Structure of a volume-regulated heteromeric LRRC8A/C channel

Volume-regulated anion channels (VRACs) participate in the cellular response to osmotic swelling. These membrane proteins consist of heteromeric assemblies of LRRC8 subunits, whose compositions determine permeation properties. Although structures of the obligatory LRRC8A, also referred to as SWELL1, have previously defined the architecture of VRACs, the organization of heteromeric channels has remained elusive. Here we have addressed this question by the structural characterization of murine LRRC8A/C channels. Like LRRC8A, these proteins assemble as hexamers. Despite 12 possible arrangements, we find a predominant organization with an A:C ratio of two. In this assembly, four LRRC8A subunits cluster in their preferred conformation observed in homomers, as pairs of closely interacting proteins that stabilize a closed state of the channel. In contrast, the two interacting LRRC8C subunits show a larger flexibility, underlining their role in the destabilization of the tightly packed A subunits, thereby enhancing the activation properties of the protein.


Introduction
The volume of a vertebrate cell is tightly linked to the osmotic state of its surroundings. While at equilibrium under isotonic conditions, the influx of water in response to a change to a hypotonic environment causes swelling, leading to a dilution of the cytoplasm and in severe cases to bursting.
To counteract swelling, cells have developed mechanisms to activate ion and osmolyte efflux pathways in a process called regulatory volume decrease 1,2 . The concomitant efflux of water causes a return of the cell to its original state. Volume-regulated anion channels (VRACs) are important participants in regulatory volume decrease 3,4 . These channels can be activated by the increase of the cell volume and the reduction of the intracellular ionic strength, although the detailed activation mechanism in their physiological context is still poorly understood 5-7 . VRACs are composed of proteins belonging to the conserved LRRC8 family, whose expression is restricted to chordates 8-10 . The family contains five closely related members in humans (termed LRRC8A-E), all consisting of an N-terminal pore domain (PD) followed by a cytoplasmic leucine rich repeat domain (LRRD) 11 . Although, upon overexpression, several family members can assemble on their own 12,13 , in a cellular environment VRACs form obligatory heteromers composed of at least two different homologs 10,14 . In these assemblies, LRRC8A constitutes an obligatory subunit, which is essential for the targeting of channels to the plasma membrane 10 .
Other subunits determine the substrate selectivity and activation properties of VRACs. Channels containing LRRC8C preferably conduct small inorganic anions and were identified to play an important role in T-cell regulation 15 , whereas the presence of LRRC8D and E extends the range of permeating substrates to osmolytes such as taurine and amino acids 16,17 . Consequently, LRRC8Econtaining VRACs in astrocytes were associated with the release of the neurotransmitter glutamate during endemic swelling and stroke leading to neurotoxic effects 18,19 , whereas the channels containing LRRC8D subunits confer the permeability of platinum compounds, which make VRAC an important uptake route of anti-cancer drugs during chemotherapy 17 .
The general architecture of VRACs was revealed from cryogenic electron microscopy (cryo-EM) structures of homomeric LRRC8A 12,20-24 , which forms a functional channel with compromised activation properties 12,25,26 . The protein assembles as a hexamer with subunits arranged around an axis of symmetry that defines the ion conduction pore. In these cryo-EM reconstructions, the PDs are generally well-defined obeying C6 or pseudo-C6 symmetry with the LRRDs showing larger conformational heterogeneity 22 . In a major population of particles, adjacent LRRDs have rearranged to maximize interactions leading to subunit pairs forming an asymmetric unit in a C3symmetric channel 12,20,21 . This conformation was found to be stabilized upon binding of a synthetic nanobody (sybody), which specifically recognizes the LRRD of the A subunit to inhibit channel activity 27 . A recent structure of a homomeric LRRC8D assembly also displayed a hexameric arrangement of subunits, arranged with lower (C2) symmetry 13 .
In contrast to LRRC8 homomers, our current understanding of heteromeric channels is scarce and restricted to a low-resolution reconstitution of a protein consisting of A and C subunits 12 . Though confirming a similar hexameric organization as found in LRRC8A channels, the similarity of subunits has prevented their identification during the classification step and the application of C3 symmetry, as a measure to improve the density, resulted in averaging out conformational differences between subunits. Consequently, the ratio and relative locations of both subunits in hexamers has remained elusive.
To gain insight into the organization of heteromeric VRACs, we engaged in structural studies of channels composed of LRRC8A/C subunits that were either obtained by overexpression or isolation of endogenous protein from native sources. Our study reveals an organization with a single predominant stoichiometry, where A subunits cluster as tightly interacting pairs with a characteristic conformational preference found in homomeric channels, whereas interspersed paired C subunits appear to increase the dynamics of the complex in line with the proposal that channel activation concomitantly increases the mobility of the LRRDs.

Distribution of LRRC8 subunits
We first set out to analyze the distribution of subunits in endogenous VRACs, which are expressed under the control of native promotors and assembled by an unperturbed cellular machinery. To this end, we used the sybody Sb1 LRRC8A (Sb1), which specifically binds the LRRD of the A subunit with nanomolar affinity 27 , for the isolation of endogenous protein from HEK293 cells. Following a tryptic digestion of the purified sample, all five LRRC8 family members were identified by liquid chromatography with tandem mass spectrometry (LC-MS/MS), confirming the described broad expression of subunits in WT cells 10,14 (Extended Data Fig. 1a). Since the observed abundance likely reflects a complex distribution of channel populations with distinct subunit composition, which would prohibit a detailed structural investigation, we turned our attention towards a modified HEK293 cell line carrying genetic knockouts of the B, D and E subunits 10,16 (LRRC8 B,D,E-/-, generously provided by T. J. Jentsch) in an attempt to reduce sample heterogeneity. However, if binomially distributed in hexameric channels, both subunits could still form twelve distinct assemblies of LRRC8A/C heteromers (Fig. 1a). Similar to HEK293 cells, LRRC8 B,D,E-/cells mediate VRAC currents in response to swelling that slowly inactivate at positive voltages, which is a hallmark of channels consisting of A and C subunits 14,16 (Extended Data Fig. 1b, c). We then proceeded with affinity purification and mass spectrometry analysis of endogenous VRACs isolated from WT and LRRC8 B,D,E-/cells, from which we determined LRRC8A/C complex stoichiometries using absolute quantification with two isotopically labeled reference peptides per subunit (Extended Data Fig. 1d). By this approach, we found A and C subunits at a 2.9:1 ratio in LRRC8 B,D,E-/cells, whereas they are detected at a lower, 1.8:1 ratio, in WT HEK293 cells (Fig.   1b, Extended Data Fig. 1e). Assuming a hexameric arrangement of channels composed of A and C subunits and considering that our approach has isolated all proteins containing LRRC8A, this result reveals a predominance of this obligatory VRAC component in populations presumably containing four to six copies of the subunit. For endogenous protein isolated from LRRC8 B,D,E-/cells, this would yield an about 3:2 ratio of channels comprising either four or six LRRC8A subunits, or alternatively, an about equal distribution of channels comprising four or five A subunits. We next attempted to characterize the subunit composition of LRRC8A/C channels produced by heterologous overexpression. To this end, we expressed differentially tagged LRRC8A and C constructs either in HEK293 GnTIcells or in LRRC8 -/cells, where all five LRRC8 subunits have been knocked out 10 , and isolated protein by tandem affinity purification to obtain heteromeric channels that contain at least one copy of each subunit. To probe the variability of the subunit composition, cells were transfected with different ratios of DNA coding for either LRRC8A or LRRC8C subunits. The overexpressed protein was purified and subjected to an analogous mass spectrometry analysis as described for endogenous VRACs that allowed the quantification of subunit ratios. In case of a transfection of subunits at equimolar ratios, the analysis yielded an A:C ratio of about 2.0:1 (Fig. 1b, Extended Data Fig. 1e). Although this ratio can be slightly perturbed upon the transfection of LRRC8C-DNA at three times higher concentration compared to LRRC8A-DNA, the resulting A:C subunit ratio of 1.8:1 emphasizes that even at a large excess of former, the LRRC8A subunits prevail (Fig. 1b, Extended Data Fig.   1e). Together our results hint towards a dominating distribution of A subunits in heteromeric LRRC8A/C channels, which contrasts a recent proposal that LRRC8A would be a minor component of VRACs 28 .

Structural properties of LRRC8C homomers
In the next step, we engaged in the structural characterization of the building blocks of VRAC heteromers. By combining data from cryo-EM and X-ray crystallography, previous studies have revealed the structural properties of LRRC8A 12,20-22 , which assembles as a hexamer exhibiting characteristic features of an ion channel. In these channels, pairs of subunits form the asymmetric unit of a C3-symmetric assembly where the approximate six-fold rotational relationship of the pore breaks down at the level of the LRRDs, which have altered their mutual conformation to maximize interactions 12 . Here, we investigated whether we would find similar properties for homomeric assemblies consisting of LRRC8C subunits. We thus expressed full-length LRRC8C and its isolated LRRD (LRRC8C LRRD ) and purified both constructs for further characterization. As for the LRRD of the A subunit, LRRC8C LRRD is a monomeric protein in solution. Its structure determined by X-ray crystallography at 3.1 Å (Extended Data Fig. 2) shows similar features as the A domain, with both horseshoe-shaped proteins superimposing with an RMSD of 1.4 Å (Fig. 2a). Both domains share a sequence identity of 56%, consist of the same number of repeats and do not contain insertions in loop regions. We then set out to characterize the full-length protein and collected cryo-EM data of an LRRC8C homomer from three independent preparations (Table 1).
Unexpectedly, the 2D classes from these datasets showed in all cases a heptameric assembly, which was confirmed upon 3D reconstruction (Extended Data Fig. 3). Despite the large size of the combined dataset, the non-symmetrized reconstruction of the full-length protein did not reach high resolution, likely due to the intrinsic mobility of the complex. After application of C7 symmetry, we were able to obtain a structure extending to 4.6 Å for the full-length protein and 4.1 Å for the PD (Extended Data Fig. 3b). At low contour, the map displays an envelope for the entire protein, and at higher threshold, where the density of the more mobile LRRDs has largely disappeared, it defines the structure of the PD (Fig. 2b, c). This map allowed a molecular interpretation with subunits consisting of the structure of PD determined by cryo-EM and the X-ray structure of the LRRD to obtain a symmetric channel protein with a pore radius of 6 Å at its extracellular constriction ( Fig. 2d-f). In this assembly, close subunit interactions are restricted to the extracellular part whereas contacts within the remainder of the TM domain and between LRRDs are scarce (Fig. 2e). In case of the LRRDs, they are restricted to interactions between residues at the C-terminal part, where fenestrations between adjacent LRRDs and a large opening at the symmetry axis would permit free access of permeable substrates (Fig. 2e). With respect to its subunit organization and pore size, the TM domain of the homomeric protein closely resembles the heptameric pannexin channel which is known to conduct large substrates including ATP 29 (Fig.   2f). The reported lack of activity of this large channel is thus likely a consequence of its cellular distribution as the subunit, when expressed on its own, is not targeted to the plasma membrane 10,26 , whereas functional LRRC8C channels have been obtained in a chimera containing a disordered loop of LRRC8A that promotes expression at the cell surface 26 .

Structure of LRRC8A/C channels in complex with Sb1
To gain insight into the structural properties of LRRC8A/C channels, we have overexpressed the protein in large suspension cultures of HEK293 cells transfected with constructs coding for A and C subunits at two distinct ratios (i.e., at LRRC8A:C ratios of 1:1 and 1:3), isolated the channels by tandem affinity purification and added the sybody Sb1 at a 1.5 molar excess prior to vitrification.
For both preparations, we have collected large datasets by cryo-EM and proceeded with 2D classification and 3D reconstruction (Tables 1 and 2). Of the two datasets, only the one collected from channels obtained from a transfection with an equimolar ratio of DNA (LRRC8A/C 1:1 /Sb1), showing an A to C stoichiometry of 2:1, permitted reconstruction at high resolution (Extended Data Figs. 4 and 5). In contrast, the structure of a sample obtained from a transfection with a 1:3 ratio of A to C subunits (LRRC8A/C 1:3 /Sb1) resulting in a subunit ratio of 1.8:1 is less well ordered, which has complicated particle alignment and thus compromised the obtained resolution (Extended Data Fig. 6a-d). In agreement with previous studies, the channels in both datasets are hexameric and we did not spot any heterogeneity with respect to their oligomeric state. To compare structural properties of overexpressed samples with endogenous VRACs, we have also purified channels from LRRC8 B,D,E-/cells using a column containing immobilized Sb1 for affinity chromatography. The eluted endogenous LRRC8A/C end /Sb1 complex was frozen on carbonsupported grids and used for cryo-EM data collection ( Table 2). The poor yield of this preparation and the consequent low particle density on the grids together with the potential heterogeneity of particles suggested by our mass spectrometry analysis (Fig. 1b, Extended Data Fig. 1e) have prevented the unambiguous alignment of distinct subunits in VRAC heteromers and thus restricted our analysis to general attributes derived from a reconstruction at low resolution (Extended Data Fig. 6e-i). Despite these limitations, we find a structure carrying characteristic properties observed for the overexpressed samples. In our low-resolution map, endogenous channels show a hexameric assembly with features of their cytoplasmic LRRDs exhibiting the familiar pairwise tight arrangement that has become hallmark of A subunits (Extended Data Fig. 6f, i).
In our study, the structure of the LRRC8A/C 1:1 /Sb1 complex has defined the properties of a heteromeric VRAC at high resolution. It was determined from a large dataset consisting of a total of 26,401 micrographs without application of rotational symmetry (C1) at an overall resolution of 3.8 Å for the entire protein and 3.3 Å for the PD (Fig. 3a, Extended Data Figs. 4 and 5). Although we expected to observe a heterogeneous population of channels, upon 3D classification we found a hexameric protein with nearly uniform subunit distribution in a single predominant conformation (Extended Data Fig. 4b). In this assembly, four adjoining subunits including their cytoplasmic LRRDs are well-defined with density of bound Sb1 distinguishing them as LRRC8A chains, whereas the density of the LRRDs of the two remaining subunits is absent (Fig. 3a). The four A domains in the hexamer (denoted as A1 to A4, Fig. 3a, b) are organized as pairs with mutual tight interactions between their LRRDs, as initially observed in the homomeric LRRC8A complex 12,27 ( Fig. 3b, c). In contrast, the two pairwise interacting C subunits (denoted as C1 and C2) are more dynamic. Whereas the extracellular part of the PDs of both C subunits, consisting of the extracellular sub-domains (ESDs) and the membrane-inserted parts (TMs), are well-defined and show structural hallmarks of this paralog, the cytoplasmic sub-domain (CSD) of the C1 subunit located at the A4/C1 interface is defined poorly and that of C2 located at the C2/A1 interface is not resolved (Fig. 3a). Additionally, both LRRDs are mobile and thus not visible in the cryo-EM density (Fig. 3a). The observed organization reflects the properties of the respective homomeric structures exhibiting extended interactions between LRRC8A subunits whereas C subunits are less well packed. Within the pore domain, the ESD obeys pseudo-6-fold symmetry, which is also largely maintained for the TMs and CSDs of the A subunits, whereas the TM of C1 has undergone a slight outward movement away from the pore axis that can be described by a 3° rigid body rotation around an axis placed at the border between ESD and PD, reflecting the apparent deterioration of the interactions at A/C, C/C and C/A interfaces (Fig. 3d). Thus, despite the tight interactions at the extracellular part of the PD, the mobile cytoplasmic parts of the C subunits lead to a decreased contact area at the intracellular CSD in interfaces involving C subunits. In the structure of the A subunit in complex with Sb1 27 , the sybody has led to a rigidification of the domain structure, which is also observed in the LRRC8A/C heteromer (Fig. 3c). In this arrangement of the A subunits, the conformation of the C subunits observed in the heptameric LRRC8C structure would lead to clashes which are pronounced between the LRRDs of C2 and A1 (Fig. 3e), thus forcing them into a different conformation. However, instead of adopting an Alike domain arrangement, which would allow their accommodation in the restricted space of the hexameric protein, the LRRDs of the C domains have become mobile and are thus not defined in the density of the heteromeric channel.

Structure of LRRC8A/C channels in absence of Sb1
Since the binding of Sb1 rigidifies the LRRDs of the four A subunits in the observed conformation, thereby restricting the accessible space of the corresponding domains of the two LRRC8C subunits, we continued to characterize structures of LRRC8A/C in absence of the sybody. For that purpose, we have followed two different strategies, one involving the labeling of the PD of the A subunit to facilitate its identification and particle alignment and a second, the classification of LRRC8A/C channels without any labeling. For the first approach, we have generated a construct where we fused the 57-residue long SAM domain of human tumor suppressor p73 to the truncated first extracellular loop of LRRC8A to create the construct LRRC8A SAM (Extended Data Fig. 7a).
The replacement of the mobile loop was well tolerated and a dataset of the homomeric LRRC8A construct showed a channel with similar conformational properties as observed in the structure of the unlabeled A subunit (Extended Data Fig. 7b, Table 1). In the structure of the fusion construct, the SAM domain, which is integrated into a loop that bridges towards the neighboring subunit, does not visibly interfere with subunit interactions and is clearly recognizable in a low-resolution reconstruction, distinguishing it as a proper fiducial marker (Extended Data Fig. 7a, b). A sample prepared from a 1:1 ratio of transfected constructs (LRRC8A SAM /C) showed a similar 2:1 ratio of A:C subunits as WT LRRC8A/C obtained under equivalent conditions, as confirmed by mass spectrometry (Extended Data Fig 7c, d).  Fig. 7i). Whereas the altered subunit disposition in this channel population is likely a consequence of the fused SAM domain, which appears to mildly perturb the interaction between LRRC8A subunits, leading to the dissociation of contacts at the loose interface, the preserved 2:1 A to C stoichiometry and the pairwise organization of tightly interacting LRRC8A subunits further supports their role as building blocks in heteromeric VRACs.
An arrangement closely resembling the LRRC8A/C 1:1 /Sb1 complex and the corresponding channel population in the LRRC8A SAM /C sample was observed in a dataset of LRRC8A/C obtained in absence of Sb1 (Table 1). In the 3D reconstruction generated from this dataset, we find a consecutive arrangement of four well-defined subunits that are organized as tightly interacting pairs and two less well-defined subunits (Fig. 4, Extended Data Fig. 8). In the latter, additional density at the level of the LRRDs, that was not found in the LRRC8A/C 1:1 /Sb1 complex, can be attributed to the subunit occupying the C1 position, suggesting that the absence of Sb1 would allow for a better integration of the C subunits in the hexameric protein (Fig. 4a). In this structure, the A subunits are readily identified by their characteristic pairing observed in previous structures (Fig.   3a). The cluster of two subunit pairs resembles the arrangement found in the LRRC8A/C/Sb1 complex (Fig. 3a). However, upon closer inspection, it is apparent that the LRRDs have rearranged compared to the interactions in the LRRC8A/C/Sb1 complex (by rigid body rotations of 16°, 9°, 11°, and 10° for the respective positions A1-A4) leading to a weakening of the tight interface and the creation of a gap between interacting domain pairs (Fig. 4b, c). The described movements of the LRRDs of the four A subunits have expanded the accessible space for the respective regions of the adjacent C subunits. The consequent population of low-energy conformations, which decreases their conformational heterogeneity, is manifested in the emergence of density of the LRRD of the subunit in the C1 position and of the CSD of the less-well defined C2 position (Fig.   4a). Remarkably, the relative orientation of the LRRD in the C1 position is distinct from the conformations observed for the A subunits. It instead resembles the arrangement in the LRRC8C heptamer, except for a rigid body rotation by 17° away from the pore axis around a pivot that is located at the interface to the PD (Fig. 4c, d). A similar LRRD conformation at the C2 position would result in a clash with the contacted A1 position, requiring a moderate rearrangement that increases the domain mobility as reflected in its absent density. Together, our results emphasize the distinct conformational preferences of A and C subunits, defined in the datasets of the respective homomers, as determinants of their properties in heteromeric channels.
Collectively, the described structures of LRRC8A/C heteromers have confirmed the 2:1 subunit stoichiometry determined by mass spectrometry leading to a preferred protein arrangement where the clustering of A domains illustrates the predominance of homomeric interactions, which are maximized in tightly interacting subunits pairs burying an extended LRRD interface. This tight arrangement is interrupted by the insertion of C domains, which increase the protein dynamics and improve the activation properties of heteromers.

The anion selectivity filter
In contrast to its intracellular parts, the TM and ESDs of the C subunits in the LRRC8A/C 1:1 /Sb1 dataset are well defined and provide detailed insight into the structural properties of a heteromeric VRAC (Fig. 5a). These are particularly informative at the level of the ESDs, which form the constricting part of the channel, resembling a selectivity filter (Fig. 5b). In the extracellular halve of the PD, the pseudo-symmetry related subunits are found in a similar arrangement as observed in the homomeric channel LRRC8A (Fig. 5c). In this part of the channel, the surface area buried between contacting subunits, ranging between 2'200 and 2'500 Å 2 , is of comparable size in all interfaces (Fig. 5b). Still, subunit-specific differences, such as the replacement of residues engaged in a salt bridge in A/A interfaces (between His 104 and Asp 110) by uncharged polar residues (i.e., Gln 106 and Asn 112) might modulate the strength of the interaction (Fig. 5d, e). At the constriction, the A subunits contain an arginine residue (Arg 103), which determines the high anion over cation selectivity of the channel 12 (Fig. 5e). In case of the C subunits, this arginine is replaced by a leucine (Leu 105) whose lower side chain volume increases the pore diameter at the constriction (Fig. 5d, f), thus likely accounting for the increased single channel conductance of the A/C complexes compared to the A homomers found in a previous study 14 . The described mutation R103L also alters the polar properties of the filter by introducing a hydrophobic segment into a ring of basic residues.

Discussion
Our study provides first detailed insight into the previously unknown organization of heteromeric VRACs. In a cellular environment, these proteins consist of the obligatory LRRC8A subunit and at least another member of the LRRC8 family, to form functional channels with distinct composition-dependent properties 10,16,17 . Their hexameric architecture and the possibility to assemble proteins from five different homologs leads to a vast number of possible arrangements.
A large heterogeneity of heteromers would thus be expected in case all subunits would interact with similar affinity and their assembly would be governed by thermodynamics 23 . In such scenario, the distribution of distinct oligomers would exclusively depend on the concentration of expressed subunits within a cell and their relative disposition in the channel would be random. The high conservation between family members and their consequent close structural resemblance would hamper their distinction in a heterogeneous sample containing a diverse set of channel populations and thus prohibit a detailed structural investigation. To limit the number of possible assemblies, we have thus focused on heteromers formed by the protein chains LRRC8A and C, which in a hexameric channel could form up to 12 distinct assemblies (Fig. 1a). LRRC8A/C channels were either obtained by overexpression and tandem purification or isolation of endogenous protein from a HEK293 cell-line only expressing A and C subunits 16 . By employing absolute quantification of proteins by mass spectrometry, we found a robust 2:1 ratio of A to C homologs in the samples purified from cells transfected with equimolar amount of DNA, which is also reflected in the structural properties of the sample (Fig. 3). The higher 3:1 ratio of A to C subunits, observed in endogenous channels purified from cells where other homologues were genetically knocked-out, reflects a heterogeneous distribution where channels with 2:1 subunit ratio would constitute a major population. In these cells, the excess of A subunits could also lead to the formation of homomeric channels, reflecting the high stability of their assembly. This is unexpected considering the current view that homomeric channels would not exist in a cellular environment 10 and it remains unknown whether populations of LRRC8A homomers would also be found in native tissues. Since they remain non-responsive under physiological swelling conditions, it is also unclear whether they could be activated by a different stimulus, as it was recently achieved by the binding of potentiating sybodies 27 . The observed abundance of A subunits in heteromeric channels contradicts a previous study suggesting that LRRC8A might be a minor component of VRACs 28 .
Assuming an unbiased distribution of subunits in channels with an A to C ratio of 2:1, we would still expect to find three distinct assemblies (Fig. 1a). In such channels, the disposition of subunits would be determined by the relative difference between homo and heteromeric interactions where we would find a binomial distribution only if the interaction between subunits is energetically balanced. Conversely, we expect a clustering of subunits of the same type in case of stronger homomeric interactions or the dispersion of subunits in the opposite case 23 . In our structures of LRRC8A/C channels, we find a single distribution with A and C subunits segregating into clusters, suggesting that the affinity between homomers prevails. Differences in the conformational properties of distinct subunits underlying their observed clustering in heteromeric assemblies can already be appreciated in structures of homomers. Homomeric channels composed of the LRRC8A domain are distinguished by their compact oligomeric arrangement leading to the formation of tightly interacting subunit pairs where the comparably mobile LRRDs have rearranged to maximize interactions 12,20,21 . This conformation is further stabilized by the inhibitory sybody Sb1 and leads to the packing of subunits in C3-symmetric channels 27 . It is thus not surprising to find interacting LRRC8A pairs as invariant building blocks also in heteromeric channels. The observed tight interaction of A subunits is consistent with the poor activation properties of homomeric LRRC8A channels 12,14,25,26 , suggesting that they may stabilize a closed state of the ion conduction pore. In contrast, there are fewer contacts in case of LRRC8C homomers, which have formed a larger heptameric assembly (Fig. 2). In this case, the LRRDs show an increased mobility compared to A subunits and we did not find any indication of tight subunit interactions. Similarly, the interaction interface in the TMD is reduced and restricted to contacts in the ESD (Fig. 2). A conformation deviating from LRRC8A and C was found in a previous structure of the D subunit, which assembles into hexamers exhibiting C2 symmetry 13 . In this structure, the tight interaction between three subunits forming the asymmetric unit has disrupted the packing between trimers resulting in a distortion of the assembly. Considering the observed symmetry-mismatch, the incorporation of C subunits and potentially also of other LRRC8 homologs into heteromeric channels would perturb the tight interactions found in A homomers and thus destabilize the closed state and increase the activation properties (Fig. 6a). This is illustrated in the observed weaker density of C subunits in the LRRC8A/C 1:1 /Sb1 complex, which is pronounced at the level of their intracellular components (Fig. 3a) and the disruption of LRRD contacts in tightly interacting subunit pairs leading to a conformational change in the structure of a LRRC8A/C channel in absence of Sb1 (Fig. 4). This mechanism is consistent with the previously proposed role of potentiating sybodies to perturb LRRD packing 27 and the observed correlation between increased LRRD mobility and activation 30,31 . Although in combination with earlier studies 27,30,31 , our current data strongly support the notion of LRRDs to regulate channel activity by coupling to a physical gate, the exact location of this gate and the nature of the coupling mechanism is still poorly understood 23 . Previous studies have assigned a role of the ESD in voltage-dependent inactivation 32 and suggested the N-terminus, which projects into the pore, to be a major constituent of the gate 33 .
The N-terminus was defined in the structure of LRRC8D lining the pore at the interface between domains, which further strengthens this hypothesis 13 . In contrast, the highly similar N-termini of A and C subunits appear both mobile and are thus not defined in the structures of homo and heteromeric channels and the mechanism by which they might contribute to the inhibition of ion conduction is still unclear and requires further investigations.
The observed organization of subunits also affects the properties of the ESD constituting the anion selectivity filter, where the substitution of a constricting arginine in LRRC8A by a smaller leucine in LRRC8C has increased the pore diameter and introduced a hydrophobic segment in the filter that resembles the amphiphilic properties of equivalent regions found in unrelated chloride channels and transporters 34-36 (Fig. 5d, f, and 6b). Knowledge of the detailed filter architecture could be exploited in the design of potent and specific pore blockers binding to this region as strategy against diabetes and other VRAC-related diseases 37,38 . An increased pore diameter can also be expected for heteromeric channels containing D and E subunits 13,16,17 , although it is currently unclear whether in a hexameric organization, this increase would be sufficiently large to account for the pronounced properties of these channels that renders them permeable also to larger substrates such as amino acids osmolytes and anti-cancer drugs 17 . In that respect, the larger heptameric assembly of LRRC8C homomers described in this study is noteworthy, as it would allow for a further increase of the pore radius, although there is so far no evidence for the existence of heteromeric channels with larger oligomeric organization in a physiological context. It will thus be important to study the assembly of heteromeric VRACs composed of different family members and containing more than two distinct subunits in future studies to better understand their versatile functional properties and unique activation mechanism. However, also in channels with alternate subunit composition, we expect the general rules defined in the present study to apply.    1 and 1:3). Absolute peptide amounts calculated by spiking each sample with known amounts of stable isotope-labeled peptides were used for ratio determination. Boxplots cover the first and third quartiles from bottom to top and the whisker extends to largest/smallest value but no further than 1.5×IQR (inter-quartile range). The median ratio is indicated by a black solid line.

Expression constructs and cloning
All constructs were generated using FX-cloning and FX-compatible vectors 39  For tandem purification of heteromeric LRRC8A/C or LRRC8A SAM /C channels, a similar protocol as described for LRRC8C purification with some modifications was used, as described previously 12 . In all cases, cell pellets from typically 8 l of culture were used. After the clarification of lysates, two consecutive affinity chromatography steps were performed to ensure that final samples contain both, the His-tagged LRRC8C and SBP-tagged LRRC8A or LRRC8A SAM constructs. In a first purification step, the lysate was supplemented with 5 mM imidazole and applied to 10 ml of Ni-NTA resin (Agarose Bead Technologies) to pull-down channels containing LRRC8C subunits. The resin was washed with 30 CV of SEC buffer 1 supplemented with 5 mM imidazole and the protein was eluted with 4 CV SEC buffer 1 containing 300 mM of imidazole.
Elution fractions were applied to Streptactin superflow resin (IBA LifeSciences) to remove homomeric LRRC8C channels. All following steps were performed as described for LRRC8C purification. Final protein samples were either immediately frozen on cryo-EM grids or flash frozen and used for quantification of both subunits by mass spectrometry. The typical yield for an A/C tandem purification amounted to 10 µg/l of transfected cells.

Sybody purification
For the labeling of LRRC8A in samples used for cryo-EM, sybody Sb1 was purified from bacteria as described 27 . Briefly, the pBXnPH3 plasmid containing the construct coding for Sb1 was transformed into E. coli MC1061 bacteria and grown in Terrific Broth medium supplemented with ampicillin. Protein expression was induced with arabinose and bacteria were harvested after 19 h.
Cells were lysed and His-tagged Sb1 was purified using Ni-NTA resin (Agarose Bead Technologies). The His-tag was cleaved using HRV 3C protease and the sample was dialyzed overnight to eliminate imidazole. The dialyzed sample was subjected to Ni-NTA resin to remove the cleaved tag before concentration and further purified on a Sepax SRT-10C SEC100 column (Sepax Technologies). Sybody fractions were pooled and concentrated to 5.3 mg ml -1 , supplemented with glycerol and flash-frozen till further use. Affinity chromatography, dialysis and concentration steps were performed at room temperature.

Purification of endogenous LRRC8 channels
Purification of endogenous LRRC8 channels from HEK293T and LRRC8 B,D,E-/cells was performed by two alternative strategies. In one protocol, the respective cells were transfected with DNA coding for the SBP-tagged sybody Sb1 and endogenous LRRC8 channels were pulled-down during sybody purification. Channels purified in this way were only used for subunit quantification and not for structural characterization. In an alternative approach, endogenous LRRC8 channels were isolated from non-transfected cell pellets (grown to a density of up to 5 million cells per ml) using purified SBP-tagged Sb1 bound to Streptactin superflow resin as affinity matrix. For affinity purification from HEK293T and LRRC8 B,D,E-/cells, the pcDXc3MS plasmids containing the construct for Sb1 were transfected into HEK293 GnTIcells grown in suspension culture as

Sample preparation for LC-MS/MS
Samples for the quantification of LRRC8A and LRRC8C subunits were collected either prior or after the final size exclusion chromatography step from protein preparations described above. and NSLSVLSPK (residues 739-747) of LRRC8C. They were selected based on criteria regarding peptide length (7-25 aa's) and the absence of methionines, cysteines, and ragged ends (KR/RR).
In addition, the selected tryptic peptides covered shared sequences in the mouse and human genome and were found to exhibit a linear response in the dynamic range of the detector. Absolute quantified, stable-isotope labeled peptides (SIL) were synthesized as SpikeTides TQL at >95% purity by JPT Peptide Technology GmbH as determined by HPLC, MS and amino acid analysis.

LC-MS/MS analysis
Mass spectrometry analysis was performed on an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific) equipped with a Nanospray Flex Ion Source (Thermo Fisher Scientific) and coupled to an M-Class UPLC (Waters). Solvent composition at the two channels was 0.1% formic acid for channel A and 0.1% formic acid, 99.9% acetonitrile for channel B. Column temperature was 50 °C. For each sample 1 µl of peptides were loaded on a commercial nanoEase MZ Symmetry C18 Trap Column (100 Å, 5 µm, 180 µm x 20 mm, Waters) followed by a nanoEase MZ C18 HSS T3 Column (100 Å, 1.8 µm, 75 µm x 250 mm, Waters). The peptides were eluted at a flow rate of 300 nl/min. After a 3 min initial hold at 5% B, a gradient from 5 to 35% B in 60 min was applied. The column was cleaned after the run by increasing to 95% B and holding 95% B for 10 min prior to re-establishing loading condition for another 10 minutes.
For absolute quantification, the mass spectrometer was operated in parallel reaction monitoring mode with a scheduled (5 min windows) inclusion list using Xcalibur 4.5 (Tune version 4.0), with spray voltage set to 2.3 kV, funnel RF level at 40 %, heated capillary temperature at 275 °C, and Advanced Peak Determination on. Full-scan MS spectra (350−1'500 m/z) were acquired at a resolution of 120'000 at 200 m/z after accumulation to a target value of 3'000'000 or for a maximum injection time of 50 ms. Precursors of heavy and light peptides were selected as given in Extended Data Fig. 1d, isolated using a quadrupole mass filter with 1 m/z isolation window and fragmented by higher-energy collisional dissociation (HCD) using a normalized collision energy of 30 %. HCD spectra were acquired at a resolution of 30'000 and maximum injection time was set to Auto. The automatic gain control was set to 100'000 ions. The samples were acquired using internal lock mass calibration on m/z 371.1012 and 445.1200.
Data-dependent acquisition for the identification of endogenous LRRC8 peptides in isolated complexes was performed with full-scan MS spectra (350−1200 m/z) acquisition at a resolution of 120'000 after accumulation to a target value of 3'000'000, followed by HCD fragmentation for a cycle time of 3 seconds. Ions were isolated with a 1.2 m/z isolation window and fragmented by higher-energy collisional dissociation (HCD) using a normalized collision energy of 30%. HCD spectra were acquired at a resolution of 30'000 and a maximum injection time of 119 ms. The automatic gain control was set to 100'000 ions. Precursor masses previously selected for MS/MS measurement were excluded from further selection for 20 s, and the exclusion window tolerance was set at 10 ppm. The mass spectrometry proteomics data were handled using the local laboratory information management system 45 .

Protein and peptide identification and quantification
For the generation of spectral libraries from SIL peptides, acquired raw MS data were converted into Mascot Generic Format files (.mgf) using Proteome Discoverer 2.1 and the proteins were identified using the Mascot search engine (Matrix Science, version 2.7). Spectra were searched against a reviewed Uniprot proteome database (taxonomy 9606, version from 2019-07-09), concatenated to its reversed decoyed fasta database. Enzyme specificity was set to trypsin and modification to 13C(6)15N(2) for lysine and 13C(6)15N(4) for arginine. A fragment ion mass tolerance of 0.02 Da and a parent ion tolerance of 10 ppm were set. Scheduled parallel reaction monitoring runs of spiked samples were imported into Skyline. Identity assignments were evaluated by determining spectra similarity between endogenous and SIL peptides via dot product.
Endogenous peptide quantification was done by one-point calibration using the ratio of the endogenous and SIL peptides and is given in fmol on column 46 Fig. 2a).

Cryo-EM sample preparation and data collection
Cryo-EM grids were frozen immediately after purification. For grids of LRRC8A/C from transiently expressed proteins, samples were concentrated to a final concentration of 2-5 mg ml -1 .
Endogenous LRRC8A/C channels were concentrated to a total protein concentration of 0.075-0.2 mg/ml. For the analysis of these channels, purified, tag-free Sb1 was added to the purified complexes at a 1:1.5 molar excess (based on LRRC8 subunits) directly before grid freezing.
Homomeric LRRC8C was concentrated to 5 mg ml -1 . Aliquots were either frozen directly or after addition of Sb1 at a 1:1.5 molar excess (per LRRC8C subunit) as a negative control. For classification, a previously determined map of LRRC8A/Sb1 (ref. 27 ) was used as a reference after low-pass filtering to 60 Å. In further processing steps, the respective best maps at each stage were used as references after low-pass filtering to 40 Å. Particles subjected to 3D auto-refinement were masked with soft masks encompassing only protein density. In datasets of LRRC8C and LRRC8A/C 1:1 /Sb1, when the reported resolution reached the Nyquist limit, selected particles were re-extracted with two-fold binning (336-pixel box size, 1.302 Å/pixel) and subjected to iterative 3D auto-refinement, per-particle motion correction 56 and per-particle CTF correction 52 . To improve resolution of the LRRC8C channel, C7 symmetry was applied during 3D refinement of the full-length protein as well as the focus-refinement of the TM domain. In the LRRC8A/C 1:1 /Sb1 dataset, polished particles were subjected to further iterative 3D classification in C1 without alignment, followed by the refinement to separate assemblies of A:C ratio other than 2:1. Despite of two other low-resolution classes emerging, the predominant class displayed a 2:1 arrangement.
Masked local refinement maintaining C1 symmetry of the TM domain, the ESD containing the selectivity filter and a pair of tightly interacting LRRDs from A subunits with bound Sb1 increased the resolution of these regions compared to the resolution of the full-length channel. The same approach of masked 3D classification without particle alignment was applied for the LRRC8A SAM /C dataset, but it did not discriminate between the alternate arrangements of the A and C subunits. The resolution of all generated maps was estimated using a soft solvent mask and based on the gold standard Fourier Shell Correlation (FSC) 0.143 criterion [57][58][59][60] . The cryo-EM densities, except for the endogenous LRRC8A/C/Sb1, were also sharpened using isotropic bfactors.

Cryo-EM model building and refinement
All models of LRRC8 channels and their sybody complexes were built into cryo-EM density with features were used in the subsequent 3D classification with no symmetry applied. In each dataset, one out of four classes showed a symmetric pore domain and seven, albeit flexible, LRRDs. The distribution of particles (%) is indicated. Particles assigned to the boxed classes were used in independent refinements with C7 symmetry applied. As all three reconstructions show the same features, all particles belonging to these refined models were pooled and subjected to per-particle motion correction. Polished particles were subsequently used as input for C7-symmetrized focused refinement of the full-length channel and TM region. Insets show the masked regions during refinement. c, Angular distribution plot of all particles included in the final reconstruction of the full-length complex. The length and color of cylinders correspond to the number of particles with respective Euler angles. d, e, Final 3D reconstruction colored according to local resolution (left) and FSC plot (right) of the final refined unmasked (yellow), masked (red), phaserandomized (black) and corrected for mask convolution effects (blue) cryo-EM density map of the LRRC8C channel. The resolution at which the FSC curve drops below the 0.143 threshold is indicated. d, Full-length complex and e, masked PD. f, Cα representation of a single subunit of LRRC8C with cryo-EM density of the entire protein at 4.6 Å (contoured at 5 σ, left) and of its PD at 4.1 Å (contoured at 10 σ, right) shown superimposed as grey mesh. Fig. 4 | Structure of the LRRC8A/C channel in complex with Sb1. a, Representative 2D class averages of the LRRC8A/C 1:1 /Sb1 complex. b, Data processing workflow. To preserve the unique structural details, all 3D classification and refinement steps were done without applying symmetry in C1. Eight classes generated during 3D classification reveal a well-defined pore domain and structural heterogeneity in the LRRDs, which reflects their intrinsic mobility. Particles assigned to the boxed classes containing at least one ordered pair of LRRDs with Sb1 bound were used in further refinement. The distribution of particles (%) is indicated. 3D refinement using all selected particles as input resulted in a reconstruction with an overall resolution of 3.8 Å. To recover less abundant assemblies, multiple rounds of 3D classification without alignment step followed by 3D refinement were performed. By this approach, particles were assigned to three different classes, each showing unique assemblies. The focused refinement of the PD and the selectivity filter of the most abundant class improved its resolution to 3.3 Å and 3.1 Å, respectively. Focused refinement of a tightly interacting LRRC8A domain pair in complex with Sb1 strongly improved its density and increased the resolution of this part of the structure to 3.8 Å. The view of the domain pair is rotated by 60º with respect to the full-length refined reconstruction. Insets show the masked regions during refinement. c, Angular distribution plot of all particles included in the final reconstruction of the full-length complex. The length and color of cylinders correspond to the number of particles with respective Euler angles. d, e, Final 3D reconstruction colored according to local resolution (top) and FSC plot of the final refined unmasked (yellow), masked (red), phase-randomized (black) and corrected for mask convolution effects (blue) cryo-EM density map of the LRRC8A/C 1:1 /Sb1 complex. The resolution at which the FSC curve drops below the 0.143 threshold is indicated. d, Full-length complex and e, masked TM region. Sections of the cryo-EM density of the LRRC8A/C 1:1 /Sb1 dataset superimposed on the refined model. Subunits are labeled as in Fig. 3. a, Density of the PD at 3.3 Å superimposed on each of the six subunits. The PD of respective subunits is shown left, the first membrane-spanning α-helix TM1 right. The density is contoured at 12 σ for A1-A4 and at 8 σ for C1 and C2. b-d, Masked density around a pair of tightly interacting LRRC8A domains in complex with Sb1 at 3.8 Å (contoured at 15 σ). b, LRR Domain pair with bound Sb1. c, Leucine rich repeats 14 (left) and 7 (right) of the LRRD of subunit A1. d, LRRD-Sb1 interaction interface (left) and open book representation with the LRRD of subunit A1 shown in the center and Sb1 on the right. e-g, Selectivity filter with cryo-EM density of the masked region at 3.1 Å (contoured at 12 σ) shown superimposed. e, Pore constriction viewed from the extracellular side. f, View of the inside of the filter parallel to the membrane. Top, subunits A1, A2, and A3, bottom, subunits A4, C1 and C2. g, Interaction interface of the ESD between subunits A4 and C1. a-g The protein is shown as sticks, the cryo-EM density as grey mesh. Fig. 7 | Structure of channels containing a genetically modified LRRC8A subunit. a, Close-up of a SAM domain fused to the truncated first extracellular loop of LRRC8A, connecting residue 67 at the end of the β-strand E1β with residue 93 preceding the α-helix E1H. The inset shows the SAM domain fitted into cryo-EM density of the homomeric LRRC8A SAM channel. b, Cryo-EM density of the homomeric LRRC8A SAM construct at 6.9 Å, showing a channel with similar conformational properties as observed in the structure of the unlabeled LRRC8A subunit. The views from within the membrane (left) and the extracellular side (right) show extra density corresponding to fused SAM domains for all subunits in the C3 symmetrized map. c, Quantification of absolute peptide levels in isolated LRRC8A SAM /C channels using LC-MS/MS. Peptide amounts on the column (fmol) were calculated using the ratio of endogenous and SIL peptides (as defined in Extended Data Fig. 1d and e). d, Ratio determination of LRRC8A to LRRC8C in isolated complexes using LC-MS/MS. All pairwise ratios of LRRC8A peptides relative to LRRC8C peptides were obtained as described in Fig. 1b. Absolute peptide levels calculated by spiking each sample with known amounts of stable isotope-labeled peptides were used for ratio determination. Boxplots cover the first and third quartiles from bottom to top and the whisker extends to largest/smallest value but no further than 1.5×IQR (inter-quartile range). The median ratio is indicated by a black solid line. e, Representative 2D class averages of the LRRC8A SAM /C complex. f, Data processing workflow. To preserve the unique structural details all 3D classification and refinement steps were carried out in C1. Four classes generated during 3D classification reveal a well-defined pore domain and structural heterogeneity in the LRRDs, which reflects their intrinsic mobility. Particles assigned to the boxed class with one ordered pair of LRRDs were used in further refinement. The distribution of particles (%) is indicated. 3D refinement using all selected particles as input resulted in a reconstruction with an overall resolution of 8.6 Å (not shown). As this reconstruction indicated that both populations of the LRRC8A SAM /C complex with distinct subunit arrangment might be averaged, iterative focused 3D classification with a mask around the subunits with stable positions followed by 3D refinement was performed to separate these two populations. Despite the described efforts, the final refined map still contains an average of two assemblies. g, Angular distribution plot of all particles included in the final reconstruction of the full-length complex. The length and color of cylinders correspond to the number of particles with respective Euler angles. h, FSC plot of the final refined unmasked (yellow), masked (red), phase-randomized (black) and corrected for mask convolution effects (blue) cryo-EM density map of the LRRC8A/C complex. The resolution at which the FSC curve drops below the 0.143 threshold is indicated. i, Cryo-EM density of the LRRC8A SAM /C complex showing a channel with two populations averaged in a consensus reconstruction. Subunits are colored accordingly: A -orange, C -blue, mixture of A and C -grey. The insets show the schematic arrangement of A and C subunits in the two distinct channel assemblies.