Molecular basis for convergent evolution of glutamate recognition by pentameric ligand-gated ion channels

Glutamate is an indispensable neurotransmitter, triggering postsynaptic signals upon recognition by postsynaptic receptors. We questioned the phylogenetic position and the molecular details of when and where glutamate recognition arose in the glutamate-gated chloride channels. Experiments revealed that glutamate recognition requires an arginine residue in the base of the binding site, which originated at least three distinct times according to phylogenetic analysis. Most remarkably, the arginine emerged on the principal face of the binding site in the Lophotrochozoan lineage, but 65 amino acids upstream, on the complementary face, in the Ecdysozoan lineage. This combined experimental and computational approach throws new light on the evolution of synaptic signalling.

Glutamate is an indispensable neurotransmitter, triggering postsynaptic signals upon recognition by postsynaptic receptors. We questioned the phylogenetic position and the molecular details of when and where glutamate recognition arose in the glutamate-gated chloride channels. Experiments revealed that glutamate recognition requires an arginine residue in the base of the binding site, which originated at least three distinct times according to phylogenetic analysis. Most remarkably, the arginine emerged on the principal face of the binding site in the Lophotrochozoan lineage, but 65 amino acids upstream, on the complementary face, in the Ecdysozoan lineage. This combined experimental and computational approach throws new light on the evolution of synaptic signalling. R apid conversion of chemical to electrical signals at the cell membrane is a hallmark of the animal nervous system. This process is mediated by ligand-gated ion-channels (LGICs), membrane-spanning channels that open rapidly upon recognition of a specific transmitter (agonist). Gating of excitatory, cation-selective LGICs leads to membrane depolarization, increased neuronal signalling and muscle contraction. Conversely, gating of inhibitory, anion-selective LGICs generally leads to membrane hyperpolarization, inhibiting responses to synaptic activity. Both cation-and anion-selective pentameric LGICs (pLGICs) are descended from an ancestral pLGIC present before the prokaryote-eukaryote dichotomy 1 . Cation-selective pLGICs typically respond to acetylcholine, whereas the anion-selective pLGICs respond to an especially wide variety of agonists, predominantly GABA and glycine in vertebrates and in addition, glutamate and biogenic amines in invertebrates 2,3 . Characteristics of the original anion-selective pLGIC and the molecular details of how the current repertoire of inhibitory signalling arose remains an open question that has major implications for our understanding of the evolution of animal neuromuscular control and development 4 .
In the case of glutamate, functional homo-pentameric glutamate-gated chloride channels (GluCls) have been characterized in the Ecdysozoan 5-8 and Molluscan lineages 9 . Phylogenetic analysis suggested that these may either be descended from a single ancient glutamate receptor that subsequently diversified to also produce receptors responding to other neurotransmitters or, more likely, represent descendants from distinct origins that independently converged on glutamate binding 9 . More recently, the characterization of several homomeric GluCls in the Platyhelminth, Schistosoma mansoni 10 , supported the idea that glutamate-binding pLGICs are ancient but could not provide additional clarification of a single or multiple origins. The inference that glutamate recognition has multiple origins is based largely on experimental verification of GluCls in a few species, coupled with the assumption that GluCl subunits form well-defined phylogenetic clades that are likely to respond to the same agonist. As such, the hypothesis proposed by Kehoe et al. 9 of multiple origins for glutamate recognition awaits a mechanistic insight. We have taken the approach that identifying specific functional amino acids essential for glutamate binding to GluCls will improve the precision with which we can identify the phylogenetic position of when and where glutamate recognition arose.
The binding of glutamate to the GluCl closely reflects that of the other amino acid agonists, GABA and glycine, to their receptors (GABA A Rs and GlyRs; reviewed recently 11 ): the agonist amine is coordinated by the ''aromatic box'' of the principal subunit, at one side of the intersubunit interface; and the agonist carboxyl (one of two for  glutamate) binds to positively charged and hydroxyl side chains of the complementary subunit at the opposing side of the interface. The aromatic box and the positively charged and hydroxyl side chains (hereafter referred to as the amino acid agonist motif, ''AAA motif'') are highly conserved in agonist-binding GluCl, GlyR and GABA A R subunits, in which the mutation of these residues greatly reduces agonist potency [12][13][14][15][16][17] . Glutamate, however, is unique in that its recognition also requires the coordination of an additional carboxyl moiety. Here, we identify the amino acids of Ecdysozoan, Molluscan and Platyhelminth GluCls that underlie the coordination of this carboxyl moiety, using a combined computational, experimental and phylogenetic approach. These results confirm that glutamate recognition by the pLGIC family has evolved at least three times, with discrete parts of receptor sequence utilized for carboxyl-binding in the Lophotrochozoan and Ecdysozoan lineages.

Results
Putative determinants of glutamate recognition in diverse GluCls. Alignment of Arthropod, Nematode, Mollusk and Platyhelminth sequence from subunits that form experimentally verified GluCls revealed a highly conserved motif responsible for glutamate recognition. This includes the AAA motif, comprising three aromatic and one threonine residue from principal Loops A, B and C (cyan arrows, Fig. 1b) and one arginine and one serine residue from complementary Loops D and E (magenta arrows, Fig. 1b). This motif is present in essentially all GluCls, suggesting that the mode in which glutamate binds is similar in all GluCls, with the amino nitrogen coordinated at the principal face in the aromatic box and the c-carboxyl group coordinated at the complementary face by the Loop D arginine and Loop C and E hydroxyl side chains. This interpretation is supported by the fact that in GluCls, GABA A Rs and GlyRs that share the AAA motif, agonists bind in the same orientation, according to structural studies 18 , homology modelling 16,19 and mutagenesis data [12][13][14][15][16][19][20][21] . According to the structure of the glutamate-bound GLC-1 GluCl 18 from C. elegans, the coordination of the additional a-carboxyl moiety of glutamate is achieved by positively charged Loop F lysine and Loop G arginine side chains, present in this GluCl but absent from GABA A Rs and GlyRs 18 . Our alignment showed that the Loop F lysine is moderately conserved but the Loop G arginine is absolutely conserved in ecdysozoan GluCls (Fig. 1b). The critical role of the Loop G arginine is supported by mutagenesis in GLC-2 from C. elegans 14 . These positively charged side chains are not conserved, however, in lophotrochozoan GluCl subunits (Fig. 1b). This raises the question as to how lophotrochozoan GluCls accommodate the glutamate acarboxyl, or if indeed glutamate binds in the same mode to these channels.
To address these questions, we used Modeller software 22 to build homology models of the AVR-14B GluCl from the nematode Haemonchus contortus and the SmGluCl-2 from the platyhelminth S. mansoni, based on the C. elegans GluCl structural template 18 , as described previously 23 . The reliability of the models is enhanced by high sequence identity (.40%) with the template, particularly in the agonist-binding loops (Fig. 1b). In silico docking of glutamate to the agonist-binding site of these models, using Autodock Vina 24 , indicated that glutamate interacted with the AAA motif similarly in both models (Fig. 1c, d), supporting the notion that the binding mode is common to all GluCls. There is, however, a striking difference between Ecdysozoan and Lophotrochozoan GluCl interactions with the glutamate a-carboxyl. In the absence of a positively charged side chain in Loops F or G of the complementary face of SmGluCl-2, an arginine side chain from Loop A of the principal face approaches the glutamate a-carboxyl to within 2.8 Å (Fig. 1d). This position is occupied by an arginine in all Lophotrochozoan GluCl sequences, whereas it is occupied by serine, proline or glutamine in all ecdysozoan GluCl sequences (Fig. 1b). Based on this conservation and our Averaged glutamate dose-response data. Mean (6s.e.m.) peak glutamategated currents were normalized to maximum glutamate-gated current (n 5 3-7), except for R95A, A197E and Y199F (n 5 3-6), which were normalized to averaged ivermectin-gated current amplitude (reported in Table 1). Data are fit with non-linear regression for illustration. modelling, we therefore hypothesize that an arginine residue is responsible for coordinating the glutamate a-carboxyl at the ''bottom'' of the agonist-binding site in all GluCls; remarkably, in lophotrochozoan GluCls, this arginine comes from the opposite side of the intersubunit interface, some 65 amino acid positions downstream (see Fig. 1a) of the arginine in ecdysozoan GluCls.
Arginine at either face of the agonist binding site is sufficient for glutamate recognition. If the presence of an arginine in either position were a valid basis for glutamate recognition, we reasoned that an arginine in either Loop A or Loop G should be sufficient for glutamate recognition. We tested this idea by mutating these positions in the H. contortus AVR-14B GluCl, and measuring glutamate-gated currents when expressed in Xenopus laevis oocytes. Before experimentally testing the role of the arginine at the bottom of the site, we sought experimental verification that glutamate binds to AVR-14B as predicted computationally.
To this end, we mutated the AAA motif, generating Y199F (Loop B), T245A and Y248F (Loop C), R95A (Loop D) and S169A (Loop E) single-mutants. In our in silico docking, the phenyl rings of both Y199 and Y248 appear to sandwich the agonist amine, but the hydroxyl moiety of only Y199 is involved in direct interactions, both with the agonist c-carboxyl and with the nearby S169 hydroxyl side chain (Fig. 2a). Accordingly, removal of these hydroxyl side chains via the Y199F and S169A substitutions dramatically decreased glutamate potency, whereas the Y248F substitution had no significant effect (Fig. 2b, c; Table 1). Our results also supported a principle role for the Loop D arginine in coordinating the agonist c-carboxyl (Fig. 1b, c), as alanine substitution of R95 abolished and T245 decreased glutamate potency (Fig. 2b, c). An A197E mutation was designed to test the apparent requirement for glutamate recognition of a small, uncharged side chain at this position (initial residue in Loop B in Fig. 1b). This contrasts with a negatively charged glutamate residue at this position in GABA A Rs and GlyRs that is proposed to interact with the agonist primary amine, directly or via a water molecule 16,25,26 . Indeed, the A197E substitution abolished glutamate-gated currents (Fig. 2b, c), consistent with predictions that a glutamate residue at this position is better suited to GABA or glycine recognition 25 . Finally, K219A was designed to assess the possibility that this Loop F side chain (Fig. 1b) contributes to glutamate-binding, as is apparently the case in the C. elegans GLC-1 GluCl 18 . However, the K219A mutant showed similar glutamate sensitivity to WT (Fig. 2c; Table 1), indicating that this interaction is not a requirement. To verify that glutamate recognition, rather than struc-tural integrity, was affected by these mutations, 1 mM ivermectin was shown to gate robust currents at glutamate-insensitive mutants ( Fig. 2b; Table 1). In summary, these experiments point towards a common mode of glutamate recognition in ecdysozoan GluCls, where the agonist amine is sandwiched by the aromatic box and the c-carboxyl is oriented towards Loop D of the complementary face.
Finally, we addressed the ability of arginines at the bottom of the site to recognize the a-carboxyl of glutamate. To establish a ''latent'' GluCl without this ability, we generated the AVR-14B R76N mutant, replacing the Loop G arginine with a shorter, uncharged asparagine side chain, present in, for example, GlyR subunits (Fig. 1b). Although the R76N mutant was functionally expressed, as demonstrated by large ivermectin-gated currents (Table 1), it was not substantially gated by glutamate, even up to 100 mM (Fig. 3a, b), confirming the latency of the R76N GluCl. To then introduce a Loop A arginine, we made Q141R and K219A substitutions on the R76N background. Q141R introduces into Loop A the positively charged side chain present at this position in lophotrochozoan GluCls; the K219A mutation eliminates the possibility that in the absence of a Loop G arginine, this lysine residue could itself bind the glutamate a-carboxyl. This was apparently not the case for the R76N single-mutant, but we felt it necessary to remove this possibility, given the flexibility of lysine side chains in ligand binding 27 . At R76N/Q141R/K219A triple-mutant GluCls, glutamate gated robust currents that did not differ significantly from the original AVR-14B in terms of peak amplitude ( Fig. 3a; Table 1), indicating that a Loop A arginine does indeed confer glutamate recognition on the latent GluCl.
Glutamate potency at this triple-mutant was, however, much less than at WT AVR-14B GluCls, as indicated by an EC 50 value of 34 6 3 mM (n 5 5), almost three orders of magnitude greater than WT AVR-14B (Table 1). We also tested a GluCl with both Loop A and Loop G arginines. Notably, this Q141R single-mutant showed a similarly low glutamate potency (EC 50 5 6 6 1 mM; n 5 5) to R76N/Q141R/K219A (Fig. 3b). Furthermore, both Q141R (n H 5 0.9 1 0.1) and R76N/Q141R/K219A (n H 5 0.9 1 0.1) were the only mutants tested that showed significantly decreased cooperativity, as indicated by lower Hill coefficients than WT (Table 1). We loosely interpret the similarities between these two mutants as evidence that, at least in the AVR-14B GluCl, a Loop A arginine side chain can ''take over'' the interaction with the glutamate a-carboxyl, irrespective of the presence of the Loop G arginine. Indeed, the two arginines within the Q141R receptor are likely to be mutually repulsive, and interactions with the negatively charged D195 residue (Fig. 3c) -present in   AVR-14B but absent from SmGluCl-2 -may contribute to the fact that these mutant AVR-14B GluCls have lower glutamate affinity than WT AVR-14B or SmGluCl-2 GluCls. These data provide experimental support for our hypothesis that the presence of a principal face or a complementary face arginine residue forms the molecular basis for the convergent evolution of glutamate recognition.
Phylogenetic analysis indicates polyphyletic origins of glutamate recognition. The origin of glutamate recognition in the pLGICs was previously unclear. There remained a possibility that the Ecdysosoan-like (E-like), Schistosome-like (S-like) and Aplysia-like (A-like) GluCls were paraphyletic, sharing a common ancestor that recognized glutamate, with receptors recognizing, for example, GABA or glycine also sharing this GluCl common ancestor. Here we can conclude that the E-like GluCls have a unique, monophyletic origin based on the fact that they coordinate glutamate binding through a Loop-G arginine that does not exist in the S-like or Alike GluCls. To determine whether the S-like and A-like GluCls are paraphyletic requires examination of the phylogenetic origin of the glutamate-binding motif that we have identified. The unrooted phylogeny in Fig. 4 highlights, in blue, the three distinct subunit clades that possess the glutamate-binding motif. Subunits from nematodes, insects and vertebrates representing previously identified major anionic pLGIC classes are also included 2 . A large number of predicted anionic pLGIC subunits can be identified in the recently available genomes of the gastropods, Biomphalaria glabrata and Lottia gigantea, the bivalve, Crassostrea gigas, the oligochaete Capitella teleta, the clitellate, Helobdella robusta and the turbellarian Schmidtea mediterranea. Many of these do not possess the AAA motif or the complete glutamate-binding motif and appear to represent several new subunit classes. The A-like GluCl subunits form a monophyletic clade in the gastropods and are derived from subunits that would not have been able to bind glutamate, due to the absence of a crucial arginine and divergence from the AAA motif (Fig. 4). The relatively recent, derived position of the A-like GluCl subunits within the phylogeny confirms that the A-like and S-like GluCls converged on glutamate binding independently. The S-like GluCl clade is distinct and all subunits within share the AAA motif and Loop A arginine, with the exception of two S. mediterranea subunits that appear to have lost the Loop A arginine and are highly divergent from the other subunits (Fig. 4). All S. mediterranea subunits identified fall within the S-like GluCl clade, except for one GABA A R-like subunit, suggesting that glutamate is a major inhibitory neuromuscular transmitter of Platyhelminths. Of particular interest is a class of four subunits from the Annelida C. taleta and H. robusta that are within with the S-GluCl clade and share the lophotrochozoan glutamate-binding motif (e.g. HroP83631 in Fig. 4b). Current understanding of protostome taxonomy places Mollusca and Annelida as sister groups, descendent from a common ancestor that together with the Platyhelminthes form the Lophotrocozoa 28 . A shared S-like GluCl implies that the common ancestor of all Lophotrocozoa possessed this GluCl. The fact that to date no mollusk subunit has been identified that falls within this clade suggests that glutamate response mediated by the S-like GluCl was lost from the Mollusca. Whether this was before, after or coincident with appearance of the A-like GluCl is not clear. Analysis of this phylogeny shows three independent origins for glutamate recognition. We would expect that the S-and A-like GluCls emerged from glycine-binding subunits because the closely related subunits (orange in Fig. 4) that precede both these GluCls possess the initial Loop B glutamate residue conducive to glycine-and GABAbinding (see Arginine at either face of the agonist binding site is sufficient for glutamate recognition) and lack the closing Loop C arginine residue (Fig. 1b) required for high-affinity GABAbinding 15,29 .   Discussion There are precedents for convergent paths to agonist recognition in diverse pLGICs, although perhaps less conspicuous than the arrival of GluCls at glutamate recognition via opposite sides of the intersubunit interface. For example, cation-p interactions with the agonist amine are common to numerous pLGICs, but the interacting aromatic side chain is from Loop A in b2-containing GABA A Rs 30 , Loop B in r 1 GABA A Rs 31 and a1 GlyRs 17 , Loop C in GLC-2 GluCls 14 and Loops B and C of RDL GABA A Rs 32 . Based on the position of the glutamate amine in our computational dockings, it seems that the main cation-p interaction may involve the Loop C aromatic of AVR-14B but the Loop B aromatic of SmGluCl-2. Together with results at C. elegans GLC-2 GluCls 14 , our docking tentatively suggests that the cation-p interaction may favour Loop C when the glutamate a-carboxyl is coordinated by a complementary face arginine. Perhaps more dramatic are the different means by which murine 5-HT-gated cation channels (5-HT 3 R) and C. elegans 5-HT-gated chloride channels (MOD-1) bind 5-HT. The amine of 5-HT forms a cation-p interaction with a Loop B tryptophan in the former and a Loop C tryptophan in the latter 33,34 , such that the same agonist probably binds in very different orientations in the two receptors 34 .
One limitation of our study is the focus on homo-pentameric receptors. This was required to accurately annotate putative GluCls in relation to verified GluCls, but as such, we may have excluded subunits that could, for example, form only the principal face of a glutamate-binding hetero-pentamer and thus lack the complementary face requirements for glutamate recognition. However, two lines of evidence suggest that our analysis, in which glutamate-binding principal and complementary faces are required for annotation as a GluCl, truly reflects the evolution of GluCls. First, most GluCls identified thus far form glutamate-gated homo-pentamers when expressed recombinantly (Fig. 1). Moreover, those subunits that form hetero-pentamers -additionally or exclusively -also possess both the principal and complementary subunit motives we have discussed 6,10 . Second, pharmacological and immunocytochemical experiments suggest that native GluCls occur as homomers and heteromers 35,36 , but all subunits contributing to these characterized native receptors -including AVR-14B employed in our experiments -contain the homo-pentameric glutamate recognition motif.
In summary, this study successfully employed a combination of computational, experimental and phylogenetic analyses to unravel the molecular basis of convergent evolution of glutamate recognition in pLGICs. This approach sheds new light on the origins of synaptic signalling and should prove similarly enlightening when applied to ligand-receptor interactions in other protein families. The finding that pLGICs recognizing glutamate have arisen independently at least three times suggests that care should be taken when inferring the pharmacological profile of ancestral receptors based solely on phylogeny. The suggestion that other pLGICs specific for acetylcholine 37 and serotonin 34 have arisen by convergent evolution implies that neurotransmitter signalling is more adaptable than has been appreciated before. In fact, the recent analysis of the ctenophore genome suggests that a neuromusculature regulated by glutamate may itself have multiple independent origins 4 .

Methods
Sequence alignments and computational dockings. The amino acid sequence alignment in Fig. 1 was performed in ClustalW2 38 using sequences and agonist sensitivity information from original publications for D. melanogaster GluCla 5 and RDL 39,40 , M. domestica GluCla 41 , C. elegans GLC-1 and 22 6 , GLC-3 42 and AVR-15 43 , H. contortus GLC-5 44 , AVR-14B 45 and UNC-49B 46 , A. californica GluCl1 and 2 9 , S. mansoni GluCl-2.1 and 3 10 and Homo sapiens a1 GlyR 47 . Only portions of the ligandbinding loops were shown, in order to highlight the side chains that interact with ligands according to the GLC-1/glutamate crystal structure 18 , mutagenesis data 14 , and to our computational dockings. Portions of the channel-lining M2 helix were shown to illustrate the anion-selective motif of the channels 2 . Homology models of H. contortus AVR-14B, R76N/Q141R/K219A AVR14-14B and S. mansoni SmGluCl-2 were built on the C. elegans GLC-1/glutamate/ivermectin crystal structure (PDB 3RIF 18 ) using Modeller 22 . Simulated glutamate docking to these models was performed with Autodock Vina 24 , allowing for a limited set of flexible side chains (those illustrated in Figure 1c except for the Loop F lysine; the two residues preceding the conserved Loop C threonine; and the residue two positions upstream of the Loop D arginine), within a 20 Å -sided cube encompassing the glutamate binding site. Agonist docked poses with the lowest predicted energy were selected for further analysis (shown in Fig. 1).
Site-directed mutagenesis and electrophysiology. All mutant cDNAs were generated from AVR-14B cDNA in the T7 plasmid vector 45 by conventional PCR with Pfu DNA polymerase (Thermo Scientific, Germany) and commercially generated mutagenesis primers (Eurofins MWG Operon, Ebersberg, Germany). All cDNAs were sequenced to confirm the presence of appropriate mutations (Eurofins MWG Operon) and linearized with XbaI (Thermo Scientific). cRNAs were synthesized with the mMESSAGE mMACHINE T7 Kit (Life Technologies GmbH, Darmstadt, Germany). X. laevis oocytes were used for two electrode voltage-clamp electrophysiology as previously described 48 and under approval of the Technical University of Darmstadt (Agreement V54-19c20/15 DA8/Anz. 20). After 2-5 days, oocytes were perfused with extracellular solution (115 mM NaCl, 1 mM KCl, 1.8 mM CaCl 2 , 10 mM HEPES, pH 7.4 with NaOH) alone or containing agonist, and currents were sampled at 1 kHz and filtered at 500 Hz with a Digidata 1322A digitizer, Geneclamp 500B amplifier and Clampex software (Molecular Devices, Sunnyvale, CA, USA). Current responses to glutamate were plot against log glutamate concentration and fit with non-linear regression in Prism4 (GraphPad Software Inc., San Diego, CA, USA). Non-linear regression parameters were compared to the respective WT value with Student's unpaired t tests (Prism4).