Structural basis of Ca2+-dependent activation and lipid transport by a TMEM16 scramblase

1The plasma membranes of eukaryotic cells are organized in an asymmetric manner; at rest, polar and charged lipids are sequestered to the inner leaflet by the activity of ATP-driven pumps. Activation of a specialized class of membrane proteins – phospholipid scramblases – causes rapid collapse of this asymmetry and externalization of negatively charged phosphatidylserine molecules. As a result, extracellular signaling networks, controlling processes such as apoptosis, blood coagulation, membrane fusion and repair, are activated. The TMEM16 family of membrane proteins includes phospholipid scramblases and Cl- channels, all of which are Ca2+-dependent. Notably, TMEM16 scramblases also mediate Ca2+-dependent ion transport. Prior structural and functional analyses of the fungal nhTMEM16 scramblase from Nectria haematococca identified a membrane-exposed hydrophilic groove that serves as the translocation pathway for ions and lipids. In the TMEM16A channel, this pathway is sealed from the membrane, preventing lipid access while allowing only ion permeation.

2Phospholipid scramblases are unusual membrane proteins in the sense that their environment serves as their substrate. Activation of scramblases results in the fast and passive transbilayer movement of lipids. Thus, scramblases should affect the surrounding membrane to create a conduit between leaflets though which lipids can diffuse. Current models of lipid translocation, based on the structure of the Ca2+-bound conformation of the nhTMEM16 scramblase, postulate a mechanism in which lipid headgroups move through the hydrophilic permeation pathway while the tails remain embedded in the membrane core. However, little is known about whether or how lipids and Ca2+ binding –the physiologic activation trigger– affect the structure of the TMEM16 scramblases. Further, it is not known how these proteins affect the surrounding membrane to enable lipid scrambling. Here, we use cryo-electron microscopy (cryo-EM) and functional experiments to address these questions. We determine the structures of a functionally characterized TMEM16 scramblase, afTMEM16 from Aspergillus fumigatus, in lipid nanodiscs in an inactive (Ca2+-free) and an active (Ca2+-bound) conformation. These structures allow us to define key conformational rearrangements that underlie Ca2+-dependent scramblase activation. Additionally, we show that scrambling is inhibited by the lipid C24:0 ceramide (Cer24:0) and determine the 3.6 Å resolution structure of the Ca2+-bound afTMEM16/Cer24:0-nanodisc complex. These structures, together with functional experiments, suggest that the Ca2+-dependent conformational rearrangements described here allow the scramblase to locally thin the membrane at the opened lipid pathway to facilitate lipid transport.

3To isolate conformations of afTMEM16 relevant to its scramblase activity, nanodiscs were formed from a 3:1 mixture of POPE:POPG lipids where afTMEM16 mediates lipid scrambling while its non-selective ion transport activity is silenced. We used single-particle cryo-EM to determine the structures of nanodisc-incorporated afTMEM16 in the presence and absence of Ca2+ to resolutions of 4.0 and 3.9 Å, respectively (Figure 1). In both conditions, afTMEM16 adopts the TMEM16 fold (Figure 1; Figure 1—figure supplement 1–6), where each monomer in the dimeric protein comprises a cytosolic domain organized into a 6 α-helix (α1-α6)/3 β-strand (β1-β3) ferredoxin fold and a transmembrane region encompassing 10 α-helices (TM1-TM10) (Figure 1—figure supplement 5). The two monomers are related by a twofold axis of symmetry at the dimer interface, formed by TM10 and the cytosolic domain (Figure 1E). The helices near the dimer interface delimit two large hydrophobic cavities, called dimer cavities, which are lined by TM1, TM2 and TM10 from one monomer and TM3, TM5 from the other (Figure 1E). In both maps, the C-terminal portion of TM6 and the linker connecting it to the short cytosolic α4 helix are not well-resolved, likely reflecting their mobility within the membrane (Figure 1A,B). When processing the data without imposing symmetry between subunits, the two monomers differ in the well-resolved density of the ~22 C-terminal residues of TM6, likely reflecting the asymmetric orientation of the protein within the nanodisc (Figure 1—figure supplement 2). The maps generated by signal subtracting the nanodisc and imposing C2 symmetry are of higher resolution and nearly identical to the non-symmetrized (C1) maps, apart from the resolved portions of TM6 (Figure 1—figure supplements 2 and 3), and were therefore used for model building.

4In the presence of the functionally saturating concentration of 0.5 mM Ca2+, afTMEM16 adopts a conformation similar to that of Ca2+-bound nhTMEM16 and hTMEM16K in detergent (Figure 1—figure supplement 5). The TM3-6 from each monomer form a peripheral hydrophilic cavity that opens to the membrane with a minimum diameter of ~5 Å. This pathway was proposed to allow entry and translocation of phospholipid headgroups and is analogous to the ion pathway in the TMEM16A channel. Thus, the presence of a lipid environment does not affect the Ca2+-bound conformation adopted by the scramblase.

5To understand the changes that occur upon Ca2+ activation, we compared our Ca2+-bound and Ca2+-free structures of afTMEM16 and identified global conformational rearrangements of the lipid pathway, of the cytosolic domains and of the Ca2+ binding site (Figure 2, Video 1). In the absence of Ca2+, the cytosolic domains of afTMEM16 translate ~3 Å parallel to the plane of the membrane, away from the twofold axis of symmetry, such that the overall cytosolic domain becomes more expanded and the cytosolic α7 helices tilt toward the axis of symmetry with no noticeable movement of the transmembrane dimer interface (Figure 2, left panel; Figure 2—figure supplement 1). In the Ca2+-free conformation, the afTMEM16 lipid pathway is closed to the membrane by a pinching motion of the extracellular portions of TM4 and TM6, which move toward each other by ~7 and~3 Å, respectively (Figure 2, right panel). From the Ca2+-bound conformation, TM4 bends around two prolines (P324 and P333) and TM3 slides by ~6 Å to reach the Ca2+-free conformation (Figure 3A). Additionally, the intracellular portion of TM6 kinks around A437 by ~20o, inducing a ~ 16 Å vertical displacement of its terminal end (Figure 3A, Video 2). These rearrangements lead to tighter packing of side chains from TM4 and TM6 and result in exposure of a hydrophobic surface to the membrane core (Figure 3B). The pathway is also closed to ion entry by multiple stacked aromatic and hydrophobic side chains from TM3-7 (Figure 3C). The narrowest access point of the lipid pathway constricts from ~5 to ~1 Å in the absence of Ca2+ preventing lipid entry (Figure 3D–F). Notably, mutating residues at the interface between helices that rearrange upon Ca2+ binding results in severely impaired scrambling activity in the closely related nhTMEM16 homologue (Figure 3—figure supplement 1), highlighting the importance of these dynamic rearrangements. The global Ca2+-dependent rearrangements of the afTMEM16 scramblase contrast to the local conformational changes seen in the TMEM16A channel, where only TM6 bends upon Ca2+ions leaving the binding sites (Figure 3—figure supplement 2).

6The transmembrane region of afTMEM16 contains two Ca2+-binding sites, located between TM6, TM7, and TM8 (Figure 4A,B). In the presence of 0.5 mM Ca2+, both sites are occupied by calcium ions, as evidenced by the strong densities in the cryo-EM map and in an ‘omit’ difference map, calculated between experimental data and simulated maps not containing Ca2+ (Figure 4A). The bound Ca2+ ions are coordinated by five conserved acidic residues (E445 on TM6, D511 and E514 on TM7, and E534 and D547 on TM8), three polar residues (Q438, Q518, N539), and the unpaired main chain carbonyl of G441 (Figure 4A–C). This coordination is similar to that seen in the nhTMEM16 and hTMEM16K scramblases as well as in the TMEM16A channel, consistent with the evolutionary conservation of the Ca2+-binding residues (Figure 4C). In the absence of Ca2+, the binding site is disrupted by the movement of TM6 which displaces the three residues participating in the site (Q438, G441 and E445) (Figure 4D–F, Video 3). Additional rearrangements of TM8, displacing N539 and E543, further disrupt the Ca2+-binding site (Figure 4D–F, Video 3). No Ca2+ density was visible in the cryo-EM map (Figure 4E), confirming that the scramblase is in a Ca2+-free conformation. The movement of TM6 and TM8 in opposite directions partially relieves the electrostatic repulsion of the uncoordinated acidic side chains that form the Ca2+-binding site and opens a wide, water-accessible, conduit for ions to access the binding site from the intracellular solution.

7Because these structures are of protein/nanodisc complexes, they also provide insights into the Ca2+-dependent interactions between the scramblase and the nanodisc membrane (Figure 5). Inspection of these maps reveals that the nanodiscs containing afTMEM16 are bent along the two dimer cavities of the scramblase (Figure 5A,C) and that there is a region of low density at the open lipid pathway (Figure 5B). The nanodisc density is a convolution of the lipid molecules and the MSP1E3 protein that defines the nanodisc boundary, suggesting that presence of the afTMEM16 scramblase affects the orientation of both.The bending along the dimer cavities of afTMEM16 is seen in the 3D reconstructions from all datasets presented here, suggesting that it does not depend on the presence of Ca2+, on the conformation of the protein or processing algorithm (Figure 5A,C, Figure 5—figure supplement 1–3, Supplementary file 1). Remarkably, bending of the nanodisc is also visible in 2D classes of particles containing afTMEM16, but is absent from protein-free ones (Figure 5—figure supplement 4). In all maps, the membrane appears to be thicker around the long TM3 helix in one monomer and thinner at the short TM1 of the other (Figure 5E). This bending is visible along both dimer cavities, irrespective of the positioning of the scramblase within the nanodisc (Figure 5A,C, Figure 5—figure supplement 1). The independence of the observed membrane bending on Ca2+ is consistent with the lack of Ca2+-dependent rearrangements undergone by these cavities (Figure 3A). The membrane slant matches the tilted plane defined by the extracellular ends of the five, dimer cavity-lining helices (Figure 5E) suggesting that it is caused by the architecture of afTMEM16.

8In contrast to the membrane bending, the region of weaker density of the nanodisc membrane near the lipid pathway depends on the scramblase conformation (Figure 5B,D). In the three datasets collected in the presence of 0.5 mM Ca2+, the density between the TM4 and TM6 helices that delimit the open lipid transport pathway is weaker than in the rest of the complex (Figure 5B, insets, Figure 5—figure supplement 2). This weakening can be seen in both subunits irrespective of the distance of the pathway from the nanodisc edge (Figure 5B, insets). In the absence of Ca2+, no region of weak density is visible (Figure 5D, inset) as the conformational rearrangement of TM4 and TM6 pinches shut the permeation pathway. The weaker density suggests that the membrane is thinner and/or more disordered near the open lipid pathway of an active scramblase.

9The hypothesis that afTMEM16 alters the membrane to scramble lipids suggests that scrambling should be impaired by thicker membranes and by bilayer modifying lipids. To increase the membrane thickness, we reconstituted afTMEM16 into liposomes composed of lipids of varying acyl chain length and saturation. The rate of scrambling was determined with a dithionite-based fluorescence assay (Figure 6A). We found that afTMEM16 is equally active in liposomes formed from mixtures of POPE:POPG or POPC:POPG lipids, which have C16:0 and C18:1 acyl chains, and in liposomes formed from DOPC:DOPG lipids, where both acyl chains are C18:1 (Figure 6B, Figure 6—figure supplement 1, Supplementary file 2). This suggests that the saturation of the acyl chains does not influence afTMEM16 scrambling activity. In contrast, increasing membrane thickness by ~7 Å by forming liposomes with DEPC:DEPG lipids, with C22:1 acyl chains, reduces the scrambling rate of afTMEM16 by ~500 fold in the presence of Ca2+ and ~20-fold in the absence of Ca2+ (Figure 6A–B, Figure 6—figure supplement 1, Supplementary file 2). The observation that scrambling is inhibited in thicker membranes is consistent with our hypothesis that afTMEM16 thins the membrane to scramble lipids.

10Among naturally occurring bilayer-modifying constituents of cellular membranes, we focused our attention on ceramides, as these sphingolipids regulate cellular processes that involve activation of phospholipid scramblases, such as blood coagulation, inflammation and apoptosis. We found that addition of physiological levels of long chain ceramides potently inhibits scrambling by reconstituted afTMEM16 (Figure 6C, Figure 6—figure supplement 2A–H). Among the tested ceramides, C24:0 Ceramide (Cer24:0) inhibits scrambling ~250 fold when added at 5 mole% (Figure 6C). The inhibitory effect depends on ceramide concentration, with minimal effects at 1 mole%, as well as acyl chain saturation, as Cer24:1 is nearly inert (Figure 6C, Figure 6—figure supplement 2). The length of the ceramide acyl chain is also important as Cer22:0 is as potent as Cer24:0 while Cer18:0 has minimal effect at 5 mole% (Figure 6C, Figure 6—figure supplement 2F). The long chain ceramides also inhibit scrambling in the absence of Ca2+ by ~10-fold (Figure 6—figure supplement 2G). The inhibitory effects of ceramides or thicker bilayers do not reflect impaired reconstitution of the protein (Figure 6—figure supplement 2H). We used a gramicidin-based fluorescence quench assay to investigate whether Cer24:0 and Cer24:1 differentially affect bulk membrane properties, such as thickness and fluidity. The assay monitors alterations in the gramicidin monomer↔dimer equilibrium, which varies with changes in membrane thickness and elasticity. Addition of Cer24:0 or Cer24:1 comparably reduces gramicidin activity (Figure 6—figure supplement 2I–K), indicating that both ceramides stiffen and/or thicken the membrane to a similar extent. The comparable effects of Cer24:0 and Cer24:1 on membrane properties suggest that their distinct effects on afTMEM16 scrambling might reflect specific interactions with the scramblase and/or their differential ability to form gel-like microdomains in membranes.

11To understand how long-tail ceramides affect scrambling, we determined the 3.6 Å resolution structure of Ca2+-bound afTMEM16 in nanodiscs containing 5 mole% Cer24:0 (Figure 7; Figure 1—figure supplement 1–4, Figure 7—figure supplement 1), a concentration that drastically inhibits activity (Figure 6C). The protein adopts a conformation where the lipid pathway is open to the membrane, both Ca2+-binding sites are occupied, and is nearly identical to the Ca2+-bound active state, with an overall Cα r.m.s.d. <1 Å (Figure 7A). No individual lipids are resolved within the pathway, while several densities that we attributed to partial acyl chains are visible in the dimer cavity (Figure 7—figure supplement 1). In the Cer24:0-inhibited structure, the nanodisc membrane bends along the dimer cavities in a similar manner to what was seen in the Ca2+-bound and Ca2+-free structures, (Figure 7B). Interestingly, the density near the lipid pathway in the presence of Cer24:0 appears to be less weakened (Figure 7C) than that seen in the Ca2+-bound structure (Figure 5B), even though the pathway is open in both structures. While direct comparisons of the densities between the two structures are difficult because of their different resolutions, it is tempting to speculate that the density of the nanodisc membrane in this area correlates with the activity of the protein. Further work will be required to test this possibility. The finding that afTMEM16 adopts a Ca2+-bound conformation with open lipid pathways suggest that long chain ceramides do not inhibit scrambling by preventing the Ca2+-dependent opening conformational transition. Rather, our findings suggest that the physico-chemical properties of the membrane determine whether lipid scrambling actually occurs.

12Despite recent advances, the molecular mechanisms underlying the Ca2+-dependent activation of the TMEM16 scramblases and their interactions with the surrounding membrane lipids remain poorly understood. Here, we use cryo electron microscopy to determine the structure of a functionally well-characterized TMEM16 family member, afTMEM16, in a membrane environment. Our results show that the afTMEM16 scramblase reconstituted in lipid nanodiscs undergoes global conformational rearrangements upon Ca2+ binding (Figures 1–4). These rearrangements involve the closure of the pathway via a pinching motion of TM4 and TM6, the upward motion of TM3 and the dilation of the Ca2+-binding site. The global nature of these rearrangements differs greatly from those seen in the TMEM16A channel, where only TM6 moves in response to Ca2+ binding (Figure 3—figure supplement 2). The movements we observe in afTMEM16, bear striking similarities to those predicted by MD simulations for the nhTMEM16 scramblase upon removal of Ca2+. Notably, similar rearrangements are observed in the human TMEM16K scramblase, supporting their evolutionary conservation.

13Our structural and functional experiments provide detailed insights into the activation mechanism of TMEM16 phospholipid scramblases (Figures 1–4) and how these proteins alter the surrounding membrane to facilitate the transfer of lipids between leaflets (Figures 5–7). The Ca2+-free and Ca2+-bound structures of afTMEM16 define the extremes of the Ca2+-dependent activation process and suggest that opening of the lipid pathway is primarily controlled by two structural elements, TM4 and TM6 (Figure 8). Without Ca2+, TM4 and TM6 are bent, sealing the pathway from the lipid membrane (Figure 8A). The first step in activation is presumably Ca2+ binding, which facilitates the transition of TM6 to a straight conformation and its disengagement from TM4, allowing TM6 to move toward TM8 and complete the formation of the Ca2+-binding site (Figure 8B). The resulting proposed conformation is similar to that of Ca2+-bound TMEM16A, where TM6 is straight but lipid access is prevented by a bent TM4. Notably, an intermediate, Ca2+-bound conformation with a closed lipid pathway has been recently observed for the hTMEM16K scramblase. Straightening of the TM4 helix opens the lipid pathway to enable lipid translocation, as seen in Ca2+-bound afTMEM16 and nhTMEM16 (Figure 8C). To complete the gating scheme, we propose a state where TM4 is straight and TM6 is bent (Figure 8D), which would give rise to a partially opened lipid pathway. This conformation, while not yet observed experimentally, would account for the low, basal activity of reconstituted scramblases in the absence of Ca2+ (Figure 6—figure supplements 1 and 2). Within this gating mechanism, Ca2+-activated TMEM16 channels would naturally arise from scramblases via mutations that render straightening of TM4 unfavorable, while maintaining the Ca2+-dependent rearrangement of TM6 (Figure 8A,B). Indeed, in the Ca2+-free afTMEM16 scramblase, the pathway and cytosolic domain adopt conformations similar to those seen in the TMEM16A channel (Figure 8—figure supplement 1). Mutations that convert TMEM16A into a scramblase might re-enable the TM4 transition.

14Our structures of afTMEM16 scramblase/nanodisc complexes suggest that lipid scrambling is enabled by two features of the TMEM16 architecture: the dimer cavities and the lipid pathway (Figures 5 and 7). The helices lining the dimer cavity are of different heights; TM3 and TM5 of one monomer are longer than the TM1 and TM2 from the other (Figure 5E) and the upper leaflet of the membrane bends to track this sloping. The shortest helices (TM1 and TM2) are unusually rich in aromatic side chains, which anchor TM segments to membrane/solution interfaces, creating a favorable environment for both phospholipid headgroups and tails (Figure 5E, Figure 8—figure supplement 2). This architecture is conserved in the nhTMEM16 and hTMEM16K scramblases (Figure 8—figure supplement 2A,B). In contrast, in the mTMEM16A channel TM1 is longer and contains fewer aromatics, and TM3 is shorter, such that the extracellular termini of these helices sit at comparable heights within the membrane (Figure 8—figure supplement 2C,D). The lesser degree of slanting in the dimer cavity-lining helices in mTMEM16A compared to afTMEM16 is consistent with the lack of membrane bending upon reconstitution of the mTMEM16A channel in nanodiscs. The observed bending is similar in the Ca2+-bound and Ca2+-free structures presented here (Figure 5), consistent with the idea that the dimer cavity does not undergo Ca2+-dependent rearrangements. In the nhTMEM16 and mTMEM16A structures, this cavity was proposed to be packed with lipids. Indeed, while we could not identify well-defined lipids in this cavity, several densities attributable to partial acyl chains are visible between the extracellular termini of TM10 and TM2 as well as between TM3 and TM5, consistent with our hypothesis of membrane thickening around these helices (Figure 7—figure supplement 1). It is worth noting that reconstitution of unrelated non-scramblase membrane proteins into nanodisc, such as the SthK K+ channel (Figure 8—figure supplement 3), do not cause distortions in the nanodisc membrane, reinforcing the idea that these effects are specific to the afTMEM16 scramblase.

15The Ca2+-dependent conformational rearrangements undergone by the pathway-lining helices correlate with changes in the density of the membrane near this conduit. In the Ca2+-bound conformation, the pathway is open and the density between TM4 and TM6 is weaker than in the rest of the nanodisc (Figure 5). We propose that this weakening reflects a thinning of the membrane in this region, as the hydrophilic residues lining the open pathway provide an energetically unfavorable environment for the acyl tails. This would cause the lipids to rearrange, to allow their headgroups to interact with the polar pathway and locally thin the membrane. Indeed, similar thinning and rearrangements in lipid orientation have been proposed based on MD simulations and on the ability of afTMEM16 to scramble lipids conjugated to PEG molecules with diameters larger than the width of the lipid pathway. In the Ca2+-free conformation, TM4 and TM6 rearrange to close the pathway preventing lipid access (Figure 3). The membrane exposed surface of the closed pathway is hydrophobic preventing thinning. Our finding that in the presence of the Cer24:0 lipid, the pathways of afTMEM16 are open while scrambling is inhibited suggests that its activity is not only regulated by Ca2+-dependent gating of the protein, but that other factors, such as the physico-chemical properties of the membrane, are also critical determinants of lipid scrambling. We propose that modulation of bilayer properties and composition might constitute secondary layers of regulatory control for the in vivo activation of scramblases.

16Based on these observations, we propose a mechanism for lipid scrambling where the dimer cavities ‘prime’ the membrane by bending the outer membrane leaflet in opposite directions at the two sides of an open lipid pathway. This creates a membrane region that is highly curved, thin and disordered, all of which will facilitate lipid transfer between leaflets through the conduit formed by the open hydrophilic pathway (Figure 8E). In the Ca2+-free conformation of the scramblase, the closed pathway prevents lipid entry and membrane thinning (Figure 8F). Similar mechanisms, where hydrophobic mismatches induce local distortion of membranes to lower the energy barrier for lipid movement through hydrophilic grooves, could be generally applicable to other scramblases.

17afTMEM16 was expressed and purified as described. Briefly, S. cerevisiae carrying pDDGFP2 with afTMEM16 were grown in yeast synthetic drop-out medium supplemented with Uracil (CSM-URA; MP Biomedicals) and expression was induced with 2% (w/v) galactose at 30° for 22 hr. Cells were collected, snap frozen in liquid nitrogen, lysed by cryomilling (Retsch model MM400) in liquid nitrogen (3 × 3 min, 25 Hz), and resuspended in buffer A (150 mM KCl, 10% (w/v) glycerol, 50 mM Tris-HCl, pH8) supplemented with 1 mM EDTA, 5 μg ml−1 leupeptin, 2 μg ml−1 pepstatin, 100 μM phenylmethane sulphonylfluoride and protease inhibitor cocktail tablets (Roche). Protein was extracted using 1% (w/v) digitonin (EMD biosciences) at 4°C for 2 hr and the lysate was cleared by centrifugation at 40,000 g for 45 min. The supernatant was supplemented with 1 mM MgCl2 and 10 mM Imidazole, loaded onto a column of Ni-NTA agarose resin (Qiagen), washed with buffer A + 30 mM Imidazole and 0.12% digitonin, and eluted with buffer A + 300 mM Imidazole and 0.12% digitonin. The elution was treated with Tobacco Etch Virus protease overnight to remove the His tag and then further purified on a Superdex 200 10/300 GL column equilibrated with buffer A supplemented with 0.12% digitonin (GE Lifesciences). The afTMEM16 protein peak was collected and concentrated using a 50 Kdmolecular weight cut off concentrator (Amicon Ultra, Millipore).

18Liposomes were prepared as described, briefly lipids in chloroform (Avanti), including 0.4% w/w tail labeled NBD-PE, were dried under N2, washed with pentane and resuspended at 20 mg ml−1 in buffer B (150 mM KCl, 50 mM HEPES pH 7.4) with 35 mM 3-[(3-cholamidopropyl)dimethylammonio]−1- propanesulfonate (CHAPS). afTMEM16 was added at 5 µg protein/mg lipids and detergent was removed using four changes of 150 mg ml−1 Bio-Beads SM-2 (Bio-Rad) with rotation at 4°C. Calcium or EGTA were introduced using sonicate, freeze, and thaw cycles. Liposomes were extruded through a 400 nm membrane and 20 µl were added to a final volume of 2 mL of buffer B + 0.5 mM Ca(NO3)2 or 2 mM EGTA. The fluorescence intensity of the NBD (excitation-470 nm emission-530 nm) was monitored over time with mixing in a PTI spectrophotometer and after 100 s sodium dithionite was added at a final concentration of 40 mM. Data was collected using the FelixGX 4.1.0 software at a sampling rate of 3 Hz. All experiments with added ceramide lipids were carried out in the background of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1'-rac-glycerol) (POPG) at a ratio of 3:1. Ceramides tested include: N-stearoyl-D-erythro-sphingosine (Cer18:0), N-behenoyl-D-erythro-sphingosine (Cer22:0), N-lignoceroyl-D-erythro-sphinganine (Cer24:0), and N-nervonoyl-D-erythro-sphingosine (Cer24:1) all of which were tested at 1 mole% and 5 mole%. Chain length experiments were done in the background of 7PC:3 PG due to the availability of the long chain lipids. Lipids used include 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC, 16:0-18:1C), POPG (16:0-18:1C), 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC, 18:1C), 1,2-dioleoyl-sn-glycero-3-phospho-(1'-rac-glycerol) (DOPG, 18:1C), 1,2-dierucoyl-sn-glycero-3-phosphocholine (DEPC, 22:1C) and 1,2- dierucoyl-phosphatidylglycerol (DEPG, 22:1C).

19Quantification of the scrambling rate constants by afTMEM16 was determined as recently described. Briefly, the fluorescence time course was fit to the following equationwhere Ftot(t) is the total fluorescence at time t, LiPF is the fraction of NBD-labeled lipids in the inner leaflet of protein free liposomes, γ=γ’[D] where γ’ is the second order rate constant of dithionite reduction, [D] is the dithionite concentration, f0 is the fraction of protein-free liposomes in the sample, α and β are the forward and backward scrambling rate constants, respectively, and

20The free parameters of the fit are f0, α and β while LiPF and γ are experimentally determined from experiments on protein-free liposomes. In protein-free vesicles a very slow fluorescence decay is visible likely reflecting a slow leakage of dithionite into the vesicles or the spontaneous flipping of the NBD-labeled lipids. A linear fit was used to estimate the rate of this process was estimated to be L=(5.4 ± 1.6).10−5 s−1 (n > 160). For WT afTMEM16 and most mutants the leak is >2 orders of magnitude smaller than the rate constant of protein-mediated scrambling and therefore is negligible. All conditions were tested side by side with a control preparation in standard conditions. In some rare cases, this control sample behaved anomalously, judged by scrambling fit parameters outside three times the standard deviation of the mean for the WT. In these cases, the whole batch of experiments was disregarded.

21MSP1E3 was expressed and purified as described. Briefly, MSP1E3 in a pET vector (Addgene #20064) was transformed into the BL21-Gold (DE3) strain (Stratagene). Transformed cells were grown in LB media supplemented with Kanamycin (50 mg l−1) to an OD600 of 0.8 and expression was induced with 1 mM IPTG for 3 hr. Cells were harvested and resuspended in buffer C (40 mM Tris-HCl pH 78.0, 300 mM NaCl) supplemented with 1% Triton X-100, 5 μg ml−1 leupeptin, 2 μg ml−1 pepstatin, 100 μM phenylmethane sulphonylfluoride and protease inhibitor cocktail tablets (Roche). Cells were lysed by sonication and the lysate was cleared by centrifugation at 30,000 g for 45 min at 4° C. The lysate was incubated with Ni-NTA agarose resin for 1 hr at 4°C followed by sequential washes with: buffer C + 1% triton-100, buffer C + 50 mM sodium cholate +20 mM imidazole and buffer C + 50 mM imidazole. The protein was eluted with buffer C + 400 mM imidazole, desalted using a PD-10 desalting column (GE life science) equilibrated with buffer D (150 mM KCl, 50 mM Tris pH 8.0) supplemented with 0.5 mM EDTA. The final protein was concentrated to ~8 mg ml−1 (~250 μM) using a 30 kDa molecular weight cut off concentrator (Amicon Ultra, Millipore), flash frozen and stored at −80°C.

22Reconstitution of afTMEM16 in nanodiscs was carried out as follows, 3POPE:1POPG lipids in chloroform (Avanti) were dried under N2, washed with pentane and resuspended in buffer D and 40 mM sodium cholate (Anatrace) at a final concentration of 20 mM. Molar ratios of 1:0.8:60 MSP1E3:afTMEM16:lipids were mixed at a final lipid concentration of 7 mM and incubated at room temperature for 20 min. Detergent was removed via incubation with Bio-Beads SM-2 (Bio-Rad) at room temperature with agitation for 2 hr and then overnight with fresh Bio-Beads SM2 at a concentration of 200 mg ml−1. The reconstitution mixture was purified using a Superose6 Increase 10/300 GL column (GE Lifesciences) pre-equilibrated with buffer D plus 5 mM EDTA or 0.5 mM CaCl2 and the peak corresponding to afTMEM16-containing nanodiscs was collected for cryo electron microscopy analysis.

233.5 μL of afTMEM16-containing nanodiscs (7 mg mL−1) supplemented with 3 mM Fos-Choline-8-Fluorinated (Anatrace) was applied to a glow-discharged UltrAuFoil R1.2/1.3 300-mesh gold grid (Quantifoil) and incubated for one minute under 100% humidity at 15°C. Following incubation, grids were blotted for 2 s and plunge frozen in liquid ethane using a Vitrobot Mark IV (FEI). For the +C24:0 Ceramide/+Ca2+and EDTA samples as well as +Ca2+Dataset D (Supplementary file 1) micrographs were acquired on a Titan Krios microscope (FEI) operated at 300 kV with a K2 Summit direct electron detector (Gatan), using a slid width of 20 eV on a GIF Quantum energy filter and a Cs corrector with a calibrated pixel size of 1.0961 Å/pixel. A total dose of 62.61 e–/Å2 distributed over 45 frames (1.39 e–/ Å2/frame) was used with an exposure time of 9 s (200 ms/frame) and defocus range of −1.5 μm to −2.5 μm. For the +Ca2+datasets B and C (Supplementary file 1), micrographs were acquired on a Titan Krios microscope (FEI) operated at 300 kV with a K2 Summit direct electron detector with a calibrated pixel size of 1.07325 Å/pixel. A total dose of 69.97 e–/Å2 distributed over 50 frames (1.39 e–/ Å2/frame) was used with an exposure time of 10 s (200 ms/frame) and a defocus range of −1.5 μm to −2.5 μm. For all samples, automated data collection was carried out using Leginon.

24For all datasets except dataset D (in 0.5 mM Ca2+, Supplementary file 1), motion correction was performed using MotionCorr2 and contrast transfer function (CTF) estimation was performed using CTFFIND4 both via Relion 2.0.3. After manually picking ~2000 particles, the resulting 2D class-averages were used as templates for automated particle picking in Relion. The particles were extracted using a box size of 275 Å with 2xbinning and subjected to 2D classification ignoring CTFs until the first peak. For the Ca2+-free and ceramide data sets, particles selected from 2D classification (245,835 from Ca2+-free, and 185,451 from ceramide) were subjected to 3D classification using the nhTMEM16 crystal structure low-pass filtered to 40 Å as an initial model. For the 0.5 mM CaCl2 sample, datasets B and C were used combined to make dataset A; particles selected from the first round of 2D classification from each dataset were combined and subjected to a second round of 2D classification and the resulting 302,092 particles were subjected to the same 3D classification procedure. Particles from 3D classes with defined structural features (100,268 CaCl2, 70,535 Ca2+-free, 90,709 + C24:0 ceramide,) were combined and re-extracted without binning and refined without enforcing symmetry using an initial model generated in CryoSPARC.

25Initial refinement of the afTMEM16 dataset in 0.5 mM CaCl2 (Dataset B, C, which processed together form dataset A; Supplementary file 1) resulted in a map with a resolution of ~7 Å. The protein was symmetric, with the exception of the resolved portion of TM6 in each monomer (Figure 1—figure supplement 2). The complex was, however, not two-fold symmetric due to the off-center placement of the protein within the nanodisc (Figure 1—figure supplement 2). Therefore, data processing was carried out in parallel with C2 symmetry and without enforcing symmetry (in C1 symmetry). The particles from first C1 refinement were subjected to an additional round of 3D classification, using a mask that excluded the nanodisc and maintaining the particle orientations determined by the previous refinement. The best class from 3D classification with 27,948 particles was selected for further refinement and particle polishing. Masked refinement following particle polishing resulted in a 4.36 Å final map. To refine with C2 symmetry, the particles were polished and the nanodisc was removed using signal subtraction in Relion and the subtracted particles were refined using C2 symmetry, resulting in a 4.5 Å map. Using this map, a similar procedure to the C1 processing was carried out in which the best two classes from 3D classification without alignment applying a mask including the protein (37,146 particles) were selected for further refinement. Masked refinement of these classes yielded a 4.05 Å final density map. The C1 and C2 density maps were extensively compared and determined to be nearly identical except for the resolved portion of TM6 (Figure 1—figure supplement 2). The C2 map was used for model building while the C1 map was used for analysis of the afTMEM16/nanodisc complex.

26For the Ca2+-free data set (referred to as ‘0 Ca2+”), the first refinement resulted in a map with a resolution of ~6 Å. As with the +Ca2+ sample, the protein was two-fold symmetric with the exception of the resolved portion of TM6 and the overall afTMEM16/nanodisc complex was not symmetric (Figure 1—figure supplement 2). Therefore, data was processed in parallel using both C1 and C2 symmetries as described above. The C1 map was classified and the best class from 3D classification with a mask excluding the nanodisc (38,550 particles) was selected for further refinement and particle polishing. Masked refinement following particle polishing resulted in a 4.00 Å final density map. Masked, C2 refinement following polishing and signal subtraction resulted in a 3.89 Å map. The C2 map was used for model building while the C1 map was used for analysis of the afTMEM16/nanodisc complex in 0 Ca2+.

27For the data set in the presence of 0.5 mM CaCl2 and 5 mole% C24:0 Ceramide the selected particles were refined without symmetry, which resulted in a 4.2 Å resolution map. These particles were further classified in 3D with applied mask excluding the nanodiscs and maintaining angular information from the previous 3D refinement, from which two classes with 45,021 particles were selected for masked refinement which generated a final map of 3.74 Å. The resulting refinement showed that the nanodisc and the protein were C2 symmetric, therefore, further processing was completed with C2 symmetry enforced. Refinement resulted in a map with a resolution of 3.89 Å. An additional round of 3D classification was carried out using a mask excluding the nanodisc and maintaining the particle orientations determined by the previous refinement. The highest resolution class with 24,602 particles was selected for further refinement and particle polishing. Masked refinement following particle polishing resulted in a final map with a final average resolution of 3.59 Å.

28The final resolution of all maps was determined by applying a soft mask around the protein and the gold-standard Fourier shell correlation (FSC) = 0.143 criterion using Relion Post Processing (Figure 1—figure supplement 1G). BlocRes from the Bsoft program was used to estimate the local resolution for all final maps(Figure 1—figure supplement 1F).

29The two other +Ca2+datasets (B and D, Figure 5—figure supplement 2, Supplementary file 1) mentioned were processed as described above up to the first refinement step due to the limited resolution (Figure 5—figure supplement 2). For processing +Ca2+dataset D in cryoSPARC, particles were picked using Difference of Gaussian picking implemented through the appion program suite. For analysis of the final +Ca2+ dataset (A) in cyroSPARC, the particles originally picked in Relion were used. Extracted particles were classified in 2D and classes with structural features consistent with afTMEM16-nanodisc complexes were selected for ab initio model generation and classification. For each dataset, the best model and the associated particles were selected for homogeneous refinement followed by b-factor sharpening. Only the C24:0 dataset had a resolution better than 6 Å from cryoSPARC so these models were not used for analysis other than to confirm the observed membrane bending and thinning (Supplementary file 1, Figure 5—figure supplement 3). For processing in cisTEM, particles from 3D classes selected in Relion were used to create an mrc stack and imported. Several rounds of automatic refinement and manual refinement with both local and global searches were carried out. Using auto-masking and manual refinement the final resolution for the +Ca2+dataset was~5 Å and was therefore not used for an analysis other than to confirm the observed effects on the membrane (Supplementary file 1, Figure 5—figure supplement 3).

30The maps were of high quality and we built atomic models for each structure for the transmembrane domain, most of the cytosolic region and connecting loops (Figure 1—figure supplement 4), as well as identifying seven partial lipids in the +C24:0 Ceramide structure (Figure 7—figure supplement 1). The model of afTMEM16 in the presence of ceramide was built first, due to the higher resolution of the map. The crystal structure of nhTMEM16 (PDBID 4wis) was used as a starting model and docked into the density map using chimera and the jiggle fit script in COOT (https://www2.mrc-lmb.cam.ac.uk/groups/murshudov/index.html), and mutated to match the afTMEM16 sequence. The final model contains residues 13–54, 60–119, 126–259, 270–312, 316–399, 418–461, 490–597, 602–657 and 705–724 and the following residues were truncated due to missing side chain density: L43, Q49, R53, K69, K70, E94, K100, Q102, K129, H130, D137, K257, E258, L316, E461, H555, F705, K713, E714, and E717. The model was improved iteratively by real space refinement in PHENIX imposing crystallographic symmetry and secondary structure restraints followed by manual inspection and removal of outliers. The model for the +Ca2+structure was built using the refined +C24:0 Ceramide structure (PDBID 6E1O) as an initial model, that was docked into the +Ca2+density using the jigglefit script in COOT (https://www2.mrc-lmb.cam.ac.uk/groups/murshudov/index.html), manually inspected, and refined using real space refinement in PHENIX as above. The final model contains residues 13–54, 60–119, 128–258, 270–312, 316–400, 420–460, 490–594, 602–659 and 705–724, and the following residues were truncated due to missing side chain density: D17, Q49, R53, K100, Q102, E104, K129, H130, N135, D137, E164, H247, K254, K257, H460, K550, H595, R710, K713, E714, E717, and L720. The model for the 0 Ca2+ structure was built using the +Cera model as a starting point and regions where the density differed were built manually (TM3, TM4 and TM6) or rigid body refined where smaller rearrangements were observed. The final model contains residues 13–54, 60–119, 128–258, 270–312, 316–400, 420–460, 490–594, 602–659 and 705–724, and the following residues were truncated due to missing side chain density: D17, F36, R53, K70, E94, K100, Q102, E132, K241, H247, R250, K257, E310, K317, K634, K642, Y654, R704, R708, E714, E717. For all three structures, the density in the regions of TM3 and TM4 and the connecting loop (residues 274–352) was less well-defined than remainder of the structure and the model may be less accurate in this area.

31To validate the refinement, the FSC between the refined model and the final map was calculated (FSCsum) (Figure 1—figure supplement 1H). To evaluate for over-fitting, random shifts of up to 0.3 Å were introduced in the final model and the modified model was refined using PHENIX against one of the two unfiltered half maps. The FSC between this modified-refined model and the half map used in refinement (FSCwork) was determined and compared to the FSC between the modified-refined model and the other half map (FSCfree) which was not used in validation or refinement. The similarity in these curves indicates that the model was not over-fit (Figure 1—figure supplement 1H). The quality of all three models was assessed using MolProbity and EMRinger, both of which indicate that the models are of high quality (Figure 1—figure supplement 1H).

32To compare the maps resulting from C1 and C2 processing in the 0 Ca2+ and +Ca2+ structures, we calculated a difference map between the two volumes using the volume operation subtract function in chimera (Figure 1—figure supplement 2). We used ‘omit density’ to assign the placement of several lipids in the 0 Ca2+ and +C24:0 Ceramide structures and the Ca2+ ions in the +Ca2+and + Cera structures which were calculated using the ‘phenix.real_space_diff_map’ function in PHENIX. Briefly, completed models without the ligand in question were used to generate a theoretical density map which was subtracted from the experimental density map. In the subtracted map, areas of ligand density appeared as positive density.

33The gramicidin channel-based fluorescence assay for monitoring changes in lipid bilayer properties was implemented using large unilamellar vesicles (LUVs) incorporating gramicidin (gA) and loaded with 8-aminonaphthalene-1,3,6-trisulfonic acid, disodium salt (ANTS), following published protocols. Vesicles composed of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-1'-rac-glycerol (POPG) in 3:1 (mol/mol) proportion in chloroform were mixed with a gA to lipid molar ratio of 1:40,000. An additional 0, 1, or 5 mole% of C24:0 ceramide (d18:1/24:0) or C24:1 ceramide (d18:1/24:1) in chloroform were added to the gA-phospholipid mix. The lipid mixtures were dried under nitrogen to remove chloroform and further dried in vacuum overnight. Lipids were rehydrated with fluorophore-containing buffer solution (25 mM ANTS, 100 mM NaNO3, and 10 mM HEPES), vortexed for 1 min, and incubated at room temperature for at least 3 hr while shielded from light. The solutions were sonicated for 1 min at low power and subjected to six freeze-thaw cycles. The samples then were extruded 21 times with an Avanti Polar Lipids mini-extruder (Avanti) and 0.1 mm polycarbonate filter. The samples were extruded at 35–40°C. Extravesicular ANTS was removed with PD-10 desalting column (GE Healthcare). The average LUV diameter was ~130 nm, with an average polydispersity index (PDI) of 10%, as determined using dynamic light scattering. LUV stock solutions were stored at room temperature with an average shelf life of 7 days.

34The time course of the ANTS fluorescence was measured at 25oC with an SX-20 stopped-flow spectrofluorometer (Applied Photophysics), with an LED light source. Excitation was at 352 nm and emission was recorded above 450 nm with a high-pass filter; the deadtime of the instrument is ≈ 2 ms with a sampling rate of 5000 points/s. Samples were prepared by diluting the stock lipid concentration with buffer solution (140 mM NaOH and 10 mM HEPES) to 125 μM LUV; each sample was incubated for at least 10 min before several 1 s mixing reactions. Each sample was first mixed with the control buffer (no Tl+), followed by mixing with the quench solution (50 mM TINO394 mM NaNO3 and 10 mM HEPES). Experiments are conducted with two independently prepared populations of vesicles per lipid/ceramide combination and traces are analyzed in MATLAB (MathWorks).

35The fluorescence quench rate therefore was determined as described by fitting the time course to a stretched exponential function:

36F(t) denotes the fluorescence intensities at time t; β (0 < β ≤1) is a parameter that accounts for the dispersity of the vesicle population; and τo is a parameter with dimension of time. F(0), F(∞), β and τo were determined from a nonlinear least squares fit of Eq. 1 to each individual fluorescence trace, and the quench rate was determined:evaluated at t = 2 ms. The ceramide-induced changes in the quench rate then was evaluated aswhere the subscripts ‘ceramide’ and ‘cntrl’ denote the rates observed in the presence and absence of ceramide.

37The three-dimensional cryo-EM density maps of the calcium-bound, calcium-free, and Ca2+-bound in the presence of C24:0 Ceramide afTMEM16/nanodisc complexes have been deposited in the Electron Microscopy Data Bank under accession numbers EMD-8948, EMD-8931, and EMDB-8959, respectively. The deposition includes corresponding masked and unmasked maps, the mask used for the final FSC calculation, and the FSC curves. Coordinates for the models of the calcium-bound, calcium-free, and Ca2+-bound ceramide inhibited states have been deposited in the Protein Data Bank under accession numbers 6E0H, 6DZ7, and 6E1O respectively. All other data are available from the corresponding author upon reasonable request.