Cotranslational folding cooperativity of contiguous domains of α-spectrin

Significance

In living cells, proteins are produced in a sequential way by ribosomes. This vectorial process allows the growing protein chain to start to fold before translation has been completed. Thereby, cotranslational protein folding can be significantly different than the folding of a full-length protein in isolation. Here, we show how structurally similar repeat domains, normally produced as parts of a single long polypeptide, affect the cotranslational folding of their neighbors. This provides insight into how the cell may efficiently produce multidomain proteins, paving the way for future studies in vivo or with chaperones. We also provide an estimated magnitude of the mechanical force on the nascent chain generated by cotranslational folding, calculated from biochemical measurements and molecular dynamics simulations.

Abstract

Proteins synthesized in the cell can begin to fold during translation before the entire polypeptide has been produced, which may be particularly relevant to the folding of multidomain proteins. Here, we study the cotranslational folding of adjacent domains from the cytoskeletal protein α-spectrin using force profile analysis (FPA). Specifically, we investigate how the cotranslational folding behavior of the R15 and R16 domains are affected by their neighboring R14 and R16, and R15 and R17 domains, respectively. Our results show that the domains impact each other’s folding in distinct ways that may be important for the efficient assembly of α-spectrin, and may reduce its dependence on chaperones. Furthermore, we directly relate the experimentally observed yield of full-length protein in the FPA assay to the force exerted by the folding protein in piconewtons. By combining pulse-chase experiments to measure the rate at which the arrested protein is converted into full-length protein with a Bell model of force-induced rupture, we estimate that the R16 domain exerts a maximal force on the nascent chain of ∼15 pN during cotranslational folding.

Classically, protein folding has been studied in vitro using purified proteins, providing a detailed picture of how the biophysical properties of a given polypeptide chain affect folding pathways. However, studies of purified proteins do not reflect the vectorial nature of biological protein synthesis, where each amino acid is added sequentially on a timescale that allows the growing protein to explore its continually expanding energy landscape (1).

The discovery (2, 3) and engineering (4, 5) of translational arrest peptides (APs) have provided us with a tool to examine how proteins fold cotranslationally. To date, the folding of several small proteins and protein domains has been studied using the force-sensitive AP from the Escherichia coli SecM protein, establishing a method we have called force profile analysis (FPA) (6–15). Recently, we demonstrated that FPA faithfully picks up cotranslational protein folding events observed by direct biophysical measurements (15).

Spectrin, the primary structural component of the cellular cytoskeleton (16–19), is composed of two heterodimers of α- and β-spectrin that come together in a coiled tetramer (20). The 285-kDa nonerythroid αII isoform of spectrin contains 20 three-helix bundle domains, which provide interesting model systems for protein folding. This is because, while they are structurally nearly identical (21), they vary in stability (22–25), folding rate (26, 27), and folding mechanism (28). In particular, folding of the 15th, 16th, and 17th α-spectrin domains (herein called R15, R16, and R17, respectively) has been extensively studied in vitro (24, 27, 29–32).

Previously, using FPA, we demonstrated that R15 and R16 are able to fold cotranslationally and that the point at which this folding begins does not seem to be influenced by the in vitro folding rate (9). Specifically, while R15 folds very quickly in vitro, cotranslational folding of R16 starts at shorter chain lengths than does folding of R15. Indeed, even when the folding core of R15 was substituted into R16, which gives R16 the folding properties of R15 in vitro (33), the protein was still able to fold earlier than R15 during translation. When the minimal, five-residue R15 folding nucleus (34) was substituted into R16, this led to late cotranslational folding, but early folding was restored when this mutant was expressed in tandem with R15, as it would be in vivo. This demonstrated that the mutant R16 variants fold cotranslationally via early intrachain contacts, at chain lengths where the R15 folding nucleus cannot yet form. This was not observed in vitro because the purified mutant R16 protein would preferentially fold using the “fast” R15 nucleus rather than the slower R16 contacts. Cotranslationally, R16 folds using the first available pathway, i.e., the slower R16 folding contacts.

Here, we have extended our studies on spectrin to determine how the presence of neighboring domains affect the folding of R15 and R16. We find that the cotranslational folding of these two domains is differently affected by their respective upstream and downstream neighbor domains, a previously unappreciated feature of spectrin folding. We were also able to observe that the synthesis of native, likely helical, structures downstream of the folding spectrin domain increases the observed folding force and further stabilizes the fold even in the presence of amino acid substitutions that destabilize the final native spectrin structure.

While it has been shown previously that the cotranslational folding of proteins can provide a mechanical force to release SecM-mediated ribosomal stalling (6), the magnitude of the force has never been directly measured by FPA. Using pulse-chase experiments, we have now measured the rate at which arrested protein is converted to full-length protein under standard experimental conditions. These rates were fit to a Bell model (35) for force-induced rupture, using parameters elucidated by molecular dynamics (MD) simulations (12), to provide an estimate of the magnitude of the pulling force exerted on the nascent chain by cotranslational protein folding. The determined force is in good agreement with previously published measurements using optical tweezers (6).

Results

The FPA Assay.

APs are short stretches of polypeptide that interact with the ribosome exit tunnel in such a way that translation is stalled when the last codon in the portion of the mRNA that codes for the AP is located in the ribosomal A-site (36). APs generally have regulatory functions in cells, controlling, e.g., translation initiation on downstream open reading frames in polycistronic mRNAs (36). Interestingly, the stalling efficiency of many APs has been shown to be sensitive to pulling forces exerted on the nascent polypeptide chain, with high pulling force leading to reduced stalling (6, 37, 38). APs can therefore be used as force sensors, reporting on cotranslational events that in one way or another generate pulling forces on the nascent chain (7, 38, 39). The factors influencing such force-generating events were recently examined using molecular simulations and statistical mechanics models, in particular the effect of translation speed (40). Although AP measurements of force do not include the effect of translation speed directly, this could be accounted for by using a suitable kinetic model for coupled folding and translation (41).

In FPA, a force-generating domain in a polypeptide is placed at increasing distances upstream of an AP (Fig. 1A), and the degree of translational stalling is measured for the corresponding series of protein constructs. Fig. 1B shows how FPA can be applied to study the cotranslational folding of soluble protein domains, e.g., tandem spectrin domains. In the construct shown on the left, the chain is long enough for one of the two spectrin domains to have already folded, while the second spectrin domain is largely buried in the exit tunnel and thus cannot fold when the ribosome reaches the C-terminal end of the AP. Therefore, little force is generated on the AP, and stalling is efficient. In the right-hand construct, the chain is so long that both spectrin domains have folded prior to the ribosome reaching the C-terminal end of the AP, and again little force is generated on the AP. In the middle construct, however, the chain is just long enough that the second spectrin domain can begin to fold if the linker that connects it to the AP is stretched beyond its equilibrium length. Under this condition, some of the free energy gained upon folding will be converted to tension in the linker and generate a pulling force on the AP, resulting in reduced stalling and increased production of full-length protein.

Fig. 1.

FPA. (A) Schematic representation of a typical construct. A “force generator,” which here is one or more spectrin domains, is connected by a linker sequence to the E. coli SecM arrest peptide (AP). Following the AP is an unrelated protein sequence from the E. coli LepB protein that is added to allow the arrested (A) and full-length (FL) forms to be separable by SDS/PAGE. (B) Schematic representation of cotranslationally arrested ribosome–nascent chain complexes with increasing linker lengths (Left to Right). Represented here is the R15R16 nL construct (colored as in Fig. 2A). (Left) At short linker lengths, when the folding of the R16 domain (red) has not yet begun, the force acting on the nascent chain is minimal, and the arrest is efficient (F ∼ 0, f_FL ∼ 0). (Middle) At linker lengths that allow the folding of the R16 domain, the force acting on the nascent chain is increased and the arrest is less efficient (F > 0, f_FL > 0). (Right) At linker lengths that arrest the ribosome after the folding of the R16 domain has already occurred, the force acting on the nascent chain is again minimal (F ∼ 0, f_FL ∼ 0). (C) Determination of f_FL by SDS/PAGE. Lane 1 is the R15 nL39 construct, and lane 2 is the R15 nL39 full-length control construct. The control construct has the final (critical) proline residue of the AP mutated to alanine, preventing arrest and producing only full-length protein. (D) A plot showing the calculation of L_onset, L_max, and L_end. The data points used for the calculation are denoted by black, yellow, and gray circles, respectively. The interpolated midpoint between each pair of selected point gives the L_onset (vertical dotted line), L_max (dashed line), and L_end (dash-dot line) values.

The cotranslational folding process can hence be followed by measuring the amount of full-length and arrested protein product at each linker length (Fig. 1C). Folding transitions will appear as peaks in a “force profile” plot of the fraction full-length protein (f_FL) vs. linker length (L) (Fig. 1D) and can be described by the peak amplitude and the linker lengths that define the onset (L_onset), maximum (L_max), and end (L_end) of the peak, as shown.

Cotranslational Folding of Tandem Spectrin Domains.

In the intact spectrin protein, the C-terminal α-helix in one domain is continuous with the N-terminal α-helix in the next domain (Fig. 2A). Consecutive domains are thus intimately connected to each other and can influence each other’s thermodynamic stability (24, 27, 31, 42). In order to better understand the cotranslational folding of spectrin, we decided to systematically evaluate the effects of upstream and downstream domains on the force profiles of the R15 and R16 spectrin domains by analyzing the two- and three-domain combinations shown in Fig. 2B. For each domain combination, a force profile was recorded by generating a series of constructs where the 17-residue SecM(Ec) AP (FSTPVWISQAQGIRAGP) from the E. coli SecM protein was placed at different linker lengths L from the C-terminal end of the R15 or R16 domain, translating each construct for 20 min in an in vitro translation system (43) in the continuous presence of [³⁵S]methionine, separating the translation products by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS/PAGE), and calculating f_FL from the intensities of the bands corresponding to the full-length and arrested forms of the protein. The full amino acid sequences of all constructs are given in SI Appendix, Table S1.

Fig. 2.

Construct design. (A) Structure of the R15 (blue), R16 (red), and R17 (orange) domains of chicken brain α-spectrin (Protein Data Bank ID code 1U4Q) (24). The arrows indicate the domain boundaries and the A, B, and C helices of each domain are labeled. (B) Constructs analyzed in this study. A solid black line represents sequence derived from the LepB protein, and dashed boxes represent truncated spectrin sequences. The ends of the arrested (A) and full-length (FL) forms are indicated.

In order to distinguish which spectrin repeat is being assayed in the tandem constructs, we use the following naming convention. Constructs named “nL” (for LepB linker) are used to follow the folding of most C-terminal spectrin repeat (i.e., R15R16 nL follows the folding of R16) and constructs named “nT” (for Truncated spectrin linker) follow the folding of the penultimate spectrin repeat (i.e., R15R16 nT follows the folding of R15).

The results are shown in Fig. 3 A–C. These include previous measurements collected for R16 nL and R15R16 nL (9) along with additional replicates and newly collected data for R15 nL. To facilitate comparison of the results, the L_onset, L_max, and L_end values extracted from each force profile are summarized in SI Appendix, Fig. S1; note that L_max values for force profiles that reach saturation (ƒ_FL ∼ 1) cannot be accurately determined. As seen in Fig. 3B, the cotranslational folding of the R15 domain is unaffected by the presence of the upstream R14 domain (dark blue and light blue curves), while the presence of the N-terminal part of the downstream R16 domain induces a reduction in L_onset from 36 to 34 residues and a marked increase in the amplitude of the peak (compare the dark blue and purple curves). Thus, the onset of folding of R15 is sensitive to the presence of the early parts of the N-terminal α-helix of the downstream R16 domain but is not affected by the presence of the upstream R14 domain.

Fig. 3.

Cotranslational folding of spectrin domains. (A) Force profiles obtained with the SecM(Ec) AP for the R15 nL and R16 nL constructs. ƒ_FL is the fraction full-length protein produced, and L is linker length (i.e., the number of amino acid residues between the C-terminal end of the spectrin domain and the C-terminal Pro residue in the AP). Data for R16 nL constructs from ref. 9 along with additional replicates are presented. (B) Force profiles obtained with the SecM(Ec) AP showing the influence of R14 and R16 on the folding of R15. (C) Force profiles obtained with the SecM(Ec) AP showing the influence of R15 and R17 on the folding of R16. Data for R15R16 nL from ref. 9 along with additional replicates are presented. The reason for the abnormally low ƒ_FL value for the R16R17 nT construct at L = 47 is unknown. (D) Force profiles obtained with the SecM(Ec-sup1) AP showing the influence of R15 and R17 on the folding of R16. Note that the R15R16 nL and the R15R16R17 nT profiles overlap perfectly in the interval L = 23 to 29 residues.

For the R16 domain, the effects of the neighboring R15 and R17 domains are more dramatic (Fig. 3C). The presence of the upstream R15 domain leads to a reduction in L_onset from 30 to 28 residues (compare the red and magenta curves), as shown before (9). The presence of the downstream R17 domain does not appreciably affect L_onset but leads to an increase in amplitude and a shift to a higher L_end value (compare the red and orange curves). Finally, when flanked by both the R15 and R17 domains, the R16 folding transition starts at a lower L_onset and ends at a higher L_end than for the isolated R16 domain (compare the red and brown curves).

In order to more precisely define L_max for constructs where f_FL approaches 1, we substituted two amino acids in the relatively weak SecM(Ec) AP to create the stronger SecM(Ec-sup1) variant (FSTPVWISQAPPIRAGP) (44) in the R16 nL, R15R16 nL, R16R17 nT, and R15R16R17 nT constructs (Fig. 3D). Again, the presence of R15 reduces L_onset and L_max by approximately two residues (note that the R15R16 nL and the R15R16R17 nT profiles overlap perfectly in the interval L = 23 to 31), and the presence of R17 increases L_end by six to seven residues (SI Appendix, Fig. S2). Notably, the bimodal shape of the R15R16R17 nT profile with peaks at L = 29 and 37 residues is an almost perfect match to the sum of the R15R16 nL and R16R17 nT profiles, SI Appendix, Fig. S2F, suggesting that the R16 part folds first (stabilized by the R15 C helix; cf. Fig. 2A), followed by a second folding event involving the R17 A helix (stabilized by the R16 C helix).

Interestingly, both R16 folding constructs that include parts of R17 (R15R16R17 nT and R16R17 nT) show an increase in f_FL at the longest linker length (L = 61 residues) (Fig. 3C). At this point, the entire R17 A helix, the loop, and the very beginning of R17 helix B have been translated (SI Appendix, Table S1). This increase in f_FL may thus signal an early interaction between helices A and B in R17.

As a control, we introduced mutations that are known to inhibit folding of R15 and R16 in vitro (45, 46) into the R15 nL, R15R16 nT, R16 nL, R16R17 nT, and R15R16R17 nT constructs (Fig. 4 and SI Appendix, Fig. S3). As expected, f_FL values for the “nonfolding” R15(nf) nL and R16(nf) nL constructs were strongly reduced, but surprisingly, f_FL values remained high for linker lengths between L_max and L_end for R15(nf)R16 nT, R16(nf)R17 nT, and R15R16(nf)R17 nT constructs. Since the difference between the nL and nT series of constructs is the presence of a C-terminal “linker” derived from the N-terminal part of R16 or R17, respectively (replacing unstructured segments from LepB), this suggested that the linker itself, perhaps together with C-terminal parts of the upstream spectrin domain, might fold in the nf-mutants.

Fig. 4.

Linker effects in R15 folding. (A) Force profiles for R15 nL (blue, unbroken line) and R15R16 nT (purple, unbroken line) data from Fig. 3B compared with profiles for a R15 nonfolding (nf) mutant (F18D + I55D) in R15(nf) nL constructs (blue, dashed line) and R15(nf)R16 nT constructs (purple, dashed line). (B) R15R16 nT spectrin structure (L = 40) with the A–C helices indicated and the destabilizing mutations F18D and I55D in the R15A and C helices shown as yellow spheres. (C) Graph showing f_FL values (L = 40) for constructs with increasingly long R15 parts fused to the N terminus of the 19-residue R16 nT linker (black circles; points 1, 2, 3, 4, and 5). R15 nL L = 39 (point 6) and R15(nf) nL L = 41 (point 7) are shown for comparison. Structural representations of each construct (spectrin helices are colored as in B and LepB linker sequences colored green) surround the plot. For points 1 to 4, the effect of helix-breaking insertions in the R16 A helix (SI Appendix, Table S1) in f_FL values are shown as open circles (Gly–Gly), open dotted circles (Pro), and open dashed circles (Gly–Ser–Gly–Ser).

To investigate this possibility, we started from the R15R16 nT, L = 40 construct [whose corresponding nf-mutant has a much higher f_FL value than the R15(nf) nL version; Fig. 4A] and successively removed helices A, B, and C from the R15 part (i.e., N-terminal truncations) (Fig. 4B). We found that removing helix R15-A or both helices R15-A and R15-B decreased f_FL, but only to 0.47 and 0.36, respectively (Fig. 4C) (compare points 4, to 3, and 2). Only with the removal of all of R15, leaving only 19 amino acids of the R16 A helix, did f_FL decrease to baseline (point 1). Thus, the cotranslational formation of a continuous helical structure encompassing parts of the R15 C and R16 A helices appears to cause an increase in f_FL at L = 40; indeed, it is known from previous work that helices can form cotranslationally in the exit tunnel (47). The introduction of helix-breaking residues into the middle of the R16 A helix (SI Appendix, Table S1) in the R15-ABC/R16-A (point 4) and R15-BC/R16-A (point 3) constructs decreased the f_FL to the same level as R15-C/R16-A (point 2) (Fig. 4C, open circles). Likewise, comparison of constructs where the R16 A helix in the R15R16 nT and R15(nf)R16 nT at L = 40 was replaced by a nonhelical segment from LepB at a comparable length (L = 39 and L = 41, respectively) led to reductions in f_FL (compare points 5 and 6 to 4 and 7). Similar behavior was also observed for the R16R17 nT, L = 43 constructs (SI Appendix, Fig. S3), implying that the R17 A helix forms a cotranslational folding intermediate together with the R16 C helix, as suggested above.

We conclude that the cotranslational folding of the R15 domain is affected by its downstream but not by its upstream neighbor domain, and that folding of R16 is affected both by its upstream and downstream neighbors. Different spectrin domains thus not only fold via different folding mechanisms, but their cotranslational folding is also differently impacted by their upstream and downstream neighbor domains, a previously unappreciated feature of spectrin folding. We attribute the contribution of the downstream neighbor to the formation of helical structure encompassing helix C from the upstream domain and helix A from the downstream domain.

A Quantitative Relation Between f_FL and Pulling Force.

In all FPA studies published to date, the quantitative relation between the calculated f_FL values and the underlying pulling force acting on the nascent chain has remained undefined [although attempts have been made to derive it from simulations or other kinds of theoretical modeling (11, 39)]. Using the PURE translation system, Goldman et al. (6) showed that the interactions between the SecM(Ec) AP and the ribosome exit tunnel can be disrupted by a mechanical force applied through optical tweezers, and that the rate by which the translational stall induced by the SecM(Ec) AP is released, k_R, increases in step with the amount of force, F, applied to the nascent chain, in a way that can be approximated by the Bell model (35) for force-induced rupture:

$$k_R\approx k_0\exp \left[\frac{F\mathrm{\Delta }{x}^{‡}}{k_{B}T}\right],$$ [1]

where k₀ = 3 × 10⁻⁴ s⁻¹ (95% confidence interval [CI]: 0.5 × 10⁻⁴ s⁻¹, 20 × 10⁻⁴ s⁻¹) is the release rate at zero pulling force, and Δx^‡ = 0.4 nm (95% CI: 0.1 nm, 0.8 nm) is the distance to the transition state.

The work by Goldman et al. (6) suggested to us that approximating k_R with the rate of conversion of the arrested form of a given construct to the full-length form as measured in a pulse-chase experiment would allow us to estimate the corresponding pulling force using the relation in Eq. 1 between F and k_R. To this end, we did pulse-chase experiments on a range of spectrin R16 and ADR1a (7) constructs that have f_FL values between 0.2 and 0.9, and fit the pulse-chase data to a first-order kinetic equation (SI Appendix, Fig. S4).

To derive an analytical relation between f_FL and F, we reasoned that if f_FL were measured at a single delay time (Δt) in a pulse-chase experiment, rather than under the standard continuous-labeling experimental conditions, the release rate (and hence F, according to Eq. 1) could be estimated from the following:

$$f_{FL}=1-e^{-k_R\mathrm{\Delta }t}.$$ [2]

Since in the standard experiment translation can be initiated at any time during the 20-min labeling period, the delay time Δt varies from one ribosome–nascent chain complex to another. Although many factors could in principle contribute to the distribution of delay times, it turns out empirically that using an average delay time Δt = 550 s, approximately equal to half of the total incubation time, describes remarkably well the relation between the standard f_FL values (20-min continuous [³⁵S]Met labeling) and the release rates measured by pulse-chase experiments (SI Appendix, Fig. S5).

Combining Eqs. 1 and 2, one can solve for the pulling force F:

$$F\approx \frac{k_{\mathrm{B}}T}{\mathrm{\Delta }{x}^{‡}}\ln \left[\frac{\ln \left(1-f_{FL}\right)}{-k_0\mathrm{\Delta }t}\right].$$ [3]

To obtain good estimates of the Bell parameters k₀ and Δx^‡ we started from the values obtained in ref. 6 and performed a local optimization by requiring that k₀ (the release rate at zero force) is close to k_R measured for the “zero-force” construct R16 nL27 (Fig. 5A), and that the titin I27, spectrin R16, and ADR1a force profiles in ref. 12 are well reproduced by the MD simulation also described in ref. 12. These conditions are both fulfilled by setting k₀ = 3 × 10⁻⁴ s⁻¹ (i.e., the same value as in ref. 6 and equal to k_R for R16 nL27; Fig. 5A) and $x^{‡}$ = 0.65 nm (well inside the 95% CI from ref. 6) (SI Appendix, Fig. S6). With these parameter values and Δt = 550 s, Eq. 3 nicely captures the relation between the force estimated from the measured k_R and standard f_FL values (Fig. 5B) and can hence be used to predict F from f_FL. Over the interval 0.2 < f_FL < 0.9, the experimental data are also well approximated by the simple linear fit F = 22 f_FL – 2.2 (blue line). Obviously, Eq. 3 holds only for the 17-residue SecM(Ec) AP used here, and not for other APs of different stalling strengths (39). Presumably, Eq. 3 may be applied also in other contexts where the SecM AP is used to measure pulling forces, such as during membrane protein synthesis or translocation of charged residues across energized membranes (38, 39), although the parameters would first need to be verified to hold for in vivo experiments. We note that another approach to determining Bell parameters under a given set of conditions would be to repeat the FPA experiments using multiple APs with different resistance to force. A global fit of f_FL obtained for each AP as a function of linker length L would allow simultaneous determination of the forces exerted by the protein (independent of AP) as well as AP-dependent Bell parameters. In SI Appendix, Fig. S7, we illustrate this approach using data for translocon-mediated transmembrane helix insertion into the inner membrane of E. coli from in vivo AP measurements.

Fig. 5.

Release rates and estimation of pulling forces. (A) The rate of release (k_R) obtained from pulse-chase experiments (SI Appendix, Fig. S4 and Dataset S2), the fraction full-length protein (f_FL) measured under standard experimental conditions (20-min incubation in PURExpress in the continuous presence of [³⁵S]Met), and the pulling force F calculated using Eq. 1 ($k_0=3.0\times {10}^{-4}{\mathrm{s}}^{-1}$, $\mathrm{\Delta }{x}^{‡}=0.65\mathrm{nm}$). The constructs are from Kudva et al. (12) and are colored to match those in B and SI Appendix, Figs. S4 and S5. (B) F values calculated from Eq. 1 plotted against the standard f_FL values, with constructs colored as in A. The least-squares fit line is indicated by the blue line, and the analytic relation Eq. 3 between F and f_FL, assuming an average delay time Δt = 550 s (approximately equal to half the standard incubation time), is shown as a red curve.

The force F estimated from pulse-chase measurements or f_FL values represents the constant force that would have the same effect on the escape rate from arrest as the combined effect of the forces from both folded and unfolded states. We can compare this inferred force with the ensemble average force calculated from MD simulations, $\langle F_{MD}\rangle =p_U{F}_U+p_F{F}_F$, where $p_{U\left(F\right)}$ and $F_{U\left(F\right)}$ are the population of, and force exerted by, the unfolded (folded) state in the simulation, respectively. In previous work (12), we calculated both f_FL as well as $\langle F_{MD}\rangle$ directly from MD simulations. In SI Appendix, Fig. S8A, we show the relation between $\langle F_{MD}\rangle$ and f_FL from our previous simulations (12), which is in good agreement with the relation between F and f_FL calculated from Eq. 3. This agreement also confirms that the pulling forces are small enough that the force F calculated from simulated f_FL values using Eq. 3 is close to the ensemble-average force $\langle F_{MD}\rangle$ determined by MD simulations; the two forces are not equal in general because of the nonlinear relation between release rate and force, such that states exerting a larger force (e.g., the folded state) contribute disproportionately to the average release rate and hence to F (a direct comparison between $\langle F_{MD}\rangle$ from the MD simulations and F calculated from Eq. 3 is given in SI Appendix, Fig. S8B).

We note that the forces exerted by a protein folding as it exits the ribosome are qualitatively different from the tensile forces such as those exerted when proteins fold in atomic force microscopy (AFM) experiments [as have been performed on spectrin before (48)]. Therefore, the force magnitudes probed by the two experiments cannot be directly compared, despite their apparent resemblance. For example, even a small force exerted on the termini of an unfolded protein by an AFM can massively slow the refolding rate because of the large distance the protein must contract against this force in order to fold. The same force magnitude would have a much smaller effect on the refolding rate of a protein attached at one end to the ribosome (11).

Discussion

α-Spectrin contains over 20 repeat domains and is produced as a single, long polypeptide. Previously, we observed that the R16 domain starts to fold while a part of its C-terminal α-helix is still in the ribosome exit tunnel (9). We now find that the cotranslational folding of R16 is affected both by the presence of its upstream R15 domain and when the N-terminal part of the following R17 domain is present in the exit tunnel. In contrast, folding of the R15 domain is not affected by the upstream R14 domain, but starts at shorter linker lengths when the N-terminal end of the following R16 domain is present.

In vitro, R16 is stabilized by ∼1.7 kcal/mol by the presence of R15 (49), meaning that, during cotranslational folding, R16 can “sacrifice” a few interactions in the folding nucleus and start to fold at a shorter linker length in the R15R16 nL constructs (9).

For R15, the situation is a bit different because, in contrast to R16, its folding nucleus involves residues located at the C-terminal end of helix C (45). The presence of the upstream R14 domain thus may have little impact on L_onset, because the whole R15 domain anyway must have emerged from the exit tunnel before folding can start. On the other hand, the R15 C helix is stabilized by the presence of the early parts of the R16 A helix, possibly allowing the folding nucleus to form at a shorter linker length and reducing L_onset in the R15R16 nT constructs.

The picture that emerges is that the spectrin domains fold one after the other as they emerge from the exit tunnel, but not completely independently of one another. As seen for the R16 domain, the presence of an already folded N-terminal upstream neighbor domain not only can increase the thermodynamic stability of the folded state of an emerging domain, but can also allow the emerging domain to start folding while still partly buried in the exit tunnel. Likewise, when the C-terminal α-helix in the emerging domain is extended by a part of the N-terminal α-helix in the following domain, the folding transition can persist to longer linker lengths (as seen for R16) or start at shorter linker lengths (as seen for R15). Finally, the increase in f_FL seen for the R16R17 nT and R15R16R17 nT constructs at L = 61 residues (Fig. 3C) is suggestive of an early folding intermediate in R17. Further work is required to investigate this, but it may be possible that the regular helical structure of the spectrin repeats drives a nearly continuous folding reaction, punctuated by periods when the nascent chain is lengthened but no residues are added to the folded part emerging from the ribosome. In this scenario, at most a short stretch of unfolded polypeptide would be exposed outside the ribosome at any one time, which could minimize the need for protection of the nascent chain by chaperones.

Finally, a pulse-chase analysis has allowed us to measure the release rate k_R from the stalled state for different R16 and ADR1a constructs, making it possible to estimate the magnitude of the pulling forces exerted on the AP for different linker lengths (Fig. 5A). This in turn makes it possible to derive expressions for how the magnitude pulling force F depends on f_FL (as obtained from our standard continuous-labeling experimental protocol) (Eqs. 2 and 3). In general, we find that the folding of protein domains such as spectrin R16 and ADR1a can generate a maximal force of 15 to 20 pN on the AP (Fig. 5), in line with theoretical estimates based on MD simulations (40, 50).

Materials and Methods

All of the enzymes used for molecular biology were obtained from New England Biolabs. The PUREfrex in vitro transcription-translation system was produced by Eurogentec and purchased via BioNordika. The PURExpress in vitro transcription–translation system was purchased from NEB. GeneJet polymerase chain reaction (PCR) clean-up kit was purchased from Thermo Fisher Scientific. Precast NuPAGE gels and running buffers were purchased from Invitrogen. Homemade gels were cast using acrylamide-bisacrylamide mix, Tris, and glycine from VWR. Filter paper for gel drying was from Whatman. All other chemicals were purchased from Sigma.

Cloning.

The cDNAs for spectrin R15 and R16 domains were kindly provided by Jane Clarke, University of Cambridge, Cambridge, United Kingdom, and the spectrin R17 and R14 domains were ordered as DNA fragments from Eurofins Genomics GmbH. All protein sequences used in this publication are presented in SI Appendix, Table S1. We used two different construct types: those with an unrelated linker sequence, labeled nL, and those with a truncated spectrin linker sequence, labeled nT (Fig. 2B). For nL constructs, the respective spectrin domain or domains were cloned into the pET19b plasmid upstream from a Ser–Gly–Ser–Gly sequence attached to a linker sequence derived from the P2 domain of leader peptidase (LepB) followed by the 17-residue AP from the E. coli SecM protein (FSTPVWISQAQGIRAGP) herein referred to as SecM(Ec), and a further 23 amino acids derived from LepB. The shortest constructs (nL/nT 21) include only the SGSG sequence fused to the SecM(Ec) AP. For nT constructs, the respective spectrin domains were cloned into the same pET19b plasmid without a LepB linker sequence, but including the SGSG sequence immediately before the SecM(Ec) AP. The most C-terminal spectrin domain was then sequentially truncated via partially overlapping inverse PCR. For some constructs, a full-length (FL) control was created by changing the critical Pro at the C-terminal end of the SecM AP to Ala, thereby abolishing arrest (51) and yielding only full-length protein. The SecM(Ec-sup1) variants were created by site-directed mutagenesis of the constructs described above. N-terminal truncations of R15 and R16 were carried out by partially overlapping inverse PCR, and insertion of the helix breaking insertions was carried out by site-directed mutagenesis where the inserted residues were included in the primer sequence. All sequences were confirmed by DNA sequencing (Eurofins Genomics GmbH).

In Vitro Expression for FPA.

Expression and analysis was carried out as described previously (9). Briefly, a linear DNA product is created from each construct plasmid by PCR using Q5 polymerase with forward and reverse primers that anneal to the T7 promoter and terminator regions, respectively. Following PCR cleanup (using the manufacturer’s instructions), the product is confirmed by agarose gel electrophoresis. In vitro transcription and translation are carried out in either the PUREfrex of PURExpress commercial systems (mixed according to the manufacturer’s recommendations). One microliter of the PCR product and ∼10 µCi of [³⁵S]methionine are mixed for a 10-µL PUREfrex reaction or 0.5 µL of PCR product and ∼5 µCi of [³⁵S]methionine are mixed for a 5-µL PURExpress reaction, followed by incubation at 37 °C for 20 min at 600 rpm shaking. Translation is halted by the addition of an equal volume of ice-cold 10% trichloroacetic acid (TCA), followed by incubation on ice for at least 30 min. Total protein is sedimented by centrifugation at 4 °C for 5 min at 20,000 × g. The supernatant is carefully removed and the pellet is resuspended in a suitable volume of 1× SDS/PAGE sample buffer (62.5 mM Tris⋅HCl, pH 6.8, 10% glycerol, 2.5% SDS, 5% β-ME, 0.02% bromophenol blue, and 25 mM NaOH [NaOH is added to neutralize any remaining TCA]) by shaking at 37 °C and 1,000 rpm for at least 5 min. The prolyl-tRNA that remains attached due to SecM arrest is digested by the addition of 4 µg of RNase I (2 µL of a 4 µg/µL solution), followed by incubation at 37 °C and 600 rpm for 30 min.

Quantifications.

Following a brief centrifugation to remove any remaining insoluble material, the sample is loaded onto an appropriate SDS/PAGE gel (12% Tris-glycine gels were used for two- and three-spectrin domain constructs and 16% or 18% Tris-tricine gels were used for single-spectrin domains and the N-terminal truncations). Following electrophoresis, the gels are dried onto thick filter paper by heating under vacuum (Bio-Rad model 583 or Hoefer GD 2000), a radioactive molecular weight ladder included in the gel is visualized by spotting the filter paper with a ∼1:1,000 solution of [³⁵S]methionine in 1× SDS/PAGE sample buffer, and the gel is exposed to a phosphorimager screen for 12 to 72 h depending on the strength of the signal. The screen is imaged using a Fujifilm FLA9000 (50-µm pixels), and densitometry analysis on the resultant raw image (TIFF format) file is carried out using FIJI (ImageJ) software. The densitometry values are quantified using our in-house EasyQuant software and the fraction full-length protein, $f_{FL}=I_{FL}/\left(I_{FL}+I_A\right)$, is calculated from the intensities of the full-length (I_FL) and arrested (I_A) bands. See SI Appendix, Fig. S9, for examples of gels. Independent replicate in vitro translation reactions were carried out for all spectrin constructs in the folding peaks (Dataset S1 [191 unique constructs and 405 independent data points]). The majority of the data points in the folding peaks were collected in triplicate, with the remaining points collected in duplicate. A small number of data points outside of the folding peaks were collected as single measurements to save costs.

Pulse-Chase Experiments.

The rate at which translation recommences following the arrest of various constructs was measured for five R16 constructs and two ADR1 constructs (7) chosen to represent the range of f_FL values measured during typical experimental conditions. Pulse-chase experiments were carried out using the PURExpress system. The 75-µL reactions were mixed according to the manufacturer’s instructions, with the addition of [³⁵S]methionine. Following a 5-min incubation at 37 °C (pulse), an excess of unlabeled methionine was added, and incubation at 37 °C was continued. At discrete time points, a 10-µL aliquot of the reaction was removed and mixed with 15 µL of ice-cold 10% TCA and further processed as above. Three independent replicates were collected for each construct (Dataset S2). The rate calculated is for the conversion of arrested protein to full-length protein:

$$A\stackrel{k_R}{\to }FL.$$

The rate of this irreversible conversion can be calculated using a first-order equation:

$$f_A\left(t\right)=f_A\left(0\right)e^{-k_{R}t},$$

where f_A(t) is the fraction arrested protein at a given time, $t$, and f_A(0) is the fraction arrested protein at $t=0$. Prism 8 (GraphPad Software) was used to calculate the nonlinear regression fit of the kinetic equation to each dataset [R16 nL-27, -29, -31, -33, and -37, and ADR1 L25 and 27 (7)]. The rates were used to calculate the force exerted by the folding domain on the arrested nascent chain as detailed in the main text.

Optimization of the Bell Parameters.

The coarse-grained MD simulations of cotranslational folding of R16 and ADR1a constructs in WT, ΔuL23, and ΔuL24 ribosomes used here were originally reported in ref. 12. For each construct and each linker length, these simulations were used to calculate the ensemble average force $\langle F_{MD}\rangle =p_U{F}_U+p_F{F}_F$, where $p_{U\left(F\right)}$ and $F_{U\left(F\right)}$ are the population of, and force exerted by, the unfolded (folded) state in the simulation, respectively. In order to determine the Bell parameters k₀ and $\mathrm{\Delta }{x}^{‡}$ in Eq. 1 in the main text, we set k₀ = 3 × 10⁻⁴ s⁻¹ (i.e., the same value as estimated in ref. 6, and equal to k_R for the “zero force” construct R16 nL27; Fig. 5A) and explored a range of $\mathrm{\Delta }{x}^{‡}$ values around the approximate value $\mathrm{\Delta }{x}^{‡}$ = 0.4 nm given in ref. 6. For each $\mathrm{\Delta }{x}^{‡}$ value, we calculated an average k_R over unfolded and folded states as $k_R=p_U{k}_0\exp [F_U\mathrm{\Delta }{x}^{‡}/k_{\mathrm{B}}T]+p_F{k}_0\exp [F_F\mathrm{\Delta }{x}^{‡}/k_{\mathrm{B}}T]$and then obtained f_FL (setting Δt = 550 s) from Eq. 2 in the main text. The simulated f_FL values (i.e., the simulated force profiles) obtained in this way were then compared with the experimental force profiles reported in ref. 12 for the I27 and R16 domains expressed with WT, ΔuL23, and ΔuL24 ribosomes, and for the ADR1a domain expressed with WT and ΔuL24 ribosomes. The parameter values k₀ = 3 × 10⁻⁴ s⁻¹, Δx^‡ = 0.65 nm were found to give the best fit between the simulated and experimental force profiles (SI Appendix, Fig. S7).

Data Availability.

The quantified raw data points are provided in Datasets S1 and S2.