Conserved Dimeric Subunit Stoichiometry of SLC26 Multifunctional Anion Exchangers*

The SLC26 gene family encodes multifunctional transport proteins in numerous tissues and organs. Some paralogs function as anion exchangers, others as anion channels, and one, prestin (SLC26A5), represents a membrane-bound motor protein in outer hair cells of the inner ear. At present, little is known about the molecular basis of this functional diversity. We studied the subunit stoichiometry of one bacterial, one teleost, and two mammalian SLC26 isoforms expressed in Xenopus laevis oocytes or in mammalian cells using blue native PAGE and chemical cross-linking. All tested SLC26s are assembled as dimers composed of two identical subunits. Co-expression of two mutant prestins with distinct voltage-dependent capacitances results in motor proteins with novel electrical properties, indicating that the two subunits do not function independently. Our results indicate that an evolutionarily conserved dimeric quaternary structure represents the native and functional state of SLC26 transporters.

The solute carrier 26 (SLC26) 5 gene family was defined ten years ago by expression cloning of a Na ϩ -independent SO 4 Ϫ transporter from rat liver (1). Subsequently, many homologs were found in mammals, non-mammals, plants, fungi, and bacteria (2). The functional importance of this family is highlighted by several inherited human diseases caused by mutations in SLC26 genes (3,4), such as diastrophic dysplasia (SLC26A2), congenital chloride diarrhea (SLC26A3), Pendred syndrome (a syndrome comprising sensorineural deafness and enlarged thyroid (SCL26A4)), and inner ear deafness (SLC26A5 or pres). The symptomatic variety of these diseases illustrates the diversity of cellular functions performed by this class of transport proteins.
Mammalian SLC26s are the members of this family that have been functionally studied the best. Most of them operate as anion exchangers, and distinct isoforms differ significantly in anion specificity. Some transport monovalent and divalent anions, whereas others transport only monovalent anions (1,(5)(6)(7)(8)(9). There are also family members that do not function as anion exchangers, most notably, SLC26A5 or prestin, the motor protein in the outer hair cells (10). Outer hair cells in the mammalian cochlea change length in response to acoustic signals, and this mechanical amplification enhances the hearing sensitivity by Ͼ40 db. The basis for these length changes are voltage-dependent conformational changes of prestin that are transformed into somatic length changes of outer hair cells (11). Moreover, recent work suggested that SLC26A7 and SLC26A9 are not anion carriers, but anion channels (12,13). The SLC26 family thus exhibits an amazing variety of functions, yet the molecular basis of this diversity is poorly understood. As a fundamental step to understand structure-function relationships, we decided to study the subunit stoichiometry of various SLC26 homologs from humans, rat, zebrafish, and Pseudomonas aeruginosa. We demonstrate that all tested isoforms exhibit a dimeric subunit stoichiometry.

Expression of His-or YFP Fusion Proteins in Xenopus
Oocytes or in Mammalian Cells-pGEMHE-hSLC26A3-His CT (polyhistidine tag at the carboxyl-terminal end) encoding the human SLC26A3 (GenBank TM accession number NM_000111) (14), pGEMHE-rprestin-His CT encoding the rat prestin (GenBank TM accession number NM_030840), pGEMHE-zfprestin-His CT encoding prestin from Danio rerio (zebrafish) (GenBank TM accession number BX571796) (15), and pGEMHE-His NT -hClC-1 (polyhistidine tag at the amino-terminal end; GenBank TM accession number NM_000083) plasmids were generated by adding a cDNA fragment encoding six histidine residues carboxyl-or amino-terminal to the coding region by PCR and subcloning into pGEMHE vector (16). A construct encoding a rat prestin concatamer was generated by covalently linking two prestin copies in a single open reading frame by a cDNA sequence containing four repeats of the tripeptide AGS (17). PASulP (GenBank TM accession number NP_250338) was identified by searches for rat prestin homologues using BLAST searches at the National Center for Biotechnology Information (NCBI). It was amplified by PCR from genomic P. aeruginosa DNA and subcloned into pGEMHE using BamHI and HindIII restriction sites. Transcription of cRNAs and handling of oocytes was performed as described (18).
To create YFP fusion proteins, the YFP cDNA was excised from pEYFP-N1/pECFP-N1 (Clontech) and inserted into pcDNA3.1 (Invitrogen) connected to the coding sequences of rat and zebrafish prestin and human SLC26A3 by short linker sequences in a single open reading frame. SLC26 proteins were expressed in tsA201 cells as described (19). For some experiments, stable inducible cell lines, generated by selecting Flp-In T-REx cells (Invitrogen) transfected with pcDNA5/FRT/TO-YFP-rprestin, were used after 24-h incubation with 1 g/ml tetracycline. Point mutations were introduced using the QuikChange TM method. All constructs were verified by restriction analysis and DNA sequencing. Supplemental Fig. S1 gives an alignment of the SLC26 isoforms characterized in this study.
Functional Characterization of SLC26A3 and Prestin-Radioactive chloride uptake into oocytes was studied 4 -5 days after cRNA injection. Oocytes were preincubated for 10 min in chloride-free medium (98 mM potassium gluconate, 1.8 mM calcium gluconate, 5 mM HEPES-Tris, pH 7.5) and then transferred to the uptake solution (100 mM potassium gluconate, 5 mM HEPES-Tris, 3 mM 36 Cl, pH 7.5) above 300 l of mineral oil. After various time periods, 36 Cl uptake was terminated by centrifuging the oocytes into the mineral oil layer. Scintillation counting was performed after lysis of the oocytes in scintillation counting tubes containing 100 l of 0.5% SDS. For each incubation period, uptake was also determined for at least two uninjected oocytes. These control values were averaged and subtracted from radioactive uptake levels measured on injected oocytes. Additional control experiments were performed with oocytes expressing SLC26A3 transporters incubated in the uptake solution supplemented by 100 M DIDS. DIDS was dissolved in Me 2 SO at a concentration of 10 mM and diluted into the uptake solution.
Transfected HEK293 or tsA201 cells were studied through whole cell patch clamping using an Axopatch 200B amplifier. In experiments measuring voltage-dependent capacitances, the external solution contained (in mM): 140 NaCl, 4 KCl, 2 CaCl 2 , 1 MgCl 2 , 5 HEPES, at pH 7.4, and the intracellular solution contained (in mM): 120 NaCl, 2 MgCl 2 , 5 EGTA, 10 HEPES, at pH 7.4. The initial pipette resistances were 1-2.5 megohms. Stray pipette capacitance was neutralized before establishing the whole cell configuration. All data acquisition and analysis were performed with the Windows-based patch clamp program, jClamp (SciSoft, Ridgefield, CT). Non-linear charge movement was calculated using a continuous high resolution (2.56-ms sampling) two-sine stimulus protocol (10-mV peaks at both 390.6 and 781.2 Hz) superimposed onto voltage steps (150-ms duration) from either Ϫ200 to ϩ80 mV or Ϫ160 to ϩ120 mV (20). The voltage dependence of the non-linear capacitance was fit to the first derivative of a two-state Boltzmann function (21).
where ␤ ϭ z(e o ␦/k B T). In addition, Q max is the maximum charge transferred across the membrane; V1 ⁄ 2 is the potential at half-maximal charge transfer; z is the number of elementary charges, e o , displaced across a fraction, ␦, of the membrane dielectric; k B is the Boltzmann constant, and T is the absolute temperature.
In co-transfected cells, the voltage dependence of the nonlinear capacitance was fit with, where p D154N is the percentage of subunits carrying the D154N mutation. C D154N , C D342Q , and C Heterodimer are voltagedependent capacitances of homogenous populations of homodimeric mutant prestin or of heterodimeric prestin. Q max was assumed to be identical for each non-linear capacitance term.
To Either immediately after the pulse or after an additional chase period, the radiolabeled oocytes were extracted with digitonin (1.0%) in 0.1 M sodium phosphate buffer, pH 8.0. His-tagged proteins were isolated by metal affinity chromatography using Ni 2ϩ -NTA-agarose (Qiagen), as detailed previously (18), with the modification that iodoacetamide was routinely included at 10 mM and 1 mM in the lysis and washing buffers, respectively (22). In some experiments, 1% perfluoro-octanoic acid, ammonium salt (PFO, Sigma-Aldrich) was used as detergent for membrane protein extraction and purification in 0.1 M sodium phosphate buffer, pH 8.0, with and without 10 mM iodoacetamide as indicated. Proteins were eluted from Ni 2ϩ -NTA-agarose with PFO elution buffer consisting of 250 mM imidazole/HCl and 1% PFO at pH 7.4 and stored at 0°C until analysis later on the same day.
Chemical Cross-linking-His-tagged SLC26 protein bound to Ni 2ϩ -NTA beads was washed in triplicate with imidazolefree sodium phosphate buffer (pH 8.0) supplemented with 0.2% digitonin. The Ni 2ϩ -NTA beads (packed volume, ϳ15 l) were resuspended in 50 l of 0.2 M triethanolamine/HCl (pH 8.5), 0.5% digitonin. The cross-linking reaction was initiated by adding glutardialdehyde (Roth Chemicals) or dimethyl adipimidate (Pierce) from freshly prepared solutions in distilled water or 0.2 M triethanolamine/HCl (pH 8.5), respectively. After 30 min at room temperature, the cross-linking reaction was terminated by washing the Ni 2ϩ -NTA-agarose beads twice with imidazolefree sodium phosphate buffer, 0.2% digitonin. Proteins were eluted from Ni 2ϩ -NTA-agarose with non-denaturing elution buffer, consisting of 250 mM imidazole/HCl and 0.5% digitonin at pH 7.4, and then stored at 0°C until analysis later on the same day.
Purification of YFP Fusion Proteins from Mammalian Cells-tsA201 or HEK293 cell dishes were washed with cold phosphate-buffered saline and lysed with 800 l of 0.1 M sodium phosphate buffer containing 1% digitonin, protease inhibitors, and 50 mM iodoacetamide. For lysis, the whole dish was incubated 15 min on ice, followed by complete transfer of the cell lysate into a microtube and subsequent quadruplicate vortexing during incubation on ice for another 15 min. SLC26 proteins were isolated by metal affinity chromatography using Ni 2ϩ -NTA-agarose (Qiagen). In some experiments, 1% PFO in PFO extraction buffer was used for cell lysis.
SDS-PAGE, BN-PAGE, and PFO-PAGE-[ 35 S]Methioninelabeled proteins were denatured for 10 min at 56°C with SDS sample buffer containing 20 mM dithiothreitol (DTT) and electrophoresed in parallel with 14 C-labeled molecular mass markers (Rainbow, GE Healthcare Biosciences, Freiberg, Germany) on linear SDS-polyacrylamide gels. BN-PAGE was performed as described (18) immediately after protein purification. We loaded the protein purified from the equivalent of 0.5 oocytes in each lane. Molecular masses were determined by comparison with the defined membrane protein complexes generated by partial denaturing of the homopentameric ␣1 GlyR (23). PFO-PAGE was performed as described using freshly poured Tris-glycine (Laemmli) gels without SDS and Tris-glycine running buffer supplemented with 0.5% (w/v) PFO instead of SDS (24,25). The homopentameric ␣1 GlyR was used as a mass marker as above.
YFP-tagged proteins were visualized by scanning the wet PAGE gels with a fluorescence scanner (Typhoon, GE Healthcare Biosciences). PAGE gels with radioactive proteins were fixed and dried, exposed onto a phosphorscreen, and scanned with a Storm 820 PhosphorImager (GE Healthcare Biosciences). Individual bands were quantified with the Image-QuaNT software. To investigate the glycosylation state of the proteins, samples were treated for 2 h with either endoglycosidase H (EndoH) or PNGase F (New England Biolabs, Beverly, MA) in the presence of reducing SDS sample buffer and 1% (w/v) Nonidet P-40 to counteract SDS inactivation of PNGase F.
Data Analysis-Protein intensities were quantified with the ImageQuaNT software (GE Healthcare Biosciences). Each experiment was performed at least in triplicate. Data are given as mean Ϯ S.E.

His and YFP Tags Leave Function of SLC26A3 and Prestin
Unaltered-We added carboxyl-terminal hexahistidine tags to various SLC26 isoforms to purify the proteins from Xenopus oocytes by a single Ni 2ϩ -metal affinity chromatography step. Moreover, SLC26 isoforms were expressed as YFP fusion proteins in mammalian cells, to visualize protein bands using a fluorescence scanner. To test whether these sequence alterations affect functional properties, we compared chloride transport by WT and His-tagged SLC26A3, non-linear charge movement by WT and His-or YFP-tagged rat prestin, and Cl Ϫ / SO 4 2Ϫ antiport by WT and His-or YFP-tagged zebrafish prestin. SLC26A3 was initially identified as a potential tumor suppressor that was down-regulated in adenoma and abundant in normal colonic mucosa (14). It was demonstrated to be a DIDSsensitive anion exchanger (27,28). We expressed tagged and untagged SLC26A3 in Xenopus oocytes and measured radioactive chloride uptake with and without added DIDS (27). Oocytes expressing WT or His-tagged SLC26A3 accumulated chloride with similar time courses and magnitudes (Fig. 1A). In both cases, 36 Cl uptake was significantly higher in oocytes expressing SLC26A3 than in control oocytes, and incubation with 100 M DIDS caused a reduction of the chloride uptake to ϳ50%.
Rat prestin performs voltage-induced conformational changes that can be monitored as voltage-dependent capacitance changes (10,29). We expressed WT and His-or YFPtagged rat prestin in tsA201 cells and measured voltage-dependent capacitance using a software-based lock-in technique (phase tracking) in the whole cell patch clamp technique. Neither the absolute amplitude of the non-linear capacitance (data not shown) nor the voltage dependence of rat prestin (Fig. 1B)  FEBRUARY 15, 2008 • VOLUME 283 • NUMBER 7 was affected by any of the added tags. Rat prestin charge movement is not affected by 100 M DIDS (data not shown). However, non-linear capacitances by WT and tagged prestin were eliminated by salicylate (29) (Fig. 1B).

Subunit Stoichiometry of SLC26 Multifunctional Anion Exchangers
Zebrafish prestin is functionally different from rat prestin. In addition to mediating voltage-dependent charge movements (30), it functions as Cl Ϫ /SO 4 2Ϫ exchanger with 1:1 stoichiometry (31). We measured currents in cells expressing WT or YFP or His fusion proteins of zebrafish prestin. The three tested proteins displayed anion currents with identical reversal potential and therefore unchanged transport stoichiometry (Fig. 1C). Mean current amplitudes were comparable (data not shown), and 10 mM salicylate (31) reduced current amplitudes to ϳ50% in all cases (Fig. 1C). The addition of the His tag or YFP tag did not affect the function of the tested SLC26 isoforms. We conclude that none of the tags prevents oligomerization processes of SLC26 that are of functional importance.

SDS-PAGE Analysis of Mammalian SLC26 Isoforms Heterologously Expressed in Xenopus Oocytes-Expression of SLC26
isoforms in Xenopus oocytes resulted in protein amounts that are below the resolution limit of Coomassie or silver staining. We therefore used metabolic labeling with [ 35 S]methionine for signal enhancement to visualize analytical quantities of recombinant proteins.
His-tagged SLC26A3 and rat prestin were expressed and metabolically labeled in Xenopus oocytes. After extraction with 1% (w/v) digitonin, proteins were purified by metal affinity chromatography. Both proteins express well in Xenopus oocytes and are metabolically stable during a sustained chase. When resolved by reducing SDS-PAGE with a low percentage of acrylamide, as in Fig. 2A, the SLC26A3 and prestin polypeptides migrated at 15-20% lower masses than calculated from the amino acid sequences (85 kDa for His-SLC26A3 and 82 kDa for His-prestin). Both polypeptides shifted to higher apparent molecular masses at higher acrylamide concentrations (results not shown). This indicates that the mass deviations are due to anomalous migration in SDS-PAGE gels and not to post-translational processing of the polypeptide chains. Anomalous fast electrophoretic mobility has frequently been observed with highly hydrophobic membrane proteins (32).
Human SLC26A3 harbors four sequons for N-glycosylation, 153 NAT, 161 NNS, 164 NNS, and 165 NSS, in the predicted second ectodomain loop, which is flanked by the transmembrane segments TM3 and TM4. Deglycosylation with either EndoH or PNGase F reduced the molecular mass by ϳ8 kDa (Fig. 2B, lanes 1-3), suggesting the presence of three N-linked high mannose-type oligosaccharides of 2-3 kDa each. Substitution of Asn 153 by glutamine resulted in a di-glycosylated polypeptide (lanes 4 -6), and additional glutamine substitution of Asn 161 , Asn 164 , and Asn 165 resulted in a non-glycosylated polypeptide (lanes 10 -12). Thus, the three sequons that are in close proximity to one another acquire only two oligosaccharides, indicating that either Asn 164 or Asn 165 is sterically inaccessible for the oligosaccharyltransferase (32,33). Rat prestin carries five potential glycosylation sites, with two adjacent sequons Ϫ 163 NAT and 166 NGT Ϫ located in the second ectodomain. Deglycosylation produced a ϳ5-kDa mass shift (Fig. 2C, lanes 2-3), suggesting that two sites carry N-glycans. Glutamine substitution of Asn 163 led to a monoglycosylated polypeptide (lanes 4 -6), and no mass shift by PNGase F was observed when both sites, Asn 163 and Asn 166 , were replaced by glutamines (lanes 7-9). The three remaining N-glycosylation sequons, 589 NATV, 603 NATK, and 736 NATP, were not used.
SLC26A3 and Rat Prestin Migrate as Discrete Dimers in BN-PAGE Gels-To examine whether SLC26 transporters assemble into stable higher order structures, we natively purified human SLC26A3 or rat prestin from Xenopus oocytes and resolved it on BN-PAGE gels. The non-denatured hSLC26A3 exchanger migrated as a distinct band of ϳ200 kDa (Fig. 3A, lane 1). The size of the native SLC26A3 oligomer was assessed by comparing it to a molecular weight standard produced by partial dissociation of the homopentameric ␣1 GlyR receptor (Fig. 3A, lane 8). This procedure results in the occurrence of oligomers consisting of one to five GlyR ␣1 subunits (molecular mass ϳ52 kDa), representing a reliable size standard for membrane proteins of up to 260 kDa (32). To display the number of subunits incorporated in the SLC26A3 transporter complex, we weakened non-covalent subunit interactions by treating the natively purified protein with increasing concentrations of SDS. Incubation of the SLC26A3 exchanger with 0.01% SDS resulted in the complete disappearance of the 200-kDa band in favor of one faster migrating band of ϳ100 kDa (Fig. 3A, lane 4). The similarity with the calculated mass (91 kDa including the 6-kDa mass contributed by two N-linked glycans) and the absence of a faster migrating band in the presence of a 10-fold higher SDS concentration (Fig. 3A, lane 6) indicate that the 100-kDa band represents the monomeric state of SLC26A3, and the 200-kDa band represents the dimeric state. In the absence of the denaturant SDS, the 200-kDa band was resistant to 1-h incubation at 37°C with Coomassie dye (Fig. 3A, lane 2) or Coomassie dye plus the strong reductant DTT (Fig. 3A, lane  3). Non-glycosylated SLC26A3 transporters also existed exclusively as dimers (results not shown). Other multimerization states were virtually absent (Fig. 3). Natively purified rat prestin migrated at three positions equal to ϳ97, ϳ190 kDa, and ϳ250 kDa in BN-PAGE gels (Fig. 3B, lane 3). Denaturing by SDS resulted in a complete dissociation of the oligomeric prestin into the ϳ90-kDa form (Fig. 3B, lanes 5 and 6), closely similar to the calculated mass of the rat prestin monomer of 87 kDa (protein core of 82 kDa including His tag plus 5 kDa of N-linked carbohydrate). No faster migrating protein species was observed, indicating that the 90-kDa band represents the monomer and the ϳ180-kDa band the dimer of rat prestin.
The ϳ250-kDa band (indicated by a star) was consistently observed when non-denatured prestin was resolved in BN-PAGE gels. Its intensity was about 10 -30% of that of the ϳ190-kDa band, as determined by quantitative PhosphorImager scanning. The non-glycosylated 163 Q 166 Q-prestin mutant migrated only at two positions in the BN-PAGE gel corresponding to ϳ90 kDa and ϳ190 kDa (Figs. 3B,  lane 7). This suggests that the ϳ250-kDa band is related to the presence of N-linked glycans, and that both the ϳ190-kDa band and the ϳ250-kDa band represent dimeric states. Quantification of the relative intensity of the different bands revealed that dimers represented 52% of the total prestin in this experiment.
To rule out the possibility that the lowest molecular band corresponds to an unusually stable dimer and the higher molecular mass band to a tetrameric conformation, we engineered a concatenated rat prestin construct by linking two coding regions in a single open reading frame. After expression in Xenopus oocytes, two major bands were observed in BN-PAGE (Fig. 3B,  compare lanes 9 and 3), corresponding to (prestin-prestin) and (prestin-prestin) 2 . Both conformations are stable and occur with comparable probability, further corroborating the dimeric subunit stoichiometry of prestin. Under denaturing conditions, the ϳ190-kDa band of the covalently linked dimer became more prominent (Fig. 3B, lane 10), obviously due to the dissolution of the dimerized tandem dimer and higher mass aggregates. In addition, a faint band appeared that migrated at the same position as the prestin monomer (Fig. 3B,  indicated by a cross in lane 10). This monomeric form arises most likely from a proteolytic cleavage in the linker region of the concatenated prestin dimer, as has been observed previously with all other concatamers studies in our laboratories (17,32). Reducing SDS-PAGE resolved the full-length prestin concatamer as a 140-kDa polypeptide corresponding approximately to twice the apparent 63 kDa of the prestin monomer (Fig. 3C). In addition, proteolysis-derived monomers are visible as weak bands (indicated by a cross).
Our conclusions are in conflict with a recent report that postulated that gerbil prestin exists as a tetramer formed by association of two disulfide-linked dimers (34). To identify possible reasons for these differences in experimental results, we performed three experiments. We tested the effect of reducing  FEBRUARY 15, 2008 • VOLUME 283 • NUMBER 7 agents on the oligomeric state for all studied SLC26 isoforms (Figs. 3D, 5C, and 5G). Next, we demonstrated that BN-PAGE correctly displays the dimeric structure of ClC channels, a protein family firmly established to form dimers (Fig. 3E). Lastly, we repeated some experiments on rat prestin using PFO as detergent (34) (see Fig. 7).

Subunit Stoichiometry of SLC26 Multifunctional Anion Exchangers
Prestin Multimers Are Not Covalently Linked by Disulfide Bonds-Recombinant rat prestin migrated entirely as monomers in BN-PAGE gels after denaturing with low concentrations of SDS in the absence of added reductant (Fig. 3B, lanes  4 -6). Comparative resolution of non-reduced and DTT-reduced prestin on SDS-PAGE gels did not provide evidence for disulfide-linked multimers (Fig. 3D, lanes 3 and 4). Similar results were obtained with SLC26A3 (Fig. 3D, lanes 1 and 2) and zebrafish prestin (see Fig. 5C). We conclude that SLC26 dimers are held together by non-covalent forces rather than by covalent disulfide bonds.
BN-PAGE Displays the Dimeric Structure of the Human Muscle Chloride Channel hClC-1-The BN-PAGE technique has been shown to correctly display the quaternary state of various ion channels and transporters, including the pentameric structure of Cys-loop receptors (23,35,36) and the trimeric structure of glutamate transporters (32,37,38) and P2X receptors (18). To validate this experimental approach for a dimeric membrane protein, we expressed a member of the ClC family of voltage-gated chloride channels, the human muscle chloride channel (hClC-1), in Xenopus laevis oocytes (39,40).
Non-denatured recombinant hClC-1 channels purified by metal affinity chromatography from [ 35 S]methionine-labeled oocytes migrated as a single sharp band in the BN-PAGE gel (Fig. 3E, lane 1) at the same position as the non-denatured 260 kDa ␣1 GlyR. SDS led to a concentration-dependent appearance of a faster-migrating protein band of ϳ130 kDa (Fig. 3E,  lane 2), closely similar to the calculated mass of the hClC-1 monomer (including a single 3-kDa oligosaccharide side chain at Asn 430 (41)). BN-PAGE is therefore able to display correctly the quaternary structure of a dimeric membrane protein. The identity of the oligomeric states of both classes of proteins can be illustrated by directly comparing hClC-1 and SLC26A3 on the same BN-PAGE gel (Fig. 3E).
Cross-linking Generates Covalently Linked SLC26 Dimers-Next we used chemical cross-linking as an alternative approach to determine the oligomeric state of SLC26 proteins. Glutardialdehyde efficiently cross-linked natively purified rat prestin to dimers (Fig. 4A). Quantification of the protein bands showed that 66 -71% of prestin migrated as covalently linked dimers in the reducing SDS-PAGE gel when incubated at Ն1 mM glutardialdehyde. The non-glycosylated prestin mutant was crosslinked to dimers with the same effectiveness as WT prestin (results not shown). Cross-linking by glutardialdehyde increased the fraction of dimeric proteins from 60% (no cross-linker, Fig. 4B, lane 1) to up to 80% in BN-PAGE gel (10 mM glutardialdehyde, Fig. 4B,  lane 8). Thus, a certain fraction of non-covalently linked prestin dimers dissociates into monomers during overnight BN-PAGE electrophoreses. The imidoester dimethyl adipimidate, which has a longer spacer arm than glutardialdehyde (8.6 Å versus ϳ5 Å), was much less efficient in generating covalently linked dimers, as visualized by reducing SDS-PAGE (Fig. 4A, lanes  6 -8) and BN-PAGE (Fig. 4B, lanes 10 -13). SLC26A3 (Fig. 4C, lanes 1 and 2) could also be cross-linked to dimers. Glutardialdehyde cross-linking did not affect the migration of the SLC26 dimers in the BN-PAGE gel (Fig. 4D). However, cross-linking completely prevented the SDS-induced dissociation of the dimers into monomers (compare lanes 2 and  4). The co-analyzed concatenated rat prestin dimer migrated at exactly the same position as the glutardialdehyde-cross-linked prestin on the BN-PAGE gel (lanes 9 and 10). No adducts larger than dimers were observed for any tested SLC26 isoform.
Non-mammalian Eukaryotic and Prokaryotic SLC26 Paralogs Are Also Assembled as Dimers-The SLC26 family contains paralogs in mammals, non-mammals, plants, fungi, and bacteria (2). To assess whether the oligomeric state is evolutionarily conserved in non-mammalian eukaryotes and prokaryotes, we studied the subunit stoichiometry of zebrafish (zf) prestin (15) and a bacterial isoform from P. aeruginosa (PASulP).
SLC26 paralogs were expressed as His fusion proteins in Xenopus oocytes and metabolically labeled. The zebrafish paralog acquired EndoH-resistant complex-type carbohydrates during a sustained pulse (Fig. 5A). Complex-glycosylation resulted in an ϳ25 kDa increase in the apparent molecular mass (indicated by a star). Elimination of two putative N-glycosylation sequons, 161 NGT and 164 NSS, by glutamine substitution of Asn 161 and Asn 164 prevented N-glycosylation and thus complex glycosylation (Fig. 5B). Comparative analysis under nonreducing and reducing conditions provided no evidence for the existence of disulfide-bonded zebrafish prestin oligomers (Fig. 5C).
In BN-PAGE, the non-denatured zebrafish prestin migrated predominantly as a dimer (ϳ240 kDa) (Fig. 5D, lane 1) that dissociated into the monomeric ϳ120 kDa form when treated with SDS (Fig. 5D, lanes 2 and 3). The existence of dimeric prestin was further verified by cross-linking with glutardialdehyde (Fig. 5E). In contrast to the small amount of monomeric prestin visible in the BN-PAGE gel (Fig. 5F, lane 1), no such monomer could be detected in the cross-linked sample (Fig.  5F, lane 2). We conclude that zebrafish prestin exists as a homogenous population of dimers when expressed in Xenopus oocytes.
A prokaryotic SLC26 is also assembled as dimer (Fig. 5G). After expression in Xenopus oocytes, PASulP migrated as one distinct band of ϳ160 kDa in its non-denatured form (Fig. 5G,  lane 1). Incubation for 1 h at 37°C in the presence of Coomassie dye led to the additional appearance of a ϳ82-kDa band (Fig.  5G, lane 2). Denaturing with SDS resulted in one distinct band of ϳ62 kDa, close to the calculated molecular mass of 64 kDa of the PASulP monomer. We conclude that the 160-and 82-kDa bands represent the PASulP dimer and monomer, respectively, and that SDS increases the mobility of the PASulP monomer in the BN-PAGE gel.
Subunit Stoichiometry of SLC26 Is Not Affected by the Expression System-To test whether oligomerization of eukaryotic SLC26s might be affected by the cellular environment, we resolved rat prestin, zebrafish prestin, and SLC26A3 by BN-PAGE after heterologous expression as YFP fusion protein in mammalian cells. The expressed proteins were extracted with 1% (w/v) digitonin and subjected to BN-PAGE either by application of the full lysate (rat and zebrafish prestin) or after purification by metal affinity chromatography (His-SLC26A3) (Fig.  6). In all cases, YFP-tagged SLC26 transporters migrated predominantly as bands of ϳ260 kDa (Fig. 6, lanes 3, 6, and 9) in BN-PAGE gels. The carboxyl-terminally GFP-tagged rP2X 1 receptor served as a mass marker (Fig. 6, lanes 1 and 2). Incubation in SDS (Fig. 6, lanes 4 and 5, 7 and 8, and 10 and 11) led to a disappearance of the ϳ260-kDa SLC26 proteins and the appearance of one band with about half of the molecular mass, in agreement with a dimeric structure also in mammalian cells.  (34). To assess the consequences of this difference, we purified WT and glycosylation-deficient mutant rat prestin from oocytes using digitonin and PFO as detergents. When resolved by PFO-PAGE, WT, and glycosylation-deficient rat prestin migrated entirely as monomers, irrespective of the detergent used for purification (Fig. 7A, lanes 2 and  3, and 8 and 9). On PFO-PAGE gels, prestin dimers were only observed when digitonin-purified prestin was cross-linked with glutardialdehyde before application to the gel (Fig.  7A, lanes 6 and 7). Hence, the dimeric state of prestin is preserved during purification with digitonin, but lost during PFO-PAGE unless stabilized by cross-linking. PFO not only dissociates prestin oligomers, but also GlyR pentamers (Fig. 7A,  lane 1). The appearance of GlyR monomers in the non-denatured GlyR sample that was co-analyzed as a mass marker also indicates a significantly stronger denaturing effect of PFO than of digitonin. On BN-PAGE gels, GlyR behaves as a very stably assembled homopentamer; monomers are only observed after denaturation with SDS (22). We next extracted a YFP fusion protein of rat prestin from stably transfected mammalian cells using digitonin or PFO. Digitonin-extracted YFP-rprestin migrated in  the PFO-PAGE gel exclusively in the monomeric state (Fig. 7B,  lane 2). Two proteins bands were resolved by PFO-PAGE and could be identified by deglycosylation analysis as the predominant complex-glycosylated (Endo H-resistant) form of prestin and a less abundant (Endo H-sensitive) core-glycosylated form (Fig. 7D). The dimeric state that is predominant on BN-PAGE gels (Fig. 6) is not present under these conditions. PFOextracted YFP-prestin migrated neither in BN-PAGE gels (data not shown) nor in PFO-PAGE gels (Fig. 7B, lane 1) as oligomer. Neither the addition of DTT nor of iodoacetamide resulted in the appearance of protein bands with lower molecular weight, verifying the absence of disulfide-bonded YFPrprestin oligomers (Fig. 7C). We conclude that PFO does not conserve the oligomeric state of rat prestin.

PFO-PAGE Does Not Display a Multimeric Structure of Rat Prestin-Previous
Functional Interaction between Prestin Subunits-In a dimeric assembly, each subunit can be functionally independent, or alternatively, the two neighboring subunits can cooperate in performing a certain protein function. To study whether the two subunits of the motor protein prestin interact functionally, we co-expressed two mutant proteins Ϫ D154N and D342Q (29). In agreement with an earlier report (29), D154N shifts the voltage dependence of the non-linear capacitance to more negative voltages and D342Q shifts the voltage depend-ence to more positive ones (Fig. 8A). The voltage dependence of the non-linear capacitance of prestin can be fit with a derivative of a Boltzmann equation, providing values for the potential at which the half-maximal charge transfer has occurred (V1 ⁄ 2 ) and the number of elementary charges displaced across a fraction of the membrane during the voltage-dependent conformational change (Table 1).
In case of a parallel action of the two subunits without functional interactions, the voltage dependence of the non-linear capacitance of heterodimeric prestin will be the addition of the respective voltage dependences of each homo-dimeric transporter. In contrast, any cooperative interaction during the voltage-dependent conformational change will cause a deviation of the voltage dependence of the heterodimeric motor protein from this superposition. Fig. 8B shows the simulated voltage dependences of the non-linear capacitance (C nonlin -V) curves, assuming no interaction between D154N and D342Q prestin in a heterodimeric motor protein. All the simulated C nonlin -V curves showed two clearly separated maxima with amplitudes that correspond to the relative amplitude of the individual components.
Non-linear capacitances from co-transfected cells are different from the simple sum of the capacitances generated by the two mutant motor proteins (Fig. 8, C-E). The C nonlin -V curves exhibit a single peak that is significantly broader than the ones observed in cells expressing only a single type of prestin. Heterodimeric motor proteins thus undergo conformational changes whose voltage dependence is jointly determined by both subunits. Such a behavior can only be found in case of a functional interaction of the two subunits. Conformational changes of one prestin molecule are modified by the contralateral subunit in the dimeric motor protein.
This functional assay did not only provide evidence for a functional interaction between subunits but also further support for a dimeric structure of prestin. For a dimeric protein, cells co-transfected with two plasmids are expected to express three populations, homodimeric D154N, homodimeric D342Q, and heterodimeric motor proteins consisting of one D154N and one D342Q subunit. We thus fitted the experimentally observed voltage dependences of non-linear capacitances with the sum of three first derivatives of a two-state Boltzmann function. For the two of the three that corresponded to the homodimeric populations, V o and z␦ were fixed to the values obtained from cells only transfected with one mutant, and only the relative amplitude Q max was adjusted as a fit parameter. For the third term, V o , z␦, and Q max were optimized as fit parameters. For all tested cells, the values for V o and z␦ of the third fitted component were very similar (Table 1). Such behavior is in full agreement with a dimeric assembly. It is inconsistent with a tetramer, for which the third capacitance component would represent three hetero-multimeric components, whose relative contribution would depend on the relative number of the two mutant subunits in the oligomer.

DISCUSSION
SLC26 is a gene family encoding anion transporters, anion channels, and membrane-bound motor proteins. The first evidence for SLC26 multimerization came from fluorescence res-  onance energy transfer experiments that demonstrated a close proximity between fluorescently tagged carboxyl termini of two prestin subunits (42)(43)(44). The number of subunits necessary to form a functional motor protein was first addressed by Zheng et al. (34). In that study, native and recombinant prestin were shown to form stable oligomers that resisted dissociation into monomers by SDS. Chemical cross-linking and PFO-PAGE combined with immunoblotting and affinity purification suggested a tetrameric subunit stoichiometry of prestin. SDS dissociated the tetramer into dimers that could be converted to monomers by hydrophobic reducing agents but not by the hydrophilic ones. The authors concluded that prestin monomers are covalently linked to dimers by disulfide bonds located in the hydrophobic membrane core and that these covalently linked dimers associate to form a tetramer.
Our results are in contrast to those published findings. BN-PAGE and chemical cross-linking demonstrated that neither prestin nor any other tested SLC26 isoform was linked by disulfide bonds, and all SLC26s formed dimers as predominant oligomeric state. Oligomers dissociated entirely into monomers under non-reducing conditions in the presence of low concentrations of SDS. The use of a tandem concatamer as a mass marker corroborated the correct assignment of BN-PAGE bands to certain oligomeric states of rat prestin. Not only all tested eukaryotic, but also a prokaryotic SLC26 exhibits a dimeric subunit stoichiometry. After expression in Xenopus oocytes, a paralog from P. aeruginosa migrates exclusively as dimer in BN-PAGE.
Because our experiments were performed with rat and zebrafish prestin and those of the other study with gerbil SLC26A5/prestin (34), these differences might be due to isoform-specific locations of cysteine residues. This is not the case: all cysteines are fully conserved in human, rat, mouse, and gerbil prestin, but not in the zebrafish isoform (supplemental Fig.  S2). It appears furthermore possible that SLC26s exhibit both oligomerization states, dimers as well as tetramers, as reported for a member of another anion transporter family (SLC4) (45). However, the complete absence of tetramers in all our assays argues against this possibility.
We can thus only speculate about the reasons for the different results of our group and those by Zheng et al. (34). We solubilized prestin with digitonin, whereas Zheng et al. used PFO (34). Digitonin has a strong record as an efficient and gentle detergent for the extraction of membrane protein complexes prone to dissociation by treatment with other detergents (46,47). The combination of digitonin solubilization and BN-PAGE has been demonstrated to faithfully display the subunit stoichiometry of various membrane proteins, such as the mitochondrial protein import machinery, oxidative phosphorylation supercomplexes (47), various ligand-gated ion channels (18,32,35), transporters (32,48), and membrane-bound enzymes (49). In contrast, PFO-PAGE dissociates membranebound enzymes (50), ion channels (ionotropic ␥-aminobutyric acid (24) and glycine receptors (Fig. 7)), as well as transporters (a bacterial glutamate transporter 6 ). PFO also dissociates pres-tin (Fig. 7, A and B). To explain the occurrence of oligomers resistant to both SDS and DTT (see Zheng et al. Fig. 6, A and B), we suggest that prestin may have aggregated during or after protein solubilization to form random oligomers. Alternatively, intersubunit disulfide bond formation could have occurred during purification in those experiments in which iodoacetamide was not added. Even small amounts of such non-physiological higher order oligomers can be detected by the highly sensitive Western blot technique (34). In contrast, we used metabolic labeling or labeling with fluorescent proteins, both permitting a quantification of the relative amounts of several protein fractions.
The difference in detection might also explain the dissimilar outcome of the chemical cross-linking. In the previous study, dimers were the predominant state, although trimers and tetramers were observed (34). Cross-linking combined with Western blotting provides semi-quantitative data about the abundance of the distinct oligomeric states. Moreover, at high membrane densities, which are likely to occur in the mammalian cochlea, unspecific cross-linking can occur between neighboring protein complexes.
A limitation of BN-PAGE analysis in determining the native oligomeric state of proteins is that solubilization of the protein might dissociate weak protein-protein interactions. It thus appears possible that SLC26s form stable dimers that associate to higher order oligomers in the native membrane, i.e. tetramers that are dissociated during solubilization or gel electrophoresis. The results of our cross-linking experiments (Figs. 4, 5E, and 7A) and the functional analysis of cells co-transfected with two mutant prestin proteins with different voltage dependences of the non-linear capacitance (Fig. 8) argue against this possibility. Co-transfection resulted in the formation of heterooligomers with distinct voltage dependences. The voltage dependence of the non-linear capacitance of co-transfected cells could be well fit assuming only a single population of hetero-oligomers without evidence for a cooperative interaction of more than two subunits. These data demonstrate that a dimeric assembly is the functional unit of prestin.
Heterodimeric D154N-D342Q prestin exhibits voltage-dependent capacitances that are different from D154N and D342Q homodimeric proteins. The two mutations thus do not exert an additive effect, as expected in case of two independently functioning subunits. The two subunits rather jointly determine the voltage dependence of conformational changes of the dimeric protein. This inter-subunit interaction explains the recent finding of a missense mutation in the pres gene found in a heterozygous hearing-impaired patient (51). The mutation results in a shift of the voltage dependence of non-linear capacitance. It can result in a dominant inheritance mode, because heterodimeric prestin consisting of WT and mutant prestin exhibits an intermediate phenotype. At present, it is unknown if inter-subunit interactions occur in SLC26 transporters and channels. For all inherited diseases linked to these SLC26 genes, the inheritance mode is recessive (3).
The interaction between prestin subunits is weak. BN-PAGE of rat prestin reveals a significant percentage of monomeric proteins (Fig. 3), and cross-linking increases the percentage of dimers to only ϳ80% (Fig. 4). Hence, a certain fraction of mono-mers exists in intracellular compartments or in the surface membrane.
Many known transporter families have been shown to form oligomeric proteins. Members of the SLC1 family are assembled as trimers (32,38), whereas SLC4, SLC6, and ClC transporters form dimers (40,52,53). In oligomeric transporters two different forms of intersubunit interactions are possible. The subunits could jointly contribute to a central carrier or motor protein domain, or each subunit might be capable of functioning by itself. SLC1 (38), SLC4 (52), and ClC (40) transporters exhibit multibarreled structures with two or three carrier domains each formed by a single subunit. The ClC family contains channels and transporters, and for ClC channels cooperative gating steps have been reported (54). In SLC1 transporters, each subunit appears to function independently of each other (55). Because prestin dimers are not very stable (Fig. 3) and because the inheritance mode of SLC26-linked diseases is in general recessive, the formation of a common carrier or motor protein domain appears unlikely. A likely explanation for all our results is that two SLC26 subunits work in parallel but interact allosterically.
Two different transmembrane topology models of prestin are currently considered. Both place the amino and the carboxyl termini at the intracellular membrane side, but differ in the number of transmembrane helices. Although Zheng and colleagues postulated 12 transmembrane helices (56), a revised topology model contains only 10 transmembrane domains (42). We here confirm that rat prestin is a glycoprotein with two N-glycans at Asn 163 and Asn 166 (26) and demonstrate that zebrafish prestin and human SLC26A3 carry two N-glycans at corresponding localizations. SLC26A3 harbors one additional N-glycan in position Asn 153 , which is also located in the predicted second extracellular loop. Three N-glycosylation sequons in the rat prestin sequence remain unused: 589 NATV, 603 NATK, and 736 NATP. The non-usage of 736 NATP does not provide topological information, because a proline in the ϩ4 position is known to prevent N-glycosylation. In contrast, the lack of glycosylation of 589 NATV and 603 NATK supports the cytosolic localization of these residues. The glycosylation states of various SLC26s strongly support the topology model with 12 transmembrane domains.
We conclude that a dimeric subunit stoichiometry is general to prokaryotic and eukaryotic SLC26 isoforms. Three different approaches demonstrated a dimeric subunit stoichiometry for one mammalian anion exchanger, one mammalian motor protein, one zebrafish prestin homologue, and one prokaryotic paralog. Isoform-specific functional differences therefore do not originate from differences in the quaternary structure.