FPA takes advantage of the ability of APs to bind in the upper parts of the ribosome exit tunnel and thereby pause translation when their last codon is in the ribosomal A-site (Ito and Chiba, 2013). The duration of an AP-induced pause is reduced in proportion to pulling forces exerted on the nascent chain (Goldman et al., 2015; Kemp et al., 2020), that is, APs can act as force sensors and can be tuned by mutation to react to different force levels (Cymer et al., 2015a). In an FPA experiment, a series of constructs is made in which a force-generating sequence element (e.g., a TMH) is placed an increasing number of residues away from an AP (reflected in N, the number of residues from the start of the protein to the end of the AP), which in turn is followed by a C-terminal tail (Figure 1a). In constructs where the TMH engages in an interaction that generates a strong enough pulling force F on the nascent chain at the point when the ribosome reaches the last codon of the AP, pausing will be prevented and mostly full-length protein will be produced during a short pulse with [³⁵S]-Met (Figure 1b, middle). In contrast, in constructs where little force is exerted on the AP, pausing will be efficient and more of the arrested form of the protein will be produced (Figure 1b, left and right). The fraction full-length protein produced, f_FL = I_FL/(I_FL+I_A), where I_FL and I_A are the intensities of the bands representing the full-length (FL) and arrested (A) species on an SDS-PAGE gel (Figure 1c and Figure 1—figure supplement 1), can therefore be used as a proxy for F in a given construct (Kemp et al., 2020; Niesen et al., 2018; Leininger et al., 2019). A plot of f_FL versus N – a force profile (FP) – thus can provide a detailed picture of the cotranslational process in question, as reflected in the variation in the force exerted on the nascent chain during translation. FPs can be recorded with up to single-residue resolution by increasing N in steps of one residue (corresponding to a lengthening of the nascent chain by ~3 Å).

We chose EmrE as an example of a small, relatively simple 4-TMH protein. EmrE is a dual-topology protein, that is, the monomers integrate into the inner membrane in a 50–50 mixture of N_in-C_in and N_out-C_out topologies; two oppositely oriented monomers then assemble into an antiparallel dimer (Chen et al., 2007; Rapp et al., 2007). To avoid potential complications caused by the dual topology, we used EmrE(C_out), a mutant version that adopts the N_out-C_out topology (Rapp et al., 2007), and further used the relatively weak SecM(Ec) AP (Ismail et al., 2012) and included an HA tag for immunoprecipitation (Figure 2a). A series of EmrE(C_out)-AP constructs (see Supplementary file 1 for sequences) was used to obtain the FP shown in Figure 2b (orange curve), at 2–5 residues resolution. Also shown is an FP derived from a CGMD simulation (CGMD-FP, gray; Van Lehn et al., 2015); a hydrophobicity plot (HP) is included in Figure 2—figure supplement 1.

We have previously shown that a model TMH composed of Ala and Leu residues generates a peak in an FP recorded with the SecM(Ec) AP that reaches half-maximal amplitude (N_start) when the N-terminal end of the TMH is ~45 residues away from the polypeptide transferase center (PTC) (Ismail et al., 2012), and a recent real-time FRET study of cotranslational membrane integration found that the N-terminal end of the first TMH in a protein reaches the vicinity of the SecYEG translocon when it is 40–50 residues away from the PTC (Mercier et al., 2020). For EmrE(C_out) TMH1, this would correspond to constructs with N ≈ 50. However, the f_FL values are hardly above background in this region of the FP. Due to the functionally important E¹⁴ residue, TMH1 is only marginally hydrophobic and does not become firmly embedded in the membrane until the protein dimerizes (Seurig et al., 2019). To ascertain whether the lack of a peak in the FP corresponding to the membrane integration of TMH1 is because of its low hydrophobicity, we mutated E¹⁴ to L. Indeed, in the FP obtained for EmrE(C_out,E¹⁴L) (Figure 2b, green curve), a clear peak appears at the expected chain length N_start ≈ 50 residues. Mutation E¹⁴A yields an f_FL value intermediate between EmrE(C_out,E¹⁴L) and EmrE(C_out) at N = 55 (Figure 2c), while f_FL for the mutants EmrE(C_out,E¹⁴D) and EmrE(C_out,E¹⁴Q) is the same as for EmrE(C_out).

Peak II has N_start ≈ 76, corresponding to a situation where the N-terminal end of TMH2 is ~45 residues from the PTC (Figure 2d). The double mutation I³⁷I³⁸→NN in TMH2 reduces f_FL at N = 80 and 85 (magenta triangles), as expected. Unexpectedly, however, the E¹⁴L, E¹⁴A, and E¹⁴Q (but not the E¹⁴D) mutations in TMH1 increase f_FL at N = 85 (Figure 2c), showing that a negatively charged residue (D or E) in position 14 in TMH1 specifically reduces the pulling force generated by TMH2 at N = 85, that is, when about one-half of TMH2 has integrated into the membrane. Likewise, f_FL values at N = 115 and 130 (but not at N = 105, included as a negative control) are specifically affected by mutations in E¹⁴: at N = 115 (one-half of TMH3 integrated), all four mutations in position 14 increase f_FL relative to E¹⁴, while at N = 130 (beginning of TMH4 integration) the E¹⁴A and E¹⁴L mutations decrease f_FL (Figure 2c). FPA thus reveals long-range effects of mutations in E¹⁴ on three specific steps in the membrane integration of the downstream TMHs. This implies that TMH1 remains in the vicinity of the translocon and that E¹⁴ makes specific interactions with residues in the TMH2–TMH4 region during the membrane integration process. Further studies will be required to pinpoint these interactions and understand the role played by the slow dynamics of TMH1 integration (Seurig et al., 2019).

Peak III has N_start ≈ 102 residues, with the N-terminal end of TMH3 ~45 residues from the PTC (Figure 2d). Peak IV is difficult to locate precisely in the FP because f_FL values are high throughout the TMH3–TMH4 region, but is seen at N_start ≈ 132 residues when the strong SecM(Ec-sup1) AP (Yap and Bernstein, 2009) is used (blue curve), again with the N-terminal end of TMH4 ~45 residues from the PTC (Figure 2d). As shown in Figure 2e, the TMHs cease generating a pulling force when their C-terminal ends are ~45 residues away from the PTC, indicating that they are fully integrated at this point.

We next studied GlpG, a medium-sized monomeric 6-TMH rhomboid protease with an ~60 residue cytoplasmic N-terminal domain (NTD) (Sherratt et al., 2012; Wang et al., 2006) (Figure 3a),  a protein that allows us to follow the cotranslational folding of a soluble domain and integration of a membrane domain in the same experiment.

The FP is shown in Figure 3b (orange curve). It was obtained at 5-residue resolution, except for the portion N = 168–224, which we measured with single-residue resolution. For unknown reasons, constructs with N ≈ 140–190 residues gave rise to a slowly migrating band on the gel that was difficult to interpret (Figure 1—figure supplement 1j,k); this problem did not arise when the NTD (GlpG residues 1–60) was replaced by residues 1–58 of the LepB protein (Figure 3a), and the corresponding f_FL values are shown in the FP (N = 131–224). The LepB part contains an N-terminal, N_out-C_in-oriented TMH (Wolfe et al., 1983; von Heijne, 1989), that interacts with the signal recognition particle Ffh (Schibich et al., 2016) and hence targets the LepB-GlpG constructs to the SecYEG translocon before GlpG TMH1 is translated. This could in principle affect the FP; however, because the C-terminal end of the LepB part is ≥70 residues away from the C-terminal end of the SecM AP in these constructs, LepB is far outside the ribosome exit tunnel and therefore unlikely to exert a strong effect. Indeed, f_FL values for GlpG (calculated either including or excluding the slowly migrating band in I_FL) and LepB-GlpG are very similar in the peak III region (N = 166–231) of the FP (Figure 3—figure supplement 1a). N_start and N_end values for peaks II–VII are indicated in Figure 3—figure supplement 1c,d.

Peak I, at N_start ≈ 84 residues, is conspicuously close to what would be expected for the folding of the NTD from previous studies of cotranslational folding of small globular domains in the ribosome exit tunnel (Farías-Rico et al., 2018). To verify that the peak indeed represents folding of the NTD, we recorded an FP for the NTD by in vitro transcription-translation in the PURE system (Shimizu et al., 2005) and further made a destabilizing point mutation (F¹⁶E) in the core of the NTD (Figure 3c). The FP obtained in vitro (magenta) overlaps peak I in the in vivo FP, and the mutation strongly reduces f_FL values for peak I both in vivo (green) and in vitro (black). Given that the NTD has a relative contact order of 15% and is predicted to fold on the ms time scale (Plaxco et al., 1998) while the elongation cycle on the ribosome takes ~100 ms/codon (Young and Bremer, 1976), the NTD has ample time to equilibrate between the unfolded and accessible folded states at each elongation step (Kemp et al., 2019). We conclude that the ~60 residue NTD folds inside the ribosome exit tunnel when its C-terminal end is 25–30 residues from the PTC, well before synthesis of the membrane domain has commenced.

Peaks II–VII in the FP correspond reasonably well to the CGMD-FP (gray) and HP (Figure 3—figure supplement 1b). The unexpectedly low N_start value for peak III seems to be caused by an upstream periplasmic surface helix (Figure 3f) (see below). Likewise, peak VI-a likely reflects the membrane integration of a hydrophobic, membrane-associated cytoplasmic segment located just upstream of TMH5 (Figure 3—figure supplement 1c). In contrast, the unexpectedly high N_start value for peak IV indicates that integration of TMH3 commences only when its N-terminal end is ~52 residues away from the PTC, possibly because of the tight spacing between TMH2 and TMH3.

As peak III saturates at f_FL ≈ 0.9 over a rather wide range, we sought a more detailed view by using the strong SecM(Ec-Sup1) AP (Yap and Bernstein, 2009) (Figure 3b,d, blue) and the medium-strong SecM(Ec-Ms) AP (Farías-Rico et al., 2017) (Figure 3d, green). The SecM(Ec-Sup1) FP allows a precise determination of N_max = 200, at which point the middle of TMH2 (L¹⁵⁵) is located 45 residues from the PTC (Figure 3f). The SecM(Ec-Ms) FP reveals additional detail: peak III is now seen to be composed of three subpeaks III-a, III-b, and III-c. III-a has N_start = 182, coinciding with the N-terminal end of the periplasmic surface helix reaching 45 resides away from the PTC. For III-b, N_start ≈ 190, with the N-terminal end of TMH2 ~45 residues from the PTC. The major subpeak III-c at N ≈ 197–204 finally corresponds well to the peak seen in the SecM(Ec-Sup1) and the CGMD FPs, and therefore represents the membrane insertion of the most hydrophobic part of TMH2. Taken together, subpeaks III-b and III-c are reminiscent of the biphasic pulling force pattern previously recorded for a model hydrophobic transmembrane segment using the medium-strong SecM(Ms) AP (Ismail et al., 2012), which is closely related to the SecM(Ec-Ms) AP used here.

We further recorded a SecM(Ec-Ms) FP (magenta) for the triple mutation Y¹³⁸F¹³⁹L¹⁴³→NNN (Figure 3e) that renders the periplasmic surface helix less hydrophobic: the mutation strongly reduces the amplitude of peak III-a, has only a small effect on peak III-b, and both reduces the amplitude and shifts N_start and N_max for peak III-c by approximately four residues (Figure 3d,f). Thus, the periplasmic surface helix engages in hydrophobic interactions already during its passage through the translocon, presumably by sliding along a partly open lateral gate (Cymer et al., 2015b). It also adds to the force generated by the membrane integration of TMH2, possibly by partitioning into the periplasmic leaflet of the inner membrane at approximately the same time that TMH2 enters the translocon.

Finally, we studied BtuC, a vitamin B12 transporter with 10 TMHs, as an example of a large, multispanning protein with a complex fold (Hvorup et al., 2007). In order to improve expression, we added the N-terminal part of LepB to the BtuC constructs (Figure 4a) and used a LepB antiserum for immunoprecipitation. The N_out-C_in orientation of LepB TMH1 ensures that the N_in-C_in topology of BtuC will be maintained, and constructs that we could measure without the LepB fusion gave similar f_FL values as those seen for the LepB fusions (Figure 4—figure supplement 1b).

We identified 11 peaks in the FP (Figure 4b, orange), one more than could be accounted for by the 10 TMHs. Since it was not possible to provide an unequivocal match between the BtuC FP and the CGMD-FP (or HP, Figure 4—figure supplement 1a), we did two sets of controls. First, we chose constructs at or near peaks in the FP and CGMD-FP and mutated multiple hydrophobic residues (Leu, Ile, Val, Met) located 40–50 residues from the PTC to less hydrophobic Ala residues (Figure 4—figure supplement 2). The mutations caused significant drops in f_FL (p < 0.01, two-sided t-test), except for construct N = 191 that is mutated at the extreme N-terminus of TMH5. The mutation data allowed us to identify the membrane integration of TMHs 1, 2, 3, 4, 5, 7, 8, 9, and 10 with peaks I, II, III, IV, V, VIII, IX, X, and XI, respectively; the overlapping peaks VIII and IX appear to represent the concerted integration of the closely spaced TMH7 and TMH8. However, peak II (corresponding to TMH2) is shifted to unexpectedly high, and peaks V (corresponding to TMH5), X (corresponding to TMH9), and XI (corresponding to TMH10) to unexpectedly low, N_start values (Figure 4c). To confirm these assignments, we obtained FPs for the isolated TMH2 (dashed green), TMH8 (dashed light blue), and TMH10 (dashed pink) sequences (Figure 4b) by introducing them into the periplasmic domain of LepB such that they maintained their natural N_out-C_in orientation (Figure 4d); the FPs for the individual TMHs overlap the corresponding peaks II, IX, and XI in the full FP. Likewise, an FP obtained for a construct lacking TMH1-TMH4 overlaps the full FP, except that peak V is shifted to a higher N_start value (Figure 4—figure supplement 3), more in line with the peak seen in the CGMD-FP. The low N_start value for the N_in-C_out-oriented TMH5 in full-length BtuC may result from an early interaction between a positively charged patch (RFARRHLSTSR) just upstream of TMH5 and negatively charged lipid headgroups (note that only two of the four Arg residues are present in the ΔTMH1-TMH4 construct; Figure 4—figure supplement 3), while the low N_start values for peaks X and XI are likely caused by the short upstream hydrophobic segments LCGL and LAAALEL (Figure 4c,f), similar to peak III in GlpG.

Remarkably, the N-terminal end of the isolated TMH2 is ~45 residues away from the PTC at N_start, suggesting that upstream sequence elements present in full-length BtuC delay the integration of TMH2 by ~10 residues (compare II* and II in Figure 4c). The most conspicuous feature in the upstream region of TMH2 is the presence of three positively charged Arg residues, an uncommon occurrence in a periplasmic loop (Heijne, 1986). Indeed, when these residues are replaced by uncharged Gln residues in LepB-BtuC, peak II (dashed black in Figure 4b,e) becomes almost identical to the FP for the isolated TMH2; a similar behavior is seen when the CGMD-FP simulation is run without an electrical membrane potential (Figure 4e). Upstream positively charged residues thus delay the membrane integration of the N_out-oriented TMH2, possibly because of the energetic cost of translocating them against the membrane potential (Ismail et al., 2015), or because they are temporarily retained in the negatively charged exit tunnel (Mercier et al., 2020).

Neither peak VI nor VII seems to represent the integration of TMH6, but instead flanks the location expected from the CGMD-FP and HP and apparently corresponds, respectively, to the membrane insertion of a short periplasmic re-entrant helix and of a short cytoplasmic surface helix (Figure 4c,h). Mutation of three hydrophobic residues to Ala in the latter significantly reduces the amplitude of peak VII (Figure 4—figure supplement 2, construct N = 259). Further, the FP for the isolated TMH6 (Figure 4b,g, dashed dark blue) peaks in the location expected from the CGMD-FP, between peaks VI and VII, and the FP for the isolated TMH5-6 part that includes the re-entrant helix but lacks the downstream surface helix is intermediate between the LepB-BtuC and the TMH6 FPs (Figure 4g, dashed green). Thus, the membrane interactions of the periplasmic re-entrant helix and the cytoplasmic surface helix exert a strong effect on the membrane integration of the intervening TMH6.

All enzymes used in this study were purchased from Thermo Fisher Scientific (USA) and New England Biolabs (USA). Oligonucleotides were from Eurofins Genomics (Germany). DNA isolation/purification kits and precast polyacrylamide gels were from Thermo Fisher Scientific (USA). L-[³⁵S]-methionine was obtained from PerkinElmer (USA). Mouse monoclonal antibody against the HA antigen was purchased from BioLegend (USA). Protein G-agarose beads were manufactured by Roche (Switzerland). All other reagents were from Sigma-Aldrich (Germany).

Competent E. coli MC1061 (Dalbey and Wickner, 1986) or BL21 (DE3) cells were transformed with the respective pING1 (BtuC, GlpG) or pET Duet-1 (EmrE) plasmid, respectively, and grown overnight at 37°C in M9 minimal medium supplemented with 19 amino acids (1 μg/ml, no Met), 100 μg/ml thiamine, 0.4% (w/v) fructose, 100 mg/ml ampicillin, 2 mM MgSO₄, and 0.1 mM CaCl₂. Cells were diluted into fresh M9 medium to an OD₆₀₀ of 0.1 and grown until an OD₆₀₀ of 0.3–0.5. Expression from pING1 was induced with 0.2% (w/v) arabinose and continued for 5 min at 37°C. Expression from pET Duet-1 was induced with 1 mM IPTG and continued for 10 min at 37°C. Proteins were then radiolabeled with [³⁵S]-methionine for 2 min (1 min for BtuC constructs lacking the N-terminal LepB fusion) at 37°C before the reaction was stopped by adding ice-cold trichloroacetic acid (TCA) to a final concentration of 10%. Samples were put on ice for 30 min and precipitates were spun down for 10 min at 20,000 g at 4°C in a tabletop centrifuge (Eppendorf, Germany). After one wash with ice-cold acetone, centrifugation was repeated and pellets were subsequently solubilized in Tris-SDS buffer (10 mM Tris-Cl pH 7.5, 2% [w/v] SDS) for 5 min while shaking at 1400 rpm at 37°C. Samples were centrifuged for 5 min at 20,000 g to remove insoluble material. The supernatant was then added to a buffer containing 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.1 mM EDTA-KOH, 2% (v/v) triton X-100, and supplemented with Pansorbin (Sigma-Aldrich) (BtuC constructs) or Protein-G-agarose (Roche) (all GlpG and EmrE constructs, and BtuC constructs lacking the N-terminal LepB fusion). After 15 min incubation on ice, non-specifically bound proteins were removed by centrifugation at 20,000×g (when Pansorbin was used) or 7000×g (when Protein-G-agarose was used). The supernatant was used for immunoprecipitation of BtuC constructs using Pansorbin and LepB antisera (rabbit) (in-house), or immunoprecipitation of GlpG/EmrE constructs using Protein-G-agarose and Anti-HA.11 Epitope Tag Antibody (mouse) (BioLegend). The incubation was carried out at 4°C while rolling. After centrifugation for 1 min, immunoprecipitates were washed with 10 mM Tris-Cl pH 7.5, 150 mM NaCl, 2 mM EDTA, and 0.2% (v/v) triton X-100 and subsequently with 10 mM Tris-Cl pH 7.5. Samples were spun down again and pellets were solubilized in SDS sample buffer (67 mM Tris, 33% [w/v] SDS, 0.012% [w/v] bromophenol blue, 10 mM EDTA-KOH pH 8.0, 6.75% [v/v] glycerol, 100 mM DTT) for 10 min while shaking at 1400 rpm. Solubilized proteins were incubated with 0.25 mg/ml RNase for 30 min at 37°C and subsequently separated by SDS-PAGE on Bis-Tris gels (Thermo Fisher Scientific). Gels were fixed in 30% (v/v) methanol and 10% (v/v) acetic acid and dried by using a Bio-Rad gel dryer model 583 (Bio-Rad Laboratories, US). Radiolabeled proteins were detected by exposing dried gels to phosphorimaging plates, which were scanned in a Fujifilm FLA-3000 scanner (Fujifilm, Japan). Band intensity profiles were obtained using the FIJI (ImageJ) software and quantified with our in-house software EasyQuant. A_c and/or FL_c controls were included in the SDS-PAGE analysis for constructs where the identities of the A and FL bands were not immediately obvious on the gel. Data was generally collected from three independent biological replicates for EmrE and BtuC, and for two or three replicates for GlpG, and averages and SEM were calculated (see Source data 1). Note that for two replicates plotting the average ± SEM is equivalent to plotting the average ± error bars representing the two experimental measurements.

In vitro transcription/translation was performed using the commercially available PURExpress system (New England Biolabs). Reactions were mixed according to the manufacturer’s recommendations by the addition of 2.2 μl of linear DNA of each construct giving a final volume of 10 μl. Polypeptide synthesis was carried out in the presence of [³⁵S]-methionine at 37°C for 15 min under 700 rpm shaking. Translation was stopped by the addition of TCA to a final concentration of 5% and incubated on ice for at least 30 min. Total protein was sedimented by centrifugation at 20,000 g for 10 min at 4°C in a tabletop centrifuge (Eppendorf, Germany). The pellet was resuspended in 2× SDS/PAGE sample buffer, supplemented with RNaseA (400 μg/ml) to digest the stalled peptidyl-tRNA, and incubated at 37°C for 15 min under 1000 rpm agitation. The samples were resolved on 12% Bis-Tris gels (Thermo Fisher Scientific) in MOPS buffer. Gels were dried on Hoefer GD 2000 dryer (Hoefer, US), exposed to a phosphorimager screen for 24 hr, and scanned using the Fujifilm FLA-9000 phosphorimager for visualization of radioactively labeled protein species.

Computer simulations of cotranslational membrane integration were carried out using a previously developed and validated CGMD model in which nascent proteins are mapped onto CG beads representing three amino acids (Niesen et al., 2017; Niesen et al., 2018). The nascent protein interacts with the Sec translocon and the ribosome via pairwise interactions that depend on the hydrophobicity and charge of the beads of the nascent protein. The interaction parameters are unchanged from previous work (Niesen et al., 2017). The lateral gate of the translocon switches between the open and closed conformations with probability dependent on the difference in free energy between the two conformations. The structures of the ribosome and translocon are based on cryo-EM structures and, aside from the lateral gate of the translocon, are fixed in place during the simulations. The lipid bilayer and cytosol are modeled implicitly. The positions of the nascent protein beads are evolved using overdamped Langevin dynamics with a timestep of 300 ns and a diffusion coefficient of 253 nm²/s. Membrane potentials are included by adding an electrostatic energy term to the simulations, as previously described (Niesen et al., 2018).

To simulate protein translation, new amino acids are added to the nascent chain at a rate of five amino acids per second. Simulations of EmrE, GlpG, and BtuC begin with 12 amino acids translated. Translation continues until the nascent protein reaches the desired length, at which point translation is halted and forces on the C-terminus of the nascent chain are measured every 3 ms for 6 s. This methodology has been found to accurately reproduce experimental FPs (Niesen et al., 2018). Forces are measured starting at a nascent protein length of 18 amino acids for EmrE and BtuC, and 70 for GlpG. The computational force profile (CGMD-FP) is then obtained by measuring the forces at lengths incremented by four amino acids. Simulations at different lengths are performed independently and repeated 100 times. Because the ribosomal exit tunnel is truncated in the CGMD model, a shift in the protein index is required to compare simulated and experimental results. Shifts of −12, –5, and −5 residues are used for EmrE, GlpG, and BtuC CGMD-FPs, respectively. The shifts are estimated by aligning the computational and experimental FPs and are in line with what is expected given the length of the truncated exit tunnel. Variation in the shift may reflect different degrees of compaction of the nascent chain. Although previous work provides a framework to estimate the experimentally observed fraction full length from simulated forces given a specific AP (Niesen et al., 2018; Tian et al., 2018), forces are reported directly to facilitate comparison between experiments performed with different APs.

The relative contact order for the GlpG NTD was calculated using the Contact Order server at https://depts.washington.edu/bakerpg/contact_order/contact_order.cgi.