To examine the effect of anion binding to the dynamics of prestin and SLC26A9, we dialyzed the proteins purified in Cl⁻ into a HEPES buffer lacking other anions. Cl⁻ removal resulted in distinct stability changes for prestin and SLC26A9, manifested by significant HDX acceleration for prestin while mild HDX slowing for SLC26A9 (Figure 1, Figure 1—figure supplement 1). These HDX effects indicate that anion binding induced global stabilization for prestin while slight destabilization for SLC26A9 (Figure 1).

Among the observed HDX responses for prestin, the HDX acceleration at the anion-binding pocket appeared to be the most pronounced and indicates local stabilization induced by anion binding (Figure 2A). In detail, HDX accelerated by 20-fold for the N-termini of both TM3 (Region_136–142: Peptide_134–142 + 9 other peptides) and TM10 (Region_394–397: Peptide_392–397) (Figure 2A, Figure 1—figure supplement 2.22–2.31). This HDX change translates to a difference in free energy of unfolding (ΔΔG) by at least 1.8 kcal/mol; ΔΔGCI bindingbindingsite=1.8kcal/mol. At least four residues in the middle of TM1 exhibited faster HDX (Region_90–101: Peptide_88–101 + 10 other peptides), collectively by 350-fold;ΔΔGCI bindingTM1=3.5kcal/mol (Figure 1A, Figure 1—figure supplement 2.6–2.16). The TM1 region with accelerated HDX included L93, Q97, and F101, residues that are known to participate in the binding pocket (Bavi et al., 2021; Ge et al., 2021; Schaechinger et al., 2011).

SLC26A9 exhibited similar stability as prestin in Cl⁻ for the majority of the TMD, except for the N-terminal TM3 (Region_131–134: Peptide_129–134) which exchanged at least 100-fold slower than that of prestin’s (Region_136–142: Peptide_134–142) (Figure 2). This difference in HDX points to a relatively unstable anion-binding site of prestin as compared to SLC26A9; ΔΔGSLC26A9−prestinN−terminalTM3=2.8kcal/mol, and was also seen in the site-resolved protection factors (PFs) that were obtained by deconvoluting the HDX-MS data using PyHDX (Smit et al., 2021; Figure 2—figure supplement 1).

Compared to the 20- to 350-fold HDX acceleration observed at prestin’s binding site upon Cl⁻ removal, HDX of SLC26A9’s binding pocket was only affected mildly (Figure 2B). These included a slight slowing in HDX for the N-termini of TM3 (Region_131–134: Peptide_129–134) and TM10 (Region_392–395: Peptide_390–395) (Figure 2B). The TM1 (Region_72–92: Peptide_83–92 + 10 other peptides) continued to remain undeuterated even after 27 hr (Figure 2, Figure 1—figure supplement 3.4–3.14), emphasizing the intrinsic high stability of SLC26A9’s anion-binding pocket.

Although the anion-binding pocket is highly conserved and structurally similar across members of the SLC26 family and SLC26A5 families (Figure 2C), mammalian prestin is the only member capable of displaying eletromotility (Santos-Sacchi and Navaratnam, 2022). Hence, the distinct stability responses we observe for dolphin prestin and mouse SLC26A9 point to a prestin’s unique adaptation as a motor protein.

In addition to the binding pocket, we observed stability changes in various regions of the TMDs for prestin and SLC26A9 that may explain their distinct functions. For prestin, anion binding resulted in stabilization for the intracellular cavity but destabilization for regions facing the extracellular milieu (Figure 1C, Figure 1—figure supplement 1A). The stabilizing effects for the cytosol-facing regions were manifested by HDX acceleration upon Cl⁻ removal at the linker between TM2 and TM3, and the intracellular portions of TM7, TM8, and TM9 (Region_128–135, Region_284–294, and Region_354–375) (Figure 1—figure supplement 2.22–2.27, 2.58–2.64, 2.82–2.87). In contrast, HDX slowed for the regions facing the extracellular environment, namely the extracellular ends of TM5b, TM6, and TM7 (Region_250–262 and Region_309–316) (Figure 1—figure supplement 2.45–2.51, 2.66–2.68, 2.71). However, for SLC26A9, anion binding destabilized the cytosol-facing regions, as HDX slowed by ~fivefold upon Cl⁻ removal for the intracellular ends of TM8, TM9, and TM12 (Region_351–369 and Region_440–455) (Figure 1C, Figure 1—figure supplement 1B, Figure 1—figure supplement 3.62–3.63, 3.77–3.82). The distinct thermodynamic consequences of anion binding for prestin and SLC26A9 point to a distinct molecular basis underlying their different functions as a motor and a transporter, respectively.

For prestin in HEPES, which adopted the putative apo state, the 20-fold HDX acceleration for the binding site (Figure 2A) is consistent with a process of local destabilization, even unfolding, or increased solvent accessibility as the region becomes exposed to the intracellular water cavity (Bavi et al., 2021). To investigate these possibilities, we measured prestin’s HDX in response to a chaotrope, urea, which destabilizes proteins by interacting with backbone amides (Lim et al., 2009). In a background of 360 mM Cl⁻, the addition of 4 M urea accelerated HDX for the N-terminus of TM3 (Region_137–140: Peptide_134–140) by 20-fold (Figure 3A), suggesting that this region was destabilized and accessible to urea in its exchange-competent state. In apo prestin, however, the PF at the N-terminus of TM3 was unaffected by urea (after accounting for the known ~50% slowing of the k_chem; Lim et al., 2009; Figure 3A), arguing that this region was already unfolded prior to the addition of urea (Ramesh et al., 2019).

We note that in apparent contradiction to our inference that the N-terminus of TM3 was unfolded in the apo state, its HDX was ~100-fold slower than k_chem. Such apparent PF for an unfolded region has been reported when it is located inside an outer membrane β-barrel, rationalized by the region having a lower effective local concentration of the HDX catalyst, [OD⁻], than in bulk solvent (Zmyslowski et al., 2022; Lin et al., 2022). For prestin, we propose that detergent molecules in the micelle can restrict the access of OD⁻ to amide protons, leading to a local effective pD lower than the bulk solvent and hence producing the apparent PF for the unfolded N-terminus of TM3.

The folding reversibility of the anion-binding site was evaluated by tracking the HDX for 5 min after titrating in Cl⁻ to apo prestin. Deuteration levels for the N-terminus of TM3 (Region_137–140) decreased with increasing Cl⁻ concentration (Figure 3B), suggesting reversible folding upon Cl⁻ binding. Similar behavior was seen in the middle of TM1 (Region_86–101: Peptide_84–101) as Cl⁻ binding stabilized the binding pocket (Figure 3B).

We also examined prestin’s stability in its intermediate states, obtained by replacing Cl⁻ anions with SO₄²⁻ and salicylate (Bavi et al., 2021; Ge et al., 2021). When SO₄²⁻ is the major anion, prestin’s HDX was nearly identical as in Cl⁻, except for a slightly faster HDX at the anion-binding site (Region_128–142 and Region_394–397) at labeling times longer than 10³ s (Figure 4). This mild HDX response suggested a slightly destabilized binding site while the remaining regions retained normal dynamics as in Cl⁻. In the presence of salicylate, HDX slowed across the TMD, with the greatest effect seen at the anion-binding site (10-fold; ΔΔGsalicylate−Cl−bindingsite=1.4kcal/mol) (Figure 4), indicating that salicylate binding to prestin globally stabilized the TMD, primarily at the anion-binding site. These stability changes provide a thermodynamic context to the cryo-EM structures (Bavi et al., 2021).

We chose to measure the HDX of prestin in GDN micelle to match the cryo-EM conditions (Bavi et al., 2021; Ge et al., 2021; Butan et al., 2022). Structures of GDN-solubilized prestin in Cl⁻ obtained by three research groups are indistinguishable (Cα RMSD <1 Å), demonstrating the reproducibility and robustness of the system.

To evaluate the dynamics in a more native membrane environment, we measured HDX of prestin reconstituted in nanodisc (porcine brain total lipid extract). Except for TM6, HDX for prestin in nanodisc highly resembled that in micelles including the anion-binding pocket (Figure 5). Such high agreement between the folding stability in these two membrane mimetics suggest that our findings on prestin’s anion-binding site and its folding equilibrium are pertinent to prestin residing in a lipid bilayer. This is not surprising because structures for human prestin in GDN and nanodisc are shown to be nearly identical (Cα RMSD = 0.2 Å) (Ge et al., 2021).

Interestingly, nanodisc-embedded prestin displayed a less stable TM6 for the intracellular portion than prestin in micelles, manifested by the 10-fold HDX increase (Region_275–282: Peptide_273–282 + 2 peptides) (Figure 5, Figure 1—figure supplement 2.53–2.55). TM6 defines the interface between prestin and the lipid bilayer, and has been proposed to mediate area expansion through helical bending (Bavi et al., 2021). The exact role of TM6 in regulating prestin’s conformational cycle is currently under investigation.

Our broad HDX time range and dense sampling allowed us to observe effects at the residue level. In particular, the binding site of prestin (Region_128–140 and Region_394–397) exhibited a broad deuterium uptake curve in Cl⁻, indicative of helix fraying where exchange of deuterium occurs from multiple states that differ by one hydrogen bond (Figure 6A). Such HDX pattern is consistent with the associated residues undergoing sequential unfolding with distinct PFs (i.e., stability). Site-resolved PFs obtained using PyHDX (Smit et al., 2021) support that the stability increased residue-by-residue for TM3 for amide protons located further away from the substrate (Figure 6—figure supplement 1). This gradual increase in residue stability along the helices is indicative of helix fraying starting from prestin’s binding site.

In contrast, we observed much more cooperative unfolding in prestin’s lipid-facing helices, with exchange occurring from one or a few high energy states where a set of hydrogen bonds are broken concertedly. Cooperatively exchanging residues have similar PFs and a characteristic sigmoidal deuterium uptake curve for the associated peptide, as seen in prestin’s intracellular portion of TM6 (Region_275–282: Peptide_273–282 + 4 peptides) (Figure 6A, Figure 1—figure supplement 2.53-2.57).

To characterize the degree of cooperativity for the HDX at the N-terminus of TM3 (Peptide_134–140) and the intracellular portion of TM6 (Peptide_273–282), we fit the deuterium uptake curves as a sum of exponentials (Hamuro, 2021), D(t)=∑i=1n(1−e−kit), where ki is the exchange rate and n is the number of exponentials, ranging from one to the number of exchange-competent residues. The value of n was determined by the quality of the fit, evaluated by χ² and having a relative error smaller than one (i.e., standard deviation for ki less than ki itself). HDX data for the N-terminus of prestin’s TM3 (Peptide_134–140) were fit with four well-separated exponentials with rates spanning five log units for the four residues (Figure 6A). The need to individually fit each site indicates a lack of cooperativity and helix fraying. In contrast, the peptide representing the intracellular portion of TM6 (Peptide_273–282) has eight residues yet it could be fitted with only three rates spanning less than two log units (Figure 6A). This rather concerted deuterium uptake was independent of the anion substrate identity and also observed for TM1, TM5b, the intracellular portion of TM7, and the N-terminus of TM8 (Figure 1—figure supplement 2.6-2.16, 2.42–2.43, 2.45–2.51, 2.58–2.63, 2.78–2.79).

We define a parameter σ_ΔG to quantify the degree of folding cooperativity. The value of σ_ΔG is calculated as the standard deviation for the free energies of unfolding (∆G) for exchange-competent residues comprising the peptide. When a region folds 100% cooperatively, σ_ΔG is zero as all residues have the same ∆G. As the diversity increases (lower cooperativity), the σ_ΔG value becomes larger. The accuracy of the ∆G determination at residue level can be increased by comparing uptake curves for overlapping peptides and/or deconvoluting isotope envelopes (Hamuro, 2021). Here, we assigned exchange rates (ki), obtained from the fitting method mentioned above, to residues based on the directionality of helix fraying, with residues closer to the end of a helix having faster rates. When there is ambiguity on which rate to assign to a given residue, the geometric mean of the rates was used (Materials and methods). We found that prestin’s intracellular portion of TM6 (Peptide_273–282) has σ_ΔG = 1.1, indicating mild cooperativity, whereas the non-cooperative N-terminal TM3 (Peptide_134–140) has a σ_ΔG = 2.9. This significant decrease in folding cooperativity for helices directly involved in the Cl⁻-binding site likely has functional consequences related to prestin’s electromotility, as discussed below.

We associate the HEPES solvent condition to an apo state as the HEPES anion is too big to fit into the chloride site and prestin has a right-shifted NLC (Rybalchenko and Santos-Sacchi, 2008), To investigate further whether HEPES anion binds, we determined the structure of prestin in the HEPES-based buffer using cryo-EM. Our initial attempt to solve the structure in the complete absence of Cl⁻ was unsuccessful due to sample aggregation under cryogenic conditions. Aggregated particles were greatly reduced in the presence of 1 mM Cl⁻, and we were able to solve the structure of GDN-solubilized prestin in 1 mM Cl⁻, containing 190 mM HEPES, at a nominal resolution of 3.4 Å (Appendix 3, Appendix 3—figures 1 and 2, Appendix 3—table 1). Surprisingly, under these conditions prestin adopted a ‘compact’ conformation, virtually identical to the previously reported Cl⁻-bound ‘Up’ state (Bavi et al., 2021), displaying a clear density at the anion-binding site (Appendix 3—figure 1). This density is incompatible with the placing of a HEPES molecule, and we reason it to be a small population of Cl⁻-bound prestin resulting from a weak Cl⁻ affinity (e.g., EC₅₀ = 6 mM¹¹ implies 17% bound). This result is consistent with the fundamental role of bound anions in the conformational stability of prestin and supports a new role for the folding equilibrium of the anion-binding site in the mechanism of voltage sensing. Ultimately, however, understanding the underlying mechanism for prestin’s electromotility necessitates consideration of physiological elements such as membrane potential, kHz frequency, and protein–lipid interactions.

We applied HDX-MS to the study of prestin’s electromotility and identified folding events that are likely critical for function but had escaped detection by cryo-EM. The folding equilibrium of the Cl⁻-binding site and its dependence on Cl⁻ concentration appears to rationalize the conundrum of how an anion that binds in proximity to a positive charge (R399), can enable the NLC of prestin. We directly compared the dynamics of prestin in nanodisc to in micelles and identified TM6 as a potential mechano-sensing helix. We observed fraying of the helices forming the anion-binding site, which contrasts with cooperative unfolding of the lipid-facing helices. We believe that the non-cooperative fraying of the helices involved in voltage sensing may allow for fast charge movements within the electric field. This heightened sensitivity of the voltage sensor then induces large-scale motions of the lipid-facing helices, enabled by their cooperativity (or allostery), thereby altering the cross-sectional area of prestin. These principles warrant further investigation.

Generation of the dolphin prestin and mouse SLC26A9 constructs, protein overexpression, and purification were conducted using the same protocol as previously described (Bavi et al., 2021). The HEK293S GnTI⁻ cells in suspension that were used for protein expression and purification were obtained and authenticated from ATCC (CRL-3022). Following the cleavage of the GFP tag, the protein was purified by size-exclusion chromatography (SEC) on a Superose 6, 10/300 GE column (GE Healthcare), with the running buffer being either the ‘Cl⁻/H₂O Buffer’ or the ‘SO₄²⁻/H₂O Buffer’, including 1 μg/ml aprotinin and 1 μg/ml pepstatin (Bavi et al., 2021). The ‘Cl⁻/H₂O Buffer’ contained 360 mM NaCl, 20 mM Tris, 3 mM dithiothreitol (DTT), 1 mM Ethylenediaminetetraacetic acid (EDTA), and 0.02% GDN at pH 7.5. The ‘SO₄²⁻/H₂O Buffer’ contained 140 mM Na₂SO₄, 5 mM MgSO₄, 20 mM Tris, 0.02% GDN, and 10–15 mM methanesulfonic acid to adjust the pH to 7.5. Peak fractions containing the sample were concentrated on a 100 K MWCO centrifugal filter (Millipore) to 2–3 mg/ml, flash-frozen in liquid nitrogen, and kept at −80°C until use.

For prestin reconstitution into nanodiscs, porcine brain lipid extract (Avanti) was first solubilized in ‘Cl⁻/H₂O Buffer’ supplemented with 3% GDN to a 10 mg/ml concentration through successive rounds of sonication and freeze–thaw cycles. MSP1E3D1 was purified as previously described (Alvarez et al., 2010.Clark et al., 2022). Reconstitution was performed using the ‘on-column’ method (Clark et al., 2022) using a 1:5:600 (prestin:MSP:lipid) molar ratio, with the final buffer consisting of 20 mM Tris–HCl and 150 mM NaCl at pH 7.5.

For structure determination by cryo-EM, prestin purification was performed in buffers containing 360 mM NaCl as previously described (Bavi et al., 2021), except that the SEC running buffer now consisted of 190 mM HEPES, ~95 mM Tris-base (used to adjust the pH to 7.5), 1 mM NaCl, 3 mM Dithiothreitol (DTT), 0.02% GDN, 1 μg/ml aprotinin, and 1 μg/ml pepstatin. Peak fractions containing the sample were concentrated to 3 mg/ml and used immediately for grid preparation.

Table 1 provides biochemical and statistical details for HDX in this study per recommendations by Masson et al., 2019. HDX reactions, quench, and injection were all performed manually. Prior to HDX, proteins purified in Cl⁻ and SO₄²⁻ were buffer exchanged to the same H₂O buffer without protease inhibitor using 7 K MWCO Zeba spin desalting columns (Thermo 89882). For HDX conducted in HEPES, proteins purified in Cl⁻ were dialyzed against the ‘HEPES/H₂O Buffer’ (150 mM HEPES, 0.02% GDN, pH adjusted to 7.5 by HEPES acid or base) using a 10 K MWCO dialysis device (Thermo Slide-A-Lyzer MINI 69570) with three times of buffer exchange for near-complete Cl⁻ removal. HDX in a solution of 93% deuterium (D) content was initiated by diluting 2 μl of 25 μM prestin or SLC26A9 stock in an H₂O buffer into 28 μl of the corresponding buffer made with D₂O (99.9% D, Sigma-Aldrich 151882). HDX was conducted in one of the two conditions: (1) pD_read 7.1, 25°C; (2) pD_read 6.1, 0°C. The D₂O buffers contained the same compositions as the corresponding H₂O buffers, except that Tris was replaced with phosphate for the ‘SO₄²⁻/D₂O Buffer’ for both HDX conditions, and the ‘Cl⁻/D₂O Buffer’ for HDX performed at pD_read 6.1, 0°C. The pD_read was adjusted to 7.1 or 6.1 by DCl for the ‘Cl⁻/D₂O Buffers’ and by NaOD for other D₂O buffers. For HDX in the presence of salicylate, 50 mM salicylate acid was added to the ‘SO₄²⁻/D₂O Buffer’ and the pD_read was adjusted by NaOD. For HDX in the presence of urea, 4 M urea was added to the ‘Cl⁻/D₂O Buffer’ or the ‘HEPES/D₂O Buffer’, with accurate urea concentration determined to be 4.16 and 4.54 M, respectively, by refractive index using a refractometer (WAY Abbe) (Warren and Gordon, 1966).

HDX was quenched at various times, ranging from 1 s to 27 hr, by the addition of 30 μl of ice-chilled quench buffer containing 600 mM glycine, 8 M urea, pH 2.5. For HDX in the presence of urea, urea concentration in the quench buffer was adjusted to reach a 4 M final concentration. For HDX on prestin in nanodisc, the quench buffer also included 3 μl of 0.8% GDN and 3 μl of 300 mg/ml aqueous suspension of ZrO₂-coated silica (Sigma-Aldrich, reference no. 55261-U). The resulting mixture was incubated on ice for 1 min to remove lipids and solubilize prestin in GDN before being filtered through a cellulose acetate spin cup (Thermo Pierce, Waltham, MA, reference no. 69702) by centrifugation for 30 s at 13,000 × g, 2°C. Quenched reactions were immediately injected into a valve system maintained at 5°C (Trajan LEAP). Non-deuterated controls and MS/MS runs for peptide assignment were performed with the same protocol as above except D₂O buffers were replaced by H₂O buffers, followed by the immediate addition of the quench buffer and injection.

HDX reactions were performed in random order. No peptide carryover was observed as accessed by following sample runs with injections of quench buffer containing 4 M urea and 0.01% GDN. In-exchange controls accounting for forward deuteration toward 41.5% D in the quenched reaction were performed by mixing D₂O buffer and ice-chilled quench buffer prior to the addition of the protein. Maximally labeled controls (All D) accounting for back-exchange were performed by a 48-hr incubation with the ‘Cl⁻/D₂O Buffer’ at 37°C, followed by a 30-min incubation with 8 M of deuterated urea at 25°C.

Upon injection, the protein was digested online by a pepsin/FPXIII (Sigma-Aldrich P6887/P2143) mixed protease column maintained at 20°C. Protease columns were prepared in-house by coupling the protease to a resin (Thermo Scientific POROS 20 Al aldehyde activated resin 1602906) and hand-packing into a column (2 mm ID × 2 cm, IDEX C-130B). After digestion, peptides were desalted by flowing across a hand-packed trap column (Thermo Scientific POROS R2 reversed-phase resin 1112906, 1 mm ID × 2 cm, IDEX C-128) at 5°C. The total time for digestion and desalting was 2.5 min at 100 μl/min of 0.1% formic acid at pH 2.5. Peptides were then separated on a C18 analytical column (TARGA, Higgins Analytical, TS-05M5-C183, 50 × 0.5 mm, 3 μm particle size) via a 14 min, 10–60% (vol/vol) acetonitrile (0.1% formic acid) gradient applied by a Dionex UltiMate-3000 pump. Eluted peptides were analyzed by a Thermo Q Exactive mass spectrometer. MS data collection, peptide assignments by SearchGUI version 4.0.25, and HDX data processing by HDExaminer 3.1 (Sierra Analytics) were performed as previously described (Zmyslowski et al., 2022; Lin et al., 2022).

In our labeling convention, we name capitalized regions according to the third residue of each peptide since the first two residues have much faster k_chem (Bai et al., 1993; Nguyen et al., 2018) and hence, exhibit complete back-exchange. Labeling times for HDX performed in pD_read 6.1, 0°C were corrected to those in pD_read 7.1, 25°C by a factor of 140, determined by the ratio of the k_chem for full-length proteins in the two conditions – prestin: τchempDread 6.1, 0 °C = 10 s, τchempDread 7.1, 25 °C = 74 ms; SLC26A9: τchempDread 6.1,  0∘C = 9 s, τchempDread 7.1,  25∘C = 65 ms. Deuteration levels were adjusted with the 93% D content but not with back-exchange levels except for Figure 6A. For HDX in the presence of urea (Figure 3A), D contents of 76% and 74% were used to account for the volumes of 4.16 and 4.54 M urea in the ‘Cl⁻/D₂O Buffer’ and the ‘HEPES/D₂O Buffer’, respectively.

HDX rates, PFs, and changes of free energy of unfolding (ΔΔGs) in Results were estimated by fitting uptake curves of each peptide, after correcting for back-exchange levels, with a stretched exponential as described by Hamuro, 2021, except for discussion relevant to Figure 6A. Peptides with less than 10% D at the longest labeling time (~10⁵ s) were not used for fitting. The residue-level ΔΔG values presented in the full-length proteins in Figure 1C were estimated by averaging ΔΔG values for peptides containing the residue. We note that this stretched exponential method is only a rough approximation to extract ΔΔGs and our major conclusions are not dependent on this fitting method.

For fitting HDX data and extracting rates relevant to Figure 6A, HDX data were first normalized to in- and back-exchange levels. Given the helices are likely to exchange by fraying, the exchange rate for each residue was assigned based on their distance from the end of the helix. When a residue could not be assigned to a single rate, the geometric mean of the possible rates was used, for example, the HDX data for the 8-residue Peptide_273–282 were well fit with three exponentials, and the three associated rates, k₁, k₂, and k₃, were assigned to the 8 residues according to k₁, k₁, (k₁k₁k₂)^1/3, k₂, k₂, (k₂k₂k₃)^1/3, k₃, and k₃. These rates were used to calculate folding stability according to ΔG=−RTln(kchem/ki−1).

A hybrid statistical analysis used to generate Figure 1—figure supplement 1 was performed as described by Hageman and Weis, 2019, with significance limits defined at α = 0.05.

Simulations were conducted in our Upside molecular dynamics package (Jumper et al., 2018a; Jumper et al., 2018b) using a membrane thickness of 38 Å. Missing residues for prestin (PDB 7S8X) were built using MODELLER (Sali and Blundell, 1993) and the placement within the bilayers was accomplished using Positioning of Proteins in Membranes webserver (Lomize et al., 2022). Local restraints in the form of small springs between nearby residues were used to maintain the native structure of cytosolic domains, as well as to the TM13–TM14 helices. Also, the distance between the two TMDs was held fixed. We ran 28 temperature replicas between 318 and 360 K.

The cryo-EM data were collected from three separate samples and microscope sessions. For dataset 1, 3.5 μl of prestin sample was applied to Quantifoil 200-mesh 1.2/1.3 Cu grids (Quantifoil) that were plasma cleaned for 30 s at 20 W. For datasets 2 and 3, 3.5 μl of prestin sample was applied to UltrAuFoil 300-mesh 1.2/1.3 grids (Quantifoil UltrAuFoil) that were plasma cleaned for 40 s at 20 W. The remaining sample preparation and imaging conditions were kept constant for all three samples. Grids were blotted at 22°C and 100% humidity with a blot time of 3.5  s and a blot force of 1, and flash-frozen into liquid ethane using a Vitrobot Mark IV (Thermo Fisher). Grids were imaged at The University of Chicago Advanced Electron Microscopy Facility on a 300 kV Titan Krios G3i electron microscope equipped with a Gatan K3 camera in CDS mode, a GIF energy filter (set to 20  eV) and with magnification set to ×81,000, corresponding to a physical pixel size of 1.068 Å. Movies were acquired at a dose of 1.2 e⁻/Ų for 50 frames (corresponding to a total dose of 60 e⁻/Ų) and a defocus range of −0.7 to −2.1 μm. 2153 movies were collected for dataset 1, 1928 movies for dataset 2, and 6665 movies for dataset 3.

Cryo-EM data processing was performed using Relion-4.0 (Kimanius et al., 2021) and cryoSPARC v4.1 (Punjani et al., 2017). Movies from the different datasets were motion-corrected independently in Relion with a bin-1 pixel size of 1.068 Å. The motion-corrected micrographs were then combined and imported into cryoSPARC for Patch CTF Estimation. Unless otherwise mentioned, the following steps were performed in cryoSPARC. Particles were picked using template-based particle picking. The initial picks were curated using Inspect Picks and extracted at a box size of 256. Two rounds of 2D Classification were performed to filter out additional ‘junk’ particles, resulting in a set of 328,033 particles. Two consecutive rounds of ab initio reconstruction, followed by heterogeneous refinement with C1 symmetry were performed to obtain a stack of 170,313 particles. These particles were used as input for an initial round of non-uniform and local CTF refinement using C2 symmetry and exported to Relion for Bayesian Polishing. Additional 3D Classification was performed in Relion to remove additional ‘junk’ particles, resulting in a set of 170,313 particles. However, no additional classes were found. The final set of polished particles was then imported back into cryoSPARC and subjected to a final round of non-uniform and CTF refinement, resulting in a map with a nominal resolution of 3.4 Å, according to the gold-standard 0.143 Fourier shell correlation (FSC) criterion (Rosenthal and Henderson, 2003). Local resolution was estimated using local resolution estimation in cryoSPARC.

A previous model of dolphin prestin (PDB 7S8X) was roughly fit into the density map and used as a template for model building. Initially, only a monomer was considered for model building, and the fitting was improved by running Phenix real space refinement (Afonine et al., 2018) with secondary structure restraints, morphing, and simulated annealing enabled. Subsequently, the monomer model was iteratively refined by manual inspection in Coot (Brown et al., 2015) and real space refinement without morphing and simulated annealing in Phenix. After several rounds of refinement, the second monomer was added using the apply_ncs tool in Phenix with C2 symmetry, and the resulting dimer model was subjected to an additional round of manual inspection in Coot. A chloride ion was manually added into the density observed in the anion-binding pocket in Coot. All figures related to the cryo-EM structure were prepared using UCSF ChimeraX (Pettersen et al., 2021).