The calcium-activated chloride channel TMEM16A is a ligand-gated anion channel that opens in response to an increase in intracellular Ca2+ concentration1,2,3. The protein is broadly expressed4 and contributes to diverse physiological processes, including transepithelial chloride transport and the control of electrical signalling in smooth muscles and certain neurons5,6,7. As a member of the TMEM16 (or anoctamin) family of membrane proteins, TMEM16A is closely related to paralogues that function as scramblases, which facilitate the bidirectional movement of lipids across membranes8,9,10,11. The unusual functional diversity of the TMEM16 family and the relationship between two seemingly incompatible transport mechanisms has been the focus of recent investigations. Previous breakthroughs were obtained from the X-ray structure of the lipid scramblase of the fungus Nectria haematococca (nhTMEM16)12,13, and from the cryo-electron microscopy structure of mouse TMEM16A at 6.6 Å (ref. 14). Although the latter structure disclosed the architectural differences that distinguish ion channels from lipid scramblases, its low resolution did not permit a detailed molecular description of the protein or provide any insight into its activation by Ca2+. Here we describe the structures of mouse TMEM16A at high resolution in the presence and absence of Ca2+. These structures reveal the differences between ligand-bound and ligand-free states of a calcium-activated chloride channel, and when combined with functional experiments suggest a mechanism for gating. During activation, the binding of Ca2+ to a site located within the transmembrane domain, in the vicinity of the pore, alters the electrostatic properties of the ion conduction path and triggers a conformational rearrangement of an α-helix that comes into physical contact with the bound ligand, and thereby directly couples ligand binding and pore opening. Our study describes a process that is unique among channel proteins, but one that is presumably general for both functional branches of the TMEM16 family.
We purified mouse TMEM16A (mTMEM16A) in the absence and presence of Ca2+, confirmed its functional integrity by reconstitution (Extended Data Fig. 1a, b) and determined structures for each condition by single-particle cryo-electron microscopy (cryo-EM) (Fig. 1, Extended Data Figs 1 and 2). The resolutions of the electron microscopy maps of the Ca2+-bound and Ca2+-free protein are 3.75 and 4.06 Å, respectively, with large parts of the transmembrane domain being substantially better resolved (Extended Data Figs 1g, h and 2f, g). Both datasets are of high quality and define the architecture of the channel in distinct states (Fig. 1, Extended Data Figs 3, 4, 5, Supplementary Videos 1 and 2). The refined Ca2+-bound and Ca2+-free structures display conformations that are very similar across most of the protein, except for changes in the region surrounding the Ca2+-binding sites (Fig. 1, Extended Data Figs 3a and 6a, b, Supplementary Video 3).
The Ca2+-bound structure brings into focus molecular details that have been postulated on the basis of low-resolution data14 (Fig. 1). The protein is a homodimer that resembles the lipid scramblase nhTMEM16, with respect to its general architecture (Extended Data Fig. 6c, d). Each subunit contains cytosolic N- and C-terminal domains, a transmembrane unit consisting of ten membrane-spanning α-helices and an extracellular component (Extended Data Fig. 5). The comparably small dimer interface, burying only 1,006 Å2 (1.3%) of the combined molecular surface, is formed exclusively by interactions between residues located in the outer part of α-helix 10 (α10) (Extended Data Fig. 5d). Below the contact region, the separation of both symmetry-related helices opens up a large membrane-accessible cavity that is presumably filled with lipids. On the extracellular side, long stretches of amino acids that connect α-helices α1–α2, α5–α6 and α9–α10 interact to form a folded domain, consisting of coil regions with interspersed elements of secondary structure that are stabilized by four disulfide bridges (Extended Data Figs 3b and 5e). On the intracellular side, the N-terminal domain displays a ferredoxin-like fold that resembles the equivalent part in the scramblase nhTMEM16 (ref. 13, Extended Data Fig. 5f), whereas the preceding 115 amino acids, which are only present in the a isoforms of the channel1, are not defined in the mTMEM16A structure. Close to the dimer axis, the N-terminal domain participates in an intra-subunit interaction with the only structured part of the cytoplasmic C terminus, helix Cα1 (Fig. 1, Extended Data Figs 5b and 6a). The same location comprises the poorly defined, and thus presumably inherently flexible, extended α2–α3 loop that also contains a short α-helical region (α2′, Extended Data Figs 1f, 2d, 4a and 5a, b, g). This loop carries an insertion in c-splice variants1, with mutations that influence the activation properties of the channel15,16,17 (Extended Data Fig. 5a, g).
The ion conduction pore of TMEM16A is contained within each subunit and is formed by α3–α7 (Fig. 2a). Its shape resembles an hourglass, with a small extracellular and a large intracellular vestibule bridged by a narrow neck region that is about 20 Å long (Fig. 2a–d). Because the pore is continuous, the observed conformation is presumably close to a conducting state. With a diameter of only 2.5 Å at its constriction, however, it would have to expand to accommodate permeating anions, which have sizes that range between 3.6 and 4.1 Å in case of Cl− and I−, respectively (Extended Data Fig. 7a, b). Such dilation could be attained by the movement of the inherently flexible α3 (Extended Data Figs 1f and 2d, e) and local changes in side-chain conformations. Both cone-shaped vestibules are highly hydrophilic and are lined by ionizable residues (Fig. 2b, d). The excess of basic amino acids confers a positive electrostatic environment throughout the pore (Extended Data Fig. 8a, b), thus lowering the energy barrier for anion conduction14. On one side the extracellular vestibule is delimited by the α1–α2 and α5–α6 loops, and on the other side by the mobile region that connects α3 and α4 (Fig. 2a). The α5–α6 loop contains residues that were previously proposed to be part of the pore-lining region18,19 (Fig. 2b). In contrast to the vestibules, the narrow neck is amphiphilic and does not contain charged amino acids (Fig. 2c). It is confined by α3–α6 on its outer half and α4–α7 on its inner half (Fig. 2a, c). The narrow conduit is shielded from the membrane by extended interactions between α4 and α6, which are detached in the scramblase nhTMEM16 (ref. 14, Extended Data Figs 6c and 7c). Below the neck, α4 and α6 separate and thereby generate a spacious intracellular vestibule (Fig. 2a, d, Extended Data Fig. 7a, b). This separation opens up a gap that might be partly accessible to the lipid bilayer and provides space for the rearrangement of α6 in the Ca2+-free state (Extended Data Fig. 7a, d, g). Conserved residues of α6–α8 form a Ca2+-binding site at the surface of the wide intracellular vestibule (Fig. 2e). This site is well-defined in our experimental dataset, except for the carboxylate groups of interacting acidic residues, which are frequently not detected in cryo-EM densities (Extended Data Fig. 4b). Distinct peaks reveal the presence of two Ca2+ ions residing at equivalent locations to those in nhTMEM16 (ref. 12, Fig. 2e). Analogously, both ions are coordinated by five acidic residues (Fig. 2e), all of which have been previously described as affecting Ca2+ activation in mTMEM16A12,18,20. In contrast to nhTMEM16, in which a single asparagine in α6 contributes to the coordination of the upper Ca2+, the corresponding Ca2+ ion in mTMEM16A interacts with three polar residues (N650, N651 and, slightly further apart, N730; Fig. 2e), and mutation of these residues to alanine lowers the potency of Ca2+ (ref. 12, Extended Data Fig. 8e–g). The unpaired main-chain carbonyl of Q646 provides an additional interaction, at a somewhat larger distance (Fig. 2e, Extended Data Fig. 4b). Comparing the channels TMEM16A and TMEM16B to other family members reveals an insertion of a single residue in this region (Extended Data Fig. 5c), which leads to a partial unwinding of α6. The energetic penalty caused by the π-helix bulge that is introduced by this insertion is probably stabilized by interactions with the bound ligand.
Cryo-EM data obtained for mTMEM16A purified in the absence of Ca2+ define the structure of the protein in a ligand-free state (Extended Data Figs 2 and 3). In this structure, substantial changes (compared to the Ca2+-bound conformation) are confined to the pore region (Figs 1, 3a–c, Extended Data Fig. 6a, b, Supplementary Videos 3 and 4). In contrast to the ligand-bound structure, no density attributable to ions is found in the Ca2+-binding site (Extended Data Fig. 4c). Although comparably small differences are also observed in the narrow neck region for α3, α4 and α6, the most pronounced conformational changes concern the inner half of α6 (Extended Data Figs 6a, b and 7d–g). On release of Ca2+ from its binding site, this part of the helix has detached from its position in the Ca2+-bound state, in which it closely interacted with α7 (Fig. 3a, b, Extended Data Fig. 7g). Its movement towards α4 closes the gap between both helices that is observed in the Ca2+-bound state at the intracellular part of the pore. At the same time, it opens up a large aqueous access pathway to the Ca2+-binding site (Fig. 3b, Extended Data Fig. 7g). The rearrangement of α6 can be described by a hinge movement around G644, along with the unfolding of α6 at positions proximal to G656, which moves approximately 15 Å (Fig. 3a, Extended Data Fig. 6b, Supplementary Video 4). The remainder of the helix is poorly defined in the cryo-EM density, and is therefore presumably mobile (Extended Data Fig. 4c). The movement around G644 is accompanied by the relaxation of the strained π-helix and the formation of a canonical α-helix (Fig. 3c). Although the movement of α6 does not lead to an occlusion of the intracellular vestibule, it narrows the neck of the pore and therefore potentially affects the entry of permeating ions (Extended Data Fig. 7d–g). Notably, the absence of Ca2+ renders the electrostatic environment in this region negative, which in turn increases the barrier for anion conduction (Fig. 3d, Extended Data Fig. 8c, d).
To further characterize the role of G644 for channel activation, we replaced it with amino acids with reduced conformational flexibility and investigated activation by electrophysiology. As the residue is not directly involved in ligand interactions, we expect mutations at this position to disturb the equilibrium between open and closed conformations21,22. This is observed for mTMEM16A(G644A) and mTMEM16A(G644P), which both caused a left-shift in the half-maximal effective concentration (EC50) (Fig. 4a, b, Extended Data Fig. 9a, b). In the case of mTMEM16A(G644P), the constitutive current observed in the absence of Ca2+ indicates basal activity of the mutant, which is not observed in macroscopic recordings of the wild-type protein (Fig. 4b, Extended Data Figs 8e and 9b). These basal currents are strongly outwardly rectifying (Extended Data Fig. 9c), consistent with an electrostatic barrier at the location of the vacant binding site that impedes anion permeation from the cytoplasm (Fig. 3d, Extended Data Fig. 8d). The movement of α6 and the accompanying change in its environment is also consistent with the functional behaviour of mutants with alterations at positions that are not involved in Ca2+ binding and that are distant from the hinge. Although shifts in the EC50 were small for mutations of G656, which forms the pivot for the unfolding of the subsequent part of the helix, decreased potency was observed for mutations of the conserved P658, which is located a helix-turn closer to the cytoplasm (Fig. 4c, d, Extended Data Figs 5c, 9d–g). Next, we investigated whether the transition to a closed state on the release of Ca2+ impedes access to the intracellular vestibule, by monitoring the state-dependent modification of residues by thiol-reactive methanethiosulfonate (MTS) reagents. The time dependence of modification of a cysteine mutant of K588, which is located deep in the vestibule close to the intracellular entrance to the neck, is very similar in the presence or absence of Ca2+. This indicates that the accessibility of this position is not substantially affected by channel closure, as predicted from the structure (Fig. 4e–g, Extended Data Figs 7g and 9h, i). By contrast, the corresponding alteration of S592, which is located one helix-turn towards the extracellular side (Fig. 4e, h, Extended Data Fig. 9i–l), is not modified, which reflects the inaccessibility of the site to MTS reagents owing to the restricted geometry of the neck. Nevertheless, we expect changes in this narrow region to have an important role in gating, as is shown by mutation of a pore-lining residue (mTMEM16A(I550A)), which increases the potency of Ca2+ and gives rise to basal currents (Fig. 4e, i, Extended Data Fig. 9m).
Collectively, our data provide detailed insight into the mechanism of activation of mTMEM16A, which occurs independently in each subunit of the dimeric protein23,24. During this process, the inner half of α6 acts as a gating element (Fig. 5). In the closed channel, the helix is in a relaxed state, allowing Ca2+ from the cytoplasm to access the vacant ligand-binding site. The binding of (presumably) two Ca2+ ions to four acidic residues on α7 and α8 precedes pore opening (Figs 3b and 5a). The resulting high positive charge density creates an interaction platform that attracts residues located on the distant α6. This is sufficient to shift the conformational equilibrium of the helix to a position in which it makes stabilizing contacts with α7 and α8, as defined by the Ca2+-bound structure. The concurrent reorientation of E654 on interaction with the bound ligand causes the formation of a π-helix, which is stabilized by the interaction of the unpaired backbone carbonyl of Q646 with the upper Ca2+ ion (Figs 3c and 5b). The energetic penalty associated with this conformational transition is reflected by the stability of the closed state in the absence of ligand, as evidenced by the undetectable basal activity of the channel. The conformational transition during activation closes the aqueous path to the Ca2+-binding site, consistent with the stronger voltage dependence observed during channel closing compared to channel opening15,25. In contrast to the dynamic lower part, the upper pore region forms a narrow conduit for which structural changes appear to be comparably small during activation. Nevertheless, we think that small conformational changes within the neck that constrict the channel (Extended Data Fig. 7e–g), combined with the presence of an electrostatic barrier in the absence of Ca2+ (Fig. 3d, Extended Data Fig. 8d), contribute to the impediment of anion permeation in the ligand-free state. In the open state, the neck region provides a favourable environment for selective ion permeation. Anions, which have presumably shed most of their hydration shell, interact with the residues of an amphiphilic pore, whereas positively charged side-chains located at both ends of the neck contribute to an attractive electrostatic environment (Extended Data Fig. 8b) without creating localized strong binding sites14,26, and thereby account for the strong selectivity for anions over cations14,23,24,27. In this narrow conduit, polar residues compensate for the removal of coordinating water molecules, whereas hydrophobic residues increase the energetic penalty for smaller ions; this is consistent with the observed lyotropic permeability sequence, which favours larger anions26,28. Ion–protein interactions in the neck might also account for the observed stabilization of the open state by permeating ions15,29,30. Owing to the conservation of the Ca2+-binding site and of other sequence elements located in α6, it can be assumed that activation proceeds by a similar mechanism in lipid scramblases of the TMEM16 family11,12,13.
In summary, our work describes an activation process for TMEM16A that is distinct from other ligand-gated ion channels. Instead of binding to a separate structural unit, the ligand in TMEM16A directly interacts with the pore region, thereby immobilizing a transmembrane helix involved in gating and influencing ion conduction by altering the electrostatic properties of the pore.
The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.
Protein expression and purification
HEK293 cells constitutively expressing mTMEM16A(ac) were cultured and collected as described previously14. To improve the sample quality, the duration of purification was reduced from 27 h to 12 h. All steps were carried out at 4 °C. Cells from 23 l of adhesion culture (in 2,300 10-cm dishes) were resuspended in 230 ml of buffer A (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM CaCl2) containing 1% digitonin (AppliChem) and protease inhibitors (cOmplete, Roche). Membranes were solubilized by gentle agitation in batch for 2 h. The sample was centrifuged at 22,000g for 30 min and subsequently filtered through a 5-μm filter (Minisart, Sartorius) to remove the insoluble fraction. The supernatant was applied to 3 ml of streptavidin UltraLink resin (Pierce, Thermo Fisher Scientific) and protein was bound to the beads by incubation in batch for 1.5 h. The resin was subsequently washed with 60 column volumes of buffer B (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM CaCl2, 0.12% digitonin, Calbiochem). Protein was eluted with 3 column volumes of buffer B with 3 mM biotin (Sigma). Protein-containing fractions were deglycosylated for 2 h with PNGaseF, subsequently concentrated (100 kDa cut-off, Amicon Ultra) and applied to a Superose 6 column (GE-Healthcare) equilibrated in buffer B. The fractions from the main peak were pooled and concentrated yielding 70 μg of pure mTMEM16A. The sample at a final concentration of 3.5 mg ml−1 was immediately used for cryo-EM sample preparation.
The purification of mTMEM16A in the absence of Ca2+ was performed as described above, except with the following buffer compositions: buffer A–Ca2+ (20 mM HEPES pH 7.5, 150 mM NaCl, 10 mM EGTA) and buffer B–Ca2+ (20 mM HEPES pH 7.5, 150 mM NaCl, 5 mM EGTA, 0.12% digitonin, Calbiochem). All the buffers were prepared using calcium-free water for molecular biology (Millipore). Only plastic materials were used throughout the purification, and when needed pre-cleaned with 10 mM EDTA pH 8.0 to minimize Ca2+ contamination. Purification from 19 l of adhesion culture (in 1,900 10-cm dishes) yielded 60 μg of pure mTMEM16A. The sample at a final concentration of 3.3 mg ml−1 was immediately used for cryo-EM sample preparation.
Reconstitution and transport assays
For reconstitution, mTMEM16A was purified from 10 l of adhesion culture in the absence of Ca2+ as described above, except that the concentration of EGTA was decreased to 1 mM at the gel-filtration step. The reconstitution was performed as described31. In brief, the liposomes were prepared as a 3:1 mixture of Escherichia coli polar lipids and egg PC (Avanti polar lipids), and resuspended in reconstitution buffer (100 mM KCl, 10 mM HEPES pH 7.4, 1 mM EGTA). After detergent destabilization, the purified protein was added to the liposome suspension at a ratio of 1:75 (w/w). For the control sample, the equivalent amount of the protein buffer was added to the liposome suspension. The detergent was removed by addition of SM2 bio-beads (Bio-Rad). The ion flux assay was performed similarly to previous reports32. After collection, the buffer inside the liposomes was exchanged by freeze–thawing to assay buffer A (100 mM Na2SO4, 10 mM HEPES pH 7.4, 1 mM EGTA) containing either 0 or 1 μM of free Ca2+. The liposomes were flash-frozen and stored at −80 °C at a concentration of 10 mg ml−1 until further use. On the day of the measurement, the liposomes were sonicated and diluted 50-fold into assay buffer B (10 mM HEPES pH 7.4, 125 mM NaCl, 2 μM ACMA, 1 mM EGTA) containing either 0 or 1 μM free Ca2+. Fluorescence intensity was recorded at 5-s intervals on a TECAN Infinite M1000 plate reader, with λex and λem of 410 nm and 490 nm, respectively. The ion transport was initiated after 90 s by addition of the proton ionophore CCCP from a 100-μM stock in DMSO to a final concentration of 1 μM. The data were normalized and displayed as F/Fmax.
Electron microscopy sample preparation and imaging
Two μl of purified mTMEM16A at a concentration of 3.5 mg ml−1 in the presence of 0.5 mM CaCl2, or at a concentration of 3.3 mg ml−1 in the absence of calcium ions, was pipetted onto glow-discharged 200- and 400-mesh gold Quantifoil R1.2/1.3 holey carbon grids (Quantifoil). Grids were blotted for 2–4 s with a blotting force of 1 at 15 °C and 100% humidity, and flash-frozen in liquid ethane using an FEI Vitrobot Mark IV (FEI). Cryo-EM data were collected on a 300-kV FEI Titan Krios electron microscope using a post-column quantum energy filter (Gatan) with a 20-eV slit and a 100-μm objective aperture (Extended Data Figs 1, 2 and 10). Data were collected in an automated fashion using SerialEM33 on a K2 Summit detector (Gatan). For the dataset in the presence of calcium ions, cryo-EM images were collected at a pixel size of 0.5375 Å in super-resolution mode, a defocus range between −0.5 and −3.0 μm, an exposure time of 10 s and a sub-frame exposure time of 125 ms (80 frames) with an approximate electron dose of 1 e− per Å2 per frame. To improve data quality, an extensive effort was made for this dataset to screen different areas within the grid and only regions that provided an estimated resolution of the contrast transfer function (CTF) fit better than 4 Å were selected for data collection. The dataset in the absence of calcium ions was obtained at a pixel size of 0.6825 Å in super-resolution mode, a defocus range between −0.5 and −3.9 μm, an exposure time of 15 s and a sub-frame exposure time of 150 ms (100 frames) with an electron dose at the specimen level of 0.75–0.8 e− per Å2 per frame. The total accumulated dose on the specimen level for both datasets was approximately 80 e− per Å2.
For the dataset collected in the presence of calcium ions, a total of 4,342 dose-fractionated super-resolution images were recorded, 2 × 2 down-sampled by Fourier cropping (final pixel size 1.075 Å) and subjected to motion correction and dose-weighting of frames by MotionCor234. The CTF parameters were estimated on the movie frames by ctffind4.135. Images showing a strong drift, defocus higher than −3.0 μm or a bad CTF estimation were discarded, resulting in 2,997 images used for further analysis with the software package RELION2.1b136. Particles were picked automatically using 2D-class averages from the previously obtained TMEM16A cryo-EM map (EMD-3658) as reference14, providing an initial set of 629,679 particles. After extraction with a box size of 300 pixels, false positives were eliminated manually or through a first round of 2D classification, resulting in 368,162 particles that were further subjected to several rounds of 2D classification to remove particles belonging to low-abundance classes. The remaining 252,577 particles were sorted during 3D classification, a C2 symmetry was imposed and the low-resolution TMEM16A cryo-EM map was used as initial model. The best class, comprising 147,368 particles from a total of 2,012 images, was subjected to auto-refinement and particle polishing in RELION, with a running average window of 5, a s.d. of 1 pixel for translations and 200 pixels for particle distance. The final polished and auto-refined map used for model building had a resolution of 4.6 Å before masking and 3.75 Å after masking, and was sharpened using an isotropic b-factor of −106 Å2 (Extended Data Figs 1 and 10). For the dataset in the absence of calcium ions, a similar workflow for image processing was applied. From a total of 4,738 images (final pixel size 1.365 Å), 1,349,821 particles were extracted with a box size of 256 pixels after auto-picking and an initial round of 2D classification. Initial rounds of 2D and 3D classification resulted in a set of 467,286 particles, which were subjected to particle polishing and used for final rounds of 3D classification. The final polished and auto-refined map was calculated from 195,465 particles derived from 4,726 images with a resolution of 4.86 Å before masking and 4.06 Å after masking, and was sharpened using an isotropic b-factor of −116 Å2 (Extended Data Figs 2 and 10). To identify flexible regions, both final maps (in the presence and absence of calcium ions) were subjected to a final round of 3D classification. In both cases, three predominant classes were identified that were virtually identical, except for the density corresponding to transmembrane α3 and its connecting loops to transmembrane α2 and α4 (Extended Data Figs 1f and 2d, e). As similar results were obtained for both datasets, this region appears to be inherently flexible. The maps had a final resolution of 4.03 Å, 4.35 Å and 4.19 Å for the classes in the presence of calcium ions, and 4.48 Å, 4.85 Å and 4.36 Å for the classes in the absence of calcium ions. Local resolution estimates were calculated by BlocRes from the Bsoft software package37. All resolutions were estimated using the 0.143 cut-off criterion38 with gold-standard Fourier shell correlation (FSC) between two independently refined half maps39. During post-processing, the approach of high-resolution noise substitution was used to correct for convolution effects of real-space masking on the FSC curve40.
Model building and refinement
The model of the Ca2+-bound form of mTMEM16A was built in COOT41 using the structure of nhTMEM16 (PDBID 4WIS) as a template. The electron density was of high quality and allowed interpretation with a model consisting of residues 117–130, 165–259, 267–466, 488–668 and 683–910 (Extended Data Figs 3 and 4). Compared to most parts of the structure, the density of the α2–α3 loop (residues 441–466), the first part of α3 (residues 488–497) and the α3–α4 loop (residues 523–534) is not as well-defined and the structure may therefore be less accurate in these regions. The model was improved by real-space refinement in Phenix42, in which secondary structure elements and the symmetry between both subunits of the dimeric protein were constrained. Reciprocal space refinement was performed with Refmac43,44 incorporated in CCPEM45. Coordinates were manually edited in COOT after each refinement cycle. The structure of the Ca2+-free form of mTMEM16A was built using the refined Ca2+ bound structure as a starting model, and by rebuilding regions that changed its conformation in COOT. Major conformational changes were restricted to the second half of α6. The structure was subsequently refined in Phenix and Refmac as described above. Both models are of high quality (Extended Data Fig. 10a). For validation of the refinement, FSCs (FSCsum) between the refined model and the final map were determined (Extended Data Fig. 10b). To monitor the effects of potential over-fitting, random shifts (up to 0.3 Å) were introduced into the coordinates of the final model, followed by refinement with Refmac against the first unfiltered half-map. The FSC between this shaken–refined model and the first half-map used during validation refinement is termed FSCwork, and the FSC against the second half-map, which was not used at any point during refinement, is termed FSCfree. The marginal gap between the curves describing FSCwork and FSCfree indicate no over-fitting of the model (Extended Data Fig. 10b).
The electrostatic potential was calculated by solving the linearized Poisson–Boltzmann equation in CHARMM46,47 on a 200 Å × 140 Å × 190 Å grid (1 Å grid spacing) followed by focusing on a 135 Å × 100 Å × 125 Å grid (0.5 Å grid spacing). Partial protein charges were derived from the CHARMM36 all-hydrogen atom force field. Hydrogen positions were generated in CHARMM. The protein was assigned a dielectric constant (ε) of 2. Its transmembrane region was embedded in a 30-Å-thick slab (ε = 2), representing the hydrophobic core of the membrane, and two adjacent 15-Å-thick regions (ε = 30), representing the headgroups. This region contained a cylindrical hole around the water-filled intracellular vestibule of one subunit and was surrounded by an aqueous environment (ε = 80) containing 150 mM of monovalent mobile ions.
The wild-type mTMEM16A(ac) construct used in this study has been described previously14,23. mTMEM16A(ac) cDNA containing only the six extracellular cysteines indispensable for channel function18 (6C) was cloned into the same vector using the FX cloning system48. Mutations were introduced using a modified QuikChange method49 and were verified by sequencing.
HEK293T cells (ATCC CRL-1573) were transfected with 3 μg DNA per 6-cm Petri dish using the calcium phosphate co-precipitation method, and were used within 24–96 h after transfection. Inside-out patches were excised from cells expressing the mTMEM16A construct of interest after the formation of a gigaohm seal. Seal resistance was typically 4–8 GΩ or higher. Patch pipettes were pulled from borosilicate glass capillaries with an outer diameter of 1.5 mm and an inner diameter of 0.86 mm (Sutter) and were fire-polished using a microforge (Narishige). Pipette resistance was typically 3–8 MΩ when filled with recording solutions. Voltage-clamp recordings were performed using Axopatch 200B and Digidata 1550 (Molecular Devices). Analogue signals were filtered at a cut-off frequency of 5 kHz through the in-built 4-pole low-pass Bessel filter of Axopatch 200B and were sampled at 20 kHz. Data acquisition was performed using Clampex 10.6 (Molecular Devices). Solution exchange was performed using a theta glass pipette mounted on a high-speed piezo stepper (Siskiyou) for concentration–response experiments. For accessibility experiments, a three-barrelled glass pipette mounted on the SF-77B fast-step stepper (Warner Instruments) was used. The position of the steppers for solution delivery was controlled via analogue voltage signals, programmed in Clampex 10.6. Liquid junction potential was not corrected as this was found to be consistently negligible given the ionic composition of the recording solutions14,23.
Recordings were performed under symmetrical ionic conditions. Stock solution with Ca2+–EGTA contained 150 mM NaCl, 5.99 mM Ca(OH)2, 5 mM EGTA and 10 mM HEPES at pH 7.40, which corresponds to a free Ca2+ concentration of 1 mM. ‘EGTA only’ stock solution contained 150 mM NaCl, 5 mM EGTA and 10 mM HEPES at pH 7.40. Free Ca2+ concentrations were adjusted by mixing the stock solutions at the required ratios to yield the corresponding total Ca2+ and EGTA concentrations, which were calculated using the WEBMAXC program (http://web.stanford.edu/~cpatton/webmaxcS.htm).
For concentration–response experiments, current responses at different free Ca2+ concentrations were recorded using a rundown correction protocol23. In brief, a reference concentration of Ca2+ was delivered before and after the test pulse separated by an identical time interval. The response to the test pulse was measured as the ratio of the average of the responses to the pre- and post-pulses. For constructs that did not display noticeable basal activity, background current recorded in the absence of Ca2+ (5 mM EGTA) was subtracted before analysis. This was not possible for the mTMEM16A(G644P) and the mTMEM16A(I550A) constructs, as they display basal activity. The current responses of these mutants were thus analysed without background subtraction. The presence of minimal leak current can be inferred by the strong rectification of the basal current (Extended Data Fig. 9c), which resulted in inward current with an absolute amplitude of typically less than 30 pA in the dataset analysed.
To analyse the accessibility of the intracellular vestibule, we incubated inside-out patches with the small thiol-reactive reagent MTSEA (positively charged) or MTSES (negatively charged) in the presence or absence of calcium, and subsequently monitored changes in the current–voltage relationships of activated channels in the presence of calcium. MTSEA bromide and sodium MTSES (Anatrace) were dissolved in anhydrous DMSO (Sigma) at 1 M, aliquoted and stored under argon at −20 °C in a desiccated container. Immediately before the experiment, stock MTS reagents were diluted to the required concentration in the NaCl-based recording solutions described above, with 1 mM Ca2+ or no Ca2+ (5 mM EGTA). MTSEA was used within 15 min and MTSES, which has a reported half-life of 370 min50, was used within 2 h. Modification experiments were performed in a similar manner to a described protocol51. MTS reagents, or control solutions without MTS reagents, were applied repeatedly to inside-out patches as 2-s pulses at regular time intervals in the absence or presence of Ca2+. The progress of modification was subsequently monitored at 1 mM Ca2+ by recording I–V relations using a voltage ramp protocol. The I–V relation of the background current recorded in the absence of Ca2+ immediately before the test pulse was subtracted before analysis. In accessibility experiments, we made use of the strong outward rectification of the mutants K588C and K588Q/S592C, which disappears on reaction with MTSEA, as the covalently bound reagent compensates for the lost positive charge in the mutants (Extended Data Fig. 9h, i). Rectification was quantified as the ratio of the absolute current value at −100 mV to that at 120 mV (rectification index). Current rundown correction was not required as the readout is already normalized. In the case of MTSES (where the absolute magnitude was also measured), the rundown time course was monitored by applying ten pulses of control solution before switching to a solution that contained MTSES. Concentration–response data were fitted to a Hill equation. The time course of MTSEA modification was fitted to a sum of two exponentials. Data were analysed using Clampfit 10.6 (Molecular devices), Excel (Microsoft) and Prism 5 (GraphPad) and are presented as average ± s.e.m. unless otherwise indicated.
Statistics and reproducibility
No statistical methods were used to predetermine sample size. Electrophysiology data were repeated multiple times from different transfections with very similar results. Conclusions of experiments were not changed on inclusion of further data. In the case of constructs not showing basal activity, leaky patches were discarded. In case of mutants showing basal activity, basal currents were strongly rectifying.
The three-dimensional cryo-EM density maps of calcium-bound and calcium-free mTMEM16A have been deposited in the Electron Microscopy Data Bank under accession numbers EMD-3860 and EMD-3861, respectively. The deposition includes corresponding maps calculated with higher b-factors, both half-maps and the mask used for the final FSC calculation. Coordinates for the models of the calcium-bound and calcium-free state have been deposited in the Protein Data Bank under accession numbers 5OYB and 5OYG, respectively. All other data are available from the corresponding author upon reasonable request.
Electron Microscopy Data Bank
Protein Data Bank
We thank O. Medalia and M. Eibauer, the Center for Microscopy and Image Analysis (ZMB) of the University of Zurich, and the Mäxi foundation for support and access to electron microscopes. J. D. Walter is acknowledged for comments on the manuscript and all members of the Dutzler laboratory for help at various stages of the project. This research was supported by a grant from the European Research Council (number 339116, AnoBest). C.P. was supported by a postdoctoral fellowship (Forschungskredit) of the University of Zurich.
Extended data figures
cryo-EM density map of the mTMEM16A ion channel obtained in presence of calcium ions with the modelled structure superimposed. Only one subunit is shown and the two bound Ca2+ are coloured in blue.
cryo-EM density map of the mTMEM16A ion channel obtained in absence of calcium ions with the modelled structure superimposed. Only one subunit is shown.
Superimposition of the mTMEM16A ion channel structure in the Ca2+-bound (green) and the Ca2+-free (violet) state.
Morph between the Ca2+-bound and the Ca2+-free state (violet) superimposed on the structure in presence of calcium (green, Ca2+ shown as blue spheres). The structure is viewed from within the membrane rotated by approximately 70 degrees about the y axis compared to Figure 1.
About this article
Nature Communications (2018)