Evolution of bilaterian central nervous systems: a single origin?

The question of whether the ancestral bilaterian had a central nervous system (CNS) or a diffuse ectodermal nervous system has been hotly debated. Considerable evidence supports the theory that a CNS evolved just once. However, an alternative view proposes that the chordate CNS evolved from the ectodermal nerve net of a hemichordate-like ancestral deuterostome, implying independent evolution of the CNS in chordates and protostomes. To specify morphological divisions along the anterior/posterior axis, this ancestor used gene networks homologous to those patterning three organizing centers in the vertebrate brain: the anterior neural ridge, the zona limitans intrathalamica and the isthmic organizer, and subsequent evolution of the vertebrate brain involved elaboration of these ancestral signaling centers; however, all or part of these signaling centers were lost from the CNS of invertebrate chordates. The present review analyzes the evidence for and against these theories. The bulk of the evidence indicates that a CNS evolved just once – in the ancestral bilaterian. Importantly, in both protostomes and deuterostomes, the CNS represents a portion of a generally neurogenic ectoderm that is internalized and receives and integrates inputs from sensory cells in the remainder of the ectoderm. The expression patterns of genes involved in medio/lateral (dorso/ventral) patterning of the CNS are similar in protostomes and chordates; however, these genes are not similarly expressed in the ectoderm outside the CNS. Thus, their expression is a better criterion for CNS homologs than the expression of anterior/posterior patterning genes, many of which (for example, Hox genes) are similarly expressed both in the CNS and in the remainder of the ectoderm in many bilaterians.

The evidence leaves hemichordates in an ambiguous position – either CNS centralization was lost to some extent at the base of the hemichordates, or even earlier, at the base of the hemichordates + echinoderms, or one of the two hemichordate nerve cords is homologous to the CNS of protostomes and chordates. In any event, the presence of part of the genetic machinery for the anterior neural ridge, the zona limitans intrathalamica and the isthmic organizer in invertebrate chordates together with similar morphology indicates that these organizers were present, at least in part, at the base of the chordates and were probably elaborated upon in the vertebrate lineage.

Keywords: Central nervous system evolution, Hemichordate, Urbilaterian, Amphioxus, Tunicate, Vertebrate brain, Nerve cord

There is general agreement that the relatively complex central nervous system (CNS) characterizing most higher metazoan animals can be traced back through evolution to a nerve net in a cnidarian-like ancestor. However, it is highly controversial whether the nervous system of the next evolutionary stage (the urbilaterian) still consisted solely of a nerve net or included a CNS. If the urbilaterian had only a nerve net, then the CNSs of protostomes and deuterostomes likely evolved independently. In contrast, the view that the urbilaterian had a CNS is consistent with the view that the CNSs of all metazoans are homologous. At present, opinion is still divided, with the majority advocating a single evolutionary origin for the CNS [1-8] and the minority favoring an urbilaterian with a nerve net [9,10].

This controversy about CNS evolution is intimately related to issues of homology, and it is useful to outline current thinking about homology at the outset. It is important not to conflate the concepts and recognition criteria for homology. There are currently three concepts underlying homology – biological [11], taxic/cladistic [12,13] and historical [14-16]. What matters for the last, which is the most familiar and most germane for the nervous system controversy, is the historical continuity of descent from a common ancestor. The historical concept requires one to be explicit about what is being compared [17]; for example, bird wings and bat wings are homologous as vertebrate forelimbs, but not as wings. Importantly, historical homologies can become very different through divergence. The three chief criteria for recognizing homology are relative position to other body parts, special quality, and transitional stages [18]. A developmental criterion, introduced by Haeckel [19], proved difficult to apply and has been largely submerged into the previous three criteria. Importantly, there are differing views concerning the hierarchical distribution of homology across levels of biological organization. In the view of Striedter and Northcutt, homology at one level (say behavior) does not necessarily connote homology at another (say morphology) [20]. In contrast, Wagner argued that structures descended from a common ancestor are homologous even if they have diverged and have no clear morphological similarity [16]. The problems raised by the hierarchical nature of homology have been heightened by the discovery of developmental gene conservation [21] and are especially noticeable in the discussion of CNS evolution.

In the past 20 years, it has been found that developmental genes and core signaling pathways are typically conserved across phyla and that gene expression patterns during development can often be used as characters for inferring homologies. Thus, although the majority view had been that the urbilaterian had a nerve net, the balance was tipped towards an urbilaterian with a CNS by the discovery that bone morphogenic protein (BMP)/decapentaplegic genes were expressed dorsally in Drosophila and ventrally in vertebrates with the BMP antagonists chordin/short gastrulation expressed on the opposite side [22] (Figure 1). In this view, which is consistent with the CNSs of all higher metazoans being homologous, a dorso/ventral (D/V) inversion occurred either in basal protostomes or in the deuterostome lineage [3,5,23]. However, in the last 10 years, studies of gene expression and function in an enteropneust (acorn worm; phylum Hemichordata) have been interpreted as evidence that the ancestral deuterostome and, by extension, the urbilaterian had a nerve net and no CNS [9,24,25]. Thus, while a CNS would have arisen close to the base of the protostomes, the evolution of a CNS in deuterostomes did not occur until the base of the chordates. In the present review, we examine the detailed evidence on both sides of the controversy and evaluate its interpretations. We conclude that a stronger case can be made for the initial appearance of the CNS at the level of the urbilaterian than for independent evolution of the CNS in more than one line of metazoan descent.

Although several features of the ancestral bilaterian in addition to the presence or absence of a CNS are widely debated, a range of molecular, developmental and comparative morphological evidence indicates that this animal was bilaterally symmetrical, with distinct anterior and posterior ends, dorsal and ventral surfaces, and left and right sides. It almost certainly had defined muscle, derived from mesoderm, allowing active locomotion and a gut with either a single opening or a separate mouth and anus [30]. Whether or not this animal had a CNS, an ectodermal nerve net or some combination of the two has been hotly debated (reviewed in [31]) (Figure 1).

One difficulty in deciding whether the ancestral bilaterian had a CNS is that the ectoderm in bilaterians is broadly neurogenic. Therefore, the distinction between the CNS and the remainder of the relatively neurogenic ectoderm is not always clear-cut. In chordates, arthropods and annelids, the distinction is most clear as there is a fully internalized concentration of neurons, axons and supporting cells along the anterior/posterior (A/P) axis (that is, a CNS) that integrates information from sensory cells both associated with the CNS (for example, eyes) and with other portions of the ectoderm and coordinates behavior. Importantly, the CNS in these organisms has an anterior concentration of discrete neural centers or “brain”, which coordinates sensory inputs and responses. At the other extreme are “diffuse ectodermal nerve nets” such as in cnidarians. However, such nerve nets are not uniform; specific types of neurons may be regionally localized [32]. An additional problem in understanding the evolution of CNSs comes with the Ambulacraria (echinoderms and hemichordates), as they have both ectodermal nerve nets and nerve cords. It is controversial whether echinoderm and/or hemichordate nerve cords, neither of which has a concentration of neurons that could be termed a brain, and the CNS of chordates have a common evolutionary origin [33,34]. Here we will use the term CNS for a nervous system that is derived from ectoderm, includes both axons and neurons and is specialized along the A/P axis with an anterior concentration of neural centers (brain), and the term “nerve cord” more broadly to include axonal tracts with few or no neurons and lacking a discrete brain. The diversity of animal nervous systems and paucity of data from some species may blur this distinction on occasion; however, we will be explicit in such instances.

It is generally agreed that bilaterians evolved from radially or bi-radially symmetrical animals, comparable in some ways to modern cnidarians. Adult cnidarians have an ectodermal nerve net with a concentration of neurons around the single gut opening (Figure 2). Therefore, if the ancestral bilaterian had already evolved a CNS, it would presumably have arisen as a concentration or amplification of neurons along one side of this nerve net, perhaps together with a reduction in numbers of neurons elsewhere in the ectoderm.

Unfortunately, no extant animal is a good stand-in for this ancestral bilaterian. Extant animals that are thought to have diverged from the bilaterian lineage before it radiated into the protostomes (Ecdysozoa and Lophotrochozoa) and the deuterostomes do not show an intermediate condition between a nerve net and a CNS (Figure 2). The best candidates for such early bilaterian offshoots are the acoel and nemertodermatid flatworms, and the xenoturbellids, which in some studies have been placed basal to the deuterostomes plus protostomes but in others are placed basal in the deuterostome lineage [35,36]. Acoels have a concentration of neurons, or a “brain”, anteriorly with up to six tracts of axons extending posteriorly [37]. In contrast, xenoturbellids have an intraepithelial nerve net that lacks aggregations of neurons or axonal tracts [38]. As a result of the lack of a clear intermediate, scenarios for evolution of CNSs are necessarily based on similarities in gene expression and neuroanatomy in the two main lineages of bilaterians: protostomes (Ecdysozoa plus Lophotrochozoa) and deuterostomes (Figure 1).

Because the CNSs in protostomes and deuterostomes are in different positions, develop rather differently and are morphologically somewhat diverse, possible homologies between them have been highly contentious. Complicating the picture is that some of the genetic mechanisms for specifying A/P positions in the CNS are common to the entire organism, including the general ectoderm exterior to the CNS, and are therefore not entirely useful for inferring homologies of CNSs. For example, some genetic mechanisms mediating A/P patterning in the CNS were clearly inherited from a cnidarian-like ancestor in which they patterned the entire body axis. Thus, Six3/6 and Irx are expressed in the aboral region of the planula larva of the sea anemone Nematostella vectensis, opposite the blastopore [39] and in the anterior end of the brain of both protostomes and deuterostomes – Six3/6 in the anterior tip of the CNS and Irx genes a little more posteriorly [24,40-44]. In N. vectensis the domains of these two genes are initially congruent, while in the CNS of bilaterians the Six3/6 domain is anterior to that of Irx. Therefore, although it is most parsimonious to propose that these genes were coopted into the CNS of an ancestral bilaterian, it cannot be ruled out that they were coopted independently into the CNS of protostomes and deuterostomes.

Hox genes are another example of A/P patterning genes that are not entirely useful for inferring homologies between the protostome and chordate CNS. The problem is that although they do mediate A/P patterning of the CNS in bilaterians [45,46], they mediate A/P patterning of other tissues as well [47-54]. Thus, while their expression patterns have been used to infer homologies between the CNS in insects and vertebrates, it remains possible that they patterned the entire body axis of the Urbilaterian and were independently coopted into the CNSs of protostomes and deuterostomes. It is not clear when a role for nested expression of Hox genes in regionalization of the A/P axis evolved. They do not appear to be involved in A/P patterning in cnidarians [55,56]. Comparisons of Hox genes in protostomes with up to 10 or 11 Hox genes and invertebrate deuterostomes with up to 15 indicate that the ancestral bilaterian had at least eight to 10 Hox genes [57], while cnidarians have up to six depending on the species and acoel flatworms have three, which are more or less regionally expressed in the surface ectoderm along the A/P axis [58,59] with later expression in putative neural precursors [49]. Thus, acoels either arose before a large Hox cluster evolved or they lost some Hox genes. Even so, a role for an expanded array of Hox genes in specification of A/P positions in the ectoderm was evidently present in the ancestral bilaterian. Thus, although expression of the Drosophila melanogaster Hox1 gene labial in a stripe at the posterior end of the tritocerebrum within the unpg (Gbx) domain has been likened to nested expression of Hox genes in the vertebrate hindbrain, with Hoxb1 being expressed in a stripe in rhombomere 4, the possibility that Hox genes were independently coopted into the CNS in protostomes and deuterostomes cannot be ruled out.

Stronger support for a single origin of the CNS comes from similar expression in the CNSs of protostomes and chordates of genes that are not expressed in comparable patterns in other tissues. Thus, Reichert and colleagues used gene expression patterns to support the perhaps surprising idea that the three parts of the Drosophila brain – protocerebrum, deutocerebrum and tritocerebrum [5,45] – are homologous to the forebrain, midbrain and hindbrain of vertebrates (Figure 2) [60]. For example, in D. melanogaster, the Otx homolog Otd is expressed throughout the protocerebrum and deutocerebrum, while unpg (homologous to Gbx) is expressed in the tritocerebrum, the subesophageal ganglion and the ventral nerve cord [5]. The domains of the two abut at the boundary between the deutocerebrum and tritocerebrum, similar to the abutting domains of Otx2 and Gbx2 at the midbrain/hindbrain boundary (MHB) in vertebrates [5,61]. In addition, although some domains of Pax2/5/8 genes are not similar between the CNS of flies and chordates, Pax2/5/8 is expressed at high levels in the posterior part of the deutocerebrum (just anterior to the deutocerebrum/tritocerebrum boundary) in D. melanogaster, while the three vertebrate Pax2/5/8 genes are expressed at high levels at the MHB [5,62]. Moreover, in third instar larvae of D. melanogaster, the earmuff gene (homologous to Fezf) is broadly expressed in the anterior brain with a posterior boundary at the protocerebrum/deutocerebrum boundary [63]. The domain is just anterior to that of mirror, one of the three Irx homologs. Similarly, Irimia and colleagues showed that in chordates, the posterior limits of Fez genes (Fez and Fez-like) abut the anterior limit of Irx1 in the forebrain [64]. In vertebrates, this is the zona limitans intrathalamica (ZLI) [65].

Compatible with a single origin of the CNS, expression of the genes mediating D/V patterning within the CNS is also conserved between protostomes and deuterostomes [66] (Figure 3). These genes are not comparably expressed in cnidarians, suggesting that they were recruited for roles in D/V patterning the CNS of an ancestral bilaterian. Notably, homologs of some key genes expressed mediolaterally in the neuroectoderm of D. melanogaster embryos are expressed in comparable domains in the vertebrate CNS. Thus, the msh gene is expressed laterally in the D. melanogaster neuroectoderm, with ind expressed in an intermediate longitudinal domain and vnd expressed in a medial stripe of neuroblasts (reviewed in [7,67]). Vertebrate homologs of these three homeobox genes are comparably expressed in the developing neural tube. Two of the three msh orthologs (Msx1, Msx2, Msx3) are expressed dorsally (that is, laterally) in the roof plate of the CNS, one of the two ind orthologs (Gsh1) is expressed in the adjacent zone (alar plate), and one of the two vnd orthologs (Nkx2.2) is expressed more ventrally (that is, medially) in the basal plate.

Additional evidence for homology of protostome and chordate nerve cords, and thus a bilaterian ancestor with a CNS, comes from neuroanatomy, neuronal function and gene expression. Strausfeld and Hirth found striking parallels between the central complex in the arthropod protocerebrum and the basal ganglia in the ventral forebrain of vertebrates [3]. In particular, the vertebrate striatum and pallidum have similar organization as, respectively, the insect fan-shaped body and ellipsoid body. Both the types of neurons and their connections and the functions of these regions are similar in the two organisms. Taken together, the data from comparative gene expression and anatomy provide relatively strong support for a single origin of the CNS in insects and chordates.

Additional evidence for a single origin of the CNS comes from comparisons between annelids and vertebrates. Not only have parallels been drawn between patterning the Drosophila and vertebrate brains, but Arendt and colleagues have also noted similarities between the genetic mechanisms patterning the nervous systems of the annelid Platynereis dumerilii and vertebrates [2,81] (reviewed in [4]). The annelid brain varies from species to species, with the brains of some species lacking clear compartments but many others having such features as complex, neuron-rich mushroom bodies (a comparatively large part of the brain in insects and annelids that integrates olfactory information) [82]. Extensive comparisons of gene expression have been used to argue for homology between the mushroom bodies and the pallium of the vertebrate brain [2,81]. For example, Bf-1 (FoxG1) is expressed in the anterior part of the vertebrate telencephalon and the pallium as well as in the tip of the annelid brain, while Wnt5/8 is expressed in the vertebrate pallium and in the annelid mushroom bodies, flanking more medial expression of Hh in both [2].

Furthermore, in P. dumerilii Six3 and Otx are expressed anteriorly in the CNS (the peristomium) with the Six3 domain extending anterior to that of Otx[42]. The posterior limit of the Otx domain abuts that of Gbx in the first larval segment, while the anterior boundaries of Hox1 and Hox4 are in the second and third larval segments [48]. Six3/6 and Otx are similarly expressed in acoel flatworms [26,83,84], and in D. melanogaster all three genes are expressed in similar patterns as in P. dumerilii. Therefore, the annelid cerebral ganglion has been homologized with the insect protocerebrum. In addition, similar to the CNS in Drosophila and vertebrates, the neuroectoderm in P. dumerilii is divided into a series of domains with outer/dorsal expression of Msx and Pax3/7 (gooseberry), intermediate expression of Nk6 and Pax6, and medial expression of Nkx2.1/Nkx2.2[1].

Together with anatomical similarities, these data showing distinct similarities in expression of genes patterning the CNS both anteriorly/posteriorly and medio/laterally between both major lineages of protostomes (Ecdysozoa and Lophotrochozoa) and vertebrates support a single origin of the CNS in the bilaterian ancestor (Figure 2). The counterargument would be that the CNS in protostomes evolved independently coopting A/P and D/V patterning mechanisms from an ancestor that used them to pattern a body axis. However, this would mean that the extensive similarities in neuronal architecture between chordates, arthropods and annelids would have been convergently evolved, which seems most unlikely.

If the CNS evolved just once, then a D/V inversion must have occurred during evolution of either protostomes or deuterostomes (reviewed in [27,85]). At present, the chief theories are as follows. The first is Anton Dohrn’s idea that a D/V inversion occurred either at the base of the protostomes or within the deuterostomes [86]. The second is the idea most recently articulated by John Gerhart, Christopher Lowe and colleagues that the ancestral deuterostome was hemichordate-like with dorsal and ventral nerve cords and an ectodermal nerve net, with the chordate CNS arising directly from the nerve net [9,24,27] or alternatively, as proposed by van Wijhe [87] and more recently by Nomaksteinsky and colleagues [88], from the dorsal nerve cord. A third theory that the ancestral bilaterian had multiple nerve cords, with one evolving into the protostome CNS and another into the deuterostome CNS, was suggested by Gerhart [27] but has received little attention.

Major evidence supporting a D/V inversion in either basal deuterostomes, basal protostomes or basal chordates is that genes involved specifying polarity of the D/V body axis are expressed in opposite orientations in protostomes and chordates. Sasai, de Robertis and colleagues found that in both groups, BMP signaling is involved in establishing D/V polarity and in neural specification, with suppression of BMP signaling being a prerequisite for formation of a CNS [22,23,89]. In agreement with a D/V inversion having occurred in either the deuterostome or protostome lineages [23], BMPs are expressed dorsally in protostomes and hemichordates and the BMP antagonist short gastrulation (= chordin in deuterostomes) is expressed ventrally, while in chordates it is the opposite – BMPs are expressed ventrally and chordin dorsally. In most bilaterians, D/V orientation of the body and position of the nerve cord are coupled; however, Hejnol and Martindale have noted that expression of BMP2/4 dorsally (opposite the future mouth) in an acoel with neurite tract(s) dorsally as well as laterally [26] supports the idea that a role for BMP/chordin in axial patterning may have preceded a role in neural patterning. Another line of evidence supporting D/V inversion comes from analysis of genes involved in left–right patterning. For example, two key regulators of this distinction, Nodal and Pitx, are expressed on the left side of chordates, but on the right in echinoderms and in some molluscs [90,91].

In summary, conserved expression of some genes along the longitudinal axis of cnidarians and in the CNS and general ectoderm of bilaterians indicates likely cooption of roles for these genes in patterning the CNS. However, similar expression of genes involved in both D/V patterning of the CNS and in A/P regionalization of the brain in chordates and protostomes together with neuroanatomical parallels provides considerable support for the idea that the bilaterian ancestor had a CNS, which was modified or possibly lost in various protostome and deuterostome lineages.

Despite considerable evidence in support of a single origin of the CNS, data from hemichordates have been interpreted as indicating that the ancestral deuterostome had a nerve net, and therefore the CNSs in chordates and protostomes evolved independently. Hemichordates and echinoderms form a clade, the Ambulacraria, which branched off the deuterostome tree as a sister group to chordates. Indirect developing members of both groups have similar pelagic larvae with an apical tuft of cilia. Echinoderms, which have pentamerous symmetry, typically have an ectodermal nerve net plus radial nerves and a circumoral nerve ring. While Haag proposed that the sea urchin radial nerves are homologous to the chordate CNS [33], most authors disagree [92,93]. Importantly, echinoderm nerve cords do not express Hox genes [51,94,95], and an extensive screen by Sly and colleagues for expression of neural patterning genes in a juvenile sea urchin failed to find evidence that the nerve ring or radial nerves are homologous to any part of the brain or nerve cord in bilaterians [93]. Moreover, Engrailed is very broadly expressed in the nervous systems and other tissues of the juvenile starfish and not in localized domains as in the chordate CNS [94]. Therefore, echinoderms are currently not considered relevant to the question of evolution of chordate nerve chords. In contrast, the worm-like enteropneust hemichordates, which have longitudinal nerve cords as well as a nerve net, have figured prominently in discussions of the evolution of chordates [10,96-98].

Inferring homologies between chordate nerve cords and hemichordate nervous tissues has been complicated by large differences in morphology. Adult hemichordates have three distinct regions: proboscis, collar and trunk. There are two classes of hemichordates – enteropneusts and pterobranchs. All four families of enteropneusts (Harrimaniidae, Spengelidae, Ptychoderidae, Torquaratoridae) have an ectodermal nerve net, located in all three regions, plus dorsal and ventral nerve cords, suggesting that this organization is a basal hemichordate characteristic. In contrast, the sessile pterobranchs have anterior tentacles and a concentration of neurons at the base of the tentacles that has been termed a brain, as well as several concentrations of neurites and associated neurons extending into the tentacles, the stalk and between the gill slits [99]. Although Romer and others argued that pterobranchs were basal hemichordates [100,101], recent molecular phylogenetic analyses do not distinguish which family is basal [102,103], leaving open the possibility that pterobranchs are derived. Indeed, fossil tube-dwelling enteropneusts from the Cambrian were recently discovered [104].

Most of the work on neural development in hemichordates concerns indirectly developing ptychoderids and the direct developing harrimaniid Saccoglossus kowalevskii (reviewed by Röttinger and Lowe [105]). Miyamoto and colleagues showed that the larval nervous system in indirect ptychoderids does not carry over into the adult; in late larvae, the larval nervous system is gradually replaced by the adult one [106]. Therefore, it is the development of the adult nervous system that is pertinent for understanding evolution of the CNS.

There are two competing theories concerning the evolutionary relationship between the nerve net and nerve cords of hemichordates and the chordate CNS (Figure 1). One theory, most recently articulated by Kaul and Stach [107], proposes that one of the hemichordate nerve cords, typically the dorsal one, is homologous to the chordate CNS. This theory implies that the ancestral deuterostome and perhaps also the ancestral bilaterian had a CNS. The chief basis for this idea is that the collar nerve cord neurulates, suggestive of neurulation in vertebrates [88,108]. Anterior and posterior to the collar, the nerve cord is continued by basiepithelial tracts of neurites, which are concentrated dorsally [99]. However, there is nothing that resembles a brain. In the direct developing S. kowalevskii, neurulation in the collar nerve cord progresses from posterior to anterior, and there are posterior and anterior neuropores [107]. The nerve cord continues posteriorly as a superficial tract of nerve cell bodies overlying nerve cell fibers and rostral of the anterior neuropore as a wide, superficial tract of both neurons and nerve fibers [88]. In addition to the longitudinal nerve cords, there is a peribranchial nerve ring, which develops from ventral to dorsal, as well as a collar nerve ring at the collar–trunk boundary. Although initially neither nerve cord was thought to contain nerve cell bodies, studies with electron microscopy and with specific nerve cell markers have demonstrated nerve cell bodies and glia in the dorsal nerve cord and at least some neurons associated with the ventral one [107,109,110]. Ventrally in the dorsal cord, there is a neuropil. Bullock [111] and Brown and colleagues [110] have suggested that the large neurons may be homologous to Mauthner cells of the lamprey and Rhode cells of amphioxus.

The developing collar and ventral nerve cords as well as the peripharyngeal cord of both Ptychodera flava and S. kowalevskii express nerve cell-specific genes including Elav, synaptogamin and also genes for peptides and proteins specific for subsets of nerve cells including VAChT, serotonin, Hb9, Drg11 and GABA. Serotonergic neurons are restricted to the peripheral nervous system, while those labeling with Drg11, Hb9 and cholinergic neurons are preferentially in the collar nerve cord [88,112].

Although most of the ptychoderid ectoderm is non-neural [88], basiepithelial nerve cells are moderately numerous in the proboscis [106]. Nomaksteinsky and colleagues suggested that the more even distribution of neurons in the basiepithelial nerve net of S. kowalevskii might represent a transient larval nervous system and that part of the diffuse nervous system of developing S. kowalevskii larvae, especially in the proboscis, will become the peripheral nervous system [88]. Based on the morphological and gene expression data, they concluded that it was ‘implausible that the enteropneust skin is homologous to the chordate CNS’. Instead, they argued that the relatively few neurons in the adult “non-neural” ectoderm constituted a peripheral nervous system, and that either the dorsal or ventral nerve cord (they could not decide which one) was homologous to that of chordates [88]. A comprehensive study of developmental gene expression in the developing nerve cords of hemichordates is sorely needed; to date only a few pictures showing expression of genes including Dlx, several Hox genes, Tbx2/3, PoxN, Pitx and Olig in the dorsal and/or ventral midline of embryos of S. kowalevskii have been published [9,24,25,113], but whether the tissue expressing these genes is the developing nerve cord or overlying ectoderm is not clear. There are no studies of developmental gene expression in indirectly developing species such as P. flava due to a long pelagic larval period [114,115].

A problem with homologizing the dorsal nerve cord of hemichordates with the chordate CNS is the finding by Lowe and colleagues [113] that, as in protostomes, BMP2-4 and BMP5-8 are expressed dorsally in S. kowalevskii, while chordin is expressed ventrally. Consistent with a role in D/V patterning, excess BMP4 protein radializes the embryos and eliminates chordin expression, indicating an evolutionarily conserved role of BMP in D/V patterning. This suggests that if either nerve cord in hemichordates is homologous to the chordate CNS, it is the ventral nerve cord, which does not neurulate, and a D/V inversion occurred at the base of the chordates. This is consistent with the gill slits being dorsal and the stomochord, a dorsal/anterior extension of the gut, having been shown to be unrelated to the notochord [116]. Confusing the issue further, in amphioxus and vertebrates, Nodal expression dorsally acts in opposition to BMP expression ventrally [117], while in sea urchin embryos, BMPs and Nodal oppose each other in patterning the oral/aboral axis (Nodal ventralizes; BMP dorsalizes), suggesting that a role for Nodal in opposing BMPs was present at the base of the deuterostomes and that a D/V inversion occurred in chordates. To some extent this is similar in S. kowalevskii, in that perturbation of Nodal signaling results in D/V patterning defects [118]. However, treatment with the Nodal inhibitor SB431542 eliminates expression of both BMP2/4 and chordin and anteriorizes embryos, indicating that Nodal posteriorizes embryos, the opposite of the situation in chordates [118]. These results suggest that the role of Nodal [119] may have been altered in S. kowalevskii. Whether a role for BMP/Nodal opposition in D/V patterning was present in the ancestral bilaterian is uncertain.

Nodal is involved in left/right patterning in a mollusk [120], but possible roles in D/V or A/P patterning in protostomes have apparently not been investigated. In summary, since it neurulates, the dorsal nerve cord of hemichordates has been proposed as homologous to the chordate CNS. However, the rather scanty data on gene expression are more compatible with homology of the ventral nerve cord and the chordate CNS. More data are clearly needed on both anatomy and gene expression in the hemichordate nerve cords.

In spite of the evidence supporting the idea that the ancestral bilaterian had a CNS, there is an alternative theory – namely that the ancestral bilaterian and the ancestral deuterostome had ectodermal nerve nets, from which the chordate CNS evolved. The dorsal and ventral nerve cords of hemichordates are therefore not only a hemichordate invention, but are unrelated to the chordate CNS (reviewed in [34]). This theory, most recently articulated by Lowe and colleagues [9,24,113], is based on developmental gene expression and gene interactions in the direct-developing hemichordate, S. kowalevskii. It proposes that the chordate CNS evolved from the ectodermal nerve net of a hemichordate-like ancestral deuterostome and maintains that the hemichordate nerve net contains signaling centers evolutionarily related to the anterior neural ridge (ANR), the ZLI and the isthmic organizer (ISO) in vertebrates. As a corollary, part or all of these signaling centers have been lost in the invertebrate chordates (amphioxus and tunicates) [9]. This idea deserves careful consideration because it not only argues that the considerable similarities of gene expression in protostome and chordate nerve cords represent convergent evolution, but it assigns a key position to hemichordates in evolution of the vertebrate CNS.