We reasoned that, due to their unique properties, ion channels provide an extraordinary opportunity to investigate the effect of the protein microenvironment on the (time-averaged) pK_as of ionizable residues located within pore domains. In principle, individual protonation/deprotonation events, which would cause the net charge of the pore to oscillate by one unit, should be evident in a single-channel recording as current fluctuations between two levels, one corresponding to the ‘neutral’ pore, and the other one corresponding to the pore carrying the extra unit charge. For instance, in the case of a single basic residue (i.e., lysine, arginine, or histidine) engineered in a cation-selective pore, the ‘deprotonated channel’ is expected to have roughly the same conductance as the wild-type protein, whereas the channel with the protonated residue (i.e., positively-charged) is expected to conduct cations at a somewhat lower rate. Protonation/deprotonation events occurring while the channel is open should thus be manifest as oscillations of the current between two discrete values. The ratio between these two levels would be a measure of the proximity of the protonatable group to the pore’s long axis, and the kinetics of the current fluctuations would reflect directly the rates of proton transfer and the side-chain’s pK_a.

That these individual proton-transfer events can actually be detected is, of course, not a foregone conclusion. The rates of proton binding and unbinding might be too fast, for example, exceeding the temporal resolution of existing single-channel recording techniques (a few microseconds under optimal conditions¹⁷). Alternatively, the extent to which a charge added to the pore affects the channel’s conductance might be too small for any proton-transfer event to be detected as a discrete step-change in the current. Even more elementarily, since buried charges can reduce the stability of folded proteins¹⁸, engineered ionizable side chains could prevent the correct folding and insertion of the protein into the membrane. The literature on ion channels, however, contains at least two cases for which this phenomenon has indeed been recorded at the single-channel level, namely L-type Ca²⁺ channels¹⁹, and cyclic nucleotide-gated (CNG) channels²⁰. In these two cases, the relevant ionizable groups corresponded to the naturally-occurring rings of four acidic residues present in the P-loop region of these ion channels. Although proton-transfer rates were measurable (these were exceedingly fast in aqueous solutions but were slowed down to measurable levels by changing the solvent to deuterium oxide), the protonation/deprotonation of the individual carboxylate/carboxylic groups could not be identified. Rather, the four acidic side chains seemed to somehow interact to form a single (Ca²⁺ channel) or two (CNG channel) proton-binding sites.

As a step toward understanding the effect of the microenvironment on the pK_a-values of membrane-protein residues, we systematically engineered protonatable side chains into the pore domain of the adult-type muscle nicotinic acetylcholine receptor (AChR), a member of the Cys-loop receptor superfamily of neurotransmitter-gated ion channels (Fig. 1a and Supplementary Fig. 1). This receptor offers a number of technical advantages for this sort of study, including a large conductance (~84 pS), the presence of a single copy of three of its four different subunits, and an invariant stoichiometry ((α₁)₂β₁δ). In this superfamily of pentameric receptor-channels, the pore is formed by the lateral association of five α-helical transmembrane segments (M2; Ref. ²¹), the rotation of which has been hypothesized to underlie the closed open transition^22–24. Acidic and basic residues flank the M2 segments of all the members of the Cys-loop superfamily where, in addition to having topogenic effects that are common to all transmembrane segments, they are critical determinants of charge-selectivity, conductance, and rectification^5,9,25. In marked contrast, protonatable residues are very rare within the membrane-spanning portion of M2 in these channels, being completely absent in the particular case of nicotinic receptors. Although protonation states for some of these naturally-occurring residues have been suggested on the basis of electrostatic-model calculations²⁶, the need for experimental approaches is obvious.

Figure 1 Protonation of lysines causes partial channel block

First, we introduced single lysines, one at a time, into a continuous stretch of 32 residues of the AChR’s δ subunit starting with Pro 250 (−7’ position; Fig. 1a), at the N-terminal end of the M1-M2 loop, and ending with Thr 281 (25’ position), close to the C-terminal end of the extracellular portion of the M2 helix. The naturally-occurring lysines occupying positions 0’ and 20’ (Fig. 1a) were mutated to alanine, whereas Asp -5’, Glu -1’, and Arg 21’ were mutated to both alanine and lysine. All constructs were successfully expressed in HEK-293 cells, and the current-blocking effect of single protonated lysines was indeed observed, and analyzed in detail. Figure 1b-d summarizes the effect of the NH₂/NH₃⁺ group of lysine on the AChR’s single-channel conductance. Insofar as the extent of channel block (Fig. 1d) can be taken as an approximate measure of distance between the engineered charges and the pore’s long axis, this plot can be interpreted in structural terms. Thus, an uninterrupted α-helical pattern is apparent between positions 1’ and 17’ (perhaps even between −2’ and 18’), with the narrowest constriction being formed by positions −2’ and −1’, and with the electrostatic effect of positive unit charges on the permeating cations decreasing steeply outside the presumed membrane-spanning region. Figure 1d also suggests a particular rotational angle for the M2 α-helix in the open-channel conformation. Positions −2’, −1’, 2’, 6’, 9’, 13’, and 16’ appear to form the lumen-facing side of the α-helix in the open state, as lysine substitutions have their greatest blocking effect when engineered at these locations. The orientation of the other residues can be similarly inferred from this plot. The secondary structure of M2 beyond the stretch between, approximately, −4’ and 18’ could not be ascertained because the blocking effect of engineered charges becomes vanishingly small.

On the basis of whether protonation/deprotonation events of the introduced single lysines could be observed as fluctuations of the open-channel current between two levels of different conductance (Figs 1b and 2a), all probed positions were classified in three groups: a) the open-channel current appears to be fixed at a lower-than-wild-type conductance level at all pH-values tested (6.0, 7.4, and 9.0), with excursions to a higher-conductance level becoming somewhat more apparent, yet remaining marginally quantifiable, as the pH is raised, b) the open-channel current fluctuates between a wild-type-like (‘main’) and a lower-conductance (‘sub-’) level in a pH-dependent manner, with the occupancy of the main level increasing as the pH increases, and c) the open-channel current appears to remain fixed at, or very close to, the wild-type level at all pH-values tested (6.0, 7.4, and 9.0). Positions −2’, −1’, 2’, 6’, 9’, 13’, 16’, 17’, 18’, 19’, and 20’ fall into the first group (circles in Fig. 1d). The reduced conductance and the failure to detect open-channel current fluctuations are consistent with the introduced lysines being predominantly protonated, whereas the pH insensitivity in the 6.0 – 9.0 pH range (Fig. 3a) suggests that the pK_a-values are high, as high or perhaps even higher than that of the lysine side chain in bulk water (~10.4). Hence, precise pK_a-values could not be estimated at these positions using the NH₂ group of lysine as a probe. Positions −4’, 1’, 3’, 5’, 7’, 8’, 10’, 11’, 12’, 14’, and 15’ fall into the second group (triangles in Fig. 1d). The observed pH-dependence of the main-level sublevel current fluctuations (Fig. 3c) is compelling evidence that this phenomenon results from the protonation/deprotonation of individual lysine side chains. The mere fact that proton-transfer events involving lysines (pK_{a, bulk} 10.4) can be observed indicates that the local microenvironment around these positions lowers the pK_a of the NH₂ group, bringing it closer to the 6.0 – 9.0 pH range used in the experiments. From the estimated rates of proton association and dissociation, we calculated the side-chain pK_a-values at all these locations (Fig. 2 and Table 1; see also Supplementary Figs 2 and 3c, d, and Supplementary Methods), with the exception of positions −4’, 5’, and 15’, in which cases the current fluctuations were difficult to resolve. Positions −7’, −6’, −5’, 0’, 4’, and the continuous stretch from 21’ to 25’ fall into the third group (squares in Fig. 1d). A wild-type-like conductance and the absence of open-channel current fluctuations are consistent with a very acidic pK_a for the lysine side chain, such that the probability of being protonated, and hence of blocking the cation current, is negligibly small. However, this behavior is also consistent with an orientation (e.g., large distance from the pore’s axis) and/or a local microenvironment (e.g., water with bulk-like properties) that markedly screen the effect of the NH₃⁺ charge on the passing current, such that protonation/deprotonation events would go undetected even if they occurred. These two quite different scenarios, namely, a highly-downshifted pK_a or an attenuated electrostatic effect, cannot be distinguished easily on the basis of our data. We favor the notion that the latter might be the case for positions −7’, −6’, and −5’, located on the distal end of the cytoplasmic M1–M2 linker, and for the stretch between 21’ and 25’ located on the portion of M2 that protrudes above the membrane into the extracellular space²³. The situation of positions 0’ and 4’ is less obvious, though. If one assumed that the periodicity of the blocking effect (Fig. 1d) should be maintained without interruption throughout the α-helical segment, then a highly-depressed pK_a-value can explain the failure of these two lysines to block the current. If this were the case, then the red-square symbols in Fig. 1d show the approximate extent of channel block that protonated lysines at 0’ and 4’ would cause. On the other hand, we cannot dismiss the possibility that, due to the long (~8 Å) and flexible side chain of lysine, the 0’ and 4’ NH₂ groups reach into the bulk-like intracellular water and become protonated, in a process akin to the ‘snorkeling’ of lysines and arginines located at the ends of lipid-embedded α-helices²⁷. The screening of the NH₃⁺ charge by the highly-polarizable environment would, thus, account for the lack of observable block. However, the finding that lysines at positions 1’ and 3’ do not appear to ‘snorkel’ (i.e., their extent-of-block values are not zero; Fig. 1d), even when residing in rather hydrophobic microenvironments (Fig. 2b, c and Table 1), reduces the likelihood of the ‘snorkeling’ scenario. This seems particularly true for the 4’position, which is located more deeply into the membrane. Clearly, more experiments and electrostatic calculations are needed to settle this point.

Figure 2 Experimental estimates of pKa-values

Figure 3 pH-dependence of proton-transfer reactions

Table 1 Position-dependence of rates of proton transfer and pKas at −100 mV

To extend our pK_a measurements to locations where the introduced lysine side chains appeared to be permanently protonated, we engineered histidines at positions −2’, −1’, 2’, 6’, 9’, 13’, 16’, and 17’ (positions 18’, 19’, and 20’ were not further investigated because the small extent of block (Fig. 1d) renders the analysis unreliable). Since the ‘macroscopic’ pK_a of the imidazole ring of histidine in bulk solution is ~6.4, the imidazole imidazolium⁺ interconversion should be detected at these ‘bulk-like’ positions in the 6.0 – 9.0 pH range. Indeed, pH-dependent main-level sublevel fluctuations (Fig. 3b) were recorded for all of these mutants, and the corresponding pK_a-values were calculated from the kinetics of proton transfer (Fig. 2 and Table 1; see also Supplementary Figs 2 and 3a, b, and Supplementary Methods). As qualitatively expected from the corresponding lysine substitutions, the imidazole pK_as are not depressed, consistent with these positions facing the aqueous lumen of the permeation pathway. In fact, at most of these positions, the pK_as are higher than the bulk value of 6.4, which indicates that it is actually easier to charge a histidine in this location within the pore than in bulk water. Provided this is also the case for the NH₂ group of lysine, this explains our, initially puzzling, observation that the subconductance levels of lysine constructs at these positions were oblivious to pH, even (as was the case for the 6’ and 9’ substitutions) when exposed to pH-values as high as 10.5. We also introduced histidines at −4’, 5’, and 15’, three positions where lysine-induced current fluctuations were observed, but were difficult to quantify reliably. The corresponding histidine-pK_as and rates of proton transfer are included in Table 1. Finally, we also engineered a histidine in 12’. The local microenvironment around this position lowers the affinity of the lysine side chain for protons by a factor of ~30 relative to that in bulk water (ΔpK_a 8.9 – 10.4 = −1.5; Fig. 2 and Table 1). Thus, to the extent that the side chains of histidine and lysine probe similar microenvironments, and that their pK_as have similar sensitivities to them, a histidine in 12’ is expected to be mostly deprotonated (pK_a 6.4 - 1.5 = 4.9), even at pH = 6.0. Indeed, our single-channel recordings clearly confirmed this prediction (Fig. 3d), and we generalize this result to suggest that histidines engineered at positions that lower the lysine’s pK_a to the same or greater degree than position 12’ (ΔpK_a < −1.5 units; see color code in Fig. 2b, c) would be largely neutral, even at pH = 6.0.

To gain further insight into the unique microenvironment around positions 0’ and 4’, we engineered arginines at these positions. As was the case for the 0’ and 4’ lysine constructs, arginine substitutions at these two positions failed to reduce the single-channel conductance. This result confirms the notion that charging the pore at these two positions might be energetically costly (see above), to the extent that not even the guanidine group of arginine, with a bulk pK_a of ~12.0, can bind a proton at pH = 6.0. We conclude that the pK_a-shift would be of, at least, −7 units (Fig. 2 and Table 1). For comparison, we also engineered an arginine in 11’ (on the same face of the α-helix as 0’ and 4’), a position where lysine is predominantly deprotonated (pK_a 5.3; Fig. 2), and where proton binding/unbinding events take place most slowly (Table 1). Although both arginine and lysine at this position block the current to similar extents (extent-of-channel-block 0.3), the 11’ arginine remained charged. Excursions to a deprotonated state could not be detected, not even at pH = 9.0. Thus, we suggest that, at physiological pH, engineered arginines would keep the proton bound at all positions of the stretch between −4’ and 17’ with the likely exception of positions 0’ and 4’ (see above discussion), in which cases the arginine would be completely deprotonated, in spite of its high proton affinity in bulk water.

Figure 2 and Table 1 (see also Supplementary Fig. 2) summarize the δ-subunit pK_a results. Similar analysis on the α, β, and subunits (data not shown) suggests that, although with some clear differences, this general pattern of ΔpK_a-values is conserved in all four subunits.

The different proton affinities of an ionizable residue in a protein and in bulk water are related, through thermodynamic cycles, to the changes in free energy associated with the transfer of the protonated versus deprotonated forms of the residue from bulk water to its position in the protein (ΔΔG°_{bulk → protein} −1.36 ΔpK_a kcal/mol, at 22º C). Hence, a ΔpK_a-value is a measure of the extent to which interactions with the protein microenvironment (water molecules, protein dipoles, charges on ionizable residues) compensate for the loss of solvation free-energy incurred upon removal of the residue’s charge from bulk water. Since naturally-occurring ionizable residues are only present at the ends of M2, and are absent from the membrane-spanning portions of M1, M3, and M4 (Supplementary Fig. 1a, b), the contribution of charge-charge interactions to this energetic balance is expected to be small for most positions examined here. Instead, most of the ΔpK_as reported in Fig. 2 and Table 1 are expected to be largely governed by the interactions of the positive charges on the engineered basic residues with protein dipoles (backbone dipoles, polar side chains) and water molecules in the heterogeneous environment surrounding M2. An inspection of the color-coded ΔpK_a maps in Fig. 2 suggests that the observed ‘gradient’ of pK_a-shifts specifically reflects the unique distribution of water molecules around the δ-subunit M2 α-helix.

In spite of significant steps toward the elucidation of the atomic-resolution structure of the full receptor²³, and toward an understanding of some general aspects of the chemical dynamics of the gating reaction^28–30, several basic aspects of the AChR still remain unsettled, such as the exact sides of the M2 α-helices that face the pore’s lumen in the closed and open states^23,24,31,32, the structural rearrangements underlying the closed open transition^23,24,33, and the locations and modes of operation of the activation^23,34 and desensitization gates. The experimental data presented here unambiguously defines the side of the δ M2 α-helix that faces the aqueous lumen of the pore in the open state (Figs 1d and 2b, c). Since our approach probes the effect of protonation/deprotonation on single-channel conductance, the proton-transfer rates in the closed channel (and, by inference, the corresponding orientation of the α-helices) were not investigated here. Other approaches, however, have been utilized for the identification of residues that line the pore of the closed channel. Photoincorporation of affinity labels, for example, provided compelling evidence that residues 9’, 13’, and 16’ face the closed-channel lumen in a muscle-type AChR^35,36. A comparison with Figs 1d and 2b, c, thus, reveals that the inner lining of the AChR’s pore is nearly the same in the closed and open states. Together, these data are consistent with a mechanism of channel opening that simply involves a widening of the pore, with the narrowest constriction switching from a location near the middle of the membrane²³ to a location near the intracellular entrance (positions −2’ and −1’; Fig. 1d). The rotation of M2 is minimal, if any (see also Supplementary Discussion).

The protonation state of wild-type residues is another vexing issue. Indeed, questions like the net charge of the rings of ionizable residues that flank the pore of Cys-loop receptors³⁷, and voltage-dependent Na⁺ and Ca²⁺ channels³⁸, for instance, or the specific charge-stabilizing interactions that keep the pK_a of the voltage-sensing S4 arginines well above physiological pH, remain elusive. On the basis of the insight gained here, the pK_a-values of some of the native ionizable groups of the AChR, and other members of the superfamily, can be predicted (see Supplementary Discussion).

Gratifyingly, both the channel-block data (Fig. 1d) and the proton-affinity measurements (Fig. 2b, c) are remarkably consistent with one another, and with the proposed α-helical secondary structure of M2 in the closed state (Refs ^{23, 35, 39}). Other labeling methods have also been used to explore the properties of the AChR’s open-channel pore but, because in the presence of agonist the channel interconverts rapidly among a number of different allosteric states (closed, open, and various desensitized states), labeling results can seldom be ascribed entirely to the open-channel conformation. Since identification of the open state in a single-channel record is unambiguous (Figs 1b and 3), our approach circumvents this critical problem altogether. In this paper, application of this method has allowed us to infer structural information on the AChR’s open-channel pore with unprecedented detail and precision, to suggest protonation states for some of the naturally-occurring ionizable residues, to attain a deeper understanding of the pore’s dielectric properties, and to provide an extensive set of highly-shifted pK_a-values, which, as the resolution of structural models of the AChR improves, could be used as meaningful benchmarks to validate theoretical electrostatic models for ion-channel proteins. More broadly, these data remind us that basic and acidic amino acids are not charged but chargeable residues, the pK_as of which are complex functions of the local microenvironment, and that the high energetic cost of burying the charged form of a residue, or of exposing it to lipids, can be relieved by simply releasing (basic residues) or taking (acidic ones) a proton. We anticipate that many more facets of the relationship between structure, function, and electrostatics in ion channels will be illuminated by the application of this experimental approach.

This work was supported by a grant from the National Institutes of Health to C.G. We thank S. Sine for the generous gift of wild-type muscle-AChR subunit cDNAs, J. Jasielec and J. Gasser for technical assistance, S. Varma and B. García-Moreno E. for discussions, and E. Tajkhorshid for introducing us to the VMD program.