There was a long tradition for ranging types of symmetry in graded series thought to reflect increasing complexity. High degrees of symmetry (i.e. high numbers of symmetry planes or axes) were regarded as reflecting primitiveness, because “the most highly symmetrical systems are also the most random” [137], as illustrated in the non-living world by self-organising systems such as crystals (see [11] for physical aspects and [174] for the historical influence of crystallography on 19^th Century biologists and their considerations about symmetry). Evolution was expected to increase heterogeneity, thus decreasing the number of identical parts. This is well exemplified by Beklemishev's views, [15], on the evolution of protozoans (but according to this author the same principles apply to metazoans). Above protozoans showing asymmetry of indefinition (e.g. amoebae), the most primitive grade was represented by perfectly spherical taxa, evolving towards reduction of the number of symmetry axes while keeping a general spherical shape. The following grade comprised cylindrical organisms, evolving from non-polarised (“homopolar”) to bipolarised (“heteropolar”) and from a large number of planes of symmetry towards a reduction of this number. Bilateral (or even dissymetric) protozoans were considered as the most evolved forms. The conventional scenario for metazoans outlined previously has clear roots in the same philosophy, even if the series is shorter because symmetry types are much less diverse in multicellular animals than in unicellular eukaryotes.

An alternative approach to the contribution of symmetry to morphological complexity is to consider for each of the symmetry types the minimal number of degrees of freedom that constrain the variation of morphology in the three-dimensional space. Here we need to take into consideration not only the body (or general) polarity axes, but also those internal polarity axes that are inherent to each symmetry type (dashed lines in Fig. 1). The sum of internal and body polarity axes (last column in Fig. 1) constitute a full system of coordinates, meaning that morphology at any given point within the organism is unambiguously determined, once these coordinates are fixed. In flat asymmetrical animals, as in spherically symmetrical ones, morphology is determined by a single coordinate (distance with respect to the substrate/to the centre). In cylindrical symmetry, morphology varies along the oral–aboral polarity axis and along the radial internal axis (e.g. succession, from the axis to the periphery of hydrozoan polyps, of the gastral cavity, endoderm, mesoglea and ectoderm, Fig. 4B). In radial symmetry, the same two coordinates are required, but in addition morphology varies in an angular fashion. Thus, there is a third (angular) coordinate (Figs. 1, 5). Although this is not an “axis” in the proper sense, it might be called the internal “angular axis” by analogy. To describe a bilaterally symmetrical animal, three systems of spatial coordinates are needed: position along the AP and DV body polarity axes, and position along the medio-lateral internal axis (Fig. 1) (left-right disruption of the bilaterality corresponds to the addition of a fourth coordinate).

Thus, according to the number of spatial directions in which morphology varies, symmetry types can be classified as “monodirectional” (asymmetrical and spherical), “bidirectional” (cylindrical) and “tridirectional” (radial and bilateral), and this strictly reflects the contribution of symmetry to morphological complexity. In addition, this way of characterising symmetry type complexity is directly connected to developmental issues (i.e. how morphological differentiation between cells that are located in different areas is generated during development). Putting aside the case of asymmetry, there is a clear albeit imperfect inverse correlation between the number of symmetry elements (axes or planes) and the number of polarity axes (general or general + internal) (compare the last four columns in Fig. 1), in consistence with the common statement that polarity is breaking of symmetry. Interestingly enough, cylindrical and radial symmetries differ in number of coordinates, implying that they indeed correspond to two different levels of morphological organisation. In addition (and maybe counter-intuitively), bilateral symmetry does not appear intrinsically more complex than radial symmetry.

These considerations naturally lead us to ask if more complex types of symmetry necessarily require more complex developmental mechanisms for their establishment during ontogeny. Unfortunately, comparisons of developmental mechanisms across symmetry types are hampered by a critical lack of empirical knowledge, particularly concerning radial symmetry. Nevertheless, a few interesting ideas are emerging, based on our current understanding of the general features of polarity-generating mechanisms in animal development, and on recent advances in mathematical modelling.

In metazoans, the global symmetry of the body plan arises early during ontogeny within a relatively short time frame [37], contrasting with symmetrical patterns emerging through the progressive addition of similar body parts, under some specific spatial constraints, in some other multicellular eukaryotes (e.g. Fibonacci helicoidal symmetries in various vegetal structures [4,47,117]). Exceptions to this rule mostly concern geometrical patterns in animal colonies [16,63] and will not be discussed here. Animal body plan symmetry is the final outcome of a (rapidly progressing) succession of polarising (symmetry breaking) events occurring in the embryo, within single cells or within nested fields of cells. Thus, body plan symmetry by itself has no meaning in terms of developmental cellular/molecular processes, but is a by-product of a robust and invariant series of polarising mechanisms, among which the earliest defines the major body axes only because they are the earliest.

The molecular mechanisms underlying the establishment of polarity within a cell field are remarkably similar throughout the Metazoa. Leaving apart the probably derived instances of highly mosaic developments (where early cell fate choices and thus axis specification requires little cell-cell interactions; e.g. in ctenophores [123] or in Drosophila [80]), the fundamental polarity-generating mechanism involves an inductive signal mediated by one or several (often antagonising) diffusing molecules (morphogenes), which through a signalling pathway provide instructive clues for genes to be expressed differently in cells located in different spatial areas. There are a limited number of major signalling pathways, among which, for example, the well-studied Wnt, TGF-β, Hedgehog (Hh) and Notch/Delta pathways (designated by using the names of their ligands) [80]. During development, these signalling pathways are not only acting in the generation of early embryonic polarities, but are pleiotropically used and re-used in a large number of contexts and thus may determine polarities at every spatial scales and at any life stage.

Mathematical modelling have demonstrated that there is no simple relationship between the number of involved signalling molecules and the complexity of the resulting geometrical pattern. For example, in reaction–diffusion coupling models, more or less complex patterns are formed in epithelia through the dynamic interaction of only two diffusible factors, an activator and an inhibitor that is induced by the latter and diffuses faster [134,187]. Cummings [39–41] proposed a mathematical model coupling the dynamic distribution of two antagonising morphogenes and their receptors with shape changes in epithelial cell sheets. Under some parameters, he obtained a clear radially symmetrical pattern on a three-dimensional figure topologically equivalent to a gastrula (he called this figure a “Proto-Cnidaria”) (Fig. 7 in [41]). Astonishingly, what I described previously as a tridirectional type of symmetry can be developmentally reached through epithelial morphogenesis involving a single couple of antagonising diffusing molecules. This result has far-reaching implications, because exactly the same molecular device can under other constraints and parameters lead to monodirectional or bidirectional geometrical patterns, thus totally uncoupling morphological complexity from the complexity of underlying molecular developmental tools. Likewise, Cummings [41] claimed to have obtained bilateral symmetry (“Proto-protostome”) under still other parameters with the same model, but unfortunately what he called a bilateral pattern was in fact biradial (his Fig. 8), thus leaving unanswered whether or not bilaterality requires in theory more “developmental complexity” than radial symmetry does. Although at first sight it could seem evident that more signalling pathways are needed to create an additional polarity axis, this is not necessarily true since a single patterning system can work sequentially along separate axes (e.g. in vertebrates, Wnt signalling acts prior to gastrulation to define the dorsal organiser, and during gastrulation to posteriorise along the developing AP axis [133,134]).

Recent comparative genomic analyses have revealed that virtually all families of signalling ligands, receptors and transductors were already established in the common metazoan ancestor. Cnidarian genomes have a high level of genetic complexity and contain most of the molecular developmental toolkit that bilaterians use to build their body plans [135,136,163,184]. Not only are the major gene families present, but the diversity of their members is often of the same order of magnitude than in bilaterian genomes (e.g. within the Wnt family [109]). Likewise, components of the Wnt, TGF-β, EGF, IGF, RTK, Notch/Delta, Hh and Jak/Stat signalling pathways have been identified in sponges [1–3,149]. Thus, there is no relationship between genetic and morphological complexity [12,109,184], and in particular, the common ancestor of Metazoa “already” possessed many more molecular signalling systems than required to achieve even the most “complex” types of symmetry.

The inescapable conclusion is that at no stage during animal evolution were transitions from one type of symmetry to another limited by molecular possibilities. Clearly, such transitions (regardless of their direction) did not involve genetic innovation, but simply a different deployment of existing molecules during ontogeny. Based on these considerations, I suggest that the frequency of changes in symmetry types during animal evolution was mostly dictated by physical factors [148] (the types of symmetry representing various solutions in the “game of possible” of animal geometry), structural constraints (the earliest developmental events that set up the body plan being more or less “locked in” by the dependence of later patterning events upon them) and by natural selection exploring the wide range of adaptive possibilities offered by the various types of symmetry, under different environmental circumstances or life styles.

Reconstructing evolutionary transitions between the various types of symmetry requires rigorous assessment of primary homologies through comparative anatomy and their interpretation under a phylogenetic framework. The phylogenetic distribution of adult symmetry types onto the animal tree is presented on Fig. 6A (probable ancestral character state for each terminal taxon indicated on tip of the branches; see legend for justification). I use here as a working framework a consensus tree derived from rRNA molecular phylogenies [19,32,82,107,120,130,158,190], although in fact basal metazoan phylogeny remains uncertain, since various data sets have provide inconclusive or conflicting results [49,64,168]. I have decided to leave apart the placozoan Trichoplax for its position is totally unresolved (see for example [48]) and its type of symmetry is ambiguous (asymmetrical or cylindrical; see above).

The evolution of symmetry types cannot be directly inferred from parsimony-based character optimisation because of the lack of phylogenetic resolution combined with high number of character states and high level of homoplasy. I will nevertheless try to propose an evolutionary scenario (depicted in Fig. 6A), but the reasoning will be indirect and rather speculative. Instead of directly optimising the types of symmetry, it is simpler to start by considering the axial properties of the adult body, formalised into a three-state character with state 0 corresponding to asymmetry, state 1 to the presence of a single axis of symmetry and body polarity (monaxonic heteropolar organisation), and state 2 to the presence of two body polarity axes (bilaterality) (Fig. 6B). Here, parsimony optimisation favours state 1 on the common node of Metazoa (with a minimum of 3 steps, against 4 steps for states 0 or 2 on the same node). Thus, with reservations due to uncertainty on the phylogeny and to the weak deviation in step numbers between competing hypotheses, this simple reasoning indicates that the last common ancestor of Metazoa was probably symmetrical around a single polarity axis, rather than asymmetrical as postulated in the “conventional scenario”.

Accordingly, the adult asymmetrical organisation observed in most demosponges is probably derived rather than ancestral. Demosponges are just one out of four major extant sponge lineages [20], and the other three (hexactinellids, calcisponges and homoscleromorphs) have prevalent adult cylindrical or radial symmetry (Figs. 4A, 5A–D). Some calcisponges also are asymmetrical (e.g. Leuconia), but this is clearly a derived situation within the Calcispongia, with respect to a plesiomorphic monaxonic heteropolar adult organisation [118,120,121]. Of particular significance is the radial symmetry of homoscleromorph sponges (Fig. 5C–D) [67,68,93,176], a group recently separated from demosponges on the basis on molecular, histological and embryological characters [20,68,180]. The extinct sponge lineage Archaeocyatha (of uncertain affinity with respect to extant sponge taxa) is also characterised by a clear radial architecture. Thus, the common idea that sponges, supposedly the “most primitive” animals, would be devoid of any defined adult body symmetry and/or polarity axis [6,27,70,71,94,97,157,169,173,193] (concerning Ref. [27], see their matrix, pp. 894–895) must be vigourously refuted, as it is entirely attributable to taxonomic bias (the demosponges representing 95% of living sponges species) and typological reasoning (demosponge morphology having been taken as “typical” for sponges).

If the common metazoan ancestor was polarised along a single symmetry axis (state 1, Fig. 6B), that leaves two possibilities for the ancestral type of symmetry: cylindrical or radial. The crucial question is now whether the radially symmetrical anatomies (with repeated structures around the axis) occurring in various basal metazoan lineages are homologous or not. I argue below from comparative anatomy that they are clearly not. The proposed hypothesis of multiple convergences implies that each kind of radial symmetry actually represents a distinct character state, explaining why I have used different colours on Fig. 6A.

The multiradiate symmetries of hexactinellids, syconoid calcisponges and homoscleromorphs are certainly convergent. Molecular phylogeny suggested that the ancestral aquiferous system of calcisponges was asconoid (cylindrical symmetry), a derived syconoid aquiferous system (radial symmetry) being acquired several times independently within the clade [118,120,121] (but see criticism in Dohrmann et al. [55]). Hence, homology with the syconoid-like organisation observed in homoscleromorphs is unlikely. The multiradiate organisation of most hexactinellids entirely depends on the architecture of their skeleton (Fig. 5A), which itself results from the particular geometry and arrangement of their unique hexactine siliceous spicules [5,189]. These arguments also imply that instances of multiradiality in sponges are absolutely not homologous with radial anatomies occurring in ctenophores or cnidarians. Consistently, the strict determination and low value of symmetry plane number in ctenophores and cnidarians (with the exception of hexacorallian anthozoans) departs fundamentally from the high and indefinite number which characterises multiradial symmetry in sponges. The only reported instance of tetraradial symmetry in sponges, the regular disposition of four particular cells (“cellules en croix”) in the flagellated anterior pole of the calcisponge amphiblastula larva [59,188], has clearly nothing to do with the tetraradial symmetries of ctenophores or cnidarians which involve multicellular structures.

Convergence of the radial symmetry between ctenophores and cnidarians is equally likely, because their repeated structures are clearly not homologous. The symmetry of ctenophores (Fig. 5E) is in fact a unique combination of octoradial (eight comb rows, eight meridional canals), tetraradial (four balancers – mechanoreceptor structures supporting the statolith, within the apical sensory organ – four pairs of comb rows, four interradial canals) and prevalent biradial symmetry (e.g. one tentacle pair, one pair of polar fields, a flattened pharynx), the latter being disrupted by the oblique position of the two anal pores. By contrast, in the polyp, the probable ancestral body form of the Cnidaria [8,22,33,34,122], the only repeated structures are the tentacles (arranged in a crown around the mouth) and the mesenteries (or septae), endodermal infoldings of the gastral cavity wall. The latter occur in anthozoan (Fig. 7) and scyphozoan polyps, and are supposed to be secondarily lost in the simplified polyps of Hydrozoa and Cubozoa [122]. Thus, ctenophore radially repeated structures do not exist in the cnidarian body plan, and reciprocally, with the exception of tentacles. However, tentacles are clearly non-homologous between these two phyla, as indicated by their totally different position in the body plan, and by their differing structures and ontogenies [90].

In strict parsimony terms, “state 1” (monaxonic heteropolar organisation) present in the common metazoan ancestor could thus correspond to any of the various kinds of radial symmetry observed in non-bilaterian lineages, or to cylindrical symmetry. However, that the latter is the ancestral state is strongly supported by ontogeny, since the cylindrically symmetrical stage through which all metazoans (including the bilaterians) pass during their ontogeny is easily interpreted as an instance of recapitulation. This idea is extremely classical, since in the haeckelian theory of metazoan origin [88], the Gastraea was postulated to be the cylindrically symmetrical ancestor of all metazoans (following a spherical stage called the Blastaea). The primitiveness among adult metazoan body plans of a monaxonic (rather than asymmetrical) structure has always retained a minority of supporters (e.g. Beklemishev [15]). Admitting cylindrical symmetry as ancestral, it can be noticed that the only extant metazoans retaining this organisation as adults are the asconoid calcisponges (e.g. genera Clathrina, Soleneiscus, Leucosolenia…), and as such, these sponges should receive special attention in future comparative studies of polarity-generating molecular mechanisms. By contrast, the partial cylindrical symmetry of polyps in hydrozoan and cubozoan cnidarians (at the body column level) probably results from a reversion (i.e. loss of the mesenteries [122]).

Bilaterality is clearly a derived type of symmetry since it occurs only in the closely related Cnidaria and Bilateria. We thus need to compare the nature of bilaterality in these two clades to evaluate if it could have been inherited from their common ancestor, as believed by a few morphologists [86,99,193]. I will discuss recent arguments from developmental gene data later on in the following section. In-depth consideration of the anatomical, biological and phylogenetic significance of bilaterality among anthozoan cnidarians is particularly crucial, because failure to do so in the recent evo-devo literature has seriously hampered the interpretation of molecular data. Exceptional instances of bilaterality in hydrozoans (see Section 4) will not be discussed here, because they are clearly derived (several times independently) from within this class and thereby convergent with the bilaterality of the Bilateria.

Although the bilateral symmetry of anthozoans has been widely popularised through recent molecular studies on a sea anemone, bilaterality is considerably more evident in the polyp anatomy of the anthozoan order Octocorallia (Fig. 7A) [98]. In an octocoral polyp, two types of anatomical structures are affected by bilateral symmetry: the pharynx (an ectodermal invagination leading to the gastral cavity) and the mesenteries (Fig. 7A). The flattened pharynx is bilaterally symmetrical, due to the presence of a single ciliated groove or siphonoglyph (Fig. 7A), driving a water current into the gastral cavity. The eight mesenteries are globally arranged in an octoradial pattern, but they differ with each other in their constitutive elements, and identical mesenteries are bilaterally opposed. In particular, each mesentery bears on one of its faces a retractor muscle, of mesodermal origin [179], and the relative distribution of the eight retractor muscles is bilaterally symmetrical (Fig. 7A). In addition, in transverse section below the pharynx (not shown), the two mesenteries located on the side opposite to the siphonoglyph differ from the other in lacking gonads and in having ciliated ectodermal free edges (mesenterial filaments) serving to create an upward water current, instead of the cnidoglandular endodermal mesenterial filaments (whose function is digestive) associated with the six remaining mesenteries [98]. The octocorallian polyp is thus truely bilateral. Its second polarity axis, orthogonal to the main oral–aboral axis, is called the directive axis.

By contrast, in most adult sea anemones (Actiniaria), the disposition of mesenteries is absolutely radial (6n) with no trace of bilateral symmetry (Fig. 7B) and because there are two siphonoglyphs, also the pharynx is biradial. Hence, these sea anemones (e.g. Anemonia, Sagartia, Metridium…) are not at all bilateral. The same holds true for corals (Scleractinia and the akin “naked corals” Corallimorpharia), in which siphonoglyphs are absent. In these anthozoans, bilateral symmetry is only transiently apparent during development, in the disposition of the eight retractor muscles at the “Edwardsia” larval stage (Fig. 7C), before the production of additional mesenteries. However, several sea anemone lineages have diverged ecologically by adopting a burrowing life style in sediments, and correlatively their ontogeny is characterised by paedomorphosis, i.e. as adults they look like Edwardsia larvae [43]. The laboratory model Nematostella vectensis (family Edwardsiidae), whose complete genome has been sequenced, is such a highly derived sea anemone, explaining why throughout its life it retains eight mesenteries (and the concomitant bilaterality of retractor muscles), and a single siphonoglyph (hence bilaterality of the pharynx). Through paedomorphosis, the Edwardsiidae have reverted to more plesiomorphic character states found in other anthozoans such as octocorallians.

The phylogeny of Anthozoa is still largely unresolved, but recent studies based on various kinds of data sets tend to resolve the Octocorallia as the sister-group of other anthozoans (Hexacorallia), the latter comprising the Ceriantharia and a clade grouping all other taxa (Antipatharia, Zoanthidea, Actiniaria, Madreporaria) (France et al. [76] with mitochondrial 16S, Berntson et al. [18] with 18S rRNA, Won et al. [194] with morphology, Daly et al. [44] with combined morphology, 18S, 28S and 16S; reviewed in Daly et al. [45]). Since Octocorallia, Ceriantharia and Zoanthidea have a single siphonoglyph, a bilateral pharynx was probably present in the common ancestor of Anthozoa, while pharynx biradiality evolved secondarily through siphonoglyph loss in Madreporaria (= Scleractinia + Corallimorpharia) or through the addition of a second siphonoglyph in Antipatharia and Actiniaria. Furthermore, mesenteries exhibit some level of bilaterality at least during ontogeny in all hexacorallian lineages [15,17,98]. Thus, bilateral symmetry was clearly present in the last common ancestor of Anthozoa, or even (perhaps) in the common ancestor of the Cnidaria since a recent analysis of mitochondrial genomes obtained the paraphyly of anthozoans with respect to the other cnidarians (Medusozoa) [105], in conflict with rRNA phylogenies supporting their monophyly [18,33,76,105,151].

However, the bilateral symmetries of Anthozoa and Bilateria are in no way comparable. The bilaterality of anthozoans is clearly superimposed on a fundamentally radial anatomy, of which bilaterians show no trace. Whereas in the latter, the whole body plan is affected by bilaterality (general body shape, all germ-layers and virtually all organ systems), in anthozoans, only a restricted set of internal structures (pharynx and mesenteries) are bilateral, whereas the rest of the body (particularly the external morphology) is absolutely unaffected. Most significantly, all anthozoan bilateral structures are evolutionary acquisitions (synapomorphies) of the Anthozoa or Cnidaria, with no counterpart in bilaterian body plans. The mesenteries are a synapomorphy of the Cnidaria [122], while the mesentery retractor muscles and the ectodermal pharynx with a siphonoglyph are synapomorphies of the Anthozoa [122] (or of the Cnidaria, if anthozoans are paraphyletic [105]). Thus, the common ancestor of Cnidaria and Bilateria lacked all the structures that are bilaterally disposed in the Anthozoa. Also the biological significance of bilaterality is totally different between Bilateria and Anthozoa. In anthozoans, bilaterality has nothing to do with directional locomotion but is linked to the canalisation of water currents within internal cavities [15,98] and to the mode of mesentery production [17]. It is hard to imagine how this anthozoan-type internal bilaterality could have turned to become advantageous for unidirectional locomotion through exaptation in a bilaterian ancestor, as recently proposed [72]. In conclusion, there is compelling evidence for the convergent acquisition of bilaterality between the anthozoans and the bilaterians – actually the most common point of view among morphologists, including Hyman [98] and Beklemishev [15].

From the preceding overview of symmetry types across basal metazoan lineages, it is clear that the “conventional scenario” (a stepwise succession of asymmetry, radiality and bilaterality) is untenable. The most likely scenario places cylindrical symmetry as the ancestral state, while asymmetry, radial symmetry and bilateral symmetry (and exceptionnally, spherical symmetry) are derived types, each evolved several times independently, in the terminal branches, or within phyla (Fig. 6A). Multiple convergences as implied by the proposed scenario are coherent with my previous suggestion that important shifts in body symmetry did not necessitate any particular genetic or developmental innovation (thus, they could have occurred easily and repeatedly), and with the multiplicity of potential functionalities offered by each one of the symmetry types. A further notable implication of this scenario is that the multicellular ancestors of the Bilateria never passed through an asymmetrical or a radial type of symmetry.

In distantly related bilaterians, the generation of DV polarity involves a conserved protein network which comprises, among others, secreted molecules of the BMP family (e.g. decapentaplegic = dpp in Drosophila and its orthologue BMP4 in vertebrates) and their antagonist secreted ligands (e.g. short gastrulation = sog in Drosophila and its orthologue chordin in vertebrates) (Fig. 8A) [51,52,80]. This “BMP-chordin axis” functioned in the common ancestor of Bilateria, as supported by evidence not only from vertebrates and arthropods, but also from hemichordates, echinoderms and annelids [50,57,115]. Curiously, the polarity of DV signalling molecules is inverted in chordates with respect to other bilaterians (e.g. arthropods, but also annelids and hemichordates), suggesting that an exclusive ancestor of chordates turned upside down (the DV inversion hypothesis [7,51,115]).

In the planula larva of the sea anemone Nematostella vectensis, homologues of these bilaterian DV signalling molecules (e.g. dpp, noggin, chordin) and transcription factors (e.g. goosecoid) display restricted expression territories along the directive axis (Fig. 8B) [75,127,128,164]. The dpp homologue is likewise expressed asymmetrically in the coral Acropora [92]. According to Matus et al. [127], “the fact that NvChordin, NvNoggin1, NvGsc and NvDpp all are expressed asymmetrically along the directive axis at some point in the development of the juvenile body suggests that the cnidarian directive axis might be homologous to the bilaterian DV axis”. Thus, bilaterality would have evolved in a common ancestor of Cnidaria and Bilateria, and have been secondarily lost in non-bilateral cnidarians (the Medusozoa) [12,71,72,75,127,171]. This hypothesis is at odds with indications from comparative anatomy that bilaterality in these two lineages is certainly convergent (see preceding section).

However, these asymmetric gene expression patterns are not by themselves sufficient to support axis homology. A strict application of the principle of connexions to the gene expression patterns would expect orthologues of dorsal vs. ventral bilaterian molecules to be consistently expressed at one extremity and the other of the anthozoan directive axis, which does not happen to be the case (Fig. 8A, B). In chordates, dpp-BMP4 is mostly expressed ventrally, and the antagonists chordin and noggin are expressed on the dorsal side (Fig. 8A), while in the sea anemone, the three molecules are expressed on the siphonoglyph side of the directive axis [127,164], albeit in distinct germ layers (Fig. 8B). The transcription factor goosecoid, active in the dorsal organiser of chordates [80] (Fig. 8A), is expressed on both sides of the directive axis in the sea anemone (Fig. 8B). To these considerations it can be objected that signalling molecules operate at some distance from their transcription site, and that interacting diffusing factors (activators vs. inhibitors) may have differing diffusion capacities [134]. Thus, antagonist interaction between dpp and chordin proteins could be involved in the specification of the directive axis, even though their mRNAs locate on the same side. Only through in vivo functional experiments will the exact link between these molecules and directive axis polarity be understood. Injection experiments of the dpp and chordin cnidarian homologues in zebrafish [164,165] have demonstrated that the molecular antagonism is operating in the sea anemone, but these data do not indicate whether this mechanism is involved in generating the directive axis, or instead governs smaller scale polarities, such as for example the specification of germ layers on the siphonoglyph side [128].

Even if dpp signalling were actually specifying the sea anemone directive axis, this would not represent a strong case for homology with the bilaterian DV axis. Because morphologically the anthozoans are clearly bilateral, signalling molecules must operate to establish the directive vs. oral/aboral axis. The dpp system, present in the common cnidarian–bilaterian ancestor, could well have been adopted independently in these two lineages for creating a second axis (a phenomenon known as co-option). In conclusion, since available molecular data on cnidarian homologues of bilaterian DV genes are not compelling in favour of bilaterality being inherited from a cnidarian–bilaterian ancestor, convergence as supported by comparative anatomy clearly remains the most satisfying hypothesis.

The possibility remains that at least the cnidarian oral–aboral and the bilaterian AP axes would be homologous. Although admitted by most classical zoologists, this homology is not evident on anatomical grounds. The cnidarian oral region offers no resemblance with a bilaterian head (e.g. there is no brain), apart from the presence of the mouth, and cnidarians lack the anus which terminates the body of most bilaterians. Even if the axes were homologous, the cnidarian oral pole could correspond either to the anterior or to the posterior pole of the Bilateria. In the planula larva of the Cnidaria, the future aboral pole is generally located anteriorly during swimming. In some species this pole bears a sensory organ, which in the sea anemone expresses components of the FGF pathway, similarly to the anteriorly located apical organs associated with the brain of bilaterian planktonic larvae [166]. These observations favour the counter-intuitive hypothesis that the cnidarian counterpart of the bilaterian head would be the aboral and not the oral pole.

The alternative view, that the cnidarian oral pole corresponds to the bilaterian head, has been much more popularised in recent years, principally from expression data on the sea anemone homologues of the bilaterian Hox genes (in addition to earlier questionable arguments based on the expression of bilaterian “head specific” genes in the Hydra hypostome organiser [77,138]). Among the Bilateria, Hox genes have a well-known conserved role in AP patterning [80,156]. These homeobox genes are organised in a genomic cluster and in most investigated bilaterians, they are expressed along the AP axis in a sequential order which corresponds to their respective location within the cluster (a property called spatial colinearity). The common ancestor of Bilateria certainly used a “Hox code” to pattern its AP axis, although the system underwent many substantial modifications within the various bilaterian lineages [60]. Hox genes are not involved in the early establishment of the AP axis, a process involving various maternally localised determinants and signalling molecules (see later), but they act in controlling the correct positioning of structures along an already established axis.

The Hox gene family by itself is not an innovation of the Bilateria. Cnidarians have clear Hox homologues (these are called Anthox in anthozoan cnidarians and Cnox in hydrozoan cnidarians) [75,79,172]. Trichoplax lacks any Hox but possesses a ParaHox gene (called Trox2) [100,141] and since ParaHox is the supposed Hox sister cluster [78], the absence of Hox in Trichoplax could well be secondary. Hox and ParaHox genes are absent from the Amphimedon demosponge genome [110] and attempts to recover them from other demosponge and calcisponge taxa [119,167,178] have failed. There is, however, a debate in the literature to determine if the absence of Hox/ParaHox in sponges is ancestral [110] or derived [159]. Similarly, tentatives to identify Hox/ParaHox members in ctenophores have remained until now unsuccessful (but definitive proof of their absence is waiting for the availability of a complete genome). Thus, even if this requires confirmation, the Hox/ParaHox family could be a synapomorphy of (Trichoplax + Cnidaria + Bilateria).

Whereas the expression of Hox homologues in the Hydra polyp did not evoke a function in oral–aboral axis patterning [79], the expression of Anthox genes in the sea anemone Nematostella was claimed to support the fascinating conclusion that Hox-based patterning was inherited from the cnidarian–bilaterian ancestor and that the oral–aboral axis of Cnidaria was homologous to the AP axis of Bilateria, with the polyp oral extremity corresponding to the bilaterian head [75,172]. However, as for the DV gene data discussed earlier, these far-reaching evolutionary interpretations were elaborated on limited evidence. Of the 7 studied Anthox genes, only two showed restricted expression along the oral–aboral axis. The Anthox6 gene, an orthologue of the bilaterian anterior Hox gene Hox1/labial, was expressed at the oral pole. The Anthox1 gene, a putative orthologue of bilaterian posterior genes (Hox9-13/AbdB/Post), was localised at the opposite aboral extremity. All other investigated Hox homologues were expressed all along the body column (but they were localised along the directive axis) [75,172].

In conflict with these early interpretations of the Nematostella Anthox data, recently accumulated evidence strongly argues against a conserved role for Hox homologues in patterning the oral–aboral axis of Cnidaria in a way similar to the AP patterning Hox system in the Bilateria. First, orthologies between cnidarian genes and the various genes of the bilaterian Hox cluster (and of its supposed sister cluster ParaHox) are in general unclear. Among published Hox gene phylogenies (e.g. [31,79,104,172]), the only points of agreement are the existence of bona fide cnidarian orthologues of bilaterian anterior Hox genes, and the absence of cnidarian orthologues of bilaterian median genes. Other aspects are more controversial and there are conflicting views concerning the early history of the Hox/ParaHox genes (e.g. [31,69,73,78]). The second major problem is the absence of a genomic cluster grouping Hox homologues in the Nematostella genome. Anthox genes are dispersed among the genome (corresponding to an “atomised” cluster in the classification by Duboule [60]), apart from the probable ancient synteny represented by the association between eve (a “Hox like” homeobox gene) and the “anterior” Anthox genes (Anthox6, Anthox7, Anthox8a, Anthox8b). Although possibly derived, this absence of Hox clustering in Nematostella implies de facto an absence of spatial colinearity of gene expression.

Even more problematic is the strong variability of Hox gene expression along the oral–aboral axis among cnidarian lineages. Whereas the Hox1/labial “anterior Hox” homologue Anthox6 gene is orally expressed in Nematostella, its orthologue Cnox1 in the hydrozoan Podocoryne is expressed in opposite orientation [104,196]. Considerable variation among cnidarian taxa is also observed for the ParaHox gene Cnox2/Gsx [74]. Curiously, in cnidarians we are faced with the opposite situation to that observed in bilaterians: while among the later, Hox gene expression along the AP axis is conserved in spite of a tremendous disparity of body plans, their cnidarian orthologues have highly plastic expression territories along the oral/aboral axis, in animals that share the same basic body plan.

In conclusion, contrary to earlier and persisting claims [75,172], current evidence does not support a conserved role for cnidarian Hox homologues in patterning the oral/aboral axis, in a way similar to AP axis patterning in the Bilateria [104]. Although the situation observed for Hox genes in cnidarians could be derived, currently the most parsimonious interpretation is that colinearity along the AP axis was a bilaterian innovation [104], possibly linked to the novel requirements imposed by directional locomotion on the antero-posterior patterning of the central nervous system [53]. An additional (somewhat disappointing) conclusion is that Hox genes do not tell us which extremity of a cnidarian polyp is homologous to the bilaterian head, if such homology exists. Recent expression studies of cnidarian Otx and emx, two transcription factors involved in anterior patterning of the central nervous system in bilaterians, gave similarly unconclusive results with respect to cnidarian/bilaterian “head” homology [103,129].

The general trend is that homologues of transcriptions factors with conserved patterning functions along the polarity axes of the Bilateria do not show consistent localisations along the cnidarian oral–aboral axis. In addition, transcription factors belonging to both the DV and AP patterning systems of bilaterians were found expressed in regionally restricted patterns along the cnidarian oral/aboral axis [103], leading to the suggestion that the latter might be homologous to both polarity axes of the Bilateria. The AP and DV axes would thus have originated from a sort of duplication of a single ancestral axis, with subfunctionalisation of the corresponding signalling and patterning molecules between both novel axes [95,103,132,134,165].

Hox genes and other transcription factors are certainly not the best candidates if we wish to uncover deep correspondences between the polarity axes of basal metazoan lineages, since, as noted in a previous section, the earliest cell fate differences appearing in embryos are mediated by secreted ligands turning on signalling pathways, while transcription factors are acting downstream. The search for ancient polarity axis generating molecular mechanisms has recently focalised on the Wnt pathway [84,112].

It has become apparent in recent years that in the Bilateria, Wnt signalling plays a conserved and probably ancestral role in establishing and patterning the AP axis during early development [96]. In vertebrates, Wnt signalling has an early involvement in the formation of the dorsal organiser, and later acts to posteriorise during gastrulation [96,133,134]. Data from vertebrates, echinoderms and amphioxus supports the conservation among deuterostomes of a “posterior centre” secreting Wnt proteins, in addition to expressing a conserved cassette of regulatory molecules such as genes of the Notch-Delta pathway, and the transcription factors Brachyury and Caudal [36,96]. The conservation may extend to the Bilateria since various protostomes show posterior Wnt expression [177].

Recent studies on various cnidarians have highlighted the role of Wnt signalling in establishing and patterning the cnidarian oral–aboral axis. In the adult polyp of Hydra, axial polarity is constitutively under the control of the hypostome organiser, in the oral region [24]. The Wnt “canonical” pathway is active in this “head organiser”, as demonstrated by the localised expression of HyWnt and of another components of the pathway (HyTcf) in this region, by the higher level of nuclear β-catenin in the hypostome than in cells of the body column, and by the formation of ectopic oral structures following treatments with molecules (DAG, LiCl, alsterpaullone) that activate the canonical pathway [25,95,145]. Comparable experiments in another hydrozoan, Hydractinia, gave similar results [146,147,161].

Wnt signalling is also involved in the early establishment of the oral–aboral axis during embryogenesis in cnidarians. In the hydrozoan Clytia hemisphaerica, two (maternally inherited) oppositely localised Wnt receptors (Frizzled) govern oral–aboral polarity, as indicated by inhibition and over-expression experiments [139]. The Wnt canonical pathway is activated at the animal pole, the future blastopore and mouth region, as demonstrated by β-catenin detection assays. Experimental data likewise indicated activation of the Wnt canonical pathway at the oral pole in the other hydrozoan Hydractinia [161] and in the sea anemone Nematostella [113,192]. In addition, this region also expresses Brachyury [175], suggesting that the same cassette is acting at the cnidarian oral pole and in the bilaterian “posterior centre”. We are thus reverting to the hypothesis, consistent with the swimming direction of the planula, that the cnidarian oral extremity is homologous to the bilaterian posterior pole, in contradiction with suggestions based on Hox expression in the sea anemone [75,172] but in consistence with earlier proposals prompted by gene expression data in Hydra [131,133].

Not only are the Wnt mediating the early establishment of the cnidarian oral–aboral axis, but there is also some evidence that they are involved in later patterning along this axis. The sea anemone genome contains 12 different Wnt genes, belonging to 11 of the 12 families previously identified in bilaterian genomes, indicating that Wnt diversification predated the cnidarian–bilaterian divergence [84,109,112]. In the sea anemone, most of these Wnt genes are first expressed around the blastopore, and later on at the planula stage they become expressed along the oral–aboral axis, in overlapping domains, either in the ectoderm or in the endoderm, depending on the genes (Fig. 9) [109]. Wnt genes in Clytia embryos are also expressed in distinct nested territories along the developing oral–aboral axis, although there is no obvious correspondence between the territories of orthologous genes with Nematostella [140]. The global pattern of Wnt gene expression along the cnidarian oral–aboral is somewhat reminiscent of Hox gene expression in the Bilateria and has suggested the implication of a “Wnt code” in oral–aboral axis patterning in cnidarians [84,109,112], although functional data are required for explaining how this Wnt code could work.

The Nematostella data also indicates that, in addition to axial polarity, Wnt signalling is necessary for gastrulation and the specification of germ layers [109,113,192]. In fact, axial differentiation, gastrulation and germ-layer specification are concomitant and highly interdependent processes – meaning, in other words, that the specification of both spatial coordinate systems (axial and radial) previously defined for cylindrical symmetry (Fig. 1) is tightly coupled. In bilaterians, Wnt signalling determines not only axial differences in cell fate, but also gastrulation movements and germ-layer specification. For example, in vertebrates and echinoderms, the canonical Wnt pathway is required for the specification of the endomesoderm [36]. In chordates, the coordination of cell movements during gastrulation is under the control of the “planar cell polarity” (PCP) Wnt pathway; and the Brachyury transcription factor has well-known roles in the control of gastrulation and mesoderm specification [183]. The reported effect of Frizzled1 inactivation on cilia alignment and cell movement in the Clytia embryo [139] suggests that Wnt signalling through the PCP pathway may similarly act in cnidarian gastrulation. That Wnt signalling governs both axial cell fate differences and gastrulation movements leaves opened the question of which of these two roles is the most ancestral.

The origin of Wnt involvement in axial polarity has been recently pushed even earlier in time by expression data in the sponge Amphimedon, showing localised expression of Wnt and TGF-β genes at opposite poles of the sponge embryo and larva [2]. Wnt expression becomes localised at the future posterior pole of the embryo (with respect to the swimming direction of the larva) before any morphological polarity is detectable. Thus, a signalling centre similar to the bilaterian posterior centre and the cnidarian blastoporal centre might act on the sponge larva posterior pole. However, functional experiments and information on other genes of the cassette (e.g. Notch, Brachyury), as well as data on other sponge taxa, are requested for confirming these still preliminary conclusions.

Thus, with reservations due to the absence of Wnt data in ctenophores and still preliminary data in sponges, Wnt signalling could be the most ancient mechanism for establishing early cell fate differences along the main body axis [112,171] (but for a more sceptical point of view see Primus and Freeman [162]). A correlative suggestion is that the main body axes of sponges, cnidarians and bilaterians are probably homologous. It is furthermore tempting to speculate about a possible correspondence between the posterior pole of sponge and cnidarian larvae (and thus the adult sponge apical extremity and the polyp oral extremity) and the rear end of bilaterian animals, based on the position of the Wnt centre. In any case, it is clear that the acquisition of Wnt genes in an exclusive ancestor of the Metazoa (they are absent from non-metazoan eukaryote genomes) was a decisive event with respect to the later evolution of animal body plans.