We used atomistic MD simulations of Lam4S2 to explore the impact of the unique side-opening in Lam/GramD1 StARkin domains on the stability of bound sterol and its ability to exit the binding pocket. We chose cholesterol-bound Lam4S2 for these analyses because we (Gatta et al., 2015) and others (Gatta et al., 2018) have previously showed that Lam4S2 binds cholesterol. The protein also binds ergosterol, dehydroergosterol (DHE) (Gatta et al., 2015) and 25-hydroxycholesterol (25HC) (Jentsch et al., 2018). All these sterols appear to bind equivalently, as the extent of DHE binding to Lam4S2 is reduced in the presence of competing amounts of cholesterol or ergosterol (Gatta et al., 2015), and also of 25HC (Figure 1—figure supplement 2B). This point is highlighted by inspection of models of Lam4S2 loaded with cholesterol (Figure 1A; Figure 1—figure supplement 1B), ergosterol (Figure 1—figure supplement 2C) and 25HC (Figure 1—figure supplement 2D) in which all three sterols are seen to be positioned in essentially the same way in the binding pocket.

Cholesterol-bound Lam4S2 (Figure 1A) was placed in the vicinity of a membrane bilayer (Figure 1B) with the phospholipid composition of the anionic ‘Acceptor’ liposomes used for in vitro sterol transport assays (Jentsch et al., 2018), and its spontaneous binding to the membrane surface was monitored via ensemble MD simulations carried out in 10 statistically independent replicates (Stage 1 ensemble simulations, 3.2 µs cumulative time). The simulations revealed two modes by which Lam4S2 associated with the membrane (Figure 1C–E), both of which involved the interaction of clusters of positively charged amino acids on the surface of the protein (K203 at the C-terminus, K62/K90/K98, and K163/K167 (Figure 1D, E)) with the overall anionic surface of the membrane. In one mode (Mode 1, occurring with ~85% frequency [Figure 1C]), the protein interacted with lipid headgroups via its N- and C-termini (green and red; Figure 1D). In the second mode (Mode 2), observed with ~15% frequency, Lam4S2 engaged with the membrane via its Ω1 loop (purple; Figure 1C,E) and its C-terminal helix (yellow; Figure 1C,E). Consistent with the overall electrostatic properties of the protein (Figure 1—figure supplement 3A), both modes of binding are driven by electrostatic interactions between the protein and the membrane surface. Thus, Mode 1 implicates the protein face composed of the basic C-terminus region (e.g. K203) and nearby K62 (Figure 1—figure supplement 3A) as well as the positively charged N-terminal amino group, whereas Mode 2 is stabilized by contacts between the membrane and basic residues K163/K167 on the C-terminal helix and K98 in the Ω1 loop region.

Although more frequent, Mode 1 is probably a non-physiological consequence of simulating the isolated Lam4S2 domain. Indeed, the N- and C-terminal regions that stabilize this mode of association are covalently linked to the remainder of the protein chain in full-length Lam4, that is, the first StARkin domain and the transmembrane helix, respectively. Furthermore, in this mode, Lam4S2 is positioned on the membrane in such a way that the hydrophobic tail of cholesterol is directly exposed to the solvent (Figure 1D). In Mode 2, on the other hand, Lam4S2 is engaged with the membrane via structural regions that have been implicated in the function of the protein. Thus, the Ω1 loop is known to be a functionally important feature of all StARkin domains (Horenkamp et al., 2018; Gatta et al., 2018; Iaea et al., 2015), and the cationic residues K163/K167 in the C-terminal helix have been implicated in the in vitro sterol transfer activities of Lam4S2 (Jentsch et al., 2018; Gatta et al., 2018) and StARD4 (Iaea et al., 2015). Moreover, in Mode 2, cholesterol is oriented orthogonally to the plane of membrane, with its 3-β-OH group engaging Q121 in the binding pocket of Lam4S2 and its iso-octyl tail facing the membrane (Figure 1E).

We hypothesized that Mode 2, in which the protein appears to be situated correctly to release sterol into the membrane, is required for proper function. To test this premise, we considered the Lam4S2 K163D/K167D double mutant, which retains an electrostatic surface that is strongly positive at the C-terminus but is negatively charged around the C-terminal helix (compare panels [A] and [B] in Figure 1—figure supplement 3). The double mutant binds to membrane vesicles indistinguishably from the wild type protein (Figure 1—figure supplement 3C), but fails to transport sterol between vesicles, as we previously reported (Jentsch et al., 2018). These results suggest that this mutant, like the wild-type protein, binds the membrane predominantly via the unproductive Mode 1 form of association, which is unperturbed by the mutations. However, unlike the wild-type protein, the double mutant is unable to populate Mode 2. Indeed, when the K163D/K167D mutation was introduced into the computational model of Lam4S2 representing binding Mode 2 (Figure 1—figure supplement 3D), and the resulting system was subjected to ~50 ns MD simulations, we observed a spontaneous reorientation of the protein-membrane complex to the Mode 1 pose (Figure 1—figure supplement 3E). These results suggest that the K163D/K167D mutations destabilize Mode 2 and drive Lam4S2 to exclusively adopt the Mode 1 pose on the membrane surface. Importantly, given that the double mutant is functionally inactive, these observations confirm that Mode 1 is functionally irrelevant.

On the basis of these results, we chose not to explore Mode 1 further, but rather to test the premise that Mode 2 mediates release of sterol into the membrane. To this end, we enhanced the sampling of this mode of Lam4S2–membrane interaction by initiating a new set of 100 independent MD simulations (with random starting velocities) from 10 conformations of the system in which the Ω1 loop and the C-terminal helix were simultaneously engaged with lipids (Stage 2 ensemble simulations, 37.5 µs cumulative time). As described next, these trajectories revealed detailed mechanistic steps leading to spontaneous release of the protein-bound cholesterol into the membrane.

To facilitate analyses of the conformational dynamics of the membrane-bound Lam4S2–cholesterol complex in Stage 2 simulations, we used the time-structure-based independent component analysis (tICA) approach to reduce the dimensionality of the system (see 'Materials and methods'). To this end, we considered a set of collective variables (CVs) to describe the dynamics of cholesterol and relevant segments of the protein (i.e. the Ω1loop and the C-terminal helix) as well as to quantify the solvent exposure of the sterol-binding site (see 'Materials and methods' for details). All of the trajectory frames from Stage 2 simulations were projected onto the first two tICA vectors (Figure 2A), which represented ~90% of the total dynamics of the system (Figure 2—figure supplement 1A). The resulting 2D space (Figure 2A) was discretized for structural analyses into 100 microstates using the automated k-means clustering algorithm (Figure 2—figure supplement 1B). These microstates cover the conformational space of the system as the cholesterol molecule is transferred from the protein-bound state to the membrane.

Figure 2 with 6 supplements see all

Download asset Open asset

Figure 2—source data 1

Characteristics of different microstates from tICA approach.

https://cdn.elifesciences.org/articles/53444/elife-53444-fig2-data1-v2.zip

Download elife-53444-fig2-data1-v2.zip

Structural analyses of selected microstates on the tICA landscape (labeled 1–7 in Figure 2A), characterized by relevant CVs (Figure 2B) and visualized in structural snapshots (Figure 2C), describe key mechanistic steps of the sterol-release process. Microstate 1 represents an ensemble of states in which the sterol-binding cavity is occluded from both the solvent and the membrane. Thus, in Microstate 1 conformations (Figure 2B,C), cholesterol is stably bound in the protein (‘chol RMSD’ histogram [bottom panel, Figure 2B]), while the sterol-binding pocket is dehydrated (‘water count’ histogram [bottom panel, Figure 2B]) and sealed from the side by the side-chains of residues S181 and D61 that line the side-opening of the pocket (d_61–181 distance histogram [bottom panel, Figure 2B]). In addition, the Ω1 loop is positioned close to the C-terminal helix so that the C_α atoms of residues I95 in the Ω1 loop and A169 in the C-terminal helix are within ~10 Å of each other (d_95–169 distance histogram [bottom panel, Figure 2B]; see the middle structure in Figure 2C for the location of A169), thus occluding the sterol-binding pocket from below, that is the vantage point of the membrane. Indeed, in the ensemble of conformations representing Microstate 1, cholesterol has essentially no contact with membrane lipids (‘number of lipids’ histogram [bottom panel, Figure 2—figure supplement 2A and B]).

The first step in the sterol-release process involves widening of the side-entrance to the sterol-binding site, which is enabled by gradual separation of the side-chains of residues D61 and S181. This structural change on the tICA landscape can be followed in the evolution of the system from Microstate 1 to Microstates 2 and 3 (see d_61–181 distance histogram [Figure 2B]). Concomitant with the widening of the side-opening, the level of hydration (water count) of the binding pocket progressively increases (Figure 2B,C).

Cholesterol remains stably bound throughout these initial events (cholesterol RMSD is unchanged in Microstates 1–3). However, the rising level of hydration in the binding site results in destabilization of the polar interactions between the 3-β-OH group of cholesterol and the side-chain of residue Q121 as cholesterol initiates its translocation towards the membrane. Indeed, as the system transitions from Microstate 3 to Microstate 4, the RMSD of the cholesterol molecule increases (Figure 2B). Correspondingly, the minimum distance between the cholesterol oxygen and residue Q121 increases by ~4 Å (compare d_chol–121 for Microstates 1 and 4, Figure 2—figure supplement 2B). Notably, as the cholesterol molecule assumes this new position, the distance between the Ω1 loop and the C-terminal helix increases, as seen in the broadening of the d_95–169 histogram (Figure 2B), indicating initial opening of the sterol-binding pocket towards the membrane.

Cholesterol egress then proceeds through Microstates 5–7 in which the sterol-binding pocket remains open and solvated, while the Ω1 loop continues to sample conformations that position it relatively far from the C-terminal helix (Figure 2B,C). The cholesterol molecule leaves the binding pocket with its tail ‘down’, becoming gradually encapsulated by the hydrophobic chains of neighboring lipids until it fully embeds into the lipid membrane (‘number of lipids’ histogram [Figure 2—figure supplement 2B] and corresponding structural snapshots [Figure 2—figure supplement 2D]). The process of translocation is complete when the system reaches Microstate 7. The remaining part of the tICA space (corresponding to lower tIC1 and tIC2 values, that is, the bottom left region of the 2D space in Figure 2A) describes trajectory data in which Lam4S2 disengages from the membrane, after the release of the sterol, and diffuses into the solvent. Of note, the high level of hydration of the binding site in empty Lam4S2, that is after sterol egress, is recapitulated in MD simulations of the apo Lam4S2 system (initiated from the sterol-free Lam4S2 structure, PDB ID 6BYD) run under the same conditions as the Stage 1 simulations of cholesterol-bound Lam4S2 described in Figure 1 (see Figure 2—figure supplement 3 and 'Materials and methods' for more details).

The sterol translocation process outlined above was sampled in its entirety in 5 out of 100 Stage 2 simulations (trajectories highlighted in red in Figure 2—figure supplement 4). Another 19 trajectories in this set sampled the evolution of the system from Microstate 1 to Microstate 5 (trajectories marked with a green star in Figure 2—figure supplement 4), but on the time scales of these simulations, the system either did not progress further (i.e. to Microstates 6 and 7) or returned to Microstate 4 or 3 where it remained (see also below). In the remaining Stage 2 simulations, the system fluctuated between Microstates 1, 2, and 3 (unmarked trajectories in Figure 2—figure supplement 4).

The MD simulations indicate that widening of the side-opening to facilitate water penetration into the binding site (Figure 3) is a key step in the mechanism by which bound cholesterol leaves the protein to enter the membrane. To investigate in more detail the interplay between increased hydration of the sterol-binding pocket, widening of the side-entrance to the binding cavity, and stability of cholesterol within the pocket, we analyzed the dynamics of D61 and S181 and their interactions with other residues in the binding site during the simulations. We found that D61 is engaged in electrostatic interactions with residue K89 located in the β2 strand preceding the Ω1 loop. Thus, the side-chain of K89 faces the entrance of the binding pocket where it interacts with the anionic side-chain of D61 (Figure 2—figure supplement 2C, Figure 3). This interaction is maintained in the initial stages of the translocation process (Microstates 1–4), but becomes unstable as the hydration of the sterol-binding pocket reaches its highest levels after cholesterol leaves the site (Figure 3—figure supplement 1; also note sampling of a wide range of d_61–89 distances for Microstates 5–7 in Figure 2—figure supplement 2B). These data suggest that D61, S181 and K89 together participate in stabilizing the closed conformation of the side-entrance to the binding pocket.

Figure 3 with 1 supplement see all

Download asset Open asset

On the basis of these results and considering the position of the K89 side-chain near the protein-solvent interface, we hypothesized that replacing the polar and relatively long side-chain of K89 with a smaller hydrophobic moiety would promote widening of the side entrance, leading to destabilization of cholesterol in the binding pocket. Likewise, substituting D61 and S181 with residues with smaller-sized hydrophobic side-chains should also have a destabilizing effect on bound cholesterol.

To test these hypotheses, we computationally generated K89A, D61A, and S181A point-mutants of Lam4S2, and probed their dynamics using atomistic MD simulations. Specifically, we considered two snapshots taken at different time points (120 ns and 150 ns, respectively) from one of the Stage 2 trajectories (350 ns-long) of the wild-type protein system in which sterol release was observed. For the wild-type protein at these time points, the side-opening to the pocket is closed (Figure 2—figure supplement 5A), the cholesterol molecule is stably bound (Figure 2—figure supplement 5B), and the level of hydration is relatively low (between 5–10 water molecules as shown in Figure 2—figure supplement 5C). We introduced the three mutations separately into these two snapshots, and (for each construct) carried out 150-ns-long unbiased MD simulations in 10 replicates (1.5 µs total simulation time). Analysis of these trajectories revealed that for all three of the mutants, the hydration level of the sterol-binding site increased rapidly during the initial 4–5 ns of the simulations (Figure 2—figure supplement 6A). Note that in the original wild-type trajectory, reaching the same high level of hydration (>20 water molecules, Figure 2—figure supplement 6C) required a considerably longer time (~180 ns, Figure 2—figure supplement 5C). Furthermore, in the trajectories for the mutant proteins, cholesterol was destabilized in its binding pocket (Figure 2—figure supplement 6B,D). On the simulation timescales, rapid destabilization was especially notable for the K89A system in which, for all but one replicate, the sterol was unstable in its binding site (panel labeled K89A, Figure 2—figure supplement 6D).

To address the effect of the K89A mutation on cholesterol stability quantitatively, we compared the energetics of sterol release in the K89A mutant versus the wild-type system using umbrella sampling MD simulations. We constructed the potential of mean force (PMF) for cholesterol release by constraining the z-distance between the sterol hydroxyl oxygen and the C_α atom of residue Q121, d_{Z(chol–121)} to different values in the range $\in$ [2Å; 20 Å] along the release pathway (d_{Z(chol–121)} histogram (Figure 2—figure supplement 2B)). The results are shown in Figure 4A. For the wild-type system, the PMF calculations indicate that cholesterol release requires that an energy barrier of ~6 kcal/mole is overcome, and proceeds through two major steps that were also identified in our tICA analysis of Stage 2 simulations. Thus, the PMF has a global minimum at d_{Z(chol–121)} ~2 Å, corresponding to the position of cholesterol in the binding site where its polar head-group is coordinated by residue Q121 (snapshot at the top right of Figure 4A), and two local minima (LM-1 and LM-2) at d_{Z(chol–121)} ~10–14 Å and > 18 Å, respectively. The global minimum represents the ensemble of states found in Microstate 1-3 (d_{Z(chol–121)} histogram, Figure 2—figure supplement 2B), whereas LM-1 corresponds to the ensemble of states found in Microstate 5 (d_Z(chol-121) histogram, Figure 2—figure supplement 2B).

Figure 4 with 3 supplements see all

Download asset Open asset

Figure 4—source data 1

Potential of mean force for wild-type and K89A Lam4S2.

https://cdn.elifesciences.org/articles/53444/elife-53444-fig4-data1-v2.zip

Download elife-53444-fig4-data1-v2.zip

Figure 4—source data 2

Water counts in wild-type and K89A Lam4S2 corresponding to different umbrella windows.

https://cdn.elifesciences.org/articles/53444/elife-53444-fig4-data2-v2.zip

Download elife-53444-fig4-data2-v2.zip

The PMF calculations reveal that the energy barrier that separates LM-1 from the global minimum is ~5 kcal/mole (red trace in Figure 4). This high energy cost is associated with the clear change in hydration of the sterol-binding site and the concomitant opening of the side-opening to the pocket (see WT profiles in Figure 4B). Indeed, the water count increases and the D61–S181 interaction is destabilized when the system transitions from d_{Z(chol–121)}∈ [2 Å; 6 Å] to d_{Z(chol–121)} ≥ 7 Å (Figure 4, Figure 4—figure supplement 1A, C). LM-2 represents the ensemble of states in which cholesterol is on the verge of exiting the protein, that is, states in which cholesterol is mostly engaged by lipids and with its head-group on the level of the Ω1 loop (see Microstate 6 in Figure 2—figure supplement 2D). LM-1 and LM-2 are separated by an energy barrier of ~2 kcal/mole. Overall, the presence of multiple minima on the PMF plot is consistent with our findings from the tICA analysis of the unbiased MD simulations described above: in some of the Stage 2 trajectories, the system evolved from Microstate 1 to Microstate 5 (i.e. transitioned from the global minimum to LM-1 on the PMF plot), but either did not progress further to complete sterol egress (i.e. they did not reach LM2) or returned to the conformational space of the tICA landscape characterized by relatively low hydration of the sterol-binding pocket (i.e. they returned to the global energy minimum). For comparison, we also investigated the energetics of sterol release from wild-type Lam4S2 in water. In this case, the PMF profile no longer has LM-1 and LM-2 minima (Figure 4—figure supplement 2A). Instead, only one global minimum can be found, corresponding to the position of cholesterol in the binding site. The energy difference between the bound and released states on the PMF profile in water is >10 kcal/mole, at least ~2-fold higher than the energy barrier for sterol release into the membrane. These differences can be explained by the different solvent exposure of the hydrophobic tail of cholesterol during the egress process in water versus that in the membrane (Figure 4—figure supplement 2B).

Remarkably, comparison of the PMF plots for the wild-type and the K89A systems (Figure 4A) reveals that the mutation significantly lowers the barriers that must be overcome in order for transition between the different energy minima to occur. Thus, although the PMF profile for the K89A construct still has three energy minima, the energy cost of transitioning between the global minimum and LM-1 in this system is ~2 kcal/mole, and between LM-2 and LM-3 is ~1 kcal/mole, resulting in an energy barrier of only ~3 kcal/mole for the entire release process (d_{Z(chol–121)}∈ [2 Å; 20 Å]). This is approximately half of that determined for the wild-type system. This reduction in the energy cost can be explained by the greater extent of hydration of the binding site in K89A compared to that in the wild-type protein. Indeed, in wild-type Lam4S2, the binding site remains largely dehydrated until cholesterol disengages from Q121 (d_{Z(chol–121)}∈ [2 Å; 6 Å]) (Figure 4B, Figure 4—figure supplement 1A), whereas in the K89A protein, the level of solvation of the binding pocket is relatively high (>10 water molecules) even when cholesterol is interacting with Q121 (Figure 4B, Figure 4—figure supplement 1B). These trends in binding-site hydration are mirrored by changes in the d_61–181 distance along the d_{Z(chol–121)} coordinate (note destabilization of D61–S181 interactions in the K89A system versus those in the wild-type system for small d_{Z(chol–121)} values in Figure 4—figure supplement 1C, D).

Interestingly, the global minimum on the PMF profile of the K89A mutant is shifted compared to its location on the PMF plot of the wild-type system, from d_{Z(chol–121)} ~2 Å to ~5 Å (Figure 4A). We found that, at the shortest d_{Z(chol–121)} distances, cholesterol–Q121 interactions in the mutant are mostly mediated by water molecules, whereas at d_{Z(chol–121)} ~5 Å, the hydroxyl group of cholesterol is in direct contact with Q121 (see sharp peak at ~2 Å for the d_{Z(chol–121)} = 5 Å plot in Figure 4—figure supplement 1B; note that d_{Z(chol–121)} is the Z-distance between the hydroxyl and the C_α of Q121). This may also explain why the water content in the cavity is skewed towards lower values for d_{Z(chol–121)} ~5 Å (Figure 4B). Thus, for both the wild-type and K89A systems (and for the wild type simulated in water, Figure 4—figure supplement 2A), the global minimum on the PMF plot corresponds to the ensemble of states in which cholesterol is engaged in direct interactions with Q121. Overall, the PMF calculations reveal that the K89A substitution lowers the energy barrier for cholesterol release from Lam4S2 into the membrane and suggests that cholesterol is consequently less stable in the binding pocket.

Our computational studies indicate that substitution of D61, K89 or S181 with alanine affects the degree of hydration of the sterol-binding pocket and the stability of bound sterol, with the most significant effects seen for the K89A mutant. We tested the functionality of K89A and the other mutants using three types of experiments.

We previously showed that yeast cells lacking Ysp2/Lam2 (lam2Δ cells) are sensitive to the polyene antibiotic amphotericin B, and that this phenotype can be corrected by expression of a soluble GFP–Lam4S2 fusion protein (Gatta et al., 2015). We verified that this was also the case for nystatin, another polyene antibiotic (Figure 5A, compare the first two rows in which lam2Δ cells are transformed with either an empty vector (row 1) or a vector for expression of GFP–Lam4S2 (row 2, WT), and plated on media without nystatin or with different amounts of nystatin). We then tested the ability of GFP-fused Lam4S2 proteins carrying either K89A, D61A, or S181A single-point mutations (GFP–Lam4S2(K89A), GFP–Lam4S2(D61A), GFP–Lam4S2(S181A), respectively) to rescue the nystatin sensitivity of the lam2Δ cells. Figure 5A shows that lam2Δ cells expressing GFP–Lam4S2(K89A) remained nystatin-sensitive, whereas those expressing GFP–Lam4S2(D61A) or GFP–Lam4S2(S181A) became resistant to the antibiotic, similar to lam2Δ cells expressing wild-type protein. As all the Lam4S2 variants tested were expressed at equivalent levels (as revealed by SDS-PAGE immunoblotting [Figure 5B]), this cell-based assay indicates that the K89A mutant has a functional deficit, whereas the D61A and S181A proteins are able to provide cells with sufficient functionality to rescue their nystatin-sensitivity phenotype.

Figure 5

Download asset Open asset

To test explicitly the ability of the mutants to extract sterol from membranes and to catalyze sterol exchange between populations of vesicles, we expressed His-tagged versions of the proteins in Escherichia coli and purified them by affinity chromatography and size exclusion. The D61A mutant proved unexpectedly problematic on account of its low yield and apparent instability, and so we focused on the S181A and K89A mutants (Figure 6A). Like wild-type Lam4S2, these mutants displayed monodisperse profiles on size exclusion (Figure 6B) and yielded circular dichroism spectra that were indicative of well-folded structures (Figure 6C).

Figure 6

Download asset Open asset

Sterol extraction assays were performed by incubating the purified proteins with large, unilamellar vesicles containing [³H]cholesterol, and determining the amount of radioactivity and protein in the supernatant after ultracentrifugation to pellet the vesicles. Relative to the wild-type protein, the S181A mutant extracted only ~50% of sterol under our standard incubation conditions, whereas the K89A mutant had essentially no ability to extract sterol (Figure 6D).

To probe the sterol transfer activity of the Lam4S2 mutants, we performed in vitro sterol transport assays, as previously described and depicted schematically in Figure 7A. Donor vesicles containing fluorescent dehydroergosterol (DHE) were incubated with acceptor vesicles containing the Förster resonance energy transfer (FRET) acceptor dansyl-PE. Excitation of DHE results in sensitized fluorescence emission from dansyl-PE only when the two lipids are in the same vesicle. Figure 7B (see also Figure 7D) shows that under our standard conditions, the wild-type protein increases the rate of DHE exchange ~7 fold over the spontaneous rate. The performance of the S181A mutant was similar to that of the wild-type protein, whereas the K89A mutant had essentially no activity (Figure 7C,D).

Figure 7

Download asset Open asset

Overall, the three functional tests described above indicate that the K89A mutant is compromised in sterol handling—it is unable to extract sterol from membranes or to transfer it between vesicles, accounting for its inability to rescue the nystatin sensitivity of lam2Δ cells. These functional outcomes are in line with our computational prediction that cholesterol would be unstable in the binding site of the Lam4S2 K89A. Interestingly, the partial inability of the S181A mutant to extract cholesterol did not affect its ability to catalyze sterol exchange or to rescue the nystatin sensitivity of lam2Δ cells.

An all-atom model lipid membrane with the composition of ‘Acceptor’ liposomes in sterol transport assays (Jentsch et al., 2018) was prepared using the CHARMM-GUI web server (RRID:SCR_014892) (Jo et al., 2009). Thus, symmetric lipid bilayer containing 70% DOPC, 15% PI, 10% DOPE, and 5% DOPS (400 lipids in total on the two leaflets) was assembled, solvated (using water/lipid number ratio of 50) and ionized with 0.1M K⁺Cl^– salt. This system was subjected to MD simulations for 30 ns using NAMD version 2.13 (RRID:SCR_014894) (Phillips et al., 2005) and the standard multi-step equilibration protocol provided by CHARMM-GUI.

After this equilibration phase, the bilayer system was stripped of all water molecules and solution ions and the cholesterol-bound Lam4S2 domain (6BYM) was placed near the membrane surface so that the distance between any atom of the protein and any atom of the lipid molecules was ≥10 Å (see Figure 1B). The protein-membrane complex was solvated (using water/lipid number ratio of ~145) and ionized (with 0.1M K⁺Cl^– salt). The resulting system contained ~234,000 atoms in total.

The Lam4S2–membrane complex was equilibrated using a multi-step protocol (Shi et al., 2008) during which the backbone of the protein was first harmonically constrained and subsequently gradually released in three steps of 5 ns each, changing the restraining force constants from 1, to 0.5, and 0.1 kcal/ (mol Ų), respectively. This step was followed by 6-ns-long unbiased MD simulations carried out using the NAMD 2.13 package. After this short run, the velocities of all the atoms were reset and the system was simulated with ACEMD software (Harvey et al., 2009) in 10 statistically independent replicates (Stage 1 ensemble simulations), each for 320 ns, resulting in a cumulative time of 3.2 µs for Stage 1 runs.

As described in 'Results', Stage 1 simulations sampled events of spontaneous binding of Lam4S2 to the membrane. We randomly selected 10 frames from Stage 1 trajectories in which Lam4S2 was seen to be interacting with the lipid bilayer as in Figure 1E, and initiated a new set of simulations with ACEMD (Stage 2 ensemble simulations) in which the 10 chosen structures were run in 10 statistically independent replicates each (i.e. 100 independent simulations). Each of the 100 copies were simulated for 375 ns, resulting in a cumulative time of 37.5 µs for Stage 2 runs.

Simulations of the apo wild-type Lam4S2 protein (6BYD) followed the same protocol as described above for Stage 1 simulations of sterol-bound Lam4S2, with the only difference being the lipid membrane composition. Thus, in the manner identical to the sterol-bound Lam4S2, the apo protein was placed near the surface of the all-atom model lipid membrane (assembled with CHARMM-GUI) with the composition of ‘Donor’ liposomes in sterol transport assays (Jentsch et al., 2018). This symmetric bilayer contained 31% DOPC, 23% DOPE, 23% DOPS, and 23% cholesterol (400 lipids in total on the two leaflets). As the purpose of these simulations was to quantify solvation of the empty sterol-binding site, this system was only considered for Stage 1 simulations (cumulative time of 3.2 µs) and was not subjected to a subsequent (Stage 2) phase.

Using the FoldX server (RRID:SCR_008522) (Schymkowitz et al., 2005), three single mutations, K89A, S181A, and D79A in Lam4S2 were introduced into two separate frames of one of the Stage two ensemble trajectories of the wild type protein system (see 'Results'). The resulting structures (two per mutant) were energy-minimized for 100 steps and then simulated in five independent replicates each for 150 ns using ACEMD. This resulted in 10 statistically independent MD trajectories per mutant, totaling 1.5 µs.

All the simulations performed with NAMD 2.13 implemented the all option for rigidbonds, 2fs integration time-step, and PME for electrostatics interactions (Essmann et al., 1995), and were carried out in NPT ensemble under semi-isotropic pressure coupling conditions, at a temperature of 310 K. The Nose-Hoover Langevin piston algorithm (Phillips et al., 2005) was used to control the target p=1 atm pressure with the LangevinPistonPeriod set to 100 fs and LangevinPistonDecay set to 50 fs. The van der Waals interactions were calculated by applying a cutoff distance of 12 Å and switching the potential from 10 Å. In addition, the vdwforceswitching option was set to on.

The simulations carried out with ACEMD software implemented the PME method for electrostatic calculations, and were carried out according to the protocol developed at Acellera and implemented by us previously (Harvey et al., 2009; Khelashvili et al., 2015),with four fs integration time-step and the standard mass repartitioning procedure for hydrogen atoms. The computations were conducted under the NVT ensemble (at T = 310 K), using the Langevin Thermostat with Langevin Damping Factor set to 0.1.

For all the simulations, the CHARMM36 force field parameters were used for proteins, lipids, sterols, and ions (Phillips et al., 2005; Lee et al., 2016).

Biased MD simulations of cholesterol release from the wild-type and the K89A mutant Lam4S2 were performed using the umbrella sampling approach. The position of the translocated cholesterol was restrained to different locations along the translocation pathway (see 'Results') using as a collective variable the z-directional distance (d_{Z(chol–121)}along the axis perpendicular to the membrane plane), between the cholesterol oxygen and the C_α atom of residue Q121 (see Figure 1A). 19 windows spaced 1Å apart in the range of d_{Z(chol–121)}∈ [2Å; 20Å] were considered, and the dynamics of the sterol molecule in each window was restrained by applying a force constant of 2.5 kcal/mol ∙ Ų. The rest of the parameters for the umbrella sampling runs were as follows: width – 2Å, and both lowerwallconstant and upperwallconstant set to 25 kcal/mol ∙ Ų. Each umbrella window was simulated for 50 ns, which resulted in good overlap between adjacent windows (Figure 4—figure supplement 3A,B).

The potential of mean force (PMF) along the collective variable was constructed with the Weighted Histogram Analysis Method (WHAM) (Version 2.0.9; Grossfield, 2013). For the WHAM calculations, only the last 25 ns trajectory segments of each umbrella window were used. The tolerance parameter was set to 0.0001. To estimate error bars on the PMF, for each umbrella window, the first decorrelation time was calculated as a time-constant from a single exponential fit to the auto-correlation vs time data (Figure 4—figure supplement 3C,D). The error bars were then constructed with Monte Carlo bootstrapping error analysis in the WHAM software on the decorrelated data points using num_MC_trials of 1000.

Umbrella sampling MD simulations were also carried out on wild-type Lam4S2 in solution. To this end, structures of the protein–cholesterol complexes from the same 19 windows were extracted and immersed into a water box and the dynamics of the system was studied by restraining the sterol molecule in each window as described above. To prevent the protein–cholesterol complex from rotating in solution, we applied an additional harmonic bias (force-constant of 4000 kcal/mol ∙ Ų) on the protein backbone around the identity rotation matrix (1.0, 0.0, 0.0, 0.0) using the colvar ‘orientation’ module in NAMD 2.13. We note that this rotational restraint does not influence internal degrees of freedom but rather ensures that in each window, the protein–cholesterol complex remains aligned along the z axis throughout the umbrella sampling protocol.

To facilitate analysis of the process of cholesterol release from the Lam4S2 domain in the MD simulations, we performed dimensionality reduction using the tICA approach (Molgedey and Schuster, 1994; Naritomi and Fuchigami, 2011; Pérez-Hernández et al., 2013; Schwantes and Pande, 2013) as described previously (Morra et al., 2018; Lee et al., 2018; Razavi et al., 2018). To define the tICA space, we used several dynamic variables extracted from the analysis of the ensemble MD trajectories that quantify the dynamics of the cholesterol, the extent of exposure of the sterol binding site to the solvent, and the dynamics of the functionally important Ω1 loop. These variables include (see Figure 1A): (1)-the minimum distance between the hydroxyl oxygen atom of the translocated cholesterol and residue Q121 (d_chol–121); (2)-the root-mean-square deviation (RMSD) of the cholesterol molecule from its position in the binding site; (3)- distance between the hydroxyl oxygen of S181 and C_γ carbon of D61 (d_61–181); (4)-C_α – C_α distance between residue I95 in the Ω₁ loop and residue A169 in the C-terminal helix (d_95–169); (5)-number of water molecules in the interior of the protein (defined as the number of water oxygens found within 5 Å of the side-chains of the following protein residues – 189, 185, 181, 154, 152, 136, 138, 140, 142, 123, 121, 119, 117, 102, 104, 106, 108, but farther than 5 Å from the following residues – 116, 118, 109, 86, 103, 105); (6)-the number of lipid phosphate atoms with 3 Å of the translocated cholesterol molecule.

Using these six CVs as components of the data vector X, the slowest reaction coordinates of a system were found as described previously (Morra et al., 2018; Razavi et al., 2018; Razavi et al., 2017), by constructing a time-lagged covariance matrix (TLCM): C_TL(τ)=<X(t)X^T(t+τ)> and the covariance matrix C=< X(t)X^T(t)>, where X(t) is the data vector at time t, τ is the lag-time of the TLCM, and the symbol <…> denotes the time average. The slowest reaction coordinates are then identified by solving the generalized eigenvalue problem: CTLV = CVΛ, where Λ and V are the eigenvalue and eigenvector matrices, respectively. The eigenvectors corresponding to the largest eigenvalues define the slowest reaction coordinates.

The electrostatic potential for the wild-type and the K163D/K167D double mutant Lam4S2 constructs was calculated by solving the non-linear Poisson-Boltzmann (NLPB) equation with APBS software (Baker et al., 2001) as described before (Khelashvili, 2019). The protein and solvent dielectric constants were set to 2 and 78.54, respectively. The solution contained 150 mM of monovalent mobile ions. The NLPB equation was solved on a 3D cubic grid of 256 ų volume discretized into 1 ų grid elements, using the multigrid method (mg-manual) and multipole boundary conditions.

Lam4S2 and point mutants were expressed in E. coli as His-tagged proteins (Table 1), and purified by affinity chromatography on Ni-NTA resin, followed by size exclusion chromatography (SEC) using a Superdex 200 Increase 15/300 GL column. The purification procedure was as previously described (Jentsch et al., 2018), except that the proteolysis step to remove the affinity tag was omitted and SEC was carried out in 20 mM HEPES [pH 7.5] and 150 mM NaCl. The purified protein was snap frozen in small aliquots and stored at −80°C. Prior to use, aliquots were thawed and subjected to brief microcentrifugation to remove any aggregated material. Purified proteins were quantified by absorbance at 280 nm; quality control included analysis by circular dichroism (CD) as described previously (Jentsch et al., 2018), and re-analysis by SEC using the buffer conditions described above.

The assay (illustrated in Figure 7A) was performed and analyzed as previously described (Jentsch et al., 2018; Chauhan et al., 2019) using anionic donor and acceptor liposomes (donor lipid composition: DOPC, DOPE, DOPS, DHE [31, 23, 23, and 23 mol %, respectively]; acceptor lipid composition: DOPC, DOPE, liver PI, DOPS, dansyl-PE [70, 7, 15, 5, and 3 mol %, respectively]). Briefly, assays were carried out at 23°C in a quartz cuvette with constant stirring using a temperature-controlled Horiba Fluoromax Plus-C spectrofluorometer. The total sample volume was 2 ml, with 0.1 mM each of donor and acceptor liposomes (final concentration, based on measurement of inorganic phosphate after acid hydrolysis of the vesicles) and 0.05 μM or 0.1 μM Lam4S2 (final concentration) in 20 mM PIPES (pH 6.8), 3 mM KCl and 10 mM NaCl (assay buffer). Fluorescence was monitored for ~2500 s using λ_ex = 310 nm and λ_em = 525 nm and a data acquisition frequency of 1 Hz. Acceptor liposomes were added to donor liposomes in the cuvette, and after 60 s, 200 µl of Lam4S2 (or Lam4S2-mutant), diluted as needed in assay buffer, was added. For control assays, 200 µl of assay buffer was added. All traces were offset-corrected such that the fluorescence signal and time at the point of Lam4S2 (or buffer) addition were each set to zero. The maximum possible FRET signal was determined from assays using 0.1 µM wild-type Lam4S2, where the fluorescence readout reached a plateau value within 2000 s. Traces from such assays (done in replicate) were fit to a mono-exponential function, and the plateau value obtained (FRET_max) was used to constrain the mono-exponential fits of all other traces. Traces from different assays were compiled after data fitting by setting FRET_max = 1.

Sucrose-loaded liposomes were prepared as follows. Lipids (2 μmol total, of a mixture of DOPC, DOPE, DOPS, cholesterol (49, 23, 23, 5 mol %, respectively, containing a trace amount of [³H]cholesterol and N-rhodamine-DHPE) were dried in a glass screw-cap tube under a stream of nitrogen, then resuspended in 1 ml assay buffer (20 mM PIPES [pH 6.8], 3 mM KCl, and 10 mM NaCl) supplemented with 250 mM sucrose, by agitating on a Vibrax orbital shaker for 30 min at 1200 rpm. The resulting suspension was subjected to five cycles of freeze–thaw (immersion in liquid nitrogen, followed by thawing at room temperature), before being extruded 11 times through a 200 nm membrane filter using the Avanti Mini-Extruder. After extrusion, extravesicular sucrose was removed by diluting the vesicles 4x in assay buffer and centrifuging in a Beckman TLA100.3 rotor (75,000 rpm, 1 hr, 4°C). The supernatant was carefully removed from the pelleted vesicles (easily discernable because of their pink color due to rhodamine-DHPE), before resuspending the vesicles in 1 ml of assay buffer. Aliquots of the sample (5 μl) were removed at different points of preparation (after the freeze–thaw step, post-extrusion and after final resuspension) and taken for liquid scintillation counting to track lipid recovery by monitoring [³H]cholesterol.

The ability of Lam4S2 (wild-type and point mutants) to extract cholesterol from the vesicles was determined as follows. Liposomes (15 μl, ~1500 pmol cholesterol) and protein (500–750 pmol, as indicated) were combined in assay buffer to a total volume of 500 μl. The mixture was incubated at room temperature for 1 hr, before removing a 20 μl aliquot for liquid scintillation counting. The remainder of the sample was centrifuged in a Beckman TLA100.2 rotor (75,000 rpm, 1 hr, 4°C). Most of the supernatant (350 μl) was transferred to a fresh 1.5 ml tube, whereas the remainder was removed immediately and discarded. The pellet was resuspended in 100 μl assay buffer containing 5% (w/v) SDS. Duplicate aliquots (50 μl each) of the supernatant were taken for liquid scintillation counting to determine the amount of extracted cholesterol. Protein in the remainder of the supernatant (250 μl) and the resuspended pellet was precipitated by adding 1.2 ml ice-cold acetone, followed by overnight incubation at −20°C. The precipitated proteins were pelleted by centrifugation, air-dried after removal of the acetone, and dissolved in SDS gel loading buffer. The relative amount of protein in the supernatant and pellet fractions was determined by SDS-PAGE, Coomassie staining and quantification of band intensity using Image J software (RRID:SCR_003070). The data are represented as pmol cholesterol extracted/pmol protein in the supernatant.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Tim Levine and Ganiyu Alli-Balogun (Department of Cell Biology, University College London Institute of Ophthalmology) for plasmids, Trudy Ramlall (Eliezer laboratory, Weill Cornell Medical College) for help with protein purification, and Harel Weinstein for insightful comments on the manuscript. GK would like to thank Jerome Henin for his advice on umbrella sampling simulations. AKM would like to acknowledge Bob Dylan and Sam Canis for the usual reasons, as well as the unusual stimulation provided by Typhoon Hagibis. The computational work was performed using resources of the Oak Ridge Leadership Computing Facility (ALCC allocation BIP109 and Director’s Discretionary allocation) at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under contract no. DE-AC05-00OR22725, and the computational resources of the David A Cofrin Center for Biomedical Information in the HRH Prince Alwaleed Bin Talal Bin Abdulaziz Alsaud Institute for Computational Biomedicine at Weill Cornell Medical College.

© 2019, Khelashvili et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.