A Family of Acetylcholine-gated Chloride Channel Subunits in Caenorhabditis elegans*

The genome of the nematode Caenorhabditis elegans encodes a surprisingly large and diverse superfamily of genes encoding Cys loop ligand-gated ion channels. Here we report the first cloning, expression, and pharmacological characterization of members of a family of anion-selective acetylcholine receptor subunits. Two subunits, ACC-1 and ACC-2, form homomeric channels for which acetylcholine and arecoline, but not nicotine, are efficient agonists. These channels are blocked by d-tubocurarine but not by α-bungarotoxin. We provide evidence that two additional subunits, ACC-3 and ACC-4, interact with ACC-1 and ACC-2. The acetylcholine-binding domain of these channels appears to have diverged substantially from the acetylcholine-binding domain of nicotinic receptors.

Although no genes encoding ACh-gated chloride channels have been previously identified, it is likely that many invertebrate receptors with unusual properties remain to be characterized. The genomes of Caenorhabditis elegans and Drosophila melanogaster reveal numerous predicted Cys loop LGICs that do not obviously belong to any family of known ion or ligand specificity (18 -21). C. elegans in particular encodes ϳ70 LGIC subunit genes, of which fewer than 20 have been characterized pharmacologically (19). The function of such a large and diverse LGIC superfamily in a single species is unclear.
To better understand the constraints on Cys loop LGIC structure and evolution and to identify new modes of neurotransmission, we have characterized several members of a novel family of channel subunits from C. elegans. These form ACh-gated chloride channels exhibiting an unusual pharmacology that appears to reflect a unique ACh-binding site.

EXPERIMENTAL PROCEDURES
Cloning ACC cDNAs-Poly(A ϩ ) RNA was purified from adult worms (Bristol N2 strain). First strand cDNA was synthesized with oligo(dT) primer using the avian myeloblastosis virus reverse transcriptase system (Invitrogen Canada Inc., Burlington, Ontario, Canada). The open reading frame of ACC-1, -2, and -4, as predicted in Wormbase (available on the World Wide Web at www.wormbase.org/), was amplified by PCR using the following primers: ACC-1, 5Ј-GGGGACA-AGTTTGTACAAAAAAGCAGGCTCATATGAGTCATCCGGGTTGGA-TTAT-3Ј and 5Ј-GGGGACCACTTTGTACAAGAAAGCTGGGTCTAGA-TTAAGGTTGATCAATATTCACA-3Ј; ACC-2, 5Ј-GGGGACAAGTTTGT-ACAAAAAAGCAGGCTCATATGATATTTACTCTTTTATCAACACTG-CCT-3Ј and 5Ј-GGGGACCACTTTGTACAAGAAAGCTGGGTCTAGAT-TATCCGTCAACTCGATT-3Ј; ACC-4, 5Ј-GGGGTACCATATGCGACTA-ATCATATTAGTAATCT-3Ј and 5Ј-GCTCTAGATTAGATAGTTCTAAC-CAATAGTTTTCC-3Ј. PCR products were subcloned either into pDON-R201 (ACC-1 and ACC-2) via recombination reaction using the Gateway Cloning Technology kit (Invitrogen) or into pBluescript (Stratagene Inc., La Jolla, CA) using the KpnI and XbaI sites (ACC-4). ACC-3 was first amplified from cDNA using a primer to the SL-1 transpliced leader sequence 5Ј-GGGGACAAGTTTGTACAAAAAAGCAGGCTGGA-TCCTTTAATTACCCAAGTTTGAG-3Ј and a 3Ј primer corresponding to the end of the putative open reading frame, 5Ј-GGGGACCACTTTGT-ACAAGAAAGCTGGGTCTGCAGTCATGTGTTAACAGTAAGGTAAT-AT-3Ј. The resulting PCR product was cloned into pDONR and sequenced. A new 5Ј-primer corresponding to the 5Ј-end of the open reading frame was used with the previous 3Ј-primer to amplify the open reading frame from the cloned ACC-3 cDNA. Three cDNA clones of each gene were sequenced to determine the true open reading frames and to find possible mutations resulting from reverse transcription-PCR. Nonsilent mutations were fixed either by overlap extension PCR (22) or by splicing together mutation-free cDNA fragments using convenient restriction sites.
Sequence Analysis-Amino acid sequences were aligned using the ClustalW program (available on the World Wide Web at clustalw. genome.ad.jp). Transmembrane domains were predicted using the TMHMM method based on a hidden Markov model (23). SignalP 2.0, NetGlyc 1.0, and NetPhos 2.0 programs based on artificial neuronal networks were used to predict signal peptide sequences (24). All of the above mentioned prediction programs are available on the World Wide Web at www.cbs.dtu.dk/services. Expression in Xenopus Oocytes and Electrophysiology-cDNAs were subcloned into the pT7N expression vector (25). The pT7NcDNA constructs were linearized with SalI (ACC-2) or BamHI (ACC-1, -3, and -4), and capped cRNAs were transcribed using the MEGAscript Kit (Ambion, Austin, TX). Synthesized cRNAs were recovered by LiCl precipitation and resuspended in nuclease-free H 2 O at a final concentration of 1 g/ml.
Oocytes were harvested from mature female Xenopus laevis according to standard procedures (26). Oocytes were maintained at 20°C in ND96 solution (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl 2 , 1 mM MgCl 2 , and 5 mM Hepes, pH 7.5) supplemented with 100 mg/ml gentamycin and 550 mg/ml pyruvate. Oocytes were injected with 40 nl of cRNA using the Nanoject system (Drummond Scientific, Broomall, PA) and incubated for 2 days before measurements were taken.
Two-electrode voltage clamp recordings were performed using the AxoClamp 2B amplifier (Axon Instruments, Foster City, CA). Oocytes were perfused in an RC-12 recording chamber (Warner Instrument Inc., Hamden, CT) or a Maltese Cross chamber (ALA Scientific Instruments, Westbury, NY). Data were acquired at 1 kHz using Clampex software (Axon Instruments, Foster City, CA). All drugs were obtained from Sigma. Dose-response curves for agonists were generated by applying increasing concentrations of drug followed by 3-9-min washes. To determine the EC 50 and Hill coefficient, dose-response curves (shown as smooth curves in graphs) were fitted using the Hill equation as follows, f͑I͒ ϭ ͑I max ͓I͔ ⅐ /͑EC 50 ⅐ ϩ ͓I͔ ⅐ ͒͒ ϩ I min (Eq. 1) where I represents the normalized response, I max is the maximal response elicited by a saturating concentration of agonist, EC 50 is the concentration of agonist inducing half-maximal response, n is the Hill coefficient, and I min is the normalized response at lowest agonist concentration.
Oocytes were preincubated with antagonist for 1 min (10 min for ␣-BT) prior to co-application of the antagonist with either 1 M or 10 M ACh for oocytes expressing ACC-1 or ACC-2, respectively. The amplitude of response to co-application of ACh and antagonist was normalized to that of the response to ACh alone.
Because the ACC-1 and ACC-2 channels inactivate slowly in the presence of ACh, I-V curves were measured using voltage ramps of 4 mV/s in the presence of either 2 or 20 M ACh for oocytes expressing ACC-1 or ACC-2, respectively. For ion substitution experiments, sodium gluconate or arginine-Cl were substituted for NaCl in the ND96 solution. The current amplitude was normalized to the peak response in normal ND96 at Ϫ80 mV.

Cloning of Members of the ACC Subunit Family-A neighbor-joining tree of predicted C. elegans
LGIC subunit genes revealed the presence of a distinct clade of 16 LGIC genes whose only characterized member was the serotonin-gated chloride channel subunit MOD-1 (11). Members of this clade have no orthologs in Drosophila or vertebrate genomes (Fig.  1A, data not shown). The newly identified clade appeared to break down into three subgroups that we reasoned might correspond to families of channels with distinct ligand specificities. We isolated and sequenced four cDNAs from one subgroup: ACC-1 (also known as F58G6.4), ACC-2 (C53D6.3), ACC-3 (F55D10.5), and ACC-4 (T27E9.9). Open reading frames of the ACC-1 (1401 bp), ACC-2 (1338 bp), and ACC-4 (1227 bp) cDNAs corresponded to those annotated in the genome data base. The ACC-3 cDNA (1554 bp) was transpliced with an SL1 leader sequence and differed from the predicted open reading frame in the first and seventh through ninth exons. All typical features attributed to LGIC subunits, such as a signal sequence, an N-terminal extracellular domain with the Cys loop, four transmembrane domains (M1-M4), and a large cytoplasmic loop located between M3 and M4, were recognizable in the amino acid sequences of these proteins (Fig. 1B). The ACC-1 and ACC-2 putative proteins are 40% identical to each other at the amino acid level and exhibit 31 and 29% identity to MOD-1, respectively. In contrast, ACC-2 shows 23% identity to UNC-49B, a C. elegans GABA-gated chloride channel subunit (27), and is only 16% identical to UNC-38, a C. elegans nAChR subunit.
ACC-1 and -2 Form ACh-gated Chloride Channels-To determine the ligand specificity of the ACC subunits, we expressed ACC cRNAs in Xenopus oocytes and voltage-clamped the oocytes at Ϫ80 mV. ACC-1-and ACC-2-injected oocytes exhibited an ACh-elicited inward current with maximal magnitude varying from 0.8 to 3.4 A (Fig. 2, A and C). The ACC-1-dependent current showed almost no desensitization even at saturating ACh concentrations, whereas the ACC-2-dependent current desensitized. ACC-1 responded to ACh with a half-effector concentration (EC 50 ) of 0.26 Ϯ 0.01 M and an estimated Hill coefficient of 1.26 Ϯ 0.04 (Fig. 2B), whereas ACC-2 was much less sensitive, responding with an EC 50 of 9.54 Ϯ 0.11 M and a Hill coefficient of 2.64 Ϯ 0.08 (Fig. 2D). Oocytes injected with ACC-3 cRNAs exhibited a weak response (10 -30 nA) to 1 mM ACh, and those injected with ACC-4 showed no response. An ACh-induced current was not detected in distilled H 2 O-injected or noninjected oocytes. Oocytes expressing ACC cRNAs did not respond to 1 mM GABA, glutamate, glycine, histamine, or dopamine (not shown). ACC-2injected oocytes responded slightly to 1 mM serotonin (Ͻ2% of the maximal ACh response) and to octopamine (ϳ1%), but no response to these compounds was observed in oocytes expressing other ACC subunits.
ACC-1, -2, and -3 share a proline-alanine motif of the second transmembrane domain (corresponding to the intermediate ring of nAChRs) that has been shown to confer anion selectivity in vertebrate GABA and glycine receptors (28 -30) (Fig. 1B). To determine whether ACC-1 and ACC-2 are also chloride channels, we generated I-V curves. The reversal potentials in ND96 for ACC-1 or ACC-2 homomeric channels were Ϫ18.7 Ϯ 1.4 mV and Ϫ18 Ϯ 1.6 mV, respectively, consistent with the equilibrium potential for chloride (Fig. 3, A and B) (31). When the nonpermeant anion gluconate was substituted for chloride (7.6 mM final external chloride), there was a positive shift in the reversal potential of 65.7 mV (ACC-1) and 68.6 mV (ACC-2). In contrast, when arginine was substituted for sodium, the shift in reversal potential was negligible (E reversal ϭ Ϫ17.2 Ϯ 1.7 mV, ACC-1, and Ϫ17.0 Ϯ 1, ACC-2) (Fig. 3, B and C). We observed shifts of 59.8 Ϯ 4.9 and 54.4 Ϯ 3.7 mV in the reversal potential for a 10-fold change in chloride concentration for ACC-1 and ACC-2, respectively (Fig. 3C). This is in agreement with the theoretical shift of 58 mV predicted by the Nernst equation for chloride-selective channels.
The Pharmacology of the ACC Channels Reflects a Unique Ligand-binding Site-The sequence identity between the ACCs and nAChRs in their extracellular ligand-binding domains is relatively low. More specifically, the six ligand-binding loops (A-F), as predicted from lineups with nAChRs, are not well conserved between nAChRs and the ACCs (Fig. 1B). Therefore, we predicted that the ACCs would have a unique pharmacological profile. We tested agonists and antagonists of vertebrate nAChRs on the ACC receptors. The classical nAChR agonist nicotine at 1 mM was a poor agonist of the ACC channels (Fig.  4A). At 0.5 mM, nicotine was a partial antagonist of ACC-1 but not ACC-2. Cytisine, another nAChR agonist, similarly acted as a weak agonist of ACC-2 channels and as an antagonist of both ACC-1 and -2. The highest potency antagonists had the rank order: D-tubocurarine (d-TC) Ն strychnine Ͼ atropine Ͼ dihydro-␤-erythroidine (d␤e) Ͼ hexamethonium (C6) for ACC-1, and d-TC Ն strychnine Ն d␤e Ͼ C6 for ACC-2 (Figs. 2 (E and F) and 4B). At 1 mM, C6 acted simultaneously as an agonist and an antagonist of ACC-1. Interestingly, 20 M d-TC, a competitive antagonist of nAChRs and 5-hydroxytryptamine type 3 receptors, blocked ACC-1 and -2 completely. Weak activation by nicotine and cytisine and block by strychnine, d-TC, and d␤e are characteristic of the vertebrate ␣9 nAChR channels (32,33) and of the Aplysia ACh-gated chloride channel (3). However, unlike these channels, ACC-1 and -2 were not blocked by ␣-bungarotoxin (␣-BT).
Although nicotinic agonists had little effect on ACC-1 and ACC-2, arecoline, an agonist of metabotropic ACh receptors, evoked a current that was 82-89% of the maximal ACh-elicited current (Fig. 2, A and C). Arecoline evoked a slowly desensitizing current in both the ACC-1-and ACC-2-expressing oocytes with an EC 50 of 4.7 Ϯ 0.11 and 754 Ϯ 22 M, respectively (Fig.  2, B and D). A rebound inward current was observed upon the removal of arecoline in both types of receptors, an indication of agonist-dependent open channel block (34,35). We noted that the estimated Hill coefficients for arecoline of 2.66 Ϯ 0.23 and 1.59 Ϯ 0.09 for the ACC-2-and ACC-1-expressing oocytes, respectively, are not significantly different from the Hill coefficients for the response to ACh, indicating a similar degree of cooperativity of ACh and arecoline. Atropine, a nonselective antagonist of metabotropic ACh receptors and a competitive antagonist of ␣9 nAChRs, activated the ACC-2 channel with an EC 50 of 873 Ϯ 63 M with an estimated Hill coefficient of 0.98 Ϯ 0.07 (Fig. 2D) but evoked only 43% of maximal ACh-elicited current. In contrast, we detected no atropine-evoked activation (up to 1 mM) of the ACC-1-expressing oocytes. Instead, atropine blocked the response of ACC-1 channels to 1 M ACh with an IC 50 of 23 M (Fig. 2). Thus, the pharmacological profiles of ACC-1 and -2 support a distinct ACC ligand-binding site.

ACC-3 and ACC-4 May Form Obligate
Heteromers-Because ACC-3 and ACC-4 do not respond robustly to ACh as homomers, we considered the possibility that they form obligate heteromers with other ACh-gated chloride channel subunits. ACC-3 forms a functional heteromeric channel with ACC-1. Co-expression of ACC-1 ϩ ACC-3 generated a channel that exhibited a pronounced desensitization compared with the ACC-1 homomer (Fig. 5A versus Fig. 2A). Moreover, the response of ACC-1 ϩ ACC-3-expressing oocytes to ACh was more than 200-fold less potent (EC 50 ϭ 39.6 Ϯ 1.6 M) than that of homomeric ACC-1 channels. The estimated Hill coefficient of 0.78 Ϯ 0.02 indicates that the ACC-1 ϩ ACC-3 heteromer has fewer ACh binding sites than the ACC-1 homomer (Fig. 5B). Coexpression of ACC-1 with ACC-4 did not change significantly the maximal response, EC 50 (0.36 Ϯ 0.02) or Hill coefficient (1.11 Ϯ 0.07) relative to the ACC-1 homomer (Fig. 5B). Thus, there is no indication that ACC-4 interacts with ACC-1.
Oocytes coexpressing the ACC-2 cRNA with the ACC-3 or ACC-4 cRNAs either exhibited a weak response of 50 -60 nA to 1 mM ACh or did not respond, respectively (Fig. 5C); nor did ACC-2 ϩ ACC-3-nor ACC-2 ϩ ACC-4-expressing oocytes respond to 1 mM serotonin, GABA, glutamate, glycine, or histamine. Inhibition was specific for ACC-2, since ACC-3 and ACC-4 did not inhibit expression of a glutamate-gated chloride channel subunit, AVR-15, or expression of ACC-1 ( Fig. 5C; see  above). We interpreted this result as indicating that ACC-3 and -4 are able to assemble with ACC-2 in a heteromeric channel and interfere with its gating or trafficking. DISCUSSION We have identified a new family of Cys loop LGICs, the nematode ACh-gated chloride channels. We report the first molecular characterization of an anion-selective ACh receptor and show that a distinct class of Cys loop LGICs has evolved to mediate inhibitory cholinergic neurotransmission. This is also the first evidence of ACh-gated chloride channels in nematodes and suggests that fast inhibitory cholinergic neurotransmission is more widespread in the animal kingdom than previously suspected.
The ACh-gated chloride channel subunits in C. elegans belong to the superfamily of Cys loop ligand-gated ion channels. As such, we would predict that the ACC channels are pentameric. We showed that both ACC-1 and ACC-2 form homomeric channels when expressed in Xenopus oocytes but also that ACC-1 interacts with ACC-3 and ACC-2 interacts with both ACC-3 and ACC-4. The interaction of ACC-1 with ACC-3 produces a channel that could function in vivo, albeit one with a lower sensitivity to ACh than the ACC-1 homomer. The interaction of ACC-2 with ACC-3 and -4 is more problematic, since these subunits appear to inhibit ACC-2. We suspect that these subunits assemble into heteromers but require additional subunits to form a functional channel. However, we cannot rule out the possibility that, although capable of assembling into a heteromer, these subunits are prevented from doing so in vivo or that ACC-3 and -4 negatively regulate ACC-2. Determining which subunits are co-expressed in vivo will help resolve this issue. Finally, the ability of the ACC-3 and -4 to associate with ACC-1 and -2 is consistent with the proposition that the ACCs constitute a family of ACh-gated chloride channel subunits.
One of the most unusual features of these channels is the association of gating by ACh with anion selectivity. We can point to a clear structural motif that accounts for the anion selectivity. The M2 transmembrane domains line the pore of the channel and determine ion selectivity (30). At the cytoplasmic end of this domain is a Pro-Ala-Arg motif that is found in most anion channels in C. elegans. Similar motifs are found in vertebrate channels, and site-directed mutagenesis of this motif has confirmed its importance in determining anion selectivity in GABA A and glycine receptors (28 -30).
The other striking property of the ACCs is their unusual ligand binding site. The sequences of ACC-1 and ACC-2 subunits differ substantially from both nematode and vertebrate nAChRs in their extracellular ligand-binding domains (Fig. 1B, data not shown). Photoaffinity labeling and mutagenesis studies (36), confirmed by analysis of the crystal structure of acetylcholine-binding protein (37), have identified residues that define the ACh binding site at the interface of two subunits. Residues on one side of the (ϩ)-subunit (the ␣ subunit in nAChRs) contribute loops A, B, and C. Residues on the other side of the adjacent (Ϫ)-subunit (␥ and ␦ in nAChRs) contribute loops D, E, and F, which form the complementary part of the binding pocket. However, even the loops forming the ligand-FIG. 5. ACC subunits interact. A, dose-response curves of co-expressed channels. ACh is a less potent agonist of the ACC-1/ACC-3 heteromer (‚) than of the ACC-1 homomer (•). Co-expression of ACC-4 with ACC-1 (E) has no effect on the potency of ACh. n is indicated in parentheses. The error bars represent S.E. B, response of the co-expressed ACC-1 and ACC-3 showing increased rate of desensitization of the heteromeric channel. C, mean maximal current response in oocytes expressing subunit combinations. ACC-3 and ACC-4 inhibit expression of ACC-2 but not expression of ACC-1 or the glutamate-gated chloride channel formed by AVR-15B. n is indicated in parentheses.
binding pocket are not conserved between the ACCs and the nAChRs (Fig. 1B). Most notably, the adjacent cysteines of the C loop, a hallmark of the ligand-binding nAChR ␣ subunits, are absent from the ACCs. Thus, ACCs may have evolved the ability to bind ACh independently of the nAChRs.
The unusual pharmacological profiles of the ACC subunits support a unique acetylcholine-binding site. That nicotine, the defining agonist of nAChRs, and cytisine, a related agonist, are weak agonists and/or antagonists of ACC-1 and -2 distinguishes the ACCs from most nicotinic receptors. The weak nicotine response is not, however, unique. The C. elegans levamisole receptors (nAChRs) are insensitive to nicotine, and it has been shown that C. elegans has both nicotinesensitive and -insensitive cation-selective ACh receptors (38). Moreover, nicotine and cytisine are antagonists of the vertebrate ␣9 nAChRs (33). More surprising is that arecoline is an efficient agonist of both ACC-1 and -2, although with substantially lower affinity than ACh. Arecoline has been postulated to act on cation-selective nematode and insect LGICs, so this sensitivity to arecoline may be a common feature of invertebrate ACh receptors (39,40). Finally, ␣-BT, which is selective for vertebrate ␣7 and ␣9 nAChRs in the central nervous system and the nAChRs of the neuromuscular junction, does not block the ACC channels. ␣-BT binds primarily to the C loop in nAChRs (41). Therefore, the lack of effect on ACC channels presumably reflects the inability of ␣-BT to bind the ACC subunit's divergent C loop sequences.
Are the Aplysia ACh-gated ion channels orthologs of the ACCs? There are pharmacological similarities that suggest they might be (3). One of the two Aplysia channels identified desensitized slowly, a characteristic of ACC-1 homomers and ACC-1/ACC-3 heteromers. Nicotine and cytisine were also poor agonists of the slowly desensitizing Aplysia channel. Potencies of the antagonists for nematode and mollusk receptors were in the same range. However, Aplysia receptors appear to have an EC 50 for ACh of Ͼ100 M. Perhaps most importantly, the Aplysia channels were blocked by ␣-BT, indicating that they share a much greater similarity to the nicotinic receptors, at least in the C loop, than do the ACC subunits. Ultimately, whether the Aplysia channels evolved independently will have to be determined by sequence analysis.
Economically important antiparasitic nematocides target Cys loop LGICs. Levamisole is an agonist of a subset of nicotinic-type acetylcholine receptors (42), and ivermectin activates glutamategated chloride channels (12). Both drug targets have properties that are unique to nematodes, making it possible to create drugs that are relatively ineffective agonists of homologous channels in the vertebrate host. The levamisole receptors are most similar to the vertebrate nicotine-sensitive nAChRs (␣1-6) in sequence but are not nicotine-sensitive, whereas glutamate-gated chloride channels are not found in vertebrates (38). Since ACCs are also not found in vertebrates, they are promising targets for the development of highly specific nematocides.