Gain-of-Function Mutation W493R in the Epithelial Sodium Channel Allosterically Reconfigures Intersubunit Coupling

Sodium absorption in epithelial cells is rate-limited by the epithelial sodium channel (ENaC) activity in lung, kidney, and the distal colon. Pathophysiological conditions, such as cystic fibrosis and Liddle syndrome, result from water-electrolyte imbalance partly due to malfunction of ENaC regulation. Because the quaternary structure of ENaC is yet undetermined, the bases of pathologically linked mutations in ENaC subunits (cid:2) , (cid:3) , and (cid:4) are largely unknown. Here, we present a structural model of heterotetrameric ENaC (cid:2) 1 (cid:3)(cid:2) 2 (cid:4) that is consistent with previous cross-linking results and site-directed mutagenesis experiments. By using this model, we show that the disease-causing mutation (cid:2) W493R rewires structural dynamics of the intersubunit interfaces (cid:2) 1 (cid:3) and (cid:2) 2 (cid:4) . Changes in dynamics can allosterically propagate to the channel gate. We demonstrate that cleavage of the (cid:4)

The epithelial sodium channel (ENaC) 3 regulates Na ϩ absorption by epithelia, thereby maintaining essential waterelectrolyte balance (1,2). Therefore, altered ENaC activity con-tributes to pathological conditions that are partly mediated by sodium transport dysregulation, such as cystic fibrosis and Liddle syndrome (1). ENaC is composed of structurally homologous subunits ␣, ␤, and ␥. Mutations in the ␣-subunit, such as the gain-of-function mutation ␣W493R in exon 10, decrease lung functionality (3)(4)(5)(6)(7). Carrying the ␣W493R polymorphism in ␣ENaC can result in chronic bronchitis (4). Heterologous expression of ␣W493R with ␤ and ␥ subunits in Xenopus oocytes produced an ϳ4-fold increase in ENaC-mediated current and an increase in the channel open probability (8). However, the structural determinants of this activation remain unknown.
Wild type ENaC is normally proteolytically activated by intracellular furin-like convertases and extracellular trypsinlike serine proteases (9 -12). Cleavage of an inhibitory fragment in the finger domain of the ␣-subunit is thought to reorganize the extracellular domains to favor active conformations of the channel (13). Kashlan et al. (13) have elucidated the mechanism of inhibition by the cleaved fragment using synthetic peptides derived from the inhibitory tract of the ␣-subunit. Releasing the inhibitory tract by proteolysis was found to activate ENaC by eliminating interactions at the thumb and finger domains of the ␣-subunit. Moreover, previous computational studies on the mechanism of cleavage of ENaC by furin suggest high intrinsic disorder at the furin cleavage sites (9). However, the sequence of proteolytic events and conformational changes leading to channel activation by proteases are still unclear.
Structure-guided design of inhibitors for ENaC has been challenged by a lack of structures for ENaC subunits. In the ENaC/Degenerin family, only the acid-sensing ion channel (ASIC) has been crystalized as a homotrimer in different biophysical states, i.e. pH-and toxin-stimulated (14 -16). However, functional data show heteromeric assembly of different ASIC subunits. ASICs can exhibit alternative cation permeation functionality by heteromeric assembly of ASIC2a and ASIC2b subunits (17). Although ASIC and ENaC subunits can assemble to form hybrid channels (18), fully functional ENaC oligomers can form only by the assembly of the ␣-, ␤-, and ␥-subunits. However, channels composed of only ␣, ␣ and ␤, or ␣ and ␥ can still conduct sodium (12). Recently, ␣-like subunits, ␦ and ⑀, have been discovered; however, their physiological roles have not been characterized fully (19,20). Functional data obtained in Xenopus oocytes support an oligomeric model of ENaC in which one ␤-, one ␥-, and two ␣-subunits line the pore (21). Moreover, biochemical analysis of epithelial tissues iso-lated from rabbit colon and cultured kidney cells indicates a heterotetrameric stoichiometry (22). In contrast, atomic force microscopy studies suggest a heterotrimeric architecture of ENaC (23). The physiological relevance of in vitro heteromeric assembly of ENaC is unclear because proper assembly of the ␣-, ␤-, and ␥-subunits in epithelial cells requires trafficking and sorting machinery (24,25). Consequently, the stoichiometries of physiologically relevant ENaC oligomers remain uncertain.
These conflicting findings raise the following question. How reliable are classical structure determination techniques in identifying physiologically relevant molecular species? We have combined computational modeling, cysteine cross-linking, and electrophysiology experiments to derive structural evidence for tetrameric ENaC with configuration ␣ 1 ␤␣ 2 ␥. In support of previous findings, we propose a dynamics-driven model of ENaC activation by the gain-of-function mutation ␣W493R; in this model, proteolysis of the ␥-subunit is not the major determinant of activation.

Experimental Procedures
Building the Heterotetramer ␣ 1 ␤␣ 2 ␥ Model by Symmetric Docking-Previously, we built a homology model of the rat ENaC ␥-subunit based on the crystal coordinates of a homolog of ENaC, chicken acid-sensing ion channel cASIC (1,26). We used the same homology-based approach to construct structural models of the ␣and ␤-subunits. We used our in-house suite MEDUSA for threading ENaC subunit sequences on ASIC structure (27,28). Alignments of ASIC and ENaC subunits indicated non-homologous insertions and deletions. We used MODELLER to model non-homologous sequences as loops (29). These fragments were then ligated into the homology models designed by MEDUSA and minimized by short discrete molecular dynamics simulations (DMD) (30 -33). Briefly, DMD is an event-driven sampling engine that is faster than traditional molecular dynamics under conditions that do not rely on specific solute-solvent interactions. DMD has been successfully used to probe protein dynamics and design molecular switches (34 -36).
Because of the lack of a tetrameric template in the ENaC/ Degenerin family, we employed symmetric docking using the docking program SymmDock to construct possible tetrameric interfaces of the tetramer ␣ 1 ␤␣ 2 ␥ (37). First, we built a homotetrameric model of the ␣-subunit. Using this homotetramer, we performed structural alignments of the homology models for the ␤and ␥-subunits according to the configuration ␣ 1 ␤␣ 2 ␥. Finally, we used DMD to minimize and relax the model for production run simulations. SymmDock samples the homooligomeric configurational space with symmetry C n from a single monomer using a rigid body docking algorithm with symmetric constraints (37). SymmDock utilizes geometric complementarity of molecular surfaces to symmetrically dock atomic patches. Similar to most docking protocols, the algorithm consists of three stages as follows: molecular representation, patch matching, and model clustering and filtering. Models of oligomeric interfaces are grouped into clusters by two geometric metrics: the direction of axis of symmetry and the projection of center of mass of the asymmetric unit on the axis of symmetry. Clusters are then ranked according to the number of patch-matching pairs pointing to the same axis of symmetry. From the top 100 models ranked by SymmDock's score, one model formed interfaces that correctly arranged the extended conformations of the selectivity filter sequences ␣GSS, ␤GGS, and ␥SCS in a rhombus-like architecture. One drawback of SymmDock is the lack of a post-docking refinement step, which relieves the docked model from steric clashes and re-optimizes side chain geometric complementarity. Therefore, we utilized our in-house developed computational protocols of short DMD simulations for structural refinement. The DMD minimization protocol is publicly available on our Chiron server (38). Chiron can eliminate infinite energy atomic interactions that resemble nonphysical van der Waals volume overlap. Chiron uses a series of short discrete molecular dynamics simulations to overcome interatomic clashes. Briefly, Chiron minimization protocol utilizes a high heat exchange rate r Ͼ 0.2 fs Ϫ1 with the thermodynamic bath, which is briefly set to be at a high temperature T Ͼ 0.5 DMD reduced unit, ϳ250°C (27). By rapidly equilibrating the system at a high thermal energy, Chiron quenches infinite energy interactions.
DMD Protocols for ␣ 1 ␤␣ 2 ␥ Minimization and Production Run Simulations-Atomic clashes can be difficult to resolve, especially in hetero-oligomeric complexes, due to the variability in molecular surface complementarity at heteromeric interfaces. In this study, we found that adding a hard harmonic potential with force constant k Ͼ1 kcal mol Ϫ1 Å Ϫ2 to all backbone atoms helps in equilibrating the system. Because there are four different interfaces in the heterotetramer ␣ 1 ␤␣ 2 ␥, additional refinement steps were needed to resolve all steric clashes. First, we conducted a brief packing simulation, 1000 DMD time steps, at heat exchange rate 100 fs Ϫ1 and temperature 0.1, while keeping the backbone rigid by defining infinite mass to backbone atoms. Following packing, we utilized a combination of relaxation steps similar to steepest descent protocols. Keeping the temperature at 0.1, we assigned a harmonic potential of 5 kcal mol Ϫ1 Å Ϫ2 to the C␣ atoms. Then, we increased the temperature to 0.3 and equilibrated the system by decreasing the heat exchange coefficient to 0.1 fs Ϫ1 . Finally, we performed relaxation under unconstrained conditions. The system is ready for production run after minimizing its energy to a stable state. We used our in-house suite MEDUSA to substitute in our relaxed model Trp-493 for Arg-493 in the human ␣-subunit sequence (39). To simulate the structural dynamics of the constructed and minimized heterotetramer in wild type and mutant states, we assigned soft harmonic constraints to all backbone atoms of the two transmembrane domains (TM1 and TM2) in all subunits to mimic protein-lipid interactions. The applied harmonic potential had a force constant of 0.1 kcal mol Ϫ1 Å Ϫ2 . This parameter allowed the transmembrane backbone atoms to sample more confined configurations than the sampled conformations of the solvent-exposed extracellular domains.
For surface biotinylation and cysteine cross-linking, groups of 80 -120 oocytes were coinjected with cRNA encoding WT or mutant ENaC subunits ␣, ␤, and ␥ and incubated overnight in After biotinylation, oocytes were washed three times. Oocytes were lysed in 20 mM Tris, 50 mM NaCl, 50 mM NaF, 10 mM ␤-glycerophosphate, 5 mM Na 2 P 2 O 7 , 1 mM EDTA at pH 7.4 containing protease inhibitors (Complete, Roche Applied Science). Oocytes were lysed by passing them twice through a 27 1 ⁄ 2-gauge needle. Cell lysates were centrifuged at 4°C at 3200 rpm to separate supernatants from cell pigments and debris. Supernatants were then collected and transferred into new tubes and centrifuged at 4°C at 14,000 rpm to pellet membrane fractions. Membrane fractions were then solubilized in 50 mM Tris, 100 mM NaCl, 1% Triton X-100, 1% Nonidet P-40, 0.2% SDS, 0.1% sodium deoxycholate, 20 mM NaF, 10 mM Na 4 P 2 O 7 , 10 mM EDTA, and protease inhibitor mixture at pH 7.5. Solubilized membrane fractions were incubated overnight with 100 l of prewashed neutravidin beads (Pierce), while gently tumbling at 4°C. Samples were washed twice with 500 mM NaCl, 50 mM Tris at pH 7.5, and once with 150 mM NaCl, 50 mM Tris at pH 7.5. Samples were then incubated at 96°C for 5 min and separated by SDS-PAGE in a 7.5% acrylamide gel. Samples were transferred to a nitrocellulose membrane, which was then probed with the anti-␣ENaC mAb UNC1 19.2.1 recognizing an N-terminal epitope of human ␣ENaC that was produced by the University of North Carolina Immunology Core Facility (40). Bands of anti-␣-ENaC were detected by applying IRDye 680conjugated goat anti-mouse IgG (Invitrogen) or IRDye 800conjugate goat anti-rabbit IgG (Rockland Immunochemicals) and imaged using an Odyssey infrared imaging system (LI-COR Biosciences).
Electrophysiological Measurements-Recordings of amiloridesensitive sodium currents were measured as described (26). For each construct, WT, or mutant channel, oocytes were coinjected with 0.3 ng of each cRNA encoding ENaC subunits ␣, ␤, and ␥. 24 h after microinjection, two electrode voltage clamping was performed using a Geneclamp amplifier (Axon Instruments) in a constant perfusion system. Oocytes were voltage-clamped at Ϫ100 mV. Currents were digitized and recorded using a Digidata 1200 A/D converter (Axon Instruments) and AxoScope software. Recorded traces were analyzed using the pCLAMP software. Recordings were initiated in the presence of 10 M amiloride. Following amiloride wash, basal currents were recorded in frog Ringer's solution, 2.5 mM KCl, 2.5 mM CaCl 2 , 120 mM NaCl, and 10 mM Hepes at pH 7.35, to obtain I Na .
Results are from 5 to 6 oocytes per experimental group, repeated in 2-4 separate batches of oocytes.

Results
Quaternary Architecture of ENaC Tetramer ␣ 1 ␤␣ 2 ␥-In our ␣ 1 ␤␣ 2 ␥ model, ENaC subunits are symmetrically arranged around the channel pore axis (Fig. 1A). Each subunit forms contacts at intersubunit interfaces that stabilize the complex by inter-residue interactions. Although the different subunits share a common topology of highly conserved domains, such as the palm and ␤-ball domains in the core of each subunit, slight differences in sequence result in unique intersubunit interfaces (Fig. 2). Intersubunit contacts span the transmembrane domains in addition to extracellular domains, such as the knuckle-knuckle and knuckle-finger interdomain interactions. Most stabilizing intrasubunit contacts occur in the core region, with the largest surface area spanning the palm and ␤-ball domains (Fig. 2). ENaC subunits assemble by forming critical intersubunit interactions. These contacts include knuckle-finger and finger-finger interdomain interactions. The insertion of a second ␣-subunit changes the intersubunit angle from 120°in the crystallized trimeric configuration of cASIC to 90°in the tetramer model. Consequently, the modes of assembly of the pore-lining domains, TM1 and TM2, are slightly different between the trimeric and tetrameric configurations. Therefore, the conductive pore of ENaC can adopt somewhat different structures depending on the oligomeric state of the channel. In contrast, both oligomeric states result in a selectivity filter with a comparable constriction size of ϳ4 Å (Fig. 1A).
Similar to other multisubunit ion channels, the biophysical features of the heterotetramer's pore are determined by the geometric complementarity between subunits as the complex assembles from its monomers. Consequently, the pore contains fenestrations in the extracellular region occupied by water, ions, and small molecules such as amiloride or benzamil that can block the pore. Conversely, constrictions in the channel are formed by unique intersubunit contacts, such as the interfaces formed by the knuckle domains, where the site of the diseasecausing mutation ␣W493R resides (Fig. 1A). Substitutions of this tryptophan to arginine, lysine, cysteine, or glutamic acid activate the channel (8). Therefore, the structural organization created by intersubunit interactions can play key functional roles in activating ENaC or stabilizing closed states of the channel. Finally, the gate and the selectivity filter of the channel form the narrowest region along the pore (Fig. 1B). Our structural model of ENaC heterotetramer agrees with previously pub-lished cross-linking data (41). At the ␣ 2 ␤ interface, ␣Lys-477 is in close proximity to cross-link with ␤V85. Similarly, at the ␣ 2 ␥ interface, ␣Leu-120 can cross-link with ␥Glu-455 (Fig. 3, A  and B).
Asymmetry in Arg-493 Dynamics in the Knuckle Domains of the ␣ 1 -and ␣ 2 -Subunits-To investigate the effect of the gainof-function mutation ␣W493R on the structural dynamics of ENaC, we performed DMD simulations of ENaC tetramer ␣ 1 ␤␣ 2 ␥. From our simulations, we predicted key inter-residue interactions. We tested these predictions by site-directed mutagenesis and electrophysiology recordings of oocytes. In our model, ENaC subunits are arranged in ␣ 1 ␤␣ 2 ␥ clockwise order such that ␣ 1 complements ␤ and ␤ complements ␣ 2. Therefore, this model defines interactions by Trp-493 in each ␣-subunit to be physiochemically distinct. Our homology model for the ␣-subunit derived from the cASIC crystal structure positions the Trp-493 rotamer in a pocket inside the knuckle domain cavity, which is geometrically constrained or stabilized by two phenylalanine residues Phe-280 and Phe-503 (Fig. 4). Substituting Trp-493 with an arginine reveals interresidue interactions that can be either polar through H-bonding by backbone and side chain atoms or non-polar through the hydrophobic side chain of arginine.
Based on DMD simulations, we identified binding poses of the ␣ 1 Arg-493 rotamer buried in the knuckle domain pocket of the ␣ 1 -subunit at the ␣ 1 ␤ interface. We analyzed the backbone dynamics of the knuckle domain by computing the root mean square fluctuations (RMSF) of C␣ atoms. We found that the RMSF of the Arg-493 mutant knuckle domain in the ␣ 1 subunit are increased compared with those of the WT channel (Fig. 5B),   (35) and two previously proposed tetrameric configurations from functional studies of covalently linked concatemers (16). Cross-linking data from Collier et al. (41) suggest a trimeric clockwise topology of ␣␥␤ (35). These data also agree with two coexisting functional tetrameric configurations. Intersubunit angles vary slightly between trimer and tetramer models, resulting in slightly different interfaces as follows: in clockwise order, ␣␤, ␤␥, and ␥␣, in addition to two unique interfaces to tetramers ␣ 1 ␣ 2 and ␣ 1 ␥. B, schematic representation of ␣ 2 ␤ and ␣ 2 ␥ interfaces in the tetramer model ␣ 1 ␤␣ 2 ␥. Shown as sticks, ␤V85 can cross-link with ␣Lys-477, and ␥Glu-455 can cross-link with ␣Leu-120. Cross-linkers are represented as red bars. FEBRUARY 19, 2016 • VOLUME 291 • NUMBER 8 possibly due to reconfiguration of the knuckle domain pocket, in which arginine and tryptophan adopt slightly different geometries. Interestingly, the ␣ 2 Arg-493 structural dynamics appear to be different from WT in the opposite ␣-subunit that forms an interface with the ␥-subunit. In contrast, fluctuations are slightly decreased in the opposite ␣ 2 -subunit (Fig. 5A) due to stable electrostatic intersubunit interactions of ␣ 2 Arg-493 with neighboring acidic residues such as ␥Glu-348 and ␥Glu-346 (Fig. 6C).

Structural Dynamics of Pathological Activation of ENaC
Dissimilarity in the backbone dynamics of the ␣-subunits enhances fluctuations of the ␥-subunit in the palm domain around ␥Glu-348 (Fig. 5C). Altered dynamics in the ␥-subunit allosterically propagate down the channel through ␤-sheets in the palm domain to ␥Tyr-369. This tyrosine residue has been proposed previously to communicate extracellular signals to  . Structural dynamics modulation in ␣ 1 ␤␣ 2 ␥ by W493R gain-offunction. RMSF of C␣ atoms in the ␣and ␥-subunits are from nine different 50-ns DMD trajectories; standard error is plotted for each residue. RMSF analysis is applied on equilibrated trajectories after discarding the first 5 ns. A and B, W493R destabilizes the knuckle domain of the ␣ 1 -subunit near the Trp-493 site. However, in the ␣ 2 -subunit, dynamics are slightly stabilized due to electrostatic interactions between ␣ 2 Arg-493 and ␥Glu-348 at the ␣ 2 ␥ interface. C, changes in dynamics of the ␥-subunit propagate along a ␤-strand and enhance fluctuations of a loop, containing the signaling tyrosine Tyr-369, at the extracellular-transmembrane interface that communicates the activation signal (D); thus, fluctuations around the gate region containing the selectivity filter increase slightly. FIGURE 6. W493R rewires intersubunit interactions. A, DMD simulations indicate an electrostatic interaction between ␣ 2 Arg-493 (red) and ␥Glu-348 that is absent in the WT ␣ 2 -subunit (black). Inter-residue interactions are quantified by a histogram of minimum distance between ␣ 2 Arg-493 and ␥Glu-348. B, ENaC currents measured in oocytes injected with cRNA of WT and mutant ENaC subunits; basal current is normalized to trypsin-stimulated current to reflect basal activation. Perturbing the electrostatic interaction by the double mutant ␥E348R/␣W493R channels shows no measurable population of near-silent channels, shown by lack of activation by trypsinization (n ϭ 18, p Ͻ 0.0001). C, schematic representation of snapshots from DMD simulations at the ␣ 2 ␥ interface. Trp-493/Arg-493 in the ␣ 2 -subunit (light gray) and Glu-348 in the ␥-subunit (orange) are shown as sticks, and hydrogen bonding between ␥Glu-348 and ␣ 2 Arg-493 is shown by red dashed lines. D, hydrogen bond network at the ␣ 1 ␤ intersubunit interface by the acidic patch 303 QED 305 in the ␤-subunit with the knuckle domain in the ␣ 1 -subunit. E, amiloridesensitive current measured in oocytes. Neutralizing the acidic patch by alanine substitutions decreases W493R activation by ϳ2-fold (n ϭ 12, p ϭ 0.002). ␤ 303 AAA 305 slightly inhibits WT ENaC (n ϭ 12, p ϭ 0.008).
the transmembrane domain TM2 containing the selectivity filter, the rate-determining structure for sodium selectivity and permeation (41). These results show that extracellular fluctuations at intersubunit interfaces can allosterically propagate down the channel to the gating region.
Rewiring of Intersubunit Interactions Mediates ␣W493R ENaC Activation-The aspartic acids in the ␥-subunit, ␥Glu-346 and ␥Glu-348, lie on a ␤-strand in the palm domain that is thought to stabilize the overall fold of ENaC subunits (9 -11). In DMD simulations, ␥Glu-346 and ␥Glu-348 made statistically significant polar interactions with the side chain of ␣ 2 Arg-493 in the ␣ 2 -subunit (Fig. 6, A and C) (supplemental Movies 1 and 2). In addition to the di-aspartate moiety, other neighboring acidic residues make polar contacts with ␣ 2 Arg-493 at a lower frequency, ϳ10%. A histogram from time series of the minimum inter-residue distance between ␣ 2 Arg-493 and ␥Glu-348 indicates the stability of the ␣ 2 Arg-493 electrostatic interactions with ␥Glu-348 (Fig. 6A). Wild type ␣ 2 Trp-493 forms no significant contacts with ␥Glu-34; however, ␣ 2 Arg-493 forms stable electrostatic contacts indicated by an average contact distance of ϳ3 Å (Fig. 6A). To verify these interactions, we substituted ␥Glu-348 with an arginine and tested the effect of the electrostatic substitution ␥E348R on the function of wild type and ␣W493R ENaC in the heterologous expression system of Xenopus oocytes. The ␥E348R substitution cooperated with ␣ 2 W493R to fully activate the channel, indicated by no measurable response to activation by trypsin (Fig. 6B).
Because activation of ENaC by ␣W493R may be accounted for by an increase of cleavage in the ␥-subunit, we collected lysates from oocytes that were used to measure ENaC currents and performed immunoblotting with an anti-HA antibody to detect N-terminal HA-tagged ENaC. We detected no difference in endogenous cleavage between wild type and ␣W493R ␥ENaC (Fig. 7). To further validate the minimal role of proteolysis in the alternative activation mechanism by ␣W493R, we substituted the basic residues 135 RKRR 138 in the ␥-subunit with 135 QQQQ 138 , thereby eliminating cleavage by furin. Furin cleavage at this site is thought to be a rate-determining step for proteolytic activation of ENaC and a pre-requisite step for the final cleavage events by extracellular proteases such as trypsin (9,10). ␣/␤/␥ 135 QQQQ 138 and ␣W493R/␤/␥ 135 QQQQ 138 showed no trypsin response (Fig. 7). Moreover, ␣W493R/␤/ ␥ 135 QQQQ 138 currents were comparable with the basal currents measured from ␣W493R/␤/␥ channels. In agreement with our hypothesis, these protease-resistant channels still recapitulated the gain-of-function phenotype in ENaC by ␣W493R. We perfused oocytes with 20 g/ml trypsin to ensure full activation of proteolytically susceptible channels.
The double mutant channel ␥E348R/␣W493R was insensitive to further stimulation by trypsin, indicating that the channel reached its maximal level of activity (Fig. 6B). These findings  FEBRUARY 19, 2016 • VOLUME 291 • NUMBER 8 suggest that intersubunit conformational changes induced by electrostatic interactions around the channel pore facilitate opening of the channel independent of the cleavage state of the channel. Because our DMD simulations and electrophysiology experiments provided supporting evidence for the functional role of electrostatic intersubunit interactions mediating ENaC activation, we hypothesized a functional role of the local electric field around ␣ 1 Arg-493 at the ␣ 1 ␤ intersubunit interface. Therefore, we designed a mutant that neutralizes an acidic patch on the extracellular surface of the ␤-subunit. This mutant neutralizes the acidic charge on 303 QED 305 by substituting QED to AAA. The QED patch is in direct contact with the knuckle domain of the ␣ 1 -subunit (Fig. 6D). We found from our DMD simulations formation of polar contacts by hydrogen bonding between the knuckle domain of the ␣ 1 -subunit and the QED patch in the ␤-subunit (Fig. 6D). The AAA mutation resulted in a slight inhibition of wild type ENaC. Conversely, 303 AAA 305 ␤, when coexpressed with wild type ␥ and W493R ␣, negates the W493R gain-of-function effect by more than 2-fold (Fig. 6E). These findings support our hypothesis that the gain-of-function effect by W493R is mediated partly by changing intersubunit conformational changes, in addition to possibly altering the local electric field at the channel pore.

Structural Dynamics of Pathological Activation of ENaC
Cross-linking the ␣␥ Interface at ␣W493C and ␥E348C-Our molecular dynamics simulations of ENaC tetramer model and electrophysiology data suggest an interaction between ␣ 2 Arg-493 and ␥Glu-348. To further validate the spatial proximity between the two residues, we used cysteine cross-linking. We introduced cysteine substitutions at these amino acid positions and applied the thiol-reactive cross-linker M2M to intact oocytes that expressed the cysteine mutant channels. We hypothesized that by cross-linking the introduced cysteines, ␣Cys-493 and ␥Cys-348, we can biochemically resolve the ␣␥ dimer. In agreement with our hypothesis, cross-linking ␣W493C/␤/␥E348C channels produced a molecular species with a molecular mass of ϳ175 kDa that corresponds to the ␣␥ dimer (Fig. 7B). This result supports our findings from molecular dynamics simulations and electrophysiology data that suggest an interaction between ␣W493R and ␥E348R in the tetrameric ENaC complex.
Allosteric Propagation of 493R's Asymmetric Dynamics in the Knuckle Domain to ␥TM2-Previous molecular dynamics simulations investigating the structural dynamics of the ASIC homotrimer suggested the importance of an allosteric pathway, in which twisting and vibrational motions in the knuckle, finger, and thumb domains communicate extracellular signals to the channel transmembrane domains (42). We probed this hypothesis using coevolution information of residues in the subunits of ion channels in the ENaC/Degenerin family. We used ASIC as a reference sequence and structure. Each node in the coevolution network represents a residue, and each edge is quantified by a mutual information metric that was calculated by the Mutual Information Server to Confer Coevolution (MISTIC) (43). We superimposed the mutual information edges on the ASIC structure and found a mutually evolved set of residues that line an allosteric structural path that is similar to the path previously proposed in the allosteric model of activa-tion of ASIC (Fig. 8). High sequence conservation of residues on this path indicated a conserved allosteric network ( Table 1). The path contains residues in TM2, in addition to ␥Tyr-369 or Trp-288 in cASIC, that are involved in allosteric signaling by transmitting extracellular signals to the transmembrane domain bystacking interactions (26,44). Finally, a subset of that allosteric path is a linear path on the thumb domain, which is in contact with the ␤-ball and the knuckle domains (Fig. 8). Moreover, this coevolution analysis reveals possible allosteric mechanisms of a loss-of-function mutation in ENaC that is present in pseudohypoaldosteronism (PHA-1) patients. A highly conserved motif, HG, is mutated to HS in PHA-1 patients, altering the gating of ENaC, possibly by changing the conformational dynamics of the intracellular gate region (45). The HG motif, present in all members of the ENaC/Degenerin family, possesses a high number of coevolution contacts, in addition to exhibiting a high conservation score (Table 1). By integrating mutual information from coevolution and the structure of ASIC, we suggest that the allosteric parameters, which enable ENaC/Degenerin family to process extracellular clues, are encoded in the global fold of the channels.
From our DMD simulations, we suggest a rewiring in the ␣W493R mutant inter-residue interaction network, mostly mediated by H-bonding and possibly salt bridges, to propagate from the knuckle domain of the ␣-subunit to the palm domain of the ␥-subunit. The palm domain extends to a signaling tyrosine, ␥Tyr-369, which is in contact with the transmembrane domain TM2 that facilitates channel gating (Fig. 9). Dynamics information derived from DMD simulations of ENaC heterotetramer suggests an allosteric model for W493R's gain-of-function effect, in which local fluctuations of the Arg-493 rotomeric ensemble drive differential dynamics in the ␣-subunits. Arg-493 in the ␣ 1 -subunit maintains binding poses that are mostly embedded inside the knuckle domain cavity. Consequently, the knuckle domain of the ␣ 1 -subunit in the mutant channel is slightly destabilized compared with the dynamics of wild type ␣ 1 (Fig. 9). Moreover, the electrostatic interactions between ␣ 2 and ␥ stabilize the ␣ 2 ␥ interface. We have used electrophysiology experiments to test the functional effects of perturbing these intersubunit interactions. The proposed model of the structural dynamics induced by W493R supports previous findings from structure-function studies that emphasize the contribution of intersubunit conformational changes in the extracellular vestibule to the structural and dynamic activation determinants of ENaC (41,46).

Discussion
We propose a structural model of activation of the epithelial sodium channel by the pathological gain-of-function mutation ␣W493R. By integrating computational and experimental approaches, we identified critical intersubunit interactions that mediate allosteric activation of the channel. First, we revisit the previously suggested model of activation by ␣W493R, in which proteolytic cleavage in the ␣-subunit was found not to contribute to the gain-of-function of the channel (8). We further expand this model by showing that the gain-of-function by ␣W493R is independent of the cleavage state of the ␥-subunit, which is considered the rate-determining step of proteolytic activation (9, 10). Our structural model of the heterotetramer ␣ 1 ␤␣ 2 ␥ of ENaC gives atomistic insight into the allosteric determinants that mediate channel activation.
The stoichiometries of ASICs, ENaCs, and other members in the ENaC/Degenerin family have been under debate. FMRFamide gated FaNaCh has been shown to assemble as a homotetramer (47). Functional approaches suggest tetrameric and trimeric assemblies of different ASIC subunits. ASIC1a and ASIC2a can form trimers of flexible stoichiometries 2:1 and 1:2 (48). However, ASIC1a, ASIC2a, and ASIC2b can form heterotetramers (49). Crystal structures of ASIC suggest a homotrimeric architecture (9 -11). ASIC and ENaC subunits can assemble to form hybrid channels with novel functional characteristics (12). In vitro studies show heterotrimerization of ENaC (19). Conversely, functional data suggest a heterotetrameric configuration ␣ 1 ␤␣ 2 ␥ of ENaC. ␣ 1 ␤␣ 2 ␥ has been validated from structure-function characterization of ENaC oligomers using covalently linked concatemers (21). In addition, a heterotetrameric stoichiometry 2␣:1␤:1␥ has been previously determined from epithelial cell lysates extracted from rabbit colon epithelial cells and cultured kidney cells (22). It is evident that the structural plasticity in the ENaC/Degenerin family facilitates formation of channels with diverse functionalities. However, it remains a challenge to reconcile structural data of ENaC assembly from in vitro studies with functional data of physiologically relevant oligomeric interfaces of the channel. The structural plasticity in the assembly of subunits of ion channels is not unique to the ENaC/Degenerin family. The potassium channel KCNQ1 has been crystallized as a trimer and as a tetramer (50). The extracellular domain of the ligandgated ion channel GLIC has been crystalized as a pentamer and as a hexamer (51). Similar to ENaC, MscL is a mechanosensitive ion channel, which is activated by changes in the elastic energy stored in the lipid bilayer (52)(53)(54). MscL forms tetramers, pentamers, and hexamers (52,55). An elastic model of MscL suggests that different oligomeric states can

List of residues in cASIC with highest number of coevolution contacts
From the MISTIC analysis, the top 5% coevolving residues include highly conserved cysteine residues involved in disulfide bridges stabilizing the fold of subunits in the ENaC/Degenerin family. Trp-288 or Tyr-369 in ␥ENaC, which is crucial for communicating allosteric signaling between the extracellular region and the transmembrane domains (20,37), has a large number of 22 coevolution contacts. Highly conserved Gly-30 in the HG motif implicated in the PHA-1 disease also has a high coevolution contact number, 25, suggesting the structural importance of these allosteric sites in mediating the channel function in the ion channel family. A coevolution contact is considered significant by the MISTIC according to a z-score of Ͼ6.5 (36). Highly conserved cysteines are shown in boldface; the allosteric residue W288 is shown in blue; the disease-linked residue G30 is shown in red. FEBRUARY 19, 2016 • VOLUME 291 • NUMBER 8 shape the gating energy landscape and facilitate the contribution of membrane deformation energetics to the activation of the channel (55).

Structural Dynamics of Pathological Activation of ENaC
Rauh et al. (8) have suggested that ENaC gain-of-function by ␣W493R is mediated partly (ϳ40%) by a decrease in sodium self-inhibition of ENaC. Self-inhibition is an inhibitory response of ENaC to sodium binding to the extracellular acidic surfaces at high sodium concentrations that limits conformational changes that shift the conformational equilibrium of the channel toward open states. By releasing inhibitory peptides from the ␣and ␥-subunits, extracellular proteases are thought to relieve the oligomeric complex from inhibitory interactions that stabilize the closed state of the channel. In addition, chloride ions inhibit ENaC by binding at extracellular intersubunit interfaces (41). Chloride binding, proteolysis, and sodium selfinhibition seem to induce a common functional perturbation altering the probability of occupying the closed state of the channel. Similarly, ␣W493R activates ENaC by perturbing inter-residue interactions that stabilize the closed state of the channel. Rauh et al. (8) substituted ␣Trp-493 with Arg, Lys, Cys, and Glu, and all of these substitutions activated the channel, suggesting the importance of ␣Trp-493 in stabilizing the closed state of the channel. We expand this model to include the role of ␣W493R in modulating intersubunit interactions as well as perturbing the dynamics of the knuckle domain. Recently, by cross-linking intersubunit interfaces, Gwiazda et al. (56) have shown the functional importance of the knuckle domain in ASIC. Moreover, active state stabilization by electrostatics has been shown in the erythromelalgia-causing gain-offunction mutation Q875E in the voltage-gated sodium channel Nav1.7 (57).
We have used the functionally validated heteromeric configuration of ENaC as a template for building a molecular model of ENaC tetramer. We tested our model using the experimentally characterized gain-of-function mutation ␣W493R. By probing the structural dynamics mediating the activation of the channel, our study suggests a dynamics-driven model for the previously established functional role of the extracellular domains that participate in sodium and chloride binding and undergo proteolytic cleavage. We used DMD simulations to refine and simulate the molecular model generated by symmetric docking. Our experimentally supported computational approach predicts a functional role of intersubunit structural dynamics mediating the gain-of-function effect of the disease-causing mutation ␣W493R. DMD simulations highlight the role of electrostatic interactions in modulating the intersubunit fluctuations triggered by the knuckle domain dynamics. Moreover, extracellular fluctuations are predicted to propagate through ␤-strands in the palm domain, reaching the signaling tyrosine ␥Tyr-369 that contacts TM2, the sodium permeation rate-determining structure (58). We have shown previously that changing this tyrosine to a lysine slightly slows down the rate of proteolytic activation (26).
Proteolytic activation of the channel relieves inhibitory intrasubunit and possibly intersubunit interactions. To examine whether this mechanism applies to the ␣W493R's mode of action, we created an uncleavable mutant of the ␥-subunit, and complementary to previous findings, we found that ␣W493R's mechanism of action is largely independent of proteolysis. Furthermore, ␣Arg-493 is predicted in one ␣-subunit to interact with ␥Glu-348. By perturbing this interaction by changing Glu-348 to Arg-348, we fully activated ␣W493R channels and further decreased the pool of near-silent channels. We further validated the predicted inter-residue interaction by cysteine cross-linking. Interestingly, the examined electrostatic interaction exists in the tetramer model and is absent in our ENaC trimer models ␣␤␥ and ␣␥␤. We have built models for the two possible trimeric configurations of ENaC, ␣␤␥, and ␣␥␤, using the crystallized homotrimeric topology of ASIC. Intersubunit angles differ slightly between trimeric and tetrameric interfaces. However, small differences in intersubunit orientation can rewire some inter-residue interactions. 4 The suggested electrostatic intersubunit interaction is predicted to fully activate the channel by inducing intersubunit conformational changes that possibly involve repulsive forces mimicking how proteolysis relieves inhibitory interactions. In agreement with our model, Collier et al. (41) showed that, by increasing the length of intersubunit cross-linkers, ENaC was further activated. Moreover, Edelheit et al. (59) have shown the functional importance of highly conserved charged residues on the extracellular surface of ENaC.
Lack of ENaC structural models for functionally characterized oligomers hinders progress in understanding the structural mechanism(s) underlying the activation of the channel. What are the structural and dynamic determinants that facilitate channel opening? How do gain-and loss-of-function mutations implicated in disease modulate the mechanism of action of the channel in processing extracellular and intracellular clues? These questions will remain largely unanswered until functional data from electrophysiology experiments can be explained mechanistically using experimentally validated structural models of the channel. Relying on experimental methods alone for structural determination of physiologically relevant oligomeric species is impractical due to current technical limitations in producing sufficient quantities of the different ENaC subunits and ensuring proper physiological conditions for in vitro oligomerization. Therefore, combining functional and computational approaches provides an alternative tool for building models that generate testable hypotheses regarding mechanisms of activation of WT and disease-causing forms of ENaC.
Plasticity in the structural dynamics of the epithelial sodium channel facilitates the functional flexibility to sense and respond to changes in the extracellular milieu. The versatility of the channel to process various signals is defined by the threedimensional structure of its highly homologous subunits and their corresponding oligomers. Adapted by evolution, structural uniqueness confers functional specificity of signaling moieties such as proteolysis in the finger domains of the ␣and ␥-subunits. Our proposed dynamics-driven model for the gainof-function in ENaC by ␣W493R highlights the role of the knuckle domain in facilitating activating intersubunit conformational changes. Other variants of ENaC implicated in dis-ease, such as ␥L511Q, require further understanding of their molecular mechanisms to build models for specific pharmacological targeting. The synergy between our functional findings and structural modeling enhances the understanding of the etiology of ENaC mutants.