Membrane Lipids Are Both the Substrates and a Mechanistically Responsive Environment of TMEM16 Scramblase Proteins

Abstract

Recent discoveries about functional mechanisms of proteins in the TMEM16 family of phospholipid scramblases have illuminated the dual role of the membrane as both the substrate and a mechanistically responsive environment in the wide range of physiological processes and genetic disorders in which they are implicated. This is highlighted in the review of recent findings from our collaborative investigations of molecular mechanisms of TMEM16 scramblases that emerged from iterative functional, structural, and computational experimentation. In the context of this review, we present new MD simulations and trajectory analyses motivated by the fact that new structural information about the TMEM16 scramblases is emerging from cryo-EM determinations in lipid nanodiscs. Because the functional environment of these proteins in in vivo and in in vitro is closer to flat membranes, we studied comparatively the responses of the membrane to the TMEM16 proteins in flat membranes and nanodiscs. We find that bilayer shapes in the nanodiscs are very different from those observed in the flat membrane systems, but the function-related slanting of the membrane observed at the nhTMEM16 boundary with the protein is similar in the nanodiscs and in the flat bilayers. This changes, however, in the bilayer composed of longer-tail lipids, which is thicker near the phospholipid translocation pathway, which may reflect an enhanced tendency of the long tails to penetrate the pathway and create, as shown previously, a nonconductive environment. These findings support the correspondence between the mechanistic involvement of the lipid environment in the flat membranes, and the nanodiscs.

Introduction

Many physiological cell states and processes depend on maintaining and modulating the asymmetric composition of cell membranes, which often requires rapid flipping of lipids between the two leaflets of the membrane. Proteins in the TMEM16 family of phospholipid scramblases (PLS) have emerged as important performers of this physiological function. Thus, the TMEM16 PLS, together with other TMEM16 family members that function mainly as Cl⁻ ion channels, have attracted attention for their role in a wide range of physiological processes in several different species,^[1–3] and the mammalian TMEM16 PLS such as TMEM16E, TMEM16F, and TMEM16K, have been implicated in genetic disorders of blood, muscle, bone, and brain.^[1,4–6] However, the first structural insights into TMEM16 family of proteins became available more recently, from the crystal structure of a Ca²⁺-bound fungal homolog from Nectria haematococca (nhTMEM16).^[7] The structure revealed a homodimer, each subunit of which is composed of 10 transmembrane helices (TMs) (Fig. 1a). In the solved structure of nhTMEM16, which can function as a phospholipid (PL) scramblase as well as a nonselective ion channel,^[8,9] the dimeric arrangement generates several cavity regions. One of these regions consists of a hydrophobic pore-like structure across the TM region (see “hydrophobic cavities” in Fig. 1a). The other cavity region consists of a groove in each protomer, facing the membrane on the opposite side of the dimer interface (see “PL pathway groove” in Fig. 1a). This membrane-facing groove is lined by 5 transmembrane (TM) helices, TM3-TM7, and connects (through TM6) to the pair of bound Ca²⁺ ions (yellow spheres in Fig. 1a). It spans the entire thickness of the membrane, and yet is strongly hydrophilic. This intriguing and highly uncommon structural feature for a transmembrane protein suggested that the groove region in nhTMEM16 could serve as a PL translocation pathway.^[7] Lipids could be envisioned to slide through the polar groove by populating it with their hydrophilic headgroups while keeping the hydrophobic tails in the bilayer environment. Together, several breakthrough structures of fungal TMEM16 PLS brought evidence for global rearrangements induced by Ca²⁺-binding that result in the opening of a membrane-exposed pathway through which lipid headgroups can move between leaflets.^[7,10–12] This mechanism of lipid scrambling is consistent with the proposed “credit card-reader” model^[13] and indeed has been observed in several molecular dynamics (MD) simulations of Ca²⁺-bound nhTMEM16 protein^[8,14–16] in which lipids flip between the two leaflets of the membrane by engaging with various sites along the groove with their head-groups. As described,^[14,15] at these sites the headgroup of the translocated lipid is coordinated by side-chains of polar/charged amino acids, such as E352/K353 at the intracellular (IC) part of the TM4 helix, the T333/Y439 residue pair from the TM4/TM6 at the central part of the groove, and at E313/R432 residues from TM3/TM6 from the extracellular (EC) segment of the groove. That the groove in TMEM16 proteins serves as a translocation pathway for not only lipids but also for ions, is further substantiated from detailed mutagenesis studies which reveal that mutations affecting lipid scrambling and/or ion transport are localized to the groove area.^{[6,8,17–21]}

Figure 1.

Landmark features of the nhTMEM16 scramblase structure (PDBID 4WIS^[7]). a) The two monomers of the dimer protein are shown in different colors (cyan and pink). Yellow spheres represent the bound Ca²⁺ ions (two per monomer). Solid lines demarcate the relative positions of upper and lower leaflets of the membrane (i.e., idealized positions of phosphate groups on the two leaflets). The hydrophobic cavities and the phospholipid (PL) pathway are indicated by arrows, and labeled. b,c) Views of the PL translocation pathway (groove) in nonconductive (b), and conductive (c) states from the atomistic MD simulations described in Ref. [8] (Table S1, WT^ensemble) Transmembrane helices (TMs) 3–7 forming the groove region are shown in cartoon. The polar residues forming the extracellular gates (E318, E313, R432, T333, and Y439) are depicted in van der Waals representation and are labeled. Silver spheres represent water oxygen molecules at positions they occupy in the groove regions of these two conformations (i.e., closed and open groove) during simulations. Note that a continuous water pathway connecting IC and EC compartments of the membrane is formed only under conditions of open groove (panel c), whereas the closed conformation of the groove presents a constriction on the level of residues T333 and Y439.

In a long-term collaborative effort, we are investigating the molecular details of the mechanisms of TMEM16 PLS with an integrated protocol of iterative functional, structural and computational experimentation in which hypotheses generated from synergistic analyses of computational experiments and of structure–function data are tested using in vitro assays on purified fungal TMEM16 scramblases. Here we review briefly some of the salient new discoveries of mechanistic elements in the function of the TMEM16 PLS, which emerged from the computational investigations in the context of this collaboration. We focus in particular on the new understanding of the functional implications of the dual role of the membrane as a substrate and environment for TMEM16 scramblase activity. In the study of these implications it is essential to remember that structural information about the TMEM16 PLS comes primarily from Cryo-EM determinations in nanodiscs, whereas their functional environment in vivo and in vitro is more akin to flat membranes. Therefore, we took advantage of the newly developed capabilities offered by the nanodisc CHARMM-GUI builder^[22] to construct structurally accurate models of the lipid nanodiscs used experimentally, in order to study the lipid environment and its implications for TMEM16 PLS functional mechanisms in a systematic comparison to flat membranes. The initial findings offer clarifying insight into the mechanistic origin and possible functional implications of differences related to the composition and remodeling of the membrane in the different environments.

In the following section (“Individual membrane lipids are both substrates and facilitators of scrambling by the TMEM16 proteins”), we review key findings from our combined computational and experimental studies (largely based on the results presented in Ref. [8]) that led to the discovery of the lipid mediated molecular mechanism at the EC entryway of the translocation pathway in nhTMEM16 scramblase enabling lipid flipping. This is followed by an overview of recent structural and functional data regarding the role of the membrane environment in the function of TMEM16 scramblases (“The lipid membrane is a responsive environment supporting the function of TMEM16 scramblases”). In this background context, we detail in the last section (“Computational assessment of membrane deformation patterns in nanodiscs and planar lipid membranes containing a TMEM16 scramblase”), results from new computational studies of nhTMEM16 in lipid nanodiscs and flat membranes of various lipid compositions. In this comparative study of bilayer deformations around the scramblase in these different lipid environments, we examine the correspondence between the insights from structure and computation regarding the mechanistic involvement of the lipid environment in the flat membranes, and in the nanodiscs. The preceding mechanistic review supports the discussion of these comparative findings in the context of some aspects of TMEM16 scramblase function.

Individual Membrane Lipids Are Both Substrates and Facilitators of Scrambling by the TMEM16 Proteins

It was surprising to find from extensive atomistic MD simulations of the Ca²⁺-bound nhTMEM16 scramblase dimer in POPC lipid membranes^[8] that the Ca²⁺-bound structure considered to be the open-groove state that serves in the experimentally observed scrambling^[7] does not form a continuous conductive pathway for lipids. Indeed, spontaneous penetration of the lipid head-groups into the intracellular end of the groove was observed in the MD simulations and found to be mechanistically regulated by the extent of the solvation of the groove area. Thus, at its IC end, the groove observed in the X-ray structure of nhTMEM16 is remarkably wide (Fig. 1b), measuring ~35 Å between the ends of the TMs 4 and 6 (distance between the side chains of residue pair A356/F463). However, on the EC side, TM3/TM4 helix pair is in much closer proximity to TM6, due to interhelical interactions formed by network of polar residues, including E313 (in TM3), T333 (in TM4), as well as R432 and Y439 (in TM6, see Fig. 1b). Consequently, the groove is constricted to ~5 Å on the level of residues T333/ Y439. This creates an uneven, discrete distribution of water molecules inside the groove: the intracellular vestibule and the entry to the extracellular vestibule are both well hydrated, whereas the region in between them remains devoid of water.^[8] In the absence of continuous hydration channel, lipid head-groups cannot traverse the entire length of the groove but rather populate only the water enriched zones inside the groove^[8] (Fig. 1b).

The MD simulations we described recently^[8] revealed an allosteric mechanism that opens the constricted regions of the groove for lipid scrambling. The sampling of the conformational space and of lipid translocation events was carried out with an iterative simulation protocol we had also used for the analysis of lipid scrambling by a G protein-coupled receptor.^[23] This protocol consists of multiple replicas of unbiased MD simulations run in sequential stages, so that initiation of one stage was informed by output from the preceding one as shown schematically in Figure S1A and (see also Table S1, WT^ensemble and details in Ref. [8]). The translocation of lipid from the IC to the EC side of the groove was monitored from the time-evolution of its headgroup’s z coordinate (along the membrane normal axis) toward the z-position of the C_α atom of residue R432 (see Fig. S1B and structural snapshots in Fig. S1C). Such translocations (scrambling) from the inner to the outer leaflet were observed in 7/24 simulations (these simulations are highlighted in green in Fig. S1A). The detailed rearrangements characteristic of this inter-leaflet scrambling process in the consecutive Stages (Fig. S1B) shows the lipid headgroup moving into the groove and reaching about midway through it (state marked by red “i” in Figs. S1B and S1C). In the Stage 2 ensemble of simulations (red arrow in Fig. S1A), the lipid moves further toward the extracellular location (see position “ii” in Fig. S1B and S1C). During ensemble simulations of Stage 3, the lipid is seen flipping from the perpendicular-tail pose, to one aligned with the outer leaflet, and is released into this EC leaflet (see “iii” and “iv” in Fig. S1B and S1C).

Analyzed as described in detail (Ref. [8] and also Ref. [24]), tICA dimensionality reduction^[25] (using established procedures^[23,26,27]) of data from the ensemble simulations, served to extract mechanistic information regarding the scrambling process. Briefly, to define the tICA space we used eight collective variables (CV) extracted from the analysis of the ensemble MD trajectories (Fig. S2A). The time-lagged covariance matrix and the covariance matrix were constructed using these CV-s as components of the data vector. The slowest reaction coordinates were then identified by solving the generalized eigenvalue problem.^[23,26,27] As shown in Figure S2B, from the resulting eight tIC vectors, the first two contained ~90% information about the system’s dynamics. Subsequently, we focused our analysis on these two vectors by projecting the data from MD simulations onto the 2D space of the tIC 1 and 2 vectors, and discretizing this 2D space into 50 microstates (Fig. S2C) using an automated clustering k-means algorithm. These microstates cover the configurational space of the scramblase system as lipid translocation occurs from the IC to EC leaflets of the membrane, and reveal the structural characteristics of the key mechanistic stages in the translocation process (see Ref. [8] for more details). In particular, we identified structural states of the nhTMEM16 scramblase that correspond to key intermediate conformations along the transition pathway (Fig. 2, States 1–6). The tICA analysis also identifies rare conformational events (denoted by “*” in the tICA space in Figs. 2 and S2C), during which not just the headgroup, but also lipid tails, occupy the groove (Fig. S3A). Under these conditions, the pathway becomes dehydrated, which prevents lipid scrambling. We showed recently (Ref. [24]) that these lipid tail insertion events, observed in Ref. [8] as well, can promote a spontaneous transition of the groove from a Ca²⁺-bound, relatively open conformation, to a Ca²⁺-bound occluded conformation of nhTMEM16. In this occluded conformation, the middle region of TM4 is repositioned closer to TM6 (Fig. S3B) and the state resembles the intermediate state described in the recent cryo-electron microscopy (cryo-EM) studies of the nhTMEM16 and TMEM16F scramblases in lipid nanodiscs.^[10,28] The mechanistic underpinnings and functional consequences of this intermediate state are described elsewhere (Ref. [24]).

Figure 2.

Molecular mechanism of lipid scrambling by the nhTMEM16 protein. The results of tICA analysis of the extensive atomistic ensemble MD simulations of the nhTMEM16 in POPC membrane (Table S1, WT^ensemble) detailed in Ref. [8] are presented in the central panel showing a mapping of all the MD trajectories as the two-dimensional landscape in the space of the first two tICA eigenvectors (tIC1 and tIC2; see text and Ref. [8] for details). The locations of microstates 1–6 that represent the translocation of the lipid through the nhTMEM16 groove are indicated on the landscape, and representative structural representation of the microstates are given in the snapshots surrounding the central panel. In these structural models, TMs 3–7 lining the nhTMEM16 groove are shown in different colors, the advancing lipid is rendered in licorice, and mechanistically relevant groove residues are shown in space fill representations and labeled. The microstate denoted by “*” captures a rare mode of protein–lipid interaction in which the hydrocarbon tail of a lipid inserts into the groove and interferes with the translocation of the advancing lipid. The plot adjacent to the tICA landscape compares the percentage of trajectory frames in which the three EC gates, T333-Y439, E313-R432, and E318-R432, are simultaneously broken in the microstates (the T333-Y439 gate was assumed to be broken if the T333-Y439 distance was >8 Å; the E313-R432 and E318-R432 gates were assumed to be broken if the distance between carbonyl oxygen of Glu and sidechain nitrogen of Arg was >4 Å).^[8] The lower panel identifies the corresponding sequence of mechanistic events leading to the lipid flip is summarized. Reproduced from Ref. [8] with permission from the authors.

The key mechanistic steps that define the transition of a lipid between the two leaflets are presented as “Sequence of Mechanistic Events” in Figure 2. The lipid flipping process is seen to involve (1)- the simultaneous destabilization of the interactions tethering TM3/TM4 to TM6 (between E313-R432, E318-R432 and T333-Y439), and (2)-the resulting widening of the EC vestibule to allow for the release of the flipped lipid into the external leaflet.^[8] The involvement of the polar network of residues in the gating mechanism is consistent with the findings from earlier MD studies in which these residues were identified as lipid interaction sites.^[14,15] Remarkably, key mechanistic steps in the translocation of a lipid from the EC to IC direction recapitulate (in reverse order) the steps determined from the tICA analysis in the transfer from the IC to the EC leaflet, in spite of the asymme try of the entry points and composition of the pathway.

The lipid flip from the EC to IC direction was observed in microsecond-scale MD simulations of Ca²⁺-bound nhTMEM16 on Anton2^[29] as described in detail in Ref. [8] (see also Table S1, the A385W construct). In these simulations, lipid-dependent gating occurs when the headgroups approach the EC end of the pathway. The gradual rearrangement of the EC network of interactions in the triad of charged residues (E313, E318, and R432) eliminates the steric barriers delimiting the lipid- and water-depleted zone, thus overcoming the dehydration and enabling lipid transfer between leaflets.^[8] Specifically, engagement of a lipid molecule with the E313-R432 pair (Fig. 3a; see lipid in purple) results in destabilization of this ionic interaction as the R432 sidechain switches from E313 to E318, one helical turn above on TM3 (Fig. 3b). This switch releases the volume previously occupied by the R432 side chain to allow another lipid head-group to partition into this space (Fig. 3b; see lipid in green). Collectively, the head-groups of the two lipids further destabilize the EC network of interactions as R432 disengages from both E313 and E318. This allows TM3 and TM6 to move apart, as several polar interactions between them are disrupted (shown in Fig. 3c, and corresponding to time-point “C” in Figure 3e that shows the time-traces for the network distances E313-R432, E318-R432, and E313-Q436). The central constriction is eliminated (see Figs. 1b and 1c), as the Y439 side chain rotates away from T333 and toward the membrane (see the time-traces for T333-Y439 in Fig. 3e). These dynamic rearrangements generate an opening into which a third lipid partitions from the external leaflet (Fig. 3c; see lipid in blue) and proceeds to diffuse through the opened groove to the IC leaflet (Fig. 3d). The side chain of Y439 is seen to play a key role in the permeation, as it coordinates the headgroup of the flipped lipid throughout the process (Fig. 3f).

Figure 3.

The sequence of dynamic rearrangements of the groove in the phospholipid flipping process. The results are from 2 μs of MD simulations of the A385W mutant nhTMEM16^[8] (Table S1, A385W). a–d) Positioning of residues E313, E318, R432, and Y439 and of lipids neighboring the E313/R432 pair at various stages of the simulations. In (a) the E313-R432 gate is closed and a single lipid (in purple) is observed near E313-R432 pair. In (b) R432 switches interaction partner from E313 to E318. This conformational change allows the second lipid (in green) to enter the area. In (c) R432 breaks away from both E313 and E318 allowing lipid penetration deep into the groove (blue lipid); (d) shows the completion of lipid (in blue) translocation as its headgroup is on the level of the lipid headgroups in the intracellular leaflet. In panels e) and f) which show the time evolution of key distances in the mechanism, the arrows indicate the time points at which the snapshots in a–d were taken from the MD trajectory. e) Time evolution of the distances between E313-R432 (red), E313-Q436 (green), T333-Y439 (blue), and E318-R432 (black). f) Time evolution of the distance between the headgroup of the flipped lipid and the Y439 side chain. Reproduced from Ref. [8] with permission from the authors.

The key findings from these computational studies are (1)-the identification of the structural basis for the two constrictions along the lipid groove (i.e., E313/E318/R432 and T333/ Y439), and (2)-the modes of interaction of individual lipid molecules with residues at the EC end of the groove that trigger its rearrangement, allowing a (third) translocating lipid to move through the pathway toward the IC. These findings led to the mechanistic model of groove opening for lipid translocation triggered by modulation of the EC gate by interactions of individual lipid in the EC vestibule, as described in detail, together with the mutagenesis and functional validation experiments, in our recent publication.^[8]

The Lipid Membrane Is a Responsive Environment Supporting the Function of TMEM16 Scramblases

An important aspect of the gating mechanism described above is the direct involvement of individual lipids from the surrounding membrane in the gating dynamics that allow another lipid to be translocated through the groove (see sequence in Figs. 3a–3d). On the other hand, the conformational changes of the groove related to TMEM16 functions were found to range from a fully open lipid-conductive state of the groove facing the membrane, to a closed state in which the groove is occluded from the membrane and not permissive to lipid translocation. The deformation of the membrane surrounding the inserted TMEM16 observed both in the simulations and in the determined structures, led to the suggestion that the membrane environment plays a regulatory role in the functional dynamics of TMEM16 PLS,^[11] as discussed below. As the recent advances in structure determination of TMEM16 PLS in lipid nanodiscs^[10,11,28] have brought to light important structural details regarding the organization of the nanodisc membrane around the scramblase proteins, they offer invaluable insights regarding the active participation of the lipid environment in enabling the different functional states of these proteins.

In the cryo-EM structures of nanodiscs containing a TMEM16 scramblase the membrane exhibits a unique deformation pattern

Two types of lipid nanodiscs, differing in their size and lipid composition, have been utilized for structural studies on TMEM16 scramblases. The studies on afTMEM16^[11] were done using relatively small size (~120 Å in diameter) nanodiscs composed of 3:1 mixture of POPE:POPG lipids and surrounded by the MSP1E3 scaffold protein. The work on nhTMEM16^[10] and TMEM16F^[28] scramblases used larger (~185 Å in diameter) nanodiscs stabilized by a different scaffold protein, MSP2N2, and composed of either 7:3^[10] or 3:1^[28] mixtures of POPC:POPG lipids.

A striking unifying structural feature of the fungal TMEM16 containing nanodiscs is their strongly deformed shape, which seems to be determined by the interaction of the membrane with the special architecture of the TMEM16. In Figure 4a, the density map of the MSP1E3 nanodisc surrounding Ca²⁺-bound afTMEM16 is the result of the combination of the densities of lipid molecules and of the scaffolding protein that encircles the nanodisc boundary. The strong deformation of the nanodisc along the protein dimer interface (i.e., along the longest dimension of the protein) is clearly visible. It is thicker around the long TM3 and TM5 helices in one monomer and thinner at the short TM1 and TM2 helices of the other, and behaves symmetrically on the flip side, as the protomers are symmetrical (left panels in Figs. 4a and 4b). The overall slanting of the nanodisc closely follows the incline of the plane defined by the extracellular ends of the five helices lining the protein dimer interface (TMs 1, 2, 10, 3, and 5, left panel in Fig. 4a), suggesting that the main driving force for the deformations relates to alleviation of the hydrophobic mismatch between the protein and the nanodisc bilayer.

Figure 4.

Membrane bending and thinning by afTMEM16 observed in cryo-EM structures. a–c) Left: atomic model of afTMEM16 in MSP1E3 nanodiscs with 3:1 POPE/POPG lipids in the presence of Ca²⁺ (a, PDBID 6E0H), absence of Ca²⁺ (b, PDBID 6DZ7), and presence of Ca²⁺ and inhibitory ceramide 24:0 (c, PDBID 6E1O), inside the respective low-pass filtered (10 Å) nanodisc density from cryo-EM structures. Dimer cavity helices are shown as cylinders (TM1-red, TM2-orange, TM3-blue). Right: top view of left. Insets show tilted view of the permeation pathway to highlight the membrane thinning. Reproduced from Ref. [11] with permission from the authors.

Similarly, the nanodiscs surrounding the nhTMEM16 scramblase,^[10] which is architecturally very similar to afTMEM16, exhibit the analogous membrane slanting pattern. The analogy is more striking considering that the nanodisc enveloping the nhTMEM16 dimer (i.e., MSP2N2) is larger. These results indicate that the observed deformations of the nanodisc membrane is due to the specific 3D arrangement of the transmembrane helices in these scramblases. Indeed, a similar MSP2N2 nanodisc that surrounds the mouse TMEM16A (mTMEM16A) ion channel, does not exhibit such pronounced bending.^[30] This can be explained^[11] by the observation that in the mTMEM16A channel, TM1 is longer whereas TM3 is shorter compared to both afTMEM16 and nhTMEM16, such that the extracellular ends of these helices in the ion channel sit at comparable heights within the membrane, which would result in different pattern of hydrophobic mismatch along the protein-membrane boundary and lead in this case to a more regular nanodisc bilayer shape.

The relation of this membrane remodeling pattern to the function of TMEM16 proteins is underscored by the comparison of afTMEM16 (or nhTMEM16)-nanodisc complexes in the presence/absence of the activating bound Ca²⁺. Specifically, in the Ca²⁺-bound state the groove region is open to the lipid membrane, but not so in the apo state. Thus, the +/− Ca²⁺ structures differ as a result of large-scale conformational rearrangements of TM4 and TM6^[10,11,28] in gating dynamics resulting from Ca²⁺ binding. Interestingly however, the overall bending deformation of the nanodisc described above for Ca²⁺-bound scramblases, is recapitulated under Ca²⁺-free conditions (Fig. 4b). This is consistent with the attribution of the bending to the architecture of the TMs at the dimer interface, whereas the Ca²⁺-dependent structural rearrangements are primarily confined to the groove region. Indeed, the nanodisc density maps for the Ca²⁺-bound and Ca²⁺-free complexes reveals significant differences near the groove (compare insets in Figs. 4a and 4b). Consistent with the observation from MD simulations^[8,14–16] that the open conformation of the groove in the Ca²⁺-bound scramblases is permissive to lipids (see previous sections) and can result in local thinning of the bilayer around the groove (see also below), the density in this region is weaker compared to the rest of the nanodisc density (inset in Fig. 4a), indicating a thinning of the membrane in this region and/or higher mobility of the lipids in the groove region.^[11] In the Ca²⁺-free system, on the other hand, the membrane density near the groove remains as strong as in the other parts of the nanodisc (inset in Fig. 4b).

Thus, the structural evaluation of TMEM16-nanodisc complexes reveals two regions of the nanodisc where the membrane shape is distorted by the inserted TMEM16 scramblase dimer. The first is across the dimer interface of the protein, where the surrounding membrane is bent in opposite directions on the two sides of the groove, so that the groove is exposed on one side to the thinner bilayer segment, whereas the other faces the thicker one. The second region of interest is the groove region itself where, in the presence of Ca²⁺ ions under conditions of conductive protein the density map is significantly weaker which could be attributed to lipids dynamically populating this area with their lipid tails. These findings are fully consistent with the results from the MD simulations described above, led to a mechanistic hypothesis for the role of the distortions observed in the nanodisc, in the function of the scramblases.^[11] According to this hypothesis, the membrane bending along the dimer interface creates a gradient in bilayer thickness on the opposite sides of the groove and thereby primes the system for the scrambling by facilitating lipid transfer between leaflets through the translocation pathway formed by the open groove.

As this fundamental mechanistic model for the function of TMEM16 scramblases is based on structural observations in nanodiscs, we pursued directly the comparison between the observed membrane deformations and the response of the membrane in planar bilayers, and how they may be affected by putative finite size effects in the nanodisc construct. Earlier MD simulations of nhTMEM16 in POPC membranes showed overall slanting distortions of the bilayer,^[14,31] but the lipid compositions used in cryo-EM studies are different from simple POPC. Therefore, we addressed this question in comparative atomistic MD simulations of Ca²⁺-bound nhTMEM16 in lipid nanodiscs of different sizes, and in lipid membranes using lipid compositions from the cryo-EM experiments.

Computational Assessment of Membrane Deformation Patterns in Nanodiscs and Planar Lipid Membranes Containing a TMEM16 Scramblase

For the nhTMEM16-nanodisc complexes assembled from various lipid mixtures and scaffold proteins we matched the experimental setups^[10,11,28] (see Methods for more details). Figure 5 depicts nhTMEM16 embedded in the nanodiscs of 3:1 POPE/ POPG lipid composition and surrounded by either MSP1E3 (panels a, c) or MSP2N2 scaffolds (panels b, d). As shown in Table 1, these two nanodiscs differ substantially in the total number of lipids, yielding an ~63 Å difference in their diameter (~121 Å and ~184 Å for the MSP1E3 and MSP2N2 nanodiscs, respectively, obtained from CHARMM-GUI). While the radial length of the smaller MSP1E3 nanodisc is comparable to the longest dimension of nhTMEM16 (Fig. 5c), the nanodisc is still spacious enough to encapsulate the TM region of the protein. However, some regions of the scramblase, and in particular, structural segments that define the PL pathway, are situated very close to the nanodisc boundary so that only few lipids can be packed between nhTMEM16 and the MSP1E3 scaffold (Fig. 5c). In contrast, the larger MSP2N2 nanodisc accommodates nhTMEM16 with several layers of lipids between the protein boundary and the scaffold protein which demarcates the nanodisc (Figs. 5b and 5d).

Figure 5.

The lipid nanodisc systems containing Ca²⁺-bound nhTMEM16. Top views a,b) and side views c,d) of the nhTMEM16-containing nanodiscs created with MSP1E3 (a, c) and MSP2N2 (b,d). nhTMEM16 is shown in tan cartoon, whereas the two belts of the scaffolding proteins are drawn in red and blue cartoon. The lipids of the nanodisc are depicted with lines. In panel c, the locations of PL pathway (grooves) are indicated on the two sides of the dimer. The simulation of these molecular systems was done as described in Methods, in a cubic solvent box (removed for clarity). See also Table 1.

Table 1.

Listing of the molecular constructs studied with atomistic MD simulations. Shown (from left to right) are: the lipid environment, lipid composition, total number of lipids and of atoms in these systems, and the simulation time, in μs (excluding equilibration phases).

Environment	Lipid composition	Number of lipids	Overall number of atoms	Simulation time (μs)
Flat membrane	3:1 POPE:POPG	851	390,000	10
Flat membrane	7:3 DEPC:DEPG	1110	565,000	1
MSP1E3 nanodisc	3:1 POPE:POPG	228	575,000	2
MSP2N2 nanodisc	3:1 POPE:POPG	714	1,300,000	0.1
MSP1E3 nanodisc	7:3 DEPC:DEPG	215	576,000	4.0

During the MD simulations, these nanodisc systems retain their overall dimensions (Table S2) but become elliptical in shape due to the architecture of the embedded scramblase (see Table S2 showing that the diameters along the two principal directions of the nanodisc are different). Importantly, the nanodiscs develop the same slanting distortions seen in the cryo-EM density maps (see Fig. 6a for 3:1 POPE/POPG MSP1E3 nanodisc, and also Figure S4A–B; because the deformation for the 3:1 POPE/POPG MSP2N2 nanodisc is very similar, it is not shown). Indeed, the nanodisc bilayer is thinner around TM1 and TM2 and becomes thicker near TM3 (Fig. 6a) just as observed in the cryo-EM maps. Importantly, this bilayer deformation pattern at the protein-lipid interface is recapitulated in our simulations of nhTMEM16 in regular lipid membranes composed of the same 3:1 POPE/POPG mixture (Fig. 6b). However, as described next, we find that the overall bilayer deformations in the nanodisc are very different to those we observed in the corresponding MD simulations in the planar membrane systems, quite likely due to the presence of the encircling scaffold proteins.

Figure 6.

Remodeling of the lipid membrane by Ca²⁺-bound nhTMEM16 in nanodiscs and planar lipid membranes. Top panels show results from atomistic MD simulations in 3:1 POPE/POPG lipids in: a) a MSP1E3 nanodisc, and b) a planar lipid membrane (Table 1). Lower panels show the same comparison for 7:3 DEPC/DEPG lipids: c) MSP1E3 nanodisc, and d) planar lipid membrane. In the snapshots, nhTMEM16 is shown in cartoon with TM1, TM2, and TM3 helices colored in red, orange, and blue, respectively. The bound Ca²⁺ ions are shown as yellow spheres. The membrane is depicted by the locations of the headgroup phosphorus atoms of the two leaflets in the respective MD trajectories. In panel a, the phospholipid (PL) pathway are indicated by arrows and labeled.

The comparison of the thicknesses around nhTMEM16 in the 3:1 POPE/POPG MSP1E3 nanodisc, and in the lipid membrane is shown for each of the two membrane leaflets in Figures 7a–7c and 8a–8c (see Methods for more details of this analysis and Figs. S5 and S6 for statistical significance of the thickness data). In the planar membrane, changes in the upper or lower leaflet thickness are seen to be mostly confined to the protein-lipid boundary (blue and red shapes in Figs. 8a–8c), but in the rest of the membrane the thickness of the two leaflets fluctuates around its unperturbed (bulk) value (green shades in Figs. 8a–8c). In the MSP1E3 system, on the other hand, both leaflets experience strong thinning (blue shades in Figs. 7a–7c) also beyond the protein-membrane interface, due to the fact that at the edges of the nanodisc boundary the bilayer thickness is constrained (by the scaffold proteins).^[22] This edge effect is demonstrated by the results of simulations of the bare 3:1 POPE/POPG MSP1E3 nanodisc (i.e., without nhTMEM16), where the bilayer thickness (measured as the distance between the P atoms of the lipid headgroups in opposite leaflets) ranges from ~44 Å at the nanodisc center to <30 Å at the nanodisc boundary (Fig. S7), that is, to a value similar to that measured on the edge of the scramblase-containing nanodisc (in Figs. 7a–7c thickness per leaflet on the nanodisc edge is ~15 Å).

Figure 7.

Representations of the thickness changes in the upper and lower leaflets of the membrane surrounding Ca²⁺-bound nhTMEM16 in nanodiscs from the analysis of atomistic MD simulations. a,b) Average thickness of the upper (a) and lower (b) leaflets of an nhTMEM16-containing MSP1E3 3:1 POPE/POPG nanodisc from atomistic MD simulations (Table 1). The color code represents local thickness values in Å units, and the scale bar is 20 Å. c) Representation of the membrane leaflet surfaces by the average locations of headgroup phosphorus atoms of the two leaflets in the respective MD trajectories of the nanodisc membrane surrounding nhTMEM16. The leaflets are colored according to the color map in panel a. The groove helices (TMs 3, 4, and 6) are rendered in orange in both protomers. d,e) Same as in panels a and b, for 7:3 DEPC/DEPG MSP1E3 nanodisc containing nhTMEM16. f) Same as in panel c, for 7:3 DEPC/DEPG MSP1E3 nanodisc containing nhTMEM16. The leaflets are colored according to the color map in panel d.

Figure 8.

Representations of the thickness changes in the upper and lower leaflets of the membrane surrounding Ca²⁺-bound nhTMEM16 in planar lipid membranes from the analysis of atomistic MD simulations. a,b) Average thickness of the upper (a) and lower (b) leaflets of an nhTMEM16-containing 3:1 POPE:POPG membrane from atomistic MD simulations (Table 1). The color code represents local thickness values in Å units, and the scale bar is 20 Å. c) Representation of the membrane leaflet surfaces by the average locations of headgroup phosphorus atoms of the two leaflets in the respective MD trajectories of the membrane surrounding nhTMEM16. The leaflets are colored according to the color map in panel a. The groove helices (TMs 3, 4, and 6) are rendered in orange in both protomers. d,e) Same as in panels a and b, for a 7:3 DEPC/DEPG membrane containing nhTMEM16. f) Same as in panel c, for a 7:3 DEPC/DEPG MSP1E3 membrane containing nhTMEM16. The leaflets are colored according to the color map in panel d.

Overall, these results indicate that the bilayer deformation pattern along the interface of the protein with the surrounding membrane as observed in the cryo-EM maps is not an artifact of the finite size effect of the nanodisc construct, but is also observed in the simulations of periodic lipid membrane system with nhTMEM16, in line with the conclusions of the previous MD studies.^[14,31] This is consistent with the premise^[10,11,28] that the slanting distortions surrounding the protein are indeed induced by the unique architecture of the scramblase. However, our finding that the overall shape of the MSP1E3 nanodisc around nhTMEM16 is drastically different from that of the planar membrane indicates that the slanting deformations may incur different energy costs^[32] in the two lipidic settings. Such energy cost differences have been shown to have direct effects on the functional properties of membrane proteins.^[32–36]

A direct relation between the lipid environment and the functional efficiency of the afTMEM16 scramblase is reflected in the shape of the protein-membrane interface

Seeking to evaluate mechanisms underlying the relationship between the activation state of the TMEM16 scramblases by Ca²⁺ binding, and the membrane deformation in the groove region evidenced in the structural data, as well as consequences of such deformations for TMEM16 PLS function, we investigated the modulation of the protein-membrane interface by lipid composition. The effect of membrane thickness is a major factor because bilayer deformations alleviate the overall hydrophobic mismatch between the protein and the surrounding membrane. The physicochemical properties of the lipid bilayer impact the shape of the membrane around the inserted scramblase, and the dependence of the energy cost of modulation on such properties has been formulated and connected to functional properties.^{[32–34,36–41]} The length of the lipid tails in the membrane is often a strong determinant of the shape and energy cost.^[36,39,40]

Functional studies of the afTMEM16 scramblase in liposomes of different lipid compositions using an in vitro scramblase assay^[42] found that in liposomes assembled from lipids of various tail length and saturation^[11] the scrambling activity is dramatically reduced (>10²-fold) in mixtures of long-tail 22:1 DEPC and 22:1 DEPG (7:3 ratio) lipids, compared to activity in 16:0–18:1 POPC/POPG (3:1 ratio) composition.^[11] Furthermore, addition of just 5% long-tail saturated ceramide lipids (either 22:0 or 24:0 chain) to a 3:1 mixture of POPE/POPG had a similarly strong effect (>10²-fold reduction) on scrambling activity.^[11]

However, when the structure of Ca²⁺-bound afTMEM16 was determined in nanodiscs containing 5% of the inhibiting 24:0 ceramide lipid mixed with 3:1 ratio of POPE/POPG, the protein groove was found to be similar to that in the crystal structure of nhTMEM16, or the cryo-EM structure of Ca²⁺-bound afTMEM16 in 3:1 POPE/POPG nanodiscs. Given the inhibitory effect of the long-tailed lipids, the finding of an open conformation rather than the closed conformation expected from the usual forms of inactivated scramblases seemed surprising,^[11] suggesting that presence of the inhibitory long-tail lipids did not affect scrambling by closure of the groove. As shown in Figure 4c (left panel), the nanodisc still exhibits the slanting along the dimer interface in the presence of the ceramide lipids, consistent with the idea that this membrane deformation is due to the afTMEM16 architecture. However, the results showed that the density at the groove region of these 24:0 ceramide-containing nanodiscs is stronger than that seen in the functional 3:1 POPE/POPG Ca²⁺-bound afTMEM16 nanodiscs in which the lipid pathway is similarly open (compare insets in Figs. 4c and 4a). This difference suggested a direct relation between the density of the nanodisc membrane in this area and the activity of the protein.^[11] The mechanistic question regarding inhibition by these long-tailed lipids remained open, because it was still unclear what mode of protein-lipid interactions at the groove gives rise to the difference observed in nanodisc densities when comparing regular length lipid environments to the long-tailed lipid ones. We addressed this question by carrying out the comparative analyses described below.

The response of the longer tailed DEPC/DEPG lipid membrane to nhTMEM16 is different from that of POPE/ POPG membranes

The trajectories from atomistic MD simulations of Ca²⁺-bound nhTMEM16 scramblase in MSP1E3 nanodisc composed exclusively of long-tail phospholipids (7:3 mixture of DEPC and DEPG lipids) which dramatically inhibits scrambling activity,^[11] were compared to those from parallel simulations of the same protein construct and lipid mixture, but in planar membranes under periodic boundary conditions. Figures 6c and 6d shows the arrangement of the phosphorus atoms of DEPC and DEPG lipid headgroups around nhTMEM16 embedded in the MSP1E3 nanodisc (Fig. 6c), or in the planar membrane (Fig. 6d), during microsecond-scale MD simulations. The slanting distortion observed in the 3:1 POPE/POPG system is clearly visible here as well (see also Fig. S4C–D). However, as shown by comparison of the thickness profiles depicted in Figure 8, the 7:3 DEPC/DEPG membrane is overall thicker compared to POPE/POPG, as expected from the presence of longer tail lipids. Especially important are the differences in the membrane shapes in the groove region where the POPE/POPG membrane exhibits thinning in both upper and lower leaflets (blue shades in Figs. 8a–8c), but the DEPC/DEPG mixture shows notable thinning only in the upper leaflet (deformations of the lower leaflet are minimal). This indicates that the headgroups of DEPC and DEPG lipids only weakly penetrate the groove region of nhTMEM16.

Further analysis of these trajectories reveals hydrophobic tails of DEPC and DEPG lipids in the groove (see snapshots in Fig. S3C), which is therefore in a dehydrated nonconductive state.^[8] In some trajectories (in Fig. S3C see instances of short V337-V447 distance), the groove is seen to close on the penetrating lipid tails (see snapshot “i” in Fig. S3C) similar to that observed as a rare occurrence in our previous MD simulations of nhTMEM16 in POPC lipids^[8] (in Fig. 2 population of states marked with “*”). These results suggest a more frequent occurrence of lipid tails penetrating the groove in these thicker membranes. Taken together, this analysis provides a possible mechanistic explanation for the low activity of the homologous afTMEM16 in the DEPC/DEPG mixture,^[11] and underscores how the lipidic environment can modulate the functionally relevant dynamics inTMEM16 scramblases.

Membrane remodeling distal from the scramblase protein is different in nanodiscs vs flat membrane

Comparison of bilayer deformations in MSP1E3 nanodisc and in the lipid membrane—both composed of 7:3 DEPC/DEPG (Figs. 7d–7f and 8d–8f, respectively)—reveals that in this lipid mixture, the nanodisc bilayer is overall substantially thinner than in the flat membrane, as described above for the 3:1 POPE/POPG nanodisc. Especially significant (see Figs. S4 and S5) is the difference of the membrane remodeling away from the scramblase, as in the nanodiscs of both mixtures there is a similarly strong thinning of the membrane (~5 Å). We note that the larger extent of the area devoid of lipid in the DEPC/DEPG nanodisc (cf. white regions in Figs. 7e and 7b) is due to a different positioning of TM6 which, in the DEPC/DEPG system, extends on the intracellular side all the way to the nanodisc boundary (see Fig. 7f). Because the absolute thickness is greater there than in the POPE/POPG system, TM6 may be able to reach the edge.

Concluding Remarks

This study of special modes of interaction between the TMEM16 PLS and the membrane surrounding them, illustrating the special context created by the membrane lipids being the translocation substrate, and the membrane participating actively in the functional mechanism both through the actions of individual lipids and the response of the entire environment. We and others recently used MD simulation and cryo-EM approaches to reveal molecular mechanisms of activation of TMEM16 PLS that remodel the membrane in flat lipid membranes^[8,14–16] and nanodiscs.^[10,11,28] The results from these previous studies reviewed here demonstrated a dependence of PLS activity on the physicochemical properties of membranes, such as thickness or fluidity. Indeed, changes in phospholipid composition or addition of membrane-modifying lipids (e.g., long-chain ceramides) affect PLS activity in vitro.^[11] The membrane was shown to participate in the opening and stabilization of the pathway (open groove) for translocation of lipids and ions between the membrane leaflets. In parallel, the detailed description of the steps in lipid translocation from one leaflet to another, obtained from the analysis of the simulation trajectories, highlights the mechanistic hypothesis that emerged from the computational studies and was verified experimentally^[8] regarding the gating of this conduit by site specific interactions of the PLS with individual lipids. Importantly, this combination of membrane-dependent mechanisms suggests a putative physiological mechanism for the experimentally observed tuning of TMEM16 PLS activity in different cells and also different cellular compartments. This key insight suggests that the observations may reflect differences, and/or physiologically regulated changes, in the local membrane environment in which the TMEM PLS are expressed.

The simulations of TMEM16 PLS systems that sampled the role of the membrane in key conformational rearrangements of the groove to produce functionally different states, and the elucidation of testable molecular details of the interactions and mechanisms that promote these functional states, were carried out, as described here, in various flat lipid membrane systems. The first comparison of bilayer shaping from simulations of the same TMEM16 scramblases in same-composition nanodiscs presented here revealed both similarities and important differences from those observed in the flat membrane systems. The slanting of the membrane at the boundary with the protein is similar in the compared systems, which supports the idea that this observed membrane remodeling is due to the molecular architecture of the TMEM16 scramblase. As discussed in the text, however, achieving this slanting remodeling entails a different energy cost in the nanodisc and flat membranes. That such energy cost differences can lead to functional and mechanistic differences that can be evaluated computationally has been shown in a variety of other systems.^[32–36] The combination of MD simulations and structure determination of the TMEM16 PLS under various conditions is seen to provide very important mechanistic understanding of the interplay between proteins and membranes in the performance of physiological functions. Clearly, MD simulations under conditions similar to those used in cryo-EM structure determination in lipid nanodiscs will continue to be an important component in the explosive growth in the reach, and success, of this new and exciting method of structure determination.

Methods

Molecular constructs

All the new simulations of wild type nhTMEM16 described here used the crystal structure from PDBID 4WIS,^[7] with the missing structural details complemented as described.^[8] nhTMEM16- nanodisc complexes were prepared using the CHARMM-GUI web interface.^[22,43] The protein structure was inserted in either MSP1E3 or MSP2N2 nanodiscs. Two different lipid compositions were constructed for MSP1E3: (1) a 3:1 mixture of POPE and POPG lipids; (2) a 7:3 mixture of DEPC and DEPG lipids. For the MSP2N2 nanodisc only the 3:1 mixture of POPE and POPG lipid was used. nhTMEM16-nanodisc complexes were surrounded by cubic solvation boxes containing 0.15 M KCl salt solutions. The final sizes of these systems are given in Table 1.

Using the CHARMM-GUI, we also constructed a bare MSP1E3 nanodisc (i.e., without nhTMEM16) composed of 3:1 POPE/POPG lipid mixture and embedded it in a cubic water box containing 0.15 M KCl salt.

Two additional systems in which nhTMEM16 was embedded into regular planar membranes were prepared. These systems contained the same mixtures of lipids as the above nanodiscs, that is, either 3:1 mixture of POPE and POPG lipids, or 7:3 mixture of DEPC and DEPG lipids. These protein-membrane complexes were immersed in a water box containing 0.15 M KCl salt.

Molecular dynamics simulations

All MD simulations implemented the CHARMM36 force-field parameters for proteins,^[44] lipids,^[45,46] and ions^[47] and were initiated with multistage equilibration protocol provided by CHARMM-GUI for the nanodisc and planar membrane simulations, respectively. For the nanodisc systems, as well as for the DEPC/DEPG membrane system, this equilibration phase was followed by short (5–10 ns) unbiased MD simulations carried out with NAMD version 2.12.^[48] The POPE/POPG membrane duration of this unbiased MD step was longer (~380 ns). All the NAMD runs implemented the NPT ensemble and used the standard set of run parameters in CHARMM-GUI.

After this initial phase, the two MSP1E3 nanodisc systems and the two membrane systems were subjected to microsecond-scale MD simulations on Anton2^[29] (the timescales are indicated in Table 1). Due to its large-size, the nhTMEM16-MSP2N2 nanodisc system could not be accommodated on the Anton2 machine, and was therefore simulated with NAMD for additional ~100 ns. The nanodisc simulations employed isotropic pressure coupling whereas the membrane simulations were carried out using a semi-isotropic pressure coupling scheme. All other simulation parameters and details are as described recently.^[8]

Additional ensemble MD simulations were carried out on an nhTMEM16-MSP1E3 nanodisc system containing a mixture of DEPC/DEPG lipids. Starting with the initially equilibrated structure (obtained from the NPT simulations with NAMD, see above), the system was simulated in 20 independent replicates (with randomizing starting velocities), each for 100 ns. These runs were carried out with ACEMD software,^[49] using NVT ensemble and the force-fields and run parameters described in Ref ^[8]. Specifically, the following set of run parameters were used: timestep 4 fs, vdwforceswitching on, switching on, switchdist 7.5, cutoff 9, fullelectfrequency 2, langevindamping 0.1, pme on, and pmegridspacing 1.0.

Analysis of bilayer leaflet deformations

The analysis of bilayer leaflet shapes around nhTMEM16 was done on converged parts of the trajectories as assessed by monitoring the time-evolution of the protein root-mean-square deviation (RMSD) measure. Thus, for the nhTMEM16 in the 3:1 POPE/POPG membrane system, we excluded the first 3 μs; for the nhTMEM16 in the 7:3 DEPC/DEPG membrane system, the first 100 ns of MD trajectories were excluded; for the nhTMEM16 in the 3:1 POPE/POPG MSP1E3 nanodisc, we excluded the first 400 ns; for the nhTMEM16 in the 7:3 DEPC/ DEPG MSP1E3 the first 300 ns trajectory was excluded; for the 3:1 POPE/POPG MSP1E3 nanodisc without nhTMEM16, the whole 1 μs MD simulation trajectory was used for the analysis.

Time-averaged leaflet thicknesses were calculated with the algorithm described recently^[50] using grid spacing of 2 Å. The Z coordinates (along the axis perpendicular to the membrane) of phosphorous atoms (P) and the last carbon atoms (C218 and C316 in POPE/POPG system or C222 and C322 in DEPC/DEPG system) were obtained from a frame of MD simulation. For each frame, the heights of these atoms at a particular grid point were calculated by interpolation of their Z coordinates from all X-Y positions.^[51] Once all heights were obtained, the leaflet thickness for each frame at each grid point is the difference between the local height of the phosphorous atoms and the local height of the last carbon atoms. The average leaflet thickness is the time- averaged local thickness in the leaflet. The boundary between protein and membrane was calculated as a time-averaged XY plane occupation of the protein with −20 Å < Z < 0 Å for the lower leaflet and with 0 Å < Z < 20 Å for the upper leaflet. The thickness of the bilayer is defined as the height difference between the headgroup phosphorus atoms of the upper leaflet and those of the lower leaflet within the same grid.