The Effects of β3 Subunit Incorporation on the Pharmacology and Single Channel Properties of Oocyte-expressed Human α3β4 Neuronal Nicotinic Receptors*

We compared the main properties of human recombinant α3β4β3 neuronal nicotinic receptors with those of α3β4 receptors, expressed in Xenopus oocytes. β3 incorporation decreased the channel mean open time (from 5.61 to 1.14 ms, after approximate correction for missed gaps) and burst length. There was also an increase in single channel slope conductance from 28.8 picosiemens (α3β4) to 46.7 picosiemens (α3β4β3; in low divalent external solution). On the other hand, the calcium permeability (determined by a reversal potential method in chloride-depleted oocytes) and the pharmacological properties of β3-containing receptors differed little from those of α3β4. The main pharmacological difference in α3β4β3 “triplet” receptors was a 3-fold decrease in the potency of lobeline relative to acetylcholine. Nevertheless, there was no change in the rank order of potency for agonists (epibatidine >> lobeline > cytisine, 1,1-dimethyl-4-phenylpiperazinium iodide, nicotine > acetylcholine > carbachol for both receptors; measured at low agonist concentrations). Sensitivity to the competitive antagonists trimetaphan (0.2–1 μm) and dihydro-β-erythroidine (30 μm) was similar for the two combinations, with a Schild KB for trimetaphan of 76 and 66 nm on α3β4 and α3β4β3, respectively. The change in single channel conductance confirms that β3 replaces a β4 subunit in the pentamer. The absence of pronounced differences in the pharmacological profile of the triplet receptor argues against a role for the β3 subunit in the formation of agonist binding sites, whereas the changes in channel kinetics suggest an important effect on receptor gating. The shortening of the burst length of β3-containing receptors implies that any synaptic currents mediated by such channels would have faster decay kinetics.

Among neuronal nicotinic subunits, ␤3 was long considered an "orphan" as it does not form functional recombinant receptors if expressed as a classical heteromeric combination, i.e.
together with either an ␣ or a ␤ subunit. Because of the sequence similarity between ␤3 and ␣5 we tested the hypothesis that, just like ␣5, ␤3 would form functional receptors only if co-expressed together with both an ␣ and a ␤ subunit. By a reporter mutation strategy, we showed that ␤3 is indeed incorporated into functional recombinant ␣3␤4␤3 receptors (1). In ␣3␤4␤3 receptors, ␤3 is present as a single copy, which replaces one of the ␤ subunits (versus two copies each for the ␣3 and ␤4 subunits (2)); note that ␤3 coassembles with other ␤ subunits in rat native cerebellar nAChRs 1 (3).
The question now arises of whether the ␤3 subunit can change the properties of neuronal nicotinic receptors. There are several reasons for investigating this problem. First of all it is important to establish whether any such change introduced by the presence of ␤3 affects the role that these receptors may have in physiological processes in the central nervous system or in the pharmacology of tobacco addiction. Furthermore, if ␤3-containing receptors have distinctive biophysical or pharmacological properties, such receptors can in principle be recognized in native tissue by functional assays. Finally, clear changes in the receptor pharmacology and particularly in the binding affinity of competitive antagonists would be a strong indication that the binding sites of the receptor have changed and that ␤3 directly forms one of the two interface binding sites.
The shorter openings and bursts together with the increased single-channel conductance observed with the co-expression of ␤3 confirm that ␤3 is incorporated into the receptor complex. The single-channel conductance change and the stoichiometry of the receptor (2) suggest that ␤3 takes the place of a classical ␤ subunit. The calcium permeability of the receptor and its sensitivity to a range of agonists and antagonists were unchanged apart from a decrease in the relative potency of the agonist lobeline. These results suggest that ␤3 either does not participate in the formation of the agonist binding site or that the sequences of ␤3 and ␤4 in the relevant domains are too similar to allow them to be differentiated.
The extent and nature of changes introduced into the nAChR by the presence of ␤3 were not sufficient to provide tools for the identification of such receptors in native tissue. Nevertheless, the shorter burst length of these receptors implies a faster decay of any synaptic current that these channels may mediate and suggests a functional role for the ␤3 subunit in the diverse family of neuronal nicotinic receptor subunits.

EXPERIMENTAL PROCEDURES
Expression of Nicotinic Subunits cRNA in the Xenopus oocyte-cDNAs for the human ␣3, ␤3, and ␤4 (GenBank TM accession numbers Y08418, Y08417, and Y08416, respectively) containing only coding sequences and an added Kozak consensus sequence (GCCACC) immediately upstream of the start codon (4) were subcloned into the pSP64GL vector, which contains 5Ј-and 3Ј-untranslated Xenopus ␤-globin regions (5). All cDNA/pSP64GL plasmids were linearized immediately downstream of the 3Ј-untranslated ␤-globin sequence, and cRNA was transcribed using the SP6 Mmessage Mmachine Kit (Ambion). The quality and quantity of cRNAs were checked by gel electrophoresis and comparison with RNA concentration and size markers.
Female Xenopus laevis frogs were anesthetized by immersion in neutralized ethyl m-aminobenzoate solution (tricaine, methane sulfonate salt; 0.2% solution weight/volume) and killed by decapitation and destruction of the brain and spinal cord (in accordance to Home Office guidelines) before removal of the ovarian lobes. Clumps of stage V-VI oocytes were dissected in a sterile modified Barth's solution with a composition of 88 mM NaCl, 1 mM KCl, 0.82 mM MgCl 2 , 0.77 mM CaCl 2 , 2.4 mM NaHCO 3 , 15 mM Tris-HCl in high performance liquid chromatography-grade water with 50 units ml Ϫ1 penicillin and 50 g ml Ϫ1 streptomycin (Invitrogen), pH 7.4, adjusted with NaOH. The dissected oocytes were treated with collagenase (type IA, Sigma; 65 min at 18°C, 245 collagen digestion units ml Ϫ1 in Barth's solution, 10 -12 oocytes/ ml), rinsed, stored at 4°C overnight, and manually defolliculated the following day before cRNA injection (46 nl/oocyte). The oocytes were incubated for ϳ60 h at 18°C in Barth's solution containing 5% heatinactivated horse serum (Invitrogen) (6) and then stored at 4°C. Experiments were carried out at a room temperature of 18 -20°C between 2.5 and 14 days from injection.
cRNA was injected at a ratio of 1:1 in order to express ␣3␤4 receptors and at a ratio of 1:1:20 (␣3:␤4:␤3) in order to express ␣3␤4␤3 receptors in conditions that minimized the presence of pair ␣3␤4 receptors. We have previously found that with this ratio the proportion of current through ␣3␤4 receptors is too small to be detected by fitting doseresponse curves, since it is 39% when the ratio for ␣3:␤4:␤3 is 1:1:1 (1). The total amount of cRNA to be injected (in 46 nl of RNase-free water) for each combination was determined empirically, with the aim of achieving an optimal signal to noise ratio in the different experiments. Given that most of the experiments described here were carried out at low agonist concentrations, the level of expression we tried to achieve was higher than that needed for full agonist dose-response curves. The average quantity of cRNA injected in each oocyte was 20 and 30 ng for ␣3␤4 and ␣3␤4␤3, respectively.
Single Channel Recording-Recordings were obtained in the cellattached configuration from oocytes that had been stripped of their vitelline membrane after incubation in hyperosmotic solution (10 min with 200 mM sodium methyl sulfate, 20 mM KCl, 1 mM MgCl 2 , 10 mM HEPES, pH 7.4, with NaOH). Electrodes were pulled from thick-walled borosilicate glass (GC150F, Warner Instruments) coated with Sylgard® 184 (Dow Corning) and fire-polished to have a final resistance of 12-15 megaohms when filled with low divalent external solution (150 mM NaCl, 2.8 mM KCl, 0.5 mM MgCl 2 , 0.5 mM CaCl 2 , 10 mM HEPES, 0.5 M atropine , pH 7.2, with NaOH) which contained 1 M ACh as agonist. No channel openings other than those attributable to stretch channels were observed in the absence of ACh in the extracellular solution (7). Oocytes were immersed in high potassium solution (100 mM KCl, 10 mM HEPES, 10 mM EGTA, pH 7.2, with KOH) to drive the resting membrane potential of the oocyte to a value consistent and close to 0. Recordings were filtered at 3 kHz (8-pole Bessel filter) and digitized at 30 kHz. The recordings used for amplitude analysis were idealized by time-course fitting; fitted amplitude distributions were fitted with gaussian curves (SCAN and EKDIST, courtesy of D. Colquhoun, UCL; www.ucl.ac.uk/Pharmacology/dcpr95.html) to give a mean value for the amplitudes of the openings at 150-, 125-, 100-, 75-, and 50-mV holding potentials; the mean values from each patch (4 and 3 patches for ␣3␤4 and ␣3␤4␤3, respectively) were pooled for the different holding potentials to give the points in Fig. 1C, which were fitted with a straight line to obtain slope conductance values (CVFIT, courtesy of D. Colquhoun, UCL). Errors were calculated from the residuals.
Recordings at a holding potential of 100 mV were used for obtaining dwell time distributions and idealized by a threshold crossing method with a resolution of 80 s. Open period and shut time distributions were fitted with a mixture of exponential components (PClamp, Axon Instruments). An approximate correction of the mean open time for missed gaps was carried out as previously detailed (8).
Bursts were defined as groups of openings separated by shut times which were shorter than a particular value, t crit . We chose the t crit value by inspecting shut time distributions and the consistency of the time constants of their components across patches, since the time constants of gaps between bursts are expected to vary across patches because they depend on both the agonist concentration and the number of channels in a patch. The t crit value was on average 168 ms for ␣3␤4 and 5.6 ms for ␣3␤4␤3. Two-electrode Voltage Clamp Recording Conditions-Oocytes, held in a 0.2-ml bath, were perfused at 4.5 ml/min with modified Ringer solution (150 mM NaCl, 2.8 mM KCl, 10 mM HEPES, 2 mM MgCl 2 , 0.5 M atropine sulfate, pH 7.2 adjusted with NaOH) and voltage-clamped at Ϫ70 mV using the two-electrode clamp mode of an Axoclamp-2B amplifier (Axon Instruments). Electrodes were pulled from Clark borosilicate glass GC150TF (Warner Instrument Corp.) and filled with 3 M KCl. The electrode resistance was 0.5-1 megaohms on the current-passing side. Experiments were terminated if the total holding current exceeded 2 A to reduce the effect of series resistance errors. We chose a nominally calcium-free solution to minimize the contribution of calciumgated chloride conductance; this is endogenous to the Xenopus oocyte and may be activated by calcium entry through the neuronal nicotinic channels (9).
Agonist solutions (freshly prepared from frozen stock aliquots) were applied via the bath perfusion for a period sufficient to obtain a stable plateau response (at low concentrations) or the beginning of a sag after a peak (at the higher concentrations); the resulting inward current was recorded on a flat bed chart recorder (Kipp & Zonen) for later analysis. An interval of 5 min was allowed between ACh applications because this was found to be sufficient to ensure reproducible responses. To compensate for possible decreases in agonist sensitivity throughout the experiment, a standard concentration of ACh (approximately EC 20 for the particular combination used) was applied every third response. The experiment was started only after checking that this standard concentration gave reproducible responses. The average changes in the response to this ACh standard concentration observed by the end of the experiment for the different combinations ranged from Ϫ58 to ϩ60%. All the data are compensated for the response rundown. However, applying this compensation did not affect the conclusions of our work, because the results of analyzing the original data (without compensation) were similar (data not shown). For details specific to each type of experiment, see below.
Agonist Potency Ratios-These were obtained following a previously described protocol (10). The aim of this was to obtain partial (2-or 3-point) dose-response curves at the lowest agonist concentrations that gave an acceptable signal to noise ratio. Care was taken to match the response size for all the agonists tested in a given experiment. These conditions were chosen to reduce the contributions of desensitization and agonist self block and essentially ensured that the slope of the dose-response curves was similar for all the agonists (see the examples in Fig. 3 and 4). Given the large number of agonists tested, only a subset could normally be tested in one cell, but ACh was tested in every experiment and was used as a standard. All curves obtained in each experiment were fitted simultaneously by least squares (CVFIT, courtesy of David Colquhoun, UCL; see www.ucl.ac.uk/Pharmacology/ dcpr95.html) with power functions constrained to be parallel (these are the equivalent of fitting with a Hill equation in which the maximum is constrained to be very large with respect to the measured responses). In these simultaneous fits the program estimated the distances between the dose-response curves of the different agonists and that of the standard ACh. Such distances were obtained as concentration ratios with respect to ACh. The potency ratios shown in Table II are the reciprocal of the concentration ratios; errors and confidence intervals for potency ratios were calculated by Fieller's theorem (11). To check that the constraint of parallelism was justified, data were fitted without it to obtain estimates of the Hill slope for each agonist (see Table II). The maximum agonist concentrations used were 0.03 M for epibatidine, 1.5 M for lobeline, 5 M for cytisine, 10 M for 1,1-dimethyl-4-phenylpiperazinium iodide (DMPP), 20 M for ACh and nicotine, and 200 M for carbachol. Note that at 20 M, ACh produces less than 5% of its maximum response for both combinations tested.
Antagonist Experiments-The antagonist experiments were carried out to confirm the competitive nature of the antagonist action and to estimate the antagonist dissociation constant by the Schild method. Partial (two-point) concentration-response curves were obtained at low agonist concentrations first in control solution (modified Ringer ϩ 1 mM calcium) and then in the presence of an appropriate antagonist concentration; this was present both in the Ringer and in the agonist solutions. Oocytes were incubated with the antagonist for 25 min before responses to ACh were again tested. Only one antagonist concentration was tested in each oocyte in order to minimize the distortion by response rundown (which cannot be quantified in the presence of the antagonist). Only data from oocytes in which the agonist response recovered after antagonist treatment to at least 60% that of control were included in the analysis. In a manner similar to that employed for the agonist potency ratios, the dose ratio r produced by the antagonist was measured for each experiment by fitting to the partial concentration-response curves power functions constrained to be parallel. Unconstrained fits were also performed; there was no significant difference between the slope of the concentration response in control conditions (range 1.59 -1.71) and in 0.2-1 M trimetaphan (range 1.37-1.73, paired t test).
To check that the data were adequately fitted by the Schild equation, data were plotted as a Schild plot and fitted by a power function, r Ϫ 1ϭ a[B] n , where [B] is the concentration of antagonist, and a is a constant, in order to estimate the Schild slope n. Having found n to be close to 1 (see Table III), we refitted the data with the Schild equation, r Ϫ 1ϭ [B]/K B , to obtain K B , the antagonist dissociation equilibrium constant.
Calcium Permeability-We measured the effect of different calcium concentrations on the reversal potential of the current induced by 2-s voltage ramps from Ϫ40 to 40 mV holding potential. Such ramps were applied in the depolarizing direction and in the hyperpolarizing direction both in control conditions (leak) and in the presence of ACh (test) once the agonist current was steady. Current-voltage curves were averaged (n ϭ 4), and leak was subtracted using PClamp (Axon Instruments). Responses were sampled at 2 kHz (Axon Digidata 1320), filtered at 500 Hz (8-pole Bessel filter), and stored on disk for later analysis.
The contribution of the oocyte endogenous calcium-dependent chloride conductance (which would be activated by calcium entry through nicotinic channels) was minimized by preincubating the oocytes in chloride-free low calcium Ringer solution for 24 h (115 mM NaOH, 35 mM sucrose, 2.5 mM potassium gluconate, 10 mM HEPES, 1.8 mM Ca(OH) 2 , 0.5 M atropine bromide, pH 7.2, with methanesulfonic acid). Experiments were also carried out in chloride-free modified Ringer's solution. The composition of the low calcium Ringer was 1.8 mM Ca(OH) 2 , 35 mM sucrose, 115 mM NaOH, 2.5 mM potassium gluconate, 10 mM HEPES, 0.5 M atropine bromide, pH 7.2, with methanesulfonic acid; the high calcium Ringer contained 18 mM Ca(OH) 2 and no sucrose (osmolarity was 263 and 261 mosmol/liter, respectively). All recordings were carried out with electrodes filled with 2.5 M potassium acetate and 10 mM KCl using a 3 M KCl agar bridge as a reference electrode.

RESULTS
Single Channel Properties-As shown by the cell-attached records in Fig. 1, A and B, co-expression of the ␤3 subunit had a profound effect on the properties of oocyte-expressed ␣3␤4 neuronal nicotinic receptors. The traces (obtained in the presence of 1 M ACh at Ϫ100 mV) show that the openings of ␤3-containing receptors were larger and occurred in much shorter bursts (see also the expanded sweeps to the right of the continuous records).  (12,13). The slope conductance of these channels (measured by fitting current-voltage plots in Fig. 1C) was 28.8 Ϯ 0.66 picosiemens (4 patches) for the ␣3␤4 combination and 46.7 Ϯ 1.76 picosiemens for ␣3␤4␤3 receptors (3 patches).
The most pronounced effect of the presence of the ␤3 subunit in the receptor complex was on the kinetics of the channel. The continuous recordings of Fig. 1 show clearly that most of the openings of ␣3␤4 channels occur in very prolonged bursts. These long bursts were completely absent from recordings from ␣3␤4␤3 channels. Fig. 2 shows dwell time distributions for representative patches for the two subunit combinations (see Table I). ␤3 incorporation produced a consistent shift toward shorter apparent open periods (9.4 to 1.8 ms). Thus, the major component in the apparent open period distribution was 3.3 ms for ␤3containing receptors (45% of all openings) and 17 ms for ␣3␤4 (38% of all openings). Shut time distributions were all fitted with a mixture of five exponential components; the shortest components (the first 4 for ␣3␤4 and the first 2 for ␣3␤4␤3) were consistent from patch to patch (Table I) and were, therefore, classed as within burst (see "Experimental Procedures"). The faster shut time component is shorter in the ␣3␤4 patches, and the majority of events in this component will be below experimental resolution; this would result in the apparent lengthening of ␣3␤4 openings. Given that it is likely that few openings are missed at our experimental resolution of 80 s, we applied an approximate correction for missed gaps (8); this did not change the effect of ␤3 incorporation because corrected open times were 5.6 and 1.1 ms in ␣3␤4 and ␣3␤4␤3, respectively (Table I).
Measurements of average burst length confirmed the substantial shortening that is so striking from visual inspection of the traces. Thus mean burst length was 270 and 5.6 ms for ␣3␤4 and ␣3␤4␤3 receptors, respectively (Table I). As Fig. 1 shows, a potential complication is that there are signs of heterogeneity in ␣3␤4 bursts. We therefore considered the possibility that the population of shorter bursts was similar to the ␣3␤4␤3 bursts and simply became predominant with ␤3 incorporation. To assess this hypothesis, we measured mean burst length in ␣3␤4 patches after excluding long bursts (the threshold for exclusion was chosen by inspection and was 274 ms on average in the 10 patches analyzed); the average duration for the "shorter" bursts was 64 Ϯ 14 ms (n ϭ 10) and, therefore, still substantially longer than that of ␣3␤4␤3 bursts.
Agonist Potency-There were no obvious differences between ␣3␤4 and ␣3␤4␤3 receptors in the time course of agonist responses (see Fig. 3A) or the rank order of potency for an extensive series of agonists (epibatidine Ͼ Ͼ lobeline Ͼ cytisine, DMPP, nicotine Ͼ ACh Ͼ carbachol; Table II, Fig. 3B). This rank order of potency is very similar to that found for human or rat ␣3␤4 receptors by Chavez-Noriega et al. (14) and Meyer et al. (15) by EC 50 comparisons in oocytes or HEK293 cells.
Nevertheless, there was a marked decrease in the potency of lobeline with the incorporation of ␤3, from 23.0 Ϯ 3.70 to 7.14 Ϯ 1.1, relative to ACh for the ␣3␤4 and ␣3␤4␤3 receptors, respectively; see the different distances between the partial doseresponse curve to lobeline (leftmost in Fig. 4B, filled circles) and that to the standard, ACh (filled squares).
The Hill slopes for each agonist were similar on the two different combinations (see Figs. 3B and 4B, Table II, t test). A fairly wide range of Hill slope values is reported in the literature for the same ␣3␤4 receptor from values comparable with those observed here (10) to values in excess of 1.9 (14). Note that the latter were observed in the presence of low (0.18 mM) Antagonist Potency-Trimetaphan acts as a competitive antagonist on native ganglion nAChRs (16), which are likely to contain the ␣3 and ␤4 subunits. A measurement (by Schild analysis) of the dissociation constant of trimetaphan would be useful for receptor classification and binding site modeling (very few true equilibrium dissociation constants are available in the neuronal nicotinic literature) and might show whether the antagonist site (which overlaps with the ACh site) is changed by ␤3 (see "Discussion").
In both receptor combinations trimetaphan (0.2-1 M) produced a parallel rightward shift in the ACh partial concentration-response curves; its competitive mode of action was confirmed by the slope of the Schild plot (close to 1; see Table III). The incorporation of the ␤3 subunits had only a slight effect on the K B for trimetaphan, decreasing it from 75.5 Ϯ 1.8 to 66.0 Ϯ 1.7 nM (Table III).
Although not very potent on ␣3␤4-type receptors, the alkaloid dihydro-␤-erythroidine is likely to be competitive (17). In agreement with this, we observed a small, parallel shift in the ACh concentration-response curve in the presence of 30 M dihydro-␤-erythroidine, with similar dose ratios for ␣3␤4 and ␣3␤4␤3 (3.3 versus 3.9, n ϭ 2), again suggesting that there is no change in the antagonist binding site.
On ganglion-type nAChRs, mecamylamine is a channel blocker at micromolar concentrations (18) but may act competitively at nanomolar concentrations (16). In our experiments, the mecamylamine concentration needed to produce a significant antagonist action was high (1 M) and had substantial channel blocking effects (seen as a reduction in the slope of the ACh concentration-response curve); again, the effect was very similar for ␣3␤4 and ␣3␤4␤3 receptors (data not shown).
Calcium Permeability-This was similar for the two combinations. The reversal potential shift after a 10-fold increase in extracellular calcium was 7.7 Ϯ 1.5 and 5.2 Ϯ 0.8 mV for ␣3␤4 and ␣3␤4␤3, respectively (Table IV; see the value of 6.1 reported for ␣3␤4 receptors (19). DISCUSSION Our results show that the ␤3 subunit consistently incorporates into neuronal nicotinic ␣3␤4 receptors and profoundly changes their single channel properties, markedly shortening open periods and bursts and increasing single channel conductance. On the other hand, the presence of ␤3 had only small or negligible effects on the receptor sensitivity to a series of nicotinic agonists or to the competitive antagonists trimetaphan and dihydro-␤-erythroidine and on the channel permeability to calcium. None of these changes is likely to be useful in the identification of native ␤3-containing receptors; the decrease in lobeline potency is small, and the increase in single channel conductance is not helpful because neuronal nAChRs display a large range of conductances, usually with several overlapping levels for each combination (13).
The main difference observed in the single-channel records of ␤3-containing receptors was a decrease in the duration of apparent open times and burst length. In an ideal record, i.e. with perfect resolution, the duration of open times reflects only the value of the closing rate constant, ␣, and therefore, the gating properties of the receptor. The open times measured from a real TABLE I Analysis of dwell time distributions for cell-attached patches from oocytes expressing ␣3␤4 or ␣3␤4␤3 Data from a total of 10 ␣3␤4 patches and 5 ␣3␤4␤3 patches are shown. Note that the areas of the different components were averaged for the total number of patches (i.e. including values of 0 for patches that did not require that particular component in the fitting of the distribution).

␣3␤4 ␣3␤4␤3
Apparent open periods 1 (ms) 0.10 Ϯ 0.02 0.08 Ϯ 0.02 Burst length Mean burst length (ms) 270.5 Ϯ 53.7 5.6 Ϯ 0.6 (n) (10) (5) recording are, however, distorted by our finite temporal resolution, as undetected short gaps lengthen apparent openings. In our case, because it is likely that we missed only shuttings, we could apply an approximate correction for missed events to the apparent open times. This did not change much the effect of ␤3 incorporation. A potentially important consequence of the decrease in burst length we observed is that synaptic currents mediated by such channels would decay faster than those mediated by ␣3␤4 channels. A caution is that long bursts may be a feature only of recombinant ␣3␤4 receptors, particularly in the oocyte expression system (13,20). On the other hand, channel openings similar to recombinant long bursts have been recorded from the medial habenula, which expresses ␣3 and ␤4 subunits (21), so  it is possible that incorporation of the ␤3 subunit does exert this effect in native channels. Another point of interest concerns the role of the ␤3 subunit in the receptor structure. Our previous work showed that ␣3␤4␤3 receptors contain two copies each of ␣3 and ␤4 and only one copy of ␤3 (2). Our single channel data are in good accord with this stoichiometry, as the increase in conductance with ␤3 incorporation strongly indicates that the ␤3 subunit takes the place of one of the ␤4 copies in the ␣3␤4 pentamer. In nicotinic receptors conductance is mostly determined by the net charge at each of three sets of residues, at positions Ϫ4Ј, Ϫ1Ј, and 20Ј of the TM2 (second transmembrane) domain, the cytoplasmic, intermediate, and outer rings of charges (22). At the cytoplasmic and intermediate rings, all the subunits we expressed in this study are similar and contribute a negative charge (aspartate or glutamate). However, at the outer ring, ␣3 and ␤3 have a negatively charged glutamate, whereas ␤4 has a positively charged lysine. Hence, ␤3 can only affect the rings of charges if it takes the place of a ␤4 subunit; in this case, the charge on the outer ring would become more negative by two units, explaining the substantial conductance increase we observed. For instance, it has been reported that changing lysine to glutamate in the 20Ј position of one of the three copies of ␤2 is sufficient to increase the chord conductance of ␣4␤2 nAChRs by ϳ35% in divalent-free solution and using potassium as the permeant ion (23).
Incorporation of the ␣5 subunit is also known to increase single channel conductance of neuronal nAChR (13,24,25), probably by a similar effect on the outer ring (note, however, that ␣5, but not ␤3, incorporation should also reduce by one unit the negative charge on the cytoplasmic ring). If ␤3 replaced an ␣ rather than a ␤ subunit, the only change in the residues important for conductance would be a conservative serine for threonine swap in position 2Ј, which in muscle nicotinic receptors produces a small increase in conductance, detectable only if potassium is the main permeant ion (22,26). However, knowing that ␤3 replaces a ␤4 subunit still leaves open the question if ␤3 could participate directly, as the ␤, or complementary (Ϫ) subunit, to one of the two agonist binding sites at the subunit interface.
Our antagonist data discount major changes in the binding sites given that the antagonist binding was not affected by ␤3 (judging from the results of our Schild analysis for trimetaphan and dihydro-␤-erythroidine). Of course minor changes in the agonist binding residues may not be resolvable by competitive antagonists because their binding site overlaps but does not coincide with that of the agonist (note also that the range of  4. The potency of lobeline relative to ACh is lower in receptors containing the ␤3 subunit. Traces (A) show the current responses to bath application of low concentrations of ACh or lobeline to ␣3␤4 (left) or ␣3␤4␤3 (right) receptors. Concentration-response curves from such traces are plotted in B. The curves refer to lobeline, ACh, and carbachol (left to right). Note that in the experiment on triplet ␣3␤4␤3 receptors (right) the lobeline curve (filled circles) is closer to that of ACh (filled squares) than it is in oocytes expressing ␣3␤4 (left). Lob, lobeline. competitive antagonists available is relatively small). This may well be the case for a ␤4 to ␤3 swap, as the sequences of ␤4 and ␤3 in loops D, E, and F are fairly similar; note, however, that the important residues in these domains are not as well characterized as those for loops that form the ␣-side of the interface (27). Additionally, ␤3 can in principle only change one of the binding sites, not both. The Schild method measures the dissociation constant of the highest affinity site, as antagonist occupancy of one site only is sufficient to block receptor activation (muscle nAChR with one agonist and one antagonist molecule bound have very low P open ; see Sine and Taylor (28)). Thus, any ␤3-induced decrease in the site affinity for the antagonist would not be detectable.
Our data from the agonist experiments also argue against major changes in binding sites; despite testing a wide range of nicotinic agonists, the only change we observed with ␤3 incorporation was a decrease in the relative potency of lobeline. In principle, differences in agonist potency can be due to changes in either the binding or gating steps, which cannot be distinguished at macroscopic level (29). The analysis of open periods provides strong evidence of a gating impairment because of the shorter open times observed in ␤3-containing receptors. A gating change seems, therefore, to be the most likely explanation of the effects of ␤3 incorporation. Further work is needed to clarify why, as we previously reported, in receptors containing ␤3 there is no change in the EC 50 of ACh or nicotine and in the maximum response (relative to ACh) of the partial agonist nicotine (1). Our data show that the receptor binding sites are not appreciably changed by ␤3 incorporation. Thus, either ␤3 plays the role of a "structural" subunit or its contribution to the binding site cannot be detected because the sequences of the relevant domains are too similar to those of ␤4. This issue may only be decided conclusively by a different strategy, such as expressing ␤3 after the introduction of mutations aimed at increasing binding site differences for antagonists. Designing such mutants could for instance exploit the evidence on the determinants of binding for dihydro-␤-erythroidine and ␣-conotoxin MII provided by Luetje and co-workers (30 -32).