data Fig. 4). Interactions of nascent chain segments with the ribosome surface will result in transferred relaxation, and thus an increase in the CCR rate, Δηxy, relative to the isolated unfolded state. The increase, $${\Delta}\eta _{xy} = p_{\mathrm{B}}\eta _{xy}^{\mathrm{bound}}$$, is proportional to the bound fraction pB, and to the CCR rate of the bound state, $$\eta _{xy}^{\mathrm{bound}}$$. Based on the known rotational correlation time (τC) of the ribosome26, $$\eta _{xy}^{\mathrm{bound}}$$ is estimated to be 7,000 s−1.

CCR measurements were acquired for FLN5+31 wt, F3A3, A3Y3 and A3A3 RNCs, and the corresponding isolated proteins (Fig. 2c and Extended Data Fig. 4). In general, we observed that increased CCR rates were associated with reduced resonance intensities. In the wt RNCs, increases in ηxy of 15 s−1 were measured around the aromatic F3 cluster. Within the Y3 site, residues beyond T714 were broadened beyond detection in wt and A3Y3 RNCs, but examination of flanking residues indicates that ηxy probably increases substantially beyond 15 s−1. Increased relaxation was also observed for residues at the N terminus, which we ascribe to an interaction of the 6xHis tag as previously observed in α-synuclein RNCs21.

We observe some evidence of cooperativity between the F3 and Y3 clusters, both by comparison of resonance intensities and of CCR rates: the elimination of one cluster leads to a small reduction in the interaction of the neighbouring cluster (Fig. 2b,d). However, in quantitative terms the extent of interaction is weak in all cases. Within the F3 cluster, observed increases in CCR rates correspond to ribosome-bound nascent chain populations of 0.1% (Fig. 2d). While resonances for the Y3 cluster were strongly broadened in wt and A3Y3, only allowing for a partial quantitation of CCR rates, the ribosome-bound population between residues 708 and 716 also neared 0.1%. Although these estimates assume a rigid bound state, even substantial flexibility that results in an order of magnitude decrease in $$\eta _{xy}^{\mathrm{bound}}$$ would indicate a bound population of only a few percent. Such weak interactions within these regions are not sufficient to perturb the co-translational folding process.

#### Interaction of the C-terminal segment with the ribosome surface

We next sought to investigate the interaction of the C-terminal segment (N728–C747) (Fig. 1d). The elimination of the Y3 cluster in the F3A3 and A3A3 variants resulted in newly observable resonances flanking this segment. Careful inspection revealed that small chemical shift perturbations (CSPs) were observed for these resonances (V717–G725, encompassing part of the Y3 region), relative to the isolated protein, and that their magnitude increased towards the (unobserved) ribosome-binding segment (Fig. 3a,b and Extended Data Fig. 5). Similarly, CSPs were observed at the C-terminal end of the binding segment between I748 and A751. Focusing on the A3A3 variant, we found that these CSPs were substantially reduced at high ionic strength (Fig. 3c), suggesting that they are associated with an interaction mediated at least in part by an electrostatic contribution.

We then explored the effect of the RNC linker length on the observed CSPs. We hypothesized that a shorter RNC would experience a higher effective ribosome concentration20 and therefore modulate the extent of binding. Indeed, as the linker length is increased from 26 to 110 amino acids, CSPs were observed to decrease (Fig. 3d) while resonance intensities increased (Extended Data Fig. 5). The CSPs at different lengths were collinear, indicating that changes in the 1H and 15N chemical shifts were strongly correlated (Fig. 3d). These collinear, correlated CSPs are an unambiguous indication that the C-terminal segment is rapidly exchanging between a ribosome-bound and free state, such that the observed chemical shift reflects a population-weighted average of unbound and bound states27. Resonances of the C-terminal residues I749–A751 showed deviations from collinearity at short linker lengths, which we attribute to proximity to the exit tunnel. For this reason, they have been excluded from further analysis. Together, these observations provide compelling evidence for a strong ribosome interaction involving nascent chain residues between Y727 and C747.

Alanine mutations within the F3 and Y3 clusters did not substantially perturb the observed C-terminal CSPs, indicating that there is no detectable cooperativity between interactions of these aromatic clusters and the C-terminal segment (Extended Data Fig. 5). Furthermore, no modulations in chemical shifts or intensities were observed upon varying magnetic field strength (Extended Data Fig. 5). Given the fast chemical exchange behaviour observed, we can infer from the largest frequency difference, Δν, of 30 Hz (for the 15N resonance of E724 at 22.3 T) that the dissociation of the bound state is rapid, with a lifetime much less than 3 ms (τ 1/4πΔν, that is, koff ≈ τ−1 360 s−1).

#### Impact of the linker sequence composition on FLN5 interactions

We investigated the influence of the FLN6 linker on FLN5–ribosome interactions. We previously reported that replacing all FLN6 residues with a poly(GS) sequence (Fig. 3e) did not perturb co-translational folding equilibria20. Here we find that substituting poly(GS) linkers within A3A3 RNCs led to only small shifts in the CSPs and resonance intensities in the C-terminal segment, indicating that the extent of binding is not greatly perturbed (Fig. 3f and Extended Data Fig. 6), and suggesting little impact of the nature of the tether on the co-translational folding of FLN5.

### Quantification and molecular modelling of the C-terminal segment interaction

Having identified a ribosome interaction site in the C-terminal segment, we next examined its molecular basis. Although the C-terminal segment does not contain aromatic residues, it contains three basic side chains (R734, K739, K746) that we hypothesized would contribute towards the electrostatic interactions (Fig. 1e), given the reduced CSPs observed at increased ionic strength (Fig. 3c). To explore this, we designed a FLN5 construct, A3A3E6, in which six residues (surface-exposed within the folded state) were replaced with acidic glutamate residues, thus reversing the net charge within this segment from +1 to −6 (Figs. 1a and 4a). The E6 mutations were found to greatly reduce the magnitude of CSPs in the C-terminal segment (Fig. 4b), indicating that these mutations reduce the affinity of the segment for the ribosome surface. However, some CSPs and line broadening persist (Extended Data Fig. 7), suggesting that the binding interaction of this region is not abrogated completely. Similar reductions in CSPs were also observed between FLN5+31 wt and E6 RNCs, although due to the increased line broadening fewer resonances adjacent to the interacting segment could be resolved (Extended Data Fig. 7).

To quantify the effect of the E6 mutations on the binding interaction, we have analysed further the observed CSPs, which reflect a population-weighted average of unbound and bound states. These residues provide a convenient ‘ruler’ to compare the interactions of different nascent chains with the ribosome. However, to determine the absolute amount of binding, the chemical shift of the fully bound state must be determined. To achieve this, we carried out a global analysis of CSPs observed in the A3A3 and A3A3E6 RNC variants across multiple linker lengths (Fig. 4c). Two assumptions were required for this analysis. First, while the strength of interactions in these variants clearly varies as a function of linker length, we assume that the difference in free energy of binding between variants, $${\Delta}{\Delta}G_{{\mathrm{U}}_{{\mathrm{free}}} - {\mathrm{U}}_{{\mathrm{bound}}}}^{{\mathrm{A3A3}} - {\mathrm{A3A3E6}}}$$ (where Ufree and Ubound represent unfolded states with the C-terminal segment unbound and ribosome-bound, respectively), is independent of RNC length. This is equivalent to the assumption that C-terminal segments in both variants experience the same effective ribosome concentration at a given linker length, which is supported by the similar structural and dynamic properties of the A3A3 and A3A3E6 variants: no chemical shift perturbations or differences in 15N R2 relaxation rates are observed beyond the immediate vicinity of the E6 mutations (Extended Data Fig. 8). Second, while unbound resonance positions vary between A3A3 and A3A3E6 RNCs due to local sequence effects, we assume that the chemical shift change upon binding, Δδmax, is the same for both variants. Given this, the observed chemical shift perturbations of four well-resolved resonances in A3A3 and A3A3E6 RNCs were fitted to determine the chemical shift differences between free and bound states, and the difference in affinities of the two variants, $${\Delta}{\Delta}G_{{\mathrm{U}}_{{\mathrm{free}}} - {\mathrm{U}}_{{\mathrm{bound}}}}^{{\mathrm{A3A3}} - {\mathrm{A3A3E6}}}= 1.9\pm0.1\,{\mathrm{kcal}}\,{{{\rm{mol}}^{-1}}}$$ (Fig. 4d and Extended Data Fig. 8). These results indicate that at a short RNC length of 26 amino acids, 90% and 22% of the A3A3 and A3A3E6 RNCs are bound to the ribosome, respectively, then at a longer RNC length of 42 amino acids these values decreased to 60% and 7%, respectively, and by 110 amino acids, the interaction essentially disappears in the A3A3E6 variant (Fig. 4e).

Next, we used coarse-grained (CG) molecular dynamics simulations of the A3A3 and A3A3E6 RNCs to probe this interaction further, and to explore the location of nascent chain interaction sites on the ribosome surface (Fig. 4g). The strength of electrostatic interactions between the nascent chain and the ribosome surface in the CG model was calibrated using simulations of the A3A3 FLN5+42 RNC, in order to achieve 60% binding of the C-terminal segment (730–746), as determined from our NMR observations (Fig. 4e). Simulations of other RNC lengths (+31, +67 and +110), and of the +42 E6 variant, were then carried out with no further adjustment of parameters. We found that this CG model accurately identified electrostatics-based interactions of the C-terminal segment of the nascent chain that were consistent with the extent of binding measured from our CSP analysis across all lengths (Fig. 4f). The C-terminal binding segment (N728–C747) in the A3A3 FLN5+31 RNC contacted primarily the 23S rRNA region located adjacent to the exit tunnel (mainly helices H24 and H50), and also a loop of the nearby ribosomal protein, uL24 (Fig. 4h). Moreover, for many individual RNA bases in this region, in particular those within helices H24, H47, H50 and H59 (Extended Data Fig. 9), the contact probability correlated strongly with the extent of binding across nascent chain lengths as determined experimentally from our CSP analysis (Fig. 4i). These observations point to a nascent chain interaction site at the ribosome exit vestibule. We note that a similar interaction site is also observed for folded FLN5+47 RNC by cryo electron microscopy28, and that helices H24 and H50 are highly structurally conserved across Bacteria and Eukaryotes. It may be that this region on the ribosomal surface has a role in modulating the co-translational folding of FLN5 and perhaps generally for nascent chains.

### Modulation of co-translational folding by C-terminal interactions

In the previous sections, we have developed a detailed description of the interactions formed between unfolded FLN5 nascent chains and the ribosome surface. By stabilizing the unfolded state, such interactions will in general lead to an inhibition of folding (Fig. 5a), and we set out in this section to test this linkage in quantitative terms. Since interactions of the F3 and Y3 aromatic clusters are weak, we have focused our analysis on the strong interactions of the C-terminal segment, which we have previously shown is essential for folding of the domain29. A simple calculation (detailed in the Supplementary Information) relates the amount of binding within the unfolded state (pB) to its change in stability:

$$\Delta G_{{\rm{Ufree}}-{\rm{U}}} = RT \ln (1-p_{\rm{B}})$$
(1)

We have used this expression, together with our measurements of interactions in unfolded RNCs of various lengths, to predict the effect on the co-translational folding equilibria of folding-competent FLN5 RNCs in the absence of the destabilizing A3A3 mutations (Fig. 4e, right-hand axis). While this calculation assumes that the A3A3 mutations themselves do not perturb the interaction, we have previously noted that the impact of E6 mutations on interactions of wt and A3A3 RNCs was similar (Extended Data Fig. 7), and that there was no detectable cooperativity between the aromatic clusters and interactions of the C-terminal segment (Extended Data Fig. 5). On this basis, we predict a destabilization of wt FLN5 RNCs by over 1 kcal mol−1 at short linker lengths (Fig. 4e, right-hand axis). In the absence of other changes in thermodynamic stability, we also predict that the E6 variant should shift co-translational folding equilibria towards the folded state (Fig. 5a and Fig. 4e, right-hand axis).

To test the predicted effect of interactions on folding experimentally, we employed a cysteine mass-tagging approach, PEGylation30, as a reporter of folding and co-translational folding equilibria. 2D NMR observations indicated that wt and E6 FLN5 have similar structures both in isolation and as RNCs (Extended Data Fig. 10), and so the mutations C747V and A721C were introduced to replace the native, C-terminal cysteine with one that is unambiguously emerged beyond the exit tunnel at all linker lengths examined here, and that is protected from modification by PEG-maleimide unless the domain is unfolded (Extended Data Fig. 10). The isolated variants, FLN5 C721 V747 (the ‘pseudo-wild-type’) and FLN5 C721 V747 E6, were reacted with PEG-maleimide in increasing concentrations of urea and the extent of protection was determined by electrophoresis. The resulting measurements fitted closely to a two-state unfolding model and indicated that the E6 variant was destabilized relative to the wt by $${\Delta}{\Delta}G_{{{{\rm{N}} - {\rm{U}}}}}^{{{{\rm{E6}} - {\rm{wt}}}}} = 4.0 \pm 0.2\,{\mathrm{kcal}}\,{{{\rm{mol}}^{-1}}}$$ (where N and U represent native and unfolded states), in excellent agreement with equivalent CD measurements (Fig. 5b and Extended Data Fig. 10).

We next subjected FLN5 and FLN5 E6 RNCs to PEGylation following the same protocol (Fig. 5c and Extended Data Fig. 10). As ribosomes dissociate at high concentrations of urea17, we restricted our analysis to measurements below 3 M urea, with the exception of FLN5+110 RNC for which data up to 3.75 M were used (Fig. 5c, dashed lines). The stabilities of FLN5+110 wt and E6 RNCs are indistinguishable from the corresponding isolated proteins, indicating that the difference in thermodynamic stability between isolated FLN5 and FLN5 E6 is conserved on the ribosome at long linker lengths ($${\Delta}{\Delta}G_{{{{\rm{N}} - {\rm{U}}}}}^{{{{\rm{E6}} - {\rm{wt}}}}}=4.0\pm0.3\,{\mathrm{kcal}}\,{{{\rm{mol}}^{-1}}}$$). However, at short linker lengths, the difference in stability between FLN5 and FLN5 E6 was decreased ($${\Delta}{\Delta}G_{{{{\rm{N}} - {\rm{U}}}}}^{{{{\rm{E6}} - {\rm{wt}}}}}=2.3\pm0.2\,{\mathrm{kcal}}\,{{{\rm{mol}}^{-1}}}$$ for FLN5+31) (Fig. 5c,d). Consistent with our hypothesis, wt nascent chains therefore become destabilized relative to E6 under strongly interacting conditions. Based on our earlier measurements of bound populations (Fig. 4e), we can predict the change in stability relative to the isolated domains (Fig. 5d), and these predictions closely match direct observations using PEGylation (Fig. 5d). Therefore, we conclude that the C-terminal segment can indeed modulate co-translational FLN5 folding through the competition between binding and folding.

## Conclusion

For many domains, co-translational folding has been reported to be destabilized on the ribosome5,17,18,19. This effect is generally inferred to arise through interactions between emerging nascent chains and the ribosome, which have been observed for a range of different nascent chain sequences12,21,22,23,24. However, a direct link between the energetics of interactions and of co-translational folding has not previously been established. In this article we have therefore systematically examined a series of interactions between unfolded FLN5 nascent chains and the ribosome surface, in order to determine their effect on co-translational folding. While some of these interactions, between aromatic clusters and the ribosome surface, are too weak to perturb the energetics of folding substantially, we have identified a strongly interacting C-terminal segment, with over 90% bound at short linker lengths. The length of the sequestered segment (22 amino acids) is longer than the C-terminal truncation that can be tolerated by isolated FLN5 before it unfolds (12 amino acids)29, supporting the crucial role of this segment to enable native structure formation. Importantly, our analysis establishes quantitative agreement between the strength of the observed interactions and the energetics of co-translational folding itself, providing a residue-specific demonstration of the ability of the ribosome surface to directly modulate co-translational folding, effectively acting as a holdase.

A notable consequence of our analysis is that strong interactions are required to appreciably perturb the co-translational folding landscape, for example, a destabilization of 1 kcal mol−1 requires over 80% binding (equation (2)). Such interactions were indeed observed for the C-terminal segment at short linker lengths, but the bound population decreases sharply with increasing linker length, to below 50% at linker lengths beyond 47 amino acids (Fig. 4e). This rapid, short-range effect may provide a mechanism by which the engagement of molecular chaperones with the emerging NC is regulated.

We observe that the molecular determinants of ribosome interactions, that is, positively charged and aromatic residues, are similar to those recognized by other molecule chaperones and processing complexes31. These include signal recognition particle32, the ribosome-associated chaperone trigger factor33,34,35, and SecB36, both of which function as holdases for nascent polypeptide chains, and other downstream chaperones such as DnaK37. Indeed, over the past few years NMR studies have been instrumental in revealing with exquisite detail the mechanisms underlying substrate recognition and chaperone action by these molecules35,36,38,39,40,41,42.

During biosynthesis, substrates emerge from the ribosome in their high-energy unfolded states, and so in contrast to chaperones that act post-translationally, interactions with holdases carry no energetic cost. This is reflected in the ATP independence of both trigger factor and SecB holdases—as well as in interactions with the ribosome surface itself. We speculate that this short-range holdase activity could have a number of functional roles: sequestration of hydrophobic segments until later residues have been synthesized; to reduce the risk of misfolding, particularly in tandem repeat proteins such as FLN29,43; or to delay folding ahead of co-translational assembly44,45,46 or the engagement of downstream chaperones such as TF and SecB, which may be involved in secretory pathways35,36. This suggests that the ribosome is more than an inert hub that orchestrates interactions of auxiliary factors and chaperones47, and in fact is itself an active participant in the co-translational folding process. In conclusion, we demonstrate the holdase effect of the ribosome as a bespoke form of regulation over co-translational folding.

## Methods

### Sample preparation and quality control

DNA constructs of tandem immunoglobulin domains FLN5 and FLN6, and of poly(GS) glycine serine repeat sequence variants and Y719E variants, were described previously20. Aromatic and glutamate mutants were generated using site-directed mutagenesis. A stronger SecM variant was used for preparation of [2H,13CH3-Ile, Leu and Val (ILV)]-labelled RNCs for PEGylation experiments, with sequence ‘FSTPVWIWWWPRIRGPPPPWT’. 15N and [2H,13CH3-ILV]-labelled proteins were grown and purified as described previously49, except in the case of disordered FLN5 variants where the final chromatography step was performed using a HiLoad 16/600 Superdex 200 pg (GE Healthcare) column in the presence of 6 M urea, followed by buffer exchange into Tico buffer25. 15N and [2H,13CH3-ILV]-labelled RNCs were generated in BL21(DE3) Escherichia coli as described previously25, where the only modification was the replacement of the final step of purification (sucrose gradient chromatography) with an additional sucrose cushion step followed by hydrophobic affinity chromatography using a HiTrap FF butyl column (GE Healthcare). The occupancy and integrity of RNC samples were monitored as described previously25.

### NMR spectroscopy

NMR data were acquired on 500 and 700 MHz Bruker Avance III, and 800 and 950 MHz Bruker Avance III HD spectrometers, all equipped with TCI cryoprobes. Data were processed and analysed with nmrPipe, Sparky, CCPN Analysis and MATLAB50,51. All samples were prepared at nascent chain concentrations of between 7 and 13 µM in Tico buffer, pH 7.5, containing 10% D2O and 0.001% (w/v) DSS25. For 15N-labelled samples, 1H,15N SOFAST-HMQC experiments52 were recorded with typical acquisition times of 50.4 ms and 29.5 ms in the direct and indirect dimensions, respectively, and a 100 ms recycle delay. Transverse cross-correlated relaxation was measured using 1H,15N BEST-TROSY-CCR experiments (Extended Data Fig. 4), acquired at 950 MHz with a recycle delay of 200 ms and an acquisition time of 49 ms in the direct dimension. For measurements of isolated proteins, a relaxation delay of 111 ms was used, while shorter relaxation delays were used for measurements of RNCs (22 ms for FLN5+31; 44 ms for FLN5+31 F3A3 and FLN5+31 A3A3). An acquisition time of 43 ms was used in the indirect dimension for all measurements except for FLN5+31, for which a 21 ms acquisition time was used. Samples were doped with 15 mM NiDO2A (nickel(II) 1,4,7,10-tetraazacyclododecane-1,7-bis(acetic acid)) to enhance sensitivity (Extended Data Fig. 4)53. All experiments were interleaved with SORDID diffusion measurements54 with a diffusion delay of 190 ms, an acquisition time of 49 ms, and 4 ms trapezoidal gradient pulses with strengths of 0.058 and 0.387 T m−1. Ribosome background labelling was assessed using 15N filtered/edited difference spectroscopy25. Chemical shift assignments for FLN5 variants were verified using 1H,15N NOESY-HSQC and TOCSY-HSQC experiments, with 200 ms and 70 ms mixing times, respectively.

For [2H,13CH3-ILV]-labelled samples, 1H,13C HMQC spectra were acquired at 800 MHz with acquisition times of 100 ms and 7 ms in the direct and indirect dimensions, respectively. Sample integrity was assessed through 13C-edited 1H STE diffusion measurements25 and 1H R2 relaxation measurements (13C-edited and incorporating a filter to select slow relaxing coherences55), in which signals detected after a 100 ms relaxation delay were interpreted as indicating nascent chain release.

1H R2 rates of isolated amino acids (200 µM in D2O) were measured in the presence and absence of 1 µM 70S ribosomes, using a 500 Hz PROJECT pulse train (700 MHz, 283 K)56.

### CD spectroscopy

CD spectra were acquired using a Chirascan-plus CD spectrometer (Applied Photophysics). All samples (20 µM) were incubated for ≥3 h at 283 K prior to measurement in Tico buffer (12 mM Hepes, 30 mM NH4Cl, 6 mM MgCl2, pH 7.5). CD signals at 211 nm were fitted globally using MatLab to a two-state unfolding model with a common m value, 〈mD–N〉 (ref. 57):

$$\begin{array}{*{20}{c}} {y = \alpha _{\mathrm{N}} + \alpha _{\mathrm{D}}\frac{{e^{\frac{{\left\langle {m_{{\mathrm{D - N}}}} \right\rangle \left( {\left[ {\mathrm{D}} \right] - \left[ {\mathrm{D}} \right]_{50{{{\mathrm{\% }}}}}} \right)}}{{RT}}}}}{{1 + e^{\frac{{\left\langle {m_{{\mathrm{D - N}}}} \right\rangle \left( {\left[ {\mathrm{D}} \right] - \left[ {\mathrm{D}} \right]_{50{{{\mathrm{\% }}}}}} \right)}}{{RT}}}}}} \end{array}$$
(2)

where [D] is the denaturant concentration, [D]50% is the midpoint of folding, and α terms represent baselines for native (N) and denatured states. Stabilities were then calculated as ΔGD–N = mD–N[D]50%.

### Molecular dynamics

Simulations of FLN5+31, +42, +67 and +110 RNCs were run in Gromacs 4.5.758 using Cα structure-based models generated with SMOG 2.059,60, extended to represent RNA using three beads located at P, C4′ and N3 atoms. A rigid ribosome model was created based on the structure 4ybb61 and including only nascent chain-accessible residues, that is, the exit tunnel and the surrounding surface. The A3A3 variant was modelled by removing all contacts involving the mutated sites, and the simulation temperature was then tuned so that isolated wt and A3A3 variants were folded and unfolded, respectively. Electrostatic interactions were introduced using Debye–Hückel theory62, with parameters chosen to reproduce the experimentally observed bound population of the FLN5+42 RNC and then applied without change to the remaining lengths. Three independent simulations were carried out for each system, with 2 × 108 steps per run.

### PEGylation

Samples (100–200 pmol isolated proteins or 8 pmol of RNC) were incubated in increasing urea concentrations for at least 2 h at 283 K prior to measurements. Methoxypolyethylene glycol maleimide (PEG-Mal) was added to a final concentration of 2 mM and incubated for 10 min before quenching the reaction with 0.23 M DTT. Samples were separated on denaturing 12% (w/v) polyacrylamide Bis-Tris gels (pH 5.7) which were subsequently Coomassie stained (for isolated proteins) or analysed by Western blot (for RNCs). Quantitation was performed using ImageStudio (Licor). As a control, Western blots of dilutions of a FLN5+47 RNC variant showed excellent linearity (R2 = 0.98) over the concentration range used in analyses (Supplementary Fig. 14). Populations of unfolded (PEGylated) nascent chains were determined relative to the total nascent chain concentration (PEGylated and unPEGylated) and fitted to equation (2) to determine the thermodynamic stability as described above.