Coe Genes Are Expressed in Differentiating Neurons in the Central Nervous System of Protostomes

Genes of the coe (collier/olfactory/early B-cell factor) family encode Helix-Loop-Helix transcription factors that are widely conserved in metazoans and involved in many developmental processes, neurogenesis in particular. Whereas their functions during vertebrate neural tube formation have been well documented, very little is known about their expression and role during central nervous system (CNS) development in protostomes. Here we characterized the CNS expression of coe genes in the insect Drosophila melanogaster and the polychaete annelid Platynereis dumerilii, which belong to different subgroups of protostomes and show strikingly different modes of development. In the Drosophila ventral nerve cord, we found that the Collier-expressing cells form a subpopulation of interneurons with diverse molecular identities and neurotransmitter phenotypes. We also demonstrate that collier is required for the proper differentiation of some interneurons belonging to the Eve-Lateral cluster. In Platynereis dumerilii, we cloned a single coe gene, Pdu-coe, and found that it is exclusively expressed in post mitotic neural cells. Using an original technique of in silico 3D registration, we show that Pdu-coe is co-expressed with many different neuronal markers and therefore that, like in Drosophila, its expression defines a heterogeneous population of neurons with diverse molecular identities. Our detailed characterization and comparison of coe gene expression in the CNS of two distantly-related protostomes suggest conserved roles of coe genes in neuronal differentiation in this clade. As similar roles have also been observed in vertebrates, this function was probably already established in the last common ancestor of all bilaterians.


Introduction
The generation of neurons and glial cells is a complex and multi-step process, on which rely both the architecture and the activity of the bilaterian central nervous system (CNS). From the undifferentiated progenitors of the neural epithelium to the fully functional CNS cells, a large set of transcription factors and their combinatorial expression are required to control the different steps of commitment, patterning and differentiation. Among these transcription factors, collier/olfactory/early B-cell factor (COE) proteins form a specific family of helix-loop-helix (HLH) proteins, characterized by the presence of three highly conserved domains: a unique DNA binding domain (DBD); an immunoglobulin / plexin / transcription (IPT) domain putatively involved in both DNA/protein and protein/protein interactions; and an atypical HLH dimerization motif [1,2,3]. coe genes have been identified in the three major bilaterian lineages (deuterostomes and the two groups of protostomes, Ecdysozoa and Lophotrochozoa), as well as in the cnidarian Nematostella vectensis, the ctenophora Mnemiopsis leidyi, and the sponge Amphimedon queenslandica [3,4,5,6], but not in plants or fungi, indicating that these genes form a metazoan specific family which would have arisen during early animal evolution. Members of the coe family are present as single genes in most studied species except for vertebrates which possess up to 4 homologs [3,5], probably due to the two rounds of genome duplication which occurred during early vertebrate evolution [7].
Ebf1 and Olf-1, the two first identified members of the family, were originally described as a mouse transcriptional regulator of early B-cell differentiation [8], and a rat nuclear protein regulating the expression of the olfactory marker protein OMP [9], respectively. Several orthologs were subsequently identified in various vertebrate species, such as mouse, chicken, Xenopus, and zebrafish, and have been named either Ebf or coe [10,11,12,13,14]. A common theme for all the coe genes is a predominant expression in the developing nervous system, mostly in differentiating postmitotic neural cells -an expression is nevertheless also found in non neural tissues such as the adult mouse spleen and the developing limb buds. In the mouse, Ebf1, Ebf2 and Ebf3 expression is detected in both peripheral nervous system (PNS) cells (olfactory sensory neurons, facial brachiomotor neurons and their respective precursors) and differentiating post-mitotic neurons of the CNS [15]. This latter profile is shared by all their orthologs in other vertebrates, except the zebrafish Zcoe2 gene which is also expressed in progenitors of the primary neurons of the spinal cord, prior to the proneural gene neurogenin [11]. Extensive studies of vertebrate coe genes function have indicated their requirement at several crucial steps of neurogenesis, such as the coupling of cell cycle exit and migration from the subventricular proliferative zone toward the mantle layer of the neural tube, and proper neuronal differentiation [14,16]. Ebf1 is also required for axonal guidance of thalamic neurons projections, possibly by establishing the molecular patterning of the basal ganglia area [16,17]. A coe/ebf gene isolated in a close-relative of vertebrates, the amphioxus Branchiostoma floridae, is expressed in many cells of the developing CNS, as well as in lateral isolated cells which may correspond to epidermal sensory neurons [18]. An expression associated to the developing nervous system is also found in the echinoderm Strongylocentrotus purpuratus [6], thus suggesting conserved functions for coe genes among deuterostomes.
In Drosophila, a single coe gene, collier (col), has been identified and shown to be required for the formation of the head/trunk intercalary segment [19,20]. It has later been extensively studied for its involvement in several other developmental processes, such as the control of embryonic muscle identities [21,22], wing patterning [23,24], and differentiation of immune cells in response to parasitization [25,26]. In addition, col is also expressed in the developing nervous system, both in cells of the ventral nerve cord (VNC) and in a few PNS neurons, a subset of the so-called multidendritic neurons, where it has been shown to control the elaboration of the characteristic dendritic arborization [27,28,29]. In the CNS, it specifies the identity of the peptidergic dorsal Apterous expressing (dAp) neurons, by triggering and participating to an original ''feedforward cascade'' of transcription factors at an early postmitotic stage [30]. Despite this latter study, the identity of most of the cells expressing col in the VNC and its functions during CNS formation remain largely unknown. A coe gene, known as unc-3, has also been identified in another protostome, the nematode C. elegans, where it is involved in the axonal outgrowth of motor neurons [31] and in the specification of the identity of some chemosensory neurons [32]. Recently, an expression of coe genes in the developing nervous system has also been found in three other protostomes, a mollusc and two annelids [6].
In this article, we characterized the expression pattern of coe genes in the CNS of Drosophila and the annelid worm Platynereis dumerilii, species which belong to two different branches of the protostomes, Ecdysozoa and Lophotrochozoa respectively. In addition Platynereis displays vertebrate-like mechanisms ensuring CNS formation and patterning with key differences when compared to those found in Drosophila [33,34,35]. We first show that Col is expressed in about 10 to 15% of all Drosophila embryonic CNS neurons, corresponding exclusively to interneurons, as seen by coexpression with markers of a wide variety of interneuron types, while it is excluded from all known motor neurons. We also demonstrate that col is required for the proper differentiation of one population of interneurons, the even-skipped (eve)-lateral (EL) neurons. We then report the isolation of a single coe gene in Platynereis (referred to as Pdu-coe) and show that it is exclusively expressed in the nervous system (VNC, brain and PNS). In the VNC, Pdu-coe is only expressed in post-mitotic cells from which arise various neuronal subtypes. Taken together, our data confirm the remarkable conservation of COE protein structure among metazoans, as well as their predominant expression in the nervous system, almost exclusively in post-mitotic cells. However, striking differences are observed in the other expression domains of coe genes in the different bilaterian lineages [6], suggesting a plasticity of coe functions during evolution. We discuss a putative ancestral role for the coe gene family in neuronal differentiation in bilaterians, and hypothesize that a combinatorial mode of action involving interactions with several co-factors may explain the surprising discrepancy between high conservation of COE protein sequence and the plasticity of their exerted functions.

Results
Spatial and temporal analysis of Drosophila Col expression in the trunk CNS Previous studies showed an expression of the Drosophila col gene in the developing CNS [19], however without a detailed identification of the cells that express col, except for a few thoracic interneurons [30]. Here we used antibodies raised against Col [21,30] and a col-gal4 line [26] to further characterize its expression pattern in the trunk CNS. Col expression in the VNC was first detected at early stage 12 in about 20 cells per hemisegment and increased to approximately 50 cells per hemisegment at stage 14 and later ( Figure 1A-A0 to C-C0 and Figure S1 B, B9). The position and size of these cells, as well as the timing of Col expression, suggest that they are postmitotic neural cells. To confirm this hypothesis, we performed double-labelings with antibodies against Col and the phosphorylated form of histone H3 (PH3) which reveals cells undergoing mitosis. In the ventral layer of the VNC of stage 12 embryos, we typically found two to three Col + PH3 + cells per hemisegment probably ganglion mother cells (GMCs), but none at later stages ( Figure 1A-A0, B-B0), indicating that Col is mainly (at stage 12) or only (at stage 14 and later) expressed in postmitotic neural cells.
To find out whether Col is expressed in neurons and/or glial cells, we performed double labelling with the neuronal marker Elav [36] and the glial marker Repo [37]. At stage 15, we found co-expression with Elav ( Figure 1D-D0), but no overlap with Repo ( Figure 1E-E0), indicating that Col is expressed in neurons but not in glial cells. Accordingly, in col-gal4 / UAS-CD8-GFP embryos, GFP decorates many neurites, confirming the neuronal expression of col ( Figure S1). In addition, many VNC neurons retained col expression into larval stages (L2, see Figure S1; L1 and L3, not shown). We therefore conclude that col is mainly expressed in postmitotic differentiating and differentiated neurons.

Drosophila Col is expressed in various populations of interneurons, but not in motor neurons
We next proceeded to the identification of the neurons expressing Col at stage 15 using a large set of markers widely covering the neuronal diversity of the Drosophila VNC. To find out whether Col is expressed in motor neurons, we first performed a double labelling using antibodies against Col and the phosphorylated form of Mad (pMad), a general motor neuron marker [38,39], and found no-colocalization (Figure 2A-A0, B-B0). We confirmed this observation using several other motor neuron markers. Zfh1 is expressed in all somatic motor neurons as well as in one identified interneuron per hemisegment, the dAp neuron [39]. We found co-localization between Col and Zfh1 expression only in one dorsal neuron, much probably the dAp neuron (where Col expression has been previously characterized; [30]), but not in the Zfh1 expressing motor neurons (Figure 2 C-C0; Figure S2). Accordingly, Col is not expressed in the islet1 + and HB9 + RP1-3-4-5 motor neurons that innervate ventral muscles through the intersegmental nerve b (ISNb) [40,41], nor in the Eve + aCC, RP2 and U/CQ motor neurons innervating the dorsal muscles through the intersegmental nerve dorsal most (ISNdm) [42,43,44], nor in the BarH1 + motor neurons innervating lateral muscles through the SNa nerve [45] (Figure 2 D-D0 to G-G0; Figure S2). Taken together, these observations show the absence of Col expression in motor neurons.
Col is therefore expressed in some interneuron populations that we characterized further using several molecular markers. In the very dorsal part of the VNC, Col is reproducibly expressed in a pair of neurons in each hemisegment, one of which we suspected to be the Zfh1 + dAp neuron (Figure 2 C-C0). We confirmed this assumption by showing the co-expression of Col and apterous in this particular neuron ( Figure 3A9-A0). The second neuron of the dorsal pair co-expresses Col and islet ( Figure 2D-D0) as well as dachshund (dac) ( Figure 3B-B0) which is exclusively expressed in interneurons in the VNC [46] -we therefore named this neuron the dCID neuron (dorsal Collier/Islet/Dac neuron). In the ventral part of the VNC, Col expression is detected in several interneurons, two to three of which also express dac ( Figure 3C-C0). Four to five of the other Col + neurons belong to the EL (Evelateral) cluster, which contains 9-10 neurons expressing Even-Skipped (Eve) [47] (Figure 3D-D0). EL neurons also express eagle [48] and we found a co-expression of Col and eagle in 4-5 EL neurons ( Figure 3E-F0). We confirmed by a triple-labelling that some EL neurons express Eve, Col, and eagle. Finally, we found Col expressed in each hemisegment in 2 Engrailed + interneurons [49] (Figure 3G-G0) and in 1-2 Hb9 + interneurons ( Figure S2). By contrast, no co-expression was found between Col and dbx, whose expression marks a large set of GABAergic interneurons [50] ( Figure S3). We then looked for a possible correlation between Col expression and neurotransmitter phenotypes in the VNC. To define whether Col is expressed in glutamatergic, cholinergic, or GABAergic neurons, we performed double-labellings with Col and vGlut Gal4 (vesicular Glutamate transporter), ChAT (Choline Acetyltransferase), and GAD1 (glutamic acid decarboxylase 1), respectively. In late stage 16 embryos Col and Chat were found to be co-expressed in only one neuron in each hemisegment ( Figure 4A-A0) and we did not detect any co-expression of Col with GluT and Gad1 ( Figure 4B-C0). In first instar larvae, a stage where neurons are fully differentiated (at least at the electrophysiological level), we found that 8 to 10 Col expressing neurons per hemisegment also express Chat ( Figure 5A  Drosophila VNC also contains peptidergic neurons, most of which expressing the dimmed gene [53]. In late stage 16 co-expression of dimmed and Col is observed in 6-7 cells ( Figure 4E-E0), indicating that only a small part of the Col-expressing neurons are peptidergic, and that most peptidergic neurons do not express Col at this stage. In L1 larvae only one dorsally located neuron, namely the dAp neuron, is found in each hemisegment to co-express Col and dimmed ( Figure 5D 1 -D 6 ).
During the course of this analysis double labelling between Col and dimmed showed that Col is expressed, in each hemisegment, in a neuron immediately adjacent to one of the well-characterized dimmed-expressing neurons, dMP2 ( Figure 4E9 arrowheads). These neurons are produced by very peculiar neuroblasts, named MP2, that divide only once to produce two neurons: dMP2 and vMP2. The first one extends its axon posteriorly in the CNS whereas the second one extends its axon anteriorly towards the brain and both act as pioneering neurons during embryogenesis [54]. In the most posterior segments, dMP2 neurons survive during larval stages, innervate the hindgut and produce insulin-like peptides [55]. The specification of dMP2 versus vMP2 identity involves the Notch signalling pathway and the Numb protein: during the division of MP2, Numb is partitioned into one of the daughters where it antagonizes Notch signalling, thus promoting the acquisition of the dMP2 fate -the other daughter cell which receives Notch signal becomes vMP2 [54,56,57] ( Figure S4). Judging by their position, adjacent to dMP2, we were tempted to identify some of the Col expressing cells as vMP2 neurons ( Figure 4E9 double arrowheads). To confirm this, we forced the expression of an activated form of Notch (Nintra) in the dMP2 neurons and in response found Col to be expressed in most of these neurons ( Figure S4), thus indicating that, in the wild-type, Col is expressed in vMP2 under the control of Notch signalling.
All together our data indicate that Col is not expressed in motor neurons, but rather in several subpopulations of interneurons (together constituting 10 to 15% of all CNS neurons), some of which are well characterized. This includes the dAp peptidergic interneurons, part of the eve-lateral (EL) cluster, and the vMP2 interneurons in which col expression relies on Notch signalling. Moreover, neurons expressing Col do not produce glutamate, GABA, dopamine, nor serotonin, and only a subpopulation of them are peptidergic or cholinergic. Finally, we identified a previously uncharacterized interneuron, dCID, that can be recognized by its characteristic dorsal position in the close vicinity of dAp and by the expression of a unique combination of molecular markers (Col, islet, and dac).
col is required for the proper differentiation of some Eveexpressing interneurons in the Drosophila VNC Previous studies have shown that col is required for the proper differentiation of the dAp interneurons in the VNC [30], and a subset of the so-called multidendritic neurons in the PNS [27,28,29]. Here, we questioned whether a function for col in neuronal differentiation can be extended to other neurons. We focused on the well characterized EL cluster which contains, in the wild-type, 9-10 Eve + neurons (100% of the hemisegments, n = 50; Figure 6) among which 4-5 express Col ( Figure 3D-D0). We used two different col loss-of-function mutants, col 1 (null mutation) and col 2 (strong hypomorphic allele) [20], to assess the function of col in the differentiation of the EL neurons. With both alleles, we found a significant reduction of the number of EL neurons expressing Eve ( Figure 6). In col 1 homozygous embryos, we found 4-5 Eve + neurons in 58% of the hemisegments and 2-3 Eve + neurons in 42% of the hemisegments (n = 73); in col 2 homozygous embryos, we found 4-5 Eve + neurons in 89% of the hemisegments and 2-3 Eve + neurons in 11% of the hemisegments (n = 48). We did not observe any effect on other Eve expressing neurons (which do not express Col in the wild-type), such as the aforementioned Eve + motor neurons and the Eve + pCC interneurons ( Figure 6).
The number of missing Eve + neurons in col mutants is superior to the number of cells expressing col in the wild-type in a significant fraction of embryos. In col 1 mutants, in 42% of the cases, there are 7 to 8 missing Eve + neurons while Col is expressed at most in 5 EL neurons in the wild-type. One obvious explanation for this paradoxical observation is that Col would be expressed in the neuroblasts or GMCs that produce the EL neurons and might therefore be involved in the generation of these neurons. To test this hypothesis, we performed a triple labelling with antibodies against Col, Eve, and PH3 on embryos from early stage 12 to stage 13 (when proliferating cells such as GMCs are present) to define whether some of the Col expressing cells in the EL cluster may be dividing cells. We did not find any co-localization between Col (or Eve) with PH3 indicating that Col (and Eve) are only expressed in post-mitotic neurons within this lineage ( Figure 7A to 7D-D9), thus ruling out the possibility of a function of col in neuroblasts or GMCs. Interestingly we noticed that Col is expressed in all the early born EL neurons (up to five EL neurons in stage 12; Figure 7A to 7C). One possibility would be that the EL neurons expressing Col from stage 13 onwards ( Figure 7D-D9, E-E9) would be the late born EL neurons and therefore that Col would be transiently expressed in most or all EL neurons. This view is further reinforced by the fact that we noticed faint Col expression in some Eve+ EL neurons in stage 13 embryos (these neurons are indicate by arrowheads in Figure 7D, D9 and express lower level of Col compare to the strong Eve+ EL neurons denoted by arrows; see Figure 7F for a summary of Col expression in the EL cluster).
We therefore conclude that col is required for the proper differentiation of the EL neurons. Combined with previously published studies [27,28,29,30], our data indicate an involvement of col in the differentiation of various types of neurons both in the CNS and the PNS.
An ortholog of col in the annelid Platynereis, Pdu-coe, is exclusively expressed in the developing nervous system In order to determine whether the involvement of col in neural differentiation found in Drosophila can be extended to other protostomes, we cloned an ortholog of col in the polychaete annelid Platynereis dumerilii, by PCR using degenerated primers followed by RACE PCR with gene-specific primers. Sequence analysis and multiple alignments with known coe genes from representative metazoan species revealed that the cloned Platynereis gene encodes a protein displaying the three conserved domains found in all COE proteins, including the specific DBD domain ( Figure S5). Protein sequence analyses show that the Platynereis ortholog clusters with the coe genes from other lophotrochozoans in the phylogenetic tree of the Coe family ( Figure S5). We therefore named the cloned gene Pdu-coe (Platynereis dumerilii member of the Col/Olf1/ EBF family of transcription factors). We isolated only a single Coe gene in Platynereis; however, as the genome of this annelid has not been fully sequenced yet, we cannot exclude that others may exist. Nevertheless, as members of the family have only been found as single genes in all non vertebrate species studied so far [3,4,5,6], we expect that the one we cloned corresponds to the only coe gene present in Platynereis.
We then analyzed the expression pattern of Pdu-coe during development by whole mount in situ hybridization (WMISH) at five developmental stages, ranging from 24hpf (hours post fertilization) to 72hpf, during which the VNC is formed. In 24hpf embryos, before closure of the blastopore, Pdu-coe expression is detected in two bilaterally-symmetrical areas in ventral neurectoderm which will give rise to the VNC ( Figure 8A, B). From 33hpf to 72hpf, Pdu-coe expression is found only in the deep layers of the neurectoderm. The pattern extends posteriorly and laterally in correlation with extensive neurogenesis, marked by the expression of the neuronal markers Pdu-elav and Pdu-syt [34]. At 55hpf and 72hpf, transcripts are also detected in cells located across the ventral midline ( Figure 8G, J; white arrowheads), in areas corresponding to the commissures of the VNC ( Figure 8I, L). Moreover, a few scattered Pdu-coeexpressing cells are detected in the head area ( Figure 8A, D, E, F, H, J, K; asterisks) belonging to the Platynereis brain [58]. Finally, Pdu-coe is also expressed in isolated lateral cells of the trunk ( Figure 8H, black arrowheads) whose position suggests that they belong to the PNS. In contrast to what has been observed in other animals, we did not find any expression outside the nervous system and in particular in the developing mesoderm.
Platynereis, like most polychaete annelids, presents a biphasic development, with an early embryonic and larval development giving rise to a small 72hpf worm that bears only three chaetal segments (i.e. with appendages). The subsequent post-larval development involves the posterior addition of many chaetal segments (including their neural elements), from a subterminal growth zone containing stem cells, in a sequential manner throughout most of the life of the animal. This process is known as posterior growth and is mimicked by the capacity of juvenile worms to regenerate segments after a cut [59]. We thus analyzed Pdu-coe expression during segment regeneration by WMISH and found a pattern similar to that observed during larval development, i.e. expression in a large bilaterally symmetrical expression domain in the developing VNC and in commissural neurons ( Figure 8M; white arrowheads). We also found a few Pdu-coe-expressing cells in the lateral ectoderm ( Figure 8N; arrows and 8O double arrowheads), located in both the ventral and dorsal parts of the parapodes (the worm appendages) and probably belonging to the PNS. No expression in other tissues such as the mesoderm was observed. Pdu-coe expression is therefore, both during embryonic/larval and post-larval development, strongly associated with neurogenesis as transcripts are detected exclusively in developing neural tissues.  Pdu-coe is expressed in diverse populations of neurons in the Platynereis VNC We next tried to define more precisely the identity of the cells expressing Pdu-coe in the larval VNC. It has been shown that, consistently with what is observed in the vertebrate neural tube, the ventral neurectoderm of Platynereis is subdivided into several longitudinal progenitor domains defined by the partially overlapping expression of pax and nk genes, and from which originate conserved neuronal subtypes. From medial to lateral, we can thus define four progenitor domains characterized by the expression of Pdu-nk2.2+Pdu-nk6, Pdu-nk6+Pdu-pax6, Pdu-pax6+Pdu-pax3/7, and Pdu-pax3/7 alone, respectively. Each domain will give rise to specific types of neurons: for example, the Pdu-nk2.2+Pdu-nk6expressing progenitors will give rise to serotoninergic neurons while the Pdu-nk6+Pdu-pax6-expressing progenitors will produce Pdu-hb9 + cholinergic motor neurons [33]. In addition to this medial-lateral axis, the existence of an apico-basal (superficialinternal) organization of the ventral neurectoderm has also been described [34,35]. The most superficial (apical) layers of the ventral neurectoderm contain proliferative and undifferentiated progenitors expressing proneural genes such as Pdu-achaete-scute and Pdu-neurogenin, whereas intermediate ones are composed of post-mitotic cells undergoing differentiation and expressing early differentiation markers such as Pdu-elav. Finally, the deepest layers contain fully differentiated neurons expressing late differentiation markers such as Pdu-syt.
We used several of these markers to precisely characterize the expression pattern of Pdu-coe along both the medial/lateral and the apical/basal axes of the VNC neurectoderm. To avoid the many technical problems usually encountered with double fluorescent in situ hybridizations (such as the weakness and non-reliability of the fluorescent signals), we used a technique of pattern registration Figure 5. Col expression and neurotransmitter phenotypes in larvae. L1 larvae stained with Col together with Chat Gal4 (A 1 -A 6 ), vGlut Gal4 (B 1 -B 6 ), GAD1 (C 1 -C 6 ), Dim Gal4 (D 1 -D 6 ) and TH Gal4 (E 1 -E 2 ). There is no co-labeling between Col and vGluT, nor Col and GAD1, nor Col and TH, indicating that Col is not expressed in glutamatergic, gabaergic nor dopaminergic neurons. Only the dorsally located dAp neuron is found co-expressing Col and Dim (arrows in D 1 ) and a subset of Col expressing neurons are Chat positive (8 to 10 neurons per hemisegment; arrowheads in A 1 -A 6 ). doi:10.1371/journal.pone.0021213.g005 allowing in silico alignments of gene expression patterns, by taking advantage of the important reproducibility of Platynereis development and the highly stereotypic organization of the axonal scaffold of the larvae [58,60,61]. This method consists in independently acquiring images of the expression of different genes by whole-mount reflection confocal laser-scanning microscopy [62] on distinct larvae of the same batch/stage, reconstructing the patterns in 3D, and aligning them in silico, using the axon network of the VNC (revealed by immunostaining) as a reference (see Methods for further details). As validation of our method9s accuracy, we observed a good overlap between signals obtained for the same gene in different individuals and found a good agreement between double-labelling obtained by the in silico registration method and those obtained by doublefluorescent WMISH ( Figure S6). While the expected variability of patterns and morphology between two distinct larvae prevents single cell resolution analysis; we can easily and accurately visualize complementarity and/or overlapping expression domains.
Our data first indicate that the ventral Pdu-coe expression domain is restricted to the developing VNC, as it completely overlaps with the Pdu-elav expression domain ( Figure 9A-A-), and directly abuts that of Pdu-olig ( Figure 9B-B-) which is expressed in the lateral ectoderm immediately adjacent to the VNC [35]. We then compared the expression domain of Pdu-coe with those of the nk and pax genes. It is important to note that Pdu-coe is expressed in the deep internal layers of the VNC (therefore probably in neurons) whereas the pax and nk genes are expressed mainly in the most superficial ones (therefore in progenitors), as revealed by transverse sections (Figure 10A,C,D). Overlaps of expression domains, as viewed in ventral views, will thus not reflect a real coexpression at the cellular level, but indicate from which progenitor domain(s) Pdu-coe-expressing cells are likely to be produced. We found that the expression domain of Pdu-nk2.2 directly abuts that of Pdu-coe on its most medial border; the patterns are thus complementary except for the few Pdu-coe expressing cells localized along the axonal commissures ( Figure 9C-C-). Reciprocally, the Pdu-coe domain almost entirely overlaps with that of Pdu-pax6 (itself complementary to nk2.2; [33]) and extends slightly more laterally ( Figure 9D-D-). The Pdu-nk6 and Pdu-pax3/7 expression domains partially overlap with Pdu-coe, in medial and lateral region, respectively ( Figure 9E-E-, F-F-). Pdu-coe-expressing cells thus emerge from several different progenitor domains known to produce different types of neurons.
Given its expression in the deep layers of the neurectoderm, and according to previous observations about the organization of this tissue [33,34], Pdu-coe seems to be expressed in the post-mitotic daughter cells of paxand nk-expressing progenitors, and thus in various neuronal types. To assess this hypothesis, we first addressed whether Pdu-coe-expressing cells are indeed post-mitotic cells. We detected cells in S-phase using the EdU (59-ethynyl-29deoxyuridine) proliferation assay on 55hpf larvae. This very sensitive method is much less harmful to the tissues than the most commonly used BrdU [63], thus allowing to perform both WMISH and antibody staining in addition to EdU labelling. Using this original technique, we directly compared the expression domain of Pdu-coe with the cell proliferation profile of the ventral neurectoderm and found no overlap ( Figure 10B), indicating that Pdu-coe is only expressed in post-mitotic cells. We then compared the Pdu-coe expression pattern with that of several neuronal differentiation markers [33], and found extensive overlaps with the expression domains of all these genes, namely the general neuronal marker Pdu-syt ( Figure 9G-G-), the interneuron markers Pdu-chx10 and Pdu-gata123 (henceforth simplified as Pdu-gata) ( Figure 9H-H-,I-I-), and the putative motor neuron markers Pdu-hb9, Pdu-islet and Pdu-VAChT ( Figure 9J-J-to 9L-L-). Contrary to pax and nk genes, these genes are expressed by differentiating neuronal cells located in the deep layers of the neurectoderm that co-express Pducoe ( Figure 10A,E-J).
Taken together, our data indicate that Pdu-coe is exclusively expressed in the developing nervous system. In the VNC, its expression is only found in post-mitotic cells originating from several progenitor domains, and expressing various neuronal differentiation markers. Pdu-coe is therefore expressed, as its Drosophila ortholog col, in several subpopulations of neurons and may therefore be involved in their differentiation.

Discussion
In this article, we report the fine characterization of coe gene expression in the CNS of two protostomes, the insect Drosophila melanogaster and the annelid Platynereis dumerilii. To our opinion, the use of these two species provides particularly pertinent data to envision coe gene evolution, as they belong to the two subgroups of protostomes (Ecdysozoa and Lophotrochozoa). Together with previously published vertebrate studies, our study thus allows the comparison of data from the three main bilaterian lineages. In addition Drosophila and Platynereis are both segmented animals that display different modes of development. Indeed, while in Drosophila the segments that constitute the whole CNS of the larvae are all formed in a single phase during embryogenesis, in Platynereis only three segments are first formed within a three days period. Afterwards the others segments are progressively added over a period of several months until the worm reaches its sexual maturity. Finally, the two species show differential modes of CNS patterning: whereas in Platynereis the CNS is subdivided into large domains from which emerge specific neural cell types (similarly to vertebrates) [33,34,35], Drosophila CNS patterning is mainly based on cell-to-cell specification processes in which every neuroblast (even neighbouring ones) express different combinations of developmental genes that control their identity [64].
col is involved in the differentiation of interneurons in the Drosophila VNC While col, the single Drosophila member of the coe family, has been isolated and shown to be expressed in many cells of the VNC   15 years ago [19], very little was known about the identity of the cells that express this gene and its function during CNS development, besides the fact that it is expressed in the thoracic dAp peptidergic interneurons and required for their proper differentiation [30]. Here we provide a detailed characterization of the expression pattern of Col in the VNC and show that it is mostly expressed in neurons, from early stage 12 until larval stages -at stage 14 we found Col to be expressed in about 50 neurons per hemisegment, i.e., about 10% of the total number of neurons. We observed that Col expression is excluded from glial cells and all known motor neurons, indicating that it is only expressed in interneurons. We used a large set of neuronal markers to define the identity of these neurons and, found Col to be expressed in diverse populations of interneurons, as seen by its co-expression with various neuronal-specification genes such as Eve, eagle, dachshund, Engrailed, apterous, islet, and Zfh1 (Figures 2, 3 and 11A).

C) Col expression overlaps with Eve respectively in 4 and 5 EL interneurons per hemisegment (arrows). Note that during these stages Col expression does not overlap with PH3 indicating that Col is only expressed in postmitotic cells in the intermediate region of the VNC. (D, D9) In the dorsal layer of the intermediate region (denoted Interm-dorsal) of stage 13 embryos two EL interneurons express high levels of Col (arrows) and two EL interneurons express weak levels of Col (arrowheads). In the ventral layer of the intermediate region (Interm-ventral) three EL interneurons express high levels of Col (arrows) and three EL interneurons express weak or undetectable levels of Col (arrowheads). As found for earlier stages Col expression does not overlap with PH3 in any cells. (E, E9) In stage 15 embryos most if not all EL interneurons express eagle (using eagle-gal4). Arrows indicate the EL interneurons that express
We pointed out that some of the Col expressing cells correspond to previously characterized interneurons, such as the dAp neurons [30] in which Col is co-expressed with apterous and Zfh1, the EL neurons [47,48] in which Col is co-expressed with Eve and eagle, and the vMP2 neuron [54] for which Col becomes a reliable marker. Besides, we also identified a previously uncharacterized Col expressing neuron, we named dCID, located close to the dAp neuron and characterized by the co-expression of Col, islet, and dachshund. The cell bodies of Col expressing neurons occupy diferent locations in the VNC and using col-gal4 they are seen extending their axons to various locations within the CNS confirming both that they belong to highly different types of interneurons and that they are not motor neurons. We wondered whether these neurons may nevertheless share some common features, such as their neurotransmitter phenotype. We addressed this hypothesis using a large set of molecular markers (Figures 4 and 5), and found that 8-10 Col expressing neurons per hemisegment are cholinergic, as seen by the expression of ChAT [65] which encodes the choline acetyltransferase enzyme, and 6-7 are likely to be peptidergic, as seen by their expression of dimmed, a gene controlling peptidergic differentiation [53]. Although the neurotransmitters used by the remaining Col expressing neurons are still unknown, we can conclude that there is no evidence for a correlation between Col expression in a neuron and the acquisition of a specific neurotransmitter phenotype. Our data therefore indicate that Col is expressed in diverse subpopulations of interneurons that are highly heterogeneous. Indeed Colexpressing interneurons display differential properties such as: i) positions of their cell bodies in the VNC, ii) axonal projection patterns within the CNS, iii) molecular identities and iv) neurotransmitter phenotypes ( Figure 11A). Whether these neurons may have some shared properties, beyond their expression of Col, remains to be defined. This raises the question of the function of Col in the developing VNC. We focused on the EL interneurons which can be easily visualized by their expression of Eve [47]. Using col loss-offunction mutants, we found that col is required for the expression of Eve in these neurons (Figure 6), and thus probably necessary for their proper differentiation. This is reminiscent to what has been observed in the case of the dAp neurons [30] : in these neurons, col is required for the expression of several genes (apterous, eye absent, dimmed, dopamine D1 receptor, and Neuropeptide like precursor protein 1), which are themselves required for specific differentiation features including axonal pathfinding and the capacity to produce active neuropeptides. As a consequence, in col mutant embryos, dAp differentiation is impaired [30]. col function has also been described in a third type of neurons, a particular class (class IV) of multidendritic (md) sensory neurons, which display the most extensive and complex dendritic arborizations among md neurons [27,28,29]. There, col regulates the expression of at least two target genes, pickpocket and spastin, which are required to increase dendritic arbor outgrowth, a specific differentiation property of class IV as compared to other md neurons. Taken together, these data suggest that col controls the acquisition of specific features by acting on the expression of different sets of target genes in various populations of differentiating neurons. This capability to regulate the expression of different genes in different cells is probably due to the fact that, as suggested by previous studies [21,30], Col may act mainly in combination with other transcription factors and therefore regulate its different targets in a context-dependent manner, depending on the presence/absence of its partners.

Pdu-coe is expressed in differentiating neurons in the VNC of the annelid Platynereis
After having established the neuronal expression pattern of the Drosophila col, we aimed to delineate whether reminiscent expression patterns may be found in another segmented organism that exhibits a different mode of development. We thus identified and cloned a member of this family in the annelid Platynereis dumerilii.
This gene, named Pdu-coe is expressed exclusively in developing neural territories of both embryos and juvenile worms. We combined WMISH and EdU labelling to show that Pdu-coe is exclusively expressed in post-mitotic cells in the developing VNC, and noticed that the Pdu-coe expression domain covers most of the post-mitotic VNC area ( Figure 11B). To precisely characterize the expression pattern of Pdu-coe in the VNC, we have used a method of in silico alignment of gene expression patterns obtained by WMISH that allowed us to compare the expression of Pdu-coe with that of many genes previously shown to be involved in the development of the Platynereis VNC [33,34,35]. We found that Pducoe is expressed in many cells, probably neurons as they also express neuronal differentiation genes, such as Pdu-elav and Pdu-syt, which arise from different neural progenitor domains and express several different neuronal-specification genes. Indeed, it has been previously shown that the Platynereis neurectoderm can be subdivided into several domains characterized by the expression of specific nk and pax genes : from medial to lateral, there are 4 domains, Pdu-nk2.2 + Pdu-nk6, Pdu-nk6 + Pdu-pax6, Pdu-pax6 + Pdu-pax3/7, and Pdu-pax3/7 alone [33]. We found Pdu-coe expressed in neural cells emerging from the three most lateral domains, but excluded from medialmost ones (Pdu-nk2.2 + Pdu-nk6), except for neurons located along the commissures (Figure 11 B9). However, our observations did not allow us to determine whether these neurons actually originate from the Pdu-nk2.2 + Pdu-nk6 domain or if they differentiate in more lateral domains and migrate toward the midline to achieve their final position.
Each of these four domains has been shown to produce specific neuronal types characterized by the expression of neuronal specification genes, such as Pdu-chx10, Pdu-hb9, Pdu-islet [33], and Pdu-gata. We found that the Pdu-coe expression domain overlaps with those of all these genes ( Figure 10). Although our method of in silico pattern registration does not allow to demonstrate whether the studied genes are actually expressed in the same cells, it is much likely that Pdu-coe is co-expressed with each of these genes in, at least, some of the VNC cells. Our observations therefore suggest that Pdu-coe is expressed in most neurons of the VNC (some of the most medial cells expressing Pdu-gata and possibly Pdu-hb9 are not included in the Pdu-coe domain), and that these neurons constitute diverse subpopulations characterized by the expression of specific neuronal specification genes ( Figure 11B9). Given our observation that in Drosophila Col expression is excluded from motor neurons, an important issue is nevertheless to know whether Pdu-coe is expressed in Platynereis motor neurons. Unfortunately, we are unable to give a definitive answer to this question. Based on comparisons with vertebrates, it has been suggested that Pdu-hb9 + cholinergic neurons emerging from the Pdu-nk6 + Pdu-pax6 domain are the Platynereis motor neurons [33]. We found that some Pdu-coe expressing cells derive from the Pdu-nk6 + Pdu-pax6 domain and likely co-express Pdu-hb9 and Pdu-VAChT (a marker of cholinergic neurons), suggesting that Pdu-coe may be expressed in motor neurons. However, we have to consider that all cholinergic neurons are not necessarily motor neurons (some of them may be interneurons as it is the case in Drosophila), that all motor neurons are not necessarily cholinergic (motor neurons are glutamatergic in Drosophila for example), and that hb9 is expressed in motor neurons and interneurons, both in Drosophila [41,66] and in vertebrates [67,68]. Further studies would therefore be required to define the exact position and number of motor neurons in Platynereis and to determine whether these neurons express Pdu-coe.

Comparative analysis of coe genes in animals: conservation of ancestral functions versus multiple recruitments
One striking property of the coe gene family is its strong conservation in animals: coe genes have been found in all animal species in which these genes have been searched, usually as single gene, and encode proteins with a very high sequence similarity [3,5,6]. It is therefore tempting to hypothesize that these genes also share well conserved functions. One such conserved function might be during neurogenesis as all coe genes studied to date are expressed in the developing nervous system. However, neurogenesis is a complex process that involves many steps and many genes -to define whether coe genes may play homologous roles during nervous system development in various animal lineages requires defining their function more precisely during the neurogenesis of different animals. Our work combined with some previously published data sheds light on a striking correlation between coe genes and one particular aspect of the neurogenesis process, namely the neuronal differentiation. As discussed above, Drosophila col is likely involved in the differentiation of various types of interneurons. In Platynereis, Pdu-coe is expressed in most neuron classes of the VNC and might participate in their differentiation. In another annelid, Capitella teleta, coe is also broadly expressed in the developing VNC [6] in a pattern reminiscent to the one found in Platynereis. unc-3, the C. elegans coe gene, is required for some aspects of the differentiation of both CNS and PNS neurons [31,32]. In vertebrates, Ebf genes, the vertebrate coe orthologs, are expressed in most post mitotic differentiating neurons of the CNS and required for at least some aspects of their differentiation [10,11,12,13,14,15,16,17]. Closely related to the vertebrate lineage, the ascidian Ciona intestinalis (urochordate) shows expression of coe in a few cells of the neural plate [69] and all differentiating neural precursors of the visceral ganglion (a structure proposed to be homologous to the vertebrate spinal cord) [70]. A recent study also suggests that the coe gene in the sea urchin (echinoderm) Strongylocentrotus purpuratus is involved in the differentiation of some neurons belonging to the apical ganglion of the larva [6]. Our data therefore reinforce the hypothesis that an involvement of coe genes in neuronal differentiation is ancestral to bilaterians. As coe genes are likely to be also expressed in differentiating neuronal cells in non bilaterians, such as a cnidarian and a ctenophore [4,6], a role in neuronal differentiation is probably an ancestral feature of the coe transcription factors family, which may have arisen early in metazoan evolution. coe genes have also been identified in the sponge Amphimedon queenslandica and the placozoan Trichoplax adhaerens [5,6] which lack true neurons -it would be interesting to define the expression pattern of coe genes in these two species.
Another possible conserved function of coe genes may be during mesoderm development [6]. Drosophila col is known to be expressed in a few muscles and required for their subtype specification [21], as well as in the lymph gland where its function is required for the maintenance of hematopoietic progenitors, a prerequisite to the formation of lamellocytes (immune cells involved in response to parasites) [25]. In mice, ebf genes are key factors controlling B-cell lymphopoiesis [71,72], however, while mouse ebf genes have a cellautonomous function in B-cell development, Drosophila col is not expressed in the lamellocytes and controls their differentiation in a non-cell-autonomous manner [25]. In addition, the determination of homologies between vertebrate and non vertebrate blood cells is still challenging [73]. Nevertheless, vertebrate ebf genes can also act in a non cell-autonomous manner as shown during mouse bone development, where ebf2 is expressed in osteoblast progenitors and acts as a regulator of osteoblast-dependent osteoclast differentiation [74]. Another important mesoderm related function of coe genes has been recently uncovered in the urochordate Ciona intestinalis. Here, Ci-coe acts as a critical determinant in the specification of atrial siphon muscles, a cell lineage that arises from the same progenitors as cardiomyocytes [75]. Furthermore, expression of the Xebf2 gene in the anterior lateral mesoderm of Xenopus tropicalis embryos suggests that coe genes may be a conserved determinant of cardiopharyngeal mesoderm development in chordates [75]. However, mesodermal expressions were not found in another chordate, the Amphioxus [18] or in the non-chordate deuterostome, the sea urchin S. purpuratus [6]. Regarding protostome species, a recent study reports the expression of coe genes in mesodermal derivatives in two annelids and a mollusc [6]. While in the annelid Capitella teleta, this expression seems to be localized in the mesodermal stem cells of the trunk, in the other annelid Chaetopterus and in the mollusc Haliotis asinina the identities of the mesodermal coe-expressing cells are unknown. In this study expression of Pdu-coe in the mesoderm was not detected in Platynereis. It is therefore impossible so far to define whether the existence of some mesodermal expression of coe genes is ancestral to lophotrochozoans (and in this case, this expression would have been lost in Platynereis or would be endowed by another uncharacterized coe gene) or whether mesodermal expressions have been independently acquired in some lophotrochozoan lineages. Further studies will therefore be required to test the hypothesis of ancestral function of coe genes in the development of mesodermal derivatives and particularly the stimulating hypothesis of coe genes being involved in the differentiation of some ancestral bilaterian immune cells [25].
More generally, members of the coe family are involved in the differentiation of an astonishing diversity of cell types among bilaterians and even within one species. In Drosophila, for example, in addition to CNS and PNS cells, muscles, blood cells, col is also involved in the patterning of the wing disc downstream of the Hedgehog signalling pathway [23,24]. Such involvement in appendage patterning and/or Hedgehog pathway is not found in other species. We therefore propose that several recruitments occurred independently during bilaterian evolution, reflecting a high plasticity of Coe proteins related to their combinatorial mode of action. We hypothesize that Coe proteins are transcription factors that have been co-opted several times in pre-existing regulatory networks to ensure robust gene expression regulation. In this view, we would expect to find species -or lineage-specific expression sites of coe genes when looking at additional speciesthe expression in the foregut and mesodermal stems cells of Capitella teleta [6], not found in the other annelid Platynereis, may be such a specific expression. At the molecular level, our hypothesis would imply that target genes of Coe proteins may vary largely in different cells and in different species and that the specificity in Coe target gene recognition would depend on the interaction with other transcription factors or other protein cofactors. While some data in agreement with this proposal have been found in Drosophila [21,30], further test of this hypothesis awaits more systematic searches for Coe target genes and Coe interactors.
Platynereis dumerilii: worms were obtained from a breeding culture established in the Institut Jacques Monod (Paris), according to the protocol of Dorresteijn et al [86]. Larvae and regenerating worms were fixed in 4% PFA in PBS + 0.1%Tween20 and stored at 220uC in methanol.
Isolation of Pdu-coe and phylogenetic analyses PCR using degenerated primers (described in [12]) on cDNAs from mixed larval stages was used to clone a 560 bp fragment of Pdu-coe. Larger fragments were subsequently amplified using SMART TM RACE cDNA amplification procedures with genespecific primers. PCR products were TA cloned into the PCR2.1 vector (Invitrogen), and sequenced. The largest Pdu-coe clone (916 pb), corresponding to a partial cDNA, was then used as templates to produce RNA antisense probes for WMISH. Despite extensive efforts, we were unable to clone a full-length Pdu-coe cDNA. Accession number for Pdu-coe is GU169416. Multiple sequence alignments were built with ClustalW ( Figure S7), using a large set of COE orthologs identified by BLAST (mainly from animal whose genome has been fully sequenced). Neighbour-joining (NJ) reconstructions were performed with the PAUP 4.0 program using the BioNJ algorithm and 1000 bootstrap replicates [87,88]. Maximum likelihood (ML) analyses were performed with PhyML using the WAG model of amino acid substitutions [89] and 150 bootstrap replicates. Bayesian interference was performed using the Markov chain Monte Carlo method as implemented in MrBayes, using the WAG model. Two independent Markov chains were run, each including 3,000,000 generations Monte Carlo steps. The presented tree is a consensus of those obtained with methods mentioned above. Marginal probabilities at each internal branch were taken as a measure of statistical support.

Microscopy, registration and images processing
Drosophila melanogaster. For all fluorescent immunostaining a Zeiss LSM 5 Confocal microscope (Zeiss, Oberkochen, Germany) was used to collect data. Confocal stacks were manipulated using the LSM 5 software and Adobe Photoshop (Adobe Systems, San Jose, California, United States). Double-labelled images were false colored using Photoshop for the benefit of colorblind readers.
Platynereis dumerilii. Bright field images were taken on a Leica microscope using DIC optics. Adjustment of brightness/ contrast and Z projections were performed using the ImageJ software. Confocal images were taken with a Leica SP5 confocal microscope with a 63x oil immersion objective. For 3D pattern registration, confocal pictures were independently acquired on two distinct larvae of the same litter and stage. Signals acquired included whole mount reflection [62] and acetylated tubulin antibody staining marking the VNC network. Both signals were 3D reconstructed using the Amira software. Embryos were then aligned using the registration tool included in Amira, using only the reproducible VNC network antibody staining as a reference. Images presented in figure 8 are snapshots of the 3D reconstructed and aligned embryos, with only the reflection signals displayed (see Figure S6 for Registration tests).