Molecular mechanism of prestin electromotive signal amplification.

CONTACT FOR REAGENT AND RESOURCE SHARING

Requests for reagents may be directed to the Lead Contact, Eric Gouaux ( gouauxe@ohsu.edu ).

EXPERIMENTAL MODEL AND SUBJECT DETAILS Construct and cell culture

The gene encoding full-length human prestin (Uniprot ID: P58743 –1) was synthesized and subcloned into the pEG BacMam vector ( Goehring et al., 2014 ). The construct contains a C-terminal 3C protease cleavage site, an enhanced green fluorescent protein (eGFP), and a StrepII tag. Sf9 cells (ThermoFisher 12659017) were used for BacMam virus amplification and were cultured in Sf-900 III SFM medium at 27 °C. To express prestin for structural studies and functional analysis, HEK293 GnTI − cells were cultured in suspension using FreeStyle medium supplemented with 2% (v/v) fetal bovine serum at 37 °C. Both cell lines are routinely tested for mycoplasma contamination using CELLshipper Mycoplasma Detection Kit M-100 from Bionique and are mycoplasma free. No misidentified cell lines were used. Prestin expression and purification HEK293 GnTI − cells cultured at 37°C with a density of ~2.5 × 10 6 ml − were infected with prestin BacMam virus at an MOI of 1:1. A final concentration of 10 mM sodium butyrate was added 12 hours after infection, and cells were cultured at 37°C for another 48 hours. Cells were harvested and lysed in buffer containing 20 mM Tris pH 8.0, 150 mM NaCl, 2% (w/v) digitonin, and protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 0.8 μM aprotinin, 2 μg/ml leupeptin and 2 μM pepstatin A) for three hours at 4 °C. The insoluble material was removed by centrifugation at 186,009 g for one hour. The supernatant was filtered through a 0.45 um filter and incubated with Strep-Tactin resin, washed with 20 mM Tris HCl pH 8.0, 150 mM NaCl, 0.02% (w/v) glycol-diosgenin (GDN), 0.004% (w/v) CHS, and eluted with the same buffer supplemented with 5 mM desthiobiotin. The eluate was concentrated and treated with 1:100 (w/w) 3C protease to remove the C-terminal GFP for 2 hours at 4°C. For prestin bound with chloride (Pres-Cl), the concentrate was loaded onto a size-exclusion chromatography (SEC) column (Superose 6 Increase 10/300 GL) equilibrated in a buffer containing 20 mM Tris HCl pH 8.0, 150 mM NaCl, 0.02% GDN. For prestin in the presence of sulfate and the absence of chloride (Pres-sulfate), an SEC buffer containing 20 mM Tris sulfate pH 8.0, 60 mM Na 2 SO 4 , 0.02% GDN was used to replace the chloride. Peak fractions from both conditions were pooled and concentrated for cryo-EM grid preparation. For the salicylate-bound state (Pres-Sal), the concentrated Pres-sulfate sample was incubated with a final concentration of 10 mM sodium salicylate for 10 minutes before cryo-EM grid preparation. For prestin in lipid nanodisc (Pres-Cl-nanodisc), porcine brain total lipid extract was dissolved in chloroform, dried in a rotary evaporator, and kept in a vacuum chamber overnight. The lipids were rehydrated to a final concentration of 20 mM with buffer containing 20 mM Tris pH 8.0 and 150 mM NaCl by several freeze-thaw cycles. MSP1E3D1 scaffolding protein, which yields nanodiscs with an estimated outer membrane diameter of ~12 nm, was purified as previously described ( Alvarez et al., 2010 ). The lipid suspension was dissolved with a final concentration of 1% (w/v) GDN and 0.2% (w/v) CHS for 30 min and then incubated with prestin and MSP1E3D1 for 1 hour with a molar ratio of 1:5:600 (prestin dimer: MSP: lipid). Three batches of biobeads were added for detergent removal and nanodisc formation at 4 °C, with the first two batches incubating for 2 hours and the last one overnight. For proteolysis, 3C protease (1:100 w/w) was combined with the last batch of biobeads to remove the GFP. The material was further purified by SEC (Superose 6 Increase 10/300 GL column) with a buffer containing 20 mM Tris-HCl pH 8.0 and 150 mM NaCl. Peak fractions were harvested and concentrated for cryo-EM grid preparation.

Cryo-EM grid preparation and data collection

Aliquots of prestin (3.5 ul at ~3 mg/ml) in either detergent micelles or lipid nanodiscs were applied onto Quantifoil holey carbon grids (R2/2, 200 mesh, gold) that had been glow discharged at 15 mA for 60 s. The grids were blotted with a Vitrobot mark IV at 100% humidity using a blotting time of 4s and a blotting force of 0 at 8 °C. The grids were plunge-frozen into liquid ethane cooled by liquid nitrogen. Micrographs were collected on 300 kV Titan Krios electron microscopes equipped with an energy filter and a Gatan K3 camera. SerialEM ( Mastronarde, 2005 ) was used for automated data collection. Data collection for all datasets followed a similar strategy with a total dose of 50 e − /Å, 50 frames, and a defocus range of −0.8 to −2.5 μm. The Pres-Cl and Pres-Cl-nanodisc datasets were collected with a dose rate of 18 e/pix/s and a pixel size of 0.651 Å/pixel without the correlated double sampling (CDS) mode. The Pres-Sal and Pres-sulfate datasets were collected with the CDS mode, a dose rate of 4.5 e/pix/s, and a pixel size of 0.676 Å/pixel. The data collection statistics are summarized in Table S1 .

Image processing

Raw micrographs in super-resolution mode were binned by 2x and corrected for both full-frame and local motion with the patch motion correction in cryoSPARC ( Punjani et al., 2017 ; Rubinstein and Brubaker, 2015 ). Contrast transfer function (CTF) parameters were estimated with patch CTF estimation in cryoSPARC. Particles were initially picked from a small fraction of the Pres-Cl dataset with Gaussian blob picking and subjected to 2D classification. Class averages showing reasonable prestin features in various orientations were selected as templates for template-based particle picking. Particles from these class averages generated an ab initio model in cryoSPARC. For each full dataset, extracted particles were down-sampled by two times to either 1.301 Å (Pres-Cl and Pres-Cl-nanodisc) or 1.352 Å (Pres-Sal or Pres-sulfate), followed by one round of 2D classification to remove ‘junk’ particles. For the Pres-Cl dataset, 3D classification into eight classes with the ab initio model as a reference was performed using either C1 or C2 symmetry. Both symmetries resulted in one good class out of eight with apparent C2 symmetry, indistinguishable structural features, and similar particle distribution. Therefore, C2 symmetry was used for subsequent image processing. 3D refinement with the bin-2 particles from the single good class resulted in a reconstruction at an overall resolution of 2.71 Å. Particles were re-extracted with the bin-1 pixel size of 0.651 Å and a box size of 512 pixels. 3D refinement with the re-extracted particles resulted in a reconstruction at 2.46 Å resolution. One more round of 3D classification and multiple rounds of local CTF refinement and non-uniform (NU) refinement ( Punjani et al., 2019 ) were performed using cryoSPARC, improving the resolution to 2.34 Å. The particle stack was further subjected to one round of 3D classification without alignment in RELION 3.1 and reconstructed using NU refinement in cryoSPARC, resulting in a 2.31 Å resolution reconstruction. Detailed data processing of Pres-Cl dataset can be found in Figure S3 . For Pres-Cl-Nanodisc and Pres-Sal datasets, the same data processing strategies were used except that no 3D classification step in RELION was performed, resulting in reconstructions at resolutions of 2.70 Å and 3.43 Å, respectively. For the Pres-sulfate dataset, bin-2 (1.352 Å/pix) particles after 2D classification in cryoSPARC were exported to RELION 3.1 for global 3D classification into eight classes. One out of the eight classes had promising prestin features. Refinement of the single class resulted in a 7 Å resolution reconstruction in RELION 3.1. One round of local 3D classification followed by refinement of the best class further improved the resolution to 5.2 Å. After bin-1 (0.676 Å/pix) particles were re-extracted, a reconstruction at ~5 Å resolution was obtained from RELION 3.1. The particles were moved to cryoSPARC for one round of homogeneous refinement and multiple rounds of local CTF refinement and NU refinement, improving the reconstruction to ~4.3 Å. Detailed data processing of Pres-sulfate dataset can be found in Figure S9 . The resolutions were estimated by FSC = 0.143, calculated by both half maps with a soft solvent mask in cryoSPARC. The angular distribution of particles was plotted with the python script star2bild.py in the UCSF pyem package ( Asarnow, 2019 ). The local resolution was calculated in cryoSPARC. Detailed data processing statistics and map quantifications are listed in Table S1 , Figure S4 . Structure determination and model building The 2.3 Å resolution cryo-EM map of Pres-Cl allows for de novo model building with high confidence in COOT ( Emsley et al., 2010 ). Most regions are well resolved with clear side chain features for residue registration. The N-terminal 1–11 residues and C-terminal 726–744 residues could not be built due to the lack of density in the map. We built a ~50 amino acid stretch of the intervening sequence (IVS) region (~80 residues in total), which was previously considered unstructured and flexible. The first part of the IVS (residues 561–580) forms the Cα2 helix with residues 564–580. The second part of the IVS (residues 614–640) was built based on the unique side chain features of residues 616–622 (YPPIVIK), which also helped to determine the loop orientation and the location of the remaining residues. In the extracellular loop region, residues 162–166 were less well resolved and poly-alanines were built. Lipid-like densities were fitted with alkanes with various number of carbons. Solvent-like densities were modeled with water molecules that were automatically added by Phenix (version 1.19rc5–4047) ( Liebschner et al., 2019 ) and inspected in COOT. The model was refined in real space with Phenix (version 1.18.2–3874). For model building of the other structures, the Pres-Cl structure without lipids and waters was fit into cryo-EM maps in ChimeraX ( Pettersen et al., 2020 ) and adjusted in COOT. For Pres-Cl-Nanodisc, the chloride was built similarly to Pres-Cl. For Pres-Sal, the salicylate was fit into the oval shape of density that appeared in a similar position as the chloride anion in Pres-Cl. The orientation of salicylate was determined by the density features, the surrounding residues and the molecular dynamics simulation ( Supplemental Information ). All final models were subjected to real-space refinement and validated in Phenix (version 1.18.2–3874). Map and structure figures were generated using Pymol or ChimeraX. Electrophysiology HEK293 GnTI − cells were transfected with the construct (pEG Bacmam CMV-hPrestin-GFP) for the cryo-EM studies and used for recordings 24–48 h after transfection. The recording pipettes had resistances of 2.5–3.5 MΩ when filled with the Cl − internal solution containing (in millimolar) 135 CsCl, 5 MgCl 2 , 2 Na 2 ATP, 10 HEPES and 10 EGTA or with the SO 4 2- internal solution containing (in millimolar) 88 Na2SO4, 2 MgSO4, 10 HEPES, 3 MgATP, and 10 EGTA (pH adjusted to 7.35). Cells with a membrane resistance greater than 500 MΩ after rupturing were recorded in a bath filled with external solution containing (in millimolar) 120 NaCl, 2 CoCl 2 , 1 CaCl 2 , 2 MgCl 2 , 20 TEACl, 10 HEPES (pH adjusted to 7.35). Recordings are performed with Cl − internal buffer if not otherwise mentioned. All solutions were adjusted to 310 ± 2 mOsm/L. For salicylate and 5-methylsalicylate inhibition experiments, 10 mM salicylate or 10 mM 5-methylsalicylate was added to the external solution (adjusted to pH 7.35, 310±2 mOsm/L). For NLC measurement with the “Sine+DC” method, HEK cells on coverslips were mounted on an upright microscope (BX51WI, Olympus) and viewed at 60x magnification using direct interference contrast (DIC). Transfected cells were identified using epifluorescence (Filter wavelength). Images were acquired with an ORCA Flash 4.0LT CMOS camera operated with CellSens software (Olympus). Lockin C m measurements were performed using an EPC10 amplifier and Patchmaster software (HEKA; Germany). Voltage-clamp C m recordings were done with a sinusoidal wave (2khz; 30mV peak-to peak) superimposed on holding potential (0 mV). Incremental voltage steps were given from −150mV (dV=10mV; 200ms). C m traces were analyzed and plotted in IgorPro software (Wavemetrics, Lake Oswego). C m was measured by averaging the last 150ms of each voltage step. For NLC measurement with the dual-sine methodology, recordings were carried out at room temperature under the whole-cell patch clamp mode using an Axopatch 200A amplifier. Cells were placed in either external solution only or external solution supplemented with 10 mM salicylate (or salicylate analogs) for 2 min before each recording. Whole-cell capacitance was measured at 0 mV holding potential using a two sine wave voltage stimulus protocol (10 mV amplitude at both 390.6 and 781.2 Hz), followed by the fast Fourier transform-based admittance analysis using jClamp ( www.scisoftco.com ). Series resistance was compensated before recording and the residual was corrected with the built-in function in jClamp. The voltage-dependent capacitance was fit with the first derivative of the two-state Boltzmann function: C m = Q m a x ( z e k T ) exp ⁡ z e k T ( V m − V h ) × ( 1 + exp ⁡ ( z e k T ( V m − V h ) ) 2 + C l i n Where Q max is the maximum charge transfer, z is the valence of charge movement, e is electron charge, k is the Boltzmann’s constant, T is the absolute temperature, V m is the membrane potential, V h is the voltage at peak capacitance, C lin is the linear capacitance. The fitted parameters for prestin WT and F101Y mutant are as follows: Q max = 0.112 ± 0.009 pC, z = 0.64 ± 0.02; V h = −54.0 ± 3.2 mV, C lin = 10.94 ± 0.75 pF, Q max /C lin = 10.7 ± 1.1 fC/pF for WT (n=14), and Q max = 0.066 ± 0.008 pC, z= 0.63 ± 0.02; V h = 5.7 ± 1.9 mV, C lin = 7.98 ± 0.43 pF, Q max /C lin = 8.6 ± 1.1 fC/pF for F101Y (n=20). Data are presented as means ± standard errors of means.

Computational Methods

System preparation Cl − -, salicylate-, and sulfate-bound prestin cryo-EM structures were prepared for molecular dynamics (MD) simulations following the procedures described below. Missing side chains and hydrogen atoms were added to the experimental, cryo-EM models, using the PSFGEN plugin ( https://www.ks.uiuc.edu/Research/vmd/plugins/psfgen/ ) in VMD ( Humphrey et al., 1996 ). The experimental models are missing a disordered region in the intracellular domain, dividing the structures into two polypeptide segments (residues 13–580 and 614–724, respectively). Neutral N-terminal and C-terminal ‘caps’ were added to the first and last residues of each polypeptide segment, respectively, using PSFGEN. The coordinates of the bound Cl − and sulfate ions were based on the modeled Cl − in the cryo-EM structure, whereas the bound salicylate was placed in the binding pocket using our in-house docking protocol, GOLEM (see ‘ Computational search for docking poses of salicylate ’ for more details). Following pKa estimation by PROPKA ( Olsson et al., 2011 ; Sondergaard et al., 2011 ), all titratable residues were modeled in their default titration states, which was modeled in its protonated (neutral) form. All protein structures were internally hydrated using the DOWSER ( Gumbart et al., 2009 ; Zhang and Hermans, 1996 ) plugin of VMD. The two cholesterol molecules modeled in the experimental structures were included in the Cl − - and salicylate-bound systems. The lipid bilayer used to embed the proteins was constructed in CHARMM-GUI ( Jo et al., 2008 ). The orientation of the protein in the bilayer was obtained from the OPM (Orientations of Proteins in Membranes) database ( Lomize et al., 2006 ). The protein, along with the bound cholesterol molecules, was then inserted into a heterogeneous lipid bilayer, followed by removing sterically clashing lipid molecules. The bilayer is composed of palmitoyl-oleoyl-phosphatidyl-choline (POPC), palmitoyl-oleoyl-phosphatidyl-ethanolamine (POPE), palmitoyl-oleoyl-phosphatidyl-glycerol (POPG), palmitoyl-oleoyl-phosphatidyl-inositol (POPI), palmitoyl-sphingomyelin (PSM), and cholesterol (CHOL) lipids at a molar ratio of 32:16:5:11:16:20. The protein-membrane system was then solvated with water including 150 mM NaCl in VMD (system size: 335k atoms). To further reduce any bias from the initial lipid configuration and increase the sampling, two independent membrane systems (with randomly placed initial lipids) were generated for Cl − -bound prestin.

Computational search for docking poses of salicylate

The salicylate binding pose in our simulations was obtained using an in-house docking package called GOLEM ( Zhao and Tajkhorshid, 2021 ) This software tool employs a genetic algorithm to explore a ligand’s conformational, orientational, and positional space, with explicit consideration of water displacement, as well as bridging water molecules’ position and orientation, to optimize the fitness, which is defined as the combination of the system’s energy and fitting to the cryo-EM density. Despite the stochasticity of the genetic part of the algorithm, given the large number of iterations. GOLEM robustly reproduces the ligand’s docked pose with negligible differences in repeated trials. The resulting docked pose can be further validated by equilibrium MD simulations, as done in this study. Equilibrium molecular dynamics (MD) simulation All MD simulations were performed using NAMD simulation package ( Phillips et al., 2020 ) and employing the fully atomistic CHARMM36m ( Huang et al., 2017 ) and CHARMM36 ( Klauda et al., 2010 ) force fields for the protein and lipids, respectively. The TIP3 model was used for water molecules ( Jorgensen et al., 1983 ). The salicylate was parameterized using the CHARMM general force field (CGenFF) webserver ( https://cgenff.paramchem.org ) ( Vanommeslaeghe et al., 2010 ; Vanommeslaeghe et al., 2012 ) and further optimized employing the Force Field Toolkit (ffTK) plugin ( Mayne et al., 2013 ) of VMD. Sulfate parameters were adopted from a previous study employing ab initio calculations ( Cannon et al., 1994 ). A 12 Å cutoff was used for short-range, non-bonded interactions, with switching starting at 10 Å. Long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) algorithm ( Darden et al., 1993 ) with a grid density of 1 Å −1 , and a PME interpolation order of 6. All bonds involving hydrogen atoms were kept rigid using the SHAKE algorithm ( Ryckaert et al., 1977 ). Temperature was maintained at 310 K using Langevin thermostat with a damping coefficient of 1.0 ps -1 . Pressure was maintained at 1 atm by the Nosé-Hoover Langevin piston barostat ( Feller et al., 1995 ; Martyna et al., 1994 ), with period and decay of 100 and 50 fs, respectively. All systems were simulated in a flexible cell allowing the dimensions of the periodic cell to change independently while keeping the cell aspect ratio in the xy plane (membrane plane) constant. The simulation timestep was set to 2 fs. Lennard-Jones and PME forces were updated at one and two timesteps, respectively. Atomic coordinates were saved every 10 ps. The two Cl − -bound systems were simulated using the following steps: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, in which harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) were applied to the position of protein’s heavy atoms modeled in the cryo-EM structure, as well as to the heavy atoms of the two modeled cholesterol molecules and the bound Cl − ions. Furthermore, the z position (normal to the membrane plane) of phosphorus atoms in phospholipids and oxygen atoms of bulk cholesterols were also restrained in this step using a harmonic potential ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with only the protein backbone, heavy atoms of the two modeled cholesterol molecules, and the bound Cl − ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ); ( 3 ) 5 ns of equilibration in which all the restraints from the previous step were removed by decreasing the force constant stepwise from 10 to 0 kcal.mol −1 .Å −2 ; ( 4 ) production equilibrium runs for 1.5 μs without any restraints. The simulation of the salicylate-bound system was performed to check the stability of the docked pose of the inhibitor, and therefore, followed a different protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns of equilibration with harmonic restraints ( k = 10 kcal.mol −1 .Å −2 ) on the all the protein’s heavy atoms modeled in the cryo-EM structure, and on the heavy atoms of the bound salicylates, water molecules placed close to the salicylates by the the docking protocol, the two modeled cholesterols. In addition, z positional restraints were applied ( k = 5 kcal.mol −1 .Å −2 ) to phosphorus atoms of phospholipids and oxygen atoms of bulk cholesterols; ( 2 ) 20 ns of equilibration in which the protein backbone atoms were restrained ( k = 10 kcal.mol −1 .Å −2 ), while the salicylate and protein side chains were free to move; ( 3 ) production run extending the simulation in Step 2 to 250 ns. The simulation of sulfate-bound structure was conducted to obtain an equilibrated system for free energy perturbation (FEP) simulations used to calculate the binding affinity of sulfate in the binding pocket, and therefore, followed a shorter protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, while applying harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) to the protein’s heavy atoms modeled in the cryo-EM structure, and to the modeled sulfate ions in the binding pockets. Furthermore, the z position of phosphorus and oxygen atoms in phospholipids and cholesterols were also harmonically restrained, respectively ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with the protein backbone, and bound sulfate ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ). Prestin conformational transition between the states Two non-equilibrium MD simulations were performed to drive transitions between the contracted and expanded states and to monitor the response of the membrane to these transitions, one in the forward (contracted to expanded) and one in the backward (expanded to contracted) direction. The Cl − -bound cryo-EM structure represented the contracted state, while the salicylate-bound conformation was used to represent prestin in its expanded state. The same protocol as described in ‘System preparation’ was employed to construct the contracted and expanded systems. Since the goal was to investigate the coupling between the protein’s conformational transitions and the membrane, we did not include the bound Cl − and salicylate in these simulations. These transition simulations follow the same driving protocols (but in opposite directions): (1) 10,000 steps of minimization, followed by 5 ns equilibration while restraining the protein’s heavy atoms modeled in the cryo-EM structure, as well as the heavy atoms of the two modeled cholesterol molecules, using a harmonic potential ( k = 10 kcal.mol −1 .Å −2 ). Moreover, harmonic z position restraints were applied to phosphorus and oxygen atoms of phospholipids and bulk cholesterols, respectively ( k = 5 kcal.mol −1 .Å −2 ); (2) 20 ns of equilibration with harmonic restraints on the protein backbone, as well as on the heavy atoms of the two modeled cholesterol molecules ( k = 10 kcal.mol −1 .Å −2 ); (3) keeping the same restraints only on the protein backbone and further equilibrating the system for 150 ns; (4) a 50 ns non-equilibrium simulation, with time-dependent biases applied to drive the conversion from the initial to the target state using the collective variables (colvars) module of NAMD ( Fiorin et al., 2013 ). This transition was achieved by applying biasing forces (k = 1.5 kcal.mol −1 .Å −2 ) on the protein C α atoms to decrease their RMSD with respect to those in the target structure; (5) maintaining the protein in the target state for additional 150 ns by applying the final set of biases (not moving this time) introduced in step (4) . During the transitions between the contracted and expanded forms (the non-equilibrium phase of the simulations in Step 3), the x and y dimensions of the simulation box (membrane dimensions) were recorded every 10 ps. The resulting xy plane area reflects the total cross-sectional area of the membrane-protein system.

Analysis of molecular dynamics trajectories

Molecular images from MD trajectories are generated with VMD ( Humphrey et al., 1996 ). The membrane deformation induced by prestin is quantified by recording the z component of the center of mass of the lipids’ phosphate (PO 4 − ) groups during the last 500 ns of the Cl − -bound trajectories. The origin of the z axis is defined as the bilayer midplane. The resulting heatmap (binned in 2×2 Å 2 bins) in each leaflet captures the spatial distribution of the height of the lipid head groups within the membrane. The dynamics of the Cl − ions at their binding sites were monitored by measuring their displacement from the original (experimentally modeled) positions. When analyzing Cl − dynamics in each binding site, to average out the motion of the binding site itself, the trajectories were aligned using C α atoms of the binding-site residues (Q97, F101, P136, F137, L397, S398, and R399) with respect to those in the cryo-EM structure. Then, the distance between the ion (either from the initial model or from the bulk solution) to the modeled Cl − in the cryo-EM structure was recorded. The distance plots were smoothed using a sliding window of 1 ns. The stability of the docked salicylate in the two subunits was monitored by calculating the displacement and change in the orientation of the ligand throughout the simulations. The displacement was obtained by calculating the center of mass displacement of salicylate from its docked pose in each frame of the trajectory. The orientational change was defined as the angular change of the normal vector to the salicylate’s benzene ring plane with respect to that of the docked pose. Both the displacement and orientation plots were smoothed using a sliding window of 0.2 ns. Free energy calculation of ligand binding affinity FEP simulations ( Cournia et al., 2017 ) were performed on the Cl--, salicylate-, and sulfate-bound structures to compare relative binding affinities of the three ligands. During the employed FEP protocol, the ligand is annihilated from the binding pocket of one prestin protomer while being created at a location in bulk solution 80 Å away from the binding pocket. The FEP calculations of Cl--, salicylate-, and sulfate-bound structures were initiated from the last frame of the Step 2 of the equilibration protocol (see Equilibrium molecular dynamics simulation for more details). During FEP calculations, the protein backbone heavy atoms were restrained (k = 10 kcal.mol-1.Å-2) to the corresponding experimentally modeled positions to avoid undesired structural deviations. All other simulation parameters were the same as in the aforementioned equilibrium simulations. The alchemical transformations were performed in both forward (λ increasing from 0 to 1; ligand annihilation in the binding site and creation in solution) and backward (λ decreasing from 1 to 0; ligand creation in the binding site and annihilation in solution) directions to assess the reversibility. A soft-core van der Waals radius-shifting coefficient of 5 Å was chosen. In the forward transformation, the vdW and electrostatic interactions of the annihilated particles were linearly decoupled to the simulation over a λ range of 0–1 and 0–0.5, respectively, whereas in the backward transformation these interactions were linearly coupled to the simulation in the corresponding λ range. Each FEP calculation along the λ reaction path was divided into 50 equally-spaced consecutive windows to ensure gradual transformation. Within each window, the performed simulation consisted of 5000 steps of minimization, an equilibration phase of 100 ps, followed by 1 ns simulation for data collection at the frequency of every 20 fs, resulting in an accumulated simulation time of 110 ns for each system. To prevent the dissociation of the ligand during the alchemical transformation, harmonic restraints were applied to the annihilated and created ligands (k = 10 kcal.mol-1.Å-2). All the simulations were performed, employing the FEP module of NAMD ( Chen et al., 2020 ). The outputs were analyzed using the ParseFEP ( Liu et al., 2012 ) plugin in VMD, with the statistical error estimated using the implemented Bennett Acceptance Ratio (BAR) method ( Bennett, 1976 ).

EXPERIMENTAL MODEL AND SUBJECT DETAILS Construct and cell culture

The gene encoding full-length human prestin (Uniprot ID: P58743 –1) was synthesized and subcloned into the pEG BacMam vector ( Goehring et al., 2014 ). The construct contains a C-terminal 3C protease cleavage site, an enhanced green fluorescent protein (eGFP), and a StrepII tag. Sf9 cells (ThermoFisher 12659017) were used for BacMam virus amplification and were cultured in Sf-900 III SFM medium at 27 °C. To express prestin for structural studies and functional analysis, HEK293 GnTI − cells were cultured in suspension using FreeStyle medium supplemented with 2% (v/v) fetal bovine serum at 37 °C. Both cell lines are routinely tested for mycoplasma contamination using CELLshipper Mycoplasma Detection Kit M-100 from Bionique and are mycoplasma free. No misidentified cell lines were used. Prestin expression and purification HEK293 GnTI − cells cultured at 37°C with a density of ~2.5 × 10 6 ml − were infected with prestin BacMam virus at an MOI of 1:1. A final concentration of 10 mM sodium butyrate was added 12 hours after infection, and cells were cultured at 37°C for another 48 hours. Cells were harvested and lysed in buffer containing 20 mM Tris pH 8.0, 150 mM NaCl, 2% (w/v) digitonin, and protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 0.8 μM aprotinin, 2 μg/ml leupeptin and 2 μM pepstatin A) for three hours at 4 °C. The insoluble material was removed by centrifugation at 186,009 g for one hour. The supernatant was filtered through a 0.45 um filter and incubated with Strep-Tactin resin, washed with 20 mM Tris HCl pH 8.0, 150 mM NaCl, 0.02% (w/v) glycol-diosgenin (GDN), 0.004% (w/v) CHS, and eluted with the same buffer supplemented with 5 mM desthiobiotin. The eluate was concentrated and treated with 1:100 (w/w) 3C protease to remove the C-terminal GFP for 2 hours at 4°C. For prestin bound with chloride (Pres-Cl), the concentrate was loaded onto a size-exclusion chromatography (SEC) column (Superose 6 Increase 10/300 GL) equilibrated in a buffer containing 20 mM Tris HCl pH 8.0, 150 mM NaCl, 0.02% GDN. For prestin in the presence of sulfate and the absence of chloride (Pres-sulfate), an SEC buffer containing 20 mM Tris sulfate pH 8.0, 60 mM Na 2 SO 4 , 0.02% GDN was used to replace the chloride. Peak fractions from both conditions were pooled and concentrated for cryo-EM grid preparation. For the salicylate-bound state (Pres-Sal), the concentrated Pres-sulfate sample was incubated with a final concentration of 10 mM sodium salicylate for 10 minutes before cryo-EM grid preparation. For prestin in lipid nanodisc (Pres-Cl-nanodisc), porcine brain total lipid extract was dissolved in chloroform, dried in a rotary evaporator, and kept in a vacuum chamber overnight. The lipids were rehydrated to a final concentration of 20 mM with buffer containing 20 mM Tris pH 8.0 and 150 mM NaCl by several freeze-thaw cycles. MSP1E3D1 scaffolding protein, which yields nanodiscs with an estimated outer membrane diameter of ~12 nm, was purified as previously described ( Alvarez et al., 2010 ). The lipid suspension was dissolved with a final concentration of 1% (w/v) GDN and 0.2% (w/v) CHS for 30 min and then incubated with prestin and MSP1E3D1 for 1 hour with a molar ratio of 1:5:600 (prestin dimer: MSP: lipid). Three batches of biobeads were added for detergent removal and nanodisc formation at 4 °C, with the first two batches incubating for 2 hours and the last one overnight. For proteolysis, 3C protease (1:100 w/w) was combined with the last batch of biobeads to remove the GFP. The material was further purified by SEC (Superose 6 Increase 10/300 GL column) with a buffer containing 20 mM Tris-HCl pH 8.0 and 150 mM NaCl. Peak fractions were harvested and concentrated for cryo-EM grid preparation.

Cryo-EM grid preparation and data collection

Aliquots of prestin (3.5 ul at ~3 mg/ml) in either detergent micelles or lipid nanodiscs were applied onto Quantifoil holey carbon grids (R2/2, 200 mesh, gold) that had been glow discharged at 15 mA for 60 s. The grids were blotted with a Vitrobot mark IV at 100% humidity using a blotting time of 4s and a blotting force of 0 at 8 °C. The grids were plunge-frozen into liquid ethane cooled by liquid nitrogen. Micrographs were collected on 300 kV Titan Krios electron microscopes equipped with an energy filter and a Gatan K3 camera. SerialEM ( Mastronarde, 2005 ) was used for automated data collection. Data collection for all datasets followed a similar strategy with a total dose of 50 e − /Å, 50 frames, and a defocus range of −0.8 to −2.5 μm. The Pres-Cl and Pres-Cl-nanodisc datasets were collected with a dose rate of 18 e/pix/s and a pixel size of 0.651 Å/pixel without the correlated double sampling (CDS) mode. The Pres-Sal and Pres-sulfate datasets were collected with the CDS mode, a dose rate of 4.5 e/pix/s, and a pixel size of 0.676 Å/pixel. The data collection statistics are summarized in Table S1 .

Image processing

Raw micrographs in super-resolution mode were binned by 2x and corrected for both full-frame and local motion with the patch motion correction in cryoSPARC ( Punjani et al., 2017 ; Rubinstein and Brubaker, 2015 ). Contrast transfer function (CTF) parameters were estimated with patch CTF estimation in cryoSPARC. Particles were initially picked from a small fraction of the Pres-Cl dataset with Gaussian blob picking and subjected to 2D classification. Class averages showing reasonable prestin features in various orientations were selected as templates for template-based particle picking. Particles from these class averages generated an ab initio model in cryoSPARC. For each full dataset, extracted particles were down-sampled by two times to either 1.301 Å (Pres-Cl and Pres-Cl-nanodisc) or 1.352 Å (Pres-Sal or Pres-sulfate), followed by one round of 2D classification to remove ‘junk’ particles. For the Pres-Cl dataset, 3D classification into eight classes with the ab initio model as a reference was performed using either C1 or C2 symmetry. Both symmetries resulted in one good class out of eight with apparent C2 symmetry, indistinguishable structural features, and similar particle distribution. Therefore, C2 symmetry was used for subsequent image processing. 3D refinement with the bin-2 particles from the single good class resulted in a reconstruction at an overall resolution of 2.71 Å. Particles were re-extracted with the bin-1 pixel size of 0.651 Å and a box size of 512 pixels. 3D refinement with the re-extracted particles resulted in a reconstruction at 2.46 Å resolution. One more round of 3D classification and multiple rounds of local CTF refinement and non-uniform (NU) refinement ( Punjani et al., 2019 ) were performed using cryoSPARC, improving the resolution to 2.34 Å. The particle stack was further subjected to one round of 3D classification without alignment in RELION 3.1 and reconstructed using NU refinement in cryoSPARC, resulting in a 2.31 Å resolution reconstruction. Detailed data processing of Pres-Cl dataset can be found in Figure S3 . For Pres-Cl-Nanodisc and Pres-Sal datasets, the same data processing strategies were used except that no 3D classification step in RELION was performed, resulting in reconstructions at resolutions of 2.70 Å and 3.43 Å, respectively. For the Pres-sulfate dataset, bin-2 (1.352 Å/pix) particles after 2D classification in cryoSPARC were exported to RELION 3.1 for global 3D classification into eight classes. One out of the eight classes had promising prestin features. Refinement of the single class resulted in a 7 Å resolution reconstruction in RELION 3.1. One round of local 3D classification followed by refinement of the best class further improved the resolution to 5.2 Å. After bin-1 (0.676 Å/pix) particles were re-extracted, a reconstruction at ~5 Å resolution was obtained from RELION 3.1. The particles were moved to cryoSPARC for one round of homogeneous refinement and multiple rounds of local CTF refinement and NU refinement, improving the reconstruction to ~4.3 Å. Detailed data processing of Pres-sulfate dataset can be found in Figure S9 . The resolutions were estimated by FSC = 0.143, calculated by both half maps with a soft solvent mask in cryoSPARC. The angular distribution of particles was plotted with the python script star2bild.py in the UCSF pyem package ( Asarnow, 2019 ). The local resolution was calculated in cryoSPARC. Detailed data processing statistics and map quantifications are listed in Table S1 , Figure S4 . Structure determination and model building The 2.3 Å resolution cryo-EM map of Pres-Cl allows for de novo model building with high confidence in COOT ( Emsley et al., 2010 ). Most regions are well resolved with clear side chain features for residue registration. The N-terminal 1–11 residues and C-terminal 726–744 residues could not be built due to the lack of density in the map. We built a ~50 amino acid stretch of the intervening sequence (IVS) region (~80 residues in total), which was previously considered unstructured and flexible. The first part of the IVS (residues 561–580) forms the Cα2 helix with residues 564–580. The second part of the IVS (residues 614–640) was built based on the unique side chain features of residues 616–622 (YPPIVIK), which also helped to determine the loop orientation and the location of the remaining residues. In the extracellular loop region, residues 162–166 were less well resolved and poly-alanines were built. Lipid-like densities were fitted with alkanes with various number of carbons. Solvent-like densities were modeled with water molecules that were automatically added by Phenix (version 1.19rc5–4047) ( Liebschner et al., 2019 ) and inspected in COOT. The model was refined in real space with Phenix (version 1.18.2–3874). For model building of the other structures, the Pres-Cl structure without lipids and waters was fit into cryo-EM maps in ChimeraX ( Pettersen et al., 2020 ) and adjusted in COOT. For Pres-Cl-Nanodisc, the chloride was built similarly to Pres-Cl. For Pres-Sal, the salicylate was fit into the oval shape of density that appeared in a similar position as the chloride anion in Pres-Cl. The orientation of salicylate was determined by the density features, the surrounding residues and the molecular dynamics simulation ( Supplemental Information ). All final models were subjected to real-space refinement and validated in Phenix (version 1.18.2–3874). Map and structure figures were generated using Pymol or ChimeraX. Electrophysiology HEK293 GnTI − cells were transfected with the construct (pEG Bacmam CMV-hPrestin-GFP) for the cryo-EM studies and used for recordings 24–48 h after transfection. The recording pipettes had resistances of 2.5–3.5 MΩ when filled with the Cl − internal solution containing (in millimolar) 135 CsCl, 5 MgCl 2 , 2 Na 2 ATP, 10 HEPES and 10 EGTA or with the SO 4 2- internal solution containing (in millimolar) 88 Na2SO4, 2 MgSO4, 10 HEPES, 3 MgATP, and 10 EGTA (pH adjusted to 7.35). Cells with a membrane resistance greater than 500 MΩ after rupturing were recorded in a bath filled with external solution containing (in millimolar) 120 NaCl, 2 CoCl 2 , 1 CaCl 2 , 2 MgCl 2 , 20 TEACl, 10 HEPES (pH adjusted to 7.35). Recordings are performed with Cl − internal buffer if not otherwise mentioned. All solutions were adjusted to 310 ± 2 mOsm/L. For salicylate and 5-methylsalicylate inhibition experiments, 10 mM salicylate or 10 mM 5-methylsalicylate was added to the external solution (adjusted to pH 7.35, 310±2 mOsm/L). For NLC measurement with the “Sine+DC” method, HEK cells on coverslips were mounted on an upright microscope (BX51WI, Olympus) and viewed at 60x magnification using direct interference contrast (DIC). Transfected cells were identified using epifluorescence (Filter wavelength). Images were acquired with an ORCA Flash 4.0LT CMOS camera operated with CellSens software (Olympus). Lockin C m measurements were performed using an EPC10 amplifier and Patchmaster software (HEKA; Germany). Voltage-clamp C m recordings were done with a sinusoidal wave (2khz; 30mV peak-to peak) superimposed on holding potential (0 mV). Incremental voltage steps were given from −150mV (dV=10mV; 200ms). C m traces were analyzed and plotted in IgorPro software (Wavemetrics, Lake Oswego). C m was measured by averaging the last 150ms of each voltage step. For NLC measurement with the dual-sine methodology, recordings were carried out at room temperature under the whole-cell patch clamp mode using an Axopatch 200A amplifier. Cells were placed in either external solution only or external solution supplemented with 10 mM salicylate (or salicylate analogs) for 2 min before each recording. Whole-cell capacitance was measured at 0 mV holding potential using a two sine wave voltage stimulus protocol (10 mV amplitude at both 390.6 and 781.2 Hz), followed by the fast Fourier transform-based admittance analysis using jClamp ( www.scisoftco.com ). Series resistance was compensated before recording and the residual was corrected with the built-in function in jClamp. The voltage-dependent capacitance was fit with the first derivative of the two-state Boltzmann function: C m = Q m a x ( z e k T ) exp ⁡ z e k T ( V m − V h ) × ( 1 + exp ⁡ ( z e k T ( V m − V h ) ) 2 + C l i n Where Q max is the maximum charge transfer, z is the valence of charge movement, e is electron charge, k is the Boltzmann’s constant, T is the absolute temperature, V m is the membrane potential, V h is the voltage at peak capacitance, C lin is the linear capacitance. The fitted parameters for prestin WT and F101Y mutant are as follows: Q max = 0.112 ± 0.009 pC, z = 0.64 ± 0.02; V h = −54.0 ± 3.2 mV, C lin = 10.94 ± 0.75 pF, Q max /C lin = 10.7 ± 1.1 fC/pF for WT (n=14), and Q max = 0.066 ± 0.008 pC, z= 0.63 ± 0.02; V h = 5.7 ± 1.9 mV, C lin = 7.98 ± 0.43 pF, Q max /C lin = 8.6 ± 1.1 fC/pF for F101Y (n=20). Data are presented as means ± standard errors of means.

Computational Methods

System preparation Cl − -, salicylate-, and sulfate-bound prestin cryo-EM structures were prepared for molecular dynamics (MD) simulations following the procedures described below. Missing side chains and hydrogen atoms were added to the experimental, cryo-EM models, using the PSFGEN plugin ( https://www.ks.uiuc.edu/Research/vmd/plugins/psfgen/ ) in VMD ( Humphrey et al., 1996 ). The experimental models are missing a disordered region in the intracellular domain, dividing the structures into two polypeptide segments (residues 13–580 and 614–724, respectively). Neutral N-terminal and C-terminal ‘caps’ were added to the first and last residues of each polypeptide segment, respectively, using PSFGEN. The coordinates of the bound Cl − and sulfate ions were based on the modeled Cl − in the cryo-EM structure, whereas the bound salicylate was placed in the binding pocket using our in-house docking protocol, GOLEM (see ‘ Computational search for docking poses of salicylate ’ for more details). Following pKa estimation by PROPKA ( Olsson et al., 2011 ; Sondergaard et al., 2011 ), all titratable residues were modeled in their default titration states, which was modeled in its protonated (neutral) form. All protein structures were internally hydrated using the DOWSER ( Gumbart et al., 2009 ; Zhang and Hermans, 1996 ) plugin of VMD. The two cholesterol molecules modeled in the experimental structures were included in the Cl − - and salicylate-bound systems. The lipid bilayer used to embed the proteins was constructed in CHARMM-GUI ( Jo et al., 2008 ). The orientation of the protein in the bilayer was obtained from the OPM (Orientations of Proteins in Membranes) database ( Lomize et al., 2006 ). The protein, along with the bound cholesterol molecules, was then inserted into a heterogeneous lipid bilayer, followed by removing sterically clashing lipid molecules. The bilayer is composed of palmitoyl-oleoyl-phosphatidyl-choline (POPC), palmitoyl-oleoyl-phosphatidyl-ethanolamine (POPE), palmitoyl-oleoyl-phosphatidyl-glycerol (POPG), palmitoyl-oleoyl-phosphatidyl-inositol (POPI), palmitoyl-sphingomyelin (PSM), and cholesterol (CHOL) lipids at a molar ratio of 32:16:5:11:16:20. The protein-membrane system was then solvated with water including 150 mM NaCl in VMD (system size: 335k atoms). To further reduce any bias from the initial lipid configuration and increase the sampling, two independent membrane systems (with randomly placed initial lipids) were generated for Cl − -bound prestin.

Computational search for docking poses of salicylate

The salicylate binding pose in our simulations was obtained using an in-house docking package called GOLEM ( Zhao and Tajkhorshid, 2021 ) This software tool employs a genetic algorithm to explore a ligand’s conformational, orientational, and positional space, with explicit consideration of water displacement, as well as bridging water molecules’ position and orientation, to optimize the fitness, which is defined as the combination of the system’s energy and fitting to the cryo-EM density. Despite the stochasticity of the genetic part of the algorithm, given the large number of iterations. GOLEM robustly reproduces the ligand’s docked pose with negligible differences in repeated trials. The resulting docked pose can be further validated by equilibrium MD simulations, as done in this study. Equilibrium molecular dynamics (MD) simulation All MD simulations were performed using NAMD simulation package ( Phillips et al., 2020 ) and employing the fully atomistic CHARMM36m ( Huang et al., 2017 ) and CHARMM36 ( Klauda et al., 2010 ) force fields for the protein and lipids, respectively. The TIP3 model was used for water molecules ( Jorgensen et al., 1983 ). The salicylate was parameterized using the CHARMM general force field (CGenFF) webserver ( https://cgenff.paramchem.org ) ( Vanommeslaeghe et al., 2010 ; Vanommeslaeghe et al., 2012 ) and further optimized employing the Force Field Toolkit (ffTK) plugin ( Mayne et al., 2013 ) of VMD. Sulfate parameters were adopted from a previous study employing ab initio calculations ( Cannon et al., 1994 ). A 12 Å cutoff was used for short-range, non-bonded interactions, with switching starting at 10 Å. Long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) algorithm ( Darden et al., 1993 ) with a grid density of 1 Å −1 , and a PME interpolation order of 6. All bonds involving hydrogen atoms were kept rigid using the SHAKE algorithm ( Ryckaert et al., 1977 ). Temperature was maintained at 310 K using Langevin thermostat with a damping coefficient of 1.0 ps -1 . Pressure was maintained at 1 atm by the Nosé-Hoover Langevin piston barostat ( Feller et al., 1995 ; Martyna et al., 1994 ), with period and decay of 100 and 50 fs, respectively. All systems were simulated in a flexible cell allowing the dimensions of the periodic cell to change independently while keeping the cell aspect ratio in the xy plane (membrane plane) constant. The simulation timestep was set to 2 fs. Lennard-Jones and PME forces were updated at one and two timesteps, respectively. Atomic coordinates were saved every 10 ps. The two Cl − -bound systems were simulated using the following steps: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, in which harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) were applied to the position of protein’s heavy atoms modeled in the cryo-EM structure, as well as to the heavy atoms of the two modeled cholesterol molecules and the bound Cl − ions. Furthermore, the z position (normal to the membrane plane) of phosphorus atoms in phospholipids and oxygen atoms of bulk cholesterols were also restrained in this step using a harmonic potential ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with only the protein backbone, heavy atoms of the two modeled cholesterol molecules, and the bound Cl − ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ); ( 3 ) 5 ns of equilibration in which all the restraints from the previous step were removed by decreasing the force constant stepwise from 10 to 0 kcal.mol −1 .Å −2 ; ( 4 ) production equilibrium runs for 1.5 μs without any restraints. The simulation of the salicylate-bound system was performed to check the stability of the docked pose of the inhibitor, and therefore, followed a different protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns of equilibration with harmonic restraints ( k = 10 kcal.mol −1 .Å −2 ) on the all the protein’s heavy atoms modeled in the cryo-EM structure, and on the heavy atoms of the bound salicylates, water molecules placed close to the salicylates by the the docking protocol, the two modeled cholesterols. In addition, z positional restraints were applied ( k = 5 kcal.mol −1 .Å −2 ) to phosphorus atoms of phospholipids and oxygen atoms of bulk cholesterols; ( 2 ) 20 ns of equilibration in which the protein backbone atoms were restrained ( k = 10 kcal.mol −1 .Å −2 ), while the salicylate and protein side chains were free to move; ( 3 ) production run extending the simulation in Step 2 to 250 ns. The simulation of sulfate-bound structure was conducted to obtain an equilibrated system for free energy perturbation (FEP) simulations used to calculate the binding affinity of sulfate in the binding pocket, and therefore, followed a shorter protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, while applying harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) to the protein’s heavy atoms modeled in the cryo-EM structure, and to the modeled sulfate ions in the binding pockets. Furthermore, the z position of phosphorus and oxygen atoms in phospholipids and cholesterols were also harmonically restrained, respectively ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with the protein backbone, and bound sulfate ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ). Prestin conformational transition between the states Two non-equilibrium MD simulations were performed to drive transitions between the contracted and expanded states and to monitor the response of the membrane to these transitions, one in the forward (contracted to expanded) and one in the backward (expanded to contracted) direction. The Cl − -bound cryo-EM structure represented the contracted state, while the salicylate-bound conformation was used to represent prestin in its expanded state. The same protocol as described in ‘System preparation’ was employed to construct the contracted and expanded systems. Since the goal was to investigate the coupling between the protein’s conformational transitions and the membrane, we did not include the bound Cl − and salicylate in these simulations. These transition simulations follow the same driving protocols (but in opposite directions): (1) 10,000 steps of minimization, followed by 5 ns equilibration while restraining the protein’s heavy atoms modeled in the cryo-EM structure, as well as the heavy atoms of the two modeled cholesterol molecules, using a harmonic potential ( k = 10 kcal.mol −1 .Å −2 ). Moreover, harmonic z position restraints were applied to phosphorus and oxygen atoms of phospholipids and bulk cholesterols, respectively ( k = 5 kcal.mol −1 .Å −2 ); (2) 20 ns of equilibration with harmonic restraints on the protein backbone, as well as on the heavy atoms of the two modeled cholesterol molecules ( k = 10 kcal.mol −1 .Å −2 ); (3) keeping the same restraints only on the protein backbone and further equilibrating the system for 150 ns; (4) a 50 ns non-equilibrium simulation, with time-dependent biases applied to drive the conversion from the initial to the target state using the collective variables (colvars) module of NAMD ( Fiorin et al., 2013 ). This transition was achieved by applying biasing forces (k = 1.5 kcal.mol −1 .Å −2 ) on the protein C α atoms to decrease their RMSD with respect to those in the target structure; (5) maintaining the protein in the target state for additional 150 ns by applying the final set of biases (not moving this time) introduced in step (4) . During the transitions between the contracted and expanded forms (the non-equilibrium phase of the simulations in Step 3), the x and y dimensions of the simulation box (membrane dimensions) were recorded every 10 ps. The resulting xy plane area reflects the total cross-sectional area of the membrane-protein system.

Analysis of molecular dynamics trajectories

Molecular images from MD trajectories are generated with VMD ( Humphrey et al., 1996 ). The membrane deformation induced by prestin is quantified by recording the z component of the center of mass of the lipids’ phosphate (PO 4 − ) groups during the last 500 ns of the Cl − -bound trajectories. The origin of the z axis is defined as the bilayer midplane. The resulting heatmap (binned in 2×2 Å 2 bins) in each leaflet captures the spatial distribution of the height of the lipid head groups within the membrane. The dynamics of the Cl − ions at their binding sites were monitored by measuring their displacement from the original (experimentally modeled) positions. When analyzing Cl − dynamics in each binding site, to average out the motion of the binding site itself, the trajectories were aligned using C α atoms of the binding-site residues (Q97, F101, P136, F137, L397, S398, and R399) with respect to those in the cryo-EM structure. Then, the distance between the ion (either from the initial model or from the bulk solution) to the modeled Cl − in the cryo-EM structure was recorded. The distance plots were smoothed using a sliding window of 1 ns. The stability of the docked salicylate in the two subunits was monitored by calculating the displacement and change in the orientation of the ligand throughout the simulations. The displacement was obtained by calculating the center of mass displacement of salicylate from its docked pose in each frame of the trajectory. The orientational change was defined as the angular change of the normal vector to the salicylate’s benzene ring plane with respect to that of the docked pose. Both the displacement and orientation plots were smoothed using a sliding window of 0.2 ns. Free energy calculation of ligand binding affinity FEP simulations ( Cournia et al., 2017 ) were performed on the Cl--, salicylate-, and sulfate-bound structures to compare relative binding affinities of the three ligands. During the employed FEP protocol, the ligand is annihilated from the binding pocket of one prestin protomer while being created at a location in bulk solution 80 Å away from the binding pocket. The FEP calculations of Cl--, salicylate-, and sulfate-bound structures were initiated from the last frame of the Step 2 of the equilibration protocol (see Equilibrium molecular dynamics simulation for more details). During FEP calculations, the protein backbone heavy atoms were restrained (k = 10 kcal.mol-1.Å-2) to the corresponding experimentally modeled positions to avoid undesired structural deviations. All other simulation parameters were the same as in the aforementioned equilibrium simulations. The alchemical transformations were performed in both forward (λ increasing from 0 to 1; ligand annihilation in the binding site and creation in solution) and backward (λ decreasing from 1 to 0; ligand creation in the binding site and annihilation in solution) directions to assess the reversibility. A soft-core van der Waals radius-shifting coefficient of 5 Å was chosen. In the forward transformation, the vdW and electrostatic interactions of the annihilated particles were linearly decoupled to the simulation over a λ range of 0–1 and 0–0.5, respectively, whereas in the backward transformation these interactions were linearly coupled to the simulation in the corresponding λ range. Each FEP calculation along the λ reaction path was divided into 50 equally-spaced consecutive windows to ensure gradual transformation. Within each window, the performed simulation consisted of 5000 steps of minimization, an equilibration phase of 100 ps, followed by 1 ns simulation for data collection at the frequency of every 20 fs, resulting in an accumulated simulation time of 110 ns for each system. To prevent the dissociation of the ligand during the alchemical transformation, harmonic restraints were applied to the annihilated and created ligands (k = 10 kcal.mol-1.Å-2). All the simulations were performed, employing the FEP module of NAMD ( Chen et al., 2020 ). The outputs were analyzed using the ParseFEP ( Liu et al., 2012 ) plugin in VMD, with the statistical error estimated using the implemented Bennett Acceptance Ratio (BAR) method ( Bennett, 1976 ).

Computational Methods

System preparation Cl − -, salicylate-, and sulfate-bound prestin cryo-EM structures were prepared for molecular dynamics (MD) simulations following the procedures described below. Missing side chains and hydrogen atoms were added to the experimental, cryo-EM models, using the PSFGEN plugin ( https://www.ks.uiuc.edu/Research/vmd/plugins/psfgen/ ) in VMD ( Humphrey et al., 1996 ). The experimental models are missing a disordered region in the intracellular domain, dividing the structures into two polypeptide segments (residues 13–580 and 614–724, respectively). Neutral N-terminal and C-terminal ‘caps’ were added to the first and last residues of each polypeptide segment, respectively, using PSFGEN. The coordinates of the bound Cl − and sulfate ions were based on the modeled Cl − in the cryo-EM structure, whereas the bound salicylate was placed in the binding pocket using our in-house docking protocol, GOLEM (see ‘ Computational search for docking poses of salicylate ’ for more details). Following pKa estimation by PROPKA ( Olsson et al., 2011 ; Sondergaard et al., 2011 ), all titratable residues were modeled in their default titration states, which was modeled in its protonated (neutral) form. All protein structures were internally hydrated using the DOWSER ( Gumbart et al., 2009 ; Zhang and Hermans, 1996 ) plugin of VMD. The two cholesterol molecules modeled in the experimental structures were included in the Cl − - and salicylate-bound systems. The lipid bilayer used to embed the proteins was constructed in CHARMM-GUI ( Jo et al., 2008 ). The orientation of the protein in the bilayer was obtained from the OPM (Orientations of Proteins in Membranes) database ( Lomize et al., 2006 ). The protein, along with the bound cholesterol molecules, was then inserted into a heterogeneous lipid bilayer, followed by removing sterically clashing lipid molecules. The bilayer is composed of palmitoyl-oleoyl-phosphatidyl-choline (POPC), palmitoyl-oleoyl-phosphatidyl-ethanolamine (POPE), palmitoyl-oleoyl-phosphatidyl-glycerol (POPG), palmitoyl-oleoyl-phosphatidyl-inositol (POPI), palmitoyl-sphingomyelin (PSM), and cholesterol (CHOL) lipids at a molar ratio of 32:16:5:11:16:20. The protein-membrane system was then solvated with water including 150 mM NaCl in VMD (system size: 335k atoms). To further reduce any bias from the initial lipid configuration and increase the sampling, two independent membrane systems (with randomly placed initial lipids) were generated for Cl − -bound prestin.

Computational search for docking poses of salicylate

The salicylate binding pose in our simulations was obtained using an in-house docking package called GOLEM ( Zhao and Tajkhorshid, 2021 ) This software tool employs a genetic algorithm to explore a ligand’s conformational, orientational, and positional space, with explicit consideration of water displacement, as well as bridging water molecules’ position and orientation, to optimize the fitness, which is defined as the combination of the system’s energy and fitting to the cryo-EM density. Despite the stochasticity of the genetic part of the algorithm, given the large number of iterations. GOLEM robustly reproduces the ligand’s docked pose with negligible differences in repeated trials. The resulting docked pose can be further validated by equilibrium MD simulations, as done in this study. Equilibrium molecular dynamics (MD) simulation All MD simulations were performed using NAMD simulation package ( Phillips et al., 2020 ) and employing the fully atomistic CHARMM36m ( Huang et al., 2017 ) and CHARMM36 ( Klauda et al., 2010 ) force fields for the protein and lipids, respectively. The TIP3 model was used for water molecules ( Jorgensen et al., 1983 ). The salicylate was parameterized using the CHARMM general force field (CGenFF) webserver ( https://cgenff.paramchem.org ) ( Vanommeslaeghe et al., 2010 ; Vanommeslaeghe et al., 2012 ) and further optimized employing the Force Field Toolkit (ffTK) plugin ( Mayne et al., 2013 ) of VMD. Sulfate parameters were adopted from a previous study employing ab initio calculations ( Cannon et al., 1994 ). A 12 Å cutoff was used for short-range, non-bonded interactions, with switching starting at 10 Å. Long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) algorithm ( Darden et al., 1993 ) with a grid density of 1 Å −1 , and a PME interpolation order of 6. All bonds involving hydrogen atoms were kept rigid using the SHAKE algorithm ( Ryckaert et al., 1977 ). Temperature was maintained at 310 K using Langevin thermostat with a damping coefficient of 1.0 ps -1 . Pressure was maintained at 1 atm by the Nosé-Hoover Langevin piston barostat ( Feller et al., 1995 ; Martyna et al., 1994 ), with period and decay of 100 and 50 fs, respectively. All systems were simulated in a flexible cell allowing the dimensions of the periodic cell to change independently while keeping the cell aspect ratio in the xy plane (membrane plane) constant. The simulation timestep was set to 2 fs. Lennard-Jones and PME forces were updated at one and two timesteps, respectively. Atomic coordinates were saved every 10 ps. The two Cl − -bound systems were simulated using the following steps: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, in which harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) were applied to the position of protein’s heavy atoms modeled in the cryo-EM structure, as well as to the heavy atoms of the two modeled cholesterol molecules and the bound Cl − ions. Furthermore, the z position (normal to the membrane plane) of phosphorus atoms in phospholipids and oxygen atoms of bulk cholesterols were also restrained in this step using a harmonic potential ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with only the protein backbone, heavy atoms of the two modeled cholesterol molecules, and the bound Cl − ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ); ( 3 ) 5 ns of equilibration in which all the restraints from the previous step were removed by decreasing the force constant stepwise from 10 to 0 kcal.mol −1 .Å −2 ; ( 4 ) production equilibrium runs for 1.5 μs without any restraints. The simulation of the salicylate-bound system was performed to check the stability of the docked pose of the inhibitor, and therefore, followed a different protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns of equilibration with harmonic restraints ( k = 10 kcal.mol −1 .Å −2 ) on the all the protein’s heavy atoms modeled in the cryo-EM structure, and on the heavy atoms of the bound salicylates, water molecules placed close to the salicylates by the the docking protocol, the two modeled cholesterols. In addition, z positional restraints were applied ( k = 5 kcal.mol −1 .Å −2 ) to phosphorus atoms of phospholipids and oxygen atoms of bulk cholesterols; ( 2 ) 20 ns of equilibration in which the protein backbone atoms were restrained ( k = 10 kcal.mol −1 .Å −2 ), while the salicylate and protein side chains were free to move; ( 3 ) production run extending the simulation in Step 2 to 250 ns. The simulation of sulfate-bound structure was conducted to obtain an equilibrated system for free energy perturbation (FEP) simulations used to calculate the binding affinity of sulfate in the binding pocket, and therefore, followed a shorter protocol: ( 1 ) 10,000 steps of minimization, followed by 5 ns equilibration, while applying harmonic position restraints ( k = 10 kcal.mol −1 .Å −2 ) to the protein’s heavy atoms modeled in the cryo-EM structure, and to the modeled sulfate ions in the binding pockets. Furthermore, the z position of phosphorus and oxygen atoms in phospholipids and cholesterols were also harmonically restrained, respectively ( k = 5 kcal.mol −1 .Å −2 ); ( 2 ) 20 ns of equilibration with the protein backbone, and bound sulfate ions restrained harmonically ( k = 10 kcal.mol −1 .Å −2 ). Prestin conformational transition between the states Two non-equilibrium MD simulations were performed to drive transitions between the contracted and expanded states and to monitor the response of the membrane to these transitions, one in the forward (contracted to expanded) and one in the backward (expanded to contracted) direction. The Cl − -bound cryo-EM structure represented the contracted state, while the salicylate-bound conformation was used to represent prestin in its expanded state. The same protocol as described in ‘System preparation’ was employed to construct the contracted and expanded systems. Since the goal was to investigate the coupling between the protein’s conformational transitions and the membrane, we did not include the bound Cl − and salicylate in these simulations. These transition simulations follow the same driving protocols (but in opposite directions): (1) 10,000 steps of minimization, followed by 5 ns equilibration while restraining the protein’s heavy atoms modeled in the cryo-EM structure, as well as the heavy atoms of the two modeled cholesterol molecules, using a harmonic potential ( k = 10 kcal.mol −1 .Å −2 ). Moreover, harmonic z position restraints were applied to phosphorus and oxygen atoms of phospholipids and bulk cholesterols, respectively ( k = 5 kcal.mol −1 .Å −2 ); (2) 20 ns of equilibration with harmonic restraints on the protein backbone, as well as on the heavy atoms of the two modeled cholesterol molecules ( k = 10 kcal.mol −1 .Å −2 ); (3) keeping the same restraints only on the protein backbone and further equilibrating the system for 150 ns; (4) a 50 ns non-equilibrium simulation, with time-dependent biases applied to drive the conversion from the initial to the target state using the collective variables (colvars) module of NAMD ( Fiorin et al., 2013 ). This transition was achieved by applying biasing forces (k = 1.5 kcal.mol −1 .Å −2 ) on the protein C α atoms to decrease their RMSD with respect to those in the target structure; (5) maintaining the protein in the target state for additional 150 ns by applying the final set of biases (not moving this time) introduced in step (4) . During the transitions between the contracted and expanded forms (the non-equilibrium phase of the simulations in Step 3), the x and y dimensions of the simulation box (membrane dimensions) were recorded every 10 ps. The resulting xy plane area reflects the total cross-sectional area of the membrane-protein system.

Analysis of molecular dynamics trajectories

Molecular images from MD trajectories are generated with VMD ( Humphrey et al., 1996 ). The membrane deformation induced by prestin is quantified by recording the z component of the center of mass of the lipids’ phosphate (PO 4 − ) groups during the last 500 ns of the Cl − -bound trajectories. The origin of the z axis is defined as the bilayer midplane. The resulting heatmap (binned in 2×2 Å 2 bins) in each leaflet captures the spatial distribution of the height of the lipid head groups within the membrane. The dynamics of the Cl − ions at their binding sites were monitored by measuring their displacement from the original (experimentally modeled) positions. When analyzing Cl − dynamics in each binding site, to average out the motion of the binding site itself, the trajectories were aligned using C α atoms of the binding-site residues (Q97, F101, P136, F137, L397, S398, and R399) with respect to those in the cryo-EM structure. Then, the distance between the ion (either from the initial model or from the bulk solution) to the modeled Cl − in the cryo-EM structure was recorded. The distance plots were smoothed using a sliding window of 1 ns. The stability of the docked salicylate in the two subunits was monitored by calculating the displacement and change in the orientation of the ligand throughout the simulations. The displacement was obtained by calculating the center of mass displacement of salicylate from its docked pose in each frame of the trajectory. The orientational change was defined as the angular change of the normal vector to the salicylate’s benzene ring plane with respect to that of the docked pose. Both the displacement and orientation plots were smoothed using a sliding window of 0.2 ns. Free energy calculation of ligand binding affinity FEP simulations ( Cournia et al., 2017 ) were performed on the Cl--, salicylate-, and sulfate-bound structures to compare relative binding affinities of the three ligands. During the employed FEP protocol, the ligand is annihilated from the binding pocket of one prestin protomer while being created at a location in bulk solution 80 Å away from the binding pocket. The FEP calculations of Cl--, salicylate-, and sulfate-bound structures were initiated from the last frame of the Step 2 of the equilibration protocol (see Equilibrium molecular dynamics simulation for more details). During FEP calculations, the protein backbone heavy atoms were restrained (k = 10 kcal.mol-1.Å-2) to the corresponding experimentally modeled positions to avoid undesired structural deviations. All other simulation parameters were the same as in the aforementioned equilibrium simulations. The alchemical transformations were performed in both forward (λ increasing from 0 to 1; ligand annihilation in the binding site and creation in solution) and backward (λ decreasing from 1 to 0; ligand creation in the binding site and annihilation in solution) directions to assess the reversibility. A soft-core van der Waals radius-shifting coefficient of 5 Å was chosen. In the forward transformation, the vdW and electrostatic interactions of the annihilated particles were linearly decoupled to the simulation over a λ range of 0–1 and 0–0.5, respectively, whereas in the backward transformation these interactions were linearly coupled to the simulation in the corresponding λ range. Each FEP calculation along the λ reaction path was divided into 50 equally-spaced consecutive windows to ensure gradual transformation. Within each window, the performed simulation consisted of 5000 steps of minimization, an equilibration phase of 100 ps, followed by 1 ns simulation for data collection at the frequency of every 20 fs, resulting in an accumulated simulation time of 110 ns for each system. To prevent the dissociation of the ligand during the alchemical transformation, harmonic restraints were applied to the annihilated and created ligands (k = 10 kcal.mol-1.Å-2). All the simulations were performed, employing the FEP module of NAMD ( Chen et al., 2020 ). The outputs were analyzed using the ParseFEP ( Liu et al., 2012 ) plugin in VMD, with the statistical error estimated using the implemented Bennett Acceptance Ratio (BAR) method ( Bennett, 1976 ).

Supplementary Material Supplementary Information Movie S1 Movie S1. Related to Figure 4 ; Conformational changes between the Pres-Cl and Pres-Sal structures. Figures S1-S14 Figure S1. Electrophysiology of prestin and mutants, Related to Figures 1 and 2 . A-C . Prestin NLC measurement with the lockin “sine+DC” method. ( A ) Example whole-cell recordings from transfected (prestin+; black) and non-transfected (prestin-; grey) HEK cells. Top panel shows current responses from voltage-clamp experiments, shown in magenta, voltage stimulation with sinusoidal wave removed for visual clarity. Bottom panel shows resulting membrane capacitance (C m ) traces, showing nonlinear capacitance changes prestin+ but not prestin-cells. ( B ) Representative images of transfected prestin+ (top) and prestin- (bottom) HEK cells during whole-cell recording. (C) Plots of NLC relative to membrane voltage. Voltage dependent NLC was observed in prestin+ (black; n=11) but not in prestin-cells (grey; n=5). Open circles represent mean ± SEM. Solid lines represent Maxwell-Boltzmann fit ( Supplemental Information ). D-L . NLC measurement of prestin with the dual-sine method. ( D ) A representative NLC curve of WT prestin and inhibition by salicylate. ( E ) A representative NLC curve of WT prestin and inhibition by 5-methylsalicylate. ( F ) NLC measurement of WT prestin recorded with SO 4 2- internal solution (n=5). ( G ) A representative NLC curve of prestin F101Y mutant and inhibition by salicylate. ( H ) A representative NLC curve of prestin F101Y mutant and inhibition by 5-methylsalicylate. ( I ) NLC measurement of prestin F101Y mutant recorded with SO 4 2- internal solution (n=5). ( J-L ) Changes in NLC ( J ), charged density ( K ), voltage at peak capacitance ( L ) between the WT and F101Y mutant. Detailed fitted parameters can be found in Method section. Figure S2. Biochemistry, cryoEM-raw images and representative 2D averages of prestin, Related to Figures 1 and 2 . A. FSEC analysis of prestin in various detergents. B. SDS-PAGE gel of the pulled peak fractions of Pres-Cl and Pres-Cl-nanodisc after gel filtration. C. Size exclusion chromatography of Pres-Cl. D. Size exclusion chromatography of Pres-Cl-nanodisc. E-H. Representative raw images and 2D averages of Pres-Cl ( E ), Pres-Cl-nanodisc ( F ), Pres-Sal ( G ) and Pres-sulfate ( H ). Figure S3. Cryo-EM data processing of the Pres-Cl dataset, Related to Figures 1 and 5 . A total of 4.6 million particles were automatically picked from 14 k motion corrected micrographs in cryoSPARC. After one round of 2D classification to remove empty picking and obvious junks, 1.4 million bin-2 particles were left for 3D classification into 8 classes, with either C1 or C2 symmetry. 3D Classification with either symmetry results in one good class out of eight, with clear two-fold symmetry and similar particle distribution Particles were refined and re-extracted with bin-1 pixel size (0.651 Å/pix), followed by multiple rounds of local CTF correction and non-uniform (NU) refinement in cryoSPARC, resulting in a 2.34 Å resolution reconstruction. After an additional round of 3D classification with no alignment in Relion 3.1 ( Zivanov et al., 2018 ), local CTF correction and NU refinement in cryoSPARC, the resolution of the reconstruction was improved to 2.31 Å. Figure S4. Local resolution, FSC curve and angular distribution of prestin datasets, Related to Figures 1 , 2 , 5 . Left panel, cryo-EM maps colored according to local resolution estimations. Middle panel, FSC curves of the unmasked (blue) and masked (red) cryo-EM maps, and between the model and the cryo-EM map (purple). Right panel, Euler angle distribution of particles for cryo-EM reconstructions. Figure S5. Local cryo-EM densities of the prestin structures, Related to Figure 1 . A. Local cryo-EM densities of Pres-Cl. B. Cryo-EM densities near the anion binding site in the Pres-Cl structure. Density representing Cl − is colored in pink. C. Cryo-EM densities near the anion binding site in the Pres-Sal structure. Density representing salicylate is colored in pink. Figure S6. Sequence alignment of prestin and homologs, Related to Figure 1 . Key residues in the anion binding pocket are shown in the red boxes. Residues in the extracellular loop that are critical for electromotility are shown in the purple box. N- or C- terminal truncations that abolish the NLC are labeled with blue box. The annotations for the secondary structure is colored according to Fig. 1B . Figure S7. Structural comparison between prestin and homologs, Related to Figure 1 . A . Overall structural comparison between prestin and SLC26A9 (PDB: 7CH1). The structural alignment is based on STAS domain dimer. B. Structural comparison between the STAS domain protomers from the full-length human prestin, full-length human SLC26A9, truncated rat prestin and truncated chicken prestin. Top panel, IVS region is shown with green outline. The conserved region of the STAS domain between species is shown with black outline. Bottom panel, the sphere in red orange color is the Cl − observed in the crystal structure of the truncated rat STAS domain, but absent in other structures. C. Electrostatic potential of prestin. Positive interfaces of the arm domain and IVS-Cα2 near the membrane/cytosol interface were labeled in dash lines. D. Positive residues in the arm domain and IVS-Cα2, as outlined in panel C. E. Positively (blue) and negatively (red) charged residues in the TMD. F. Locations of positively charged (blue), negatively charged (red) and neutral (yellow) residues previously reported to affect the prestin NLC. Detailed residues in TMD region V499, Y501 ( Homma et al., 2013 ); R130, E207, R211, E280, K364, D370, K449, D457, R463, K255, K276, K359 ( Bai et al., 2009 ); S398, R399 ( Gorbunov et al., 2014 ); L93, F101, P136 ( Schaechinger et al., 2011 ); D154, D342 ( Oliver et al., 2001 ); and in cytoplasmic region E516, D518, D522, D524, E527, E528, E531, R571, R572, K573, R576, K577, K580 ( Bai et al., 2006 ). Figure S8. MD simulations investigating the binding stability of Cl − and salicylate, Related to Figures 2 and 3 . A. Two individual runs for the stability of the Cl − binding using the dimeric structure of Pres-Cl. In the left panel, a Cl − unbinding and binding event was observed. B. The pathway for the Cl − pathway in the process of Cl − unbinding and binding in simulation. C. MD simulation for the stability and orientation optimization of the salicylate binding. The initial position and angle is same as it in the cryo-EM structure of Pres-Sal. D-E. Binding free energy of three examined ligands to the binding pocket. (D) The free energy changes of transferring salicylate, Cl − , or sulfate from the binding pocket to the bulk solution. The free energies at λ=0 (ligand-bound) are used as the reference (0 kcal/mol). The unbinding free energies (ΔGunbinding), namely, the free energy changes from λ=0 to λ=1 (ligand-unbound), are 23.28, 13.18, and 8.56 kcal/mol, for salicylate (cyan), Cl- (red), and sulfate (green), respectively. The top-right inset highlights the free energy changes from λ=0.9 to λ=1. (E) The relative unbinding free energies (ΔΔGunbinding) of ligands with respect to sulfate. The binding affinities of salicylate and Cl − are 14.72 kcal/mol and 4.62 kcal/mol higher than sulfate, respectively. Figure S9. CryoEM data processing of Pres-sulfate dataset, Related to Figure 2 . A total of 3.2 million particles were automatically picked from 17 k motion corrected micrographs in cryoSPARC. After one round of 2D classification to remove false positive and junk particles, 880 k particles were left and subjected to global 3D classification with C2 symmetry in Relion 3.1. The single good class was selected and subjected to local 3D classification with 4 classes in Relion 3.1. 3D refinement of the best class results in a 5.2 Å resolution reconstruction in Relion 3.1. Bin-1 (0.676 Å/pix) particles were re-extracted and transferred to cryoSPARC for local CTF correction and NU refinement, resulting in a reconstruction at an overall resolution of 4.36 Å. Figure S10. Prestin structures in intermediate occluded-like states revealed by solvent-excluded surface analysis, Related to Figure 3 A-E. Solvent-excluded surface analysis of Pres-Cl (A), Pres-Sal (B), Pres-sulfate (C), SLC26A9 in inward-facing state (PDB: 7CH1) (D), and AE1 in outward-facing state (PDB: 4YZF) (E). Anion binding site in prestin, SLC26A9 and AE1 structures are labeled with black circle. Volume of the anion binding pocket of the prestin structures are annotated. F. Superimposition of the anion site forming helices TM3 and TM10 of Pres-Cl (yellow), Pres-Sal (cyan), Pres-sulfate (green), SLC26A9 (pink), and AE1 (blue). Alignment was performed based on TM10. The Cα position of the residue equivalent to S398 in prestin, which is critical for anion binding, is indicated by a sphere. Figure S11. Comparison of prestin structures, Related to Figure 3 , 4 A. Alignment and root-mean-square deviation (RMSD) based on all Cα atoms of Pres-Sal (orange) and Pres-Cl (green). B. Alignment and RMSD based on Cα atoms of the STAS domain. C. Alignment and RMSD based on Cα atoms of the core domain. D. Alignment and RMSD based on Cα atoms of the gate domain. E-G. Motion of the core domain after alignment based on the STAS domain. The axis of the rigid body rotation is shown in red. Panels E and F show the motion in context of the full dimer in front and side view. Panel G shows a projection along the rigid body rotation axis (red dot). H-J. Motion of the gate domain after alignment based on the STAS domain. The axis of the rigid body rotation is shown in red. Panels H and I show the motion in context of the full dimer in front and side view. Panel G shows a projection along the rigid body rotation axis (red dot). K Alignment and RMSD based on all Cα atoms of Pres-Cl and Pres-Cl-Nanodisc L. Alignment and RMSD based on all Cα atoms of Pres-Cl and Pres-sulfate M. Alignment and RMSD based on all Cα atoms of Pres-Sal and Pres-sulfate Figure S12. Structural expansion of prestin in the salicylate or sulfate expanded state, Related to Figure 4 A. Cartoon model representations of the prestin structures in the presence of Cl − (grey), salicylate (cyan), or sulfate (pink). The structures are aligned based on the intracellular STAS domain. B. Difference in cross section of the prestin dimer in different conformations along the two-fold axis. The cross section is calculated based on a thresholded gaussian map. Compared to the Pres-Cl structure, the Pres-Sal and Pres-sulfate structures have an average of 205 Å 2 and 245 Å 2 expansion in membrane region. C-E. The cross section calculation based on various thresholds. ( C ) A representative cross section with various thresholds. ( D ) Cross section of Pres-Cl dimers with various thresholds. ( E ) Cross section difference of Pres-Sal and Pres-Cl in various thresholds. F-G. MD simulations calculating the membrane area change from Pres-Cl to Pres-Sal ( F ) or from Pres-Sal to Pres-Cl ( G ). Figure S13. Movement of charged residues in the prestin TMD going from contracted to expanded state, Related to Figure 4 A. Table listing charged residues in the prestin TMD moving towards the cytoplasmic side in Pres-Sali compared to Pres-Cl B. Table listing charged residues in the prestin TMD moving towards the extracellular side in Pres-Sali compared to Pres-Cl C. Arrows indicating the movement of charged residues listed in A and B superimposed on Pres-Cl. The length of the arrows is exaggerated and corresponds to five times the actual movement. Residues with a movement >1 Å along the two-fold axis are annotated. Figure S14. Comparison of bound lipid molecules in different prestin structure, Related to Figure 5 A-B. Comparison of bound lipids between Pres-Cl (A) and Pres-Cl-nanodisc (B). In both panels prestin is shown in grey cartoon representation and the lipid positions modeled in Pres-Cl is shown in red stick representation. The electron-density map corresponding to these lipids are shown in green mesh for Pres-Cl and blue mesh for Pres-Cl-nanodisc. C-D. Comparison of bound cholesterol molecules in Pres-Cl (C), Pres-Cl-nanodisc (D), and Pres-Sal (E). In all panels one protomer of prestin is shown as grey cartoon representation and the cholesterol molecules (which are modeled identically in all structures) are shown in red stick representation. The electron-density maps modeled as cholesterol in Pres.Cl (green), Pres-Cl-nanodisc (blue), and Pres-Sal (orange) are shown in mesh representation.