Contribution of transmembrane regions to ATP-gated P2X2 channel permeability dynamics.

ATP-gated P2X(2) channels undergo activation-dependent permeability increases as they proceed from the selective I(1) state to the I(2) state that is readily permeable to organic cations. There are two main models about how permeability changes may occur. The first proposes that permeability change-competent P2X channels are clustered or redistribute to form such regions in response to ATP. The second proposes that permeability changes occur because of an intrinsic conformational change in P2X channels. In the present study we experimentally tested these views with total internal reflection fluorescence microscopy, electrophysiology, and mutational perturbation analysis. We found no evidence for clusters of P2X(2) channels within the plasma membrane or for cluster formation in response to ATP, suggesting that channel clustering is not an obligatory requirement for permeability changes. We next sought to identify determinants of putative intrinsic conformational changes in P2X(2) channels by mapping the transmembrane domain regions involved in the transition from the relatively selective I(1) state to the dilated I(2) state. Initial channel opening to the I(1) state was only weakly affected by Ala substitutions, whereas dramatic effects were observed for the higher permeability I(2) state. Ten residues appeared to perturb only the I(1)-I(2) transition (Phe(31), Arg(34), Gln(37), Lys(53), Ile(328), Ile(332), Ser(340), Gly(342), Trp(350), Leu(352)). The data favor the hypothesis that permeability changes occur because of permissive motions at the interface between first and second transmembrane domains of neighboring subunits in pre-existing P2X(2) channels.

(2), with some displaying high Ca 2ϩ fluxes (15). P2X 2 , P2X 4 , and P2X 7 channels also display permeability dynamics, whereby the channel pore dilates in an ATP activation-and time-dependent manner from a fairly selective I 1 state to the I 2 state that is also permeable to organic cations (16 -24). The transition from I 1 to I 2 typically takes several seconds. Permeability changes have been studied most for recombinant P2X 2 , P2X 4 , and P2X 7 channels (2,22), but there is also evidence that permeability changes occur for natively expressed channels in cells throughout the body, including neurons (16,19,20,(25)(26)(27)(28).
Activation-dependent changes in ionic selectivity for P2X channels imply that the selectivity filter may be dynamic and undergo conformational re-arrangements to switch its preference between ions (22,29). Two different models are proposed to explain the ability of P2X channels to undergo permeability changes in response to ATP (22). The first, which we term the cluster model, suggests that I 2 state-competent P2X channels represent clusters of accreted P2X channels/ subunits, perhaps of higher oligomeric state, and that ATP triggers this process (22). The essential features of this hypothesis were first proposed in a pioneering study on mast cells (30). This has recently drawn some support based on the observation of higher molecular weight, presumably P2X subunit-containing, bands on protein gels (6,9) and the finding that permeability changes in P2X channels are dependent on channel density (31). First principles also suggest that the pore diameter would increase with the number of helices lining the pore (32). However, in the case of P2X 7 channels there is experimental evidence to suggest that channel redistribution and cluster formation may not be needed for pore dilation (23). The second hypothesis, which we term the gating model, proposes that permeability changes occur because of an intrinsic conformational change in the channels. This draws on findings with mutagenesis (16,17) and ion replacement experiments (33) as well as work providing evidence for a permissive conformational state with FRET 1 (34). The gating model also draws indirect support from studies of mechanosensitive channels that are known to dilate from ϳ2 to ϳ30 Å because of an intrinsic conformational change involving helix tilting (35). Calculations also show that for channels with regular helices and ideal oligomeric symmetry an increase in pore diameter can be achieved with an increase in helical tilt (32). Thus both the cluster and gating models have their merits, but no study has tested the models directly under a fixed set of experimental conditions or in the case of the cluster model, with approaches that have sufficient resolution to provide direct information about P2X 2 channels in the plasma membrane.
We tested the cluster model by imaging fluorophore labeled P2X 2 channels in the plasma membrane with total internal reflection fluorescence microscopy (34,36,37). Our experiments provided no evidence to support the cluster model. We next sought to identify P2X 2 subunit regions permissive for permeability changes in an effort to provide insight and constraints on the gating model. Ab initio there was no way to deduce which parts of the protein may be most important because there is no structural information on P2X channels, no related channels of known structure, and no channels with comparable primary sequence (1,2). Because of this quandary we sought to identify regions crucial for the gating model empirically with single site Ala substitutions and analysis of state-specific perturbations. We focused on the transmembrane segments because of their importance for channel function (11-16, 39, 40) and because no detailed study of the TM segments and their role in permeability changes exists. Overall the data are consistent with the gating model and the hypothesis that permeability changes occur because of spatially diverse, and perhaps extensive, molecular rearrangements in the transmembrane, pore, and cytosolic domains (34) of pre-existing and stably expressed P2X 2 channels.

EXPERIMENTAL PROCEDURES
Molecular Biology-The cell lines used were HEK-293 cells transiently expressing appropriate wild type (wt), mutant, and fluorescently labeled channels. Some cDNAs were available from previous work (12,18,34,40), some were from R. Alan North (Sheffield University), and others were generated using standard procedures. cDNAs were propagated in DH5␣ Escherichia coli, and plasmids were purified using standard techniques. Plasmids encoding mem-Y were purchased from Clontech (pEYFP-Mem). For transient expression in HEK-293 cells 0.5-1 g of plasmid cDNA was used (Effectene, Qiagen).
TIRF Microscopy-Briefly, we used an Olympus IX70 scope equipped with a Princeton Instruments cooled I-PentaMAX camera with a High Blue Gen III Intensifier (Roper Scientific). The control of excitation and image acquisition was achieved using MetaMorph software and drivers (Universal Imaging), shutters, filter wheels, and Proscan II control box (Prior). The beams of 488/515 nm argon (150 milliwatt) and 442 nm helium-cadmium (12.5 milliwatt) lasers (Melles Griot, Carlsbad, CA) at Ͻ5% power (constant for all experiments reported) were combined and controlled with an IX2-COMB (Olympus), Uniblitz shutters (Prior), and acoustoptical tunable filter and controller (AA Optoelectronics, Les Chevreuse, France) and fed into an optical fiber (FV5-FUR; Olympus) for entry into the TIRF condenser (IX2-RFAEVA-2; Olympus). Cells were plated onto glass-bottom Petri dishes (170 nm thick; Willco Wells BV, Amsterdam, Netherlands) 24 -48 h before imaging and were viewed with a 60ϫ oil immersion objective lens with a numerical aperture of 1.45 (Olympus). The gain was adjusted for maximum signal-to-noise for each cell and kept constant for all image acquisitions.
Electrophysiology-HEK-293 cells were used for recordings 24 -48 h posttransfection, gently mechanically dispersed, and plated onto glass coverslips 2-12 h before use. We included this step (11) to ensure adequate voltage clamp during reversal potential measurements from single spherical cells. The extracellular recording solution was comprised of (in mM) NaCl, 150; MgCl 2 , 1; CaCl 2 , 1 (or 0.1); HEPES, 10; and glucose, 10 (pH 7.4), and the pipette solution was comprised of (in mM) NaCl (or CsCl or CsF), 154; EGTA, 11; and HEPES, 10. For some experiments, where stated, we replaced Na ϩ with equimolar substitutions of NMDG ϩ (16,17). Whole-cell voltage clamp recordings were made with 5 M⍀ borosilicate glass electrodes (Sutter Instruments), using an Axopatch-1D amplifier controlled by a computer running pCLAMP 8.1 software via a Digidata 1366 interface (Axon Instruments). Data were filtered at 0.5-2 kHz and digitized at 1-5 kHz. Drugs were applied to single cells by pressure-ejection using a Picospritzer (Intracell Ltd.), or in the bathing medium, which flowed at 2-3 ml/min. For pressure application (ϳ20 p.s.i.; 1 p.s.i. ϭ 6.89 kPa), we used 2-4 M⍀ pipettes as described previously (18); solution exchange occurred in ϳ10 ms. Solution exchange with the bath application occurred in ϳ10 -20 s, but no agonists were applied in this way. Voltage ramps (0.5-s duration) were usually applied every 0.5-1 s, but in some cases 0.1-s duration ramps every 0.1 s were also tested (34). On the basis of initial experiments the voltage range was adjusted to be ϳ40 -60 mV either side of the reversal potential for each construct.
Data Analysis-Data were analyzed using published approaches (29,33,41). Calcium permeability (pCa 2ϩ /pNa ϩ ) was determined under bi-ionic conditions, as described (42). Ion permeability ratios were calculated from shifts in reversal potentials using the function pNMDG ϩ / pNa ϩ ϭ exp(⌬E rev F/RT), where pNMDG ϩ is the permeability to NMDG ϩ , pNa ϩ is the permeability to Na ϩ , ⌬E rev is the shift in reversal potential, and F, R, and T have their usual meaning (33). In some of the electrophysiological traces we used adjacent point averaging. All ramps shown in Figs. 3 and 5 are leak-subtracted. Electrophysiological analysis was performed with Clampfit (Axon Instruments) or Origin 6.1 (OriginLab Corp.), and all statistical tests were run in GraphPad Instat 3.0 (GraphPad Software). Desensitization was quantified as the percent current lost after 20 -30 s of ATP (100 M) relative to the peak response (Fig. 6). We used this simple approach to quantify desensitization because the decay kinetics could not always be adequately described by exponential functions with 1 or 2 components. Data in the text and graphs are shown as mean Ϯ S.E. from n determinations as indicated.

RESULTS
We studied homomeric P2X 2 channels because they show robust permeability changes in all HEK cells, they can be labeled innocuously with GFP variants (18,34), and they likely underlie the high permeability to organic cations measured for natively expressed P2X channels in neurons (16,17,19,27). For the Ala mutants we used ␣ helical representations of the first (TM1) and second (TM2) TM segments (40,43). ␣ helices are a common feature of the TM domains of ion channels (32). Cysteine and alanine scanning mutagenesis experiments also suggest that TM1 and TM2 of P2X 2 are helical (11)(12)(13)(14)40), although there are differences between groups (2).
No Evidence for Cluster Formation of Fluorophore-labeled P2X 2 Channels-Recent work on Xenopus oocytes suggests that P2X 2 channels may form clusters permissive for the I 2 state when channel density is as low as ϳ25 channels/m 2 (31). Previous work indicates that P2X 2 channels expressed in mammalian cells are at a channel density of ϳ1-3 channels/m 2 (18,34). Available data suggest that P2X channels have a total of six transmembrane helices and based on electron microscopy studies of transmembrane proteins (45) may form a particle of ϳ8.4 nm 2 in cross-sectional area. With a channel density of ϳ1-25 channels/m 2 they thus occupy a minor part of the plasma membrane expanse in which they sit. Assuming a uniform distribution of ion channel coordinates, and a membrane protein diffusion coefficient of 10 Ϫ10 cm 2 ⅐s Ϫ1 , estimates suggest any two random channels (at three per m 2 ) will on average stay ϳ0.5 m apart, but every ϳ100 s they will come within ϳ1 diameter of each other for ϳ3 ms. For mammalian cells these considerations lead to the conclusion that clustering cannot occur randomly simply because there are too few channels for even the highest expression levels where permeability changes have been measured (16,26,31,33,34). If on the other hand clustering occurs because of specific molecular interactions then regions of high channel density must exist in an ocean of plasma membrane with no or few channels. This is because the total number of channels is too few to support clustering everywhere, which presumably requires several thousand per m 2 (29). If clustering has any relation to permeability changes, then preformed clusters must either exist in a background of plasma membrane relatively devoid of channels, or they must form in response to ATP. We tested these predictions with TIRF microscopy.
We imaged CFP-or YFP-tagged P2X 2 channels (18, 34) to determine whether they formed clusters in the plasma membrane of HEK cells with TIRF microscopy (Fig. 1A), which affords the excitation and imaging of channels within ϳ100 nm of the plasma membrane (36). Using TIRF microscopy we found no evidence of P2X 2 -C or P2X 2 -Y channel clusters for areas of HEK cells in close apposition to glass, an area called the footprint (Fig. 1, B and C). The higher intensity region in the center of the footprint in Fig. 1C does not represent clustering but rather shows the region of the cell in closest adherence to glass as the edges of the cells curl away out of the evanescent field. Indeed the qualitative appearance of P2X 2 -Y fluorescence, and quantitative measures of P2X 2 -Y hot spots, were similar to those observed for two different membrane-targeted YFP proteins (mem-Y and PH-Y) and with T18A P2X 2 -YFP channels that are known to lack the I 2 state (18). In contrast P2X 4 -Y channels were more clustered in the plasma membrane ( Fig. 1, B and D) and showed significantly more hot spots per footprint (Fig. 1B). Examination of line profiles across the highest intensity regions of the footprints (three examples for P2X 2 -Y and P2X 4 -Y are shown in Figs. 1, C and D) indicated that P2X 2 -Y channels displayed a smooth distribution in the plasma membrane with no abrupt peaks corresponding to hot spots that were frequently observed for P2X 4 -Y-expressing cells (arrows in Fig. 1D). Because we could readily observe and count hotspots for P2X 4 -Y channels we feel confident in suggesting that P2X 2 -Y channels do not cluster. Given that all cells expressing P2X 2 or P2X 2 -Y channels undergo permeability changes (114/114 cells; Table I) this implies that cluster formation is not a prerequisite for permeability changes. Seemingly P2X 2 channels are quite evenly distributed in the plasma membrane of HEK cells (46). We next asked whether P2X 2 -Y channels become more clustered in response to ATP. To this end we imaged P2X 2 -Y channels and rapidly applied ATP (100 M). We chose 20-s applications because permeability changes are complete within this time (17,33). Again we found no evidence that P2X 2 channels move in response to ATP, because the images of P2X 2 -Y footprints were indiscernible before and during ATP, both in terms of appearance ( Fig. 2A) and in terms of intensity ( Fig. 2D; Ͻ5% change, n ϭ 8). The negative data with P2X 2 -Y channels showing that they do not cluster in response to ATP warranted several controls to be confident that we could measure movement of fluorescence, if it occurred. First, we monitored the access of a fluorescent dye (5 M Lucifer yellow, n ϭ 4) to the footprint. The dye flowed freely and completely into the area between the cell and glass coverslip within 1 s (Fig. 2, B and D). Second, we imaged footprints from cells expressing PH-Y and hP2Y 4 receptors. In response to 50 M UTP application to cleave phosphatidylinositol 4,5-bisphosphate (47) we measured marked decreases in the intensity of fluorescence within the footprint (Fig. 2, C and D; n ϭ 4). We interpret this to indicate that PH-Y moved into the cytosol due to its greater affinity for phosphatidylinositol 1,4,5-trisphosphate (48). Together, these experiments suggest that access of molecules to the footprint is  (36). Laser light was focused onto the back focal plane of the objective lens, and its position adjusted so that it emerged into the immersion oil at an angle shallower than the critical angle. This created an evanescent field of illumination of depth ϳ100 nm into the cell, which was adhered to the coverslip. B, summary graph of the numbers of fluorescent hot spots per footprint from HEK cells expressing the various constructs as indicated. C, image of a footprint from a HEK cell expressing P2X 2 -Y channels. The lower plots show line profiles from 3 footprints. D, as in C but for P2X 4 -Y channels; the arrows indicate hot spots of fluorescence. The scale bar in C and D is 10 m.
relatively fast (ϳ1 s). We interpret these data to indicate that P2X 2 -Y channels do not move in response to ATP to form regions of markedly higher channel density with a time course relevant to permeability changes.
Alanine Scanning Mutagenesis and Perturbation Analysis of I 1 and I 2 States-Given the lack of evidence in favor of the cluster model for P2X 2 channels we next tested the gating model and sought to identify transmembrane regions that may be important for permeability changes. Examination of the first and second TM domain sequences for P2X 1 -P2X 7 subunits provided little further insight (Fig. 3, A and B), a trend reminiscent of findings with Ca 2ϩ flux (15) underscoring the neces-TABLE I Properties of I 1 and I 2 for wt and mutant P2X 2 channels Class 1 channels displayed reduced I 2 NMDG ϩ permeability, class 2 channels displayed no time-dependent increase in NMDG ϩ permeability and no detectable I 2 state, class 3 channels were not significantly different to wt P2X 2 , and class 4 displayed an I 1 state with significantly higher pNMDG ϩ /pNa ϩ . In the case of P2X 2 -C/Y recordings were made from cells expressing CFP and/or YFP labeled channels; the data have been pooled because there were no differences between them (31 of 31 cells). a Indicates no, or negligibly small, ATP-evoked currents; these mutants were not studied further.

P2X 2 Transmembrane Gating Regions
sity for experimental work. One way to empirically identify important transmembrane regions is to assay channels for functional effects where every residue in a given stretch has been mutated to Ala, one residue at a time. Ala scanning allows one to assay the effect of deleting an amino acid side chain beyond the C ␤ carbon atom on a given property of the protein (49). The method assumes that important residues repack as the channels proceed from one state to another. If a particular residue contributes to repacking, then substitution to a hydrophobic Ala is expected to disturb the equilibrium between the two states and results in altered function of one or both states (50 -53). By comparing across domains within a protein it should be possible to identify regions that are particularly important for gating to a particular state and in the most ideal cases infer secondary structure of these regions (51,52). The method has been widely used in studies of other channels and recently in studies of P2X 2 channels with an emphasis on the ability of ATP to bind or open the channel (40, 54).
We assayed mutants for permeability changes where every amino acid in TM1 and TM2 of P2X 2 was exchanged for Ala (40). Four mutants (V45A, Y47A, V51A, and D349A) were expressed in the plasma membrane but were non-functional and not studied further (40). The remaining mutants all produced functional responses in HEK cells when challenged with 100 M ATP (Table I), a concentration chosen to be near maximal for the mutants and wt P2X 2 (40). We tested all the mutants for initial pNMDG ϩ /pNa ϩ (the I 1 state) and for changes in this parameter as they enter the I 2 state (Table I).
Representative traces for five mutants and wt P2X 2 channels are shown in Fig. 3, C-H. For both transmembrane domains the Ala mutants produced only subtle effects on the I 1 state measured within 0.2-1 s of applying ATP (Table I; L41A, L42A, Y43A, F44A, Q52A, I328A, I332A, L338A, S340A, G342A, and G344A). More profound effects were observed for the I 2 states measured 10 -30 s into the ATP application period (Table I; Table I and from an analysis of variance: across TM1 and TM2 variance for I 1 was 7.9 and 12.4% but 52.1 and 48.0% for I 2 (Table I). Clearly I 2 is readily perturbed by Ala substitutions, with 12 and 13 residues affected in TM1 and TM2.
To obtain a measure of the mutants that affected the I 2 state only, we analyzed the magnitude of the shift in reversal potential (⌬E rev ) from I 1 to I 2 . In Fig. 4 the residues highlighted in blue are those that showed a ⌬E rev that was significantly smaller than wt P2X 2 , whereas the residues in red are those that showed no ⌬E rev at all. The colored residues (blue and red) thus provide an upper estimate on the number of Ala mutants that may affect the I 2 state. We focused on those mutants that totally abolished the I 2 state (red in Fig. 4; no ⌬E rev ), because subtle changes in reversal potential may arise because of cellto-cell variability and expression levels (31,33). These criteria revealed 6 residues in TM1 and 6 in TM2 that lacked the I 2 state (Fig. 4). In TM1 they were F31A, R34A, Q37A, Y43A, F44A, and K53A. In the case of TM2 the 6 residues were I328A, I332A, S340A, G342A, W350A, and L353A. We next tested these mutants for normal membrane expression by comparing the peak ATP-evoked currents in physiological solutions with those measured for wt P2X 2 channels. This analysis revealed that ATP-evoked currents for Y43A and F44A were significantly smaller than wt P2X 2 channels (Ϫ46.8 Ϯ 9.3 and Ϫ64.5 Ϯ 2.3 pA/pF, respectively, versus Ϫ576.4 Ϯ 116.7 pA/pF for wt P2X 2 ; Fig. 6C). Peak current data for all TM1 and TM2 mutants that showed perturbed I 2 states are shown in Fig. 6C and for all the mutants reported in this study in a recent paper by Li et al. (40). Because membrane expression levels are important for permeability changes (31), we excluded these residues from the helical wheel plots in Fig. 4 that emphasize residues that lacked the I 2 state but showed membrane expression equal to wt P2X 2 channels. All these residues fell on one-half of helical wheel representation of TM1 or TM2 (Fig. 4). The results with S340A should be interpreted with caution because this amino acid is known to affect cation flux through the pore (15,42). Thus for TM1 and TM2, 4 and 6 residues contribute the most to the I 1 to I 2 transition, and they fall on one-half of helical wheel representations of TM1 and TM2.
Disulfide Formation between V48C and I328C-Previous studies suggest that channels formed by expressing V48C and I328C mutants form a disulfide bond between these two residues, indicating that these TM1 and TM2 residues lie within ϳ8.4 Å of each other (14,55). Biochemical experiments further suggested that the disulfide was formed between the TM1 and TM2 domains of neighboring subunits (8). Remarkably when interpreted from this perspective all except one of the Ala residues that perturb the I 2 state lie on the same half of an ␣ helical view of TM1 as V48 and as I328 for TM2 (Fig. 7A). The exception was S340A, and it seems possible that this mutant directly affects the selectivity filter (15,42) rather than gating to the I 2 state. These data suggest that the Ala mutants may perturb the I 2 state by hindering rearrangements that occur at the interface between TM1 and TM2 of neighboring subunits (8,14). We attempted to test this with V48C and I328C mutants.  A and B, respectively). C, representative traces for wt P2X 2 channels expressed in HEK cells and bathed in NMDG ϩ solutions. The upper panels show the steady state current at Ϫ60 mV, whereas the lower panels show current-voltage relations determined every 500 ms after the peak response (I 1 ) and finishing at the steady state current-voltage relationship (I 2 ). D-H, as in C for two mutants with normal I 1 states but no I 2 states (D, E), for two mutants with normal I 1 and I 2 states (F, G), and one mutant with normal I 1 states and impaired I 2 states (H). Consistent with past work (8,14,55), the reducing agent dithiothreitol (DTT) (10 mM) produced no effect on 100 M ATP-evoked currents recorded from cells expressing wt P2X 2 or the single V48C and I328C mutant P2X 2 channels (Fig. 5,  A-D). However, DTT augmented 6-fold ATP-evoked currents recorded from channels that contained V48C and I328C mutants (Fig. 5, C and D). The magnitude and time course of the DTT effect were consistent with past data (8,55). We next asked whether V48C and I328C channels undergo permeability changes and display the I 2 state. Channels formed by expressing V48C or I328C mutants showed significant increases in NMDG ϩ permeability and a robust I 2 state (Fig. 5, E and F). It is noteworthy that the shift in reversal potential for these single mutants was reduced by about half in relation to wt P2X 2 (Table I). We next repeated these experiments for channels formed by co expressing V48C and I328C double mutants because they should form a disulfide between subunits (8). For these channels the I 1 and I 2 state were not significantly different from either V48C or I328C mutants (Fig. 5, G and H; Table  I). When interpreting this result it is important to consider that co expression of two subunits is expected to result in a mixed population of channels in the membrane. For these reasons it is problematic to draw comparisons between I 1 and I 2 states formed from channels upon co-expression of V48C and I328C FIG. 4. I 2 state-specific Ala mutants. A, upper graph shows the effect of single site Ala mutants on the shift in reversal potential measured from cells containing extracellular NMDG ϩ for TM1, whereas the middle panel shows a helical net representation of TM1 with the colored residues indicated positions where the I 2 state was abolished. B, as in A but for TM2. For A and B residues listed in blue showed significantly reduced shifts in reversal potential, those in red showed no shifts in reversal potential (over 30 s), and the remainder were no different from wt P2X 2 . The lower panels in A and B are helical wheel plots of TM1 and TM2. The highlighted residues are those that lacked I 2 states but expressed at levels equal to wt P2X 2 channels. with those of channels formed in cells expressed either mutant alone (Table I). The most reliable test to determine whether a disulfide formed between V48C and I328C upon coexpression is to assay the effect of a reducing agent to break the disulfide. If motions between V48C and I328C were needed to allow permeability changes to occur, then one would expect breaking the disulfide with DTT would affect the ability of V48C and I328C channels to undergo permeability changes. If on the other hand motions at, or near, V48C and I328C are not needed, then breaking the disulfide should have no effect. Remarkably, application of DTT (10 mM for 2-5 min; Fig. 5C) produced significant effects on channels formed by coexpressing V48C and I328C channels. The channels opened to an I 1 reversal potential much more depolarized than before DTT (Fig. 5, G-I), as though once the disulfide had been broken the channels readily opened to an I 2 -like state and then dilated further by ϳ20 mV to a larger I 2 state. If one assumes that the true I 1 state for these DTT-treated channels has the same reversal potential as the average of that measured for channels formed by V48C, I328C, and coexpression of V48C and I328C in the absence of DTT (mean Ϫ65.7 Ϯ 2.5 mV; Table I), then the net shift in reversal potential for channels formed by co expressing V48C and I328C as they enter the I 2 state after DTT is ϩ42.5 Ϯ 2.4 mV, which is significantly larger than that measured in the   FIG. 5. V45C and I328C mutants. A, whole-cell ATP-evoked currents (100 M) recorded from HEK cells expressing V45C/I328C mutants. ATP was applied every 2 min for 10 applications, and DTT (10 mM) was applied during the times indicated by the bar. B, representative traces from the panels in A on an expanded time scale. C, average data from experiments like those shown in A; 10 mM DTT was applied for the time indicated. D, summary bar graph for experiments such as those shown in C for the channels as indicated. E, I 1 and I 2 reversal potentials measured in NMDG ϩ solutions for V48C mutants. F, same as described for E but for I328C mutants. G, same as in E but for channels formed by co expressing V45C and I328C mutants. H shows experiments identical to those shown in G but after 5-min application of DTT (10 mM) to the bathing medium. The "true I 1 E rev " is the average measured for V48C, I328C, and V48C and I328C channels without DTT. I and J, summary data for experiments like those in G and H.
absence of DTT at ϩ20.0 Ϯ 1.7 mV (n ϭ 6 and 10; p Ͻ 0.05). The data favor the view that a disulfide forms between V48C and I328C leading to a normal I 1 but reduced I 2 state. DTT reduces this disulfide leading to a restored I 2 state. Presumably V48C/ I328C channels in the absence of DTT undergo restricted permeability changes because the permissive rearrangements between TM1 and TM2 are impaired. Our data indicate that permeability changes in P2X 2 channels can be impaired by perturbing the interface between TM1 and TM2 by 1) Ala mutants along the interface for TM1 and TM2 (Table I) and 2) by a disulfide formed between residues that span the interface.
Channels That Lack I 2 Tend to Desensitize Rapidly-We recently suggested that the P2X 2 I 2 state gives rise to currents that desensitize slowly over 10 s, whereas channels that lack this state desensitize somewhat more rapidly in physiological solutions (34). This supports previous work on P2X 4 and P2X 7 channels, which also display permeability changes, and in Na ϩ solutions show biphasic or slowly decaying currents in response to ATP (2,16,26,56). Collectively, these data imply that one physiological correlate of the I 2 state may be enhanced ion flow reflected as slowly desensitizing channels. We tested this hypothesis in the present study by comparing desensitization kinetics for all those mutants that lacked the I 2 state and were expressed at levels equivalent to wt P2X 2 (Fig. 6). Of these ten (F31A, R34A, Q37A, K53A, I328A, I332A, S340A, G342A, W350A, and L352A), eight displayed significantly faster desensitization than wt P2X 2 measured at 20 -30 s into the ATP application period (Fig. 6). The two outliers were Q37A and I328A; these lacked I 2 and expressed at normal levels in the membrane but showed desensitization kinetics similar to wt P2X 2 . Overall, together with previous reports of sustained and biphasic currents (2,16,56), these data support the hypothesis that one physiological correlate of the I 2 state is enhanced ion flow associated with slow desensitization.
Might the Ala mutants affect the pore directly and complicate our interpretations? There is no way to totally exclude this possibility, but if the Ala mutants do markedly affect the selectivity filter then one expects the relative permeability of another cation to be affected as well. We tested this prediction by examining the pCa 2ϩ /pNa ϩ of two mutants from each transmembrane domain (G31A, H33A, P329A, and L353A), as well as T18A, which we have shown previously lacks I 2 (34). There were no significant differences in pCa 2ϩ /pNa ϩ as compared with wt P2X 2 (Table II). These data provide reassurance that the selectivity filter in the Ala-substituted mutants retains its ability to select Ca 2ϩ over Na ϩ with a fidelity comparable with wt P2X 2 channels (15,42) and support the hypothesis that Ala mutants produce their perturbing effects by affecting channel gating (40). DISCUSSION The main finding of the present study is that both transmembrane domains contribute to permeability changes in P2X 2 channels and that the interface between transmembrane domains appears important for conformational rearrangements that allow opening to the I 2 state. P2X 2 channel permeability changes are known to depend on plasma membrane channel density (31). One interpretation, FIG. 6. Desensitization kinetics in physiological solutions of Ala mutants that lack I 2 . A, representative traces for ATP-evoked (100 M) inward currents from cells expressing wt P2X 2 , G342A, and F31A mutants, with ATP applied for the times indicated by the solid bars. B, summary bar graph for experiments such as those illustrated in A for Ala mutants that showed no I 2 state. C, relationship between the shift in NMDG ϩ reversal potential over time (Fig. 5) and the peak current measured in Na ϩ solutions. Note that two mutants displayed currents that were significantly smaller from the others when tested with Tukey's analysis of variance. These were Y43A and F44A. D, relationship between the shift in NMDG ϩ reversal potential over time and the percent desensitization measured in physiological solutions from experiments like those shown in A. Nine of 11 channels that lacked I 2 also showed significantly faster desensitization.
consistent with the cluster model for permeability changes (see Introduction), is that clusters of channels are needed in order for P2X 2 channels to undergo permeability changes. Presumably, these hypothetical clusters are either preformed or form in response to ATP. We attempted to test this view by imaging channels formed by CFP-or YFP-labeled channels (P2X 2 -C, P2X 2 -Y). This is a valid approach because the intrinsic fluorescence of the GFP variants provides a measure of channel location in living cells (57) and because tagged P2X 2 channels function in a manner identical to wt P2X 2 channels (18,58). By using high resolution TIRF microscopy we found no evidence for clusters of P2X 2 channels in the plasma membrane of HEK cells, whereas we could readily detect clusters of P2X 4 channels. The stationary P2X 4 channel clusters may represent points of capture from, or release of P2X 4 channels into, the membrane, whereas the mobile P2X 4 -Y clusters may represent vesicles located just underneath the membrane (58). Our experiments did not seek to discriminate between these possibilities, but they do serve to illustrate that the microscope used has sufficient resolution to detect clustered fluorescence had it occurred for P2X 2 -Y channels. Moreover, we found no evidence that P2X 2 channels move in response to ATP to form clusters in HEK cells. In contrast we could readily detect movement of PH-Y from the membrane to the cytosol. The images of P2X 2 -Y fluorescence support recent work indicating that these channels are evenly and stably expressed in the plasma membrane of HEK cells, due to a membrane stabilization sequence (46). Because P2X 2 -Y and wt P2X 2 channels undergo permeability changes in every HEK cell (Table I), and because we never detected P2X 2 -Y clusters before or during ATP, we conclude that clustering is not an obligatory requirement for permeability changes. The cluster model has its origins in pioneering work on ATP-evoked permeability changes in mast cells (30), which are likely due to the P2X 7 subunit (20). Permeability changes at P2X 7 channels display several features that are different to those recorded for P2X 2 channels (2). Most notable among these is the finding that the P2X 7 pore dilates to a size much larger than that formed by P2X 2 channels (26,33). However, the data from the present study along with previous work (23) suggests that clustering in not needed for P2X channels to undergo permeability changes.
Past work on the cytosolic domain has been interpreted to indicate that it changes conformation as permeability increases occur for P2X 2 channels (34). Notably, there was a decrease in a measure of FRET between the tips of the C tail domain, as though these were splayed during permeability changes. The decrease in FRET is inconsistent with the cluster model, because in the simplest case if channels clustered there would be an increase in FRET between labeled channels. The previous work was in accord with an intrinsic conformational change that allowed P2X 2 channel permeability changes to occur (34). If so, binding of ATP to the extracellular portions of the receptor must be communicated to the cytosolic domain, and we hypothesized the existence of regions that undergo I 2 statespecific conformational rearrangements (34). In the present study we focused on transmembrane segments to identify such regions in an attempt to reveal features consistent with the gating model for permeability changes. We selected the Ala mutants for I 2 state specificity (see "Results"), in an attempt to minimize direct effects on the pore. The impelling assumption is that the sensitivity of the I 2 state to mutations is informing about gating changes associated with this state (50). Remarkably, the I 2 state was far more readily perturbed by Ala substitutions than the I 1 state, suggesting that these mutants shift the equilibrium in favor of opening to the I 1 state, either by stabilization of this state or destabilization of the I 2 state. Based on the adage that there are more ways to destabilize rather than stabilize protein structures (50) we interpret these data to indicate that there is a greater propensity of mutants to destabilise the I 2 state and favor the I 1 state conformation.
With the exception of S340A, for both TM1 and TM2, the residues that abolished the I 2 state fell on one-half of ␣ helical wheel representations, although there were no contiguous stretches. Our experiments do not provide a precise understanding of Ser 340 because the Ala substitution may affect gating that is thought to occur as the pore dilates (33) or the selectivity filter itself given that S340A is known to affect the ability of P2X 2 channels to select for Ca 2ϩ ions (15,42). With the exception of this residue the Ala hits in TM1 and TM2 all fall on one-half of a helical wheel representation of TM1 and TM2. What do the differences between our work and that of Li et al. (40) on an Ala scan across TM1 and TM2 for measures of agonist potency tell us about the motions that occur in P2X channels? The previous work measured EC 50 values for agonist-evoked currents over a time course of 1-2 s (40). In contrast we measured permeability values for agonist applications of ϳ30 s. This latter time course provides information about the I 2 state, whereas measurements over 1-2 s are too brief to accurately reflect the I 2 state (33). Also, EC 50 values and permeability ratios by definition report distinct aspects of channel function and are not readily comparable. Furthermore, the final data set used in this study is for mutants that abolished the I 2 state and so a close correspondence with the work of Li et al. (40) is not expected. On the other hand if one compares all the Ala hits on I 2 (Table I) to the data of Li et al. (40) there is good agreement for TM1 and TM2 (Leu 41 , Gln 37 , Phe 44 , Ile 50 , Tyr 43 , Ile 328 , Leu 338 , Thr 339 , Ser 340 , Gly 342 , Gly 344 , and Ser 345 are common).
Insight into TM2 residues that contribute to the selectivity filter has been gained by examining the effect of mutations on relative ion permeability and flux (15,42). The most crucial TM2 residues in determining the ability of P2X 2 channels to select for Ca 2ϩ over Na ϩ were Thr 336 , Thr 339 , and Ser 340 , and these sit on the same half of an ␣ helical representation of TM2 (15). Assuming that the selectivity filter has to project into the pore these residues allow us to orientate TM2 from the perspective of the present experiments. Using this as a starting point, and the necessity for V48C and I328C to be close to one another (8,55), we mapped the I 2 state-specific hits onto these helical wheel representations. Our hypothetical view of the simplest possible arrangement that can describe the data presented in this study and elsewhere (8,15,42) is shown is Fig.  7 and in Fig. 7B if this unit is repeated around the central axis to form a channel with three subunits. This hypothetical view encompasses several features that are noteworthy. First, Val 48 and Ile 328 face each other across an interface, as would be expected if they formed a disulfide between subunits (8). Second, polar Thr 336 , Thr 339 , and Ser 340 residues face the central pore as would be expected if they were part of the selectivity filter (15,42). Third, many of the residues that face the central cavity have been identified previously in substituted cysteine accessibility mutagenesis experiments (11)(12)(13)(14). Fourth, a hydrophobic residue (Val 343 in P2X 2 ) always follows a conserved Gly (Gly 342 of P2X 2 ): in the present view Val 343 projects into the pore and is a good candidate to form a hydrophobic gate (35). Fifth, six helices line the pore, thus overcoming physical constraints on the pore diameter possible with three helices (32). Sixth, the view shows a trimeric assembly, consistent with the best available biochemical data (6,9,10). Seventh, the Ala hits appear to sit at the interface between subunits, which is consistent with a basic assumption of the approach, namely that the hydrophobic Ala affects protein-protein interaction surfaces (49,50). Further work on permeability changes is needed, but presumably the permeation pathway could dilate by ϳ3-4 Å (33) by helix tilting, rotation, or bending as is thought to be the case for other ion channels (35). Past and present data focus attention on a conserved glycine in TM2 (Gly 342 of P2X 2 ) as a point of possible local flexibility. Glycines are known to occur in high-resolution models of potassium channels at regions of presumed flexibility (35). In the case of mechanosensitive channels data indicate that the pore can dilate from ϳ2 to 34 Å by the tilting of five pore lining helices (35,59,60). In this case an intrinsic conformational change, driven by changes in membrane tension, can drive dilation of the pore. It is tempting to speculate that relatively more subtle tilting motions could allow the P2X 2 pore to dilate by ϳ3 Å for entry to the I 2 state. In relation to other transmitter-gated ion channels, such as the Cys loop and glutamate-gated family, our understanding of P2X channels is somewhat limited. For example, although progress has been made, key aspects such as the precise determinants of the ATP binding site and pore remain unclear. Another unresolved question is how the pore of some P2X channels undergoes rearrangements to switch its preference between ions. The present experiments indicate that important motions likely occur at the interface between neighboring subunits and offer a schematized/hypothetical view of the pore and thus several hypotheses that can now be tested with direct structural methods and optical approaches that detect motions of single amino acids and domains (34,38,44). FIG. 7. Hypothetical view of how the P2X 2 pore may form. A, an ␣ helical TM1 domain of one P2X 2 subunit is shown near the TM2 domain of a neighbor. The residues highlighted in red are I 2 statespecific Ala hits, whereas those with a thick green edge are determinants of the selectivity filter. B, by arranging three views of A around a central pore one can obtain an idea of how the P2X 2 pore may form. In this view, the residues that form the selectivity filter project centrally into the pore, TM1 and TM2 are appropriately positioned to line the pore, Val 45 and Ile 328 are close, and the Ala hits identified in this study sit at the interface between neighboring subunits. This hypothetical view is the simplest one that can describe all of the data (see "Discussion").