Folding speeds of helical membrane proteins

The biogenesis of multipass helical membrane proteins involves four primary steps [1]: (1) the targeting of a nascent polypeptide chain to eukaryotic endoplasmic reticulum (ER) or bacterial plasma membrane, (2) the insertion of the chain into lipid bilayers with an appropriate topology and formation of transmembrane (TM) helices, (3) the folding into the functional tertiary structure mainly through the lateral packing of TM helices, and, (4) if any, the assembly of the tertiary folds into a higher-level complex. The middle two steps are commonly regarded as constituting the core folding process. In the co-translational folding pathway [1–5], the helix insertion is mainly mediated by the Sec translocon complex while being translated from a ribosome bound to the complex (the first-stage folding). The helices then dissociate from the ribosome–translocon complex and associate with each other until adopting the compact native fold (the second-stage folding).

The helix packing interactions may also initiate early on the ribosome–translocon complex [3,4]. Unless there is a strong packing cooperativity across the entire protein, the two folding steps would occur simultaneously along the pathway [2]. This coupling is likely a prominent feature, particularly for large membrane proteins with clearly separated, multiple TM domains [6–9]. The extent of coupling between the two stages and the resulting time scales of the entire folding process will depend on various factors, including the rates of translation and insertion, rate and stability of helix packing interactions, protein size and independence of folding domains, and specific local/global folds like re-entrant helices and knots. Accessory proteins such as insertases and chaperones can participate in the overall folding process as essential components, including those like Oxa1 superfamily insertases and calnexin [1,10]. In this review article, I will focus on simplified in vitro folding studies that offer quantitative evaluations of folding speeds of helical membrane proteins.

The folding process of helical membrane proteins may manifest in three primary scenarios based on the timing and stability of helix packing interactions (Figure 1). In the sequentially accrued folding, the inserted TM helices accumulate early and steadily within lipid bilayers, maintaining stable structures throughout the translation and insertion (Figure 1A). For this case, the translation event would serve as a principal limiting factor on overall folding speeds, given the slow elongation of polypeptide chains at 2–17 amino acids per second (6–50 s for 100 amino acids) [11–18]. The second layer of the folding speed constraint involves the helix insertion into lipid bilayers. The insertion rate is slightly faster than the elongation rate, roughly ranging from 10 to 20 amino acids per second (1–2 s for a TM helix) [3,4,13,19]. The time scale of seconds required for helix insertion is still much slower than a hundred of milliseconds needed for the formation of a helical hairpin, which is a preferred folding unit [2,6,20], and its subsequent packing on the structured template within lipid bilayers [21].

In case of the strongly cooperative folding for all TM helices, the helix packing and folding transition occurs as a single step after the complete translation and insertion of all helices (Figure 1B). In contrast, in a more realistic scenario of the transiently assembled folding, the TM helices frequently form less stable intermediate structures that rapidly dissociate during the translation and insertion until ultimately adopting the final stable fold (Figure 1C). For these folding scenarios, the entire folding times would not be solely limited by the earlier events of translation and insertion, but likely be delayed compared with what is expected for the sequentially accrued folding. The second-stage folding can be further set back by various other factors, such as alterations in initial helix topology [22–27], back-tracking events [28], the need for interplay with chaperone proteins and/or cofactors [1,7,29–31], and the cooperativity enhancement and retardation effect by lipid bilayer properties [32–37].

Previous in vitro folding studies offer intriguing insights suggesting that the prevalence of delayed and/or slow second-stage folding may be a characteristic feature in many helical membrane proteins. For instance, a rhomboid protease GlpG mainly comprising six TM helices was shown to undergo cooperative folding from the unfolded helical state [38–40]. The entire folding time for GlpG was notably slower compared with the helical hairpin formations within lipid bilayers, by a factor of 20–30 ([21,39]; see the section Folding speeds under mechanical unfolding approaches for more details). Despite its relatively small size with four-TM helices, peripheral myelin protein 22 (PMP22) was observed to undergo remarkably slow folding, taking up to an hour at very low denaturant concentrations [41,42]. This slowness may be attributed to intrinsic slow folding kinetics and/or marginal conformational stability [41,42]. The entire half domain of lactose permease (LacY) can be flipped across the lipid bilayer even after the translation and insertion are completed, taking up to several minutes [25].

These illustrative cases evoke the original two-stage folding model [43,44], where the insertion of helices is followed by their association with clean separation, as a practically useful framework in folding studies despite its simplicity. Several in vitro methods have characterized the folding time scales of selected model membrane proteins, with a primary focus on the second-stage folding or spontaneous insertion coupled with helix formation and packing (Figure 2 and Table 1). As detailed methodological reviews can be found elsewhere [6,48–50], only their major features, relevant folding steps, and caveats in interpretations are briefly outlined in the following sections. The cell-free expression approaches employing surface-enhanced infrared absorption spectroscopy (SEIRAS) [29,51] are omitted in this review, as their exceptionally prolonged time scales of several hours may involve stochastic, non-synchronized initiation of protein translation, spontaneous insertion, and structure formation.

In chemical denaturation methods (Figure 2A), membrane proteins in detergent micelles, mixed detergent/lipid micelles, or vesicle bilayers are initially denatured by chemical denaturants like sodium dodecyl sulfate (SDS) or urea [10,52–55]. The denatured proteins with full or partial helical content are then refolded by diluting the denaturant concentration. The observed rate constants at different denaturant concentrations are determined by monitoring the transitions through (un)folding reporters, such as intrinsic fluorescence, protein absorbance, circular dichroism spectra, and pulse proteolysis. These values are extrapolated to zero denaturant concentration. Folding time scales for helical membrane proteins with four to seven TM helices, including disulfide bond formation protein B (DsbB), GlpG, and bacteriorhodopsin (bR), were estimated to fall within the range of 1–10 s in detergent micelles or mixed detergent/lipid micelles [40,45–47] (Table 1). The interpretation of kinetics can sometimes be complicated by a less well-defined, chemically denatured state, especially concerning the locations of unstructured parts and the topology of helices in the reconstituted solvent space. The extrapolation analysis may introduce non-negligible errors in the estimated folding speeds, especially for bR, as it is almost completely folded at low denaturant concentrations, requiring a long extrapolation from high denaturant concentrations [38,56].

The chemical denaturants themselves can impact the folding kinetics by preferentially binding to specific protein sites and altering the solvent space reconstituted for membrane proteins [2,6,57]. The smaller size of mixed detergent/lipid micelles was shown to enhance the folding rates, possibly by constraining the conformational search space during folding [58]. Additionally, the global properties of lipid bilayers can influence the folding speeds as well as the stability of native folds and denatured state ensembles [34,59–61]. For example, an increase in lateral chain pressure has been suggested to slow the formation rates of folding intermediates of bR from SDS-denatured states [34–37]. A recent magnetic tweezer study also indicates that the high viscosity of lipid bilayers significantly retards the helix packing interactions [32], likely serving as a physical origin for the chain pressure effect (see the section Folding speed limit of helical membrane proteins for more details).

Employing fluorescence resonance energy transfer (FRET) as a spectroscopic ruler [62,63], Krainer et al. observed remarkably rapid folding within time scales of 35–670 μs for a membrane-associating protein Mistic in various detergent micelles [63]. This process initiated from the urea-denatured state with partial helical content. While the ultrafast folding of Mistic is intriguing, its underlying mechanism may not be relevant to helix packing interactions within the membrane mimetics, as the protein is more likely to interact with the membrane surfaces [64]. The absence of characteristic signals representing the correct folding in this approach also raises a question about how non-native states with close contacts can be excluded when assigning the native state.

Single-molecule force spectroscopy (SMFS) can more precisely determine the structural states during unfolding or refolding by analyzing molecular extension with force-dependent polymer models [6]. Mechanical force as a denaturant is also decoupled from the chemical forces driving folding and possible chemical disruption of lipid bilayers [2,6]. In atomic force microscopy (AFM)-based SMFS (Figure 2B), a single membrane protein in a supported lipid bilayer is partially unfolded to an unstructured polypeptide by vertically pulling the attached AFM tip. The partially unfolded protein, with the last helical hairpin still embedded in the bilayer, is refolded by lowering the tip to the membrane surface within 5–10 nm and then waiting for up to tens of seconds with the tension relaxed. For the Na⁺/H⁺ antiporter NhaA, the preferred folding unit of a helical hairpin was observed to refold in a time scale of 1–3 s, estimated as the mean waiting time [65]. A helical segment containing the Na⁺-binding site showed a much faster folding rate of ∼20 ms. A high-speed AFM probed the rapid refolding transitions of three-residue segments of bR with tens of μs under high force conditions at ∼100 pN [66]. In these AFM approaches, the measured time scales are relevant to the spontaneous insertion of unstructured polypeptide portions coupled with helix formation and packing onto the remaining structured template [6,50]. The biased vertical pulling at only one end of the N- or C-terminus introduces variations in refolding pathways and overall folding time scales. The C-terminal pulling design is considered a better mimic of the native insertion and folding process [6].

In single-molecule magnetic tweezers (Figure 2C), the force is applied laterally along the membrane surface at both the N- and C-terminal ends, thereby excluding the potential effect of the biased vertical pulling geometry in AFM. The applied force can remain nearly constant during refolding due to an extremely small spring constant $(k=ΔF/Δz∝ΔF≪1)$⁠, even in a low force range of a few to tens of pN [6]. These inherent system properties enable the capture of the unfolded helical state, where the helices adopt a zigzag configuration within a lipid bilayer at 5–12 pN [21,32]. With this structural state, our focus can be primarily directed towards the helix packing interactions uncoupled from the initial insertion events during refolding, although the helices would not fully traverse the lipid bilayer to the opposite side.

To estimate folding speeds using the tweezer method, mean dwell times at a particular state before transitioning to another state were measured under various forces and then extrapolated to zero force using force-dependent kinetic models. GlpG was shown to predominantly fold in a helical hairpin unit within a bicelle lipid bilayer, exhibiting a time scale of 110–140 ms for each transition [21] (Table 1 and Figure 3A). Despite the applied low force level of 5–7 pN, potential errors in the estimated kinetics may arise from the narrow force window of only 3 pN. Moreover, during the helix packing interactions, the helices effectively rotate within the membrane and the soluble parts fully cross the membrane, possibly influencing the relevant kinetics [32].

The entire folding time of GlpG is notably slower than the intermediate packing transitions, extending to ∼3 s [21,39] (Table 1). The large difference in time scales implies repetitive occurrences of formation and dissociation of transient intermediate structures before the final transition to the stable native fold (Figure 3A). Indeed, time-resolved extension traces of GlpG in the low force range demonstrated the frequent, conformational dynamics that eventually disappeared once the protein correctly folded [21]. Based on dissociation time scales, the first intermediate structure of the N-terminal helical hairpin (I₁) would dissociate five times more frequently than the second intermediate structure with two helical hairpins packed (I₂). Especially, the dissociation time scale of I₂ (∼3.5 s) is comparable to the entire folding time scale (∼3 s), suggesting that once the I₂ state is formed, it serves as a template for packing the last C-terminal helical hairpin (Figure 3A). However, the reverse transition from I₂ to I₁ is still much faster than the translation and insertion, supporting a possibility that the case of transiently assembled folding is more plausible for GlpG among the three possible folding scenarios (Figure 1C). Furthermore, adjusting the lipid composition from DMPC:DMPG = 70:30 mol% to 100 mol% DMPC resulted in an ∼10-fold increase in the entire folding time of GlpG [21,39] (Table 1). Although the alteration in lipid head group may primarily impact the folding of soluble and interface parts, the exact mechanism of the retardation remains uncertain.

It has been consistently observed that large transporters with two TM domains, such as LacY, melibiose permease (MelB), and glucose transporter 3 (GLUT3), are prone to misfolding and/or poor folding in the absence of Oxa1 superfamily insertases [7]. There are hardly any systematic studies examining how the insertases modulate the folding speeds of helical membrane proteins. In studies using AFM, LacY and MelB showed increased overall yields of coupled insertion and folding in the presence of the YidC insertase [68,69]. Yet, it is unclear whether the increased yields with YidC arise from the acceleration of the insertion and folding, prevention of trapping in misfolded states, or a combination of both factors. Likewise, in a study using magnetic tweezers, the folding yield of GLUT3 significantly increased with another Oxa1 superfamily member, the ER membrane protein complex (EMC), in conjunction with phosphatidylethanolamine (PE) lipids [8]. The first passage time from the unfolded state to the fourth intermediate state decreased as the concentration of EMC increased, suggesting that the insertase expedited relevant transitions. However, it is not clear which event was more predominantly affected by EMC during the transitions, such as the helix rotation, soluble part crossing, and helix packing, which were described in the previous section.

The so-called folding speed limit has been extensively explored for water-soluble proteins, ranging roughly from 1 μs to 1 ms [70], whereas only recently has it been studied for membrane protein folding [32]. In the Kramers rate framework, represented as τ_folding = τ_ω·exp(ΔG^†/k_BT) [71,72], the pre-exponential factor (τ_ω) offers a rough estimate of the fastest possible folding time, serving as the upper limit of protein folding speeds [70,73]. Recently developed magnetic tweezers with highly stable molecular tethering have estimated a τ_ω value for helical membrane proteins, corresponding to their folding speed limit [32].

Using the robust tweezer method, Kim et al. probed persistent transitions between four structural states of a designer four-TM helical bundle (scTMHC2) over an extended 9-h period at a constant force [32]. From the observation of numerous transitions, the authors independently characterized the energy barrier heights (ΔG^†) and the folding times (τ_folding) between the structural states, while correcting for possible instrumental and molecular system errors. The application of the Kramers formula yielded a τ_ω time scale of ∼20 ms for the preferred folding unit of a helical hairpin, representing the speed limit of helical hairpin formation within the lipid bilayer of DMPC bicelles [32] (Table 1). As described in the following paragraph, the 20-ms time scale would likely be applicable to other helical hairpins with different residue sequences. It may also be regarded as an estimate for the folding speed limit of entire helical membrane proteome under the specific membrane mimetics.

The estimated time scale of ∼20 ms corresponds to an exceedingly slow range compared with that of soluble proteins of similar size by a factor of ∼10⁴ [32,70], likely attributable to the highly viscous nature of lipid bilayers [32]. Since solvent friction is expected to be the primary determinant for τ_ω in the highly viscous medium [74], the speed limit estimation would be less sensitive to residue sequences, although slight sequence-dependent modulations may exist. Additionally, a helical hairpin serves as the minimal folding unit for the tertiary structures of helical membrane proteins [2,20,62]. Larger proteins will thus fold through the packing of additional helical segments onto the initial structural template [2,8,21], taking longer times for complete folding. The helical hairpin formation and additional packing interactions are not physically distinct processes that both occur within lipid bilayers. Lastly, the earlier events of translation and insertion and the other delaying factors as mentioned above can further impede the overall folding process.

The same time scale is also applicable to the speed limit of the reverse unfolding transitions, as a similar τ_ω value of ∼20 ms was obtained for helical harpin dissociation [32]. The viscosity of the DMPC bicelles and the resultant speed limit estimates would not be substantially altered for other lipids with the same acyl chain and similar bilayer-forming tendency, such as DMPG [60,61,67]. It should be noted, however, that other membrane mimetics and lipid compositions with different viscosity levels may non-negligibly change the speed limit time scales.

The prior simplistic application of a speed limit range for soluble protein folding likely led to an overestimation of the energy barrier heights associated with (un)folding in helical membrane proteins [21,39,75]. For instance, an energy landscape reconstruction for the second-stage folding of GlpG revealed three transition states with substantial barriers between four structural states [21] (Figure 3B). Each transition primarily involves the formation or dissociation of a helical hairpin (Figure 3A). By adopting the more accurate value of 20 ms for the speed limit time scale, specifically estimated for helical membrane proteins [32], the barrier heights of the transition states are significantly reduced by ∼7.6 k_BT (Figure 3C). Consequently, the previously recorded folding times for helical membrane proteins belie the architecture of relevant free energy landscapes. The folding of GlpG within intact lipid bilayers is mostly much slower compared with soluble protein folding (Table 1 and [76]). As mentioned earlier, this sluggish folding process is thought to mainly result from the high membrane viscosity retarding the helix packing interactions, rather than the presence of high energy barriers and/or ruggedness over the barriers. It still remains uncertain, however, whether the refined folding energy landscape is maintained, even during the co-translational folding involving the ribosome–translocon complex.

• Membrane protein folding is a fundamental biophysical problem that needs to be addressed to enhance the treatment of conformational disorders arising from misfolding. While unraveling the kinetic details is crucial for a comprehensive mechanistic understanding, only a limited body of research is available for helical membrane proteins, mainly due to technical limitations across various facets.

• The folding speeds of helical membrane proteins are primarily limited by the earlier steps of translation and insertion but can be substantially delayed by other factors, such as strong packing cooperativity and post-insertion topology changes. The viscosity of lipid membranes emerges as a notable factor that decelerates helix packing interactions, significantly altering the current picture of folding energy landscapes.

• A promising direction for future research involves mechanistic studies aimed at understanding how the precise timing of insertion and folding steps coordinates the entire co-translational folding process while mitigating the risks of misfolding and aggregation. An essential question in this context pertains to when and where insertases and chaperones participate in the folding process and how they contribute to reorganizing the sequence and timing of intermediate steps.

The author declares that there are no competing interests associated with this manuscript.

This work was supported and funded by the National Research Foundation of Korea [2020R1C1C1003937].

The author thanks Seoyoon Kim in his laboratory for help with preparing a figure.

AFM

atomic force microscopy

bR

bacteriorhodopsin

CHAPS

3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate

CHAPSO

3-[(3-cholamidopropyl)dimethylammonio]-2-hydroxy-1-propanesulfonate

DDM

n-dodecyl β-d-maltoside

DM

n-decyl β-d-maltoside

DMPC

1,2-dimyristoyl-sn-glycero-3-phosphocholine

DMPG

1,2-dimyristoyl-sn-glycero-3-phosphorylglycerol

DsbB

disulfide bond formation protein B

EMC

endoplasmic reticulum membrane protein complex

ER

endoplasmic reticulum

FRET

fluorescence resonance energy transfer

GLUT3

glucose transporter 3

LacY

lactose permease

MelB

melibiose permease

MT

magnetic tweezers

PE

phosphatidylethanolamine

PMP22

peripheral myelin protein 22

SDS

sodium dodecyl sulfate

SEIRAS

surface-enhanced infrared absorption spectroscopy

SMFS

single-molecule force spectroscopy

TM

transmembrane

Trp

tryptophan