Folding and misfolding of potassium channel monomers during assembly and tetramerization

Significance

Ion channels exist in all organisms. Here, we connect the folding of potassium channel monomers to the kinetics of tetramerization. Rather than adopting a native-like conformation once inserted into the bilayer, monomers initially exist as a structurally heterogeneous ensemble in a protein-dense region. This early clustering of monomers may be a general phenomenon that assists in the assembly of multimeric membrane proteins by pre-localizing the subunits. Folding can occur along fast or slow (misfolded) pathways that can be modulated with mutations that trap monomers in a native-like state. In spite of its name, the C-terminal “tetramerization” domain in KcsA does not enhance tetramerization, suggesting it may play another role in channel function.

Abstract

The dynamics and folding of potassium channel pore domain monomers are connected to the kinetics of tetramer assembly. In all-atom molecular dynamics simulations of Kv1.2 and KcsA channels, monomers adopt multiple nonnative conformations while the three helices remain folded. Consistent with this picture, NMR studies also find the monomers to be dynamic and structurally heterogeneous. However, a KcsA construct with a disulfide bridge engineered between the two transmembrane helices has an NMR spectrum with well-dispersed peaks, suggesting that the monomer can be locked into a native-like conformation that is similar to that observed in the folded tetramer. During tetramerization, fluoresence resonance energy transfer (FRET) data indicate that monomers rapidly oligomerize upon insertion into liposomes, likely forming a protein-dense region. Folding within this region occurs along separate fast and slow routes, with τ_fold ∼40 and 1,500 s, respectively. In contrast, constructs bearing the disulfide bond mainly fold via the faster pathway, suggesting that maintaining the transmembrane helices in their native orientation reduces misfolding. Interestingly, folding is concentration independent despite the tetrameric nature of the channel, indicating that the rate-limiting step is unimolecular and occurs after monomer association in the protein-dense region. We propose that the rapid formation of protein-dense regions may help with the assembly of multimeric membrane proteins by bringing together the nascent components prior to assembly. Finally, despite its name, the addition of KcsA’s C-terminal “tetramerization” domain does not hasten the kinetics of tetramerization.

Many homo- and hetero-tetrameric ion channel monomers contain two transmembrane (TM) helices connected by a re-entrant pore loop segment (p) and adopt a modular structure with a “TM-p-TM” topology (Fig. 1) (1). The ion permeation pathway is created by the juxtaposition of the p-loop of each monomer along the tetramer’s symmetry axis. The highly conserved p-loop contains a short 3.5-turn helix and the ion selectivity filter (1, 2). Mutations in this region are detrimental to potassium channel trafficking (3) and are associated with diseases such as weaver (4–6) and long-QT syndrome (3, 7, 8) (the QT interval is the time from the start of the Q wave to the end of the T wave).

Fig. 1.

KcsA TM Domain [Protein Data Bank (PDB) ID 3EFF]. (A and B) Top, side views. (C) Monomer: TM helices TM1, TM2, pore helix, turret, and selectivity filter. (D) Surface residues (polar, basic, and acidic residues are colored green, red, and blue, respectively).

The folding of helical membrane proteins is thought to proceed in a two-stage manner, with insertion of stable TM α-helices followed by lateral packing within the bilayer (9–12). A more intricate three-stage model has also been proposed, where insertion and lateral packing of α-helices is followed by folding of loops, ligand binding, and insertion of peripheral domains (10). The folding of multimeric membrane proteins also requires the assembly of independent subunits, “building blocks,” that may need to be at least partly folded before oligomerization. This raises the question: How folded are the ion channel’s subunits prior to tetramerization? In vivo, monomers are independently synthesized and inserted into the lipid bilayer from the ribosome–translocon complex. Then, for a K⁺ channel, four monomeric subunits associate to form a functional tetrameric channel (13). The tetramerization process may be assisted by cytosolic tetramerization domains that could help the monomers fold or assist in their association (13, 14). However, prior studies have also shown that the pore domain can form tetramers on its own (13, 15–18) (Fig. 1 A and B).

Several biophysical properties of K⁺ channels are relevant to their dynamics. The ion-conducting pore of K⁺ channels, formed by the re-entrant segment located between the two hydrophobic TM helices (termed TM1 and TM2 in KcsA and S5 and S6 in Kv1.3), is lined with polar and charged residues (Fig. 1D) (19). If the helices retain the same structure and orientation as they do in the tetramer, the polar residues along with the non–hydrogen-bonded backbone atoms would be exposed to lipids in the channel’s monomeric form. As these interactions are energetically unfavorable, the dominant conformation of individual monomers prior to tetramerization is unclear. Previous thiol-labeling experiments have shown that for Kv1.3, the monomers maintain the helical structure of the re-entrant pore helix in a native-like orientation at the water–lipid interface (20, 21). Likewise, previous molecular dynamics (MD) simulations indicated that the monomers’ native conformation (i.e., the conformation observed in the assembled tetramer) was surprisingly stable on a timescale of ∼1 μs (21).

In this paper, the dynamics of KcsA and Kv1.2 pore domain monomers are first investigated using MD and NMR. Then, we connect their dynamics to the kinetics of tetramerization using a gel-based refolding assay and fluoresence resonance energy transfer (FRET) on KcsA. The monomers are seen to associate in the bilayer prior to tetramerization, suggesting that the liposome provides a poor solvent environment for KcsA’s TM helices. A comparison between wild-type (WT) KcsA and an enhanced folding construct containing a disulfide bond linking TM1 and TM2 (A29C-A109C) indicates that the arrangement of the TM helices can influence the kinetics of folding and misfolding. Finally, the lack of concentration dependence in folding kinetics points to a late rate-limiting step occurring after monomer association.

Results

Monomer Dynamics.

In a previous 650-ns MD simulation, the WT Kv1.2 pore domain monomer in a POPC lipid bilayer was stable in a native-like state with a Cα-RMSD below 3 Å (21). However, this simulation is relatively short compared with the micro- to the millisecond timescale of membrane protein dynamics (22). To further explore the monomer’s dynamics, we carried out 16.2 μs of simulations at T = 353 K using the Anton supercomputer (23). The relatively high temperature was chosen to accelerate sampling while still reproducing the thermodynamics of membrane protein folding (24).

During the first 3 μs of the simulation, the monomer’s structure was stable with a Cα-RMSD to the initial (native-like) state below 4 Å (Fig. 2A and SI Appendix, Fig. S1). Nevertheless, the pore helix became more parallel to the membrane during the first microsecond with the tilt angle increasing from 47° to 80° relative to the surface normal (SI Appendix, Fig. S1). The average tilt angle over the entire 16.2 μs was 80° ± 10°. At 4 μs, the two TM helices, S5 and S6, started to separate laterally with slightly different tilt angles due to their different lengths, leading to an overall Cα-RMSD above 4 Å. Between 6 and 8 μs, the three helices separated (SI Appendix, Fig. S1), with the pore helix remaining helical and staying nearly parallel to the bilayer. The parallel orientation allowed the polar and charged residues in the pore loop region to become more solvent exposed. This result is qualitatively consistent with the previous thiol-labeling results, which indicated that the pore helix remains helical and resides at the water–lipid interface (20). Based on short simulations (650-ns) and thiol-labeling data, the Kv1.3 monomer was inferred to retain native-like tertiary contacts (21). In our much longer simulations, however, the native-like arrangement of the three helices was only stable for the first 3 μs (RMSD < 4 Å).

Fig. 2.

MD simulation and Markov state modeling of a Kv1.2 pore domain monomer. (A) Selected snapshots from a 16.2-μs simulation run on Anton. The gray beads represent phosphate atoms in the POPC lipid headgroup. (B) MSM built on 394 μs of total simulation time at 303 K. The MSM built with a lag time of 20 ns is projected onto TIC1 and TIC2, which are folding coordinates determined through Time-structure Independent Component Analysis (TICA) (SI Appendix, SI Materials and Methods). The size of the circle is proportional to the population of each microstate, and the color of the circle represents the RMSD of each microstate to the native structure. (C) Salt bridge between D381 (Green) and K398 (Yellow) shown from the top with lipid phosphate atoms shown as gray spheres (Top) and the salt bridge shown from the top view on the protein alone (Bottom).

After 8 μs of simulations, residue D381 formed a salt bridge with K398 located at the top of the carboxyl–terminal S6 helix (Fig. 2C). This salt bridge persisted for the remaining 8 μs and stabilized the interactions between S6 and the pore helix. The amino terminal S5 helix eventually drifted toward the complex formed by the other two helices to produce an alternate nonnative arrangement of the three helices. Interestingly, the salt bridge formed between the pore helix and S5 mimics the salt bridge that is formed between the pore helix of one monomer and the S5 helix of an adjacent monomer in the native tetramer (which is impossible to form as there is only a single monomer in our simulations). Salt bridges are known to be overly stabilized in simulations (25), so the longevity of the D381–K398 bridge may be unrealistic. Nevertheless, the monomer spent a majority of time with the three helices in a nonnative arrangement.

Markov State Analysis.

To obtain a better estimate of the conformational ensemble of Kv1.2 monomers, an analysis based on a Markov state model (MSM) framework was performed using three sets of simulations (SI Appendix, Table S1): 1) 16.2-μs Anton simulation at T = 353 K; 2) 10 independent 9-μs long simulations starting from the native state at T = 303 K; 3) 100 rounds of adaptive sampling simulations at T = 303 K (SI Appendix, SI Materials and Methods). A total of 394 μs of simulations was accumulated and used for Time-structure Independent Component Analysis (TICA) and MSM analysis (26–30).

Based on the MSM analysis, the population of native-like structures (Cα-RMSD < 4 Å) converged at 18% (Fig. 2B and SI Appendix, Fig. S2, Bottom). This result further supports the direct visual observation from the simulations that the two TM helices and the pore helix retained the majority of their helical content with minimal fraying, but a native-like conformation similar to that in the folded tetrameric channel was not the dominant conformation. The pore helix remained parallel to the water–lipid interface, which is consistent with the thiol-labeling results (21). Overall, the monomer existed as a heterogeneous ensemble of contacting and noncontacting helices.

To compare with the Kv1.2 simulations and to enable a direct comparison with our experiments, simulations also were conducted on KcsA pore domain monomers without the C-terminal tetramerization domain [Protein Data Bank (PDB) ID: 1R3J, residues 22 through 124]. The pore domain of KcsA and Kv1.2, comprising the pore loop and the two TM helices, have 31% sequence identity. This suggests that the gross dynamical features of the two proteins should be qualitatively similar. Five 5-μs trajectories were initiated from the native state at T = 353 K. To compare with the Kv1.2 simulations, the KcsA trajectories were projected onto the same set of Time-structure Independent Components (TICs) obtained from the Kv1.2 simulations and also aggregated with the Kv1.2 simulations to create a new set of common microstates. Although the sampling was less extensive as compared to Kv1.2, the general behavior of KcsA was similar. The KcsA monomer adopted a variety of native-like (44%) and nonnative structures, albeit with the three helices separated less often than Kv1.2’s helices (SI Appendix, Figs. S5 and S6).

Locking the Two TM Helices into a Native-Like Conformation.

As the TM helices often were in a nonnative configuration in the simulations and, as discussed in the next paragraph, the quality of the NMR spectra were poor for the WT monomers lacking the tetramerization domain (“KcsA Δ125”) (SI Appendix, Fig. S7), we designed a disulfide bonded “CC” variant to stabilize the TM helices in a native-like conformation. Two Ala-to-Cys substitutions were introduced at a helix–helix contact near the intracellular end of the helices (A29C, A109C). The spontaneous formation of a disulfide bond was confirmed by mobility shift with SDS–polyacrylamide gel electrophoresis (SDS-PAGE) (SI Appendix, Fig. S7A). A trajectory and MSM analysis of simulations of the CC variant focusing on the Cα-RMSD and number of TM helix contacts indicated that the variant adopts a much more restricted conformational ensemble than the WT version (SI Appendix, Figs. S5 and S6).

WT KcsA Δ125 monomers were reconstituted into MSP1D1 Δh5 nanodiscs containing DMPG lipids as well as into bicelles made of DMPC and DHPC with q = 0.3. The WT monomers incorporated into nanodiscs and bicelles behaved poorly, forming soluble aggregates, and produced a weak and poorly dispersed ¹H-¹⁵N Heteronuclear Single Quantum Coherence (HSQC) NMR spectrum (SI Appendix, Fig. S7 B and C). In contrast, the CC version was stable in bicelles and exhibited a well-resolved HSQC spectrum indicative of a structurally homogenous ensemble (SI Appendix, Fig. S7D). NMR T1 relaxation measurements verified that the CC monomer was monomeric (SI Appendix, Fig. S8). Consistently, when DTT was added to CC mutant monomers in bicelles, the monomers formed soluble aggregates similar to those observed for the WT monomers. Although the exact aggregation mechanism is unknown, the TM helices may interact with their counterparts on other bicelles leading to soluble aggregates.

Kinetics of Folding.

The kinetics of tetramerization of KcsA Δ125 channels were examined by tracking the formation of native tetramers using an SDS-resistance assay conducted at a sufficiently high SDS level where only native tetramers remained folded and could be distinguished from monomers on a gel (Fig. 3B) (31). The refolding protocol started with tricholoroacetic acid–precipitated monomers solubilized with 17 mM (∼0.5% wt/vol) SDS at pH 6.5 (SI Appendix, Fig. S9). To initiate tetramerization, the monomers were diluted 10-fold into refolding buffer containing ∼14-mM asolectin liposomes. Control experiments employing dynamic light scattering verified that the liposomes remained intact after mixing with the SDS-solubilized KcsA Δ125 or the 17-mM SDS buffer (SI Appendix, Table S2). To measure refolding, aliquots of the protein–liposome mixture were removed over 25 min and quenched in 350 mM (∼10%) SDS buffer to arrest tetramer formation. At this high SDS level, liposomes were disrupted, and only native tetramers persisted, whereas weakly associated species were broken up and ran as monomers on SDS-PAGE gels (15, 16, 31, 32). The folding kinetics were quantified from the change in the gel band intensities. The similarity in the thermal denaturation profiles of KcsA refolded with this protocol and folded channels extracted from native membranes indicates that the folding assay produces a native or near-native species (33) (SI Appendix, Fig. S10).

Fig. 3.

The presence of a constraining disulfide bond results in faster folding and higher yields. (A) Location of A29C and A109C mutations are shown on top of a KcsA monomer structure taken from PDB ID 1R3J. (B) Examples of raw data for the refolding of WT Δ125 and CC Δ125 are shown, with monomer (M) and tetramer (T) bands indicated. (C) Fraction tetramer as a function of time for WT Δ125 and CC Δ125 constructs, which were both carried out with a protein concentration of 10 μM in 14-mM asolectin vesicles, pH 6.5. Solid lines represent double exponential global fits to both WT Δ125 and CC Δ125 kinetic traces, assuming common fast and slow rates. (D) Concentration dependence of folding of WT and CC constructs either with (FL, full length) or without (Δ125) the carboxyl–terminal tetramerization domain. The concentration dependence between 1 and 10 μM of the folding kinetics is shown for WT Δ125, CC Δ125, WT FL, and CC FL. At 1- to 10-μM monomer concentration, there are between 30 and 300 monomers per liposome, respectively.

For KcsA Δ125 monomers at 10-μM monomer concentration, two nearly equal refolding populations were observed, one that folded in under a minute and another that folded on the 10-min time scale (Fig. 3 B and C). Since the buildup of native tetramers is directly measured, the fast and slow appearance of populations of native tetramers implies that there are multiple routes to the native state rather than each phase representing a step on a sequential pathway. The latter possibility would have resulted in a 10-min lag phase in the buildup of native tetramers, which was not observed.

However, for the disulfide bonded CC construct, essentially all the monomers tetramerized at the fast rate (Fig. 3 B and C). This difference in folding behavior between WT and the CC construct implies that the presence of unconstrained TM helices in the WT monomer resulted in the formation of a slow, misfolded species. The kinetic data for WT and CC were fit globally, assuming the rates of the two phases were the same for the two versions. The resulting time constants were τ_fast = 40 ± 2 s and τ_slow = 1,500 ± 100 s, with 28 ± 6% and 74 ± 6% fast-folding population fractions for the WT and CC constructs, respectively. WT and CC data were also fit separately for comparison, and the resulting time constants were τ_fast = 88 ± 10 s and τ_slow = 2,000 ± 100 s and τ_fast = 32 ± 5 s and τ_slow = 811 ± 80 s, respectively (SI Appendix, Tables S3 and S4)

Next, the concentration dependence of the folding rates was studied. Interestingly, the tetramerization process, both in rate and branching ratio, are concentration independent for KcsA Δ125 WT and CC from 1 to 10 μM within the accuracy of our measurements (Fig. 3D and SI Appendix, Table S3). This striking result indicates that the rate-limiting step is unimolecular, despite the native state being a tetramer. If tetramerization were limited by the association of four monomers or two unstable dimers, one would have observed a 1,000-fold decrease in folding rate for a 10-fold decrease in concentration. As no measurable concentration dependence was found, we propose that the oligomerization process occurs early and is fast on both routes, with the rate-limiting step representing a productive folding (fast pathway) or error-correction step (slow pathway). This proposal and the nature of the fast oligomerization process are investigated further with FRET, as discussed below.

Influence of the Tetramerization Domain.

Several types of potassium channels have N- or C-terminal tetramerization domains that are believed to promote the association of their TM segments. For Shaker Kv channels, the T1 tetramerization domain may facilitate the proper association of TM segments (13). For KcsA, there is a moderate-size C-terminal segment, containing eleven positive and seven negative residues, that is similarly implicated in channel folding (13, 14, 34). Based on these studies, the 35-residue, helical C-terminal tetramerization domain was added to test whether it could enhance the folding process of the WT or CC variants (“WT FL” and “CC FL”).

For both the WT FL and CC FL variants, the addition of the C-terminal tetramerization domain resulted in less efficient and concentration-dependent folding behavior (Fig. 3D and SI Appendix, Table S3). At 10-μΜ protein concentration, WT FL constructs had folding kinetics similar to those of the WT Δ125 constructs. However, as the protein concentration was decreased to 1 μM, the fast-folding population disappeared, and only ∼15% of the molecules formed tetramers within 25 min. For the CC mutant, only 20 to 55% folded rapidly, depending upon protein concentration. In addition, only 20 to 70% were folded after 25 min. Thus, the C-terminal segment did not enhance the folding of the TM domains under our folding conditions.

Kinetic Studies of Tetramerization Using FRET.

As described above, the folding kinetics of both the WT and CC constructs of KcsA Δ125 had minimal concentration dependence, implying that the rate-limiting step was a unimolecular process. This step seems likely to occur after oligomerization in liposomes, although in principle, the insertion of monomers into liposomes could have been rate limiting. To test whether oligomerization was fast and occurred before the rate-limiting step, ensemble FRET measurements were carried out under the same folding conditions as used in the gel-based folding assay.

Dyes were attached using thiol-labeling at a single position at the amino terminus of TM2 using a KcsA Δ125 L86C variant (SI Appendix, Fig. S11). Equal mixtures of donor- (Cy3) and acceptor- (Cy5) labeled KcsA Δ125 in SDS micelles were mixed and diluted 10-fold into the liposome mixture to initiate folding exactly as in our gel-based folding assays, and the transfer of energy between fluorophores was monitored by fluorescence emission (Fig. 4). In the initial 17-mM SDS condition, the emission spectrum of KcsA was dominated by that of the donor, implying that SDS-solubilized KcsA was monomeric prior to folding in liposomes. Upon dilution into liposomes, the donor emission maximum at 570 nm was quenched by 71% within the 10-s manual mixing dead-time while the acceptor emission maximum at 680 nm increased eightfold. During the next 25 min, over which tetramer formation occurred, only a 4% increase in intensity was observed across the entire emission spectrum. The large initial FRET response suggests that prior to tetramerization, the monomers rapidly associate into a protein-dense region within the liposomes. A variety of measurements were performed to test this interpretation.

Fig. 4.

FRET measurements of KcsA tetramerization. (A) Ensemble FRET measurement of KcsA Δ125 in SDS and in liposomes as a function of time. (B) Double-jump FRET measurements of KcsA refolding quenched with 10% SDS overlaid with “in SDS” and “10s in liposome" time points from A. Spectra are normalized to have the same value at 641 nm, an empirical iso-emissive point. Measurements are conducted at a monomer concentration of 10 μM.

To examine the degree of monomer and tetramer association in the liposomes at various time points during refolding, we transferred aliquots of the folding reaction of FRET-labeled KcsA into 350 mM SDS where only native tetramers persist, done in a double-jump folding assay. For both the WT and CC variants, acceptor emission levels increased over time in SDS but at substantially reduced values compared to those in the intact liposomes (Fig. 4B). Even by 1,440 s, once most of the molecules had folded, the acceptor emission levels in SDS were only about 1/4 of the value observed when the protein remained in the liposomes. The large decrease in FRET upon SDS quenching suggests that the majority of the FRET signal in liposomes was due to Donor/Acceptor pairs on different but laterally associated folded tetramers, as previously observed (35, 36). Furthermore, the similarity between FRET levels after 10 and 2,000 s in liposomes indicates that upon insertion, the monomers rapidly clustered with similar donor/acceptor separation distances as those present in laterally associated tetramers, further supporting the presence of a protein-dense region.

We next investigated the possibility that the large initial increase in FRET levels was simply the result of the monomers being closer on average to other monomers in liposomes as compared to the SDS-solubilized state rather than the formation of an actual protein-dense region. To test this possibility, we examined the effects of adding 15 μM of unlabeled monomers to 2.5 μM of both donor- and acceptor-labeled monomers (SI Appendix, Fig. S12). If colocalization in the liposomes is the cause of the increased FRET signal, there would be no change in FRET, but if the monomers clustered together in a protein-dense region, the mean distance between dye-labeled monomers would increase due to the presence of intervening unlabeled monomers, and FRET would decrease. Consistent with the formation of a protein-dense region, we observed upon transfer to liposomes that the FRET level was lower in the presence of the unlabeled molecules (1/3 versus 1/2).

To examine the possibility that FRET occurred within protein aggregates forming outside of liposomes, SDS-solubilized KcsA monomers were diluted 10-fold into water in the absence of liposomes. In this control, the overall donor fluorescence decreased 4%, while the acceptor fluorescence exhibited only a minor increase (SI Appendix, Fig. S13), especially as compared to folding in liposomes. The decrease in acceptor emission indicated that some fraction of monomers was no longer in solution. However, and more importantly, the lack of significant FRET indicates that the large observed FRET changes in the presence of liposomes came only from membrane-solubilized proteins.

We observed some differences in refolding kinetics with the dye-labeled proteins in the gel-based assay, although none were apparent with the CC version at 10 μM (SI Appendix, Fig. S14). As discussed above, unlabeled monomers were present in the dense region. Therefore, the major finding based on our FRET measurements—the early formation of a dense region—is unlikely to be the result of dye–dye interactions.

Upon transfer from SDS to liposomes, the insertion of the TM helices into the bilayer is likely to be quick relative to observed folding times. Valiyaveetil and coworkers have shown with cysteine labeling that KvAP’s six TM helices fully insert into the bilayer within their 30-s measurement dead-time (37), while simulations find that insertion of TM helices occurs in microseconds (24). In addition, we find that the FRET level directly after transfer to the liposome is similar to that of laterally associated tetramers (in which TM1 and TM2 are inserted into the bilayer). Based on these results, we believe that the TM helices rapidly insert into the bilayer rather than lie on its surface.

Discussion

Our investigation of the folding of potassium channel monomers and their role in the assembly of ion channels revealed a number of salient features (Fig. 5). The MD simulations and MSM analysis indicate that for both Kv1.2 and KcsA pore domains, isolated monomers form a heterogeneous ensemble of native and nonnative states with all three helices folded and lying within the membrane. While the population of native-like states of the monomer is nonnegligible (18% for Kv1.2 and 44% for KcsA), it is nevertheless striking that a considerable number of conformations exist where the two TM helices are separated despite the substantial conformational restriction imposed by the environment; namely, the secondary structure of all three helices is retained, and the two TM helices remain inserted within the planes of the bilayer. Based on prior studies of hydrophobic matching (38–41), the different lengths of TM helices presumably contribute to their tendency to separate while remaining within the bilayer.

Fig. 5.

Proposed model for KcsA folding and tetramerization. SDS-solubilized monomers enter the liposomes and undergo rapid association into a protein-dense region within the membrane prior to tetramerization. Oligomers can form with a native or nonnative TM helical arrangement, which fold on the 1- or 20-min time scale, respectively. The rate-limiting step on the faster pathway may involve the insertion of the selectivity filter and pore helix to stabilize the tetramer in its native conformation.

This overall picture is consistent with our NMR data. Our attempts to generate well-resolved NMR spectra for the WT KcsA monomer inserted into nanodiscs and bicelles were unsuccessful. Suspecting that the separation of the TM helices was the critical feature, a double-cysteine variant was engineered to have a disulfide bond at the cytosolic ends of the two TM helices locking them into a native-like arrangement. This CC variant yielded a well-dispersed ¹H-¹⁵N HSQC spectrum, supporting the view that the WT’s TM helices were often separated in the monomers and formed a heterogeneous ensemble. A combined thiol-labeling/MD study on Kv1.3 (20, 21) demonstrated that the pore helix is folded and near the bilayer surface, although the MD simulations, which were conducted for less time than ours (0.6 versus 400 μs), indicated that the TM helices remain in contact, whereas we find that they often are separated.

FRET-monitored refolding measurements of monomers passing from SDS into liposomes indicated that both WT Δ125 and CC Δ125 variants lacking the tetramerization domain associated well before the appearance of native tetramers. The SDS-resistance assays and FRET measurements support a model in which the WT KcsA monomers assembled into native tetramers via two kinetic paths with time constants of 40 ± 2 and 1,500 ± 100 s. The CC variant largely if not fully lacked the slow phase. We posit that slow folding is the result of nonnative packing arrangements of the TM helices that must ultimately be corrected for tetrameric assembly to proceed. The observation of slow- and fast-folding routes has been seen in many soluble proteins where the initial collapse step leads to species with native-like topology on a direct pathway or to a species containing some misfolded structure that is slow to correct (42–44).

Despite the channel being a tetramer, folding of the WT Δ125 and CC Δ125 constructs was concentration independent from 1 to 10 μM for both the fast- and slow-folding pathways, both in rate and fraction. This observation implies that the rate-limiting step on both pathways is unimolecular. Because the FRET measurements indicated that monomer insertion and oligomerization are relatively quick processes, the rate-limiting step on the fast pathway likely occurs going from an oligomer to a native tetramer. This complex transition requires the formation of a slightly twisted arrangement of the eight TM helices and the energetically costly opening of an aqueous channel. The final stages in the formation of native tetramers may occur in a concerted step with the association of correctly folded monomers already having the pore helices and selectivity filters positioned in a native or near-native orientation. Alternatively, this event may occur in two distinct steps, with the pore helices and selectivity filter segments adopting their native positions only after the eight TM helices have adopted a native arrangement. The energetics of dissociation from the protein-dense region prior to tetramerization also may contribute to the overall rate. Further studies will be needed to resolve the nature of the rate-limiting step.

Tetramerization Domain.

In eukaryotic channels such as the Shaker voltage-activated potassium channel, the presence of the T1 tetramerization domain improved the rate of folding and assembly (13). Whereas the carboxyl–terminal domain of KcsA forms a four-helix bundle that contributes to the stability of the folded tetrameric channel (45), its role in the assembly process is unclear. In prior studies, the KcsA tetramerization domain aided tetramerization in a pH- and concentration-dependent manner (46, 47) as well as increased the stability of tetramers (34).

To our surprise, we found that the presence of the C-terminal segment domain, which forms the tetramerization domain, did not improve the folding kinetics of KcsA. For both the WT and CC variants, tetramerization became concentration dependent in a complex manner with both a reduced fast-folding population and lower overall yield of tetramers. Each of the four C-terminal segments is highly charged with 11 positively and 7 negatively charged residues. These charged moieties may hinder the domain from crossing from the bilayer into the interior of the liposome (48) and produce complicated kinetic behavior. Further studies are required to better understand this behavior.

Folding Behavior, Protein-Dense Region, and Solvent Quality.

The folding of potassium channels displays similar behavior to other α-helical membrane proteins, having a transition state close to the native state (9–11). For example, in the force-induced unfolding studies of GlpG where the unfolding forces are parallel to the bicelle surface, the transition state was closer to the native state than that observed in SDS-driven folding studies in solution (49, 50). In other SDS-based refolding studies on bacteriorhodopsin, DsbB, and GlpG, the transition states were expanded (51–53). The observed difference between the force- and the SDS-driven unfolding studies may be due to the difference in the mode of denaturation as well as folding conditions (micelle or bicelle versus liposome). Regardless, KcsA in liposomes appears to have a transition state near the native state.

The two-stage folding model, proposed nearly 30 y ago, has provided a useful framework for discussing membrane protein folding. The model, which posits insertion of all TM helices followed by lateral packing, was proposed based on studies observing that bacteriorhodopsin cut into a few pieces could still be reassembled (54–56). To account for more diverse folding behaviors, a more sophisticated three-stage model (10) was proposed where insertion and lateral packing of the TM helices could be followed by ligand binding, loop folding, or peripheral domain insertion. This extra step is relevant to K⁺ channels, in which the insertion of the four pore helices and the selectivity filter may be required to finalize the folding process.

For KcsA, the first two folding stages likely correspond to insertion and formation of a protein-dense region in the membrane before folding into native tetrameric channels. This observation motivates a question in regard to membrane protein folding in general, namely, should the membrane be considered a good or poor solvent of TM helices? In other words, are helix–lipid interactions stronger or weaker, respectively, than helix–helix interactions? With detailed packing, even hydrophobic helices can associate into stable TM proteins (57), implying that a bilayer that behaves as a good solvent can still support folding. Using FRET, we observed a protein-dense region having a heterogeneous packing of monomers; however, the two TM helices can dissociate from each other according to NMR (in bicelles) and MD simulations (POPC bilayers). Hence, for potassium channel monomers, we observe aspects of both poor and good solvent behavior.

Potentially, in the folding of other membrane proteins, nonspecific or quinary interactions between TM helices also may give rise to a protein-dense region. These interactions can also occur in native structures, as seen with KcsA tetramers undergoing lateral association in our and earlier studies (35, 36). Two studies found that phase separation can occur near or on membrane surfaces, which increases signaling fidelity (58, 59). Several studies have shown that membrane proteins can alter lipid packing in the fluid liquid crystalline phase, which can in turn cause proteins to associate nonspecifically in order to minimize the perturbation on the membrane (60, 61). Generally, lipids may act as a marginally poor solvent for TM helices if the helices contain polar amino acids that are less soluble in the hydrophobic bilayer (62–68) or have a hydrophobic mismatch (69). These mixed results, along with variability of the hydrophobicity in the TM helices and the physical properties of the bilayer (e.g., composition, thickness, curvature, and lateral pressure), suggest that solvent quality is likely to be system dependent.

These issues have implications to in vivo folding and assembly where helices partition between the chamber of the translocon, the lipid–water interface, and the membrane (70). Potentially, the assembly of multimeric membrane proteins may be aided by the formation of a protein-dense region that increases the local concentration of the nascent components. This colocalization reduces the kinetic challenge of simultaneously bringing all the components together as would normally be required in a fourth-order reaction when the monomers can otherwise freely diffuse within the bilayer.

Materials and Methods

MD Simulation Setup.

For Kv1.2 simulations, the monomer structure was taken from the crystal structure with PDB ID 2A79 (residues 323 through 421) (19), and for KcsA, the monomer structure was taken from crystal structure, 1R3J. All systems were prepared by using CHARMM-GUI’s Membrane Builder module (https://www.charmm-gui.org/) (71–75). Further details are listed in SI Appendix, SI Materials and Methods.

Protein Purification.

All KcsA constructs were expressed in XL10-GOLD cells and purified as described in SI Appendix, SI Materials and Methods.

Data Availability

Data are available upon request.