Aminosulfonate Modulated pH-induced Conformational Changes in Connexin26 Hemichannels*

Gap junction channels regulate cell-cell communication by passing metabolites, ions, and signaling molecules. Gap junction channel closure in cells by acidification is well documented; however, it is unknown whether acidification affects connexins or modulating proteins or compounds that in turn act on connexins. Protonated aminosulfonates directly inhibit connexin channel activity in an isoform-specific manner as shown in previously published studies. High-resolution atomic force microscopy of force-dissected connexin26 gap junctions revealed that in HEPES buffer, the pore was closed at pH < 6.5 and opened reversibly by increasing the pH to 7.6. This pH effect was not observed in non-aminosulfonate buffers. Increasing the protonated HEPES concentration did not close the pore, indicating that a saturation of the binding sites occurs at 10 mm HEPES. Analysis of the extracellular surface topographs reveals that the pore diameter increases gradually with pH. The outer connexon diameter remains unchanged, and there is a ∼6.5° rotation in connexon lobes. These observations suggest that the underlying mechanism closing the pore is different from an observed Ca2+-induced closure.

the effects of these modulators, there is not much information at the structural level as to the conformational changes that occur in closing the pore in response to these stimuli.
Each connexin (Cx) channel is composed of two hexamers (connexons) that dock at their apposed extracellular surfaces. The cyclic arrangement of the subunits within the hexamers suggests that gating can occur by a rotation and translation of the transmembrane segments within all six monomers. It has been postulated that gating occurs as a "camera iris" shutter (2). An alternate hypothesis has been proposed in which intra-connexin associations occur to produce either a particle-receptor blockage at the cytoplasmic surface (3,4) or as a physical gate near the extracellular surface ("loop gate") (5). Whether these proposed mechanisms correlate to the closure of fast and/or slow gates that have been characterized by electrophysiological methods (see Ref. 6) remain to be determined.
Gating by intracellular acidification is one way that connexin channels open and close in response to stimuli. Experimentally determined decreases in intracellular pH are known to decrease junctional electrical coupling in cardiomyocytes and in Purkinje fibers (7)(8)(9)(10) as well as in teleost and amphibian embryos (11). Stergiopoulos et al. (12) showed that many, but not all, connexins close in a pH-sensitive manner when tested in the paired Xenopus oocyte system. For example, Cx26 channels are pH-sensitive, but Cx32 channels are much less sensitive. Homomeric hemichannels also displayed this pH regulation (13). Differences in the pH regulation of gap junctions were attributable to the diversity of the primary sequence, particularly in certain regions such as the C-terminal tail because the pH sensitivity of the dodecamer channels could be modified only when they were composed of heterotypic combinations (13). However, it is important to note that these experiments were done in whole cells and cannot distinguish between gating of the channel because of protonation of the connexin or protonation of modulators or ligands that bind to the connexin and then close the channel.
Bevans et al. (14) used another assay system to test for functional pore size called the transport-specific fractionation method (TSF). Heteromeric Cx32/Cx26 connexons reconstituted into liposomes showed pH-dependent channel activity as measured by a permeability assay when suspended in aminosulfonate buffers (e.g. HEPES, TAPS, MES). One of the simplest of these aminosulfonate compounds is taurine, a naturally occurring ubiquitous cytoplasmic component (15,16). This pH sensitivity was directly attributed to binding of protonated aminosulfonates to Cx26, because homomeric Cx32 channels did not show this pH sensitivity (17,18). However, it should be noted that Cx46 hemichannels in excised patches appear to display pH sensitivity in the absence of any added cytosolic material (19).
Previously, we had shown using atomic force microscopy (AFM) that force-dissected connexin26 (Cx26) gap junction hemichannels reversibly open and close in response to Ca 2ϩ acting as ligand (20). These conformational changes could be only observed on extracellular hemichannel surfaces because the cytoplasmic GJC surface appears too flexible to be imaged at sufficiently high resolution to assign structural changes. This flexibility of the cytoplasmic surface is well documented not only by AFM studies (20 -22) but also by electron microscopy and by other structural and biochemical methods. (For a complete review of flexibility in the cytoplasmic domains see Ref. 23.) In addition to our previous work, we have been able to image the extracellular surface at higher spatial resolution, allowing insight into the tertiary conformations of polypeptide loops connecting the transmembrane ␣-helices lining the connexon pore. Remarkably, gating events are well visualized at the extracellular surface. Taking the advantage of AFM to work under freely adjustable physiological conditions, we directly observed the reaction of hemichannels to pH changes of the buffer solution. All experiments were performed at ambient temperatures with fully hydrated proteins under buffer conditions. This simplistic system allows for the in vitro correlation of adding modulating agents and the resulting conformational changes directly imaged with AFM. The high-resolution AFM topographs suggest that the extracellular connexin domains undergo an aminosulfonate-modulated conformational change that closes the connexon channel at the extracellular surface. This conformational change was fully reversible. From these high-resolution AFM topographs, we propose that the mechanism for closure of the extracellular "loop" gate is different from the Ca 2ϩ -induced closure observed previously.

EXPERIMENTAL PROCEDURES
Cx26 Gap Junction Preparation-Gap junctions were purified from overexpressing stably transfected HeLa cells grown to confluency by following a previously published procedure (24) with one modification. For Cx26 gap junctions, HEPES buffer was used instead of Tris buffer because preparations contained more gap junctions and less contaminating material as judged by conventional negative stain transmission electron microscopy.
AFM Imaging-AFM topographs were recorded in buffer solution using contact mode. The AFM (Nanoscope III, di-Veeco) was equipped with a fluid cell and oxide-sharpened Si 3 N 4 cantilevers (OMCL TR400PSA, Olympus, Japan), which had a nominal spring constant of Ϸ0.09 N/m. Prior to imaging, gap junction membranes were adsorbed to freshly cleaved mica as described (20,25). High-resolution topographs were recorded at minimal contact forces of Յ50 pN, which were manually adjusted to compensate for thermal drift (20,26). Proportional and integral gains were adjusted manually to minimize the error (deflection) signal and to maximize the height signal (27). When approaching a lateral resolution of Ϸ1 nm the scanning speed of the AFM tip was between 500 and 1.500 nm/s. Only topographs showing identical structural features scanning the same sample area in trace and retrace direction were selected for further analysis. Topographs showing asymmetric particles or indicating any kind of tip artifacts were not analyzed.
Image Processing and Averaging-Topographs (512 ϫ 512 pixel) were selected by the structural details of the protein imaged reproducibly and by comparing the simultaneously monitored height profiles acquired in trace and retrace direction. Correlation averaging was performed using the SEMPER image processing system (28). A well preserved unit cell was selected from the raw data and cross-correlated with the topograph (29). Unit cells were extracted according to the peak coordinates of the cross-correlated topograph. Single particle averages were generated by translationally and rotationally aligning the unit cells to a reference connexon and then averaged. This correlation average was used as reference for refinement cycles (30). Correlation-averaged unit cells were 6-fold symmetrized. To assess the standard deviation k,l , individual unit cells were extracted according to the coordinates of their correlation peaks and were aligned angularly as well as translationally before single particle averaging (31). The S.D. was then calculated from the averaged topograph k,l for each pixel (k, l) for x i particles (32) as shown in Equation 1.
These S.D. maps are displayed as an image in a one-to-one pixel correspondence with the correlation-averaged topograph. The values range from 0.1 (black) to 0.4 (white) nm with the color table continuously ranging from black to white.

RESULTS
Previously, we had shown that high-resolution AFM topographs of the extracellular surface could be obtained reproducibly. We also imaged the cytoplasmic surface; but because of the enhanced flexibility of the cytoplasmic domains (23), topographs could be obtained only at very low resolution, which would not allow the revealing of delicate conformational changes such as those observed on the relatively rigid extracellular surface of the hemichannel (20).
High-Resolution Imaging of the Extracellular Connexon Surface-Force dissection with AFM imaging provides highresolution surface views of the extracellular surface. Fig. 1A shows a gap junction plaque imaged by AFM in buffer solution. As reported previously, Cx26 gap junctions exhibited a thickness of 17.5 Ϯ 0.8 nm (n ϭ 20), while the surrounding lipid membranes were only 4.5 Ϯ 0.6-nm high (20). These gap junction plaques exposed their cytoplasmic surfaces to the AFM tip, while the extracellular surfaces were sandwiched between the membranes embedding the bridging connexons. To characterize the extracellular surface, the upper connexon layer of the plaque had to be mechanically removed by the scanning AFM tip (Fig. 1B). At minimal forces of Յ50 pN applied to the scanning AFM cantilever, the repetitive imaging is non-destructive and allows reproducible observation of substructures of pro-

pH-induced Structural Changes in Cx26 Hemichannels
teins embedded in biological membranes in their native conformation (26). However, at about 10 -20-fold increased forces, the scanning process of the AFM allowed mechanical removal of the upper layer of the gap junction plaque (20,33,34).
After removal of the upper layer, the exposed extracellular surface could be observed at high resolution (Fig. 1C). The unprocessed AFM topograph shows the 6-fold symmetry of each connexon with its characteristic central pore. Each subunit of the connexon is thought to be formed by one connexin showing substructural details (23). The correlation average of the connexon surface (Fig. 1, D and E) showed their common structural details, whereas the S.D. map (Fig. 1, F and G) marked regions exhibiting an enhanced structural flexibility.
Observing pH-induced Conformational Changes-Visualization of pH-induced conformational changes of the extracellular connexon surface was first performed under constant electrolyte concentration (2 mM EGTA, 1 mM phenylmethylsulfonyl fluoride) and at a constant HEPES concentration of 10 mM (Fig.  2). Prior to imaging, the membranes were absorbed onto the mica support using buffers containing 10 mM HEPES, 2 mM EGTA, 200 mM NaCl, and 1 mM phenylmethylsulfonyl fluoride. Connexons observed at pH 6.0 appear very different to those observed at pH 7.0 and higher. However, it could appear that single connexons show individual deviations among each other. For example, a small number of connexons (Ͻ5%) showed different channel diameters than others. To allow statistical relevant conclusions about the average structural conformation of a single connexon under a certain buffer condition, we calculated single particle averages (Fig. 2, G-J, top row). The unprocessed topographs of single connexons and their corresponding averages did not show any significant differences in connexon shape or substructure. The majority of single connexons imaged (Ͼ85%) exhibited the same structural conformation as reflected by their average. Thus, we can rule out that the connexons imaged may have represented a mixed state of several conformational states, but rather a single predominant configuration. Connexon averages were computed from AFM topographs recorded at pH 6.0 (Fig. 2G, top), 6.5 (Fig. 2H, top), 7.0 ( Fig. 2I, top), and 7.6 ( Fig. 2J, top). Averages calculated from connexons imaged at pH 8.0 and 8.5 did not show any significant deviation from that recorded at 7.6 and therefore are not shown. Regions of S.D. maps (Fig. 2, G-J, bottom row) exhibiting enhanced values indicate for structural fluctuations of the connexon surface. All averages show the 8°skew of the connexon lobes from the vertical axis characteristic of detergenttreated samples (35). In these averages, the connexon is displayed with a left skew. Whereas this overall appearance of the hexameric connexon does not change, the diameter of the central pore significantly increased with increasing pH. Besides measuring the channel diameters of averaged connexons, we analyzed those from single connexons. The histograms (Fig. 3) suggest that the average channel diameter, measured for a certain pH value reflects that of the majority (Ͼ85%) of the individual connexons. Under experimental conditions known to induce a closed channel (12,17), the channel entrance exhibited a maximum depth of 0.4 Ϯ 0.1 nm with a diameter of only 0.6 Ϯ 0.2 nm measured at full width half-maximum height (FWHM). The S.D. map of the average (Fig. 2G, bottom) shows no maxima at the channel entrance, suggesting that this area exhibited no enhanced structural flexibility. The surface structure of the connexon did not significantly change after increasing the pH to 6.5. However, the S.D. map of the closed connexon changed. The region at the channel entrance showed a slightly enhanced value of 0.15 nm (Fig. 2H, bottom), indicating that the corresponding structures now provided some structural flexibility. This may also explain the slightly increased maximum depth of the channel (0.5 Ϯ 0.3 nm). At this pH, the width of the channel entrance increased slightly to 0.9 Ϯ 0.3 nm (n ϭ 91). Further increasing the pH to 7.0 further increased the width of the channel entrance (Fig. 2I, top) to 1.3 Ϯ 0.2 nm (n ϭ 85). As a result of this opening, the AFM tip now could penetrate into the channel entrance, detecting an average maximum depth of 1.2 Ϯ 0.35 nm (n ϭ 40). Furthermore, the S.D. map of the connexon surface (Fig. 2I, bottom) increased its central maxima now indicating that the channel entrance has further increased its flexibility. At the same time, the outer regions of the connexins were observed to slightly enhance flexibility, as indicated by their S.D. increasing to 0.2 nm (n ϭ 40). Increasing the pH to 7.6 (   MARCH 23, 2007 • VOLUME 282 • NUMBER 12

pH-induced Structural Changes in Cx26 Hemichannels
ing of the channel entrance the S.D. map (Fig. 2J, bottom) showed that this structural region further increased its flexibility to a maximum S.D. of 0.35 nm (n ϭ 40). Interestingly, individual connexins showed an increased S.D. of 0.25 nm at their outer rims as well. Further increasing the pH to 8 (Fig. 2E) and 8.5 (Fig. 2F) did not significantly change the correlation averages or connexon structures of the extracellular surface (data not shown). It should be noted that "partially closed channels" had a diameter lying between that observed for fully open and closed channels. AFM topographs showed that more than 85% of the single hemichannels had no significant deviation in their channel diameter (Fig. 3). This suggests that the partially closed hemichannels adopted a functional state that reflects an intermediate channel size between fully open and closed conformations and not a mixture of solely open and closed states.
To prove whether the observed conformational change was reversible, we decreased the pH to 6.0 after the channels were fully opened at pH 9.0. The AFM topographs recorded showed that the previously opened pore now re-closed, suggesting that their conformational change was fully reversible. Cycles of pH changes were often repeated more than four times. Table 1 summarizes our data for the measurements of inner pore diameter and channel entrance depth for different pH values.
Conformational Changes Are Not Dependent on HEPES Concentration-HEPES belongs to a class of aminosulfonate compounds that have been shown to act as modulators of the Cx26 channel (17). Following the lead of Bevans and Harris (17), we tested whether the observed conformational change is due solely to pH or to the binding of aminosulfonates to Cx26, which in turn modulates a pH-induced conformational change.
We increased the HEPES concentration to 50 mM and imaged the extracellular connexon surface (Fig.  4). Surprisingly, AFM topographs recorded at pH 8.0 (Fig. 4A) and at pH 9 (Fig. 4B) showed that the open state of the connexon channels was not influenced by the increase in protonated HEPES. For both pH conditions, the averages of the inner channel diameters measured from single connexons were 1.6 Ϯ 0. Additional experiments were performed in an effort to close open channels with higher HEPES concentrations at pH Ն 7.5. The sample was absorbed to the mica surface at 50 mM HEPES, pH 7.5, and AFM imaging was performed with the same buffer. At this pH, the effective concentration of protonated HEPES would be maximal at ϳ25 mM. Under these conditions, the channels remained open. Open channels were also observed at HEPES concentrations up to 200 mM at pH 7.5 (data not shown). In each case, a positive control (10 mM HEPES, lowering the pH) was included to ensure the functionality of the sample. Alternatively, adsorption at pH 6.5 (50 mM HEPES) was performed to ensure HEPES binding and then increased the pH to 7.5 for AFM imaging. In this case, the topographs revealed open channels at pH 7.5 but partially closed channels at pH 6.5. Therefore, 10 mM may be the concentration for which binding is at saturation conditions in these experiments.
No Conformational Change Occurs in the Absence of Aminosulfonates-According to previous findings (17,18), the pH-dependent gating of Cx26 could be only observed in the presence of aminosulfonate-containing compounds. To test

pH-induced Structural Changes in Cx26 Hemichannels
this hypothesis and to prove that the conformational change observed can be indeed correlated to the gating mechanism, we imaged connexin preparations at different pH, buffered in nonaminosulfonate buffers. AFM topographs of maleate-buffered connexons showed that they did not change their conformation at pH 6.0, 6.5, or 7.0 (Fig. 5, A-C). The connexon surface did not change at pH values higher than 7.0. Similarly, the extracellular connexon surface apparently did not show a pH-dependent change in their channel diameter if the aqueous solution was buffered with potassium phosphate (KH 2 PO 4 ) (Fig. 5, D-F). All topographs (Fig. 5, A-F) and correlation averages (Fig. 5, G-N) have in common that the channel entrance appeared widely opened throughout the pH ranges imaged (compare with Fig. 2).
Quantitative Analysis of the Channel Closure-The diameter of the average connexon decreased gradually with lower pH (Fig. 6A). Whereas the depth of the channel decreased concomitant with an increase in the width of the connexon lobes, the overall diameter of the connexon itself does not change (Fig.  6B). This is also reflected in the difference image between the fully closed (pH 6.0, Fig. 7A) and fully open (pH 7.6, Fig. 7B) averages shown in Fig. 7C. In this difference image, positive differences are displayed as red and negative differences as black. The medium red level reflects no differences. It is also clear from this difference image that the extracellular region of the subunits rotate by ϳ6.5°between the open and closed states (Fig. 7, A and B) thereby changing the diameter of the pore. Our previously published hemichannel structure shows a narrowing of the connexon pore at the extracellular end that may be part of a physical gate closing upon acidification (36).

DISCUSSION
In this study, we have been examining the molecular mechanisms by which Cx26 hemichannels respond to H ϩ ions. An AFM equipped with a buffer chamber allowed us to observe Cx26 membranes under native conditions at high resolution. Analogous to the TSF system of Harris et al. (37) or the excised patch experiments of Verselis et al. (38), this is a reductionist system. We limit the number of variables in the imaging experiment, which makes our results complementary to the pH studies done in whole cells (12,39). We have found that Cx26 hemichannels close in response to acidification, but only in the presence of an aminosulfonate buffer, suggesting that the aminosulfonate cation acts as a modulator or ligand to specific sites on the Cx26 protein. In addition, our data support the hypothesis that the aminosulfonate binding site is located on the cyto-    (17) that aminosulfonate (HEPES) binding to Cx26 modulates the acidification-based closure of the hemichannel. The curves obtained for the pH-induced closure (Fig. 6A) closely follow the sigmoidal, gradual plots obtained with either electrophysiology (12) or by permeability measurements (17). It is, however, unknown whether the sigmoidal shape indicates a cooperative mechanism among the six connexin subunits (40). However, another AFM study of the Ca 2ϩ -induced closure of Cx43 hemichannels reconstituted into lipid vesicles reported an all-or-none effect (41). Topographs recorded in non-aminosulfonate buffers do not reveal any pH-dependent conformational change. Binding of aminosulfonates has been shown to occur at the C terminus (42), and incomplete binding of HEPES to the cytoplasmic domains that face down on the mica surface may explain why in our topographs recorded at low pH a minority species of channels show a wider diameter ("open" state) while the majority of channels have a smaller opening ("closed" state) at their extracellular surface. It is possible that some connexins may have bound HEPES, whereas others did not because of steric hindrance of the aminosulfonate-binding site on the C terminus. This binding event is required first before the pH effect can be activated because non-aminosulfonate buffers could not induce pH-dependent closures. If HEPES is sterically hindered from binding to   Table 1) imaged at different pH in aqueous solution buffered by 10 mM HEPES, 20 mM potassium phosphate, or 10 mM maleate. B, two-dimensional profile of the pore channels recorded at different pH values. Note that while the channel entrance becomes shallower, the connexon diameter does not change significantly.

pH-induced Structural Changes in Cx26 Hemichannels
the C-terminal tail or if the binding site has been altered by the interactions of the cytoplasmic domains with the mica support, then increasing the HEPES concentration will have no effect.
In this study, gap junctions that were absorbed to the mica at 10 mM HEPES and subsequently exposed to 50 or 200 mM HEPES did not close the pores. This is in contrast to the results of Bevans and Harris (17) where raising the protonated HEPES concentration closed the channel. The most likely explanation is that at this concentration and with homomeric Cx26 channels, HEPES binding sites are already saturated to effect a conformational change.
pH Gating of Connexins-Chemical gating has been extensively studied in cell-cell channels (43,44). Typically, CO 2 has been used as the acidification agent in whole cell experiments, and the term is often used interchangeably with pH gating (6). It has been proposed that calmodulin plays a role in modulating chemical gating (44) by acting as a "cork" to plug the channel entrance at the cytoplasmic surface. Here, we imaged the effects of direct acidification in the presence of different buffers rather than with CO 2 .
Channel closure by H ϩ ions is a common gating mechanism among the connexin family. Some members such as Cx32 channels are more pH-independent than Cx26 channels, which are highly pH-sensitive within a physiological pH range. Other isoforms such as Cx43 or Cx45 display an intermediate sensitivity to acidification (12). For cardiac connexins (Cx43, Cx40, and Cx45), Delmar and co-workers (12) demonstrated that truncation of the C terminus significantly decreased the response of the intact channels to acidification. In a study comparing Cx46 channels and hemichannels expressed in oocytes, and using excised patches of Cx46 hemichannels, Verselis and coworkers (19,38) showed that Cx46 hemichannels act similarly to intact channels and close reversibly with acidification. The site of action for H ϩ is localized to the cytoplasmic domains. These kinetics are slow (tens of milliseconds), and no additional components other than KCl buffer were necessary to induce closure in Cx46 hemichannels. It was also noted that a "pH inactivation" occurred, whereby the number of hemichannels re-opening with increased pH decreased every cycle. We could observe such an effect as well. Whereas the opening and closure of most Cx hemichannels were fully reversible, some individual channels remained in their closed state and re-opened much later than the others. These pH cycles, during which we watched the closing or opening of hemichannels, were repeated many times (Ͼ4).
Whereas hemichannels have been documented as having unique functions (reviewed in Ref. 45) separate from the ones served in docked channels, it has been proposed that the two docked hemichannels act in series (38), so the results we obtained for connexons may be extrapolated to intercellular channels. Hemichannel plaques have been imaged with AFM in isolated preparations (46) and using freeze fracture and thin section electron microscopy in Xenopus oocytes expressing exogeneous Cx50 (47). In vivo Cx26 hemichannels have been demonstrated to be expressed in endogenously and exogenously transfected tissue culture cells (48) and in retinal horizontal cells (49,50). In recent work by Tao and Harris (42), HeLa cells were stably transfected with combinations of Cx26 and Cx32 with or without a C-terminal tag. As assayed with a parachute dye coupling assay, dye transfer through native heteromeric Cx26/Cx32 or Cx26/Cx32 tag intercellular channels was significantly reduced by the presence of 10 mM taurine in the medium. HeLa cells contain a plasma membrane-bound taurine transporter so that cytoplasmic taurine levels become elevated when excess taurine is added extracellularly. Tao et al. (42) showed that tagged Cx26 heteromeric channels were unaffected by the increased taurine concentration, presumably because the taurine binding site is at the distal end of the C terminus and the added tag sterically blocks binding. HEPES blocked the taurine-induced inhibition, because it obstructs the taurine transporter and is membrane-impermeable. 4 Following up with simple scrape-dye loading assays (51) on the Cx26 overexpressing HeLa cells we used in this AFM study, we also see a decrease in dye transfer with 10 mM taurine in the medium (data not shown). Therefore, the pH gating observed here for "undocked" Cx26 hemichannels is relevant and may be comparable to that seen in Cx26 intercellular channels. Taken together, these results suggest that the binding site is most likely on the cytoplasmic side, because the extracellular domains in intercellular channels are typically inaccessible to ligands, such as peptides or antibodies.
Mechanism of Channel Closure at the Extracellular Surface Gate-Mechanisms to explain the opening and closing of connexin channels have been proposed based on the hypothesis that there are two different physical gates: one at the cytoplasmic surface and one at the extracellular surface. High-resolution topographs of the Cx26 cytoplasmic surface indicate a unique surface domain structure (20) containing both the N and C termini and cytoplasmic loop. It is not known whether the short C terminus interacts with the cytoplasmic loop, although, as previously discussed, C-terminal tagging of Cx26 in heteromeric Cx26/Cx32 hemichannels and gap junctions eliminates the aminosulfonate-modulated pH closure. Cx32, a much more pH-insensitive connexin, is not affected by aminosulfonates (17) and presumably does not contain an aminosulfonate binding site. Because the transmembrane and extracellular domains are highly conserved among connexins, the 4 A. Harris, personal communication.  MARCH 23, 2007 • VOLUME 282 • NUMBER 12 ligand binding site would most likely reside in the variable C terminus of Cx26. This part of the sequence is not conserved in Cx32. Therefore, we propose that there is a two-step process where the aminosulfonate first has to bind to a C-terminal domain before a pH-induced conformational change can occur.

pH-induced Structural Changes in Cx26 Hemichannels
Because our high-resolution topographs reveal only the topology of the extracellular surface (20), we address what may be occurring at the extracellular gate. Two mechanisms have been postulated for the gating of the connexon at the extracellular gate. The first is that of Unwin and co-workers (2,52), whereby the pore closes by a pivoting and tilting of the hexameric subunits from a stationary extracellular end. This model postulates that there is very little change in the channel opening at the extracellular end, but that rotation of the individual subunits causes the channel diameter to become more compact, and the channel length is slightly increased as well (53). The second mechanism was obtained by single channel conductance measurements in Cx46 hemichannels whereby two distinct gates, one fast reacting and the other slow reacting, were observed. It was proposed by Trexler et al. (5) that the fast gate was physically located closer to the cytoplasmic surface and the slow gate closer to the extracellular surface. The fast gate has been proposed to be the cytoplasmic gate that acts in a receptor/ligand mechanism (4). The extracellular voltage gate was postulated to form a "loop gate" in isolated hemichannels and a cytoplasmic "cell-cell channel" gating mechanism for paired, docked connexons (5). It was shown using an electrophysiological analysis, that these slow component currents seen in hemichannels were similar to those observed during early cellcell channel docking events. The term "loop gate" was coined because Trexler et al. (5) proposed that the main structural elements involved in docking are the extracellular loops. We see a 6.5°rotation of the subunits between pH 6.0 and 7.6 ( Fig. 7,  A and B). Whereas Unwin and co-workers (2,52) predicted this rotation of the subunits, we did not observe a change in the connexon outer diameter as would have been predicted by the Unwin model. Such a consistency of the connexon diameter would argue for the "loop gate model." Evidence from Substituted Cysteine Accessibility Method (SCAM) studies have shown the loop gate is extracellular to amino acid Leu 35 in Cx46, and that the gate is localized to this extracellular end (54). However, the electrophysiological data cannot discern if there would be a rotation in the subunits in concert with this extracellular gate. More recent three-dimensional reconstructions (36,55) contain a constriction of the pore ϳ3/4 from the cytoplasmic surface opening. In both these three-dimensional structures, the pore does not appear to be closed, and the physiological state of the channel is ambiguous. In the structure by Unger et al. (55), the channel has been speculated to be in a closed or partially closed state, because the crystals were obtained in the presence of oleamide, a compound that was demonstrated to close Cx43 channels in vivo (55). Our AFM topographs support the model of a physical gate close to the extracellular surface, because not only does the pore diameter decrease upon acidification, but also its depth is shallower as probed by the AFM tip (see Fig. 6B).
Connexon Extracellular Surface Is More Rigid When the Extracellular Gate Is Closed-As recently reviewed (23), there are several lines of evidence that the extracellular domain is fairly rigid. One structural model for the extracellular domain consists of two concentric ␤-barrels (56). Cryo-EM reconstructions have shown protrusions at the extracellular surface that interdigitate with those from a symmetry-related partner connexin (57), a topology supported by the AFM images presented in this work. Our AFM topographs indicate that there is some flexibility dependent on the open state of the channel and that this flexibility is not homogenous. In the open state, this flexibility is highest at the channel entrance, minimal at the adjacent structures, and higher at the outer rim of the structures again. The fact that channel flexibility increases with diameter has been observed before with Ca 2ϩ -dependent closure (20).
The Relevance of pH Gating in Tissues and Organs-Tissue cells use intracellular acidification as a mechanism for invoking regulatory processes. With respect to gap junction function, two systems are worth noting for their relevance to disease processes. The first is the pH gating of Cx43 in heart gap junctions and the second, relevant to the isoform used in this study, is the effect of pH in the inner ear with implications for hearing dysfunction or loss.
Cx43 channels close in response to acidification (40), and this has implications for the functioning of cardiac tissue where Cx43 is the most highly expressed connexin. During ischemic events, a drop in intracellular pH (pH i ) occurs in heart (58) and in brain (59). Cardiac ischemia leads to electrical uncoupling (60) as well as to dephosphorylation and internalization of Cx43 (61). Gating of connexin channels by lowered pH is likely to contribute to arrhythogenesis in acute myocardial infarction by closing connexin channels when the cytoplasm is acidified by the accumulation of lactic acid because of anaerobic metabolism, thereby eliminating electrical coupling (62). However, this uncoupling of cells may serve a more important role of isolating normal tissue in the heart from damaged heart cells, thereby limiting the spread of acute ischemic injury (63).
Cases of patients with hereditary non-syndromic deafness because of mutations in the Cx26 sequence have pointed to the important role that these gap junction channels play in homeostasis and most likely in potassium ion recycling (64,65). Cx26 and Cx30 are co-expressed in the epithelium and connective tissue of the cochlea. Cx26-based intercellular communication in the inner ear has been postulated to play a homeostatic role in hearing analogous to the function that Cx32 plays during action potential generation in nervous tissue (66,67). It has been proposed that gap junctional communication restores the ionic balance after a nerve potential has been generated and that dysfunction of Cx26 channels results in exotoxicity leading to inner ear damage (68). While the role of pH gating in the inner ear has not been well investigated, the study by Ikeda and Morizono (69) has suggested that CO 2 levels influence the acidbase regulation of inner ear fluids through the cochlear round window and that changes in CO 2 levels and pH can affect cochlear function. In severe cases, this might lead to sensorineural hearing loss that has been demonstrated in cases of patients with secretory otitis media (70).

pH-induced Structural Changes in Cx26 Hemichannels
The Importance of Taurine in Tissues and Organs and Coexpression with Cx26-The HEPES buffer we used in this series of experiments is one of a class of compounds known as aminosulfonates. The simplest of the ␤-aminosulfonates is ␤-aminoethane sulfonic acid, also known as taurine, which in many animals is one of the most abundant low molecular weight organic compounds (15). In mammals, significant cytoplasmic taurine concentrations are highly ubiquitous in their tissue distribution with the highest concentrations found in heart, brain, muscle, and particularly in retina. The concentration in these tissues is at millimolar levels; however, for HeLa cells, the expected concentration would be in the mid-micromolar range (15). In many cell types, there are active taurine transporters in the plasma membrane (71). It is interesting to note that heart and muscles do not contain Cx26; however, astrocytes and leptomeningeal cells do contain this isoform (72). This suggests that there may be an inverse relationship between tissues expressing Cx26 and those containing cytosolic millimolar amounts of taurine. Hence, tissues containing lower levels of taurine could modulate Cx26 channel function in a much more regulated manner. It has been hypothesized that taurine functions to aid in osmoregulation, especially in excitable tissues rich in membranes (73) where high, transient ionic fluxes are accompanied by osmotic imbalances. Taurine modulates many Ca 2ϩ -dependent processes (as reviewed in Ref. 15), although this is thought to occur by indirect processes. Another function for taurine as an antioxidant has been proposed (74). Harris and co-workers along with our new experiments have demonstrated a potential biological role for aminosulfonate binding to Cx26. A subject for further investigations is to explore whether other connexins highly expressed in excitable tissues show a taurine-modulated acidification-induced closure in whole cells, where taurine can exist in millimolar concentrations, and whether these isoforms can bind this compound, or whether taurine binding is unique to Cx26.