Nervous systems of the sea anemone Nematostella vectensis are generated by ectoderm and endoderm and shaped by distinct mechanisms

Cnidarians (sea anemones, corals and jellyfishes) hold a key phylogenetic position for understanding early nervous system evolution: Cnidaria is the likely sister group to the Bilateria and one of only two extant non-bilaterian clades that possess a nervous system (Fig. 1A) (Hejnol et al., 2009; Medina et al., 2001; Putnam et al., 2007; Wallberg et al., 2004; Philippe et al., 2009; Schierwater et al., 2009). Thus, comparing neural development in Cnidaria and Bilateria can help to reconstruct early events in nervous system evolution.

Cnidaria is a diverse clade in which Anthozoa form a sister group to the Medusozoa, consisting of Staurozoa, Scyphozoa, Cubozoa and Hydrozoa (Fig. 1A) (Collins et al., 2006). Cnidarians have two germ layers, the ectoderm and endoderm, separated by extracellular matrix, the mesoglea. Anthozoans, staurozoans and some hydrozoans develop through a swimming planula into a sexually mature polyp (Fig. 1B-D), whereas in many medusozoans sexual medusae develop from asexual polyps.

The nervous system of adult cnidarian polyps consists of epithelial sensory cells, basally located ganglion cells, and neurites from both cell types that constitute basiepithelial networks alongside the mesoglea (Fig. 1E) (Fautin and Mariscal, 1991; Lesh-Laurie and Suchy, 1991; Thomas and Edwards, 1991). Both sensory and ganglion cells can synapse to epitheliomuscular cells and other neurons. In contrast to most bilaterians, cnidarians often possess both endodermal and ectodermal nervous systems. The overall architecture of the adult nervous system in the hydrozoan polyp Hydra resembles a nerve net with regional differences in neuron densities (reviewed by Koizumi et al., 2004), whereas anthozoan polyps often display accumulations of neurons and neurites associated with longitudinal endodermal infoldings termed mesenteries (e.g. Batham et al., 1960). Despite local differences in neuron density and accumulations of neurites, cnidarian polyps lack morphological centralization comparable to the brains of bilaterians.

Whether the strikingly different architecture of bilaterian and cnidarian nervous systems is established by correspondingly different developmental mechanisms, or whether similar mechanisms are used in different ways, is not well understood. In bilaterians, nervous systems are almost invariably generated by the ectoderm [e.g. ecdysozoans (Hartenstein and Campos Ortega, 1984; Scholtz, 1992; Stollewerk et al., 2001), lophotrochozoans (Jacob, 1984; Kadner and Stollewerk, 2004; Meyer and Seaver, 2009) and deuterostomes (Miyamoto et al., 2010; Nakajima et al., 2004; Rao and Jacobson, 2005)]. Initially, the ectoderm becomes divided into epidermal ectoderm and neuroectoderm. Whereas sensory neurons can be generated from both areas (except in vertebrates), interneurons (morphologically equivalent to cnidarian ganglion cells) are generated by the neuroectoderm. In the pseudostratified neuroectodermal epithelium, neural precursor cells (NPCs) develop in the apical layer. NPCs and/or their progeny then move basally/internally, via cell migration and/or cell division oriented parallel to the epithelial surface, to generate mature neurons in the basal epithelium (Harris and Hartenstein, 2008).

Whether a similar mode of ectodermal neurogenesis exists in cnidarians is unclear. Hydrozoans contain a particular type of stem cell called interstitial cells (i-cells) that are thought to derive from the endoderm, based on ultrastructural, developmental genetic and embryological experimental evidence (Genikhovich et al., 2006; Martin, 1988; Martin and Archer, 1986; Summers and Haynes, 1969; Thomas et al., 1987). Interstitial cells can give rise to neurons, cnidocytes (stinging cells), gland cells and, in some species, also epithelial cells (Bosch et al., 2010; Bosch and David, 1987; David and Murphy, 1977; Muller et al., 2004; Watanabe et al., 2009). Neurogenesis occurs continuously in adult Hydra, and all neurons are derived from interstitial cells (Campbell, 1976), whereas during embryonic development of Pennaria disticha and Clytia gregaria sensory cells, but not ganglion cells, can be generated in the absence of interstitial cells (Martin and Thomas, 1981; Thomas et al., 1987). In the scyphozoan Aurelia, the earliest detectable (FMRFamide-immunoreactive) sensory cells occur in the ectoderm, suggesting an ectodermal origin (Nakanishi et al., 2008). By contrast, RFamide-immunoreactive NPCs were reported to first occur in the endoderm of late stage gastrulae in the anthozoan Nematostella vectensis and later migrate into the ectoderm to differentiate into mature neurons, indicating an endodermal origin (Galliot et al., 2009). However, based on the analysis of gene expression patterns Marlow et al. (Marlow et al., 2009) suggested that early neurogenesis might occur in the ectoderm in N. vectensis. Thus, whether ectodermal or endodermal neurogenesis (or both) is the ancestral condition within cnidarians, and whether ectodermal neurogenesis is ancestral for Cnidaria and Bilateria, remain unclear.

After the specification of neurons, the outgrowth of neurites and the formation of synaptic contacts characterize neural differentiation. In bilaterians, growing axons and dendrites are guided to final or intermediate target areas by a combination of target-derived long-range guidance signals and permissive or inhibitory locally acting cues (Burden et al., 2008; Kolodkin and Tessier-Lavigne, 2008). Whether similar guidance mechanisms act during the development of the morphologically less complex cnidarian nervous systems is not known.

Here we investigated nervous system development in the anthozoan N. vectensis with the aim of identifying similarities and differences in neural development between this cnidarian model and bilaterians. Nematostella has recently emerged as an important model system for cnidarian development that is amenable to the analysis of gene function (Pankow and Bamberger, 2007; Putnam et al., 2007; Rentzsch et al., 2008; Technau and Steele, 2011; Wikramanayake et al., 2003). N. vectensis gastrulates by invagination before developing into a free-swimming planula with an apical tuft at its aboral pole (Kraus and Technau, 2006; Magie et al., 2007). The planula transforms into a polyp with a tube-shaped body and only one opening, traditionally called the mouth (Hand and Uhlinger, 1992), surrounded by tentacles for prey capture and feeding (Fig. 1B-D). A previous report indicated that it has ectodermal and endodermal nervous systems with presumptive neural structures such as the apical organ in the planula and the oral and pharyngeal ‘nerve rings’ in the polyp (Marlow et al., 2009) (but see Discussion).

We generated a neuron-specific transgenic reporter line of N. vectensis in which the expression of the fluorescent protein mOrange (Renfer et al., 2009; Shaner et al., 2004) is driven by regulatory elements of the NvElav1 gene, an ortholog of bilaterian elav1 genes (supplementary material Figs S7, S8) (Marlow et al., 2009). Elav1 genes encode highly conserved RNA-binding proteins that have been implicated in neural differentiation across Bilateria (Akamatsu et al., 2005; Akamatsu et al., 1999; Campos et al., 1985; Denes et al., 2007; Jimenez and Campos-Ortega, 1987; Meyer and Seaver, 2009; Nomaksteinsky et al., 2009; Satoh et al., 2001). We used this transgenic line together with in situ hybridization, immunohistochemistry, confocal microscopy and embryological experiments to study the formation of the Nematostella nervous system in detail. We show that early neurogenesis begins in the ectoderm during gastrulation, followed by endodermal neurogenesis, a change in the orientation of neurite extension of ectodermal neurons during planula development, and tentacular ectodermal neurogenesis during polyp development. Furthermore, we use morpholino-mediated gene knockdown to demonstrate that the development of a subset of ectodermal neurons requires NvElav1. Our findings combined with existing comparative data imply that neurogenesis from the ectodermal epithelium evolved prior to the divergence of Cnidaria and Bilateria and support an ancient role for elav1 genes in neural development. Furthermore, the Nematostella planula appears to have almost uniform neurogenic potential, and its seemingly simple nervous system architecture is established by changing neurite guidance systems.

Nematostella vectensis were cultured as described previously (Fritzenwanker and Technau, 2002; Hand and Uhlinger, 1992).

Cloning was performed by standard procedures. For RACE reactions, the SMART RACE cDNA Amplification Kit (BD Biosciences) was used.

The NvElav1::mOrange transgenic line was generated as described (Renfer et al., 2009). A 2.43 kb region of genomic sequence upstream of the start codon of the NvElav1 gene (genomic coordinates: scaffold 210, 220684-223104; http://genome.jgi-psf.org/Nemve1/Nemve1.home.html) was cloned in front of the open reading frame of mOrange (Shaner et al., 2004) carrying a membrane-tethering CAAX domain at the C-terminus. The I-SceI-digested plasmid was injected into fertilized eggs at 20 ng/μl, animals were raised and crossed to wild-type polyps to identify carriers, and two independent lines were established. All animals used in this study are derived from incrosses of F1 heterozygotes. Where indicated in the figure legend, mOrange was visualized with anti-DsRed antibody (Clontech 632496).

Late stage larvae (mid-planula II through primary polyp) were anesthetized in 2.43% MgCl₂. Embryos and larvae were fixed as described (Rentzsch et al., 2008), except that PBS was supplemented with 0.5% Triton X-100 (PBSTr).

Fixed specimens were rehydrated and washed in PBSTr and blocked with 3% normal goat serum (NGS) in PBSTr. Primary antibodies were incubated with the specimens in PBSTr. We used antibodies against FMRFamide (1:500, Millipore AB15348), GLWamide [1:200 (Schmich et al., 1998)], DsRed (1:100, Clontech 632496), acetylated tubulin (mouse, 1:500, Sigma T6793) and tyrosinated tubulin (mouse, 1:500, Sigma T9028). Following washes in PBSTr and blocking in 3% NGS in PBSTr, specimens were incubated with secondary antibodies (Molecular Probes) conjugated with Alexa Fluor 568 (rabbit, 1:200), Alexa Fluor 488 (mouse, 1:200) and Alexa Fluor 647 (rabbit, 1:200). Nuclei were labeled using DAPI (1:1000, Molecular Probes) or TO-PRO-3 iodide (1:1000, Molecular Probes) and filamentous actin was labeled using Alexa Fluor 488-conjugated phalloidin (1:25, Molecular Probes). Specimens were incubated with fluorescent dyes together with secondary antibodies, washed in PBSTr and mounted in ProLong Gold antifade reagent (Molecular Probes). Images were recorded using a Leica SP5 confocal microscope; confocal stacks were viewed using ImageJ.

The following protocol was developed by modifying published in situ hybridization protocols (Finnerty et al., 2004; Gates et al., 2002; Nakanishi et al., 2010; Okamoto et al., 2005). Endogenous hydrogen peroxidase activity was quenched by incubating the fixed specimens in 3% hydrogen peroxide in methanol. They were then washed in ethanol and incubated in ethanol:xylene (1:1 v/v). Specimens were then washed in ethanol and methanol, followed by gradual rehydration in water, and were permeabilized in acetone (80%) at –20°C. They were washed in PBSTr, then in 1% triethanolamine in PBS, followed by addition of 0.6% acetic anhydride. Specimens were washed in PBSTr and refixed in 4% formaldehyde, followed by washes in PBSTr. The specimens were then incubated in the hybridization solution (HB: 50% formamide, 5× SSC, 1% SDS, 50 μg/ml heparin, 100 μg/ml salmon sperm DNA, 9.25 mM citric acid, 0.5% Triton-X100) at 60°C. The specimens were incubated with a digoxigenin-labeled riboprobe (MEGAscript Kit, Ambion) at a final probe concentration of 1 ng/μl for at least 60 hours at 60°C. The specimens underwent a series of washes in HB/2× SSC solutions [75/25, 50/50, 25/75 (v/v)], then in 0.05× SSC, and finally in 0.05× SSC/TNT buffer [1:1 (v/v); TNT: 0.1 M Tris-HCl pH 7.5, 0.15 M NaCl, 0.5% Triton X-100]. After blocking [TNT containing 0.5% blocking reagent (PerkinElmer FP1020)], specimens were incubated with anti-digoxygenin horseradish peroxidase (Boehringer Mannheim) together with other primary antibodies. After washing in TNT buffer, specimens were incubated in fluorophore tyramide amplification reagent (TSA Plus Kit, PerkinElmer). They were then washed in PBSTr and blocked in 3% NGS in PBSTr. Following secondary antibody incubation, specimens were mounted in ProLong Gold antifade reagent and examined using the Leica SP5 confocal microscope.

A 5-day-old late planula larva was processed and serially sectioned for TEM as described previously (Kraus and Technau, 2006). TEM was carried out at the Molecular Imaging Centre at the University of Bergen (FUGE, Norwegian Research Council).

Microinjection procedures were conducted as previously described (Rentzsch et al., 2008). Morpholinos were injected at 333 or 666 nM. MO sequences (5′ to 3′) are:

• NvElav MO1, AATTGTCGTCCATCATGTCCAACAT (targeting the start codon, underlined);

• NvElav1 MO2, ATGAACTGAAATAAGCCGTGAGCGT (targeting the 5′UTR);

• NvElav1 MO1mm, AATTCTCGTGCATGATGTGCAAGAT (control MO1); and

• Generic control MO, CATGGAGAAATCGGACTTCATATTT (control MO2).

The target sequence of NvElav1 MO2 is included in the NvElav1::mOrange construct, therefore NvElav1 MO2 could not be used for analysis in the transgenic line. NvElav1 is a single-exon gene so splice site morpholinos cannot be used.

In a search for genes with broad neural expression, we turned to the RNA-binding protein Elav, which has been shown to be expressed in a large fraction of neurons in many bilaterians and in scattered ectodermal and endodermal cells during Nematostella development (Marlow et al., 2009; Pascale et al., 2008). Moreover, regulatory elements of elav genes have been utilized to generate neuron-specific transgenic lines of Drosophila melanogaster and the zebrafish Danio rerio (Park et al., 2000; Yao and White, 1994). We first used fluorescent in situ hybridization to better characterize NvElav1-expressing cells in Nematostella. At the early to mid-gastrula stage, which is the earliest stage at which mRNA expression could be detected (Fig. 2A-D) (Marlow et al., 2009), NvElav1-expressing cells are columnar/spindle-shaped epithelial cells with middle- to apically positioned nuclei and often with apical cilia (Fig. 2D), exhibiting sensory-cell-like morphology. NvElav1 expression was not detected in dividing cells (Fig. 2A-C), and none of the NvElav1-expressing cells examined incorporated the thymidine nucleoside analog EdU (supplementary material Fig. S1). These observations suggest that early NvElav1-expressing cells are largely postmitotic differentiating sensory cells. At later stages of development, NvElav1 is expressed in a subset of differentiated neurons in the ectoderm and endoderm.

To allow visualization of the nervous system, including neurites, we generated a stable transgenic line in which a 2.4 kb fragment of genomic DNA immediately upstream of the NvElav1 start codon drives the expression of the fluorescent protein mOrange (Renfer et al., 2009; Shaner et al., 2004). Analysis of this line showed that NvElav1::mOrange expression is neuron specific, but the transgene is detectable in only 50-70% of all NvElav1 transcript-positive cells. Co-labeling with antibodies against the FMRFamide and GLWamide neuropeptides revealed that neither NvElav1 transcripts (Fig. 2E) nor the transgenic line (Fig. 2F) labels all differentiated neurons. Thus, although low-level expression in all neurons cannot be ruled out, NvElav1 does not appear to be a pan-neural marker in Nematostella. We therefore used the NvElav1::mOrange transgenic line in combination with in situ hybridization and immunohistochemistry to characterize Nematostella nervous system development in detail.

The first sign of neurogenesis is already detectable at the early to mid-gastrula stage (see supplementary material Fig. S2 for staging), when transcripts for the neuropeptide precursor NvRF (Anctil, 2009; Marlow et al., 2009) are visible in a subset of apically positioned dividing cells (Fig. 3A,B) [here referred to as neural precursor cells (NPCs)] as well as in cells with centrally positioned nuclei within the ectodermal epithelium, presumably differentiating neurons (Fig. 3B). This suggests that at late gastrula stage NPCs differentiate into sensory cells with apical cilia and basal neuronal processes, as well as into ganglion cells (insets in Fig. 3C), albeit fewer in number. At this stage, sensory cells are distributed in the ectoderm except for in the immediate vicinity of the oral and aboral poles (Fig. 3D; supplementary material Fig. S3). However, they become visible in the oral and pharyngeal ectoderm by the mid-planula I stage (Fig. 2F, Fig. 3F). Neurites of these early-born ectodermal neurons lengthen and project in the aboral direction, towards the base of the apical organ (Fig. 3E; see supplementary material Fig. S4 for details of apical organ development). Electron microscopy shows that these neurites contain opaque, dense cored vesicles with evidence of interneuronal synapses (Fig. 3G), but we failed to detect synapses between apical organ cells and underlying neurites (ten separate TEM sections analyzed).

A similar pattern of neurogenesis occurs in developing tentacles during polyp formation (Fig. 3H-J). At the tentacle bud stage, nuclei are separated into superficial and deeper layers in the ectodermal epithelium (Fig. 3I,J); compare with the late planula (supplementary material Fig. S2F). As in NPCs observed during gastrulation, a majority of NvRF-expressing cells in tentacle buds contain apically positioned nuclei (Fig. 3I,J); however, mitotic spindles were not observed in these cells. In addition, tentacle bud ectoderm contains NvRF-expressing sensory cells and ganglion cells with middle- or basally located nuclei (Fig. 3J), some of which are FMRFamide immunoreactive and extend basal neuronal processes (see Fig. 1E). Thus, putative NPCs occur apically, and their cell bodies appear to shift basally as they differentiate into mature neurons during tentacular neurogenesis, similar to the pattern of ectodermal neurogenesis during gastrulation. We conclude that neurogenesis during gastrulation and tentacle development occurs in the ectoderm.

Our observation of early ectodermal neurogenesis during gastrulation and tentacular development indicates that the ectodermal epithelium generates both sensory cells and ganglion cells in N. vectensis, in contrast to the previous claim that NPCs are derived from the endoderm (Galliot et al., 2009). In order to distinguish these two alternatives, we first generated endoderm-free larvae via transverse bisection at the early gastrula stage. Oral halves develop into small but complete primary polyps, whereas aboral halves develop into hollow spheres consisting of an ectodermal epithelium without endoderm (Fig. 4A,B) (Fritzenwanker et al., 2007; Lee et al., 2007). Four-day-old endoderm-free larvae developed morphologically normal sensory cells and ganglion cells in the ectoderm (Fig. 4C), showing that the ectodermal epithelium can autonomously (i.e. in the absence of endoderm) generate both types of neurons. Next, we analyzed whether, in the presence of endoderm, sensory and ganglion cells are indeed generated by the ectoderm. At early gastrula stage, we transplanted the invaginating endoderm and some surrounding oral ectoderm tissue of wild-type animals into NvElav1::mOrange hosts whose endoderm had been removed (Fig. 4D). mOrange-positive sensory and ganglion cells were readily detectable in the body column ectoderm of the resulting chimeric planulae (Fig. 4E). Thus, during normal development, both neural cell types are generated from the ectoderm.

The mode of ectodermal neurogenesis described above is strikingly similar to neurogenesis in bilaterians. To extend these finding to the functional level, we analyzed the role of NvElav1 in ectodermal neurogenesis. Elav is considered a conserved neural marker gene in bilaterians and functional data support a role in neural development in vertebrates and Drosophila melanogaster (Akamatsu et al., 1999; Campos et al., 1985).

To test the hypothesis that NvElav1 is required for neuronal development in Nematostella, we used two translation-blocking morpholinos (NvElav1 MO1 and MO2, see Materials and methods) and two control morpholinos (NvElav1 MO1mm and a generic control MO, referred to as ctr MO1 and MO2). NvElav1 MO1, but not ctr MO1, efficiently blocked translation of a reporter construct containing the morpholino target site (supplementary material Fig. S4). The number of NvElav1::mOrange-positive ectodermal neurons decreased in morpholino-injected 4-day mid-planulae, on average from 102 (uninjected, n=11) to 14.8 (NvElav1 MO1, n=5; P=0.0002, two-tailed Student's t-test; Fig. 5A), whereas injection of control morpholinos had little effect (ctr MO1, 99.4, n=6, P=0.839; ctr MO2, 93.6, n=4, P=0.720). Similarly, FMRFamide/GLWamide-immunoreactive neurons were reduced from 92.2 (uninjected, n=7) to 27.2 (NvElav1 MO1, n=8; P<0.001, two-tailed Student's t-test) and 25.2 (NvElav1 MO2, n=5), with little effect upon injection of control morpholinos (ctr MO1, 93.4, n=7, P=0.958; ctr MO2, 79.6, n=5, P=0.482) (Fig. 5B). The morpholino affected the NvElav1::mOrange-positive population of neurons more strongly than the FMRFamide-positive population, as the proportion of FMRFamide-positive neurons versus the total number of neurons labeled by FMRFamide and/or mOrange increased, on average, from 0.536 in uninjected controls (n=5) to 0.846 in NvElav1 MO1-injected mid-planula larvae (n=5, P<0.001), whereas injection of ctr MO1 had no significant effect (0.527, n=5, P=0.854). This is consistent with the above gene expression data showing that only a subset of FMRFamide-positive neurons express NvElav1 transcript at high levels. This result, along with the unperturbed development of the ectodermal apical organ and the endodermal longitudinal muscle fibers, support morpholino specificity. We found no evidence for increased numbers of apoptotic cells in NvElav1 MO-injected animals by TUNEL staining, nor did we observe changes in proliferation by EdU labeling (not shown). Development of the endodermal nervous system was not affected by injection of NvElav1 MO. We assume that this is due to decreased efficiency of the NvElav1 MO at the later developmental stage of endodermal neurogenesis. However, it might also reflect a differential requirement for NvElav1 in ectodermal versus endodermal neurogenesis.

In summary, these observations show that NvElav1 is required for normal development of a subset of ectodermal neurons.

The outgrowth of ectodermal neurites before the mid-planula stage occurred with unexpected polarity towards the aboral region, the site at which the apical sense organ develops (supplementary material Fig. S5). This polarized outgrowth is difficult to reconcile with the formation of what is often assumed to be a diffusely organized nerve net of cnidarian polyps. However, from the mid-planula II stage, we observed that newly born ectodermal neurons develop transversely oriented neurites (Fig. 6B). The average angle of the neurites from NvElav1::mOrange-positive ectodermal sensory cells changed from 30° (‘longitudinal’) in the mid-planula I stage to 80° (‘transverse’) in the tentacle bud stage (P=1.28×10^–11, two-tailed Student's t-test; Fig. 6A-C) and the number of ectodermal neurons in the body column (including the aboral region) increased on average from 30 to 100 (P<0.01, two-tailed Student's t-test).

This change was also observed in GLWamide-immunoreactive ectodermal neurons, the majority of which were NvElav1::mOrange negative (mid-planula I, 26.4°; tentacle bud, 72°; P=9.0×10^–6; Fig. 6D,E). Thus, the development of the ectodermal nervous system consists of two phases with different preferential orientation of neurite outgrowth.

The presence of endodermal nervous systems has been described in cnidarian polyps and medusae (e.g. Anctil, 2000; Davis, 1974; Grimmelikhuijzen, 1983; Singla, 1978), but information about the development of these endodermal nervous systems is scarce. We observed that, in Nematostella, endodermal NvElav1-expressing sensory cells become evident at the early planula stage (Fig. 7A) and differentiate during planula stages throughout the endodermal epithelium (Fig. 7B,C) (Marlow et al., 2009). After transplantation of wild-type endoderm plus surrounding ectoderm into NvElav1::mOrange host ectoderm we could not detect any mOrange-positive cells in the endoderm (Fig. 4E). Reciprocal transplantations (NvElav1::mOrange endoderm plus surrounding ectoderm into wild-type ectoderm) demonstrated that the transplanted endoderm is able to generate mOrange-positive neurons (Fig. 4F). Similarly, after integration of small pieces of NvElav1::mOrange endodermal plate into the endoderm of wild-type hosts, mOrange-positive neurons were detected in the endoderm but not in the ectoderm (Fig. 4G,H). These experiments show that there is little or no migration of NvElav1::mOrange-positive neurons or their precursors from the ectoderm into the endoderm and that endodermal neurons are generated directly from the endodermal epithelium.

The endodermal neurons form a basiepithelial network that initially does not display obvious polarity. However, once parietal muscle fibers form at the junction between the mesentery and the body wall, neurites turn either orally or aborally at this site, leading to the formation of prominent tracts of longitudinally oriented neurites along each side of these muscle fiber bundles (Fig. 7B,C,E,F). These neurite bundles converge at the base of the aboral pole, forming a mesh-like structure (Fig. 7E, inset), but a concentration of neuronal cell bodies was not observed at the point of convergence. Irregularly spaced small clusters of NvElav1::mOrange-labeled neurons can be found along the longitudinal neurite bundles (Fig. 7F). A set of neurons occurs between neighboring parietal muscle bundles and sends basiepithelial neurites that connect with each other and with parietal muscle-associated neurons (Fig. 7E,F). These endodermal neurites do not follow the circumferential musculature (supplementary material Fig. S9A). Although the number of endodermal neurons increases during further development of the polyp, the overall organization of the endodermal nervous system is maintained (Fig. S9B; see supplementary material Fig. S6 for details on the primary polyp nervous system).

The endodermal neurons that we observed initially display the elongated morphology of sensory cells and extend a cilium into the gastric cavity (Fig. 7D). However, as thinning of the endodermal epithelium occurs during planula development (supplementary material Fig. S2C-F), the cell bodies of the endodermal sensory cells appear shortened (superficially resembling ganglion cells) along the apical-basal axis at later stages (e.g. polyps).

Taken together, endodermal neurons are generated from the endodermal epithelium and the establishment of the endodermal nervous system architecture appears to be controlled by guidance systems that differ from those present in the ectoderm.

We have shown via transplantation experiments that the ectodermal epithelium generates both sensory cells and ganglion cells in N. vectensis. During gastrulation, cell bodies of dividing NPCs occur apically in the ectodermal epithelium, whereas differentiated neurons have centrally or basally located cell bodies, suggesting basally oriented translocation of postmitotic neural cell bodies. Although basal translocation of postmitotic cell bodies or nuclei of epithelial cells might not be restricted to neural cells (Meyer et al., 2011), this migration is strikingly similar to interkinetic nuclear migration described in vertebrate neuroepithelia (Taverna and Huttner, 2010) and to the basally oriented migration of differentiating neural cells/cell bodies in a broad range of bilaterians (e.g. Denes et al., 2007; Jacob, 1984; Kadner and Stollewerk, 2004; Meyer and Seaver, 2009; Stollewerk et al., 2001). In addition, the function of NvElav1 appears to be similar to that of its orthologs in Bilateria: postmitotic expression is required for development of ectodermal neurons in N. vectensis, vertebrates and Drosophila melanogaster. These parallels imply that, in the last common ancestor of Cnidaria and Bilateria, NPCs in the ectodermal epithelium divided apically and some of their daughter cells expressed elav postmitotically to differentiate into neurons during basal migration of the cells/cell bodies (Fig. 8A), probably including sensory and ganglion cells. Thus, our findings together with evidence for ectodermal neurogenesis in the hydrozoans C. gregaria (Thomas et al., 1987) and P. disticha (Martin and Thomas, 1981), as well as in the scyphozoan Aurelia (Nakanishi et al., 2008), suggest that neurogenesis from ectodermal epithelial cells was present in the last common ancestor of cnidarians and bilaterians (Fig. 8B).

This hypothesis explains why bilaterian neurogenesis is invariably derived from the ectoderm. Adult bilaterian central nervous systems are internal organs, but they are generated by the outermost layer of the embryo, the ectoderm, and need to be subsequently internalized. This is unusual because other internal organs (e.g. muscle, kidneys, gonads) are generated by the middle layer of the embryo, the mesoderm, and it has not been clear why the central nervous system derives from the ectoderm. We propose that this is simply due to the ancestry of ectodermal neurogenesis; that is, ectodermal neurogenesis predates the origin of bilaterian mesoderm.

In vertebrates NPCs originate from a restricted area of the ectoderm (the neuroectoderm) and can first be detected after gastrulation (Harris and Hartenstein, 2008). By contrast, in Nematostella we observe NPCs already at early gastrulation stages and neurons develop throughout the ectoderm during gastrulation and planula stages, with the exception of a small area at the aboral pole from which the apical organ develops. The broadly distributed ectodermal neurogenesis also includes ganglion cells/interneurons, which in most bilaterians originate only from the neuroectoderm. Together with the development of sensory and ganglion cells in endoderm-free planulae these findings suggest that vertical induction from the endoderm is not required for ectodermal neurogenesis and that the ectoderm does not become subdivided into territories of differential neurogenic potential.

The transplantation experiments show that endodermal neurons of Nematostella develop from the endodermal epithelium. This indicates that both the ectoderm and endoderm of Nematostella planulae are neurogenic. Since the widespread occurrence of endodermal nervous systems appears to be unique to cnidarians, it will be interesting to understand the genetic basis of this phenomenon. NvElav1 is expressed in both ectodermal and endodermal nervous systems during development, suggesting the possibility of deployment of pre-existing genetic mechanism(s) of ectodermal neurogenesis to neurogenesis from the endodermal epithelium. This suggests that the genetic mechanisms that control neural specification and differentiation might be similar in Nematostella and bilaterians, although it remains to be determined whether the signaling pathways that control the segregation of neural and non-neural tissues and territories in vertebrates have equivalent functions in Nematostella.

Additional genetic data comparing ectodermal and endodermal nervous system development in N. vectensis, and across distantly related cnidarian taxa, should shed further light on how this potential novelty was achieved in Cnidaria.

At mid-planula I stage, the neurites of most ectodermal neurons project aborally towards the base of the apical organ. Since the apical organ itself is typically considered a sensory structure, it could be that sensory information from the ectoderm becomes integrated around the aboral pole. However, there is no accumulation of neural cell bodies or a ganglion-like structure as is often associated with apical organs in bilaterians. Similarly, from mid-planula II stage, prominent bundles of endodermal neurites flank the developing mesenteries and form a mesh at the aboral pole area, but again no accumulation of neural cells at the aboral pole is detectable. We do observe clusters of endodermal neurons along the mesenteries, but we could not detect any regularity in their positions along the oral-aboral axis or in their distribution among the eight mesenteries. Although a cluster of sensory cells develops around the mouth opening by the tentacle bud stage (Fig. 7D), we did not observe the formation of oral or pharyngeal nerve rings.

Thus, despite clear accumulations of neurites and the occurrence of small clusters of neural cell bodies, we cannot detect any morphological structure that would be comparable to the nervous system centralization observed in bilaterians.

A drastic change in the overall ectodermal neurite orientation occurred in the body column during planula development, from longitudinal in early-born to transverse in later-born neurons. The initial aboral neurite orientation might reflect putative global attractive or repulsive signals originating from the aboral or oral pole, respectively, and could therefore be related to the patterning of the oral-aboral body axis. Alternatively, endodermal processes that penetrate the mesoglea may form synapses with ectodermal neurites and might be the source of attractive guidance cues. These processes from the endoderm are initially concentrated in the aboral region and later occur throughout the body column, correlating with the change in neurite orientation (supplementary material Fig. S10).

We were unable to identify morphologically distinguishable landmarks that transverse ectodermal neurites would project to. Moreover, endodermal neurites developing at the same time do not display preferential orientation, suggesting that the architectures of the endodermal and ectodermal nervous systems are shaped by separate guidance mechanisms. We thus recognize three different neurite guidance mechanisms (two in the ectoderm and one in the endoderm) during Nematostella development. Although we cannot exclude the possibility that additional subpopulations of neurons display other patterns of neurite orientation, our results suggest that guidance mechanisms acting more globally and probably over long distances (mainly in the early stage ectoderm) and others acting more locally (mainly in the endoderm and late stage ectoderm) contribute to the formation of the nervous system in Nematostella. These mechanisms might differ from each other in the type of the signaling molecules that are used and/or in the spatiotemporal expression patterns of the same guidance molecules.

Our results suggest that in Nematostella neurogenesis commences in the ectoderm and that neurons originate from epithelial cells. This is in contrast to the situation in hydrozoans, where most neurons originate from interstitial stem cells. The similarity between neurogenesis in Nematostella and that in bilaterians indicates that the condition in hydrozoans might be derived.

It remains to be determined to what extent other findings of our study can be generalized for cnidarians. For example, strictly aboral orientation of early ectodermal neurites in the hydrozoan Podocoryne is restricted to neurons in the oral region (Groger and Schmid, 2001; Momose and Schmid, 2006), whereas RFamide-positive ectodermal neurites in the scyphozoan Aurelia (Nakanishi et al., 2008) and the anthozoan Acropora millepora (Hayward et al., 2001) develop oral and aboral projections. Similarly, the extensive endodermal nervous system development in planulae observed in N. vectensis has not been described in other cnidarians (de Jong et al., 2006; Groger and Schmid, 2001; Martin, 2000; Nakanishi et al., 2008; Piraino et al., 2011). Additional studies of nervous system formation at high temporal and spatial resolution in these and other cnidarians are needed for a detailed reconstruction of the evolutionary histories of cnidarian neural development.

In conclusion, our data suggest that neurogenesis from ectodermal epithelial cells is ancestral for cnidarians and bilaterians and that post-transcriptional regulation of gene activity by elav1 genes is an ancient feature of neural development. We show that, in contrast to bilaterians, almost the entire tissue of the Nematostella planula is neurogenic and that, similar to the situation in bilaterians, multiple neurite guidance mechanisms are involved in the formation of the only superficially simple nervous system of Nematostella.

We thank Thomas Leitz for generously providing the anti-GLWamide antibody; Alina Rey, Tessa Bargmann and Bård Steinar Giezendanner for excellent care of the Nematostella culture; Harald Hausen for help with interpretation of electron microscopy data; Anne Nyhaug for help with sample preparation for TEM; and Roger Tsien's laboratory for allowing the use of mOrange and lab members for constructive discussions.