Evolutionarily Conserved Interactions within the Pore Domain of Acid-Sensing Ion Channels

Abstract

Despite the sequence homology between acid-sensing ion channels (ASICs) and epithelial sodium channel (ENaCs), these channel families display very different functional characteristics. Whereas ASICs are gated by protons and show a relatively low degree of selectivity for sodium over potassium, ENaCs are constitutively active and display a remarkably high degree of sodium selectivity. To decipher if some of the functional diversity originates from differences within the transmembrane helices (M1 and M2) of both channel families, we turned to a combination of computational and functional interrogations, using statistical coupling analysis and mutational studies on mouse ASIC1a. The coupling analysis suggests that the relative position of M1 and M2 in the upper part of the pore domain is likely to remain constant during the ASIC gating cycle, whereas they may undergo relative movements in the lower part. Interestingly, our data suggest that to account for coupled residue pairs being in close structural proximity, both domain-swapped and nondomain-swapped ASIC M2 conformations need to be considered. Such conformational flexibility is consistent with structural work, which suggested that the lower part of M2 can adopt both domain-swapped and nondomain-swapped conformations. Overall, mutations to residues in the middle and lower pore were more likely to affect gating and/or ion selectivity than those in the upper pore. Indeed, disrupting the putative interaction between a highly conserved Trp/Glu residue pair in the lower pore is detrimental to gating and selectivity, although this interaction might occur in both domain-swapped and nonswapped conformations. Finally, our results suggest that the greater number of larger, aromatic side chains in the ENaC M2 helix may contribute to the constitutive activity of these channels at a resting pH. Together, the data highlight differences in the transmembrane domains of these closely related ion channels that may help explain some of their distinct functional properties.

Significance

Ion channels mediate the rapid transport of ions across cell membranes in the body, and their dysfunction can result in a broad range of diseases. For example, acid-sensing ion channels (ASICs) and epithelial sodium channel, both members of a large family of trimeric ion channels, are involved in neuronal signaling and salt homeostasis, respectively. However, despite their high degree of homology in primary amino acid sequence and molecular structure, these two ion channel types are functionally very different. Based on a combination of computational and functional approaches, we identified highly conserved residues and co-evolved residue pairs whose mutations disrupt various aspects of ASIC function. This provides insight into the functional importance of residues within the ASIC transmembrane domain.

Introduction

Ligand-gated ion channels (LGICs) are membrane proteins with ligand-sensing domains and an intrinsic membrane-spanning ion channel (1). By coupling chemical signals to ionic current across the cell membrane, LGICs are indispensable in, for example, the sensation of external stimuli and rapid signaling within the nervous system. One family of LGICs, the trimeric degenerin (DEG)/epithelial sodium channel (ENaC) family, includes channels of great functional diversity. Examples include acid-sensing ion channels (ASICs), which contribute to excitatory sodium currents in the nervous system, and ENaCs, which resorb sodium in the kidneys (2, 3). ASICs and ENaCs are formed by three subunits, each consisting of two membrane-spanning helices (M1 and M2) connected by a large extracellular domain (4, 5).

Despite their co-conservation in vertebrates, ASICs and ENaCs are only distantly related members of the DEG/ENaC family (6), which is reflected in several functional differences. Although both are selective for sodium over potassium, the relative permeability (P_Na⁺/P_K+) ranges from ∼10 at most ASICs to over 100 at ENaCs (7). The channels also differ in their gating. Most ASICs are closed at a physiological pH, activated by a rapid drop in extracellular pH to 6.0–6.7 (ASIC1 and ASIC3 isoforms) or below 5.0 (ASIC2), and desensitize in the continued presence of acidic pH (8). By contrast, ENaCs are constitutively active, although this activity is modulated by sodium ions, pH, and enzymatic activity (7, 9). Much work is invested in understanding the proton-induced activation of ASICs. Postsynaptic ASICs are activated by synaptically co-released protons (10, 11, 12, 13), and pain and synaptic plasticity can be targeted by the modulation of ASIC function (14, 15, 16).

On the molecular level, the transitions of vertebrate ASIC1a between resting, activated, and desensitized states have been dissected in structural studies on a chick ASIC1 (cASIC1) construct in high pH and low pH conditions (17, 18). These models indicate a rolling motion of the large extracellular domain of each subunit during proton-induced activation, in which the upper part of the extracellular domain moves inwards (toward the center of the trimer) and the lower part moves outwards (away from the center of the trimer). This pulls membrane-spanning helices tangentially and outwards, seeing a substantial rearrangement of the channel pore with little intra-subunit re-orientation of M1 and M2 helices (18, 19). Curiously, these cASIC1 structures show a domain-swapped structure, with M2 helices unwound in the middle such that M2a from one subunit forms a pseudocontinuous membrane-spanning helix with M2b of another (in resting, active, and desensitized states), whereas others show regular, straight M2 helices (4, 17, 18, 20). M2 helices are also likely to be straight in human ENaC, according to recent cryoelectron microscopy data (5). Mutagenesis experiments suggest a crucial role for lower parts of the channel in gating and in Na⁺/K⁺ selectivity (21, 22), and resolution is limited in this part of the channel in existing x-ray and cryoelectron microscopic structures. Thus, the contributions of residues within the transmembrane domain to channel function, as well as possible differences between ASICs and ENaCs, deserve further investigation.

We therefore employed computational methods using amino acid sequences from a broad sample of DEG/ENaC genes to identify residues that are closely associated with the evolution of ASICs or ENaCs. Using site-directed mutagenesis and electrophysiology with mouse ASIC1a (mASIC1a), we provide experimental evidence that many of these residues are indeed crucial to ASIC function.

Materials and Methods

Mining for ASIC sequences in the Swiss-Prot and UniProt databases

Based on 25 ASIC sequences found in the Swiss-Prot database (23) and aligned with Clustal Omega (24), we initialized a Hidden Markov Model (HMM) for the transmembrane region of the ion channel (helices M1 and M2; residues 45–72, 426–458 of cASIC1) using HMMER (25). The model was then used to identify sequences homologous to ASIC in the UniProt database (26). Several iterations, each consisting in an update of the HMM using the newly detected sequences, were performed, and overall, 3632 putative ASIC sequences were identified. After removing identical sequences, the size of the data set was reduced to 1156. This, however, included also the ENaC sequences (annotated according to the Swiss-Prot database), indicating that a significant fraction of our hits encode for ENaCs. To filter out these sequences, we applied a two-step strategy. First, we classified some of the sequences based on the phylogenetic tree, which was built using FastTree suit (27); sequences clustered with Swiss-Prot-annotated ASIC or ENaC were correspondingly assigned as ASIC or ENaC homologs. In the second step, we classified the rest of the sequences using Jarvis-Patrick clustering algorithm (28). We additionally preprocessed the data to increase the signal-to-noise ratio; in particular, we selected 14 columns (positions) of the multiple sequence alignment (MSA), which contain most information about it (29), and performed principal component analysis (PCA, (30)) on subsequences containing only these columns. Our analysis revealed two well-separated clusters, including mostly ASIC or ENaC sequences (annotated according to the Swiss-Prot database and classified based on the phylogenetic tree). We further used this information to classify the initial MSA containing all positions of the transmembrane domain of the ion channel. Among 1156 sequences, 826 were classified as ASIC, and 330 as ENaC.

Sequence logos and relative entropy

We used sequence logos to quantify the conservation profile of the MSA (31). In this representation, each column of the MSA is rendered as a stack of symbols whose height is proportional to the relative frequency. The total height of the stack conveys the degree of conservation and is equal to the difference between the maximal attainable value of the entropy (corresponding to a uniform distribution) and the observed entropy:

$$H_i={\log }_{2}20-\left(-\sum _{a=1}^{20}{P}_a^i{\log }_2{P}_a^i\right),$$

where $P_a^i$ is the relative frequency of the amino acid a in position i.

The relative entropy, or Kullback-Leibler (KL) divergence, is an information theoretic measure that quantifies the number of extra bits that, on average, are needed to encode samples from a probability distribution P using a code that is optimal for a different distribution Q. If applied to two MSAs, the KL divergence can be used to characterize the difference between two amino acid distributions (calculated for structurally equivalent positions). In this case, the KL divergence can be conceived as the information lost if the distribution of one MSA were used to approximate the distribution of the other. As a function of the position i, it reads as follows:

$$D_{KL}^i\left(P\Vert Q\right)={\log }_2\sum _{a=1}^{20}{P}_a^i{\log }_2\frac{P_a^i}{Q_a^i},$$

where $P_a^i$ and $Q_a^i$ are the relative frequencies of amino acid a at position i for the two alignments. Because the measure is not symmetric, we report two profiles in Fig. 1; each MSA is assigned the role of reference ($Q_a^i$) in one profile and sample distribution ($P_a^i$) in the other.

Figure 1.

Detection, classification, and comparative sequence analysis of ASIC and ENaC sequences. (A) Phylogenetic tree generated using a data set containing sequences homologous to ASIC and ENaC. The branches including sequences annotated in Swiss-Prot as ASICs are colored in red, whereas the branch including sequences annotated as ENaCs is shown in blue. The scale bar shows the evolutionary time and is in units of substitutions per sequence site. (B) Identification of the most informative columns of the MSA using the feature selection algorithm CVS (29, 32). CVS builds on the notion that the relevant subsequences occur in the data set with a particular frequency distribution. The algorithm identifies those positions along the sequence that are found recurrently in relevant subsequences. Positions exceeding a threshold of 50% in the frequency of occurrence are deemed relevant. M1 and M2 denote the first and second transmembrane helices, respectively. (C) The sequence data set was projected on the first two principal components (PCA vec1 and PCA vec2). The sequences annotated in Swiss-Prot and classified based on the phylogenetic tree as ASIC or ENaC are shown in red and blue, respectively; the unclassified sequences are shown in black. The two clusters corresponding to ASIC and ENaC sequences are highlighted. (D) Sequence logos of ASICs (top) and ENaCs (bottom). M1 and M2 denote the first and second transmembrane helices, respectively. The height of each residue at a given position is proportional to its frequency at this position; the height of the overall stack of residues at a given position is inversely proportional to Shannon entropy (31). Note that the comparison of diverse DEG/ENaC channels benefits from a common numbering system, in which equivalent residues from different family members are referred to with the same name/number (21). Hereafter, we use this system, referring to the conserved M1 tryptophan as M1 W0′ and the conserved M2 glutamate residue as M2 E18′ (see Fig. S1 for details). (E) Relative entropy (D_KL). The red and blue lines show the relative entropy of ASIC (using ENaC as background distribution) and ENaC (using ASIC as background distribution), respectively.

Critical variable selection

This feature selection approach, introduced in (32), has been recently applied to the analysis of MSAs (29) and aims at identifying the most relevant variables in a multivariate probability distribution. It is based on the assumption that each sample results from the optimization of an objective function; the method evaluates the amount of information that each variable carries on such an unknown function. The algorithm works by selecting the set of L variables (for each value L) that maximizes the entropy of the distribution of frequencies H[K]:

$$H\left[K\right]=-\sum _k\frac{k{m}_k}{M}\log \frac{k{m}_k}{M},$$

where $m_k=\sum _s{\delta }_{k,k_s}$ is the number of (sub)sequences s that occurs exactly k times in the reduced MSA defined on the L positions. The rationale of the method is as follows: because the counts estimate the unknown objective function, a set of variables is maximally informative if it is characterized by the widest possible statistical range for the set of counts, a condition that is equivalent to the maximization of H[K]. Because the number of combinations of variables grows exponentially with the sequence length, the maximum is found, in practice, through stochastic optimization. Thus, multiple runs are performed for each value of L, and for each of them, a set of positions of the MSA is selected. Ultimately, the relevance of each position is quantified by the number of times it was selected during the optimization runs.

Evolutionary coupling analysis

This analysis is based on the parametrization of a generative probabilistic model that reproduces some of the statistics observed on the MSA. In particular, the model distribution is obtained by maximizing Shannon entropy with the constraints that the distribution at each position along the sequence, and the joint distribution for each pair of positions coincide with the corresponding statistics calculated on the MSA. The resulting probability distribution function is that of a Potts model (33), and, for each sequence $\underset{¯}{s}$, reads as follows:

$$P\left(\underset{¯}{s}\right)=\frac{e^{-H\left(\underset{¯}{s}\right)}}{Z}\mathrm{and}H\left(\overline{s}\right)=-\sum _{i\le j}{J}_{ij}\left(s_i,s_j\right)-\sum _i{h}_i\left(s_i\right),$$

where the parameters J_ij(s_i, s_j) represent a statistical interaction between the amino acids s_i and s_j in positions i and j, the parameters h_i(s_i) account for the compositional bias at position i, and Z is a normalization factor. The value of the parameters can be found by maximizing a likelihood function. In practice, given the computational complexity of this task, an approximate likelihood function (the pseudolikelihood) is maximized (34). The usefulness of this model lies in the identification of strongly interacting pairs (large J_ij(s_i, s_j)) with co-evolving positions along the sequence, an insight that has been shown to enable the prediction of tertiary contacts using only sequence information (33). Because we are dealing with small data sets of sequences, particular care is taken in estimating the statistical errors associated with the estimated parameters J_ij(s_i, s_j). We used the resampling strategy described in (35) to associate a z-score to each parameter J_ij(s_i, s_j) and set significance threshold z to 3, which corresponds to a one-tailed p-value of 0.001.

Two-electrode voltage-clamp experiments

The mASIC1a complementary DNA in the pSP64 vector (provided by Dr. M. Carattino, University of Pittsburgh) was used for site-directed mutagenesis and complementary RNA (cRNA) synthesis, as described previously (21). Capped cRNA was synthesized with the Ambion mMESSAGE mMACHINE kit (Thermo Fisher Scientific, Waltham, MA) and injected in 40 nl water (see Table S1 for amounts). Stage V/VI Xenopus laevis oocytes were prepared as described (21), under license 2014-15-0201-00031 from the Danish Veterinary and Food Administration. For the incorporation of naphthyl-L-alanine (Nap) into ASIC1a, Nap was protected with nitroveratryloxycarbonyl and coupled to the dinucleotide pdCpA. Nap-pdCpA was enzymatically ligated to modified Tetrahymena thermophila transfer RNA (tRNA) (36). Before injection into oocytes, nitroveratryloxycarbonyl was removed by ultraviolet irradiation, and Nap/tRNA and 40 ng W46UAG (amber stop codon) mutant cRNA were co-injected into oocytes. As a control for nonspecific incorporation of endogenous amino acids, nonaminoacylated tRNA was co-injected with 40 ng W46UAG cRNA (Fig. S4).

1 to 2 days after cRNA injection, the oocyte was placed in a recording chamber continuously perfused with bath solution containing (in mM) 96 NaCl, 1 KCl, 1.8 CaCl₂, 1 MgCl₂, and 5 HEPES (for pH > 6.0) or 5 MES (2-(N-morpholino)ethanesulfonic acid; for pH equal to or lower than 6.0), adjusted with NaOH or HCl. Currents were measured with two-electrode voltage clamp, using 3 M KCl-filled borosilicate micropipettes, OC-725C amplifier (Warner Instruments, Hamden, CT), Digidata 1550 digitizer (Molecular Devices, San Jose, CA), and pClamp software (Molecular Devices). Membrane potential was −40 mV for most experiments. For estimating reversal potentials, peak proton-gated currents were measured at −40, 0, and +40 mV, and the x intercept between the latter two points was taken as the reversal potential. Currents were acquired at 1 kHz and filtered at 200 Hz for analysis (10 Hz for display in figures).

All recordings were performed on at least two different batches of oocytes. For proton concentration response experiments, peak proton-gated currents were normalized to the maximum and fit with the Hill equation at each cell. These were averaged for each construct, yielding the reported means (±SE). In contrast, curves in figures are fit to mean ± SE data for each point. Similarly, E_rev was calculated at each oocyte and averaged for each construct; figures show lines connecting normalized mean ± SE data points. Mean values were compared with one-way analysis of variance and Tukey’s test for multiple comparisons. All data analyses were performed in Prism v7 (GraphPad Software).

Results

Building a MSA of the ASIC transmembrane domain

Our goal was to identify patterns of conservation in genes encoding for members of the ASIC family and use this information to generate hypotheses about the molecular mechanisms underlying ASIC gating and how it differs from that of ENaC. However, because of the homology of ASICs and ENaCs and the fact that few channels outside of the rodent/vertebrate subfamily have been characterized, we first dissected and classified a broad range of ASIC and ENaC sequences and then analyzed their sequence conservation profiles in the transmembrane domain.

We started by collecting a large number of putative ASIC transmembrane domain sequences and generated a MSA. To this end, we searched the UniProt database (26) for statistically significant matches with a suitably defined profile HMM (25). The latter was initialized by aligning 25 sequences from Swiss-Prot (23) and annotated as ASICs from different species, including nonvertebrate and vertebrate animals, using Clustal Omega (24) (Fig. S1). To increase the sensitivity of our search for ASIC homologs, we iteratively improved the HMM through updating it with the sequences detected in the database; overall, we identified 1556 hits.

Although sufficient for quantitative statistical analyses, these sequences are unlikely to belong to a single functional family as suggested by the presence of Swiss-Prot hits annotated as ENaC. To filter out these sequences and collect only bona fide ASIC orthologs, we used a two-step strategy combining phylogenetic and data analytics approaches. We first calculated the dendrogram for the MSA (27) and highlighted the branches containing Swiss-Prot annotated sequences (Fig. 1 A). Whereas annotated ENaCs segregate in a well-defined cluster, annotated ASICs are found in three different major branches, although gratifyingly, our analysis appeared to be suited to segregate the non-acid-sensitive bile acid-sensitive ion channels (termed ASIC5 in Fig. 1 A) from ASICs. However, most of the nonannotated sequences cannot be unambiguously grouped with either ENaCs or ASICs. Therefore, we partitioned the data set in two groups using a well-established clustering algorithm Jarvis-Patrick (Jarvis and Patrick (28)). To increase the signal-to-noise ratio, we preprocessed the sequences through feature selection and dimensionality reduction. In particular, we used the critical variable selection (CVS) approach to identify the most informative columns (positions) of the MSA (29, 32). We thus restricted the MSA to 14 positions, which contain the subsequence maximally informative about the data set (Fig. 1 B). We then performed PCA (30) to project each sequence on a two-dimensional space (Fig. 1 C). The projection makes apparent the intrinsic tendency of the sequences to cluster in two groups, each containing sequences annotated as ASICs or ENaCs. Importantly, a few ASIC-annotated sequences appeared in the ENaC group, including two from Caenorhabditis elegans (Fig. S1 B). Unlike other annotated sequences, these have not yet been studied experimentally but were classified as ASICs based on homology to mammalian ASIC sequences. Considering the absence of functional data and the relationships we identified here, we suggest that the annotation of these sequences is incorrect, and they rather belong to ENaCs.

Analysis of sequence conservation in ASIC and ENaC

Having classified each sequence (including all positions of the transmembrane domain) as either putative ASIC or ENaC family members, we proceeded to analyze the conservation profiles of the two groups of sequences. To this end, we considered the distribution of amino acids at each position of the MSA and calculated a sequence logo (31) as a graphical representation of evolutionary conservation. Interestingly, these profiles are very similar for ASICs and ENaCs, as shown in Fig. 1 D. The similarity observed in the amino acid distributions at all positions explains why discriminating between ASIC and ENaC sequences is a challenging task that required data analytics approaches. The sequence logos show that in both groups, the second transmembrane helix (M2) is significantly more conserved than the first (M1). This suggests that M2 was subject to greater evolutionary pressure compared with M1, consistent with the fact that M2 is pore lining and crucially involved in ionic conduction, selectivity, and likely gating. Most of the conserved residues in M2 are located in the center of the helix. In several ASIC structures, this region corresponds to a pore constriction and is where the helical periodicity of M2 is interrupted to allow for the domain swap to occur (17, 18). It includes four conserved glycine residues that, because of the lack of side chain, allow for a wide pore diameter and possibly act as molecular hinges for the domain swap transition. In addition to these, the amino acid at M2 18′ (see Figs. 1 D and S1 for numbering) is highly conserved. This position is almost invariably a glutamate, which is shown to contribute to ion selectivity and/or other functions in ASIC1a and ENaC (21, 37, 38, 39). In M1, only W0′ is strictly conserved in both ASIC and ENaC sequences, whereas the rest of the helix shows considerable sequence variability. Interestingly, in some ASIC structures (4, 17, 18), M1 W0′ is in close contact with M2 18′, suggesting that a physical interaction between these amino acids might be responsible for the high degree of conservation and thus have a functional role (a notion we address experimentally further below).

Many of the corresponding positions in the MSAs of ASICs and ENaCs have similar amino acid distributions (Fig. 1 D). To quantify the differences, we calculated the relative entropy between the ASIC and ENaC distributions for each position. As expected, the relative entropy is small for almost all of the positions (Fig. 1 E), consistent with the observed high sequence similarity between ASIC and ENaC. However, for two positions, M2 8′ (F441 in cASIC1) and, to a lesser extent, M2 21′ (D454), the relative entropy is large, indicating that the amino acid distributions differ significantly between ASICs and ENaCs. Indeed, residue M2 8′ is a conserved tryptophan in ENaCs but a phenylalanine in most ASICs, and residue M2 21′ is glutamate in most ENaCs but usually aspartate or tyrosine in ASICs. This divergence between ASICs and ENaCs at the M2 21′ position may be related to observations that M2 21′ mutations have different effects in the two channel types. M2 D21′N mutant ASICs show decreased ion selectivity, whereas M2 E21′R/C mutant ENaCs show unaltered ion selectivity (21, 37, 38). Also consistent with a different role of this position in ASICs and ENaCs, the D21′N mutation confers on ASIC1a neither ENaC-like levels of ion selectivity nor constitutive activity (21).

To investigate possible functional implications of the apparent divergence at the M2 8′ position, we tested mutant M2 F8’W mASIC1a channels for potential indications of ENaC-like function, such as increased Na⁺ selectivity and/or constitutive channel activity. This was done using site-directed mutagenesis and two-electrode voltage-clamp electrophysiology on Xenopus oocytes expressing mASIC1a. In oocytes expressing mutant channels, we observed a small constitutive current and a sustained proton-gated current in the continued presence of protons, although neither was significantly greater than wild-type (WT) (Fig. 2, A and B). Half-maximally activating proton concentration (pH₅₀; 6.94 ± 0.06, n = 5) and the reversal potential (E_rev) of peak proton-gated currents (44 ± 4 mV, n = 10) at M2 F8′W channels did not differ significantly from WT (6.74 ± 0.03, n = 6, and 44 ± 2 mV, n = 10), indicative of unaltered proton sensitivity and ion selectivity. Either side of the M2′ 8′ position, we also noticed that leucine or methionine residues occupy positions M2 7′ and 9′ in most ASICs and ENaCs. However, in several mammalian ENaC subunits that have been functionally characterized, these positions are occupied by a phenylalanine residue (Fig. S1 B) as reflected in the noticeable occurrence of phenylalanine at these positions for ENaCs in Fig. 1 D. We therefore generated triple-mutant L7′F/F8′W/I9′F mASIC1a channels and measured the same parameters as above, looking for potential ENaC-like function. Although the functional phenotype showed greater variability than the other mutants investigated here, this triple mutant showed obvious increases in constitutive and sustained proton-gated current amplitude (Fig. 2, A and B), which are likely to be nonselective (40, 41). Proton sensitivity was increased, with a pH₅₀ value of 7.29 ± 0.01 (n = 4, P < 0.001 cf. WT), and a significant change in E_rev of the proton-gated peak current (30 ± 4 mV, n = 13) was observed (Fig. 2, C and D). Taken together, these data tentatively suggest that the aromatic side chains at these positions in ASICs and ENaCs could contribute to channel gating and ion selectivity.

Figure 2.

Function of single-mutant M2 F8′W and triple-mutant M2 L7′F/F8′W/I9′F mASIC1a channels. (A) Example recordings of pH 6.0-gated currents at different membrane potentials. (B) Mean constitutive current and mean proton-gated current after 10 s are both normalized to peak proton-gated current (n = 8 (WT), 10 (F8′W), and 11 (L7′F/F8′W/I9′F), at least three oocytes from each of three different batches). ^∗∗p < 0.05, ^∗∗∗p < 0.001 compared with WT in one-way ANOVA with Tukey’s test. (C) Normalized, mean (±SE, n = 4–5, across two batches of oocytes) peak current responses to increasing proton concentration. Inset shows current responses to a pH of 7.6, 7.4, 7.2, 7.0, and 6.8 at an oocyte expressing M2 L7′F/F8′W/I9′F (resting pH = 8.5). F8′W data in (C) are from (21). (D) Normalized, mean (±SE, n = 10 (WT), 10 (F8′W), and 13 (L7′F/F8′W/I9′F), at least three oocytes from each of three different batches) peak proton-gated current at different membrane potentials.

Co-evolved positions in the MSA of ASICs

Having analyzed the conservation of residues at each single position of the MSA, we proceeded to consider residue pairs (i.e. conservation of putative residue-residue interactions). We performed direct coupling analysis using the pseudolikelihood method to find the optimal parameters of the pairwise statistical model (34). As is customary in this analysis, we identified statistically significant pairwise couplings (z-score larger than 3) and corresponding pairs of co-evolving residues (Fig. 3 A). We then highlighted these seven pairs in the context of two available ASIC structures, one with and one without the domain swap (4NTW and 2QTS, respectively), and observed that residues from several pairs contact each other in both structures (Fig. 3 B; upper panel, Fig. S2). Most are located in the upper half of the channel and form intrasubunit interactions M1 L11′/M2 M5′, M1 V14′/M2 Q4′, M1 R18′/M2 D0′, M1 R18′/M2 Q4′, M1 I19′/M2 I1′, and M1 Y22′/M2 L-3′. It is only M1 W0′/M2 E18′ in the lower half, where the interaction occurs across two subunits. Our finding suggests that the upper half of the channel is stabilized by a network of residue-residue interactions and therefore is likely more rigid compared with the lower half (closer to the intracellular solution).

Figure 3.

Comparison between the structures with (Protein Data Bank (PDB): 4NTW) and without (PDB: 2QTS) the domain swap. (A) Maps of distances between residues in the two structures (left), the difference between these two distance maps (middle) and superposition of pairs of contacting residues and evolutionary coupled pairs of residues (right). The grey arrows lining the distance maps highlight the location of helices M1 and M2; the tips and nocks of the arrows correspond, respectively, to the extra- and intra-cellular ends of the helices. Two residues were considered to be in contact if the distance between any of their heavy atoms was less than 5 Å. The distance maps were used to identify evolutionary coupled pairs of residues that are in contact with each other. The difference between the distance maps shows that the two structures have similar sets of contacts between the two M1 helices but different sets of contacts between two M2 helices and between M1 and M2. In the right panel, pairs of residues that are both evolutionary coupled and in contact in both conformations (swapped and nonswapped) are shown in green; evolutionary coupled pairs that are in contact only in one conformation are colored in red (swapped) and blue (nonswapped). Evolutionary coupled pairs that are not in contact in any of the two structures are shown in gray. (B) Cartoon representation of the structures with and without the domain swap and the evolutionary coupled pairs of residues (color scheme as in Fig. 3A, right panel).

In contrast to these co-evolving pairs that are in contact in both structures, six pairs were identified in either one or the other structure (Fig. 3 B, lower panel; Fig. S3). Paired residues are located only in the upper half of the channel in the domain-swapped structure, including M2 G2′/M2 L7′, M1 Y21′/M2 E-7′, M1 R18′/M2 E-7′, and M1 R18′/M2 A-5′, and only in the lower half in the straight-helix structure, including M1 W0′/M2 Y22′ and M1 W0′/M2 T15′. This structure-based analysis indicates that both conformations are necessary to explain the identified network of evolutionary coupled residues, which implies functional relevance of both structures. Whether or not this functional relevance reflects transitions through both of these conformations during channel gating remains to be established.

We sought an experimental test of the reliance on co-evolved positions by measuring the functional characteristics of mutant mASIC1a channels expressed in Xenopus oocytes. Individual amino acid residues in mASIC1a were replaced with residues unlikely to occur at these positions according to our computational analysis (Figs. S2 and S3). For example, M1 L11′ was replaced with asparagine, and M2 M5′ was replaced with glutamine as both substitutions reflect white squares in Fig. S2 (top-middle plot), leading to M1 L11′N and M2 M5′Q mutants. In other cases, we used mASIC1a mutants that were already available in the laboratory, such as M2 D0′N, which—although asparagine is tolerable in this position (pink squares in row “N” in Fig. S2 bottom-left plot)—is unlikely to occur in combination with M1 R18′ (white square in row “N” of the same plot).

In each of the seven pairs of residues that appear close together in both structures, at least one of the two mutations caused a significant decrease in peak proton-gated current amplitude (Fig. 4 A; Table S1). For three of these mutations, measurable currents were observed at oocytes injected with a greater amount of cRNA, M1 L11′N and M2 M5′ (paired) and M2 L-3′A (paired with M1 Y22′). There did not seem to be a correlation between the extent of physicochemical changes of substitutions and the presence of this low-current phenotype. For instance, the effects of the mutation M1 R18′L, which decreases side-chain length/volume and substitutes hydrophobicity for hydrophilicity, were negligible compared with the effects of partner mutation M2 D0′N, which maintains size and similar hydrophilicity but removes negative charge. In mutants involving pairs from the upper half of the channel (with measurable currents), the effects on other aspects of channel function were relatively minor. M1 L11′N was the only mutation that caused a decrease in the potency of proton-gated activation (Fig. 4, B–D). The M2 M5′Q mutation and the M2 D0′N mutations both showed a trend toward decreased Na⁺ selectivity (Fig. 4, E and F; as also reported by others for M2 D0′N (42) and M2 M5′C (43)). Mutations to the M1 W0′/M2 E18′ pair in the lower half of the channel caused substantial changes in several functional parameters (Table S1) and are discussed in detail further below. Thus, mutations that disrupt co-evolved residues that are paired in both domain-swapped and straight-helix structures seem detrimental to functional cell surface expression and, in some cases, to specific functional parameters, such as ion selectivity or channel gating by protons.

Figure 4.

Functional analysis of ASIC1a mutants. (A) Top, Example recordings of responses to a pH of 5.0 (black bars) from oocytes injected with 0.8 ng cRNA of indicated mASIC1a mutants. Bottom, Mean peak current amplitude (n = 4–11) of responses to a pH of 5.0 from oocytes injected with 0.8 ng or 40 ng (underlined) cRNA. (B) Example recordings of current responses to decreasing pH from oocytes injected with 40 ng of mutant ASIC1a cRNA. (C and D) Normalized, mean (±SE, n = 4–5) peak current responses to increasing proton concentrations at WT and mutant ASIC1a channels. (E) Example recordings of pH 6.0-gated current at different membrane potentials. (F and G) Normalized, mean (±SE, n = 4–5) peak proton-gated current at different membrane potentials. Inset in (F) shows a magnified view of WT, M2 D0′N, and M2 M5′Q data.

Regarding the six pairs of residues that appear close together in only one of the two structures, we generated additional mASIC1a mutant channels targeting the upper three of the pairs of M1 R18′/M2 E-7′, M1 R18′/M2 A-5′, and M1 Y21′/M2 E-7′. M2 A-5′Q, M1 Y21′L, and M2 E-7′C mutations caused significant but relatively mild decreases in maximal current amplitude of proton-gated currents (Fig. 4 A; Table S1), consistent with results with the M1 R18′L mutant described above. Effects on pH₅₀ and E_rev were also negligible in these mutants, although the M2 E-7′C mutation caused a significant increase in the n_H for proton-gated currents (Fig. 4 D; Table S1). The remaining three pairs are near the middle and the bottom of the channel, M2 G2′/M2 L7′, M1 W0′/M2 T15′, and M1 W0′/M2 Y22′. Their disruption has a much greater impact on channel function as evident in previously published mASIC1a mutations such as M2 G2′C and M2 T15′V, which both abolish measurable activity (21, 43), and M2 L7′A, which alters both ion selectivity and channel gating by protons (21). W0′ mutations also cause substantial changes in function and are considered at length below. Taken together, these results reveal no clear relationship between the occurrence of a pair in one structure or both structures and functional effects of the mutation. However, mutations near the top of the channel seem to either decrease maximal current amplitude or have little effect on channel gating and ion selectivity, whereas mutations near the middle of the channel and at the bottom of the channel seem to alter gating and ion selectivity in addition to affecting maximal current amplitude.

Functional test of potential M1 W0′/M2 E18′ interactions

Given the appearance of M1 W0′ and M2 E18′ in our initial sequence conservation analysis, the appearance of the M1 W0′/M2 E18′ pair in our co-evolution analysis, the noticeable effects of mutations in this pair (Table S1), and the potential role of a direct M1 W0′/M2 E18′ interaction in ASIC structure (17), we probed the functional role of these two residues further. M1 W0′F and M2 E18′Q mutations both induced constitutive currents and sustained proton-gated currents that were not observed for WT, although only significant and much greater for M1 W0′F (Fig. 5, A–C). The E_rev for peak proton-gated currents was significantly decreased from 42 ± 2 mV at WT (n = 5) to 17 ± 6 mV at M1 W0′F (n = 5) and −3 ± 1 mV at M2 E18′Q (n = 4) channels (both p < 0.001 cf. WT; Fig. 5 D; Table S1), as described previously for mASIC1a E18′Q (21) and indicative of decreased Na⁺/K⁺ selectivity. Both mutations also caused significant increases in the potency of activation by protons, with increases in pH₅₀ from 6.74 ± 0.03 at WT (n = 6) to 7.14 ± 0.02 at M1 W0′F (n = 7) and 7.08 ± 0.03 at M2 E18′Q (n = 6) channels (both P < 0.001; Fig. 5 E).

Figure 5.

Function of M1 W0′ and M2 E18′ mutants. (A) and (B) pH 6.0-gated currents from oocytes expressing M1 W0′F or M2 E18′Q mutant channels at different membrane potentials. W46F resting pH is 8.5; M2 E18′Q resting pH is 7.6. a is constitutive current, b is proton-gated current after 10 s, and c is peak proton-gated current. (C) Mean constitutive current and mean proton-gated current after 10 s are both normalized to peak proton-gated current (n = 5–6). ^∗p < 0.05 and ^∗∗∗p < 0.001 compared with WT in one-way ANOVA with Tukey’s test. (D) Normalized, mean (±SE, n = 4–5) peak proton-gated current at different membrane potentials. (E) Currents in response to increasing proton concentration from oocytes expressing WT (resting pH = 7.6) or M1 W0′F (resting pH = 8.5). (F) pH 5.5-gated currents at different membrane potentials (upper) and currents in response to increasing proton concentrations (lower) from oocytes expressing M1 W0′Nap channels (resting pH = 7.6). (G) Side-chain structure of M1 W0′ mutant substituents. (H) Normalized, mean (±SE, n = 5–7) peak current responses to increasing proton concentration.

If these residues were to interact directly via a H-bond between the M1 W0′ indole nitrogen and the M2 E18′ carboxylate, altered function of mASIC1a might be expected when the equivalent M1 W0′ residue is replaced with 2-amino-3-indol-1-yl-propionic acid (Ind), which retains the indole structure but removes any H-bonding propensity (Fig. 5 G; (44)). Ind was successfully incorporated via the nonsense suppression method (see Fig. S4). Ion selectivity and the potency of proton-gated activation of M1 W0′Ind channels was very similar to WT channels and to WT-rescue channels incorporating tryptophan (Trp) via the same method (Fig. 5, D and H). Similarly, M1 W0′F₃-Trp channels, in which the H-bond donor propensity of the indole nitrogen is increased, and in which π-electron density of the benzene ring is decreased (44) did not differ from WT in these parameters (Fig. 5, D and H). This is in stark contrast to the effects of M1 W0′F and W0′L substitutions, which both increased the potency of channel activation by protons and decreased ion selectivity, arguing that the decrease in side-chain size through these substitutions is responsible for the observed effects as opposed to altered electrostatic properties and thus a hydrogen bond with the M2 E18′. To confirm this notion, we measured ion selectivity and proton-gated activation of channels containing Nap at the M1 W0′ position, thus introducing a modestly larger side chain than tryptophan without polar atoms (Fig. 5, F and G). Ion selectivity was similar to WT (Fig. 5 D), and the potency of proton-gated activation was moderately decreased at M1 W0′Nap channels (Fig. 5 H). This confirms an inverse correlation of side-chain volume at this position and the potency of proton-gated activation and the requirement of a large aromatic side chain at this position to retain normal ion selectivity in ASIC1a. The dependency of channel function on side-chain bulk at this position explains the high conservation of tryptophan throughout the ASIC—and potentially the broader DEG/ENaC—family (Fig. 2 A).

Discussion

In this study, we adopted a computational approach to identify conserved and/or co-evolved residues in the transmembrane domains of ASICs and ENaCs to illuminate key determinants of channel function. This was complemented by functional experiments, confirming that most of the identified residues are important to functional expression, channel gating, or ion selectivity in ASIC1a. Below, we discuss how this data could help explain divergent function in the DEG/ENaC family.

Conserved residues

Our analysis of sequence conservation highlighted six residues located in the pore lining M2 helix as broadly conserved. These include M2 2′, 3′, 6′, and 10′ glycine and M2 12′ serine residues near the middle of M2 and the M2 18′ glutamate residue near the intracellular end of M2. ENaCs are obligate heterotrimers, and in α/β/γ ENaC, M2 G10′ and S12′ mutations in any subunit have been shown to alter Na⁺ conduction and/or ion selectivity (45, 46). M2 G2′, G3′, G10′, and S12′A mutations render homotrimeric ASIC1a nonfunctional (or not expressed (43)), but concatemeric ASIC1a channels containing up to two M2 S12′A mutations retain WT-like ion selectivity (21). Conversely, M2 E18′ mutations do not seem to directly affect ion selectivity in ENaC (although such mutants certainly show altered current properties (37, 38, 39)), whereas M2 E18′ is a major determinant of ion selectivity in various ASICs (21, 47). Given these results, and that M2 S12′ side-chain hydrogen bonds appear crucial to the domain swap in certain cASIC1 structures and the domain swap does not appear to be present in existing ENaC structural data (5, 17), it is conceivable that conserved mid-M2 residues contribute more to ion conduction in ENaCs, although they are more important to pore structure in ASICs. The converse could be true for M2 E18′ at the lower end of the pore.

The most broadly conserved residue is the tryptophan at the M1 0′ position. We show here that it is co-conserved with M2 E18′ and that, somewhat like M2 E18′Q, the M1 W0′F and L mutations decrease ion selectivity and show constitutive currents and sustained proton-gated currents. These results and the close proximity of M1 W0′ and M2 E18′ in at least one subunit (or subunit interface) in both straight-helix and domain-swapped cASIC1 structures point toward a direct interaction between the two that is crucial for channel function. However, our noncanonical amino acid substitutions show that a hydrogen bond between the tryptophan indole nitrogen and glutamate carboxylate is unlikely to contribute to these functions in ASIC1a. Rather, our data are consistent with the notion that side bulk at M1 W0′ might help position M2 E18′ for WT-like ion conduction and gating. Overall, highly conserved residues in the lower part of the ASIC channel clearly contribute to both channel gating and ion selectivity, but the molecular details remain to be resolved. The role of M1 W0′ in ENaC has not been probed by mutagenesis to our knowledge.

Analysis of sequence conservation also identified two M2 positions of significant divergence between ASICs and ENaCs. Also, near the bottom of the channel, the M2 21′ position is occupied by glutamate in functionally characterized ENaCs but by similarly charged but shorter aspartate side chains in ASICs. Mutation of this residue affects ion selectivity in ASIC1a but only ion conduction in ENaC (21, 37, 38), pointing again toward different contributors to ion selectivity in the two channels (39). Most noticeably, however, M2 8′ near the middle of the channel is a phenylalanine in most ASICs but a tryptophan in nearly all ENaC sequences. Similarly, M2 7′ and 9′ are often leucine or isoleucine in ASICs but larger phenylalanine residues in ENaCs. Our results with M2 F8′W ASIC1a mutants suggested that this residue is more important for gating than ion selectivity in ASIC, and in ENaC, M2 W8′ mutations have little effect on ion selectivity (gating was not measured (46)). Similarly, the more drastic mASIC1a L7′F/F8′W/I9′F triple mutant had only minor effects on relative ion permeability, whereas effects on resting and proton-gated currents were much more pronounced. Although not definitive, these findings could indicate that larger side chains in this part of M2 contribute to the gating differences observed between ASICs and ENaCs.

Co-evolved residue pairs

The direct coupling analysis performed on the MSA revealed a network of conserved residue-residue interactions between the M1 and M2 helices. Projecting this network on the available ASIC structures further showed that most of the interacting residues are located in the upper part of the transmembrane domain. A recent database-wide analysis focused on the network topology of co-evolving residues has shown that densely connected communities of couplings correspond to quasirigid domains (48). In channels of the six transmembrane helix family, these dense clusters of evolutionary couplings have been used to rationalize the different functional mechanics underlying activation of voltage-gated and transient receptor potential channels in terms of rigid and flexible regions (35, 49, 50, 51, 52). These studies have highlighted not only that densely connected regions are rigid, but also that mobile structural elements, such as the S4 helix of the voltage sensor domain, are statistically uncoupled from the rest of the sequence (52).

By analogy, we surmise that the relative position of M1 and M2 remains relatively stable in the upper half of the ASIC transmembrane domain, although they are likely to move in concert during gating (18). Whether there is greater flexibility in the lower half of the pore during the gating cycle, we cannot say with certainty, although recent comparisons of resting, active, and desensitized domain-swapped structures suggests that this might be the case (18). Indeed, some of the co-evolved pairs we identified only appear in the domain-swapped structure and some only in the nondomain-swapped structure. This suggests that both structures could reflect states that channels need to visit for normal structure and/or function, also because mutagenesis of residues involved in the network of couplings in ASIC1a affects biophysical properties of the channel. A potential caveat with this interpretation involves the only pair of co-conserved residues in the lower half of the channel, M1 W0′/M2 E18′. These are in contact in both swapped and straight-helix structures, although we cannot exclude the possibility that the functional contribution of this interaction might differ depending on the M2 conformation (inter-subunit in the domain-swapped structures and intra-subunit in the nondomain-swapped structure).

Additionally, although M2 E18′ had been implicated in ion selectivity before, to our knowledge, this is the first evidence that M1 W0′ may also contribute to ion selectivity in ASICs. The fact that M2 D0′, which was previously implicated in Na⁺ conduction (42), had a small effect on ion selectivity is not surprising, but M1 R18′L showed no such effect. Thus, the contribution of a potential M1-M2 salt bridge (M1 R18′-M2 D0′) is unclear, although M2 D0′ in particular seems important for ASIC activity. Similarly, the functional effects of the predicted interaction between M2 M5′, whose mutation has a minor effect on ion selectivity, and M1 L11′, whose mutation decreased the potency of proton-gated activation, appear complex.

Overall, we find that residues are more likely to be co-evolved at the top of the transmembrane domain, whereas mutations near the middle or the bottom of the transmembrane domain are more likely to affect ASIC gating and ion selectivity than those at the top. However, we cannot link a particular functional phenotype to mutations of coupled residue pairs occurring in one or both structural contexts. Our exploratory study thus highlights crucial residues in the ASIC transmembrane domain and supports the idea that both domain-swapped and nondomain-swapped conformations might be populated during the ASIC gating cycle. However, additional experiments and calculations will be needed in the future to provide more definite insight, which ideally would also include the regions beyond the transmembrane domains.

Author Contributions

M.A.K., T.L., V.C., and S.A.P. designed experiments. M.A.K. and D.G. conducted the computational and T.L., Z.P.S., and C.B.B. the functional experiments. All authors contributed to data analysis and manuscript writing.

Editor: Vasanthi Jayaraman.