Structural basis for anion conduction in the calcium-activated chloride channel TMEM16A

The calcium-activated chloride channel TMEM16A is a member of a conserved protein family that comprises ion channels and lipid scramblases. Although the structure of the scramblase nhTMEM16 has defined the architecture of the family, it was unknown how a channel has adapted to cope with its distinct functional properties. Here we have addressed this question by the structure determination of mouse TMEM16A by cryo-electron microscopy and a complementary functional characterization. The protein shows a similar organization to nhTMEM16, except for changes at the site of catalysis. There, the conformation of transmembrane helices constituting a membrane-spanning furrow that provides a path for lipids in scramblases has changed to form an enclosed aqueous pore that is largely shielded from the membrane. Our study thus reveals the structural basis of anion conduction in a TMEM16 channel and it defines the foundation for the diverse functional behavior in the TMEM16 family. DOI: http://dx.doi.org/10.7554/eLife.26232.001


Introduction
Calcium-activated chloride channels (CaCCs) are important constituents of diverse physiological processes, ranging from epithelial chloride secretion to the control of electrical excitability in smooth muscles and neurons (Hartzell et al., 2005;Huang et al., 2012;Oh and Jung, 2016;Pedemonte and Galietta, 2014). These ligand-gated ion channels are activated upon an increase of the intracellular Ca 2+ concentration as a consequence of cellular signaling events. Although CaCC function can be accomplished by unrelated protein architectures (Kane Dickson et al., 2014, Kunzelmann et al., 2009, the so far best-characterized processes are mediated by the protein TMEM16A (Caputo et al., 2008;Schroeder et al., 2008;Yang et al., 2008). TMEM16A is a member of the large TMEM16 family of membrane proteins, also known as anoctamins (Yang et al., 2008). The family is exclusively found in eukaryotes and contains 10 paralogs in mammals that all share considerable sequence homology (Milenkovic et al., 2010) (Figure 1-figure supplement 1). Although it was initially anticipated that all TMEM16 proteins would function as anion channels (Hartzell et al., 2009;Tian et al., 2012;Yang et al., 2008), it is now generally accepted that only two family members (the closely related TMEM16A and B) are ion channels (Pifferi et al., 2009;Scudieri et al., 2012), whereas most others work as lipid scramblases, which catalyze the passive and bidirectional diffusion of lipids between the two leaflets of a phospholipid bilayer (Brunner et al., 2016;Malvezzi et al., 2013;Suzuki et al., 2013Suzuki et al., , 2010Hartzell, 2017, 2016).
The TMEM16 family shows a new protein fold, as revealed by the structure of the fungal homologue nhTMEM16, which functions as lipid scramblase (Brunner et al., 2014). nhTMEM16 consists of structured cytoplasmic N-and C-terminal components and a transmembrane domain (TMD) containing 10 transmembrane helices. As general for the TMEM16 family, the protein is a homo-dimer (Fallah et al., 2011;Sheridan et al., 2011) with each subunit containing its own lipid translocation path located at the two opposite corners of a rhombus-shaped protein distant from the dimer interface (Brunner et al., 2014). This lipid path is formed by the 'subunit cavity', a membrane-spanning furrow of appropriate size to harbor a lipid headgroup. Since the subunit cavity is exposed to the membrane, it was proposed that its polar surface provides a favorable environment for lipid headgroups on their way across the membrane, whereas the fatty-acid chains remain embedded in the hydrophobic core of the bilayer (Brunner et al., 2014). In the vicinity of each subunit cavity, within the membrane-embedded domain, a conserved regulatory calcium-binding site controls the activity of the protein (Brunner et al., 2014). In light of the nhTMEM16 structure and the strong sequence conservation within the family, a central open question concerns how the TMEM16A architecture has adapted to account for its altered functional properties. Previous results suggested that the same region constituting the scrambling path also forms the ion conduction pore (Yang et al., 2012;Yu et al., 2012). However, in what way the distinct structural features of a scramblase, which allows the diffusion of a large and amphiphilic substrate, are altered in a channel that facilitates the transmembrane movement of a comparably small and charged anion, remained a matter of controversy.
Here we have resolved this controversy by the structure determination of mouse TMEM16A (mTMEM16A) by cryo-electron microscopy (cryo-EM) at 6.6 Å resolution and a complementary electrophysiological characterization of pore mutants. Our data define the general architecture of a calcium-activated chloride channel of the TMEM16 family and reveal its relationship to the majority of family members working as lipid scramblases. The protein shows a similar overall fold and dimeric organization as the lipid scramblase nhTMEM16. However, conformational rearrangements of helices eLife digest Cell membranes are made up of two layers of oily molecules, called lipids, embedded with a variety of proteins. Each type of membrane protein carries out a particular activity for the cell, and many are involved in transporting other molecules from one side of the membrane to the other.
The TMEM16 proteins are a large family of membrane proteins. Most are known as lipid scramblases and move lipids between the two layers of the membrane. However, some TMEM16 proteins transport ions in or out of the cell, and are instead called ion channels. TMEM16 proteins are found in animals, plants and fungi but not bacteria, and play key roles in many biological activities that keep these organisms alive. For example, in humans, ion channels belonging to the TMEM16 family help keep the lining of the lung moist, and allow muscles in the gut to contract.
The structure of a scramblase shows that two protein units interact, with each unit containing a furrow that spans the membrane, through which lipids can move from one layer to the other. However, to date, the shape of a TMEM16 ion channel has not been determined. It was therefore not clear how a protein with features that let it transport large, oily molecules like lipids had evolved to transport small, charged particles instead.
TMEM16A is a member of the TMEM16 family that transports negatively charged chloride ions. Using a technique called cryo-electron microscopy, Paulino et al. have determined the threedimensional shape of the version of TMEM16A from a mouse. Overall, TMEM16A is organized similarly to the lipid scramblase. However, some parts of the TMEM16A protein have undergone rearrangements such that the membrane-exposed furrow that provides a path for lipids in scramblases is now partially sealed in TMEM16A. This results in an enclosed pore that is largely shielded from the oily membrane and through which ions can pass. Additionally, biochemical analysis suggests that TMEM16A forms a narrow pore that may widen towards the side facing the inside of the cell, though further work is needed to understand if this is relevant to the protein's activity.
The three-dimensional structure of TMEM16A reveals how the protein's architecture differs from other family members working as lipid scramblases. It also gives insight into how TMEM16 proteins might work as ion channels. These findings can now form a strong basis for future studies into the activity of TMEM16 proteins.
lining the lipid scrambling path have sealed the subunit cavity, resulting in the formation of a protein-enclosed ion conduction pore that is for most parts shielded from the membrane but that might be partly accessible to lipids on its intracellular side.

Structure determination
We were interested in the structural properties that distinguish ion channels from lipid scramblases in the TMEM16 family and thus decided to investigate the structural properties of the chloride channel TMEM16A by single particle cryo-EM. For that purpose, we generated a stable HEK293 cell-line, which constitutively expresses the (ac) isoform of mTMEM16A, and purified the protein at a saturating calcium concentration in the detergent digitonin ( Figure 1-figure supplement 2A,B). Images of flash-frozen samples were recorded on a FEI TITAN Krios electron microscope equipped with an energy filter and a K2-summit camera (Figure 1-figure supplement 2C). The three-dimensional structure of the mammalian ion channel at a nominal resolution of 6.6 Å was reconstructed from total of 213,243 particles picked from 4178 micrographs (  Table 1). Since the resolution did not significantly improve after addition of further images, it is likely limited by the sample. In the resulting electron density map, the main features of the protein are well defined ( Figure 1A, Figure 1-figure supplement 4 and Video 1). Similarities with nhTMEM16 allowed the construction of a poly-alanine model

mTMEM16A structure
The EM-density of mTMEM16A superimposed on the model of the protein is shown in Figure 1A.
Due to the presence of Ca 2+ , it likely shows the channel in a Ca 2+ -bound conformation. In light of the irreversible rundown of TMEM16A-mediated currents observed in patch-clamp experiments at high Ca 2+ concentrations, it is at this point ambiguous whether this conformation corresponds to a conducting or a non-conducting state of the channel. Within the membrane, the overall dimensions of mTMEM16A are very similar to nhTMEM16 ( Figure 1B, Figure 1-figure supplement 4B and Video 1). All transmembrane helices are well resolved and thus, could be unambiguously allocated.
On the extracellular side, the mTMEM16A map contains a substantial amount of unassigned density that can be attributed to extended loops connecting transmembrane a-helices 1-2 (a1a2 loop) and transmembrane a-helices 9-10 (a9a10 loop), which are respectively 50 and 65 residues longer compared to nhTMEM16 ( Figure 1A and Figure 1-figure supplement 1). Both loops appear to be structured, folding into a compact extracellular domain ( Figure 2A). Notably, this domain harbors six cysteines that have been shown to be indispensable for channel activity (Yu et al., 2012) and that are thus potentially involved in disulfide bridges. On the cytoplasmic side, the N-terminal domain of mTMEM16A exhibits a similar fold and location with respect to the TMD as its counterpart in nhTMEM16 ( Figures 1B and 2B,C and Figure 1-figure supplement 4B). Consequently, there is no interaction between the N-terminal domains of adjacent subunits, which was previously proposed based on biochemical experiments (Tien et al., 2013). A 92 residue long extension in mTMEM16A that precedes the folded N-terminal domain (Figure 1-figure supplement 1) appears to be unstructured, but there is unaccounted electron density that cannot be interpreted at the current resolution of the data ( Figure 1-figure supplement 4B-D). At the C-terminus, which is 38 residues shorter than its equivalent part in nhTMEM16, the first a-helix (Ca1) is folded but it has moved away from the dimer axis and thus no longer contacts its symmetry mate ( Figure 2D). The remainder of the C-terminus is likely unstructured and, unlike in nhTMEM16, does not interact with the adjacent subunit. Hence, the interaction of the subunits within the mTMEM16A dimer differs significantly from nhTMEM16 since the cytosolic domains do not contribute to the dimer interface. Instead, interactions are established mainly at the extracellular part of transmembrane a-helix 10, which  is in a similar location as in nhTMEM16 but extends further towards the outside (Figures 1 and  2D). In the TMD, all membrane-spanning segments are well defined including two short amphiphilic a-helices at its N-terminal part that interact with the polar headgroups at the inner leaflet of the lipid bilayer ( Figure 1-figure supplement 4E). In general, the transmembrane helices are in comparable locations to their counterparts in nhTMEM16 ( Figure 1B and Figure 1-figure supplement 4B) and thus account for the overall similarity between both structures.

The pore region
The pore region of mTMEM16A, also containing the regulatory calcium-binding site, is formed by transmembrane a-helices 3-8. This region is well defined, except for the loops connecting a-helices 5 and 6 and 6 and 6' (Figure 3, Figure 3-figure supplement 1 and Video 2). Although, at the current resolution, neither the helix-pitch nor side-chains are resolved, there are several structural features that constrain the location of residues and thus allow for their approximate assignment. The placement is facilitated by conserved loops connecting a-helices 4-5, 7-8 and 8-9, which are well defined in the cryo-EM map and thus determine the register of the transmembrane segments (  Towards the intracellular side, the detachment of a4 and a6 results in the a dilation of the pore to a wide intracellular vestibule that is exposed to both the cytoplasm and the lipid bilayer

Functional properties of poremutations
A model of the pore is shown in Figure 5a. Since the current resolution of the data does not permit a quantitative analysis of its geometry, we restrict our description of the pore to its general geometric features. The wide, intracellular entrance narrows above the region constituting the regulatory Ca 2+ -binding site (Figure 4-figure supplement 1B). Under the assumption that the structure is close to a conducting state, the narrow upper part most likely requires permeating ions to shed their hydration shell. This is consistent with the observation that the anion selectivity of TMEM16A follows a type 1 Eisenman sequence (Qu and Hartzell, 2000;Schroeder et al., 2008;Yang et al., 2008), which favors larger anions with a lower solvation energy. The pore is amphiphilic Video 2. Pore region of mTMEM16A. Transmembrane a-helices 3-8 constituting the ion conduction pore and the Ca 2+ binding site of one mTMEM16A subunit (blue). EM density is superimposed. Green spheres correspond to the positions of bound Ca 2+ in nhTMEM16. The views are as in Figure 3. DOI: 10.7554/eLife.26232.015  (Läuger, 1973). Such behavior has been previously observed for mutations of Lys 588, where the removal of the positive charge has resulted in a strong outward rectification of the current (Jeng et al., 2016;Lim et al., 2016). In the model of mTMEM16A, this residue is located at the end of the funnel-shaped vestibule close to the neck of the ion conduction path ( Figure 5A and Figure 4-figure supplement 1C). In our data, the mutation K588A has resulted in a similar rectification, indicating that the truncation of the positively charged side-chain has perturbed the electrostatic interaction with permeating anions, ( Figure 5B and Figure 5-figure supplement 2B,C) effectively increasing the energy barrier of negatively charged ions to enter the pore from its intracellular side ( Figure 5-figure supplement 2A,B). A similar effect was observed for the nearby mutant K645A, which removes a positive charge from ahelix six at a position that is located slightly further towards the extracellular side ( Figure 5A,C and In between Lys 645 and Arg 535, the mutation R515A has caused a deviation from the linear current-voltage relationship in both directions ( Figure 5A,E). Thus, this positive charge most likely lowers a rate-limiting energy barrier for anion permeation halfway through the narrow part of the mTMEM16A pore ( Figure 5figure supplement 2A,B). This is consistent with the six-fold lower currents measured for this mutant, despite its robust expression at the surface of HEK cells ( Figure 5-figure supplement 1B,  C). In no case have we seen any change in the reversal potential measured in asymmetric chloride concentrations, which indicates that no single positive charge dominates the strong anion selectivity of the channel ( Figure 5-figure supplement 3). Together with our structural investigations, the electrophysiology data support the notion of a narrow pore in TMEM16A that widens towards the intracellular side.

Discussion
The present study has addressed structural relationships within the TMEM16 family. Since the majority of TMEM16 proteins work as lipid scramblases, which catalyze the diffusion of lipids between the two leaflets of a bilayer, it was postulated that the few family members functioning as ion channels may have evolved from an ancestral scramblase (Whitlock and Hartzell, 2016). However, the way in which TMEM16 channels have adapted to fulfill their distinct functional task has remained unknown. The structure of mTMEM16A reported here has now resolved this question. As anticipated from the strong sequence conservation, the general architecture of each subunit is shared between both branches of the family ( Figure 1B). A previous structure-based hypothesis suggested a possible subunit rearrangement in dimeric TMEM16 channels, where both subunit cavities come together to form a single enclosed pore (Brunner et al., 2014). Although this hypothesis was already refuted by recent functional investigations, which demonstrated that the protein contains two ion conduction pores that are independently activated by Ca 2+ (Jeng et al., 2016;Lim et al., 2016), the ultimate proof for a double barreled channel is now provided by the mTMEM16A structure, which reveals the location of two pores, each contained within a single subunit of the dimeric protein. A different proposition, referred to as the proteolipidic pore hypothesis, postulated that the ion conduction pathway in TMEM16 channels is partly composed of lipids (Whitlock and Hartzell, 2016). The authors suggested that immobilized lipid headgroups lining the membrane-exposed ion conduction pore may lower the dielectric barrier for permeating ions on their way across the lipid bilayer (Whitlock and Hartzell, 2016). Our study has also provided strong evidence against this hypothesis. Instead, the model of mTMEM16A shows that a-helical rearrangements have resulted in occlusion of the lipid pathway, while opening up a conductive pore which is largely shielded from the membrane (Figure 6 and Videos 3 and 4). The only potential access of lipids is provided on the intracellular side where the detachment of transmembrane a-helices 4 and 6 form a funnel-shaped vestibule that is exposed to the cytoplasm and the lipid bilayer ( Figures 5A and 6B). The gap between both a-helices may be a relic of an ancestral scramblase, and as suggested by the observed distortion of the detergent micelle in mTMEM16A, possibly destabilizes the bilayer (Figure 1-figure supplement  4B,D). Notably, this gap is also present in nhTMEM16, where a similar effect of membrane-bending has been proposed to facilitate scramblase activity, as suggested by molecular dynamics simulations (Bethel and Grabe, 2016). In this respect, it is noteworthy that the intracellular region connecting transmembrane a-helices 4 and 5 has recently been identified to play an important role in lipid scrambling in TMEM16F and was thus assigned the term 'scramblase domain' . Whereas TMEM16A itself does not facilitate lipid transport, scrambling activity was conferred to a chimeric TMEM16A protein carrying the 'scramblase domain' of TMEM16F  or the equivalent region of TMEM16E (Gyobu et al., 2015). Although these results emphasize the general role of the intracellular funnel region for lipid interactions, the altered structure of the 'subunit cavity', in particular the absence of a membrane-exposed polar crevice in TMEM16A, leave the mechanism of lipid scrambling in these chimeras ambiguous.
The structural view of the ion conduction path in mTMEM16A consisting of a funnel-shaped intracellular vestibule that narrows to a tight pore at the extracellular part of the membrane ( Figure 6A) is supported by our electrophysiology experiments. Analysis of mutants shows minimal influence of basic residues in the wide intracellular vestibule, but pronounced rectification upon similar replacements near the narrow neck of the pore. Remarkably, equivalent mutations of two of these residues (Arg 515 and Lys 645) have previously been described to alter the selectivity between different anions (Peters et al., 2015). Assuming that the imaged protein conformation resembles a conducting state, its pore structure suggests that permeating anions have to shed their hydration shell and interact with pore-lining residues ( Figure 6A). The low effective affinity of Clconduction indicates that there might not be a single strong site for ion coordination, but that the ions might instead weakly interact with the extended pore region (Qu and Hartzell, 2000) (Figure 5-figure supplement 1A). This is consistent with the fact that no single mutation was identified so far that weakened the strong selectivity for anions over cations ( Figure 5-figure supplement 3). Although ion conduction was previously also reported for TMEM16 family members which function as lipid scramblases (Lee et al., 2016;Malvezzi et al., 2013;Yang et al., 2012;Yu et al., 2015), it was proposed that these processes are leaks accompanying the movement of lipids , which differs significantly from the selective anion permeation described here for TMEM16 channels.
In summary, our work has unraveled how TMEM16 proteins use a similar architecture to exert substantially different functions. Both structures, namely the scramblase nhTMEM16 and the ion channel mTMEM16A, define the structural relationships within the family, whereby a hydrophilic membrane-exposed cavity in TMEM16 scramblases has changed to an aqueous membrane-shielded pore in TMEM16 channels ( Figure 6B and Video 4). Despite the unusual functional breadth of the family, this ligand-gated ion channel turns out to share its mechanism for ion conduction with other, structurally unrelated, channel proteins.

Protein expression and purification
A HEK293 cell-line stably expressing the mouse TMEM16A(ac) isoform (mTMEM16A, UniProt Q8BHY3.2) containing a 3C cleavage site, a myc-and an SBP-tag at its C-terminus was generated using the Flp-In System (Flp-In-293 Cell Line, R75007, Invitrogen). Adherent HEK cells constitutively expressing mTMEM16A were grown on 10 cm dishes (Corning) at 37˚C and 5% CO 2 in Dulbecco's modified Eagles's Medium (Sigma) containing either 10% fetal bovine serum (FBS, Sigma) for cell propagation or 5% FBS during protein production. After reaching >80% confluency, cells were harvested by centrifugation at 500 g, washed with PBS buffer (137 mM NaCl, 2.7 mM KCl, 12 mM phosphate pH 7.4) and stored at À20˚C until further use. For purification, frozen cell pellets from 7 l of adhesion culture were thawed and resuspended in 140 ml buffer A (20 mM HEPES pH 7.5, 150 mM NaCl and 0.5 mM CaCl 2 ) containing protease inhibitors (cOmplete, Roche). All further steps were carried out at 4˚C. Protein was extracted in buffer A containing about 1% digitonin (AppliChem) for 2 hr under gentle agitation. Insoluble material was removed by centrifugation at 22,000 g for 30 min. The supernatant was filtered through a 5 mm filter (Minisart, Sartorius) and incubated with 3 ml of Streptavidin UltraLink resin (Pierce, ThermoScientific) in batch for 1.5 hr. The beads were washed with 60 column volumes of buffer A containing 0.12% digitonin (Calbiochem; buffer B) and eluted with three column volumes of buffer B containing 4 mM of biotin. Protein was deglycosylated for 2 hr by addition of PNGaseF, and subsequently concentrated (Amicon Ultra, 100 k). The concentrated sample was applied to a Superdex 200 size-exclusion chromatography column equilibrated in buffer B. The following day fractions containing target protein were concentrated to obtain 15 ml of pure protein at a final concentration of 3 mg ml À1 and subsequently used for EM sample preparation.
Electron microscopy sample preparation and imaging 2.5 ml of purified mTMEM16A at a concentration of 3 mg ml À1 were pipetted onto glow-discharged 200 mesh gold Quantifoil R1.2/1.3 holey carbon grids (Quantifoil). Grids were blotted for 2-5 s with a blotting force of 1 at 20˚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 mm objective aperture. Data were collected in an automated fashion on a K2 Summit detector (Gatan) set to super-resolution mode with a pixel size of 0.675 Å and a defocus range of À0.5 to À3.8 mm using SerialEM (Mastronarde, 2005). Images were recorded for 15 s with an initial sub-frame exposure time of 300 ms (50 frames total) with a dose of 1.5 e À /Å 2 /frame, and later with a sub-frame exposure time of 150 ms (100 frames total) with a dose of 0.8 e À /Å 2 /frame, resulting in a total accumulated dose on the specimen level of approximately 80 e À /Å 2 .

Image processing
A total of 5503 dose-fractionated super-resolution images were 2 Â 2 down-sampled by Fourier cropping (final pixel size 1.35 Å ) and subjected to motion correction and dose-weighting of frames by MotionCor2 (Zheng et al., 2016). The contrast transfer function (CTF) parameters were estimated on the movie frames by ctffind4.1 (Rohou and Grigorieff, 2015). Images showing a strong drift, higher defocus than À3.8 mm or a bad CTF estimation were discarded, resulting in 4178 images used for further analysis. Image processing was performed using the software package RELION1.4 (Scheres, 2012) and at a later stage RELION2.0 (Kimanius et al., 2016). Approximately 4000 particles were manually picked to generate templates for automated particle selection. Following automated picking in RELION, false positives were eliminated manually or through a first round of 2D classification resulting in 755,348 particles. These were subjected to several rounds of 2D classification to remove particles belonging to low-abundance classes. The remaining 522,701 particles were sorted during 3D Classification with C2 symmetry imposed. A model was generated from the nhTMEM16 X-ray structure (Brunner et al., 2014) (PDBID 4WIS), low-pass filtered to 60 Å and used for the first round of classification. In an iterative mode, the best output map was used for subsequent classification or refinement rounds. Similar classes, comprising 377,371 particles, were combined and subjected to auto-refinement in RELION. The resulting map was masked and had a resolution of 7.35 Å . To further improve the quality of the density map, per-particle alignment of the frames was performed using the polishing algorithms in RELION. The best results were obtained upon inclusion of all dose-weighted frames and application of a running average window of 9, a standard deviation of 2 pixels on the translations during movie refinement and 200 pixels on particle distance during particle polishing (Scheres, 2014). Polished particles were subjected to another round of 2D and 3D classification, resulting in a selection of 213,243 particles. The final polished, auto-refined and masked map had a resolution of 6.6 Å . The final map was sharpened using an isotropic b-factor ranging between À351 Å 2 and À700 Å 2 and used for model building. Local resolution estimates were calculated within RELION. All resolutions were estimated using the 0.143 cut-off criterion (Rosenthal and Henderson, 2003) with gold-standard Fourier shell correlation (FSC) between two independently refined half maps (Scheres and Chen, 2012) (Figure 1-figure supplement 3B). During post-processing, the approach of high-resolution noise substitution was used to correct for convolution effects of real-space masking on the FSC curve (Chen et al., 2013).

Model building
A poly-alanine model encompassing the secondary structure elements of mTMEM16A was constructed based on the nhTMEM16 X-ray structure (Brunner et al., 2014) (PDBID 4WIS). For that purpose the nhTMEM16 structure was initially docked into the EM density using UCSF Chimera (Pettersen et al., 2004). The fit of certain fragments as rigid bodies was subsequently improved in Coot (Emsley and Cowtan, 2004). Long and poorly conserved loop regions and side-chains were removed from the model and residues of mTMEM16A were assigned based on a sequence alignment (Figure 1-figure supplement 1). Density for conserved short loops and bound Ca 2+ ions assisted the assignment of the register for residues of the pore region. The structure was improved locally by real space refinement in Coot (Emsley and Cowtan, 2004) followed by global real space refinement in Phenix (Adams et al., 2002;Afonine et al., 2013) maintaining strong secondary structure and symmetry constraints between the two subunits of the dimeric protein ( Table 1). The final model consists of 434 residues and includes the b-strands and a-helices of the N-terminal domain, two peripheral and 10 transmembrane spanning a-helices of the TMD, including short and conserved loop regions, and the first a-helix of the C-terminal domain. It contains residues 123-127, 167-214, 242-254, 278-282, 295-305, 315-355, 409-438, 486-520, 535-602, 633-666, 681-781, 855-885 and 892-904. The molecular surface of the pore was calculated with MSMS (Sanner et al., 1996) from coordinates where side-chain positions of residues constituting the ion conduction pore were modeled in Coot (Emsley and Cowtan, 2004). Model building was performed on the final cryo-EM map sharpened with a B-factor of À700 Å 2 , as shown in all figures except for Figure 1-figure supplement 4C,D where a B-factor of À351 Å 2 was applied. All structure calculations and model building were performed using software compiled by SBGrid (Morin et al., 2013). Structure figures ad movies were prepared with DINO (http://www/dino3d.org) or UCSF Chimera (Pettersen et al., 2004).

Electrophysiology
For electrophysiology, the mTMEM16A(ac) cDNA was cloned into a pcDNA3.1 plasmid modified for the FX-system (Geertsma and Dutzler, 2011) with a C-terminal YFP/SBP/myc tag. Mutations were introduced by a modified QuikChange method (Zheng et al., 2004) and confirmed by sequencing. HEK293T cells (ATCC CRL-1573) were transfected with 3 mg of DNA per 6 cm dish using the calcium phosphate precipitation method. Transfected cells were used within 24 to 96 hr after transfection. Inside-out patches were excised from HEK293T cells expressing WT or mutant constructs after the formation of a gigaohm seal. The seal resistance was typically 4-8 GW or higher. Recording pipettes were pulled from borosilicate glass capillaries (O.D. 1.5 mm, I.D. 0.86 mm (Sutter)) and fire-polished with a microforge (Narishige) before use. Pipette resistance was 3-8 MW when filled with recording solution. Voltage-clamp recordings were performed using the Axopatch 200B amplifier controlled by the Clampex 10.6 software through Digidata 1550 (Molecular Devices). Raw signals were analogue-filtered at 5 kHz through the in-built 4-pole Bessel filter and digitized at 20 kHz. Liquid junction potential was not corrected. Solution exchange was performed using a theta glass pipette mounted on a high-speed piezo switcher (Siskiyou).
Experiments were performed at 1 mM Ca 2+ on the intracellular side to maximize channel activation. This also minimizes interference by time-dependent relaxation of the current during a voltage step when information on the instantaneous current response is required. The pipette solution contained 150 mM NaCl, 5.99 mM Ca(OH) 2 , 5 mM EGTA and 10 mM HEPES at pH 7.4 (NaCl buffer). Rectification experiments were carried out under symmetrical ionic conditions with a bath solution having the same composition as the pipette solution. For permeability experiments, the NaCl concentration was adjusted by mixing the NaCl buffer with a (NMDG) 2 SO 4 solution containing 100 mM (NMDG) 2 SO 4 , 5.99 mM Ca(OH) 2 , 5 mM EGTA and 10 mM HEPES at pH 7.4 at the required ratio. For high ionic strength, KCl buffer, containing 150 mM KCl, 5.99 mM Ca(OH) 2 , 5 mM EGTA and 10 mM HEPES at pH 7.4, was used for both bath and pipette solutions to minimize the junction potential. For concentrations above 150 mM Cl À , KCl was dissolved in this solution at the required amounts.
Data were background-subtracted before analysis. Background current was obtained by recording in the corresponding solution in the absence of intracellular Ca 2+ . I-V data were obtained by measuring the instantaneous current after each voltage jump in a step protocol ( Figure 5-figure  supplement 1B). To correct for current rundown, the measured instantaneous currents were divided by the fraction of current remaining during the pre-pulse at +80 mV and were expressed as normalized current (I/I 120mV ). This is important as uncorrected current rundown can give rise to artificial rectification. Potential voltage offset was detected by recording in symmetrical solutions. Only patches with an offset <2 mV were accepted for analysis. The voltage offset was subtracted from the reversal potentials obtained from asymmetric ionic conditions for the same patch whenever possible. This was not possible for a minority of constructs that displayed low current and/or fast rundown. For these constructs, the averaged offset was subtracted from the averaged reversal potentials obtained in asymmetric ionic conditions. Data are presented as mean ± s.e.m..

Model of permeation
To analyze the position-dependent effect of mutations on the rectification of the current, we have employed a barrier model akin to that described by Läuger (1973). We are aware of the general limitations of barrier models for quantitative interpretations (Eisenberg, 1999) and thus only aim for a phenomenological description. The model assumes the presence of multiple hypothetical energy barriers on the ion conduction path that are not necessarily identical (Appendix Scheme 1). The equation used to fit the experimental I-V data and to determine the descriptive energy profile of the constructs is shown below.
The model contains three free parameters (n, s b and s h ) that govern the shape of the I-V relation, which, with reasonable constraints, can be reliably determined from our data ( Figure 5B-E and Figure 5-figure supplement 2A,B). A is a proportionality factor, n is the number of barriers and s b and s h are relative rates for outward flux across the innermost and internal barriers compared to the external barrier. For our fit, we used three barriers to describe the observed behavior and determined s b and s h for the mutant constructs. The relative increase of the barrier height is obtained by where E a is the activation energy corresponding to the respective rate constant. These parameters were used to construct descriptive energy profiles to illustrate the effect of the mutations and are shown in Figure 5-figure supplement 2A,B. For more details, see Appendix 1.

Accession codes
The electron density map has been deposited in the Electron Microscopy Data Bank under the accession code EMD-3658 and the coordinates of the model in the Protein Data Bank under the accession code 5NL2. Additional files

Major datasets
The following datasets were generated: