Glutamate Transporter Homolog-based Model Predicts That Anion-π Interaction Is the Mechanism for the Voltage-dependent Response of Prestin*

Background: The structure of the transmembrane domain of prestin and its mechanism of action are unknown. Results: GltPh-based model predicts that aromatic residues bind intracellular anions through anion-π interactions. Conclusion: Aromatic residues in the proposed ion tunnel confer prestin with a unique capability to perform electromechanical conversions. Significance: Anion-π interactions are identified as a novel mechanism to explain unique voltage-dependent property of prestin. Prestin is the motor protein of cochlear outer hair cells. Its unique capability to perform direct, rapid, and reciprocal electromechanical conversion depends on membrane potential and interaction with intracellular anions. How prestin senses the voltage change and interacts with anions are still unknown. Our three-dimensional model of prestin using molecular dynamics simulations predicts that prestin contains eight transmembrane-spanning segments and two helical re-entry loops and that tyrosyl residues are the structural specialization of the molecule for the unique function of prestin. Using site-directed mutagenesis and electrophysiological techniques, we confirmed that residues Tyr367, Tyr486, Tyr501, and Tyr508 contribute to anion binding, interacting with intracellular anions through novel anion-π interactions. Such weak interactions, sensitive to voltage and mechanical stimulation, confer prestin with a unique capability to perform electromechanical and mechanoelectric conversions with exquisite sensitivity. This novel mechanism is completely different from all known mechanisms seen in ion channels, transporters, and motor proteins.

Mammalian hearing is extremely sensitive and exhibits exquisite frequency selectivity. The inner ear possesses an active process that can overcome the viscous damping and structural impedance imposed by the architecture and mechanical property of the organ of Corti. The electromotile property of cochlear outer hair cells, a mammalian cellular innovation, underlies this active amplification (1)(2)(3)(4). The fundamental molecular basis of outer hair cell electromotility is the motor protein prestin (2,5,6). As a motor protein, prestin is unique in that its conformational change is dependent on voltage, and the speed of such change is in the microsecond range (7).
Prestin, encoded by the Slc26A5 gene, is a functionally distinct member of the solute carrier 26 (SLC26A) 4 anion transporter protein family (8,9). The overall tertiary structure of prestin consists of three domains: the N-terminal domain (ϳ100 amino acid residues), a highly conserved central core of hydrophobic amino acids (ϳ400 amino acid residues), and the C-terminal domain (ϳ240 amino acid residues) (8). The N-and C-terminal domains are both located in the cytoplasmic side of the plasma membrane, whereas the hydrophobic core is embedded in the cell membrane. The sulfate transporter (SulTP) signature sequence is in the hydrophobic core, whereas the sulfate transporter and anti-factor antagonist (STAS) domain is in the C-terminal domain (10). Prestin can respond to voltage change (signified by the presence of a gating current) and undergo conformational change (reflected by length change of outer hair cells). Such voltage-dependent properties lie in distinct regions within the SulTP domain (residues 70 -512) (11)(12)(13)(14). Nevertheless, the underlying mechanism and the identity of residue(s) associated with the voltage sensor are still unresolved. Studies using site-directed mutagenesis and electrophysiology have suggested that intracellular anions, namely chloride and bicarbonate, serve as an extrinsic voltage sensor (6). The alternative view is that movement of charged residues modulated by chloride ions in an allosteric manner serves as a voltage sensor (15,16). Nevertheless, the questions of which residues are the anion-binding residue(s) and how they interact with the anions are still controversial. Determination of the mechanism for voltage sensing and the residues for interaction with anions is the key to understanding how prestin works.
The SLC26A family of anion transporters has been included as a distant member in the amino acid-polyamine-organocation (APC) superfamily (17)(18)(19). Saier and co-workers (19,20) placed the sulfate permease family of transporters within the APC branch that includes the anion exchanger (SLC4), nucleobase-cation symporter-2 (SLC23), and benzoate transporter/ permease transporters. This group of transporters represents the most distant and distinct branch of the APC superfamily with the sulfate permease transporter family representing an outlying cluster. Typically, the APC transporters display a 2-fold pseudosymmetrical topology consisting of four to seven transmembrane (TM)-spanning regions (19). The benzoate transporter/permease, anion exchanger 1, and nucleobasecation symporter-2 transporters are furthermore unique within the APC superfamily hierarchy by exhibiting a 7-7 TM topology, which is supported by the available crystal structures for anion exchanger 1 (21) and UraA (22). On the basis of these data, it was suggested that sulfate permease transporters also should exhibit a 7-7 TM topology. Recently, the structure of UraA was used as a template to predict a model of the SulTP domain of prestin (23). The membrane topology of the model was validated by the substituted cysteine accessibility method. However, the UraA model does not appear to be able to explain the mechanism underlying the voltage sensing, anion binding, and piezoelectric property of prestin.
To circumvent the unavailability of x-ray crystallography data for prestin, we used a combination of computational folding recognition, homology modeling, and molecular dynamics (MD) simulations to determine the three-dimensional structure of the SulTP domain. Our computational model suggested that prestin has structural similarity to the Pyrococcus horikoshii glutamate symport protein (Glt Ph ) (10,24). Specific residues within the internal anion binding site (residues Tyr 367 and Tyr 486 ) and adjacent to the intracellular tunnel entrance (residues Tyr 501 and Tyr 508 ) participate in the voltage sensor mechanism of prestin. The non-covalent anioninteractions (25)(26)(27)(28) of anions with the aromatic rings of Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 furnish prestin with a unique capability to bind to anions and respond to voltage and mechanical stimulations.

Experimental Procedures
Fold Prediction and Homology Modeling of Prestin-Appropriate templates for the three-dimensional structural determination of the SulTP domain (residues 70 -512) of prestin were selected by submitting eutherian orthologs from rat, gerbil, and human and a consensus eutherian sequence (29) to the pGen-THREADER profile-based fold recognition program (30).
Homology models were built using the YASARA software package (31). For all modeling, the hm_build.mcr macro of the YASARA package was used with the following parameters: modeling speed, slow; number of PSI-BLAST iterations in template search, 6; maximum allowed PSI-BLAST E-value to consider template, 0.5; maximum number of templates to be used, 5; maximum number of templates with same sequence, 1; maximum oligomerization state, 1; maximum number of alignment variations per template, 5; maximum number of conformations tried per loop, 100; and maximum number of residues added to the termini, 10. The homology modeling procedure only included energy minimizations of structures in water; therefore, the resultant structures were further refined. The models were embedded in a phosphatidylethanolamine lipid bilayer, solvated with water molecules, and submitted to 1-ns MD simulation using the md_runmembrane.mcr macro of YASARA.
MD Simulation-The final homology model was submitted to three independent 100-ns MD simulations with the DESMOND molecular dynamics package (32) using the OPLS-AA/L force field (33). N-Acetyl and N-methylamide protecting groups were added to the N and C termini, respectively, to preserve the electronic structure of the backbone as in prestin. The protein was embedded in a lipid bilayer composed of 142 1-palmitoyl-2-oleoylphosphatidylcholine molecules and solvated with 15,713 SPC water molecules. The net charges were neutralized with Cl Ϫ ions. The structure of the system first was relaxed by using the MultiSim script of DESMOND that includes a series of energy minimizations and short MD simulations. The system was then equilibrated with 1-ns NPT, constant number of molecules, constant pressure (1 bar), and constant temperature (310 K) simulation. The pressure and temperature were kept constant with a Berendsen barostat (relaxation time, 2 ps) and thermostat (relaxation time, 1 ps), respectively. For short range non-bonded interactions, a 9-Å cutoff was used, and the long range electrostatic interaction was treated with the smooth particle mesh Ewald method. The equation of motion was integrated with the multistep RESPA scheme (34) with time steps of 2, 2, and 6 fs for the bonded, short range, and long range non-bonded interactions, respectively. Bonds involving H-atoms were constrained to their equilibrial values with the SHAKE method (35). For the production simulations (100 ns), the same parameters were used. Using Visual Molecular Dynamics (VMD), trajectories were converted to GROMACS (36) format and analyzed with the GROMACS analysis software suite. Cluster analyses of the trajectories were performed by the GROMOS method of clustering (37) using a backbone root mean square deviation (r.m.s.d.) cutoff of 0.15 nm. Secondary structural changes during simulations were determined by using the Dictionary of Secondary Structure of Proteins (DSSP) program (38).
Peptide Synthesis and Purification-To preserve the electronic structure of the backbone as in the whole protein, the fragments were synthesized as N-terminal acetyl-and C-terminal amide-protected peptides. The syntheses were performed on a CEM Liberty Microwave peptide synthesizer (CEM Corp., Matthews, NC) at a 0.1-mmol scale using Rink amide aminomethyl resin and a 5 M excess of N ␣ -Fmoc (N-(9-fluorenyl)methoxycarbonyl)-protected amino acids as described previously (39).
Electronic Circular Dichroism (ECD) Spectropolarimetry-ECD spectra of peptides were recorded using a Jasco J-810 spectropolarimeter. ECD spectra were obtained by averaging 20 scans from 185 to 250 nm at 100 nm/min. The peptides were dissolved either in 20 mM KH 2 PO 4 buffer (pH 7), 50% 2,2,2trifluoroethanol (TFE) in the same buffer (v/v), or 100% TFE, and their spectra were measured in a quartz cell with a 0.05-cm path length. The TM8 fragment of prestin was only soluble in 80% acetonitrile (ACN). The ECD spectra of the peptides were measured at peptide concentrations of 25, 50, and 100 M to ensure that the peptide concentration did not affect the shape of the spectrum. The solutions were kept at 4°C with a Jasco PFD-425 single position Peltier thermostated cell holder. Background spectra of the solvents were subtracted. Mean residue molar ellipticities were calculated using peptide concentrations determined by quantitative reverse phase HPLC (40). To determine the secondary structure content of peptides, CDSSTR analysis (41) was performed with data set 6 using the DichroWeb website (42).
Cell Culture and Transient Transfection-Human embryonic kidney (HEK) cells were cultured in DMEM solution (Invitrogen) supplemented with 10% fetal bovine serum. Prestin or mutants (4 g) were introduced into the dishes using 10 l of Lipofectamine (Invitrogen) following the protocol from the manufacturer.
Confocal Imaging-Membrane-associated expression of prestin mutations was verified 24 h after transfection by confocal imaging performed on an LSM 510 microscope (Zeiss, Thornwood, New York, NY) using a 40ϫ water immersion objective.
Electrophysiology-Nonlinear capacitance (NLC) measurements using whole cell voltage clamp techniques were performed 24 -48 h after transfection on cells with robust membrane-associated EGFP expression. During electrophysiological recordings, cells were bathed in an extracellular solution containing 120 mM NaCl, 2 mM MgCl 2 , 2 mM CoCl 2 , 20 mM triethanolamine, 10 mM HEPES, and 10 mM 4-aminopyridine. Recordingpipetteswerepulledfromborosilicateglasswithresistances between 2.5 and 5.0 megaohms and filled with a solution containing 140 mM CsCl, 2 mM MgCl 2 , 10 mM EGTA, and 10 mM HEPES. Membrane capacitance was measured using a two-sine wave voltage stimulus protocol (10-mV peak at both 390.6 and 781.2 Hz) with subsequent fast Fourier transform-based admittance analysis (43) from a holding potential of 0 mV. The capacitive currents were sampled at 100 kHz and low passfiltered at 5 kHz. Series resistance was compensated offline. The capacitive currents were amplified by Axopatch 200B (Molecular Devices, Sunnyvale, CA). Data were acquired using jClamp (Scisoft, New Haven, CT) and analyzed with Igor (WaveMetrics, Portland, OR).
The NLC can be described as the first derivative of a twostate Boltzmann function relating nonlinear charge movement to voltage (44,45). The capacitance function is described as follows.
where Q max is maximum charge transfer, V 1/2 is the voltage at which the maximum charge is equally distributed across the membrane or equivalently the peak of the voltage-dependent capacitance, C lin is linear capacitance, and ␣ ϭ ze/kT is the slope of the voltage dependence of charge transfer where k is the Boltzmann constant, T is absolute temperature, z is the valence of charge movement, and e is electron charge. The Q max was normalized by C lin in the results. The C lin is proportional to the surface area of the membrane (the size of the cell). Because C lin is not relevant to this study, we subtracted the linear capacitance and only presented the NLC as a function of voltage for the data presented under "Results." To compare the magnitude of NLC obtained from different cells with different sizes, we normalized the NLC by C lin . Data were collected from cells with membrane resistance greater than 500 megaohms after rupturingandexhibitingnormalmembranecapacitance(C m )andresistance (R m ) values. Series resistance was compensated offline. For each construct, NLC data were acquired from cells representing at least four separate transfection experiments. Data were analyzed with Student's t test, and significance was considered at p Ͻ 0.05.

Homology Modeling of the SulTP Domain of Prestin
Nine different structures primarily consisting of three groups of proteins with a highly significant fold characteristic similar to the SulTP domain (residues 70 -512) of prestin (p Ͻ 0.001) were identified with the pGenTHREADER three-dimensional fold recognition method (30). The first group represented proteins with a solenoid protein domain and included human cullin-1 ligase (Protein Data Bank code 1U6G), Saccharomyces cerevisiae exportin Cse1p (Protein Data Bank code 1WA5), rat clathrin-associated adaptor protein (Protein Data Bank code 2VGL), and human exportin-5 (Protein Data Bank code 3A6P). The solenoid domain most closely resembles the topological models of prestin with the up and down ␣-helical membranespanning motifs (6,46). The next group consisted of two chloride-conducting (CLC) chloride channel proteins, the Salmonella enterica serovar Typhimurium CLC chloride channel (CLCst) (Protein Data Bank code 1KPL) and the Escherichia coli CLC chloride channel (CLC-ec1) (Protein Data Bank code 1OTS). The third group represented five different transporter proteins. These were the Glt Ph sodium-dependent aspartate transporter (Protein Data Bank code 2NWL), the P. horikoshii proton-glutamate symport protein (Protein Data Bank code 1XFH), the Microbacterium liquefaciens nucleobase-cation symport-1 (Protein Data Bank code 2JLN), the Aquifex aeolicus Na ϩ /Cl Ϫ -dependent transporter (LeuT) (Protein Data Bank code 3F3A), and the Vibrio parahaemolyticus sodium/galactose symporter (vSGLT) (Protein Data Bank code 3DH4). The bovine cytochrome c oxidase polypeptide I (Protein Data Bank code 1V54) also exhibited a highly significant folding pattern similar to that of prestin. However, for our subsequent analyses, it was selected as a functionally dissimilar protein.
Using YASARA (31), SLC26A5 ortholog and paralog sequences from different species (zebrafish, chicken, opossum, gerbil, and rat) were examined for fold similarities with the above nine proteins. Attempts to model prestin paralogs and orthologs using only the CLC-ec1 structure as a template (47) did not yield any viable structural models; therefore, from the above three groups of proteins, nine templates (1WA5, 1KPL, 1OTS, 2NWL, 1XFH, 2JNL, 3F3A, 3DH4, and 1V54) were selected to obtain hybrid models of prestins. Of these templates, 1V54 and 1WA5 never had significant alignments with prestin and were discarded. The hybrid models that were built on the other seven templates had better Z-scores (Ϫ2.217) than those of the individually built models. Furthermore, rat prestin had the best Z-score (Ϫ2.014) using Glt Ph (2NWL) as a template. Although the overall sequence identity between the SulTP domain of prestin and the Glt Ph protein is 13.1%, the three-dimensional structure of the protein is more conserved than the sequence (48,49). Therefore, our final homology model of the rat prestin ( Fig. 1) was created by using four x-ray structures (2NWL, 2NWW, 2NWX, and 1XFH) of Glt Ph with a satisfactory Z-score of Ϫ1.837. The major structural features of the model are eight TM-spanning domains, TM3-TM4 extracellular loop (EL3-4), two helical pin (HP) re-entry loops, and the ICF, the loop that connects TM2 and TM3 ( Fig. 1a and supplemental Data S1). The overall structure can be further divided into an outer shell represented by TM1-TM6 and an inner core (HP1, HP2, TM7, TM8, and ICF) that contain the putative transport-associated and anion-sensing features of prestin.

MD Simulations
To validate and refine the Glt Ph -like structure of the SulTP domain of rat prestin, we performed three independent 100-ns MD simulations in an aqueous lipid-membrane environment. The structural stability of prestin was evaluated by calculating the r.m.s.d. of the backbone atoms from the starting structure along the trajectories (Fig. 2a). During an initial 40 ns of each of the three simulations, the structure of prestin modestly deviated from that of the initial homology model-based structure, and r.m.s.d. values reached equilibrium with fluctuation around an average value. These deviations from the initial homology model-based structure, however, did not result in substantial structural reorganization. All three simulations resulted in similar structural features; thus we show our results from simulation 1. Per residue root mean square fluctuations (RMSF) from the mean positions of individual C␣ atoms showed that the largest motions took place at ICF and HP2 (Fig.  2b). ICF and its movement are unique to our proposed prestin model. A similar large motion was detected at HP2 and is comparable with that in the structural template Glt Ph protein (50), indicating similar dynamic properties. Cluster analysis of the trajectories generated 19, 30, and 18 clusters for simulations 1, 2, and 3, and the largest cluster contained 58.2, 29.5, and 58.5% of all sampled structures, respectively. The relatively high portion of structures in the largest cluster of two trajectories further indicates that prestin forms a stable structure. Fig. 2c shows the central structure of the largest cluster from simula- In addition to a stable overall structure, a stable ␤-hairpin structure was formed by residues Ile 140 -Val 152 of the ICF (Fig. 2c and  Fig. 3). To show that the ␤-hairpin formation was not a computational artifact, we chemically synthesized the Ser 133 -Ile 156 fragment of prestin and used ECD spectropolarimetry to explore its conformation. ECD measurements confirmed that ICF forms a ␤-sheet conformation in an aqueous buffer ( Fig. 4 and Table 1).

Intramolecular Interactions Play Role in Chloride Ion Sensing
In the Glt Ph protein, the inner core forms a channel for substrate binding and translocation (50). We propose that, in a similar fashion, residues within the inner core of prestin play an important role for interaction with anions. Close scrutiny of the MD trajectories revealed that either invariant residues in the eutherian orthologs or conserved residues within the eutherian prestins contribute to the overall stability of the inner core and could provide a unique structural framework for sensing intracellular chloride ions. We observed three different types of interactions (Table 2): ionic, hydrogen bonding, and weakly polar interactions (51). The latter includes aromatic-aromatic (-) interactions between side chains of aromatic amino acids, -backbone (-bb) interactions between aromatic side chains and peptide backbone, and -CH interactions between aromatic side chains and CH groups of side chains. Ionic interactions Lys 77 -Glu 374 , Arg 197 -Glu 468 , Lys 283 -Glu 374 , and Lys 449 -Glu 207 (Fig. 5) stabilize the structure of the inner core, whereas Arg 399 -Asp 485 ionic interaction (Figs. 2, d and e, and 6a) positions the side chain of Arg 399 so that the ␦ CH 2 group interacts with the aromatic side chain of Tyr 486 through -CH interaction with 0.4-nm average distance between them (52) (Figs. 2, d  and e, and 6f). In turn, the location of the aromatic side chain of Tyr 486 is stabilized in space so that it interacts with the aromatic side chain of Tyr 367 though ainteraction (Figs. 2, d and e,  and 6, b and c). The distance between the center of the two aromatic rings and the angle between the planes of the two aromatic rings are around 0.4 nm and 110°, respectively, which are in agreement with x-ray protein structures in the Protein Data Bank (53). Residue Tyr 501 of TM8 is located at the intracellular tunnel entrance of the inner core tunnel and interacts with the peptide backbone of residues Gly 379 -Ile 388 through a -bb interaction (Figs. 6e and 7a). These residues connect HP1 and TM7. Residue Tyr 508 of the intracellular tail interacts with several hydrophobic residues of the ICF through -CH interactions ( Table 2 and Fig. 6, g-i).
Data mining of the Protein Data Bank for anion-interactions involving aromatic side chains of amino acid residues and inorganic anions such as Cl Ϫ , Br Ϫ , PO 4 3Ϫ , and NO 3 Ϫ revealed that Cl Ϫ ion interacted most frequently among the anions (28). Furthermore, Cl Ϫ was in close contact with His, Tyr, Phe, and Trp residues, respectively, in decreasing order of frequency. Energetic and electron density analysis showed that the interaction is weak, and nearby polar, charged, and H-bond-donating residues may enhance the interactions (26). Tyrosyl residues of HP1 and TM8 are located inside of our proposed tunnel (Fig. 7) so they can attract Cl Ϫ ions and facilitate the ion movement into and within the tunnel.

Experimental Validation of Structure
ECD Spectropolarimetry-To experimentally validate our structural model, we selected the TM8 fragment (residues Ser 478 -Arg 502 ) of the inner core of prestin (Fig. 1a) for chemical synthesis and subsequent structural analysis by ECD spectropolarimetry. In the Glt Ph structural template, residues of TM8 play an important role for substrate binding (50). We hypothesized that similar features might exist for prestin. As a control, we also synthesized the Gly 388 -Glu 418 TM8 fragment of Glt Ph . As a transmembrane segment, TM8 fragment of prestin was hydrophobic, and we were not able to measure its ECD spectrum in an aqueous buffer but were able to measure its ECD spectrum in 80% acetonitrile. The ECD spectra were also determined in 50 and 100% TFE. Using organic solvents other than a strictly aqueous buffer can be justified by the fact that the peptide fragments are out of context with the rest of the protein and   OCTOBER 2, 2015 • VOLUME 290 • NUMBER 40 its membranous environment. Therefore, it is possible that helical structures will not form spontaneously in an aqueous buffer system. Thus, it is a common practice to use the secondary structure-promoting agent TFE in ECD measurements (54). The weak signal of the prestin TM8 fragment in 80% acetonitrile solution could be attributed to either a ␤-sheet or an aggregated structure (Fig. 4). Increasing the TFE concentration induced an ␣-helical conformation in both prestin and Glt Ph TM8 fragments ( Fig. 4 and Table 1). The increasing ␣-helical conformation is consistent with structural properties of many well characterized fragments of different transmembrane proteins and indicates the correct structural assignment for TM8 in prestin.

Anion-Interactions in Voltage Sensing of Prestin
Site-directed Mutagenesis and Electrophysiology-We examined the proposed structure of prestin with site-directed mutagenesis and electrophysiological measurement from HEK cells transfected with different sets of mutant prestin. Membrane expression of the mutant prestins was examined by confocal microscopy. A unique feature of our prestin model is ICF (residues Ser 133 -Ile 156 ), which extends into the intracellular region and forms an antiparallel ␤-sheet conformation (Figs. 2c   and 3). Because ICF is outside the TM-spanning structure of prestin, we assumed that prestin can still fold into a SulTP structure without ICF. Therefore, we expressed a variant of prestin in which residues Ser 133 -Ile 156 were replaced by a Gly-Gly dipeptide. Robust membrane expression of the mutant protein in HEK cells was detected 24 h after transfection as shown in Fig. 8a. Thus, it appeared that deletion of 23 amino acid residues of the ICF did not affect membrane targeting of the truncated protein. To determine whether deletion of ICF would affect prestin function, we measured NLC from HEK cells transfected by the mutant prestin. Wild-type gPres was used as a control. NLC, reflecting voltage-dependent charge movement arising from the redistribution of charged "voltage sensors" across the membrane, is usually used to describe the voltagedependent characteristics of prestin. NLC is characterized by a bell-shaped dependence on membrane potential with a peak between Ϫ80 and Ϫ0 mV (44,45). Fig. 8a shows NLC responses measured from HEK cells transfected by gPres (n ϭ 12) and ICF-truncated prestin (n ϭ 8). As shown, gPres exhibits a robust bell-shaped dependence on membrane potential with a peak capacitance close to 9% of the total membrane capacitance near Ϫ73 mV. We obtained two parameters (V 1/2 and z value) from nonlinear curve fitting of the NLC response using the first derivative of the Boltzmann function (Fig. 4b). V 1/2 is the voltage at which half of the total charges are translocated across the membrane, and z is the valence of charge movement. These two parameters are often used to define the magnitude and voltage dependence of NLC (44,45). Curve fitting of normalized NLC of gPres yielded a V 1/2 value of Ϫ76.5 Ϯ 5.3 mV and a z value of 0.74 Ϯ 0.09. These numerical values are consistent with those reported in previous studies (5,6,16). In contrast, the NLC response of the truncated prestin was substantially reduced with a z value of 0.39 Ϯ 0.09, which is significantly less than that of gPres (p Ͻ 0.01). The peak of NLC was also shifted toward depolarization with a V 1/2 value of Ϫ32 Ϯ 5.3 mV. Taken together, these results suggest that, although deletion of ICF did not affect protein folding and membrane targeting, its voltagedependent capability was significantly reduced but not completely eliminated, indicating the functional role of ICF.
The four tyrosyl residues (Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 ) of HP1 and TM8 are the putative anion-binding residues of prestin. Consequently, their modification might affect the NLC properties of prestin. We substituted these residues with either Ala, Arg, Phe, or Trp. We first measured NLC from HEK cells transfected with plasmids with substitution of Tyr 367 and Tyr 486 with either Ala, Arg, or Phe. These residues, which are

TABLE 2 Non-bonded interactions of the inner core of prestin
Red-colored residues are invariant within the eutherian orthologs, whereas the black-colored residues are conserved only in eutherian prestins. The location of a residue in prestin is given in parenthesis.
a Types of interactions: -, between side chains of aromatic amino acids; -bb, between aromatic side chains and peptide backbone; -CH, between aromatic side chains and CH groups of side chains. b HP1 does not move. c Backbone -backbone and side chain -backbone hydrogen bonds. d Peptide bond between the two residues.   Refs. 52,53). Red lines are running averages at every 5-ns interval. a, distance between center of masses of the -␦ NHC(NH 2 ) 2 ϩ group of Arg 399 and the -␤ COO Ϫ of Asp 485 . b, angle between the planes of aromatic rings of Tyr 367 and Tyr 486 ; planes of the aromatic rings were assigned as a plane formed by the C␥, C⑀1, and C⑀2 atoms. c, distance between the center of mass of the aromatic rings of Tyr 367 and Tyr 486 . d, distance between center of masses of the aromatic ring of Tyr 508 and the -␦ CH 2 group of Leu 142 . e, distance between center of masses of the aromatic ring of Tyr 501 and the peptide bond between Gly 379 and Ile 380 . f, distance between center of masses of the aromatic ring of Tyr 486 and the -␦ CH 2 group of Arg 399 . g, distance between center of masses of the aromatic ring of Tyr 508 and the -␦ CH 2 group of Leu 151 . h, distance between center of masses of the aromatic ring of Tyr 508 and the -␥1 CH 3 and -␥2 CH 3 groups of Val 149 . i, distance between center of masses of the aromatic ring of Tyr 508 and the -␦ CH 3 group of Ile 144 . located at the top of the tunnel, are part of the internal anion binding site in our model (Fig. 7). As shown in Fig. 8b, the magnitude of NLC was significantly reduced after either Tyr 367 or Tyr 486 was replaced by either Ala, Arg, or Phe. This is reflected by a significant reduction of the z value for all mutant prestins. The values of z and V 1/2 (n ϭ 8 for each mutation) are presented in Fig. 8b. Furthermore, either Y367A or Y367R mutation resulted in almost complete elimination of the NLC, indicating a more significant role of Tyr 367 in binding to Cl Ϫ ion.
As part of the proposed internal anion binding site, Tyr 501 and Tyr 508 are located at the intracellular tunnel entrance (Fig.  7a). These residues were substituted with either Ala, Arg, Phe, or Trp, and the NLC responses of the mutant prestins were recorded. As shown in Fig. 9, a and b, mutations of either of these two residues resulted in a significant reduction of NLC and the z values. The more negative V 1/2 values for Y501R and Y508R mutants indicate that the arginyl residue binds the Cl Ϫ ion more strongly than other residues, resulting in a decreased z value. We also did double mutations of these two residues with substitution by Phe or Trp. As shown in Fig. 9c, double mutations also caused a significant reduction of the NLC response and z value.

Discussion
In the present study, a predictive three-dimensional model of the SulTP domain of rat prestin was built using x-ray structures of the Glt Ph protein as templates for homology modeling. The structural properties of the model were validated with subsequent MD simulations. On the basis of MD simulations, site-directed mutagenesis, and electrophysiological measurements, we propose a novel mechanism for the unique voltage-dependent function of prestin. Our results indicate that the aromatic side chains of the four tyrosyl residues (Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 ) form non-covalent bonds with intracellular anions through anion-interactions. In this model, anions in the tunnel form non-covalent bonds with the aromatic side chains inside the tunnel at the resting condition (Fig. 10). Depolarizing voltage drives Cl Ϫ ions toward the cytosol and disrupts the binding. Translocation of the anions and uncoupling of the non-covalent bonds results in gating current and conformational change of prestin. Conversely, hyperpolarization forces the Cl Ϫ ion to move into the tunnel to facilitate the binding. Voltage or mechanical force can easily disrupt the anioninteractions holding the anions and induce ion movement because anion-interactions are an order of magnitude weaker than ionic interactions. This mechanism may underlie the unique capability of electromechanical and mechanoelectrical (piezoelectric) conversions of prestin with high sensitivity to voltage and mechanical stimulations. The fact that both the anions and the interactions are necessary for the generation of gating current suggests that neither the intracellular anions nor the four tyrosyl residues alone are sufficient for the voltage sensing of prestin. Our model reconciles the two competing models of how prestin works. On one hand, we support the notion that there is no intrinsic voltage sensor (6), the charged residues that can move across the electric field in the same way as that seen in voltage-gated ion channels. On the other hand, we also support the idea that non-covalent bonds between anions and residues are critical for voltage sensing as proposed by Santos-Sacchi and co-workers (15,16).
Our model has revealed a number of important features unique to prestin. It suggested that prestin contains eight TMspanning segments with EL3-4, an ICF, and two HP re-entry loops. This is in contrast to previously proposed models containing 10 -14 TM regions (10,23). The structure of our Glt Phlike model is further divided into an outer shell (TM1-TM6) and an inner core (HP1, HP2, TM7, and TM8). The unique feature of the structure is the ICF, which functionally belongs to the inner core and is neighbored in the sequence by the signature sulfate transport (4) and motile eutherian SLC26A helper motifs (12). These two motifs contribute significantly to the integrity of the unique functional properties of prestin.
Analysis of MD trajectories predicted that four tyrosyl residues (Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 ) of the inner core structure are functional components for anion binding (Fig. 8). All four residues participate in weakly polar interactions, which enhance their anion binding capability (28). At the intracellular entrance of the tunnel, Tyr 508 interacts with residues of ICF, whereas Tyr 501 interacts with residues that connect HP1 and TM7. ICF has an intrinsic capability to undergo large scale motion (Fig. 2b) that may support the entry of Cl Ϫ ions into the tunnel. Thus, the magnitude of NLC was substantially decreased (Fig. 8a) when the ICF was deleted. At the top of the tunnel, Tyr 367 and Tyr 486 interact with each other, and Tyr 486 also interacts with Arg 399 . These two tyrosyl residues together with Arg 399 and Asp 485 form an occlusion of the tunnel (Fig. 7). Thus, complete anion transport through the tunnel will not likely take place.
We indicated that anions can form non-covalent bonds with an aromatic ring of Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 residues through anion-interactions. Among weakly polar interactions (51), the anion-interactions are a new type of non-covalent bonds that can stabilize protein structure (27). Through such interactions, inorganic anions (such as Cl Ϫ , Br Ϫ , PO 4 3Ϫ , and NO 3 Ϫ ) can bond to aromatic residues in the range of 3.5-4.5 Å and with a 60 -90°angle between the vector that connects the anion to the aromatic ring center and the plane of the ring (28). However, the nature and strength of these interactions in proteins are not yet well explored in part because relatively low numbers of such ions are found in the Protein Data Bank. Nevertheless, for Cl Ϫ -interactions, ՅϪ10 kcal mol Ϫ1 interaction energy is predicted by DFT and LMP2 quantum chemical calculations (26) in line with other weakly polar interactions but strong enough to bind anions. Different anions vary in affinity for aromatic rings. This can explain why prestin shows different affinities for different anions (6). For example, the EC 50 of Cl Ϫ for prestin is 6.3 mM, whereas that for HCO 3 Ϫ is 43.6 mM (6). It is now possible to explain previous experimental results using our three-dimensional model. For example, studies have shown that Val 499 and Tyr 501 are critical for NLC and electromotility (55). However, the mechanism of how mutation of these two residues can lead to significant reduction of NLC and outer hair cell motility was unclear. In our model, these two residues, which are located near the junction between the TM8 and the intracellular STAS domain, are positioned at the entrance of the tunnel (Fig. 7). Together with the ICF, Tyr 501 directly regulates the anion movement in and out of the tunnel. This explains why mutation of these two residues can cause loss of NLC and motility (55). Comparative analysis shows that prestin is the only member of the SLC26A family containing the combination of Tyr 367 , Arg 399 , Asp 485 , and Tyr 486 , thus making the substrate binding region of prestin unique. Mutation of any of these residues can result in reduction of NLC. Site-directed mutagenesis experiments have demonstrated that R399Q (56) and R399S (23) substitutions significantly altered NLC properties. D485Q mutation, however, did not alter the NLC of prestin (56). These results are consistent with our study. In our threedimensional model, Arg 399 directly interacts with Tyr 486 ; therefore, its substitution affects the polarity of the aromatic ring. The D485Q mutation most likely preserves the interaction with Arg 399 and does not induce structural change. Using the x-ray structure of the bacterial uracil transporter UraA as the template, Gorbunov et al. (23) recently proposed a new structural model with 14 TM domains to explain how prestin works. Their model identified different structural components for the voltage-dependent response of prestin. However, the UraAbased model is unable to identify the mechanism for anion transport, voltage sensing, and motor-based movement that facilitates electromotility. Although their study and ours essentially used the same methodologies, the structural components and mechanisms proposed in the two studies are fundamentally different. First of all, in the UraA-based model, residues Ser 133 - Ile 156 (identified as the ICF in our model) are in the TM3 region and constitute part of the permeation pathway. Therefore, it is reasonable to expect that the protein would not be able to be expressed in the membrane if a large portion of a TM chain was removed. Our experiment, however, clearly showed that prestin without the ICF was expressed in the membrane. The expression of truncated protein without ICF strongly suggests that these residues are not in the TM region. In addition, the adjacent signature sulfate transport (residues 127-130) (4) and motile eutherian SLC26A helper (residues 158 -178) motifs (12) further support the importance of the region encompassing the ICF structure in prestin function. Experiments that alter these motifs by site mutagenesis and/or swapping have demonstrated the functional importance of these two motifs. In the UraA-based model, these are located in the first intracellular and the second extracellular loops, respectively, and as such are unlikely to exert a functional impact. This strongly argues against the viability of the UraA-based model proposed by Gorbunov et al. (23). Second, our experiments demonstrated that Tyr 367 , Tyr 486 , Tyr 501 , and Tyr 508 are part of the structural and functional components for anioninteractions. In the UraAbased model, these residues are either located outside of the proposed anion binding sites or in the shell of the transporter structures. Based on their locations in the UraA-based model, none of the tyrosyl residues should have any impact on the physiological properties of prestin. We clearly showed that they are critical for prestin function. Thus, the UraA-based model is incompatible with our results.
Conceptually, prestin should have at least two essential functional domains: the voltage sensor that detects changes in the membrane potential and the actuator that undergoes a conformational change. Anions need to be translocated across the membrane in response to changes in the transmembrane voltage, moving toward the extracellular surface following hyperpolarization and toward the cytoplasmic side in response to depolarization. As a consequence, this translocation triggers conformational changes in the protein that ultimately alter its surface area in the plane of the plasma membrane. Although our Glt Ph -based model of mammalian prestin does not explain how anion movement can trigger conformational change of the molecule, our new model provides a major step forward in elucidating the structure-function relationships of prestin. Furthermore, although the Glt Ph protein is a trimer and prestin is known to function in homooligomeric form (57, 58), our model does not address the issue of oligomerization. Previous studies have shown that each prestin monomer is mechanically independent in the oligomer (57,58) and interacts with cytoplasmic partners to facilitate the mechanical response.
Prestin is the only known protein that can respond to mechanical force (deformation) by generating a current (59 -61). This piezoelectric property is an order of magnitude better than any man-made piezoelectric crystal (60). None of the existing models so far can explain this. In our model, anions are held by the weak non-covalent bonds inside the tunnel and can easily be expelled by both voltage and mechanical stimulation. Because the bonds are weak, conversion efficiency from mechanical stimulation to current is high. The identification of an anion binding site and the mechanism for high sensitivity to voltage and mechanical stimulations are critical to understanding how prestin works. Our threedimensional structure can contribute to the design of nanomotors for applications in medicine, protein chemistry, and biomechanical engineering.