Functional Conservation and Genetic Divergence of Chordate Glycinergic Neurotransmission: Insights from Amphioxus Glycine Transporters

Glycine is an important neurotransmitter in vertebrates, performing both excitatory and inhibitory actions. Synaptic levels of glycine are tightly controlled by the action of two glycine transporters, GlyT1 and GlyT2, located on the surface of glial cells and neurons, respectively. Only limited information is available on glycinergic neurotransmission in invertebrates, and the evolution of glycinergic neurotransmission is poorly understood. Here, by combining phylogenetic and gene expression analyses, we characterized the glycine transporter complement of amphioxus, an important invertebrate model for studying the evolution of chordates. We show that amphioxus possess three glycine transporter genes. Two of these (GlyT2.1 and GlyT2.2) are closely related to GlyT2 of vertebrates, whereas the third (GlyT) is a member of an ancestral clade of deuterostome glycine transporters. GlyT2.2 expression is predominantly non-neural, whereas GlyT and GlyT2.1 are widely expressed in the amphioxus nervous system and are differentially expressed, respectively, in neurons and glia. Vertebrate glycinergic neurons express GlyT2 and glia GlyT1, suggesting that the evolution of the chordate glycinergic system was accompanied by a paralog-specific inversion of gene expression. Despite this genetic divergence between amphioxus and vertebrates, we found strong evidence for conservation in the role glycinergic neurotransmission plays during larval swimming, the implication being that the neural networks controlling the rhythmic movement of chordate bodies may be homologous.


Introduction
Glycine is a major inhibitory neurotransmitter in the vertebrate spinal cord, brainstem, and retina, where it activates ionotropic glycine receptors (GlyR) [1][2][3][4]. In addition, glycine acts as a co-agonist of N-methyl-D-aspartate (NMDA) glutamate receptors [5,6]. Consequently, glycine has a dual role in the vertebrate central nervous system (CNS), acting as an inhibitory neurotransmitter at glycinergic synapses and as a positive modulator of excitatory glutamatergic synapses. In either case, glycinergic neurotransmission is regulated by clearance of glycine from the synaptic space by high-affinity Na + /Cl − -dependent glycine transporters belonging to the solute carrier family 6 (SLC6) [7]. SLC6 proteins are widely distributed across eukaryotes, serving as transporters of various signaling molecules in unicellular eukaryotes and fungi and as transporters of monoamine and amino acid neurotransmitters in metazoans [8].
In vertebrates, two genes encode glycine transporters, GlyT1 and GlyT2, corresponding to SLC6 members 9 and 5, respectively [9][10][11][12]. Both glycine transporter genes are alternatively spliced to generate multiple variants. In rodents, the GlyT1 protein has five isoforms: three differing in the N-terminal region (GlyT1a, GlyT1b, and GlyT1c) and two differing at the C-terminal end (GlyT1d and GlyT1e) [13][14][15]. GlyT2 protein isoforms (GlyT2a, GlyT2b, and GlyT2c) differ only in their N-terminal region [9,16,17]. GlyT1 and GlyT2 show different pharmacological and kinetic proprieties [7,18] as well as differential expression in the nervous system. GlyT1 expression is widespread in the CNS, whereas GlyT2 is mainly restricted to the hindbrain and spinal cord [19,20]. Differences in glycine transporter subtype expression are also observed at the cell type level: GlyT1 is found on astroglial cells and glutamatergic neurons, whereas GlyT2 localizes to glycinergic presynaptic terminals [5]. The observation that some anti-GlyT1 antibodies predominantly stain glial cells associated with glycinergic synapses in the caudal region of the CNS, whereas others stain both astroglia and glutamatergic neurons throughout the CNS [21], further suggests a differential expression of GlyT1 isoforms. Taken together, the expression data, combined with knockout experiments [22,23], indicate that GlyT2 is responsible for replenishing the glycine pool of the presynaptic terminal, whereas GlyT1 is responsible for terminating glycine signaling at both glycinergic and glutamatergic synapses.
Much less is known about glycine transporters, and glycinergic neurotransmission in general, in invertebrates ( Figure 1). Indeed, a functional description of glycinergic neurotransmission is only available for the sea hare Aplysia californica (phylum Mollusca), a classical neurobiology model, the main invertebrate model organism Drosophila melanogaster (phylum Arthropoda), and the tadpole of the sea squirt Ciona intestinalis (phylum Tunicata), an important model in evolutionary developmental biology [24][25][26][27]. Conversely, the roundworm Caenorhabditis elegans (phylum Nematoda) apparently lacks glycinergic transmission [28]. Importantly, biochemical evidence for the presence of a functional glycine receptor in the freshwater polyp Hydra vulgaris (phylum Cnidaria) [29] suggests that glycinergic transmission was already present in the earliest metazoans and subsequently lost in some lineages, such as nematodes. Yet, the tissue-specific distribution of glycine transporters has so far only been investigated in a single invertebrate taxon: D. melanogaster [30].
Shpak and coworkers analyzed genomic databases of selected invertebrate deuterostomes (a cephalochordate, an ascidian, and a sea urchin) and retrieved, for each species, two glycine transporter paralogs, which they called GlyT1-like and GlyT2-like. According to their phylogeny, the GlyT2-like genes of invertebrate deuterostomes are bona fide orthologs of vertebrate GlyT2, whereas the invertebrate GlyT1-like genes represent ancestral deuterostome glycine transporters, from which both the GlyT2 genes of invertebrate deuterostomes and all vertebrate glycine transporter genes evolved [31]. However, the paucity of data on glycinergic neurotransmission among invertebrates makes the evolutionary history of this neurotransmitter difficult to trace. To address this issue, we analyzed the glycine transporters complement of amphioxus, an invertebrate with a key phylogenetic position within the chordates, to define the evolutionary events that took place during the early diversification of vertebrates [32].
Amphioxus (phylum Cephalochordata) is the earliest branch in the chordate lineage, and hence the sister group of the vertebrate plus tunicate clade ( Figure 1). Amphioxus has a prototypical chordate body plan with a nervous system consisting of a CNS and a peripheral nervous system (PNS) organized into ectodermal plexuses. The CNS of amphioxus larvae is a hollow tube that is slightly expanded anteriorly to form a cerebral vesicle [33]. Despite lacking overt anatomical landmarks, molecular evidence suggests that the amphioxus CNS is vertebrate-like and regionalized into an anterior hypo-prethalamic, an intermediate dien-mesencephalic, and a posterior rhombospinal region [34]. The neurochemistry of the amphioxus nervous system is similar to that of vertebrates [35,36]. Of particular interest is the presence of evenly-spaced groups of glycinergic neurons in the rhombospinal regions, at the level of somites 2 to 5, with the first two groups also containing GABAergic neurons [35] and with astroglia-like cells being abundant in the whole rhombospinal region [37]. Here, we analyzed the genome of three amphioxus species to characterize the glycine transporter repertoire of cephalochordates. We retrieved additional sequences compared with previous work [31] and report their developmental expression patterns in the amphioxus Branchiostoma lanceolatum. Based on an updated phylogeny of deuterostome glycine transporters, this analysis revealed expression of amphioxus glycine transporters in neurons and glial cells, further strengthening the previously proposed homology of amphioxus and vertebrate glial cell populations [37]. However, we found the paralogspecific expression domains of amphioxus and vertebrate glycine transporter genes to be inversed, hinting at a particularly complex evolutionary history of this gene family. Our results nonetheless provide strong support for the evolutionary conservation, at least in chordates, of glycinergic neurotransmission in the control of larval swimming, also highlighting the need for further work on glycine signaling in other invertebrates. that the amphioxus CNS is vertebrate-like and regionalized into an anterior hypoprethalamic, an intermediate dien-mesencephalic, and a posterior rhombospinal region [34]. The neurochemistry of the amphioxus nervous system is similar to that of vertebrates [35,36]. Of particular interest is the presence of evenly-spaced groups of glycinergic neurons in the rhombospinal regions, at the level of somites 2 to 5, with the first two groups also containing GABAergic neurons [35] and with astroglia-like cells being abundant in the whole rhombospinal region [37]. Here, we analyzed the genome of three amphioxus species to characterize the glycine transporter repertoire of cephalochordates. We retrieved additional sequences compared with previous work [31] and report their developmental expression patterns in the amphioxus Branchiostoma lanceolatum. Based on an updated phylogeny of deuterostome glycine transporters, this analysis revealed expression of amphioxus glycine transporters in neurons and glial cells, further strengthening the previously proposed homology of amphioxus and vertebrate glial cell populations [37]. However, we found the paralog-specific expression domains of amphioxus and vertebrate glycine transporter genes to be inversed, hinting at a particularly complex evolutionary history of this gene family. Our results nonetheless provide strong support for the evolutionary conservation, at least in chordates, of glycinergic neurotransmission in the control of larval swimming, also highlighting the need for further work on glycine signaling in other invertebrates.  [38,39] showing evolutionary relationships of the main animal phyla. The dashed lines indicate that the position of the Xenacoelomorpha is still a matter of debate [40][41][42], which might potentially affect the monophyly of the deuterostomes [43]. The earliest branches of the tree are currently also unresolved, with different phylogenies placing either the Ctenophora or the Porifera as the sister group to all other animals [44][45][46]
Protein sequences were aligned using the MUSCLE algorithm in MEGA6 [56]. The alignment used to generate Figure 2 is provided in FASTA format as Supplementary File S1. The multiple sequence alignment was uploaded to the Phylogeny.fr platform [57], and positions with gaps were removed. A first phylogenetic tree was calculated using the Bayesian Inference (BI) method as implemented in the MrBayes program (v3.2.6) [58]. The number of substitution types was fixed to 6. A Poisson model was used for amino acid substitution, whereas rate variation across sites was fixed to "invgamma". Four Markov Chain Monte Carlo (MCMC) chains were run for 10,000 generations, sampling every 10 generations, with the first 250 sampled trees discarded as "burn-in". A 50% majority rule consensus tree was constructed. Support for internal branches was evaluated by calculating posterior probabilities. The tree was graphically rendered with TreeDyn (v198.3) [59]. The evolutionary history of glycine transporter sequences was further inferred using the maximum likelihood (ML) and neighbor-joining (NJ) methods in MEGAX [60]. The most appropriate evolutionary models for phylogenetic inference were identified using the Model Selection tool implemented in MEGAX, selecting models with the lowest BIC scores. For the ML tree, a JTT model with Gamma distribution (JTT + G) was the best fit, and the rate variation among sites was modeled with a gamma distribution using a shape parameter of 0.75 in an analysis of 33 amino acid sequences. All positions containing gaps and missing data were eliminated. A total of 538 positions were used in the final dataset. For the NJ tree, evolutionary distances were computed using the same JTT matrix-based method. For both the ML and NJ analyses, support of internal branches was evaluated by 1000 replicates of non-parametric bootstrapping.

Whole Mount In Situ Hybridization
The expression patterns of amphioxus glycine transporter genes were determined by whole mount in situ hybridization according to previously published methods [47]. Stained whole mount embryos were mounted with glycerol and photographed using an IX71 inverted microscope (Olympus, Hamburg, Germany) equipped with a ColorViewII camera (Olympus, Hamburg, Germany). Subsequently, selected embryos were retrieved from the slides, rinsed with distilled water, counterstained with 1% Ponceau S in 1% acetic acid, dehydrated with an ethanol series, and embedded in Spurr's resin (Sigma-Aldrich, St. Louis, MO, USA) [61]. Serial 3 µm sections were obtained using an RM2145 microtome (Leica, Wetzlar, Germany) equipped with a glass knife, mounted in mineral oil, and photographed. For double-label in situ hybridization, embryos were hybridized simultaneously with two probes labeled with digoxigenin or fluorescein following published protocols [61]. Staining was performed, respectively, with NBT/BCIP (Roche, Penzberg, Germany,) or Fast Red (Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturers' instructions. Whole mount embryos were mounted with glycerol and photographed using an IX71 inverted microscope (Olympus, Hamburg, Germany) equipped with a ColorViewII camera (Olympus, Hamburg, Germany) both in brightfield and in epifluorescence using a Texas Red filter to visualize fluorescent Fast Red staining. Selected embryos were retrieved from the slides, rinsed with distilled water, incubated in 20% sucrose overnight, embedded in Killik (Bio-Optica, Milan, Italy), and frozen in liquid nitrogen. Cross-sections were obtained using a CM1900UV cryostat (Leica, Wetzlar, Germany), thawed to room temperature, rinsed in PBS, mounted in 20% glycerol, and photographed.

Expression of GlyT, GlyT2.1, and GlyT2.2 in Developing Amphioxus
In early neurulae (N1 stage), GlyT was broadly expressed in the anterior two-thirds of the neural plate and the forming somites ( Figure 3A-D). In N4 neurulae, expression was mostly restricted to the CNS ( Figure 3E) and particularly to lateral cells of the cerebral vesicle and anterior rhombospinal region ( Figure 3F-H). A faint signal was also observed in the pharyngeal and caudal regions. In T1-stage embryos, GlyT was expressed in the cerebral vesicle ( Figure 3I-J) and in ventrolateral cells of the rhombospinal region up to the end of somite 6 ( Figure 3I,K-M). GlyT started also to be expressed in the developing Neighbor Joining (NJ) methods were used, and the BI tree is shown with branch support values for the BI, ML, and NJ analyses, respectively. "-" indicates branches not recovered by the corresponding tree reconstruction method. Branch lengths are expressed as amino acid substitutions per site. Caenorhabditis elegans neurotransmitter transporter snf11 was used as the outgroup. List of animal species featured in the tree: Acropora millepora (coral, cnidarian), Bos taurus (cattle, mammal), Branchiostoma belcheri (Chinese amphioxus, cephalochordate), Branchiostoma floridae (Florida amphioxus, cephalochordate), Branchiostoma lanceolatum (European amphioxus, cephalochordate), Caenorhabditis elegans (roundworm, nematode), Ciona intestinalis (sea squirt, tunicate), Danio rerio (zebrafish, teleost fish), Gallus gallus (chicken, bird), Homo sapiens (human, mammal), Petromyzon marinus (sea lamprey, jawless vertebrate), Strongylocentrotus purpuratus (purple sea urchin, echinoderm), Xenopus laevis (African clawed frog, amphibian).

Identification and Phylogenetic Analysis of Amphioxus Glycine Transporter Genes
We identified three orthologs of vertebrate glycine transporters in cephalochordates ( Table 1), two of which have previously been identified in B. floridae and named GlyT1-like and GlyT2-like [31]. However, based on our phylogenetic reconstructions (Figure 2), the name of the GlyT1-like gene was changed to GlyT. The phylogenetic tree further demonstrated that the GlyT, GlyT2.1 and GlyT2.2 sequences from B. lanceolatum, B. floridae, and B. belcheri are homologous ( Figure 2), a notion supported by an alignment of the amino acid sequences of the different glycine transporters from these three amphioxus species (Supplementary Table S2). We further found that, at least in B. floridae and B. lanceolatum, the GlyT2.1 and GlyT2.2 genes were located on the same scaffold in the genome and only about 100 kb apart ( Table 1), suggesting that the amphioxus GlyT2.1 and GlyT2.2 genes originated from a lineage-specific duplication event. Interestingly, two GlyT2 paralogs were also identified in sea urchins, but not ascidian tunicates or vertebrates ( Figure 2).
In the tree, the vertebrate GlyT1 and GlyT2 sequences were clearly separated, falling into two distinct and very strongly supported clades (Figure 2). The amphioxus GlyT2.1 and GlyT2.2 sequences were basal to the ascidian tunicate plus vertebrate GlyT2 sequences, and the echinoderm GlyT2.1 and GlyT2.2 sequences were basal to all chordate GlyT2 sequences. In accordance with previously published phylogenies [31], our phylogenetic analysis revealed that invertebrate deuterostomes lack true GlyT1 orthologs, but rather possess GlyT-like sequences that branch at the base of the clade composed of the GlyT2s and the vertebrate-only GlyT1s. We propose to refer to all members of this group as GlyTs ( Figure 2).

Expression of GlyT, GlyT2.1, and GlyT2.2 in Developing Amphioxus
In early neurulae (N1 stage), GlyT was broadly expressed in the anterior two-thirds of the neural plate and the forming somites ( Figure 3A-D). In N4 neurulae, expression was mostly restricted to the CNS ( Figure 3E) and particularly to lateral cells of the cerebral vesicle and anterior rhombospinal region ( Figure 3F-H). A faint signal was also observed in the pharyngeal and caudal regions. In T1-stage embryos, GlyT was expressed in the cerebral vesicle ( Figure 3I-J) and in ventrolateral cells of the rhombospinal region up to the end of somite 6 ( Figure 3I,K-M). GlyT started also to be expressed in the developing preoral organ. In L1 larvae ( Figure 3N-X), GlyT transcripts were present in the most anterior part of the cerebral vesicle at the level of the frontal eye complex ( Figure 3N,O), which at this stage contains glutamatergic and glycinergic neurons [62,63], and in pairs of ventrolateral cells of the rhombospinal region of the nerve cord ( Figure 3Q-U). Outside the CNS, GlyT was expressed in the preoral organ ( Figure 3N,P), in the ectoderm of the oral region ( Figure 3Q,R), in the primordium of the first gill slit ( Figure 3N,S), and in some ectodermal sensory cells of the trunk ( Figure 3S,T,W). Finally, expression was also detected around the neurenteric canal ( Figure 3N,X).
GlyT2.1 transcripts were first detected in the G5 gastrula around the blastopore and in the presumptive notochord and endoderm ( Figure 4A Figure 4N). which at this stage contains glutamatergic and glycinergic neurons [62,63], and in pairs of ventrolateral cells of the rhombospinal region of the nerve cord ( Figure 3Q-U). Outside the CNS, GlyT was expressed in the preoral organ ( Figure 3N,P), in the ectoderm of the oral region ( Figure 3Q,R), in the primordium of the first gill slit ( Figure 3N,S), and in some ectodermal sensory cells of the trunk ( Figure 3S,T,W). Finally, expression was also detected around the neurenteric canal ( Figure 3N,X).  GlyT2.2 expression was first detectable at the neurula stage (N1), where it was restricted to a small region of the anterior neural plate ( Figure 5A-B), corresponding to the prospective cerebral vesicle. In T1-stage embryos, GlyT2.2 expression was maintained in the anterior cerebral vesicle ( Figure 5C-D) and extended to the preoral organ ( Figure 5C,E) and pharyngeal endoderm ( Figure 5C,F). In L1-stage larvae, expression in the anterior part of the animal was largely consistent with that observed at the T1 stage, and a new domain of GlyT2.2 expression appeared around the neurenteric canal ( Figure 5G-K).

Glycine Transporter Expression and the Neurochemistry of the Developing Amphioxus Nervous System
To assign an identity to the cells expressing GlyT, GlyT2.1, and GlyT2.2 in the CNS of amphioxus embryos, we compared the expression of the three genes to that of known markers of neural subpopulations, namely glutamic acid decarboxylase (GAD), the vesicular GABA/glycine transporter (VGAT), and the pan-neuronal marker hu-elav [35,64]. At the T1 stage, the nerve cord contained three pairs of GABAergic neurons (defined by the co-expression of GAD and VGAT [35]), with each pair being in register with one of the three anteriormost somite pairs ( Figure 6A). In the proximity of these pairs of GABAergic neurons, we found cells expressing VGAT, but not GAD ( Figure 6B). These cells have previously been interpreted as glycinergic neurons [35]. Of the three amphioxus glycine transporter genes, GlyT ( Figure 3) and GlyT2.1 (Figure 4) were broadly expressed in the nerve cord, whereas GlyT2.2 expression was limited to the anterior cerebral vesicle ( Figure 5). Cells expressing GlyT formed two rows running along the anterior-posterior axis of the CNS located adjacent to the floor plate ( Figure 6C) in a ventrolateral position ( Figure 3K-M). GlyT expression was co-localized with that of hu-elav ( Figure 6G,G'), indicating that GlyT is expressed in neurons. Conversely, we found GlyT2.1 to be expressed in the floor plate as well as in cells lateral to those expressing GlyT ( Figures 4K-M and 6D). These GlyT2.1-expressing cells are glial cells ( Figure 7A-E) with marked similarities to vertebrate astroglia [37]. Furthermore, some cells expressing GlyT2.1 were in the proximity of cells expressing VGAT ( Figure 6E,E',F). However, the two signals did not overlap ( Figure 6E

Glycine Transporter Expression and the Neurochemistry of the Developing Amphioxus Nervous System
To assign an identity to the cells expressing GlyT, GlyT2.1, and GlyT2.2 in the CNS of amphioxus embryos, we compared the expression of the three genes to that of known markers of neural subpopulations, namely glutamic acid decarboxylase (GAD), the vesicular GABA/glycine transporter (VGAT), and the pan-neuronal marker hu-elav [35,64]. At the T1 stage, the nerve cord contained three pairs of GABAergic neurons (defined by the coexpression of GAD and VGAT [35]), with each pair being in register with one of the three anteriormost somite pairs ( Figure 6A). In the proximity of these pairs of GABAergic neurons, we found cells expressing VGAT, but not GAD ( Figure 6B). These cells have previously been interpreted as glycinergic neurons [35]. Of the three amphioxus glycine transporter genes, GlyT ( Figure 3) and GlyT2.1 (Figure 4) were broadly expressed in the nerve cord, whereas GlyT2.2 expression was limited to the anterior cerebral vesicle ( Figure  5). Cells expressing GlyT formed two rows running along the anterior-posterior axis of the CNS located adjacent to the floor plate ( Figure 6C) in a ventrolateral position ( Figure 3K- GlyT is expressed in neurons. Conversely, we found GlyT2.1 to be expressed in the floor plate as well as in cells lateral to those expressing GlyT (Figures 4K-M, 6D). These GlyT2.1expressing cells are glial cells ( Figure 7A-E) with marked similarities to vertebrate astroglia [37]. Furthermore, some cells expressing GlyT2.1 were in the proximity of cells expressing VGAT ( Figure 6E,E',F). However, the two signals did not overlap ( Figure  6E,E',F), suggesting that cells of the developing amphioxus CNS do not co-express GlyT2.1 and VGAT.

Evolution of Glycine Neurotransmission in Metazoan Animals
Glycine is an important inhibitory neurotransmitter in the nervous system of vertebrates, where it has been investigated in various species [35,[65][66][67][68][69]. Conversely, our understanding of glycinergic neurotransmission in invertebrates is much more limited (Figure 1). Among invertebrates, evidence for glycinergic neurotransmission was first documented in the gastropod mollusk A. californica, and data from this model still provide the only information on glycinergic signaling in spiralian animals, showing that glycine is required for the contraction of the ventral aorta. Subsets of neurons innervating the aorta of A. californica possess a sodium-dependent glycine uptake system, an axonal glycine transport system, and glycine-containing vesicles in their nerve terminals [24,25]. Interestingly, glycinergic neurotransmission is thought also to modulate blood pressure in vertebrates, but with the inverse role of decreasing contractions and, likely, a different mechanism of action [70].
More recently, glycine was identified as an inhibitory neurotransmitter in an ecdysozoan animal, the fruit fly D. melanogaster, where it regulates the duration of a complete circadian cycle [26]. Intriguingly, several lines of evidence strongly suggest that glycine is also involved in the regulation of biological cycles in vertebrates. In zebrafish (Danio rerio), glycine and GlyT2 are present in a subset of photoreceptor cells of the pineal organ (called pinealocytes) [71], a neuroendocrine structure that regulates circadian and seasonal behaviors. In zebrafish, the pineal organ projects to mesencephalic and diencephalic structures [72], and some of the projecting fibers are glycinergic [73]. GlyT2 and VGAT are also co-expressed in rat pinealocytes, which release glycine in response to depolarization in vitro [74,75]. However, glycine-immunoreactive cells were not reported in the pineal organs of either lampreys, catsharks or sturgeons [65,69,76,77]. Glycine thus appears to be used as a neurotransmitter and, possibly, a paracrine signal in the pineal organ of at least some vertebrate species [71]. Amphioxus larvae perform diurnal migration in the water column, a typical circadian behavior, and possess a putative pineal homolog in the dorsal part of the dien-mesencephalic region: the lamellar body [78][79][80]. The lamellar body consists of two parallel rows of lamellate cells whose cilia form elaborated stacks of membranous lamellae [78]. The number of lamellate cells increases during larval development, but then the organ disaggregates, with only scattered

Evolution of Glycine Neurotransmission in Metazoan Animals
Glycine is an important inhibitory neurotransmitter in the nervous system of vertebrates, where it has been investigated in various species [35,[65][66][67][68][69]. Conversely, our understanding of glycinergic neurotransmission in invertebrates is much more limited (Figure 1). Among invertebrates, evidence for glycinergic neurotransmission was first documented in the gastropod mollusk A. californica, and data from this model still provide the only information on glycinergic signaling in spiralian animals, showing that glycine is required for the contraction of the ventral aorta. Subsets of neurons innervating the aorta of A. californica possess a sodium-dependent glycine uptake system, an axonal glycine transport system, and glycine-containing vesicles in their nerve terminals [24,25]. Interestingly, glycinergic neurotransmission is thought also to modulate blood pressure in vertebrates, but with the inverse role of decreasing contractions and, likely, a different mechanism of action [70].
More recently, glycine was identified as an inhibitory neurotransmitter in an ecdysozoan animal, the fruit fly D. melanogaster, where it regulates the duration of a complete circadian cycle [26]. Intriguingly, several lines of evidence strongly suggest that glycine is also involved in the regulation of biological cycles in vertebrates. In zebrafish (Danio rerio), glycine and GlyT2 are present in a subset of photoreceptor cells of the pineal organ (called pinealocytes) [71], a neuroendocrine structure that regulates circadian and seasonal behaviors. In zebrafish, the pineal organ projects to mesencephalic and diencephalic structures [72], and some of the projecting fibers are glycinergic [73]. GlyT2 and VGAT are also co-expressed in rat pinealocytes, which release glycine in response to depolarization in vitro [74,75]. However, glycine-immunoreactive cells were not reported in the pineal organs of either lampreys, catsharks or sturgeons [65,69,76,77]. Glycine thus appears to be used as a neurotransmitter and, possibly, a paracrine signal in the pineal organ of at least some vertebrate species [71]. Amphioxus larvae perform diurnal migration in the water column, a typical circadian behavior, and possess a putative pineal homolog in the dorsal part of the dien-mesencephalic region: the lamellar body [78][79][80]. The lamellar body consists of two parallel rows of lamellate cells whose cilia form elaborated stacks of membranous lamellae [78]. The number of lamellate cells increases during larval development, but then the organ disaggregates, with only scattered lamellate cells being found in the adult [81]. In N4 neurulae, GlyT2.1 was faintly expressed in the dorsal part of the prospective posterior cerebral vesicle, where the lamellar body will develop. Interestingly, this stage corresponds to the initiation of lamellar body differentiation [80,82]. During subsequent development (at T0 and T1 stages), GlyT2.1 expression remained conspicuous in this region of the cerebral vesicle and, in L1 larvae, in which the first lamellate cells have already differentiated [83], GlyT2.1 expression was still detectable in the lamellar body, extending also to the dorsal part of the anterior cerebral vesicle. These observations hint at a conserved ancestral function of glycine neurotransmission in the pineal organ of chordates.
Genomic analyses have suggested that glycine does not function as a neurotransmitter in the roundworm C. elegans, a nematode and thus another ecdysozoan. The C. elegans genome contains a vesicular transporter for GABA and glycine (VGAT/unc-47), but no obvious orthologs of glycine receptors or transporters, although it does contain a GABA transporter ortholog (snf-11) [84]. However, VGAT/unc-47 is expressed not only in GABAergic neurons, but also in GABA-negative neurons, some of which have an unknown neurochemical profile [85]. It thus remains possible that nematodes use glycinergic neurotransmission and that one of the orphan ligand-gated channels in their genome is a glycine receptor [86]. However, if C. elegans indeed lacks glycinergic neurotransmission, it was likely lost secondarily, as the wide deployment of glycine as a neurotransmitter in both protostomes and deuterostomes suggests that glycinergic neurotransmission was already present in the last common ancestor of all bilaterian animals. This notion is supported by biochemical evidence for the presence of a functional glycine receptor in the freshwater polyp H. vulgaris [29], a cnidarian, and thus a member of one of the earliest-branching metazoan lineages (Figure 1).

Evolution of Glycine Transporters in Deuterostomes: Competing Scenarios
Two glycine transporter genes have previously been described in amphioxus [31], corresponding to our GlyT and GlyT2.1. We identified an additional gene, which we named GlyT2.2, and reconstructed the evolutionary history of glycine transporters using three different tree-building methods (BI, ML, and NJ). Our phylogenies strongly support the split between the GlyT1 and GlyT2 clades. Whereas the GlyT1 clade only contains vertebrate sequences, the GlyT2 clade is composed of sequences from vertebrates and invertebrate deuterostomes. Other invertebrate deuterostome GlyT sequences branch outside the GlyT1 plus GlyT2 clade, indicating that no true GlyT1 orthologs exist in invertebrates, as has previously been suggested [31].
Within the GlyT2 clade, the tunicate plus vertebrate branch is very strongly supported, leaving little doubt that the ascidian tunicate C. intestinalis possesses a true GlyT2 ortholog. Conversely, the evolutionary relationships of the other invertebrate GlyT2-like sequences are more difficult to resolve. One possible scenario ( Figure 8A) is that an ancestral glycine transporter gene was duplicated early in the deuterostome lineage, before the split of echinoderms and chordates, giving rise to one GlyT and one GlyT2-like sequence. The GlyT2-like gene was then duplicated in the amphioxus lineage, resulting in the GlyT2.1 and GlyT2.2 genes, which are located on the same scaffold in the genome, only about 100 kb apart (Table 1). Interestingly, the genome of the sea urchin S. purpuratus also contains two GlyT2-like genes (at a distance of 16 Mb from each other), which thus also originated by lineage-specific duplication. Another possible scenario for the evolution invertebrate GlyT2-like sequences ( Figure 8B) is that ancestral deuterostomes already possessed two GlyT2 paralogs, one of which was lost in the lineage leading to extant tunicates and vertebrates. However, this scenario is not supported by our phylogenies, as it predicts that sea urchin and amphioxus GlyT2.1 and GlyT2.2 stably associate in the tree. Instead, we found that only the two sea urchin GlyT2 sequences formed a highly supported clade.
Phylogenetic analyses would thus be unable to correctly assess the orthology of GlyT1 and GlyT sequences. Interestingly, in the ascidian tunicate C. intestinalis, GlyT and GlyT2 are both located on chromosome 7, at a distance of about 640 kb from each other [31]. However, while an analysis of genetic synteny of vertebrate glycine transporter loci revealed a general conservation of the genes linked to GlyT1 and GlyT2, analyses of genetic synteny between invertebrate and vertebrate glycine transporter loci remained largely inconclusive [31].

Amphioxus Glycine Transporter Paralogs are Differentially Expressed in Glia and Neurons
Analysis of the expression patterns of the three amphioxus glycine transporter genes during development revealed that they were already expressed by the mid-neurula stage, which is earlier than what has been reported in vertebrates [13,88,89]. This finding indicates that glycine transporters perform functions other than glycine neurotransmission in amphioxus, a notion supported by the fact that all three amphioxus paralogs are expressed in the pharynx and GlyT2.1 additionally in the somites.
When comparing the expression domains of the amphioxus glycine transporter genes, that of GlyT2.2 appears to be the most divergent. Excepting the signal in the anterior cerebral vesicle, GlyT2.2 is mainly expressed outside of the nervous system. Conversely, both GlyT and GlyT2.1 are widely expressed in the central nervous system, albeit in different cell types: GlyT is expressed by ventrolateral neurons, whereas GlyT2.1 transcripts are mostly localized in floor plate and mediolateral glial cells (Figure 7). This situation is inversed in vertebrates, where GlyT1 is widely expressed by glial cells and glutamatergic neurons [13,90] and GlyT2 expression is characteristic of glycinergic neurons [20,91]. This observation supports the notion that the vertebrate GlyT1 genes derived from the duplication of an ancestral GlyT2 gene very early during vertebrate diversification ( Figure 8A,B). In this scenario, an ancestral chordate GlyT2 gene was already preferentially expressed in glial cells and, after its duplication in the vertebrate lineage, this same specificity was maintained exclusively in the GlyT1 paralog. To reconcile this observation with the alternative scenario that the vertebrate GlyT1s are orthologous to the GlyTs of invertebrate deuterostomes ( Figure 8C), one would have to Regarding the origin of vertebrate GlyT1, one scenario suggested by our phylogenies is that the GlyT1s originated by duplication from an ancestral GlyT2 gene, very early during vertebrate diversification, but after the split from the tunicate lineage [87]. In this scenario, the last common ancestor of tunicates and vertebrates had a GlyT gene that was subsequently lost in the lineage leading to extant vertebrates ( Figure 8A). An alternative scenario is that the vertebrate GlyT1s might be orthologous to the GlyTs of invertebrate deuterostomes, as previously suggested [31]. However, this scenario ( Figure 8C) can only be reconciled with our phylogenies, if the vertebrate GlyT1s have experienced a substantial functional diversification with concomitant changes in their coding sequences. Phylogenetic analyses would thus be unable to correctly assess the orthology of GlyT1 and GlyT sequences. Interestingly, in the ascidian tunicate C. intestinalis, GlyT and GlyT2 are both located on chromosome 7, at a distance of about 640 kb from each other [31]. However, while an analysis of genetic synteny of vertebrate glycine transporter loci revealed a general conservation of the genes linked to GlyT1 and GlyT2, analyses of genetic synteny between invertebrate and vertebrate glycine transporter loci remained largely inconclusive [31].

Amphioxus Glycine Transporter Paralogs Are Differentially Expressed in Glia and Neurons
Analysis of the expression patterns of the three amphioxus glycine transporter genes during development revealed that they were already expressed by the mid-neurula stage, which is earlier than what has been reported in vertebrates [13,88,89]. This finding indicates that glycine transporters perform functions other than glycine neurotransmission in amphioxus, a notion supported by the fact that all three amphioxus paralogs are expressed in the pharynx and GlyT2.1 additionally in the somites.
When comparing the expression domains of the amphioxus glycine transporter genes, that of GlyT2.2 appears to be the most divergent. Excepting the signal in the anterior cerebral vesicle, GlyT2.2 is mainly expressed outside of the nervous system. Conversely, both GlyT and GlyT2.1 are widely expressed in the central nervous system, albeit in different cell types: GlyT is expressed by ventrolateral neurons, whereas GlyT2.1 transcripts are mostly localized in floor plate and mediolateral glial cells (Figure 7). This situation is inversed in vertebrates, where GlyT1 is widely expressed by glial cells and glutamatergic neurons [13,90] and GlyT2 expression is characteristic of glycinergic neurons [20,91]. This observation supports the notion that the vertebrate GlyT1 genes derived from the duplication of an ancestral GlyT2 gene very early during vertebrate diversification ( Figure 8A,B). In this scenario, an ancestral chordate GlyT2 gene was already preferentially expressed in glial cells and, after its duplication in the vertebrate lineage, this same specificity was maintained exclusively in the GlyT1 paralog. To reconcile this observation with the alternative scenario that the vertebrate GlyT1s are orthologous to the GlyTs of invertebrate deuterostomes ( Figure 8C), one would have to assume that the glycine transporters of the ancestral chordate were not differentially expressed in neurons and glia and that this specialization evolved independently in both the cephalochordate and vertebrate lineages, resulting in different paralogs being expressed in homologous cell types.

Role of Glycine and Glycine Transporters in Larval Swimming
In vertebrates, glycinergic neurotransmission in the brainstem and spinal cord has important roles in escape reflexes and in the modulation of rhythmic motor behaviors, such as locomotion and breathing. Consequently, mutation of genes encoding glycine transporter or receptor subunits results in excessive startle reflexes (hyperekplexia) in humans and swimming-related phenotypes in zebrafish [4,92,93]. Rhythmic motor patterns for locomotion are generated by spinal networks called central pattern generators (CPGs). Spinal CPGs are remarkably conserved in vertebrates, regardless of the type of locomotion (swimming, walking or flying), and have further already been described in larvae of the ascidian tunicate C. intestinalis [94,95]. CPGs typically consist of segmentally arranged motor neurons that innervate adjacent trunk muscles, glycinergic commissural interneurons, ipsilaterally-projecting inhibitory interneurons, and excitatory glutamatergic interneurons. Excitatory interneurons project to all other neuronal cell types that establish the CPG, and glycinergic inhibitory interneurons cross the midline to inhibit motor neurons and all interneuron types. In this network, contralateral inhibition by glycinergic interneurons allows the generation of alternating and finely timed motor outputs that result in repetitive movements, such as swimming [96,97]. Indeed, glycine receptor antagonists abolish the alternation between sides [27,98].
The role of glycinergic transmission in the amphioxus swimming CPG has not yet been investigated. However, serial ultrastructural data are available for the anterior part of the amphioxus larva, which have allowed the reconstruction of a putative locomotory control center in the dien-mesencephalic and rhombospinal regions [36,[99][100][101]. Startle responses result from the action of ventral compartment (VC) motor neurons [102] that innervate deep muscle fibers responsible for fast swimming. Excitatory signals to VC motor neurons come from ectodermal sensory neurons, anterior bipolar neurons, glutamatergic large paired neurons (LPNs), and cholinergic ipsilateral projecting interneurons (IPNs). Inhibitory inputs come from GABAergic commissural neurons (CNs) and a pair of multipolar neurons (MPs) located at the level of somite 2 [36]. MPs have been mapped to the first pair of VGAT+/GAD− cells in the rhombospinal region and have hence been considered as putative glycinergic neurons [35,36]. They have axons that branch into an ipsilateral ascending fiber and a descending fiber that further splits into an ipsilateral and a contralateral branch [100]. All fibers emanating from the right MP form numerous synapses, targeting IPNs and CNs in addition to VC motor neurons. Inputs to MPs come from LPNs and small, likely sensory, fibers. Amphioxus larvae thus possess two types of interneurons that can potentially inhibit motor neurons on the contralateral side: GABAergic CNs and glycinergic MPs. Glycinergic contralateral inhibition may thus be part of the swimming CPG in amphioxus larvae, similar to what has been described in ascidian tunicate larvae and vertebrates. Future work aimed at characterizing the swimming CPGs of larval amphioxus will be important not only for the insights it provides into CPG origin, but for what it tells us about the structure and function of the far more complex vertebrate CPGs.
The anatomical organization of the MP axons, highly branched with multiple serial synapses, might be the key for explaining the disequilibrium between the abundance of cells expressing glycine transporters and the paucity of putative glycinergic neurons in the amphioxus nerve cord. If our interpretation is correct, most ventral neurons of the amphioxus nerve cord receive glycinergic innervation, and GlyT thus functions as a postsynaptic transporter, comparable to the role of vertebrate GlyT1 in glutamatergic neurons. Most floor plate and astroglia-like cells, both of which are intimately associated with the neuropile [37], express GlyT2.1. In this cellular context, GlyT2.1 is hence likely involved in the modulation of glycinergic signaling, which, again, is similar to the functions of vertebrate GlyT1. Several ectodermal sensory neurons project to the nerve cord [103]. Many of these neurons are GABAergic or glutamatergic [35,54], and the expression pattern of GlyT indicates that some of them are glycinergic. Afferent peripheral neurons might thus contribute significantly to increasing the number of glycinergic synapses in the nerve cord. Notably, GlyT2.1 is also expressed in the somites until the T1 stage, when the embryo is already capable of coordinated swimming in response to external stimuli. It might thus be that the muscle fibers of amphioxus are a direct target of glycinergic neurons during the early development of the locomotory system, as has been hypothesized for the ascidian tunicate tadpole based on glycine receptor expression [27].
Amphioxus late embryos and larvae possess non-neuronal cells that, as with vertebrate astroglia, express EAAT2, a glutamate transporter responsible for modulating glutamatergic transmission [37]. We show here that virtually all the astroglia-like cells of amphioxus also express GlyT2.1, strengthening their proposed homology to vertebrate astroglia. Moreover, the co-expression of EAAT2 and GlyT2.1 in the same cell populations suggests that glycine and glutamate might co-act at the same synapse. This is known to be the case in vertebrates, where glycine is a co-agonist of NMDA glutamate receptors and GlyT1 is found at glutamatergic synapses, where it modulates the function of NMDAR regulating glycine levels in the receptor microenvironment [6,90].

Conclusions
Glycinergic neurotransmission performs well-established functions in vertebrates, but its role in invertebrates has been badly neglected. Despite the limited number of studies, glycinergic neurotransmission in invertebrates has been implicated in diverse functions, including the coordination of locomotion, blood pressure regulation, and circadian rhythmicity. Some of these functions may be conserved in vertebrates, implicating glycine as part of an ancestral neurochemical toolkit, present in early metazoans and employed widely among taxa for purposes of intercellular communication.
Perhaps significantly, at least two glycine transporter paralogs are present in the genome of all the animal species we investigated. Additionally, though the phylogenetic relationships of glycine transporter family members are complex, there is clear evidence for lineage-specific duplications of glycine transporter genes in several animal phyla. This strongly suggests that no less than two transporters, likely with different kinetic and/or stoichiometric properties, are necessary for effective glycinergic transmission. As in vertebrates, invertebrate glycine transporter paralogs likely exhibit a differential, cell type-specific expression, which is what we observed in amphioxus, where GlyT and GlyT2.1 are widely expressed in the CNS, but differentially in neurons and glia. Our data also suggest that different glycine transporter paralogs are expressed in homologous cell types in amphioxus and vertebrates, which means further studies on echinoderms and tunicates will be needed to fully explain the evolution of glycinergic neurotransmission from invertebrates to vertebrates.