Hemichordates are perhaps the most promising of the non-chordate deuterostomes groups for addressing issues of both the early deuterostome evolution and the evolution of the chordate body plan (Cameron et al. 2000; Tagawa et al. 2001; Lowe et al. 2003). The phylum is divided into two classes: the enteropneust worms and the pterobranchs. Both groups possess a similar tripartite body organization, but are characterized by distinct feeding mechanisms. The pterobranchs use a lophophore, a ciliated extension from the mesosome, to filter feed (Halanych 1995), whereas enteropneusts use their highly muscular and ciliated proboscis (prosome) for direct particle ingestion and filter feeding (Cameron 2002). I will focus almost exclusively on the body plan of enteropneusts, as there are currently no published molecular data for pterobranchs. The most recent molecular phylogenies describe two main groups of enteropneusts: the Harrimaniidae on one lineage and the Ptychoderidae and Spengelidae on the other (Cameron et al. 2000). These two lineages have major life-history differences: harrimaniids are all direct developers, whereas the spengelids and ptychoderids are indirect developers with feeding larvae, which often spend many months in the plankton before metamorphosing into juveniles (Cameron et al. 2000). Phylogenetic relationships of the various hemichordate groups remain poorly resolved, and this area is in need of further research. Pterobranchs have classically been considered as basally branching hemichordates, based largely on the proposed homology of its lophophore with that in other lophophorate groups. However, reclassification of the lophophorates as protostomes reveals that the structural similarities of lophophores are due to convergence rather than homology (Halanych et al. 1995). Further molecular phylogenetic studies have proposed that pterobranchs are perhaps nested within the enteropneusts (Cameron et al. 2000; Winchell et al. 2002), but this is weakly supported by current datasets. Clearly, this issue should be revisited with broader phylogenetic sampling.

Although it is now widely accepted that many of the developmental regulatory cascades controlling anteroposterior and dorsoventral axial patterning are probably homologous as regulatory modules (Gerhart & Kirschner 1997; Carroll 2005; Davidson 2006), the extent to which the comparative expression and functional studies involving regulatory genes are effective for testing hypotheses of morphological homology remains controversial (Arendt & Nübler-Jung 1996; De Robertis & Sasai 1996; Lowe et al. 2003, 2006; Lichtneckert & Reichert 2005; Denes et al. 2007). Most studies have focused on similarities in nervous system patterning along both dorsoventral and anteroposterior axes (Acampora et al. 2001; Reichert & Simeone 2001; Hirth et al. 2003; Lichtneckert & Reichert 2005). Classical morphological comparisons have generally converged on the hypothesis that the central nervous systems of arthropods and chordates evolved independently and that early bilaterian nervous systems were generally quite simple (Holland 2003). More recent interpretations of the molecular genetic data based on model systems with the central nervous systems lead to quite different conclusions and propose a protostome–deuterostome ancestor with a complex, centralized nervous system with a regionalized brain. The homologous suites of genes involved in the patterning of the central nervous systems of model systems are very similarly deployed spatially during development (Arendt & Nübler-Jung 1994, 1996; Finkelstein & Boncinelli 1994; Sharman & Brand 1998; Hirth et al. 2003; Lichtneckert & Reichert 2005). Along the anteroposterior axis, the hox genes are involved in patterning the nerve cords of both arthropods and chordates. The boundary of the trunk and the rest of the anterior nervous system are marked by the homeobox gene gbx or unplugged (Hirth et al. 2003; Castro et al. 2006). Gbx is expressed at the boundary between hox genes and otx and marks a morphological transition in the organization of the nervous system. Other anteriorly localized homeobox genes such as orthodenticle (otx), pax6, distalless (dlx), emx and retinal homeobox (rx) play conserved roles in brain patterning and exhibit similar relative spatial localization during the central nervous system development (Lowe et al. 2003). In the dorsoventral axis, early patterning events of the ectoderm are similarly conservative between arthropods, vertebrates and annelids (Arendt & Nübler-Jung 1994, 1996; De Robertis & Sasai 1996; Holley & Ferguson 1997; Cornell & Ohlen 2000; Denes et al. 2007). The secreted factor chordin/short gastrulation is released dorsally in vertebrates and ventrally in arthropods, protecting the ectoderm from the neural-inhibiting effects of the TgfB ligand Bmp, which is expressed ventrally in vertebrates and dorsally in flies. The interaction of these secreted ligands results in the formation of the central nervous system on the dorsal side in vertebrates and the ventral side in arthropods and annelids. These data have revived the venerable dorsoventral axis inversion hypothesis by Dohrn (1875), which proposed that the dorsoventral organization of chordates is best explained by a complete body axis inversion in the lineage leading to chordates. Further similarities have been revealed in later dorsoventral pattering of the neurectoderm of all three groups, but most closely between the annelid Platynereis dummerilii and vertebrates (Denes et al. 2007). The similarities in relative expression domains and essential functions along both the dorsoventral and anteroposterior axes during central nervous system patterning have led to the proposal that a tripartite brain is ancestral for bilaterians, implying that the protostome/deuterostome ancestor and early deuterostomes were characterized by a complex, centralized nervous system. Within a purely phylogenetic framework, based on current molecular phylogenies, the outgroups to the bilaterians are both acoel flatworms and cnidarians, both of which are characterized by a nerve net (Holland 2003). Within the bilaterians, the basiepithelial nerve net is quite common. Proposing a protostome/deuterostome ancestor with a complex central nervous system and brain implies that this organization has been lost multiple times in evolution of the bilaterian phyla.

Until recently, the vast majority of data generated in the developmental biology have been from terrestrial model systems that are characterized by central nervous systems. Recently, this bias has been redressed by a broad description of the genetic information involved in patterning the hemichordate enteropneust body plan and its diffuse basiepithelial nerve net (Lowe et al. 2003, 2006). The organization of the nerve net is based around a broad distribution of cell bodies throughout the ectoderm. Despite the general diffuse organization of the nervous system, there is a significant dorsoventral and anteroposterior polarity in its structure and organization, particularly in the dorsoventral dimension (Bullock 1945; Knight-Jones 1952). A mat of axons spreads out along the basement membrane, which is thickened in certain areas of the ectoderm, at the base of the proboscis, along the anterodorsal region of the body in the mesosome and in both the dorsal and ventral midlines of the metasome. In the proboscis ectoderm, there is a dense concentration of nerve cells that have been proposed to be primarily sensory (Bullock 1945; Knight-Jones 1952) and is particularly thick at the base of the proboscis (Brambell & Cole 1939).

Probably the most well-known aspect of hemichordate anatomy is the mid-dorsal region of the dorsal cord, or collar cord which is internalized into a hollow tube of epithelium in some species within the Ptychoderidae and in one species of the Spengelidae. However, in the harrimaniids, there is no contiguous hollow tube, but scattered blind lacunae (Bullock & Horridge 1965; Nieuwenhuys 2002; Ruppert 2005). This structure has been widely compared to the dorsal cord of chordates due not only to the superficial similarities of the hollow nerve cord but also to the collar cord forms, in some species, by a process that quite closely resembles chordate neurulation (Morgan 1891). However, the similarities have generally been over emphasized as it seems to be more of a conducting tract rather than processing centre (Ruppert 2005) as evidenced by both ultrastructural (Dilly et al. 1970) and physiological data (Pickens 1970; Cameron & Mackie 1996). Another striking feature of the dorsal cord is the presence of giant axons in some species. The cell bodies project their axons across the midline and continue posteriorly within the collar cord. It is not known where the axons finally project; Bullock (1945) proposed that they innervate the ventrolateral muscles of the trunk and suspected that their primary function is to elicit a rapid contraction of these muscles. However, several groups do not possess giant axons and yet are still able to elicit a rapid retreat, so the role of the giant axons remains uncertain (Pickens 1973).

The last group of genes includes gbx and hox genes. In vertebrates, the regulatory interaction between otx and gbx is involved in positioning the isthmus along the A/P axis, with gbx expressed posterior to otx. In amphioxus, gbx is also expressed in a mutually exclusive domain to otx in the central nervous system, suggesting a conserved interaction between the two genes (Castro et al. 2006), but is absent from the genome in ascidians. In S. kowalevskii, we observe a departure from chordates in that otx and gbx expression overlap extensively at all stages of development examined, suggesting that they do not share the same mutual antagonism as found in vertebrates. Posterior to gbx, hox genes are expressed in closely abutting domains, but roughly in order down the rest of the trunk in ectodermal rings. This work was further developed in a study of hox genes (Aronowicz & Lowe 2006) which cloned a further 6 hox genes to make a complement of 11 hox genes. There is probably a true orthologue of hox8 still to be cloned from the genome, but it is not clear whether further posterior class member duplications are present or the current complement, shared with sea urchins, represents the ancestral complement of deuterostomes, or just the Ambulacraria. There are currently no data on hemichordate cluster organization. This information will be critical to determine the full hox complement, and the hox gene order along the chromosome, to check for collinearity with gene expression as has been described in other groups (Aronowicz & Lowe 2006). Hox expression domains follow predictable nested domains with the most anterior hox genes expressed in the most anterior regions of the ectoderm, and the more posterior members in the most posterior domains. At the stages that were examined, expression of many of the genes was tightly grouped, with little evidence of difference in the anterior limits. Expression has not been examined for all genes at late developmental stages when the trunk begins to elongate and become further regionalized. Perhaps the anterior expression limits of hox genes become more markedly differentiated in later stages. Posterior hox family members were examined at later stages when the ventral post-anal tail was developing, and the expression of these posterior members was restricted to the post-anal tail in these juveniles, which is similar to the expression of their orthologues in the dorsal post-anal tail of vertebrates. These data would support the proposed homology of the chordate and enteropneust post-anal tails, although it is certainly also possible that independently evolved posterior extensions are likely to express posterior hox genes already expressed in the posterior ectoderm.

A summary of the data from the development of the A/P axis of the hemichordates is diagrammed in figure 2. These data clearly demonstrate similar relative expression of transcription factors with critical roles in anteroposterior patterning between chordates and hemichordates. Although most comparative studies and speculations on the nervous system of hemichordates have focused on the dorsal and ventral axon tracts as potential homologues of chordate central neural structures, the results from this study suggest, as was proposed by Bullock in 1945, that the appropriate comparison is with the entire net. The cords are probably local thickenings of the nerve plexus rather than integrative centres (Dilly et al. 1970). The conclusions one can draw from these data are more complicated, particularly, whether these data help to reconstruct early morphological evolution of deuterostomes. First, it is most parsimonious to conclude that the majority of the similarities in the expression of multiple genes, with such close topological similarities, is conservation of a transcriptional regulatory network between vertebrates and hemichordates, and at that level, this is a homologous feature. It is highly unlikely that all the similarities of gene expression along the A/P axes of both groups are a result of co-option of individual genes into convergently similar domains. However, the organizational difference in the nervous systems of both groups suggests that, despite this regulatory conservation, the evolutionary possibilities of the downstream morphologies have not been constrained. The nervous system, in particular, demonstrates this point effectively: the development of both the central nervous system of chordates and the basiepithelial nerve net of hemichordates is probably regulated by this conserved regulatory map (although it is important to note that this was not directly tested in Lowe et al. 2003). Clearly, the forebrain of vertebrates and the proboscis of hemichordates are not homologous structures. So this suite of genes is not a reliable marker of morphological homology between groups. By considering these data alone, we can speculate that the deuterostome ancestor, and also the protostome/deuterostome ancestor, may have been characterized by a completely diffuse or fully centralized nervous system, and all possible intermediates. Reconstructing ancestral morphologies from gene expression data can be problematic, even with such large expression datasets. These data give a unique insight into the A/P patterning of the deuterostome ancestor, revealing a degree of transcriptional complexity previously attributed to the complex nervous system of vertebrates.

The TGFB-signalling ligand, Bmp, and one of its antagonists, chordin, are involved in establishing the dorsoventral developmental axis in arthropods and chordates. This molecular axis has recently been investigated in S. kowalevskii (Lowe et al. 2006). Hemichordates occupy a key position for investigating the evolution of this developmental pathway in dorsoventral patterning of the bilaterians (Nübler-Jung & Arendt 1996; Lowe et al. 2006). The most striking feature of Bmp and chordin between vertebrates and arthropods is that their relative expression is inverted dorsoventrally with respect to each other (Arendt & Nübler-Jung 1994, 1996; De Robertis & Sasai 1996). In hemichordates, bmp2/4 is expressed along the dorsal midline throughout all stages of development, along with all the members of the Bmp synexpression group (Niehrs & Pollet 1999; Karaulanov et al. 2004). At early developmental stages, chordin is expressed ventrally and very broadly, on the opposite side to bmp2/4, almost up to the dorsal midline, but is increasingly restricted to the ventral side as developmental progresses. There are many genes that exhibit marked dorsoventrally restricted expression domains along either dorsal or ventral midlines in ectoderm (tbx2/3, dlx, olig, netrin, pitx, poxN, lim3, admp, sim), endoderm (mnx, admp, sim, nk2.3/2.5) and mesoderm (mox/gax). These data reveal a molecular dorsoventral asymmetry that perhaps underlies the morphological asymmetry along this axis. Although in hemichordates the expression of chordin and bmp in relation to the dorsoventral axis is similar to protostomes, the early developmental action of Bmp and chordin does not result in segregation of a central nervous system from the general ectoderm: there is no central nervous system, but a diffuse and broadly distributed nerve net. What is the early role of Bmp in an animal without a non-neural ectoderm? Over expression and knockdown analyses addressed this question, and two main conclusions were presented: first, over expression of Bmp did not result in the repression of neural fates; second, bmp plays a central and critical role in dorsoventral patterning (Lowe et al. 2006). In embryos incubated with recombinant vertebrate Bmp4 protein, endogenous hemichordate bmp2/4 expression was activated throughout the ectoderm, rather than localized along the dorsal midline in normal embryos. These treated embryos do not perforate a mouth, and with high levels of Bmp protein do not perforate gill slits. Additionally, in the endoderm, dorsolateral endodermal pouches, precursors to the gill slits, do not form, and the entire endoderm projects into the protocoel rather than a thin dorsal projection that would normally develop into the stomochord. Knockdown or diminished expression of bmp2/4 by injection of short-interfering RNAs (siRNA) resulted in a complementary phenotype, particularly in relation to the mouth, which normally perforates on the ventral side. In injected embryos, the mouth develops circumferentially, and eventually results in the detachment of the entire prosome. The morphological interpretation of Bmp modulation experiments suggests that over expression of Bmp dorsalizes embryos, and knockdown of Bmp ventralizes embryos. This was confirmed by further molecular analysis: markers of the dorsal midline, in both the ectoderm and endoderm, expanded into circumferential rings in Bmp ligand-treated embryos, suggesting that in normal embryos, they are activated by Bmp signalling on the dorsal midline. Some of the same dorsal markers failed to activate expression following siRNA injection, adding further support for a role of Bmp in patterning dorsal cell fates. Further experimental evidence suggested that Bmp is involved in restricting the expression of ventrally expressed genes to the ventral midline, as Bmp ligand-treated embryos failed to express ventral markers, and these same markers expand to the dorsal side in siRNA-injected embryos.