Branching out: origins of the sea urchin larval skeleton in development and evolution

Abstract

It is a challenge to understand how the information encoded in DNA is used to build a three dimensional structure. To explore how this works the assembly of a relatively simple skeleton has been examined at multiple control levels. The skeleton of the sea urchin embryo consists of a number of calcite rods produced by 64 skeletogenic cells. The ectoderm supplies spatial cues for patterning, essentially telling the skeletogenic cells where to position themselves and providing the factors for skeletal growth. Here we describe the information known about how this works. First the ectoderm must be patterned so that the signaling cues are released from precise positions. The skeletogenic cells respond by initiating skeletogenesis immediately beneath two regions (one on the right and the other on the left side). Growth of the skeletal rods requires additional signaling from defined ectodermal locations, and the skeletogenic cells respond to produce a membrane-bound template in which the calcite crystal grows. Important in this process are three signals, FGF, VEGF, and Wnt5. Each is necessary for explicit tasks in skeleton production.

Introduction

The sea urchin larva produces a beautiful, intricately patterned, calcareous endoskeleton. Enclosed within the transparent epithelium of the embryo, the skeleton gives the larva its characteristic 3-dimensional shape. Because each half of the skeleton is formed as a single crystal of calcite, the skeleton is refractive to light and prominently visible under a light microscope (Fig. 1). The form of the skeleton generally resembles an artist’s easel and for this the larva was named pluteus by Johannes Muller in 1846 (from the Greek for easel, see Hörstadius, 1973). The skeleton of the early pluteus larva has four arms extending out ventrally from the larval body (Fig. 1F). The detailed anatomy of this structure differs widely among the euechinoids (Wray, 1992), reflecting changes in the patterning inputs over time. In all species, the skeleton is produced by a small number of mesenchyme cells. These cells fuse, and share a common cytoplasm. The membranes of this syncytium surround an extracellular pocket in which the calcite skeleton grows. The final skeletal pattern is strongly dependent on signaling inputs from the surrounding ectodermal epithelium (Duboc et al., 2004; Hardin et al., 1992; Armstrong et al., 1993; Hardin and Armstrong, 1997; Guss and Ettensohn, 1997; Rottinger et al., 2008; McIntyre et al., 2013). The nature of these patterning signals, how they are localized within the ectoderm, and how the mesenchyme cells respond to them are questions considered here. This system is a fascinating and relatively simple example of how tissues collaborate to produce pattern during morphogenesis. All animals pattern the three-dimensional shape of the embryo. Here we take advantage of molecular information that has accumulated on specification of ectoderm and mesoderm in the sea urchin to understand how the skeletal patterning template emerges.

Fig. 1.

Growth of the skeleton. (A) A DIC image of the ventrolateral cluster of PMCs shows at midgastrula stage the first evidence of a calcite skeleton (arrowhead). (B) 30 min later the tri-radiate skeleton is evident as the first branching skeleton (red dot). (C) One hour after initiating skeletogenesis the second branchpoint arises, also at a location within the ventrolateral cluster (yellow dot). (D) In a side view from the left the skeletal rods grow at prism stage show skeletal rods 4 and 5 extending from the second branch and rod # 2 extending anteriorly from the first tri-radiate. Branch #1 can also be seen extending medially. (E) At midprism the third branch appears (blue dot). Finally, the early pluteus stage displays the “four-armed” skeleton (2 postoral arms (#5) and 2 anterolateral arms (#6). Later four additional arms will be added prior to metamorphosis. Skeletal elements include: 1) ventral transverse rod, 2) D-V connecting rod, 3) anonymous rod, 4) body rod, 5) postoral rod, 6) anterolateral rod, 7) recurrent rod. The three branchpoints are illustrated with a red, yellow and blue dot to indicate the first, second and third branch respectively. Arms not numbered in E and F are present but out of the plane of the images.

The echinoderm phylum consists of 5 extant classes: Crinoidea (feather stars and sea lillies), Asteroidea (sea stars), Holothuroidea (sea cucumbers), Ophiuroidea (brittle stars and basket stars) and Echinoidea (sea urchins and sand dollars) (McClay, 2011). Most echinoderms employ a biphasic life cycle that includes a planktonic larval stage followed by a benthic adult stage. The sea urchin larval skeleton is thought to serve several functions. Primarily, it protects the digestive organs and helps orient the arms upward in the water column for effective swimming and feeding (Pennington and Strathmann, 1990). A ciliary band a few cells wide runs along the larval arms and functions in locomotion and orientation, and it produces currents directing food to the mouth (Strathmann, 2000; Strathmann, 2007; Strathmann and Grunbaum, 2006). Echinoid larval skeletons are diverse between orders and even between species of the same order (Kinjo et al., 2008; Wray, 1992). The number, length, shape and decoration (e.g. spines) of the skeletal rods supporting the arms can vary widely. Likewise, the elements and basketlike shape of the body skeleton are different between species.

Fig. 1 shows a sequence of skeletal growth in Lytechinus variegatus. For orientation we will refer to the Dorsal-Ventral sides of the embryo with the mouth in the center of the Ventral side. We will also use the terms Anterior-Posterior to define the orthogonal axis with the anus or blastopore being the most posterior structure. We use this terminology throughout based on recent conventions in the literature (Wei et al., 2012; Range et al., 2013), and because the nomenclature of the skeletal elements, established early in the last century, is largely based on those terms. On the right and left side of the embryo the skeletogenic cells, also called primary mesenchyme cells (PMCs) enter the blastocoel, migrate, and form a ring around the future archenteron at the posterior end of the embryo. Cells then migrate within the ring so that more than half the 64 PMCs accumulate as the right and left ventrolateral clusters (Peterson and McClay, 2003). As will be seen below, the migratory cues that localize PMCs to the ventrolateral clusters are provided through ectodermal signaling. In response to local signals each ventrolateral cluster of PMCs produces a tri-radiate skeletal primordium (Fig. 1A,B). Shortly after the first branch (red dot in Fig. 1), the skeleton branches again (yellow dot). Later a third branch occurs in the anterior ectoderm just inside the overlying dorsal-ventral margin (DVM)(blue dot).

The basic relationship underlying skeletal patterning—that the mesenchymal cells produce skeleton under the direction of ectodermal signaling—was inferred more than one hundred years ago. Both Herbst (1893) and Driesch (1896) (in Horstadius, 1973) suggested that the ectoderm contributed inputs to patterning of the skeleton. They based their idea on the observation that the apex of the skeleton was always at the mid-dorsal ectoderm. However, for about 100 years there was little experimental evidence to support this hypothesis of non-autonomous input. Horstadius (1973) credits Boveri as the first to report that the micromeres become the skeletogenic cells (Boveri, 1901a; Boveri, 1901b). Horstadius later confirmed this lineage relationship by vitally staining micromeres and transplanting them to unstained micromereless embryos (Hörstadius, 1935). In this manner, embryologists in the first half of the 20th century made significant progress in understanding the source of mesenchyme cells in the early embryo. With this knowledge of cell fates in hand, other researchers turned to studying the behaviors of differentiating mesenchymal cells. Production of the calcite crystal rods was studied in detail by Okazaki (1960, 1975). A paper by Okazaki and Inoue (1976) was especially important in showing that while the skeleton produced in vivo was birefringent, it nevertheless was smooth (perhaps suggesting a partial amorphous calcium carbonate structure). However, if one isolated the skeleton and grew calcite crystals on it, the crystals grew in a single orientation, thus showing that despite its smooth appearance the skeleton was a single crystal. Later it was shown that the calcite crystal grows by addition of amorphous calcium carbonate granule secretion from the skeletogenic cells followed by rapid conversion of that amorphous Calcium carbonate into the single calcite crystal (Beniash et al., 1999). Gustafson and Wolpert (1961) described the behavior of skeletogenic cells as they migrated inside the blastocoel in preparation for skeletogenesis, and Okazaki showed that many of those same behaviors occurred when PMCs were placed in vitro (Okazaki, 1975). The Okazaki work and earlier work by von Ubisch and Baltzer supported the hypothesis that autonomous patterning mechanisms were programmed into the skeletogenic cells (Ubisch, 1933; Ubisch, 1939; Baltzer et al., 1959). As a consequence they showed that details of the skeletal elements such as whether or not the skeleton rods were fenestrated, smooth, or contained sharp branches, was a function of the autonomous skeletogenic program. All of these early studies, and many others, set the stage for molecular analyses of specification, migration of PMCs, patterning of the ectoderm, signaling from the ectoderm to pattern the skeleton shape, and production of anatomically correct skeletal elements. That complex process provides a model for how multiple tissues in an embryo contribute to emergence of pattern (reviewed by Ettensohn, 2009, Lyons et al., 2012).

In the last two decades much work has been dedicated to building a detailed gene regulatory network for sea urchin embryogenesis, and the skeletogenic lineage is the best understood of all cell types (Oliveri et al., 2008; McClay, 2011). In parallel, the biogenesis of the skeleton itself has been studied in detail from structural, proteomic and genomic perspectives (Killian and Wilt, 2008; Livingston et al., 2006; Wilt et al., 2003). One of the most significant advances came in 1992, when it was finally demonstrated experimentally that the ectoderm is a source of non-autonomous patterning information for the mesenchyme cells (Hardin et al., 1992). Hardin et al showed that if ectodermal patterning was perturbed there was a dramatic effect on skeletal patterning, even when the skeletogenic cells were unperturbed. Shortly thereafter, a series of chimera experiments again demonstrated that ectodermal inputs were crucial for patterning (Armstrong et al., 1993; Armstrong and McClay, 1994; Hardin and Armstrong, 1997). By transplanting mesenchymal cells between species with different skeletal patterns, it was shown that the overall shape of the skeleton depends on inputs from the host ectoderm. The skeletal rods grew then with a combination of non-autonomous inputs from ectoderm plus the autonomous patterning based on the genotype of the skeletogenic cells. The fact that skeletogenic cells of one species correctly patterned a larval skeleton even when surrounded by ectoderm of another species also suggested that many of the patterning cues were remarkably conserved.

The molecular nature of the signals that pattern the mesenchyme cells has begun to be revealed over the last several years. Recent work has shown that two orthogonal bands of ectoderm are uniquely specified at the molecular level to produce growth factors and patterning cues to the mesenchyme cells. In this review, we highlight the recent advances in the understanding of ectodermal patterning, explore how the various signals direct skeleton formation, and discuss how this system has changed during evolution. Many questions remain unresolved. In particular, while we have a gross understanding of the ways mesenchyme cells respond to signaling, many of the molecular mechanisms underlying these processes are not known. How do the skeletogenic cells move to their correct locations? How do they organize into a syncytium? How do they interpret the signals and secrete the skeleton crystal in the correct shape? And how does the protein matrix function as it is incorporated into the crystal? To address those questions it is first necessary to understand the information already known about skeletal patterning. That is the purpose of this paper.

Ectodermal Patterning Foreshadows the Embryonic Skeleton

Ectoderm patterning begins with early subdivisions of cell specification along the anterior-posterior (A-P) and dorso-ventral (D-V) axes. This process results in two bands of ectoderm. The dorsal-ventral margin (DVM) of the ectoderm separates the dorsal and ventral ectoderm (Fig. 2A, pink band). The border ectoderm (BE) encircles the embryo immediately adjacent to the endoderm. (Fig. 2A, blue band)(McIntyre et al., 2013). The DVM and ventral half of the BE eventually becomes the Ciliary Band that surrounds the ventral ectoderm. Once specified, these territories will participate in morphogenesis in several ways: 1) The DVM and the BE provide the ectodermal patterning inputs for skeletogenesis; 2) the Ciliary Band will later form along the DVM and the ventral portion of the BE. 3) the Ciliary Band will also later contain many of the neurons of the larva (Nakajima et al., 2004; Burke et al., 2006; Angerer et al., 2011). The early distribution of skeletogenic mesenchyme cells parallels the BE and then the portion that grows anterior parallels the DVM band. The right and left BE-DVM intersections coincide with the ventrolateral clusters of PMCs (Fig. 2A). We now know that cells in the BE and DVM produce growth factors that direct migration of the mesenchymal cells. VEGF and FGF are initially produced throughout the BE (at the time the PMCs initially form into a ring immediately beneath the BE), and then VEGF is up-regulated at the BE-DVM intersection, coincident with the clustering of the PMCs just beneath. Shortly thereafter skeletogenesis begins in those ventrolateral clusters (McIntyre et al., 2013; Adomako-Ankomah et al., 2013, Duloquin et al., 2007; Rottinger et al., 2008). Thus, to understand how skeletogeneic patterning works it is necessary to understand how the BE, DVM and BE-DVM intersection are precisely located in the ectoderm.

Fig. 2.

Relationship between Dorsal-Ventral Margin (DVM), Border Ectoderm (BE), and skeleton. (A) Diagram of the mesenchyme blastula stage from the left side showing the DVM and BE. The BE-DVM intersection is diagrammed and will later produce VEGF to initiate skeletal growth of the PMC ventrolateral cluster just inside the ectoderm. At that location the tri-radiate skeletal element initiates. The anonymous rod branches shortly thereafter, also beneath the region of the BE-DVM intersection. The third branch of the skeletal later occurs along the DVM in the oral hood. (B) A diagram of the skeleton with the three branch points shown on each side. On each side 5 of the 7 rods grow in the blastocoel immediately inside either the DVM or BE. The 2 rods in the anterior/oral hood produce arms that initiate at the DVM branch point (blue dot) and extend anterior and posterior.

The border ectoderm is established by short-range signaling from the endoderm. Recent work demonstrated that the BE is uniquely specified via signaling from the endoderm (McIntyre et al., 2013). This narrow band of ectodermal cells, about 4–5 cells wide, forms immediately adjacent to the endoderm prior to gastrulation (Fig. 2A). A number of transcription factors and signaling molecules mark the BE and BE-DVM intersection; these include Lim1, Nk1, Pax2/5/8, IrxA, Msx, Otp, Trim1 and FoxJ1 (Fig. 3) (McIntyre et al 2013; Czerny et al., 1997); Su et al., 2009; Saudemont et al., 2010; Di Bernardo et al., 1999). These markers are expressed dynamically such that by the beginning of gastrulation each is localized to a sub-region of the BE. Even though the pattern of the larval skeleton is species-specific, the transcription factors expressed in the BE are similar across several species of urchins including Lytechinus variegatus, Strongylocentrotus purpuratus, and Paracentrotus lividus.

Fig. 3.

The BE and DVM are special domains of gene expression. Three signals are produced in the BE-DVM intersection. A number of transcription factors are expressed in one or more of the several domains.

The BE is induced by a short-range Wnt5 signal from neighboring endodermal cells (McIntyre et al., 2013). The evidence supporting this signal includes knockdown and mis-expression of Wnt5, as well as perturbations to endoderm specification upstream of Wnt5 expression. The initial observation was that many perturbations to endoderm specification also disrupt formation of the skeleton (for example, foxa, bra, Hox11/13: Gross and McClay, 2001; Oliveri et al., 2006; Peter and Davidson, 2010). These data suggested that endoderm played some role inducing formation of the skeleton. Since VEGF and FGF were identified as ectodermal drivers of skeleton, if the endoderm functioned in skeletogenesis, it had to operate indirectly through the ectoderm. Indeed, expression of VEGF, FGF, and other BE markers were lost in embryos where hox11/13b expression was blocked by MASO (McIntyre, et al., 2013).

Wnt5 was identified as the endodermal signal because it is first expressed downstream of Hox11/13b in the endoderm, is necessary for BE marker expression, and it activates BE markers when ectopically expressed. Embryos injected with a Wnt5 MASO fail to activate BE markers and produce no skeleton. When ectopic Wnt5 mRNA was injected, BE markers were activated without restriction along the A-P axis (see below), showing that Wnt5 is capable of activating BE fates in all ectodermal cells. This result was confirmed by transplanting single blastomeres mis-expressing Wnt5 mRNA; the presence of the Wnt5 protein in an ectopic cell set up a developmental sequence: these cells attracted PMCs, and the PMCs produced ectopic skeleton. Wnt5 later is expressed in the dorsal BE after gastrulation has begun. This later expression is dependent on the earlier endodermal Wnt5 activity, and is not seen in embryos injected with the Wnt5 MASO (McIntyre, et al., 2013).

Several important conclusions emerged from these experiments. First, it was clear that BE specification was necessary for skeleton formation. In Wnt5 MASO embryos, no skeleton formed and the growth factor VEGF was not expressed at the BE-DVM intersection. Second, Although Wnt5 has been shown to be required for convergent extension movements during gastrulation in Xenopus and Zebrafish (Kilian et al., 2003; Yamanaka et al., 2002), sea urchin embryos lacking Wnt5 function form a gut normally, indicating that Wnt5 probably functions primarily as a signal to the ectoderm and not as a regulator of gut morphogenesis. Finally, endodermal Wnt5 acts as a short-range signal. All cells are able to respond to ectopic Wnt5 signal, but in vivo, BE fates are only activated in a band 4–5 cells wide immediately adjacent to the endoderm border. This is consistent with Wnt5 activity in cultured MEF cells, which appear not to release Wnt5 into the culture media (Ho et al., 2012). The use of a short-range signal to specify ectodermal fate is consistent with the use of tiered, short-range signals to specify endomesoderm in the sea urchin embryo (Davidson, 1989) and suggests it may be the major mechanism used to assign fates along the anterior posterior axis.

The above conclusion is at odds with classical embryological thinking, which hypothesized that sweeping, long-range gradients of factors within the embryo were responsible for determining fates along the A-P axis. As cleavage proceeded it was thought that a cell’s position within the gradient resulted in differential inheritance of cytoplasm, different complements of these factors and, ultimately, different fates. It was in this intellectual tradition that Horstadius (1939, 1973), interpreting his extensive results on sea urchin blastomere recombinations, posited that dual gradients of secreted factors emanating from the animal and vegetal poles of the embryo provided the information necessary to specify cell lineage. Later it was argued that an equally plausible explanation could be a series of local inductive interactions between tiered blastomeres to explain the experimental results obtained by Horstadius and others (Davidson, 1989).

In the more than twenty years since Davidson first proposed it, most of the accumulated molecular data indicate that the short-range tier-to-tier signaling hypothesis is better supported than the long-range gradient hypothesis of earlier thinking. Maternal Wnt signaling autonomously activates specification of the large micromeres, fated to produce the embryonic skeleton (Emily-Fenouil et al., 1998; Wikramanayake et al., 1998; Logan et al., 1999). Micromeres signal through Delta-Notch and an unknown signal to their macromere neighbors to regulate endomesoderm specification (Sherwood and McClay, 1997, 1999; (Sweet et al., 2002) Oliveri et al., 2003). Shortly thereafter the macromeres equatorially cleave into the Veg2 and Veg1 layers, the Veg2 layer then signals to the Veg1 layer above it (Peter and Davidson, 2011). The Wnt5 evidence above extends that short range tier to tier signaling relationship into the ectoderm since the Veg1 layer induces the BE via short range Wnt5 (McIntyre et al., 2013), and further signaling to the anterior end of the embryo (animal pole) has been suggested by additional Wnt signaling (Range et al., 2013). Thus most cells along the A-P axis likely are specified with the expression and necessary contribution of a series of short-range signals.

D-V Axis Formation and Subdivision of the BE by TGFβ Signaling

The Dorso-Ventral (aboral-oral) axis in the sea urchin is activated by asymmetric TGFβ signaling in response to what is proposed to be an earlier mitochondrial asymmetry (Bradham and McClay, 2006; Coffman and Davidson, 2001; Duboc et al., 2004). The two principle signals in this process are Nodal and Bmp2/4 (Duboc et al., 2004). Both molecules are produced on the ventral side of the embryo. Nodal activates exclusively ventral fates, whereas BMP2/4 diffuses to the opposite side of the embryo and activates dorsal fates. Several inhibitors of TGFβ signaling limit the extent of Nodal and BMP signaling, ensuring that cells in the D-V margin receive neither signal and therefore have a unique gene regulatory network state. In most ectodermal cells, Nodal and BMP2/4 serve as activators of the gene regulatory network. Without them, D-V fates are lost and the ectoderm defaults to a ciliated epithelium. Surprisingly, the BE is not activated by TGFβ; instead gene expression domains within that band are limited by Nodal and BMP. Dorsal markers in the BE are repressed on the ventral side by Nodal signaling; ventral markers of the BE are repressed in the dorsal side by BMP signaling (McIntyre et al., 2013).

Perturbations to the Nodal-BMP system in the sea urchin lead to specific defects in skeletal patterning (Hardin et al., 1992; Duboc et al., 2004). Those defects generally include loss of anterior structures including the oral hood, and radialization of posterior skeletal elements. These contrasting effects mirror the dual activities of TGFβ signaling in the border versus non-border ectoderm. Absent Nodal signaling, tri-radiates are nucleated in a radial pattern and these elongate into unstructured rods. When BMP is knocked down, embryos have only a partially radialized vegetal skeleton - extra spicules form only around the dorsal aspect of the BE. These results indicate a model where normally, growth factor signaling is initially activated at the intersection of the BE and DVM (Fig. 2). Signaling from Nodal and BMP contribute to the refinement of the BE sub-compartments so that only the BE-DVM intersection contains the correct information for VEGF and FGF production, thereby providing the site for skeletogenesis immediately adjacent to that patch of cells.

A prediction from this model is that when Nodal and BMP are mis-expressed throughout the ectoderm, BE gene expression and skeleton would be lost (because both Nodal and BMP repress BE genes). Indeed this is the case when BMP mRNA is injected into embryos (McIntyre and McClay, unpublished data). Interestingly, this is only partially true for Nodal. At high doses, Nodal does indeed block skeleton formation. However, at low doses embryos have robust, radialized skeletons. In these embryos endodermal Wnt5 apparently is able to induce the BE and that induction is sufficient for signaling to initiate skeletogenesis. The ability of Wnt5 to modify ventral ectoderm specification explains the longstanding observation that NiCl2 – treated embryos have both a ciliary band surrounding the vegetal plate of the embryo and radialized tri-radiate skeletal elements just inside the ciliary band (Hardin et al., 1992). Normally, the CB passes through the DVM, separating the ventral from dorsal ectoderm, but it also overlaps with the ventral BE. In NiCl2 – treated embryos Nodal is expressed throughout the Ventral and Dorsal ectoderm. The BE is specified, apparently because the level of Nodal expression is not high enough to block the BE induction by Wnt5. However, in that situation there is no BMP produced to separate Dorsal from Ventral, hence no DVM. Nevertheless, the BE is sufficient to allow tri-radiate skeletal elements to form in a radial pattern surrounding the embryo.

Ectodermal Signaling to the skeletogenic cells

Over the past ten years, a number of ectodermal genes have been implicated in the process of skeleton formation. These include the signaling ligands VEGF, FGF and recently, Wnt5 (Duloquin et al., 2007; Rottinger et al., 2008; Adomako-Ankomah and Ettensohn, 2013; Knapp et al., 2012; McIntyre et al., 2013), as well as two transcription factors, Otp and Pax 2/5/8 (Cavalieri et al., 2003; Saudemont et al., 2010; Rottinger et al., 2008) and the gene Trim1 (Cavalieri et al., 2011). A common feature of these genes is that they are all expressed at the BE-DVM intersection, immediately adjacent to the ventrolateral skeletogenic mesoderm clusters. Loss-of-function experiments indicate that each gene is required for the skeleton to form.

The first molecular data explaining how the ectoderm contributes to skeletogenesis was the cloning and functional analysis of VEGF (Duloquin et al., 2007). VEGF is expressed in the lateral BE-DVM intersection, and its receptor is exclusively expressed by the PMC’s; and is particularly enriched in the PMCs found within the ventrolateral clusters. Absent VEGF signaling, skeletal formation does not occur, and the PMCs appear disorganized within the blastocoel. Ectopic VEGF mRNA resulted in extra skeletal branches, consistent with a role for this molecule in skeleton branching. However, based on these data alone, it was not possible to determine the exact nature of the patterning information imparted to the PMCs by VEGF alone.

Soon after this, FGF was discovered to play a role in skeletogenesis as well. But unlike VEGF, FGF is expressed in multiple territories as the embryo develops (Rottinger et al., 2008). After an early period of expression at the vegetal pole, FGF is expressed in the BE-DVM intersection, and shortly thereafter is expressed in the PMCs themselves especially at the distal tips of the arms as the postoral rods elongate. This complex expression sequence has complicated analysis of its exact function because it is difficult to selectively perturb FGF signaling either in the DVM intersection or in the PMCs. However, when FGF expression was blocked with a MASO, no skeleton was formed and expression of Pax2/5/8 and Pea3 was lost (Rottinger et al., 2008). By contrast, a recent paper suggests that the role of FGF in skeletogenesis is more limited and may contribute little signaling information to the PMCs to initiate the tri-radiate (Adomako-Ankomah and Ettensohn, 2013).

Wnt5, discussed above as an activator of border ectoderm fates, also has a phase of expression in the BE (McIntyre et al., 2013). After gastrulation, Wnt5 expression moves from the endoderm, to the dorsal and lateral BE. As gastrulation proceeds, it is further restricted to two lateral patches of the BE corresponding to the BE-DVM intersection where VEGF and FGF are expressed. This later expression of Wnt5 depends on earlier Wnt5 signaling from the endoderm (McIntyre et al., 2013). As such it has been difficult to precisely distinguish this later Wnt5 function from the earlier inductive activity. It is possible that at this stage, Wnt5 signals directly to the PMCs though current evidence suggests that the action of Wnt5 on skeletogenesis is indirect via activation of VEGF.

Although we have referred to the BE-DVM junction as if it is a uniform patch of cells, that may not be true. VEGF, FGF and Wnt5 are each expressed in that region but it is possible that the BE-DVM intersection actually does not uniformly express the three signals. Each is expressed in this region, but because there are small embryo-to-embryo variations, it is difficult to distinguish between the BE-DVM as a uniform patch of cells, or whether there are real differences between each signal’s expression domain, and the additional possibility exists that the patch undergoes dynamic changes. For example, if VEGF were expressed slightly more ventrally than FGF, or if Wnt5 were shifted dorsally by 2 or 3 cells, they may be able to provide more detailed positional information than if they were perfectly coincident. In other words the possibility remains that the BE-DVM intersection contains subdomains that have functional significance. Recently, Adomako-Ankomah and Ettensohn (2013) examined expression patterns of VEGF and FGF mRNA using double fluorescent WMISH. At mesenchyme blastula both mRNAs were restricted to the ectoderm over the ventrolateral clusters. During gastrula stages VEGF and FGF expression overlapped briefly in the ectoderm over the ventrolateral clusters (in the BE-DVM intersection), but by late gastrula the expression of FGF ceased to be observed in the ectoderm over the ventrolateral clusters and switched to being expressed in PMCs of the ventrolateral cluster themselves (Adomako-Ankomah and Ettensohn, 2013). As often happens it is the details that matter. At the site of the ventro-lateral clusters two successive skeletal branching events occurs. The extent to which signaling from the BE-DVM controls that precise patterning remains to be clarified.

An important question yet to be addressed is how each of the signals is uniquely activated in the BE-DVM intersection. What are the immediate upstream transcriptional activators? Do different signals have the same or different activators? The logic underlying activation is of interest because it determines the timing and intensity of each signal, which in turn are needed for proper patterning. Moreover, as the skeletal pattern has changed through evolution, it is likely that the activation logic has been one target of selection. Three GRN models of ectoderm specification in sea urchins have been published (Saudemont et al., 2010; Su et al., 2009; Li et al., 2012). These network models do not include the D-V margin and border ectoderm as part of their models, and connections to VEGF, FGF and Wnt5 are either incomplete or missing. Knowledge of expression of transcription factors and the signals in the BE and DVM help complete this analysis, though the number of transcription factors involved may yet be incomplete, and perturbation analyses establishing epistatic relationships have yet to be published.

Evolution of skeletal patterning

Once the GRN operating at the intersection of the BE and DVM is understood, and once the position and pattern of the skeletal rudiment is known in greater detail, it will become possible to ask how evolution of different patterns occurred. It has been suggested that because the echinoderm crown group ancestor possessed an auricularian-like larva with no skeleton (Nakano et al., 2003), that the larval skeleton in echinoids is the result of a heterochronic shift that activates the adult skeletogenic program early in development (Gao and Davidson, 2008; Ettensohn, 2009). Somehow this skeletogenic program was integrated with existing GRNs that govern the placements of the DVM territory and the BE territory. To appreciate how this happened it would be of great value to know the expression patterns of genes in these territories in echinoid outgroups.

As yet the expression patterns of several key players including Pax2/5/8, Otp and FGF have not been examined in other echinoderms. But, the expression of VEGF and VEGFR has been examined in sea stars (class Asteroidea), which do not have larval skeletons (Morino et al., 2012). Transcripts were not detected by in situ hybridization or qPCR in sea stars before the bipinnaria larval stage. At the brachiolaria stage VEGF (expressed in the ectoderm) and VEGFR (expressed in the mesoderm) were associated with the rudiments of the adult skeleton. Thus we can hypothesize that the heterochronic activation of VEGF and VEGFR in the euechinoid ancestor was an equally essential step in the evolution of the larval skeleton as was the heterochronic activation of the skeletogenic program itself.

Brittle stars (class Ophiuroidea) also have a larval skeleton that is considered to have emerged independently from that in euechinoids (Williamson, 2003), which if true, makes them an excellent additional test case for asking how the adult skeletogenic program and ectoderm/mesoderm communication system evolved. In contrast to echinoids, the brittle star skeletogenic lineage does not arise from an asymmetric cell division at 4th cleavage. Instead, the skeletogenic cells, also called PMCs, become obvious inside the blastocoel before gastrulation and form bilateral rudiments of the larval skeleton (Yamashita, 1985). Brittle star homologs of VEGF are expressed in the ectoderm in bilateral patches very much like the pattern in sea urchins, and VEGFR is likewise expressed in PMC ventrolateral clusters (Morino et al., (2012). As in sea urchins, the expression of brittle star VEGF is initially expressed in a ring in the ectoderm before gastrulation, but then expression is restricted to the lateral patches. During prism stages VEGF and VEGFR continue to be expressed at the tips of the growing arms, in the ectoderm and mesoderm, respectively. This remarkable similarity in expression patterns of VEGF and VEGFR between sea urchins and brittle stars suggests that either the conventional wisdom that the larval skeletons are independently derived in these two groups is wrong, or suggests that heterochronic activation of this pathway was essential during independent innovations of the larval skeleton is both clades.

As more genes are examined in brittle stars, and in more species, a clearer picture will emerge. Ideally, it would be of value to assess how the whole network surrounding VEGF signaling operates relative to sea urchins. Recently Vaughn et al., (2012) generated a transcriptome of the gastrula stage of a brittle star and examined its complement of “network” genes and skeletogenic genes. A comparison of territory-specific GRN components expressed in brittle star gastrulae showed that the highest conservation between sea urchin and brittle star were those genes in the skeletogenic GRN, suggesting that convergent evolution of the larval skeleton in brittle stars may have occurred. They detected VEGF transcripts in the gastrula, but not VEGFR. Also, a number of genes that are present in the BE in sea urchins, such as Irx, Msx, Lim and Nk1, were not represented in their gastrula transcriptome. Nodal was not present in the transcriptome, but BMP2/4, Hnf6 and Foxj1 were. These differences could indicate large differences in the way the BE is established in brittle stars but until a detailed analysis of brittle star skeletogenesis and ectodermal signaling is conducted evolutionary insights will remain rudimentary.

While the phylogenetic relationship of brittle stars and sea urchins might still be under debate, it is well established that the sea cucumbers (class Holothuroidea) branch closest to the euechinoids (Janies et al., 2011; Pisani et al., 2012). Their larval form does not have an asymmetric early cleavage resulting in micromeres, nor a well-developed larval skeleton. However, in some species the auricularian larvae possess posterior spicules, which are thought to act as ballast for proper orientation in the water column (Pennington and Strathmann, 1990). McCauley et al., (2010) investigated the developmental and molecular origin of spicules in the larva of the sea cucumber Parastichopus. This species makes a very small spicule granule in the blastocoel underneath the posterior dorsal ectoderm. They showed a group of Alx1-positive cells ingress into the blastocoel, migrate away from the archenteron, form a dorsal chain of cells and then coalesce into a cluster of approximately 8 cells that corresponds to the position of the spicule. Knockdown of Parastichopus Alx1 with an antisense morpholino abolished this dorsal cluster and the spicule, suggesting that like sea urchins, Alx1 is involved in the skeletogenic fate. If true, one might expect holothurian skeletogenesis to use additional orthologs of genes used by sea urchins.

Going Forward: Building the 3D Skeleton

Once the tri-radiate skeleton is established, the skeletal rods elongate to form the 3-dimensional framework of the larva. The rods either grow parallel to the DVM or BE, or, the post-oral and anterolateral rods grow out ventrally, in each case distorting a patch of one of these bands, which expands in parallel with rod growth. Growth of each type of rod is remarkably stereotypical between embryos, but differs predictably between the various types of rods. Guss and Ettensohn (1997) measured rates of skeletal elongation and showed that most rods have a basal elongation rate of ~6um per hour at 23C. However certain rods, notably the post-oral and anterolateral rods, accelerate their rate of growth later in development. In addition, the ventral transverse rod ceases to elongate well before the end of embryogenesis. The authors concluded that mesenchyme cells use an autonomous program for rod production but extrinsic cues modify the basal activity.

Once the skeleton is initiated in the BE-DVM intersection it then appears to use both bands for much of its patterning since all skeletal elements except those in the anterior/oral hood grow parallel to these bands. How might that work? If we take all the information on localized cues (Armstrong et al., 1993; Duloquin et al., 2007; Guss and Ettensohn, 1997; Rottinger et al., 2008; McIntyre et al., 2013), and combine that information with data from studies on biomineralization (Illies et al., 2002; Killian and Wilt, 2008; Knapp et al., 2012; Livingston et al., 2006; Urry et al., 2000; Wilt, 2005), and also add in a number of earlier studies using experimental embryology (Armstrong et al., 1993; Armstrong and McClay, 1994; Hardin and Armstrong, 1997; Hardin et al., 1992; Guss and Ettensohn, 1997; Hodor and Ettensohn, 1998; Malinda and Ettensohn, 1994; Malinda et al., 1995; Peterson and McClay, 2003), we can begin to outline mechanisms used to build a skeleton. The following model is preliminary and probably misses many components of skeletal growth, but will be useful in directing future studies on how skeletal patterning may function in this embryo.

Since the pattern of the calcite crystal deposited depends on the shape of the syncytial cavity of PMC cell extensions, much of the patterning model hinges on how PMCs organize the syncitium. PMCs initially form as a ring at the posterior end of the blastocoel. Formation of the PMC ring likely is guided by VEGF, which at an early stage is expressed in a matching ring all around the BE (McIntyre et al., 2013; Adomako-Ankomah and Ettensohn, 2013). When that ring first forms the PMCs distribute themselves roughly equally around the circumference of the posterior ectoderm. It is during this stage that the PMCs extend long thin filopodia to touch each other and to touch the ectoderm beneath (the BE). They then initiate fusion to form the syncytium (Hodor and Ettensohn, 1998; Miller et al., 1995). The ventrolateral clusters form by PMCs migrating laterally from both the dorsal and ventral sides of the ring (Peterson and McClay, 2003). Formation of the ventrolateral clusters is directed by secretion of VEGF and/or FGF from the BE-DVM intersection (Duloquin et al., 2007; Rottinger et al., 2008; McIntyre et al., 2013)). If either signal, or loss of the receptor for either signal is missing, the PMCs fail to form the ventrolateral clusters. The next step is formation of the initial tri-radiate spicule. This process appears to be tightly controlled by signaling, though the mechanism by which this happens is less clear. VEGF appears to promote nucleation and branching (Knapp et al., 2012; McIntyre et al., 2013); if VEGF is knocked down embryos do not make skeletons, and overexpressed VEGF results in large numbers of ectopic, branched skeletal elements. FGF knockdown embryos produce tri-radiate skeletons but skeletal rods do not elongate (Rho and McClay, 2011), though in Paracentrotus FGF knockdown appears to eliminate even the tri-radiate (Rottinger et al., 2008).

The final embryonic skeleton emerges as the rods elongate and branch once more as the ventral/oral hood forms. The result is a bilaterally symmetric structure, each half of which is formed from a single crystal of calcite. The work of Okazaki demonstrated that each half of the skeleton was a single crystal of calcite which accounted for the unique birefringence of the skeletal rods (Okazaki, 1960, 1975; Okazaki and Inoue, 1976). The optical axis of the crystal is always aligned in the same manner, with the c-axis parallel to the post oral/body rods and with the three arms of the initial triradiate extending in the plane perpendicular to the c-axis. The precise, reproducible orientation of the crystal suggests the process is closely regulated, and it will be crucial to explain how the polarity of this crystal is established (especially given that its scale is so much smaller than that of the cluster of mesenchyme cell bodies which deposit it. Important for biomineralization, once the initial axis is formed, it is retained as the crystal grows because when added, Amorphous Calcium Carbonate (ACC) reaches the pre-existing crystal, and joins the crystal entrained by the initial crystal axes.

It is thought that the skeleton is deposited in a mold, the PMC syncytium. When the skeletal rods are observed with an electron microscope, the surface of each rod is visibly smooth, showing no edges characteristic of growing crystals (Okazaki, 1960, 1975; Okazaki and Inoue, 1976). However when such a rod is used as a nucleus for artificial growth, crystal edges readily appear as the in vitro crystal grows and the crystal growth is entirely in parallel planes indicating that the smooth rod nevertheless is a single crystal. When rods are fractured, the core of the rod appears to have concentric growth layers, much like rings on a tree. These observations suggested that the crystals are deposited in several stages. The ACC is inherently unstable, and in vitro will rapidly (within minutes) crystalize into calcite (Brecevic and Nielsen, 1989). Sea urchins and other marine invertebrates stabilize ACC with the addition of proteins, store the ACC intracellularly, and then deliver it to the growing skeleton though intracellular vesicles where it is induced to crystalize (Decker et al., 1987; Wilt, 2005; Beniash et al., 1999; Gong et al., 2012). The process of induction is not understood, but the ACC is added to the existing rod and may require changes to the proteins incorporated in the ACC. There are up to around 50 proteins incorporated into the ACC (Adomako-Ankomah and Ettensohn, 2011; Rafiq et al., 2012), so there is plenty of room for regulative activities in skeletal growth. Mahamid et al., (2011) speculate that perhaps the same molecules could be responsible for both stabilizing ACC and inducing calcite crystal formation after some type of structural change. Because of the use of ACC, the crystal can take the shape of whatever structure contains it at the time the ACC is induced to crystalize. All this suggests the crystals are NOT grown in a particular shape, but rather deposited in a mold, the PMC syncytial cavity. The PMCs appear, therefore, to act as a biological template, which is consistent with how Calcium carbonate is deposited in mollusks (Addadi et al., 2006; Wilt, 2005). In fact Addadi et al. conclude that template-directed spicule nucleation occurs in both sea urchins and mollusks. The question becomes what governs the placement of PMC chains, and tells which cells should fuse together? That question returns focus to the BE and DVM.

Important clues of function come from interspecies experiments (Armstrong et al., 1993; Armstrong and McClay, 1994). Lytechinus PMCs were inserted into the blastocoel of Tripneustes esculentus in place of Tripneustes PMCs, or the reciprocal recombination was done. In both combinations the PMCs were able to interpret the ectoderm signals of the other species to produce the 3 dimensional shape of the skeleton, i.e. the ventrolateral clusters formed at the correct location, and the skeleton grew along the BE and DVM. Thus, the ectodermal signaling inputs were well conserved over an evolutionary distance of more than 10 million years. Normally Tripneustes forms fenestrated postoral skeleton rods while all Lytechinus postoral skeletal rods are simple. In the recombinants the Tripneustes PMCs produced fenestrated postoral rods while the Lytechinus PMCs produced simple rods. Thus the actual rod type is determined by the genotype of the PMCs. Also determined by the genotype of the PMCs was the extent of the rods produced. As evidence, Tripneustes forms long recurrent rods while the recurrent rods in the Lytechinus skeleton are bent and stunted. In the recombinants the Tripneustes recurrent rods formed at the correct location and were correctly patterned (requiring ectoderm input for placement), while the Lytechinus recurrent rods were stunted when grown in Tripneustes, reflecting the autonomous readout of Lytechinus genotypic information in response to Tripneustes ectodermal positional cues. Of note, the Lytechinus ectoderm provided spatial signals to Tripneustes even though the Tripneustes skeleton grew in an area where no Lytechinus skeleton grows. This suggests that the ectoderm provides spatial signals essentially telling the skeletogenic cells where their location in the embryo (positional information), and it provides growth factors that promote skeletal growth. The PMCs genotypically provide the template for the actual skeletal element by virtue of the syncitial folds in which the skeletal rods grow: a simple template for simple rods, a fenestrated template for fenestrated rods. Thus the 3-dimensional skeleton is a combination of spatial cues and growth factor provision from the ectoderm plus a built in response mechanism that allows the PMC syncytium and the proteins produced by the PMCs to mold the correct skeletal element in any position.

This then has been the thesis of this paper: signaling cues from the ectoderm provide the spatial information for the PMC syncytium template using two belts of ectoderm, the DVM and the BE. That means ectoderm patterning ultimately is responsible for the three dimensional shape of the larval skeleton. The challenge is to understand the many levels of control that sets up the border ectoderm and D-V margin as specific territories in the ectoderm, then to understand how the DVM and BE provide the correct spatial cues to the PMCs, at the correct time, and finally, how the PMCs read, interpret, and then form the template for the skeletal element they produce.