Water for sterol: an unusual mechanism of sterol egress from a StARkin domain

Previously we identified a new family of endoplasmic reticulum membrane proteins that possess sterol-binding StARkin domains (Gatta et al. eLife 2015). These Lam/GramD1 proteins are implicated in intracellular sterol homeostasis, a function that requires them to be able to bind sterols. Here we show how these proteins exchange sterol molecules with membranes. An aperture at one end of the StARkin domain enables sterol to enter/exit the binding pocket. Strikingly, the wall of the pocket is fractured along its length, exposing bound sterol to solvent. We considered whether hydration of the pocket could mediate sterol entry/exit. Large-scale atomistic molecular dynamics simulations reveal that sterol egress involves widening of the fracture, penetration of water into the cavity and consequent destabilization of the bound sterol. The simulations also identify polar residues along the fracture that are important for sterol release. Their replacement with alanine affects the ability of the StARkin domain to bind sterol, catalyze inter-vesicular sterol exchange and alleviate the nystatin-sensitivity of lam2Δ yeast cells. These data suggest an unprecedented, water-controlled mechanism of sterol acquisition and discharge from a StARkin domain.

are implicated in intracellular sterol homeostasis, a function that requires them to be able to 23 bind sterols. Here we show how these proteins exchange sterol molecules with membranes. An 24 aperture at one end of the StARkin domain enables sterol to enter/exit the binding pocket. 25 Strikingly, the wall of the pocket is fractured along its length, exposing bound sterol to solvent. 26 We considered whether hydration of the pocket could mediate sterol entry/exit. Large-scale 27 atomistic molecular dynamics simulations reveal that sterol egress involves widening of the 28 fracture, penetration of water into the cavity and consequent destabilization of the bound 29 sterol. The simulations also identify polar residues along the fracture that are important for 30 sterol release. Their replacement with alanine affects the ability of the StARkin domain to bind 31 sterol, catalyze inter-vesicular sterol exchange and alleviate the nystatin-sensitivity of lam2D 32 yeast cells. These data suggest an unprecedented, water-controlled mechanism of sterol 33 acquisition and discharge from a StARkin domain. 34 Cholesterol, the 'central lipid of mammalian cells' (1), is the most abundant molecular 2 component of the mammalian plasma membrane (PM) where it represents one out of every 2-3 3 lipids (1,2). Like many membrane lipids, it is synthesized in the endoplasmic reticulum (ER) 4 and transported to the PM by non-vesicular mechanisms that make use of lipid transport 5 proteins (3,4). These proteins operate as molecular ferries, achieving lipid exchange between 6 membranes by reversibly extracting a lipid from the cytoplasmic leaflet of one membrane 7 bilayer, encapsulating it within a binding pocket for transfer through the cytoplasm, and 8 depositing it in the cytoplasmic leaflet of another membrane. Proteins with steroidogenic acute 9 regulatory protein related lipid transfer (StART) domains constitute a major family of 10 intracellular lipid transport proteins -the StARkin superfamily -implicated in moving 11 glycerophospholipids, ceramide and sterol between cellular membranes (5,6). Whereas these 12 proteins are generally soluble and able to diffuse freely through the cytoplasm, a new family of 13 ER membrane proteins with StARkin domains was recently identified, including six members 14 (Lam1-Lam6) in the budding yeast Saccharomyces cerevisiae and three members (GramD1a-15 GramD1c) in mammals (7-10). Members of this new sub-family have one or two StARkin 16 domains that bind sterols and catalyze sterol exchange between populations of vesicles in vitro 17 (7,9,(11)(12)(13)(14). Lam1-Lam4 localize to ER-PM contact sites in yeast (7,15) where they play a role in 18 sterol homeostasis. Thus, yeast cells lacking one or more of these proteins are hypersensitive to 19 the sterol-binding, polyene antibiotics amphotericin and nystatin, implying alterations in PM 20 sterol content and/or organization (7,16). Furthermore, they esterify exogenously supplied 21 sterols up to 3-fold more slowly than wild-type cells, indicative of a delay in some aspect of PM-22 ER sterol transport (7,16). A sterol homeostatic role has also been suggested for mouse 23 GramD1b protein which is highly expressed in steroidogenic organs. Thus, adrenal glands from 24 a GramD1b knockout mouse are devoid of lipid droplets and show a severe reduction in 25 cholesteryl ester content (14). 26 We recently reported crystal structures of the second StARkin domain of Lam4 (here 27 termed Lam4S2) in apoand sterol-bound states (11). The protein has an overall α/β helix-grip 28 fold that forms a capacious binding pocket into which the sterol appears to be admitted head-29 first, via an aperture at one end, such that its 3-b-hydroxyl head-group is stabilized by direct or 30 water-mediated interactions with polar residues ( Figure 1A). The surface of the protein near 31 the entrance to the pocket is decorated with lysine residues, accounting for the enhanced 32 ability of Lam4S2 to transfer sterol between anionic vesicles compared with neutral vesicles 33 (11); the entryway itself is partially occluded by a flexible loop, termed Ω1 ( Figure 1A), whose 34 functional importance in the StARkin family is well-documented through mutagenesis studies 35 (12,17,18). The structures of other Lam/GramD1 StARkin domains are similar (12-14), broadly 36 resembling structures of other members of the StARkin superfamily except for one striking 37 feature. The wall of the sterol binding cavity in all Lam/GramD1 StARkin domains is fractured 38 along part of its length, exposing the sterol backbone to bulk solvent (Figure 1-figure  39 supplement 1). We considered whether this unusual structural feature -henceforth termed 40 'side-opening' -might provide a mechanism to control the stability of sterol within the binding 41 pocket. We posited that the ability to load sterol into the pocket, or discharge it from the 42 pocket into the membrane, might be controlled by water permeation via the side-opening. We 43 used a combination of large-scale atomistic molecular dynamics (MD) simulations and 44 functional tests to explore this hypothesis. Analysis of extensive ensemble and umbrella 1 sampling MD trajectories revealed that sterol egress from Lam4S2 is associated with widening 2 of the side-entrance to the binding pocket, penetration of water molecules into the cavity and 3 consequent destabilization of the bound sterol. The simulations identified several polar 4 residues that line the side-entrance to the pocket and that appear to play a critical role in the 5 initial steps of the release process. The functional importance of these residues was validated 6 experimentally by showing that their replacement with alanine compromises the ability of 7 Lam4S2 to rescue the nystatin-sensitivity of lam2D yeast cells and reduces the efficiency with 8 which the purified protein is able to extract membrane-bound sterol and catalyze sterol 9 exchange between populations of vesicles in vitro. These data suggest an unprecedented, 10 water-controlled mechanism of sterol acquisition and discharge from a StARkin domain. 11 12

13
Lam4S2 associates with the membrane via its Ω1 loop and C-terminal helix 14 We used atomistic MD simulations of Lam4S2 to explore the impact of the unique lateral 15 opening in Lam/GramD1 StARkin domains on the stability of bound sterol and its ability to exit 16 the binding pocket. Cholesterol-bound Lam4S2 ( Figure 1A) was placed in the vicinity of a 17 membrane bilayer ( Figure 1B) having the composition of anionic "Acceptor" liposomes used for 18 in vitro sterol transport assays (11), and its spontaneous binding to the membrane surface was 19 monitored via ensemble MD simulations carried out in 10 statistically independent replicates 20 (Stage 1 ensemble simulations, 3.2 µs cumulative time). The simulations showed two modes by 21 which Lam4S2 associated with the membrane. In one mode, the protein interacted with lipid 22 headgroups via its N-and C-termini (green and red; Figure 1C, D). As these regions are linked to 23 adjacent parts of the protein chain in full-length Lam4, i.e. the S1 domain and the 24 transmembrane helix, respectively, this mode of association is likely a non-physiological artifact 25 of simulating the isolated Lam4S2 domain and was not pursued further. In the second mode, 26 Lam4S2 engaged with the membrane via its Ω1 loop (purple; Figure 1C, E) and its C-terminal 27 helix (yellow; Figure 1C, E). The former is a functionally important feature of all StARkin 28 domains (12,17,18), whereas the latter harbors cationic residues (e.g., K163, K167) which have 29 been implicated in the in vitro sterol transfer activity of Lam4S2 (11,17) and the protein StARD4 30 (18). In this binding mode, cholesterol is oriented orthogonally to the plane of membrane, with 31 its 3-b-OH group engaging Q121 in the binding pocket of Lam4S2 and its iso-octyl tail facing the 32 membrane ( Figure 1E). We hypothesized that in this position the protein is primed to release 33 sterol into the membrane and proceeded to test this premise by enhancing the sampling of this 34 mode of Lam4S2-membrane interaction. To this end, we initiated a new set of 100 independent 35 MD simulations (with random starting velocities) from 10 conformations of the system in which 36 the Ω1 loop and the C-terminal helix were simultaneously engaged with lipids (Stage 2 37 ensemble simulations, 37.5 µs cumulative time). As described next, these trajectories revealed 38 detailed mechanistic steps leading to spontaneous release of the protein-bound cholesterol 39 into the membrane. To facilitate analyses of the conformational dynamics of the membrane-bound Lam4S2-1 cholesterol complex in Stage 2 simulations, we used the time-structure based independent 2 component analysis (tICA) approach to reduce the dimensionality of the system (see Methods). 3 To this end, we considered a set of collective variables (CVs) to describe the dynamics of 4 cholesterol and relevant segments of the protein (i.e. the Ω1 loop and the C-terminal helix) as 5 well as to quantify solvent exposure of the sterol binding site (see Methods for details). All the 6 trajectory frames from Stage 2 simulations were projected on the first two tICA vectors, which 7 represented ~90% of the total dynamics of the system (Figure 2-figure supplement 2A). The 8 resulting 2D space (Figure 2A), was discretized for structural analyses into 100 microstates 9 using the automated k-means clustering algorithm (Figure 2-figure supplement 2B). These 10 microstates cover the conformational space of the system as the cholesterol molecule is 11 transferred from the protein-bound state to the membrane. 12 Structural analyses of selected microstates on the tICA landscape (labeled 1-7 in Figure 2A), 13 characterized by relevant CVs ( Figure 2B) and visualized in structural snapshots ( Figure 2C) 14 describe key mechanistic steps of the sterol release process. Microstate 1 represents an 15 ensemble of states in which the sterol binding cavity is occluded from both the solvent and the 16 membrane. Thus, in Microstate 1 conformations ( Figure 2B, C), cholesterol is stably bound in 17 the protein ("chol RMSD" histogram (bottom panel, Figure 2B)), while the sterol binding pocket 18 is dehydrated ("water count" histogram (bottom panel, Figure 2B)) and sealed from the side by 19 the side-chains of residues S181 and D61 that line the side-entrance to the pocket (d61-181 20 distance histogram (bottom panel, Figure 2B)). In addition, the Ω1 loop is positioned close to 21 the C-terminal helix so that the Cα atoms of residues I95 in the Ω1 loop and A169 in the C-22 terminal helix are within ~10Å of each other (d95-169 distance histogram (bottom panel, Figure  23 2B); see the middle structure in Figure 2C for the location of A169), therefore occluding the 24 sterol binding pocket from below, i.e. the vantage point of the membrane. Indeed, as shown in 25 The first step in the sterol release process involves widening of the side-entrance to the 29 sterol binding site enabled by gradual separation of the side-chains of residues D61 and S181. 30 This structural change on the tICA landscape can be followed in the evolution of the system 31 from Microstate 1 to Microstates 2 and 3 (see d61-181 distance histogram (Figure 2B)). 32 Concomitant with the widening of the side-entrance, the level of hydration (water count) of the 33 binding pocket progressively increases (Figure 2B, C). 34 Cholesterol remains stably bound throughout these initial events (cholesterol RMSD is 35 unchanged in Microstates 1-3). However, the rising level of hydration in the binding site results 36 in destabilization of the polar interactions between the 3-b-OH group of cholesterol and the 37 side-chain of residue Q121 as cholesterol initiates its translocation towards the membrane. 38 Indeed, as the system transitions from Microstate 3 to Microstate 4, the RMSD of the 39 cholesterol molecule increases ( Figure 2B). Correspondingly, the minimum distance between 40 the cholesterol oxygen and residue Q121 increases by ~4Å (compare dchol-121 for Microstates 1 41 and 4, Figure 2-figure supplement 3). Notably, as the cholesterol molecule assumes this new 42 position, the distance between the Ω1 loop and the C-terminal helix increases as seen in the 43 broadening of the d95-169 histogram (Figure 2B), indicating initial opening of the sterol binding 1 pocket towards the membrane. 2 Cholesterol egress then proceeds through Microstates 5-7 in which the sterol binding 3 pocket remains open and solvated, while the Ω1 loop continues to sample conformations that 4 position it relatively far from the C-terminal helix (Figure 2B, C). The cholesterol molecule 5 leaves the binding pocket with its tail "down", becoming gradually encapsulated by the 6 hydrophobic chains of neighboring lipids until it fully embeds into the lipid membrane 7 ("number of lipids" histogram (bottom panel, Figure 2-figure supplement 3B) and 8 corresponding structural snapshots (Figure 2-figure supplement 3D)). The process of 9 translocation is complete when the system reaches Microstate 7. The remaining part of the tICA 10 space (corresponding to lower tIC1 and tIC2 values, i.e. bottom left region of the 2D space in 11 Figure 2A), describes trajectory data in which Lam4S2 disengages from the membrane, after 12 the release of the sterol, and diffuses into the solvent. Of note, the high level of hydration of 13 the binding site in empty Lam4S2, i.e. after the sterol egress, is recapitulated in MD simulations 14 of the apo Lam4S2 system (initiated from the sterol-free Lam4S2 structure, PDBID 6BYD) run 15 under the same conditions as the Stage 1 simulations of the cholesterol-bound Lam4S2 16 described in Figure 1 (see Figure 2-figure supplement 4 and Methods for more details).

17
The sterol translocation process outlined above was sampled in its entirety in 5 out of 100 18 Stage 2 simulations (trajectories highlighted in red in The side-opening to the sterol binding pocket is a key structural element of the 27 release mechanism 28 The MD simulations indicate that widening of the side-opening to facilitate water penetration 29 into the binding site ( Figure 3) is a key step in the mechanism by which bound cholesterol 30 leaves the protein to enter the membrane. To investigate in more detail the interplay between 31 increased hydration of the sterol binding pocket, widening of the side-entrance to the binding 32 cavity, and stability of cholesterol within the pocket, we analyzed the dynamics of D61 and S181 33 and their interactions with other residues in the binding site during the simulations. We found 34 that D61 is engaged in electrostatic interactions with residue K89 located in the β2 strand 35 preceding the Ω1 loop. Thus, the side-chain of K89 faces the entrance to the binding pocket 36 where it interacts with the anionic side-chain of D61 (Figure 2-figure supplement 3C, Figure 3). 37 This interaction is maintained in the initial stages of the translocation process (Microstates 1-4), 38 but becomes unstable as the hydration of the sterol binding pocket reaches its highest levels 39 after cholesterol leaves the site (note sampling of a wide range of d61-89 distances for 40 Microstates 5-7 (Figure 2-figure supplement 3C)). These data suggest that D61, S181 and K89 41 together participate in stabilizing the closed conformation of the side-entrance to the binding 42 pocket. 43 Based on these results and considering the position of the K89 side-chain near the protein-1 solvent interface, we hypothesized that replacing the polar and relatively long side-chain of K89 2 with a smaller hydrophobic moiety would promote widening of the side entrance, leading to 3 destabilization of cholesterol in the binding pocket. Likewise, substituting D61 and S181 with 4 residues with smaller size hydrophobic side-chains should have a similar destabilizing effect on 5 bound cholesterol. 6 7 Substitution of residues D61, S181, and K89 by Ala promotes hydration of the 8 binding site and destabilizes bound cholesterol 9 To test these hypotheses, we computationally generated K89A, D61A, and S181A point-mutants 10 of Lam4S2, and probed their dynamics using atomistic MD simulations. Specifically, we 11 considered two snapshots taken at different time points (120 ns and 150 ns, respectively) from 12 one of the (350 ns-long) Stage 2 trajectories of the wild type protein system in which sterol 13 release was observed. For the wild-type protein at these time points the side-entrance to the 14 pocket is closed (Figure 2-figure supplement 6A), the cholesterol molecule is stably bound 15 (Figure 2-figure supplement 6B), and the level of hydration is relatively low (between 5-10 16 water molecules (Figure 2-figure supplement 6B)). We introduced the three mutations 17 separately into these two snapshots, and -for each construct -carried out 150 ns long unbiased 18 MD simulations in 10 replicates (1.5 µs total simulation time). Analysis of these trajectories 19 revealed that for all the three mutants the hydration level of the sterol binding site increased 20 rapidly, during the initial 4-5 ns of the simulations (

K89A-Lam4S2 has a lower energy barrier for cholesterol release 30
To address the effect of the K89A mutation on cholesterol stability quantitatively, we compared 31 the energetics of sterol release in K89A versus the wild type system using umbrella sampling 32 MD simulations. We constructed the potential of mean force (PMF) for cholesterol release by 33 constraining the z-distance between the sterol hydroxyl oxygen and the Cα atom of residue 34 Q121, dZ(chol-121), to different values in the range ∈ [2Å; 20 Å] along the release pathway (dZ(chol-35 121) histogram (Figure 2-figure supplement 3B). The results are shown in Figure 4A. For the wild 36 type system, the PMF calculations indicate that cholesterol release requires overcoming an 37 energy barrier of ~6 kcal/mole, and proceeds through two major steps that were also identified 38 in our tICA analysis of Stage 2 simulations. Thus, the PMF has a global minimum at dZ(chol-121) ~2Å 39 corresponding to the position of cholesterol in the binding site where its polar head-group is 40 coordinated by residue Q121 (snapshot at the top right of Figure  The PMF calculations reveal that the energy barrier that separates LM-1 from the global 3 minimum is ~5 kcal/mole (red trace in Figure 4A). This high energy cost is associated with the 4 clear change in hydration of the sterol binding site and concomitant opening of the side-5 opening to the pocket (see WT profiles in Figure 4B). Indeed, the water count increases and the 6 D61-S181 interaction is destabilized presence of multiple minima on the PMF plot is consonant with our findings from the tICA 12 analysis of the unbiased MD simulations described above that in some of the Stage 2 13 trajectories the system evolved from Microstate 1 to Microstate 5, i.e. transitioned from the 14 global minimum to LM-1 on the PMF plot, but either did not progress further to complete sterol 15 egress, i.e. LM-2, or returned to the conformational space of the tICA landscape characterized 16 by relatively low hydration of the sterol binding pocket, i.e. the global energy minimum. 17 Remarkably, comparison of the PMF plots for the wild type and the K89A systems ( Figure  18 4A) reveals that the mutation significantly lowers the barriers required for transitioning 19 between the different energy minima. Thus, while the PMF profile for the K89A construct still 20 has three energy minima, the energy cost to transition between the global minimum and LM-1 21 in this system is ~2 kcal/mole, and between LM-2 and LM-3 is ~1 kcal/mole, resulting in an Interestingly, the global minimum on the PMF profile of the K89A mutant is shifted 33 compared to its location on the PMF plot of the wild type system from dZ(chol-121) ~ 2Å to ~ 5Å 34 ( Figure 4A). We found that at the shortest dZ(chol-121) distances, cholesterol-Q121 interactions in 35 the mutant are mostly mediated by water molecules, whereas at dZ(chol-121) ~ 5Å the hydroxyl 36 group of cholesterol is in direct contact with Q121 (see sharp peak at ~2Å for dZ(chol-121) = 5Å plot 37 in Figure 4-figure supplement 8B, note that dZ(chol-121) is the Z-distance between the hydroxyl 38 and the Cα of Q121). This may also explain why the water content in the cavity is skewed 39 towards lower values for dZ(chol-121) ~ 5Å (Figure 4B)). Thus, for both the wild type and K89A 40 systems, the global minimum on the PMF plot corresponds to the ensemble of states in which 41 cholesterol is engaged in direct interactions with Q121. Overall the PMF calculations reveal that 42 the K89A substitution lowers the energy barrier for cholesterol release from Lam4S2 into the 43 membrane and suggests that cholesterol is consequently less stable in the binding pocket. 44 1 Alanine substitution of residues at the side-entrance to the sterol binding 2 pocket impacts the function of Lam4S2 in cells and in vitro 3 Our computational studies indicate that substitution of D61, K89 or S181 with alanine affects 4 the degree of hydration of the sterol binding pocket and the stability of bound sterol, with the 5 most significant effects seen for the K89A mutant. We tested the functionality of K89A and the 6 other mutants using three types of experiments. 7 We previously showed that yeast cells lacking Ysp2/Lam2 (lam2D cells) are sensitive to the 8 polyene antibiotic amphotericin B, and that this phenotype can be corrected by expressing a 9 soluble GFP-Lam4S2 fusion protein (7). We verified that this was also the case for nystatin, 10 another polyene antibiotic ( Figure 5A, compare the first two rows in which lam2D cells are 11 transformed with either an empty vector (row 1) or a vector for expression of GFP-Lam4S2 (row 12 2), and plated on media without or with different amounts of nystatin). We then tested the 13 ability of GFP-fused Lam4S2 proteins carrying either K89A, D61A, or S181A single-point 14 mutations (GFP-Lam4S2(K89A), GFP-Lam4S2(D61A), GFP-Lam4S2(S181A), respectively) to 15 rescue the nystatin sensitivity of the lam2D cells. Figure 5A shows that lam2D cells expressing 16 GFP-Lam4S2(K89A) remained nystatin-sensitive, whereas those expressing GFP-Lam4S2(D61A) 17 or GFP-Lam4S2(S181A) became resistant to the antibiotic, similar to lam2D cells expressing 18 wild-type protein.
As all the Lam4S2 variants tested were expressed at equivalent levels 19 (revealed by SDS-PAGE immunoblotting (Figure 5B)), this cell-based assay indicates that the 20 K89A mutant has a functional deficit, whereas the D61A and S181A proteins are able to provide 21 cells with sufficient functionality to rescue their nystatin-sensitivity phenotype. 22 To test explicitly the ability of the mutants to extract sterol from membranes and catalyze 23 sterol exchange between populations of vesicles, we expressed His-tagged versions of the 24 proteins in E. coli and purified them by affinity chromatography and size exclusion. The D61A 25 mutant proved problematic on account of its low yield and apparent instability, and so we 26 focused on S181A and K89A ( Figure 6A). Similar to wild-type Lam4S2, these mutants displayed 27 monodisperse profiles on size exclusion ( Figure 6B) and yielded circular dichroism spectra 28 indicative of well-folded structures ( Figure 6C). 29 Sterol extraction assays were performed by incubating the purified proteins with large, 30 unilamellar vesicles containing [ 3 H]cholesterol, and determining the amount of radioactivity 31 and protein in the supernatant after ultracentrifugation to pellet the vesicles. Relative to the 32 wild-type protein, the S181A mutant extracted only ~50% of sterol under our standard 33 incubation conditions whereas the K89A mutant had essentially no ability to extract sterol 34 ( Figure 6D). 35 To probe sterol transfer activity of the Lam4S2 mutants, we performed in vitro sterol 36 transport assays as previously described and depicted schematically in Figure 7A. Donor 37 vesicles containing fluorescent dehydroergosterol (DHE) were incubated with acceptor vesicles 38 containing the FRET acceptor dansyl-PE. Excitation of DHE results in sensitized fluorescence 39 emission from dansyl-PE only when the two lipids are in the same vesicle. Figure 7B (see also 40 Figure 7D) shows that under our standard conditions the wild-type protein increases the rate of 41 DHE exchange ~7-fold over the spontaneous rate. The S181A mutant was similar to the wild-42 type protein, whereas the K89A mutant had essentially no activity (Figure 7C, D). 43 Overall, the three functional tests described above indicate that the K89A mutant is 1 compromised in sterol handling -it is unable to extract sterol from membranes and transfer it 2 between vesicles, accounting for its inability to rescue the nystatin sensitivity of lam2D cells. 3 These functional outcomes are in line with our computational prediction that cholesterol would 4 be unstable in the binding site of the Lam4S2 K89A. Interestingly, the partial inability of the 5 S181A mutant to extract cholesterol did not affect its ability to catalyze sterol exchange or 6 rescue the nystatin sensitivity of lam2D cells. 7 8 Discussion 9 Lam/GramD1 StARkin domains bind sterols specifically, admitting and exporting the sterol 10 molecule through an aperture at the end of their long axis as suggested by inspection of crystal 11 structures (11)(12)(13)(14) and also seen in the MD simulations reported here. Strikingly, the sterol 12 binding pocket in these proteins is fractured along part of its length, exposing bound sterol to 13 solvent. The analyses presented here describe a potentially general mechanism by which sterol 14 egress (or entry) from Lam/GramD1 StARkin domains is controlled by the concomitant entry (or 15 egress) of water molecules via this unusual lateral fracture. 16 Lam4S2 engages membranes via its Ω1 loop and C-terminal helix, two structural regions 17 identified previously as being functionally important in StART domains (11,12,17,18). Once 18 membrane-bound, the protein adopts diverse conformations characterized by different extents 19 of widening of the side-opening to the sterol binding pocket. The side-opening of the binding 20 pocket in sterol-loaded Lam4S2 can be sealed by the polar side-chains of residues S181, D61, 21 and K89, resulting in a low level of hydration within the cavity. In this condition, the cholesterol 22 molecule is stably bound, with its hydroxyl group in hydrogen-bonding interactions with residue 23 Q121. Cholesterol egress is triggered stochastically, by gradual widening of the side-opening 24 and concomitant penetration of water into the binding site. These dynamic events destabilize 25 cholesterol in the binding site by ~4-5 kcal/mole, driving it from the binding site towards the 26 membrane. The subsequent steps of the release process are enabled by repositioning of the Ω1 27 loop away from the C-terminal helix. This fully exposes the binding pocket to the membrane, 28 i.e. widens the axial aperture, thus creating a continuous passageway to the membrane. The 29 sequence of events by which sterol exits Lam4S2 and enters the membrane is shown in Figure  30 2-movie supplement 1. 31 The overall process of cholesterol release requires overcoming an energy barrier of ~6 32 kcal/mole energy barrier. This value is in a good agreement with the ~5 kcal/mole estimate for 33 energy barrier for sterol extraction (19) based on the assumption that intermembrane sterol 34 transfer is rate limited by sterol pick-up/delivery processes and that the rate constant for this 35 process can be described by simple Arrhenius relationship. Overall, the computational findings 36 reported here reveal that the conformational state of the side-opening to the sterol binding 37 cavity in Lam4S2 StARkin domain plays a major role in regulating the energetic stability of the 38 sterol in the pocket. 39 This prediction was probed first computationally by analyzing MD trajectories of Lam4S2 in 40 which residues that line the side-entrance to the binding site were substituted with alanine. For 41 all three mutations (S181A, D61A and K89A) we found destabilization of cholesterol in the 42 binding site. Using potential of mean force calculations, we found that the K89A mutation 43 lowered the energy barrier for cholesterol release by ~2-fold compared with wild type Lam4S2. 1 Experimental tests confirmed that the K89A mutant was non-functional, whereas the S181A 2 mutant was only partially compromised in its ability to bind sterol, a defect that did not appear 3 to influence its ability to rescue the nystatin-sensitivity of lam2D cells or exchange sterols 4 between membranes in vitro. Our test of the D61A mutant was limited to a cell-based assay 5 where it performed as well as wild-type protein in rescuing the nystatin-sensitivity of lam2D 6 cells. 7 Considering the functional importance of K89, and to a lesser extent S181, we examined the 8 conservation of these residues in the Lam/GramD1 family using a previously reported structure-9 based sequence alignment (12). We found that the positions aligning with K89 and S181 were 10 among the residues with the highest conservation score. Interestingly, it was noted that the 11 side-chain of residue K910 in the S1 domain of Lam2 (Lam2S1), which aligns with K89 of Lam4S2 12 (note that K89 in Lam4S2 corresponds to K1031 in the full-length protein ( Table 1)), is 13 positioned slightly differently in the ergosterol-bound and apo structures (12). This led to a 14 speculation that a path for ergosterol movement into and out of Lam2S1 could be enabled by 15 movement of K910. Consistent with this, our study reveals that residue K89 in Lam4S2 indeed 16 repositions when cholesterol is released from the protein. Importantly, we find that this 17 movement is a part of larger-scale dynamic changes involving neighboring polar residues, D61 18 and S181, that lead to widening of the side-opening to the binding pocket.

19
Our computational analysis points to the key role that solvation of the sterol binding pocket 20 plays in the process of cholesterol release. We find that water penetration destabilizes 21 hydrogen-bonding interactions between the 3-b-OH of cholesterol and the side-chain of Q121, 22 leading to initiation of sterol egress. While the current computations have not directly 23 addressed the mechanism of sterol entry into the binding site, the PMF profile that we report 24 here suggests that the continuous water pathway connecting the binding site to the bulk 25 solution, as observed in our simulations of the Lam4S2 under sterol-free conditions, should play 26 an important role in the delivery of sterol into the binding site. In this respect, we note that 27 while in some X-ray structures of Lam/GramD1 StARkin domains the polar head-group of the 28 bound sterol is seen in direct contact with neighboring polar residues, in the others it is 29 engaged with the protein indirectly, through water-mediated interactions. The former mode is 30 observed in Lam4S2, Lam2S2 and GramD1a, while the latter mode is seen in Lam2S1.

31
Interestingly, in both Lam4S2 and Lam2S2 the head-group of the bound sterol hydrogen-bonds 32 to the side-chain of a Gln residue (Q121 in Lam4S2). In Lam2S1, on the other hand, the position 33 aligning with Q121 is occupied by small-size polar residue, Ser. Therefore, the head-group of 34 the sterol does not form a direct hydrogen bond within the binding pocket of Lam2S1 but 35 rather associates with the protein through water-mediated interactions. In GramD1a, in which 36 the residue analogous to Q121 of Lam4S2 is also Ser, the bound sterol is seen in direct contact 37 with another adjacent polar residue (Tyr). Taken together, the structural information highlights 38 the importance of polar interactions for the stability of the sterol molecule in the binding site, 39 consistent with our results demonstrating that disruption of these interactions by influx of 40 water through the cavity side-opening leads to cholesterol release. Therefore, the molecular 41 mechanism of sterol release that we have identified in Lam4S2 is likely to be generalizable to 42 the other StARkin domains.

2
Computational methods 3 Molecular constructs of wild type Lam4S2 4 The computations were based on the X-ray structures of the second StARkin domain of Lam4, 5 Lam4S2 (PDBIDs 6BYM and 6BYD) (11). In the 6BYM structure, Lam4S2 (residue sequence 4-196 6 in the numbering used in Ref. (11), i.e. residue 4 corresponds to Thr-946 in native Lam4) is in 7 complex with 25-hydroxycholesterol, which is bound in the canonical sterol binding pocket 8 identified also in the StARkin domains of other Lam proteins (12,13). In the 6BYD model Lam4S2 9 (residue sequence 4-200) is in the apo form. For the computational studies described here, the 10 oxysterol in the 6BYM structure was replaced by cholesterol and the molecular models of 11 Lam4S2 in both 6BYM and 6BYD structures were completed using modeller 9v1 (20) to add 12 respective missing residue stretches, i.e. 1-3 and 197-203 to the 6BYM structure, and 1-3 and 13 201-203 to the 6BYD structure. 14 15 Unbiased MD simulations of sterol-bound wild type Lam4S2 16 An all-atom model lipid membrane with the composition of "Acceptor" liposomes in sterol 17 transport assays (11), was prepared using the CHARMM-GUI web server (21). Thus, symmetric 18 lipid bilayer containing 70% DOPC, 15% PI, 10% DOPE, and 5% DOPS (400 lipids in total on the 19 two leaflets) was assembled, solvated (using water/lipid number ratio of 50) and ionized with 20 0.1M K + Clsalt. This system was subjected to MD simulations for 30 ns using NAMD version 2.12 21 (22) and the standard multi-step equilibration protocol provided by CHARMM-GUI.

22
After this equilibration phase, the bilayer system was stripped of all water molecules and 23 solution ions and the cholesterol-bound Lam4S2 domain (6BYM) was placed near the 24 membrane surface so that the distance between any atom of the protein and any atom of the 25 lipid molecules was ≥ 10Å (see Figure 1B). The protein-membrane complex was solvated (using 26 water/lipid number ratio of ~145) and ionized (with 0.1M K + Clsalt). The resulting system 27 contained ~234,000 atoms in total.

28
The Lam4S2-membrane complex was equilibrated using a multi-step protocol (23) during 29 which the backbone of the protein was first harmonically constrained and subsequently 30 gradually released in three steps of 5 ns each, changing the restraining force constants from 1, 31 to 0.5, and 0.1 kcal/ (mol Å 2 ), respectively. This step was followed by 6 ns long unbiased MD 32 simulations carried out using the NAMD 2.12 package. After this short run, the velocities of all 33 the atoms were reset and the system was simulated with ACEMD software (24) in 10 34 statistically independent replicates (Stage 1 ensemble simulations), each for 320 ns, resulting in 35 a cumulative time of 3.2 µs for Stage 1 runs. 36 As described in Results, Stage 1 simulations sampled events of spontaneous binding of 37 Lam4S2 to the membrane. We randomly selected 10 frames from Stage 1 trajectories in which 38 Lam4S2 was seen to be interacting with the lipid bilayer as in Figure 1E, and initiated a new set 39 of simulations with ACEMD (Stage 2 ensemble simulations) in which the 10 chosen structures 40 were run in 10 statistically independent replicates each (i.e. 100 independent simulations). Each 41 of the 100 copies were simulated for 375 ns resulting in a cumulative time of 37.5 µs for Stage 2 42 runs.

44
Unbiased MD simulations of apo wild type Lam4S2 1 Simulations of the apo wild type Lam4S2 protein (6BYD) followed the same protocol as 2 described above for Stage 1 simulations of sterol-bound Lam4S2 with the only difference being 3 the lipid membrane composition. Thus, in the manner identical to the sterol-bound Lam4S2, 4 the apo protein was placed near the surface of the all-atom model lipid membrane (assembled 5 with CHARMM-GUI) with the composition of "Donor" liposomes in sterol transport assays (11). 6 This symmetric bilayer contained 31% DOPC, 23% DOPE, 23% DOPS, and 23% cholesterol (400 7 lipids in total on the two leaflets). As the purpose of these simulations was to quantify solvation 8 of the empty sterol binding site, this system was only considered for Stage 1 simulations 9 (cumulative time of 3.2 µs) and was not subjected to subsequent (Stage 2) phase. 10 11 Unbiased MD simulations of the mutant Lam4S2 systems 12 Using the FoldX server (25), three single mutations, K89A, S181A, and D79A in Lam4S2 were 13 introduced into two separate frames of one of the Stage 2 ensemble trajectories of the wild 14 type protein system (see Results). The resulting structures (two per mutant) were energy-15 minimized for 100 steps and then simulated in five independent replicates each for 150 ns using 16 ACEMD. This resulted in 10 statistically independent MD trajectories per mutant totaling 1.5 µs.

18
Parameters and force-field for MD simulations 19 All the simulations performed with NAMD 2.12 implemented all option for rigidbonds, 2fs 20 integration time-step, PME for electrostatics interactions (26), and were carried out in NPT 21 ensemble under semi-isotropic pressure coupling conditions, at a temperature of 310 K. The 22 Nose-Hoover Langevin piston algorithm (22) was used to control the target P = 1 atm pressure 23 with the LangevinPistonPeriod set to 100 fs and LangevinPistonDecay set to 50 fs. The van der 24 Waals interactions were calculated applying a cutoff distance of 12 Å and switching the 25 potential from 10 Å. In addition, vdwforceswitching option was set to on.

26
The simulations carried out with ACEMD software implemented the PME method for 27 electrostatic calculations, and were carried out according to the protocol developed at Acellera 28 and implemented by us previously (24,27) with 4 fs integration time-step and the standard 29 mass repartitioning procedure for hydrogen atoms. The computations were conducted under 30 the NVT ensemble (at T=310 K), using the Langevin Thermostat with Langevin Damping Factor 31 set to 0.1.

32
For all the simulations the CHARMM36 force field parameters for proteins, lipids, sterols, 33 and ions (28,29) were used.

35
Umbrella sampling MD simulations of wild type and K89A Lam4S2 36 Biased MD simulations of cholesterol release from the wild type and the K89A mutant Lam4S2 37 were performed using umbrella sampling approach. The position of the translocated 38 cholesterol was restrained to different locations along the translocation pathway (see Results) 39 using as a collective variable the z-directional distance, dZ(chol-121) (along the axis perpendicular 40 to the membrane plane), between the cholesterol oxygen and the Cα atom of residue Q121 (see 41 Figure 1A). 19 windows spaced 1Å apart in the range of dZ(chol-121) ∈ [2Å; 20Å] were considered 42 and the dynamics of the sterol molecule in each window was restrained by applying a force 43 constant of 2.5 kcal/mol · Å 2 . The rest of the parameters for the umbrella sampling runs were as 44 follows: width -2Å, and both lowerwallconstant and upperwallconstant set to 25 kcal/mol · Å 2 . 1 Each umbrella window was simulated for 50 ns which resulted in good overlap between 2 adjacent windows (Figure 4-figure supplement 9A, B). 3 The potential of mean force (PMF) along the collective variable was constructed with 4 Weighted Histogram Analysis Method (WHAM) Version 2.0.9 (30). For the WHAM calculations 5 only the last 25 ns trajectory segments of each umbrella window were used. The tolerance 6 parameter was set to 0.0001. To estimate error bars on the PMF, for each umbrella window 7 first decorrelation time was calculated as a time-constant from a single exponential fit to the 8 auto-correlation vs time data (Figure 4-figure supplement 9C, D). The error bars were then 9 constructed with Monte Carlo bootstrapping error analysis in the WHAM software on the 10 decorrelated data points using num_MC_trials of 1000. 11 12 Dimensionality reduction using the time-structure based independent component analysis 13 (tICA) 14 To facilitate analysis of cholesterol release process from Lam4S2 domain in the MD simulations, 15 we performed dimensionality reduction using the tICA approach (31) , (32,33) , (34) as described 16 previously (35)(36)(37). To define the tICA space we used several dynamic variables extracted from 17 the analysis of the ensemble MD trajectories that quantify the dynamics of the cholesterol, the 18 extent of exposure of the sterol binding site to the solvent, and the dynamics of the functionally 19 important Ω1 loop. These variables include (see Figure 1A): (1)-the minimum distance between 20 the hydroxyl oxygen atom of the translocated cholesterol and residue Q121 (dchol-121); (2)-the 21 root-mean-square deviation (RMSD) of the cholesterol molecule from its position in the binding 22 site; (3)-distance between the hydroxyl oxygen of S181 and Cγ carbon of D61 (d61-181); (4)-Cα -23 Cα distance between residue I95 in the Ω1 loop and residue A169 in the C-terminal helix (d95-169); 24 (5)-number of water molecules in the interior of the protein (defined as number of water 25 oxygens found within 5Å of the side-chains of the following protein residues -189, 185, 181, 26 154, 152, 136, 138, 140, 142, 123, 121, 119, 117, 102, 104, 106, 108, but farther than 5Å from 27 the following residues -116, 118, 109, 86, 103, 105); (6)-the number of lipid phosphate atoms 28 with 3Å of the translocated cholesterol molecule. 29 Using these six CV-s as components of the data vector X, the slowest reaction coordinates 30 of a system were found as described previously (35,37,38), by constructing a time-lagged 31 covariance matrix (TLCM): respectively. The eigenvectors corresponding to the largest eigenvalues define the slowest 36 reaction coordinates.

38
Experimental methods 39 40 Lam4S2 mutants 41 Point mutants of Lam4S2 (D61A, K89A and S181A) were generated by PCR mutagenesis and 42 confirmed by sequencing. The constructs and PCR primers are detailed in Table 1 and Table 2. 43 44

Protein expression and purification 1
Lam4S2 and point mutants were expressed in E. coli as His-tagged proteins (Table 1), and 2 purified by affinity chromatography on Ni-NTA resin, followed by size exclusion 3 chromatography (SEC) using a Superdex 200 Increase 15/300GL column. The purification 4 procedure was as previously described (11), except that the proteolysis step to remove the 5 affinity tag was omitted and SEC was carried out in 20 mM HEPES, pH 7.5, 150 mM NaCl. The 6 purified protein was snap frozen in small aliquots and stored at -80°C. Prior to use, aliquots 7 were thawed and subjected to brief microcentrifugation to remove any aggregated material. 8 Purified proteins were quantified by absorbance at 280 nm; quality control included analysis by 9 circular dichroism (CD) as described (11), and re-analysis by SEC using buffer conditions as 10 above.

12
Sterol transport assay 13 The assay (illustrated in Figure 7A) was performed and analyzed as previously described (11, concentration, based on measurement of inorganic phosphate after acid hydrolysis of the 20 vesicles) and 0.05 µM or 0.1 µM Lam4S2 (final concentration) in 20 mM PIPES (pH 6.8), 3 mM 21 KCl and 10 mM NaCl (assay buffer); fluorescence was monitored for ~2500 s using λex = 310 nm 22 and λem = 525 nm and a data acquisition frequency of 1 Hz. Acceptor liposomes were added to 23 donor liposomes in the cuvette and after 60 sec, 200 µl of Lam4S2 (or Lam4S2-mutant), diluted 24 as needed in assay buffer, was added. For control assays, 200 µl of assay buffer was added. All 25 traces were offset corrected such that the fluorescence signal and time at the point of Lam4S2 26 (or buffer) addition were each set to zero. The maximum possible FRET signal was determined 27 from assays using 0.1 µM wild-type Lam4S2 where the fluorescence readout reached a plateau 28 value within 2000 s; traces from such assays (done in replicate) were fit to a mono-exponential 29 function, and the plateau value obtained (FRETmax) was used to constrain the mono-exponential 30 fits of all other traces. Traces from different assays were compiled after data fitting by setting 31 FRETmax = 1.

mol %, containing a trace amount of [ 3 H]cholesterol and 36
N-rhodamine-DHPE) were dried in a glass screw-cap tube under a stream of nitrogen, then 37 resuspended in 1 ml assay buffer (20 mM PIPES (pH 6.8), 3 mM KCl, 10 mM NaCl) supplemented 38 with 250 mM sucrose, by agitating on a Vibrax orbital shaker for 30 min at 1200 rpm. The 39 resulting suspension was subjected to 5 cycles of freeze-thaw (immersion in liquid nitrogen, 40 followed by thawing at room temperature), before being extruded 11 times through a 200-nm 41 membrane filter using the Avanti Mini-Extruder. After extrusion, extravesicular sucrose was 42 removed by diluting the vesicles 4x in assay buffer and centrifuging in a Beckman TLA100.3 43 rotor (75,000 rpm, 1 h, 4°C). The supernatant was carefully removed from the pelleted vesicles 44 (easily discernable because of their pink color due to rhodamine-DHPE), before resuspending 1 the vesicles in 1 ml of assay buffer. Aliquots of the sample (5 µl) were removed at different 2 points of preparation (after the freeze-thaw step, post-extrusion and after final resuspension) 3 and taken for liquid scintillation counting to track lipid recovery by monitoring [ 3 H]cholesterol. 4 The ability of Lam4S2 (wild-type and point mutants) to extract cholesterol from the vesicles 5 was determined as follows. Liposomes (15 µl, ~1500 pmol cholesterol) and protein (500-750 6 pmol, as indicated) were combined in assay buffer to a total volume of 500 µl. The mixture was 7 incubated at room temperature for 1 h, before removing a 20 µl aliquot for liquid scintillation 8 counting. The remainder of the sample was centrifuged in a Beckman TLA100.2 rotor (75,000 9 rpm, 1 h, 4°C). Most of the supernatant (350 µl) was transferred to a fresh 1.5 ml tube, while 10 the remainder was removed immediately and discarded. The pellet was resuspended in 100 µl 11 assay buffer containing 5% (w/v) SDS. Duplicate aliquots (50 µl each) of the supernatant were 12 taken for liquid scintillation counting to determine the amount of extracted cholesterol. Protein 13 in the remainder of the supernatant (250 µl) and the resuspended pellet was precipitated by 14 adding 1.2 ml ice-cold acetone, followed by overnight incubation at -20°C. The precipitated 15 proteins were pelleted by centrifugation, air-dried after removal of the acetone, and dissolved 16 in SDS gel loading buffer. The relative amount of protein in the supernatant and pellet fractions 17 was determined by SDS-PAGE, Coomassie staining and quantification of band intensity using 18 Image J software.

36
The funders had no role in study design, data collection and interpretation, or the decision to 37 submit the work for publication.

39
Author contributions 40 George Khelashvili, Conceptualization, Formal analysis, Investigation, Methodology, Writing-1 original draft, Writing-review and editing; Kalpana Pandey, Neha Chauhan, Formal analysis, 2 Investigation, Methodology, Writing-review and editing; David Eliezer, Conceptualization, 3 Writing-review and editing; Anant K Menon, Conceptualization, Formal analysis, 4 Methodology, Supervision, Project administration, Writing-original draft, Writing-review and 5 editing 6 7 Author ORCIDs 8 Neha Chauhan https://orcid.org/0000-0003-1497-3359 9 David Eliezer https://orcid.org/0000-0002-1311-7537 10 George Khelashvili https://orcid.org/0000-0001-7235-8579 11 Anant Kumar Menon https://orcid.org/0000-0001-6924-2698 12 Kalpana Pandey https://orcid.org/0000-0002-7048-0690 13 involved in the uptake and intracellular transport of sterols in Saccharomyces cerevisiae.    (2018)); residue numbering based on the entire Lam4 sequence is provided in the right-hand column 4 two amino acids (DV) appended to the end of the Lam4S2 sequence 1 2  collective variables (CVs) for tICA analysis. Shown in licorice and labeled are residues Q121, 7 S181, D61. Also highlighted are locations of the Ω1 loop (purple), the C-terminal helix (yellow), 8 the N-terminus (green), and the C-terminus (red). Cholesterol is shown in space filling 9 representation colored cyan except for the oxygen atom in red (B) Initial positioning of Lam4S2 10 (cartoon) near the membrane (lines). In this configuration, the distance between any atom of 11 the protein and any atom of a lipid molecule was ≥10Å. The cholesterol molecule bound to 12 Lam4S2 is shown in space-fill representation. The water box including solution ions are omitted Structural models representing selected microstates. In these snapshots, Lam4S2 is shown in 12 cartoon, and cholesterol as well as selected protein residues (Q121, D61, S181, I95, A169) are 13 shown in space fill (the residues are labeled in the snapshot of Microstate 4). Water oxygens in 14 the sterol binding site are drawn as gold spheres. protein models are representative structures from Microstates 1 and 7, respectively). In both 6 snapshots, residues D61, K89, and S181 lining the side-opening are highlighted (in space-fill, 7 colored and labeled). The gold spheres in panels A and B represent superposition of water 8 oxygens in the binding site and near the side-entrance from one of the Stage 2 trajectories 9 before (panel A) and after (panel B) the side-entrance opens. As a reference, the approximate 10 location of the cholesterol hydroxyl group is indicated (red oval, marked OH). A water pathway 11 to the binding pocket, formed under conditions of the open, but not closed, side-entrance is 12 illustrated in panel B by the yellow arrow. Cells (lam2D) were transformed with an empty vector (top row) or with a vector for expression 4 of GFP-Lam4S2 wild-type (WT) or mutants as indicated. Serial 10-fold dilutions were spotted 5 onto agar plates containing defined minimal media (-uracil) lacking (-) or containing 2 µg/ml (+) 6 or 8 µg/ml (++) nystatin. The plates were photographed after 72 h at room temperature. (B) 7 Equivalent amounts of cytosol from lam2D cells expressing GFP-Lam4S2 wild-type or mutants 8 were analyzed by SDS-PAGE and immunoblotting using anti-GFP antibodies to detect the fusion 9 proteins and anti-GAPDH to verify equivalent loading. [ 3 H]cholesterol) were incubated with 750 pmol of purified proteins for 1 h at room 10 temperature. After ultracentrifugation, the radioactivity and the protein amount in the 11 supernatant was determined, and the stoichiometry of binding was calculated. Data are 12 represented as mean ± SEM (error bars; n = 5-7). Data are normalized to the value obtained for 13 the wild-type protein (0.11 ± 0.02 pmol cholesterol/pmol protein (mean ± standard deviation 14 (n=6)). (B) Spontaneous sterol exchange between vesicles and transport catalyzed by wild-type Lam4S2 4 (0.05 µM); Traces (n=7-8) were acquired from three independent experiments and averaged. 5 The blue and dashed lines represent monoexponential fits of the averaged data; the grey bars 6 graphed behind the fits represent the standard error of the mean (s.e.m.). (C) As in panel B, 7 except that Lam4S2 mutants were tested (n=5-7). The data fits for traces corresponding to 8 spontaneous transport and transport catalyzed by the wild-type protein are taken from panel B 9 and shown for comparison. (D) Rate constants (colored symbols) obtained from mono-10 exponential fits of individual traces from the experiments depicted in panels B and C. The bars 11 show the mean and s.e.m. of the data. 12 13 (Lam4S2)) are oriented with their sterol entry/exit site at the bottom, and displaying the lateral 7 fracture (side-opening) through which the bound sterol is visible from the bulk solvent. trajectories mapped on the 2D landscape of the first two tICA eigenvectors (tIC 1 and tIC 2). 6

Figure Supplements
Shown also are locations of the 100 microstates (red squares) obtained from the k-means 7 clustering analysis of the conformational space. The lighter shades on the 2D space indicate the 8 most populated regions (see also  2 MD trajectories mapped with the tICA transformation in the space of the first two tIC vectors. 5 The lighter shades (from red to light green to yellow) indicate the most populated regions of 6 the 2D space (see the color bar). Microstates (see Methods) representing the most populated 7 states in these simulations are indicated by the numbered circles (1-7) and represent various 8 stages in the lipid translocation process. (B) Structural characteristics of selected microstates. 9 The columns from left to right record the probability distributions of dchol-121 distance, of dZ(chol- Lam4S2 (based on PDBID 6BYD). The high degree of hydration of the binding pocket seen for 5 the apo protein recapitulates the level of solvation observed in the MD simulations of the 6 cholesterol-bound Lam4S2 after sterol exit (see Microstates 5-7 in Figure 2B). movie is based on one of the Stage 2 simulations of wild type Lam4S2. The total length of the 4 trajectory is 350 ns. In the movie, Lam4S2 is shown in white cartoon, the cholesterol molecule is 5

Movie Supplements
represented in ice-blue colored space-fill, S181, D61, and K89 residues are drawn in space-fill, 6 oxygen atoms of water molecules in the sterol binding site are depicted as pink spheres, the 7 membrane leaflet to which Lam4S2 is bound is represented by the nearby lipid phosphate 8 atoms (golden spheres), and lipid molecules within 3Å of the cholesterol are shown in licorice 9 representation. The rest of the simulation box is omitted. For clarity, the trajectory frames are 10 smoothed for the movie. 11 12