Efficiently determining membrane-bound conformations of peripheral membrane proteins using replica exchange with hybrid tempering: Orientation of PMP on lipid bilayer using replica exchange

Abstract

Accurately sampling the membrane-bound conformations of peripheral membrane proteins (PMP) using classical all-atom molecular dynamics simulations (AAMD) is a formidable enterprise due to the wide rugged free energy landscape of the protein-membrane system. In general, AAMD-based extraction of binding geometry requires simulations of multiple systems with different initial user-defined binding poses that may not be exhaustive. As an alternative, advanced sampling methods are also applied to elucidate the membrane-binding mechanism of PMPs. But these techniques are generally computationally expensive and often depend on the choice of the collective variables (CV). In this work, we showcase the utility of CV-free replica exchange with the hybrid tempering (REHT) method in capturing the membrane-bound conformations of PMPs by testing it on the Osh4 amphipathic lipid-packing sensor (ALPS) motif, a 27 amino-acid membrane-binding peptide. We show that REHT samples all the membrane-bound conformations of the Osh4 ALPS peptide observed in AAMD simulations and does it in a highly efficient manner. We clearly show that, out of the two significant conformations, the peptide prefers horizontal conformations over vertical ones. In both the conformations, REHT captures all the vital residue-wise membrane contacts. The transition between the two configurations is not uncommon as our calculations reveal a ~1 kT free energy difference between the two conformations. Interestingly, from our simulations, we also find that the transition from vertical to horizontal conformation involves limited unfolding of the main helix’s last turn. From our findings, we conclude that REHT samples the membrane-bound conformations of Osh4 ALPS peptide very efficiently and also provides additional insights and information that are often not available with regular piece-wise AAMD simulations. The method can be used as an efficient tool to explore the membrane-binding mechanisms of PMPs.

1. Introduction

Peripheral membrane proteins (PMP) are a class of membrane proteins that localize at the surface of the membrane. The function of PMP depends both on its membrane affinity and membrane-bound conformations. Also, the proteins carry different membrane-interacting motifs and domains that allow them to interact with specific membranes to perform their function [1–3]. Therefore, accurate characterization of PMP membrane-binding mechanisms is crucial to understanding their function. Experimental methods such as cryo-EM, solid-state NMR, and EPR are applied, when necessary, to determine the membrane-bound structure of the PMP, but often due to the membrane and other experimental constraints, the membrane-bound conformation and dynamics information are low-resolution in nature [4–6]. Along with experimental methods, computational algorithms such as all-atom molecular dynamics (AAMD) and related simulations have been effectively used to supplement the experimental information and elucidate the residue-level membrane association information of PMP [7–12].

However, expounding the equilibrium properties of the membrane-binding characteristics of a PMP using simulations is non-trivial [13–15]. Generally, a protein–membrane system exhibits a wide conformational free energy landscape with multiple minima [16]. Hence, classical (brute-force) AAMD simulations, which can get trapped in a local minimum, may not be adequate to sample the entire conformational landscape of the membrane-binding process of a PMP [8]. Such a shortcoming could be overcome using an enhanced sampling technique like replica exchange simulations (REX) [17, 18]. A few studies have shown that replica exchange with solute tempering (REST) explores the protein–membrane interactions faster and better than classical MD or traditional replica exchange molecular dynamics (REMD) [19–22]. Recently, we developed a variant of REX called replica exchange with hybrid tempering (REHT) for the exhaustive sampling of intrinsically disordered proteins (IDPs). In this method, the differential and optimal heating of both the protein (solute) and water (solvent) helps the protein cross significant free energy and entropic barriers. Hence, the IDPs could escape metastable states quickly and explore all conformational basins [23]. On a similar note, we believe that the differential and optimal heating of membrane and membrane-bound PMP could help the latter explore the wide conformational landscape of membrane-bound PMP. Therefore, we might be able to sample the membrane-bound conformations of PMP better by carrying out REHT simulations of protein–membrane–water systems, where we heat the solvent and membrane minimally and heat the protein to a greater extent. In theory, the minimal heating of the membrane should maintain its structural integrity and expedite the lateral dynamics of lipids which would allow the membrane-bound protein to explore different states.

Recently, Klimov and co-workers applied REHT on a short helical peptide (PGLa) and showed efficient sampling of the peptide binding to anionic bilayer as compared to the REST methods [18]. Our work, which is similar to the one by Klimov and co-workers, aims to assess the efficiency and robustness of REHT in exploring protein–membrane binding poses as compared against the well-established reference data arrived at by exhaustive AAMD simulations. We also wanted to see if the REHT, which was designed to explore protein conformations, allows one to observe changes in protein secondary and tertiary structure as a function of distance from the membrane—a feature often exhibited by larger PMPs with some tertiary structure [24]. To achieve it, we have chosen the amphipathic lipid-packing sensor (ALPS) motif of Osh4, which is an oxysterol-binding protein in yeast [25–27]. The ALPS motif of Osh4 is made up of 27 residues in the N-terminal end of the protein, one of the latter’s six membrane-binding regions [26]. The binding mechanisms of Osh4 ALPS peptide to various membrane compositions were investigated using classical MD simulations by Monje-Galvan and Klauda [28]. To compare the utility of REHT, we sought to replicate the bound conformations of the peptide with the simplest model membrane used in their study.

The rest of the paper is organized as follows. In the next section, following this introduction, we discuss the methods and system setup. We then discuss various results showing the performance and accuracy of the method. We finish the paper with a short conclusion.

2. Material and methods

2.1. System setup and simulation

The membrane composition used consisted of a mixture of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS) in a 3:2 ratio in bilayers of 80 lipids per leaflet. This mimics the composition of anionic lipids in yeast membranes. The protein–membrane system was created using the Membrane builder module in CHARMM-GUI [29–31]. Osh4 ALPS motif (27 amino acids) was extracted from the PDB ID: 1zhz [32] and was horizontally placed at least 6 Å above the model membrane. Even though Monje-Galvan and Klauda used both vertical and horizontal starting orientations, we have decided to use one of them as the protein would explore different orientations across the replicas in REHT. Water patches of 25 Å consisting of the TIP3P model [33] were placed above and below this protein-membrane setup to hydrate the system. The system was neutralized with 64 potassium ions. Energy minimization and equilibration were carried out using GROMACS 2020.4 [34]. The system was well equilibrated using the CHARMM-GUI six-step protocol [31] with two NVT steps and four NPT steps for a total time of 15 ns. The protein was restrained throughout this period, and the restraint was slowly removed during this six-step process. The simulation was carried out using the CHARMM36m force field [35]. The structure at the end of the equilibration, given in Fig. 1A, was used as the starting structure for the REHT simulation.

Fig. 1.

A Starting setup for the REHT simulations. The main helix of Osh4 ALPS peptide is marked in purple color. B Horizontal and C vertical bound conformations seen in the base replica trajectory. In the peptide, α − helix, 3₁₀ helix, turns, and coils are marked in purple, blue, cyan and white colors, respectively

2.2. REHT simulation

REHT simulations were carried out with GROMACS 2021.5 patched with PLUMED 2.7.5 [34, 36, 37]. A total of 15 replicas were used. The temperature was scaled from 303.15 to 333.15 K. This scaling affects all the atoms in the system; hence, the membrane and solvent experience temperatures between 303.15 and 333.15 K across the 15 replicas. The Hamiltonian of the protein was scaled from 303.15 to 423.15 K. Therefore, the protein experiences both the Hamiltonian and temperature scaling. The simulations used the CHARMM36m force field parameters and the TIP3P water model [33, 35]. The production run was performed in NPT ensemble using a timestep of 2 fs. The hydrogen bonds were constrained using the LINCS algorithm [38]. Non-bonded forces were calculated with a 12 Å cutoff (10 Å: switching distance). Long-range electrostatic forces were calculated using the particle mesh Ewald method [39]. The temperature of the system in each replica was maintained at their respective temperature and pressure of 1 atm using a Nose–Hoover thermostat [40] and Parrinello-Rahman [41] (with semi-isotropic coupling) barostat with time constants 1.0 and 5.0 ps⁻¹, respectively. Before the exchange in REHT started, all replicas were equilibrated for 1 ns at their respective temperatures, where no exchanges were done. Following this, the production run with replica exchanges was started, and exchanges were attempted every 1 ps. Each replica was simulated for 500 ns, leading to a total simulation time of 7.5 μs. We provide a user manual as supporting information (SI) for REHT for PMP studies.

2.3. MD simulations with most populated binding conformations

As discussed later, our REHT simulation identified three major membrane-bound conformations for the Osh4 ALPS peptide. To assess the binding stability of these major conformations, five system states were taken from each of the three tilt-angle distribution peaks (see Fig. 3), and classical MD simulations were carried out with them as starting structures. Each state was simulated for 100 ns in NPT ensemble using a timestep of 2 fs. The temperature of the system was maintained at 303.15 K using a Nose–Hoover thermostat [40]. The pressure was maintained at 1 atm using Parrinello-Rahman [41] barostat by semi-isotropic coupling with time constants 1.0 and 5.0 ps⁻¹. The hydrogen bonds were constrained using the LINCS algorithm [38]. Non-bonded forces were calculated with a 12 Å cutoff (10 Å: switching distance). Long-range electrostatic forces were calculated using the particle mesh Ewald method [39]. The distance of the key membrane-binding residue, SER8, was tracked in time to assess the stability of the membrane-bound conformation in each system.

Fig. 3. Histogram (black bins) and probability density plot (red line) of tilt angle of membrane-bound helical Osh4 ALPS peptide.

The three peaks of the tilt-angle distribution are fit to three Gaussians using the Gaussian Mixture Model, and they are marked in orange, blue, and green solid lines. The scales for frequency of histogram and density of density plot are marked on the left and right sides of y-axis, respectively. Snapshots of membrane-bound peptide conformations with tilt angles of 12, 37, and 64 degrees are shown above the plot from left to right in order. The main helix of Osh4 ALPS peptide is marked in purple color and SER8 is marked in green color

3. Results and discussion

As mentioned earlier, our main aim of the study was to investigate the utility of REHT in elucidating the membrane-bound conformations of a membrane-binding peptide, Osh4 ALPS peptide. Our findings show that REHT predicts all the membrane-bound conformations reported by Monje-Galvan and Klauda [28] and captures all the vital residue-wise contacts in those populations. Furthermore, REHT was able to sample these interactions at a much lower simulation time than the classical MD simulations.

3.1. REHT performance

We performed the REHT simulation with 15 replicas, and the exchange probabilities varied between 11–13%. Although this is on the lower side of the accepted range, other analyses given below show enough exploration of all Hamiltonian temperatures (referred to as just ’temperatures’ hereafter) by each replica. The REHT technical performance was analyzed by monitoring the random walk of replicas across Hamiltonian temperatures. The standard demuxing script provided by PLUMED was used to obtain the random walk across temperatures [37]. Figure 2A shows the random walk of the first and last replicas, and they explore all the temperatures and are not trapped in any temperature for an extended time. Furthermore, Fig. 2B shows that all the replicas explore all the temperatures. Although a couple of replicas get trapped in a window of 2–3 temperatures for a few nanoseconds, they come out and explore all the other temperatures during the simulation. To provide a quantitative assessment of the random walk of replicas, the replica mixing parameter was used [42, 43]. It is defined as,

$$m(T)=1-\frac{\sqrt{{{\displaystyle \Sigma }}_{r=0}^{R-1}{t}_r^2}}{{{\displaystyle \Sigma }}_{r=0}^{R-1}{t}_r},$$ (1)

Fig. 2.

A Walk of first and last replicas over Hamiltonian scaling temperature in REHT. The first and last replicas are marked in blue and red, respectively. B Walk of all replicas over Hamiltonian scaling temperature in REHT. The colors are marked in the color bar on the right. C Mixing parameter m(T) averaged over all replicas as a function of Hamiltonian scaling temperature. The theoretical maximum value is marked in a black dotted line. Errors are given by red error bars. D Overlap of the potential energy of replicas. The potential energy distributions of each replica are plotted along the x-axis where the replica numbers are marked

where T is the Hamiltonian temperature, and t_r is the REHT simulation time spent by replica r at temperature T. If we assume an ideal random walk, m(T) reaches a theoretical maximum of $1-\frac{1}{\sqrt{R}}$, which is constant for all temperatures [43]. For our system with 15 replicas, the theoretical maximum is approximately 0.74. Figure 2C shows that the mixing parameter reaches the theoretical maximum for most temperatures except the extreme ones, which could be attributed to the boundary effects. To further establish the adequacy of exchange between temperatures, we plotted the potential energy of each replica, shown in Fig. 2D. We observe adequate overlap of the potential energy distribution of each replica with its adjacent replicas, which suggests a healthy exchange between them. Having established the good performance of REHT, we characterized the membrane-binding properties of the Osh4 ALPS peptide. All the following results were analyzed using the last 400 ns trajectory of the base replica, the replica with a system temperature of 303.15 K in the system temperature scaling.

3.2. Osh4 ALPS peptide explores both horizontal and vertical binding poses

First, the visual molecular dynamics (VMD) was used to visually examine the ensemble of structures explored in the base replica trajectory [44]. Note that it is not a continuous trajectory but an ensemble of states obtained from exchanges across replicas. Qualitatively, we observe that the Osh4 ALPS motif has many membrane-bound states throughout the 500 ns trajectory, including unfolded membrane-bound peptide conformations. However, a fully unfolded membrane-bound Osh4 ALPS peptide has not been reported [28]. At least, most of the main helix (residues 8–18) remain folded in both ALPS peptide-only or full-length Osh4 studies[27, 28]. Therefore, we only considered the membrane-bound conformations where the main helix was intact with at least two helical turns. To match this criterion and select only the frames with alpha-helix in the main-helix region, we used the Define Secondary Structure of Proteins (DSSP) algorithm [45].

After selecting the appropriate states, we quantified the binding modes using the main helix’s tilt angle, defined as the angle subtended by the line joining the C-alpha atoms of residues 8 and 18 with the membrane surface. Ideally, a tilt angle near 0° denotes horizontal binding mode, while an angle near 90° denotes vertical binding mode. We observe three major peaks in the tilt-angle distribution plot given in Fig. 3, and we classified and quantified the different peaks by fitting the data to Gaussians using the Gaussian Mixture Model (GMM). We decided to use three Gaussians to fit the tilt-angle distribution, as three was the minimum optimal number to show convergence of the Bayesian information criterion (BIC) to lower values. The mean values of the three peaks fit by GMM are 13.12 ± 6.58, 36.88 ± 8.54, and 64.47 ± 10.33 degrees, respectively. The first two peaks correspond well with the average tilt-angle values for the horizontal and vertical binding modes reported by Monje-Galvan and Klauda, which were 14.73 ± 2.19 and 39.62 ± 4.34 degrees for the horizontal and vertical biding modes, respectively[28]. Note that we have large errors compared to the classical MD findings from Monje-Galvan and Klauda. This could possibly be due to a wide conformational sampling of numerous local minima by a replica exchange method like REHT. Conversely, classical MD simulations could get trapped in local minimum [8, 17, 20]. However, we cannot rule out the possibility of using a small number of selected frames (~ 13750) for our analysis leading to large errors.

The third peak, which was previously not reported, shows the utility of REHT in the elucidation of membrane-bound conformations of peptides. However, we believe the third peak to be a transient peak between the more stable conformations of the second peak and the unbound states. There are three reasons for this argument. First, the difference in the conformational free energy between the second and third peaks is less than the free energy of thermal fluctuations (see Fig. 5B). Also, there is only a small and shallow minimum corresponding to the third peak in the free energy profile of the helical character and the center of mass distance of the peptide from the membrane, compared to deeper and larger minima for the first and second peaks. Second, the classical MD findings suggested that the vertical conformations, corresponding to our second peak, could be intermediate to the horizontal conformation [28]. Third, the more perpendicular conformations of the third peak would not be plausible physiologically in the case of the full-length protein [27]. Therefore, we decided to treat the second and third peaks together as vertical binding modes and the first peak as horizontal binding modes for all our comparisons with the findings from classical AAMD study [28].

Fig. 5.

A Interaction energy of the Osh4 ALPS peptide with the membrane (black line) as a function of tilt angle. Errors are given by black error bars. The density plot (in red) is given for comparison. B The free energy of membrane-bound helical Osh4 ALPS peptide as a function of tilt angle. The free energy is given in kT units. C Free energy map of helical character of Osh4 ALPS peptide as a function of the distance of peptide COM from the membrane center. D Helical character of the full peptide (Black line) and main helix (Green line) of the membrane-bound Osh4 ALPS peptide as a function of tilt angle. Errors are given by black and green error bars. The density plot (in Red) is given for comparison. E Frequency of contact of C-alpha atom of each residue with phosphorous atom in vertical and horizontal binding poses, given in red and blue, respectively. F Frequency of contact of the peptide with individual lipid species of the membrane. Vertical and horizontal binding poses are marked in red and blue, respectively

3.3. Stability of the membrane-bound conformations of Osh4 ALPS peptide

The population of the first peak is greater than the other two peaks, which suggests that the horizontal binding mode is the preferred membrane-bound conformation. To assess the binding stability of these major conformations, five system states were taken from each of the three tilt-angle distribution peaks, and short classical MD simulations were carried out with them as starting structures for 100 ns. We observed a stable binding without unbinding in all five systems taken from the first peak. In systems from both the second and third peaks, only two out of five systems exhibited stable binding. The peptide in all the other systems showed unbinding and binding events. To capture the stability of binding, we performed a Z-distance analysis, wherein we calculated the distance between the membrane phosphate plane and each residue. The average positions of all residues in the first and last 10 ns of the simulations are given in Fig. 4. The average positions of the main-helix residues in systems from the first peak clearly show that the peptide remained membrane-bound (Fig. 4A). In fact, the first five N-terminal residues, which were not bound to the membrane at the start of the simulation, were bound to the membrane toward the end. Conversely, we observe the opposite trend with systems from the second and third peaks (Fig. 4B, C). The average positions of most residues in systems from the latter two peaks shifted away from the membrane, owing to the unbinding events seen in some of the systems. Such a binding profile suggests that the horizontal binding mode (first peak) is indeed the most stable membrane-bound conformation.

Fig. 4.

Mean distance of all residues from the membrane plane (Z-distance) in the classical MD runs with starting configurations taken from A first peak, B second peak, and C third peak. The mean was taken across all five systems in each peak, and the values for the first and last 10 ns are given in blue and red, respectively. The mean value is plotted as a solid line, and the standard deviation is shaded around the solid line. The mean starting position in each peak is given as a solid black line. The membrane phosphate plane is marked by a black broken line at the z-position of 0 Å

To understand the difference in stability between the two binding modes, we calculated their peptide–membrane interaction energy, which includes only the short-range Lennard–Jones and electrostatic interactions. Even though we cannot use purely enthalpic component of the binding free energy to validate the population, we can make reserved comments. The average interaction energies as a function of tilt angle are given in Fig. 5A. The interaction energies decrease with decreasing tilt angle and reach a minimum average value for ~ 15° corresponding to the horizontal binding mode. For the vertical and horizontal modes, we obtained average interaction energies of −168.27 ± 64.98 and −205 ±.51 54.04 kcal/mol, respectively. In comparison, Monje-Galvan and Klauda reported average interaction energies of −60.46 ± 14.21 and −124.17 ± 9.59 kcal/mol for the vertical and horizontal modes, respectively [28]. Although both studies follow a similar trend, we report higher values than the classical MD simulations. The higher energies were expected as the base replica visited systems with higher Hamiltonian, increasing the peptide’s interaction energy with the membrane. Furthermore, the interaction energies for the vertical mode are just 0.8 times lower than the horizontal mode, unlike 0.5 times lower [28], as mentioned above. We believe that this discrepancy could be due to the better exploration of the vertical binding pose in REHT simulations. Furthermore, we calculated the free energy of binding with the tilt angle as the collective variable (Fig. 5B) from the populations we obtained from REHT simulations. We observed a global minimum corresponding to the horizontal binding mode and a flat region for the vertical binding modes. However, the difference in free energy between the two binding modes is just 1 kT, and therefore, transitioning between the two binding modes should not be restricted. Even though the difference in free energy and peptide–membrane interaction energy between the two binding modes is small, we do observe that the horizontal binding mode seems to be more stable.

3.4. REHT captures conformation variations as a function of membrane approach

The secondary structure of ALPS motifs is known to change upon membrane binding. There have been studies reporting both loss [46] and formation [47] of helix in ALPS motifs upon membrane binding. Hence, we wanted to check if REHT can capture changes in the secondary structure of the Osh4 ALPS peptide upon membrane binding. On visual examination, the striking observation was that there were more helical turns in the vertical binding mode (Fig. 1C) than horizontal mode (Fig. 1B). To quantify the helical character of the peptide as it approached the membrane, we calculated the two-dimensional (2D) free energy profile of the helical character and the center of mass (COM) distance of the peptide from the membrane. Figure 5C shows two major minima, one each for horizontal and vertical binding modes. The minimum nearer the membrane (COM distance of ~23 Å) corresponds to the horizontal binding pose and has a lower helicity than the minimum away from the membrane, which corresponds to the vertical binding mode. To differentiate the helicity between the two binding modes, we plotted the helical character of the peptide-only for the selected frames with membrane-bound peptide and intact main helix. Again, we clearly observe a difference in the helical character between the two binding modes. Figure 5D shows that the horizontal binding mode has a decreased helical character compared to the vertical binding mode, and this decrease comes majorly from the main-helix region (green plot in Fig. 5D). We observed that this loss in helicity was due to the unfolding of the last turn of the main helix toward the C-terminal side. Interestingly, such a loss of the C-terminal helical turn upon membrane binding was observed in the ALPS motif of Arf GTPase-Activating Protein 1 (ArfGAP1) protein [46]. Furthermore, Monje-Galvan and Klauda reported a small loss of helicity of the Osh4 ALPS peptide on binding to the membrane with the composition of yeast endoplasmic reticulum (ER) membrane. They observed that the last turn of the main helix partially unfolded before binding and regained helicity after binding [28]. Although they did not report such partial loss of helicity in the case of the PC-PS membrane, we do observe a loss of helicity, especially in the horizontal binding mode, as mentioned above. However, we do not observe any minimum where there is higher helicity with horizontal mode, which implies the lack of refolding seen by Monje-Galvan and Klauda [28].

3.5. REHT captures the contact frequency between peptide and lipids

To quantify the Osh4 ALPS peptide binding to the membrane, we performed the frequency of contact (FOC) analysis. A contact was said to be present if the C-alpha atom of any residue reached or went below the phosphate plane of the membrane. As expected, the two binding modes had different FOC profiles (Fig. 5E). The main-helix residues were the major membrane contacts for the horizontal binding mode. On the other hand, the first ten N-terminal residues were the predominant membrane contacts for the vertical binding mode. However, the FOC we observed was higher than the values reported by Monje-Galvan and Klauda [28]. This disparity in the fractions is because we have considered only the membrane-bound conformations with intact helix in our analysis. Nonetheless, our simulations capture the vital contacts. In both the binding modes, we correctly captured the importance of the SER 6–9 region. This region was reported to be the first to contact the membrane in case of stable binding [28]. Although we cannot comment on any temporal phenomenon from our REHT simulations, we do observe from the FOC analysis that this region seems to be important for both binding modes. Furthermore, we observed significant contacts in the TRP10-LYS15 region for the horizontal mode. Although Monje-Galvan and Klauda reported fewer contacts in the region TRP10-LYS15, they emphasized the stabilizing effect of the contacts made by TRP10, THR11, and LYS15 for the horizontal binding mode from their final stable membrane-bound conformations [28]. We also calculated the FOC of the peptide for each lipid species. Contrary to the reported results [28], we found that both binding modes established a similar number of contacts with both lipid species (Fig. 5F). In both binding modes, both DOPC and DOPS interacted with the peptide equally, emphasizing the role of electrostatic interactions in the binding process. This finding is in line with the reported values [28] and the general consensus that ALPS motifs utilize electrostatic interactions among other interactions to bind membranes [46–48].

3.6. REHT is efficient in sampling the membrane-bound protein conformations

To check the convergence of our REHT simulation, we counted the number of unique states (N_US) collected with time or frames in the base replica at the temperature of 303.15 K. The N_US with the system’s potential energy and the COM distance between the peptide and membrane as parameters (Fig. 6A) showed that the simulation converged by ~400 ns. Similarly, the N_US of tilt angle (Fig. 6B) showed convergence around ~6500 selected frames, which corresponds to ~325 ns. The convergence was further ascertained by plotting the maximum error in the probability density of each peak in time. As shown in Fig. 6C, the REHT simulation converged by ~400 ns. Furthermore, we did not observe much change in the tilt-angle density distribution after 400 ns (Fig. 6D). In total, we carried out REHT simulations of 500 ns for each of the 15 replicas, leading to a total simulation time of just 7.5 μs. In comparison, Monje-Galvan and Klauda performed six simulations with DOPC:DOPS membrane (NPT ensemble), amounting to 12 μs of total simulation time [28]. In almost half the simulation time, we clearly captured both the binding poses reported earlier. In addition, we could also explore another subset of the vertical binding mode corresponding to the third peak. Furthermore, REHT has the advantage that the binding poses are not user-defined, and the algorithm explores the different conformation of binding independently and exhaustively. Therefore, we could capture them in a single simulation trajectory (of the base replica) instead of using multiple replicates that gave the range of binding poses with Boltzmann’s weighted populations. This is an essential advantage of replica exchange as we could overcome energy barriers that prohibit shifting from one binding pose to another. Another advantage of REHT is the rapid phase-space exploration, which leads to quick contact with the membrane. The peptide took just ~25 ns to make contact with the membrane in REHT simulations compared to ~300 ns in classical MD simulations [28].

Fig. 6. Convergence of REHT simulations.

A Number of unique states (N_US) of potential energy and peptide COM distance from the membrane observed in the base replica is plotted against simulation time. B N_US of peptide tilt angle observed in the base replica is plotted against the number of selected frames. Note the convergence after around ~6500 frames, which correspond to ~325 ns. C Time evolution of maximum error in the three peaks of tilt-angle distribution. Orange, blue, and green colors represent first, second, and third peaks, respectively. D Evolution of the tilt-angle distribution from 350 ns till 500 ns. Note the convergence just after 400 ns

4. Conclusion

We tested the utility of the REHT method in exploring the equilibrium properties of the membrane-bound PMP conformations and compared it to the results from classical AAMD [28]. To do so, we used the Osh4 ALPS motif as the model system and elucidated its membrane-binding characteristics to a DOPC/DOPS model membrane. We demonstrated the two already reported major membrane-bound conformations of the peptide, namely horizontal and vertical binding modes, approximately two times faster than AAMD simulations. In addition, we could also show an additional vertical binding mode, which is most probably a transient conformation. Furthermore, the first contact between the peptide and the membrane is accelerated in REHT. REHT not only sampled the conformations efficiently but also captured the protein–membrane interactions accurately. The free energy profile obtained from the populations observed in the REHT base replica showed that the horizontal conformation was favored. Based on the equilibrium properties, the transition from vertical to horizontal binding mode most probably involves the unfolding of the last turn of the main helix. Although Monje-Galvan and Klauda observed such unfolding only with the yeast ER membrane, we showed that it happens even with the simple anionic model membrane. Our studies captured the importance of SER6-SER9 residues in initiating the membrane binding for both conformations. All the main-helix residues, especially TRP10, THR11, and LYS15, are essential to a stable horizontal bound conformation. Not only that, but REHT was also able to show the importance of contact with DOPS lipids in both major conformations. Therefore, we can conclude that REHT is efficient and accurate in characterizing the membrane-binding characteristics of peptides.

However, we must test the method for larger PMPs. We note that such a study must involve careful choice of the REHT parameters to capture the membrane-binding phenomenon accurately (please see the user manual in the supporting information file). The parameters must be chosen such that the uncharacteristic unfolding of protein or non-physiological membrane properties does not dominate the sampling process. Even though REHT can sample various protein–membrane landscapes, we must be cautious with their interpretation and choose relevant, physiological, and structural regions. In future studies, it may be difficult to know if an advanced sampling method like REHT has sampled the conformational landscape of membrane-bound peptides or PMPs exhaustively and has reached convergence. Since we had a large set of AAMD findings to compare with, we checked for the convergence of the relevant parameters and ascertained the convergence in our simulations. However, such a dataset may not be available in all cases, and thus, convergence must be checked using the relevant membrane-binding properties under consideration. Even so, stringent convergence criteria based on the system under consideration might have to be used to ensure convergence in sampling.

Data availability statement

All input files related to the REHT simulations of OSH4 systems are curated and publicly available at our laboratory Github repository: codesrivastavalab/REHT-PMP. We have also included a supporting information file as a USER MANUAL for the method.