Growth factor-mediated mesodermal cell guidance and skeletogenesis during sea urchin gastrulation

Growth factor signaling pathways play essential roles in diverse developmental processes (Hogan, 1999; Wilson and Leptin, 2000; Wu and Hill, 2009; Dorey and Amaya, 2010), including the migration and differentiation of mesoderm cells during gastrulation. In Drosophila embryos, members of the fibroblast growth factor (FGF) family, pyramus and thisbe, are expressed in the ectoderm and regulate the intercalation, spreading, migration and differentiation of mesoderm cells, which express the FGF receptor heartless (Kadam et al., 2009; McMahon et al., 2010; Winklbauer and Müller, 2011; Reim et al., 2012). A similar role is played by platelet-derived growth factor (PDGF) during Xenopus gastrulation; in this case, the ligand PDGF-A, expressed in the ectoderm, regulates the orientation and migration of mesoderm cells that express PDGFRα (Ataliotis et al., 1995; Nagel et al., 2004; Damm and Winklbauer, 2011; Winklbauer and Müller, 2011). The migration of mesoderm cells away from the primitive streak in the gastrulating chick embryo is regulated by N-cadherin through the PDGF pathway (Yang et al., 2008; Chuai and Weijer, 2009), while FGF4 and FGF8 provide directional cues by acting as attractants and repellants, respectively (Yang et al., 2002; Lunn et al., 2007; Chuai and Weijer, 2009). Additionally, in the mouse embryo, FGF signaling regulates the migration of ingressed mesoderm cells and the formation of paraxial mesoderm (Sun et al., 1999; Ciruna and Rossant, 2001; Boulet and Capecchi, 2012).

The sea urchin embryo, which is optically clear and amenable to both cellular and molecular manipulations, serves as a valuable model with which to study the role of growth factors in gastrulation. Gastrulation in the sea urchin is characterized by the directional migration of mesodermal primary mesenchyme cells (PMCs), which produce the embryonic endoskeleton (Wilt and Ettensohn, 2007; Ettensohn, 2013). PMCs, after undergoing an epithelial-to-mesenchymal transition (EMT), migrate directionally within the blastocoel by means of filopodia and form a characteristic ring pattern (the subequatorial PMC ring), which consists of two ventrolateral clusters (VLCs) of PMCs joined by oral and aboral cell strands. The PMCs fuse as they migrate, forming a single continuous syncytium within which an elaborate calcium carbonate endoskeleton is secreted. Specific cues from the ectoderm of the embryo regulate both the migration and differentiation of the PMCs, thereby determining the structure of the resulting endoskeleton (Ettensohn and McClay, 1986; Ettensohn, 1990; Hardin et al., 1992; Armstrong et al., 1993; Ettensohn and Malinda, 1993; Guss and Ettensohn, 1997). However, relatively few genes expressed in the ectoderm have been shown to play a direct role in regulating PMC behavior (Di Bernardo et al., 1999; Cavalieri et al., 2003; Duloquin et al., 2007; Röttinger et al., 2008; Cavalieri et al., 2011).

Recent work has pointed to two pathways - mediated by vascular endothelial growth factor (VEGF) and FGF - in controlling the directional migration of PMCs and the formation of the embryonic skeleton (Duloquin et al., 2007; Röttinger et al., 2008). Duloquin et al. (Duloquin et al., 2007) showed that a VEGF receptor (VEGFR-10-Ig) is expressed selectively by migrating PMCs in the sea urchin Paracentrotus lividus, and a VEGF ligand (VEGF3) is expressed in the ectoderm overlying the VLCs. Perturbation of VEGF signaling in this species led to defects in PMC migration and skeletogenesis. Röttinger et al. (Röttinger et al., 2008) identified a similar, complementary pattern of expression of an FGF ligand and receptor (FGFA and FGFR2, respectively) in P. lividus and showed that blocking the FGF pathway also led to defects in skeletogenesis. These studies suggest that FGF and VEGF might have essential (and non-redundant) functions in mesoderm cell migration and skeletogenesis in the sea urchin, and raise the question of whether these two pathways regulate the same, or different, cell behaviors during PMC morphogenesis.

In this study, we aimed to further elucidate the roles of VEGF and FGF signaling in gastrulation and skeletogenesis in the sea urchin. We used two-color, fluorescent in situ hybridization to show that, in Lytechinus variegatus, fgfa and vegf3 are expressed in distinct domains in the ectoderm. Knockdown of Lv-vegf3 caused striking defects in the migration and differentiation of PMCs, whereas PMC migration was unperturbed in Lv-fgfa morphants, although these formed truncated skeletal elements. Using a second-generation VEGFR inhibitor, axitinib, we found that VEGF signaling is essential for all phases of PMC migration. Importantly, we identified a role for VEGF signaling in biomineralization that was separable from its effects on PMC migration, and showed that VEGF regulates the expression of many biomineralization genes in the PMC gene regulatory network (GRN). This study expands our knowledge of the regulation of mesoderm migration and differentiation by growth factor signaling pathways during gastrulation.

Adult Lytechinus variegatus were obtained from the Duke University Marine Laboratory (Beaufort, NC, USA) or Reeftopia (Key West, FL, USA). Adult Strongylocentrotus purpuratus were obtained from Pat Leahy (California Institute of Technology, Pasadena, CA, USA). Spawning was induced by intracoelomic injection of 0.5 M KCl and embryos were cultured in artificial seawater (ASW) at 23°C (L. variegatus) or 15°C (S. purpuratus).

Partial Lv-fgfa and Lv-vegf3 sequences were obtained by PCR using degenerate primers (supplementary material Table S4). Remaining Lv-vegf3 and 3′ Lv-fgfa sequences were obtained by RACE using L. variegatus late gastrula cDNA, the GeneRacer Kit with SuperScript III reverse transcriptase, and the Zero Blunt TOPO PCR Cloning Kit for Sequencing (Invitrogen). 5′ Lv-fgfa sequence was obtained by screening an L. variegatus mid-gastrula cDNA library (a gift from Dr David McClay, Duke University, NC, USA) using partial sequence obtained from degenerate PCR as a probe. GenBank accession numbers for Lv-fgfa and Lv-vegf3 are KF285451 and KF285452, respectively. Comparisons of the sequences of L. variegatus, S. purpuratus and Paracentrotus lividus VEGF3 and FGFA were carried out using ClustalW (Thompson et al., 1994). A fragment of Lv-pax2/5/8 and the complete Lv-otp coding region were obtained by PCR and cloned into the pCS2+ vector using the EcoRI and XbaI restriction sites.

Embryos were fixed (Adomako-Ankomah and Ettensohn, 2011) and WMISH and fluorescent WMISH (F-WMISH) were carried out (Lepage et al., 1992; Duloquin et al., 2007; Sharma and Ettensohn, 2010) as described.

Microinjections were carried out as described (Cheers and Ettensohn, 2004). Injection solutions contained 20% (vol/vol) glycerol and 0.16% (wt/vol) Texas Red dextran. MOs (supplementary material Table S5) were obtained from Gene Tools. Lv-fgfa splice-blocking MO was complementary to the exon 2/intron 2 boundary as described (Röttinger et al., 2008) and was injected at 2 mM and 4 mM. Lv-fgfa translation-blocking MO was injected at 4 mM and Lv-vegf3 and Sp-vegf3 translation-blocking MOs were injected at 2 mM.

RT-PCR analysis was performed as previously described (Adomako-Ankomah and Ettensohn, 2011). PCR products were cloned into the pCR4-TOPO vector (Invitrogen) for sequencing.

A 5 mM stock solution of axitinib (Selleckchem, Houston, TX, USA) was prepared in DMSO and stored at -20°C. Embryos were cultured in ASW containing axitinib at a final concentration of 75 nM (L. variegatus) or 50 nM (S. purpuratus). Control embryos were cultured in equivalent concentrations of DMSO.

Immunofluorescence using monoclonal antibody (mAb) 6a9 was carried out as described (Ettensohn and McClay, 1988). Immunostained embryos were examined using a Zeiss LSM 510 Meta/UV DuoScan inverted spectral confocal microscope.

The terminal dUTP nick end labeling (TUNEL) assay was carried out using the ApoAlert DNA Fragmentation Assay Kit (Clontech) as described (Voronina and Wessel, 2001). Modifications were introduced to label PMCs with the 6e10 mAb (Hodor et al., 2000), prior to nuclear labeling with 0.2 mg/ml Hoechst stain. Embryos were observed by confocal microscopy as described above.

To assess the ability of PMCs in axitinib- and DMSO-treated (control) embryos to target to the vegetal region of the embryo, PMCs were scattered throughout the blastocoel at the mesenchyme blastula stage (prior to the invagination of the archenteron) using a microneedle (Ettensohn and McClay, 1988). Embryos were cultured at 23°C for 5 hours, then fixed in 4% paraformaldehyde (1 hour), washed twice in ASW, and post-fixed in 100% methanol (20 minutes). The distribution of PMCs in the blastocoel was assayed by 6a9 immunostaining.

To further analyze PMC migration, PMCs were removed microsurgically from late mesenchyme blastula stage embryos and placed on fibronectin-coated coverslips as described (Malinda et al., 1995) in a medium of 2% horse serum in ASW. After allowing 30 minutes for attachment, PMCs from DMSO- or axitinib-treated embryos were imaged at 5-minute intervals for 2.5 hours using an Olympus DP71 digital camera. The velocity of migrating PMCs was measured using the Manual Tracking plug-in of ImageJ (NIH). Individual PMCs were tracked either throughout the experiment or until they became untraceable due to clustering.

To quantify filopodial numbers and lengths in axitinib- and DMSO-treated embryos, PMCs were scattered in the blastocoel at the mesenchyme blastula stage, and approximately half of the cells were flushed out of the blastocoel to reduce cell density. This made it possible to visualize unambiguously the filopodia of individual PMCs. Embryos were cultured at 23°C for 1.5 hours post-surgery, and the distribution and length of filopodia present were measured in 6a9-stained embryos using ImageJ.

PMC fusion was monitored by dye transfer (Hodor and Ettensohn, 2008). One set of fertilized eggs was injected with a solution of Lv-vegf3 MO that contained 10% (wt/vol) fluorescein dextran, while another set was injected with a solution of the same MO without the dextran. At the mesenchyme blastula stage, two to six PMCs were transferred from dextran-labeled donor embryos into each unlabeled host embryo. Embryos were observed by confocal microscopy 6 and 24 hours post-surgery to assess the distribution of the dextran within the PMC syncytium.

Changes in mRNA levels were assayed using the Nanostring nCounter analysis system (Nanostring Technologies, Seattle, WA, USA) (Geiss et al., 2008). For each experiment, RNA was extracted from 200 control and 200 axitinib-treated or VEGF morphant S. purpuratus embryos using the Nucleospin RNA II Kit (Clontech). The code set of probes corresponded to ∼80 genes in the PMC GRN (supplementary material Table S2).

To compare the roles of VEGF and FGF signaling, we cloned vegf3 and fgfa from L. variegatus, a North American species. ClustalW analysis (Thompson et al., 1994) of the amino acid sequences of VEGF3 and FGFA in L. variegatus, S. purpuratus and P. lividus revealed ∼70% and 80% identity, respectively, across species (supplementary material Fig. S1).

We performed two-color F-WMISH to compare directly the expression domains of Lv-fgfa and Lv-vegf3 at various developmental stages (Fig. 1). Lv-fgfa showed a highly dynamic pattern of expression whereas Lv-vegf3 expression was steadily maintained in ectodermal territories associated with PMC accumulation and skeletal growth. Both genes were first detectable at the hatched blastula stage (Fig. 1A-A′′), when Lv-fgfa was expressed in a ring of ectoderm that spanned the equator of the embryo and in the presumptive PMCs at the vegetal pole. Lv-vegf3 was also expressed in a ring in the ectoderm, but in a territory more vegetal to, and not overlapping with, the domain of Lv-fgfa expression. This separation of expression domains persisted at the mesenchyme blastula stage, although at this stage the expression of both genes was generally restricted to the ectoderm overlying the VLCs (Fig. 1B-B′′). As the archenteron began to invaginate, the expression domains of Lv-fgfa and Lv-vegf3 in the ectoderm overlapped slightly (Fig. 1C-C′′). This overlap persisted as the archenteron elongated (Fig. 1D-D′′), but was no longer evident at the late gastrula stage (Fig. 1E-E′′), when Lv-vegf3 continued to be expressed in the ectoderm overlying the VLCs (Fig. 1E′) while Lv-fgfa was downregulated in these domains and was expressed mostly in two apical domains of ectoderm and in PMCs (Fig. 1E). Lv-fgfa expression at the prism stage was largely restricted to the PMCs, whereas Lv-vegf3 expression persisted in the ectoderm adjacent to these cells (Fig. 1F-F′′). At the pluteus stage, Lv-vegf3 was expressed in the ectoderm at specific regions overlying the elongating postoral and anterolateral rods, while Lv-fgfa was expressed in PMCs associated with these rods and in the gut (Fig. 1G-G′′). Additionally, we observed that the cognate receptors Lv-vegfr-10-Ig and Lv-fgfr-2 were expressed selectively by PMCs in patterns identical to those reported in P. lividus (Duloquin et al., 2007; Röttinger et al., 2008) (data not shown).

To compare the functions of FGF and VEGF in L. variegatus, we knocked down the expression of each ligand. Skeletogenesis was blocked in embryos injected with 2 mM Lv-vegf3 translation-blocking MO (Fig. 2A-B′), a phenotype very similar to that observed in P. lividus (Duloquin et al., 2007). Likewise, skeletogenesis was eliminated in S. purpuratus embryos injected with 2 mM Sp-vegf3 translation-blocking MO (supplementary material Fig. S2B,B′). The Lv-vegf3 and Sp-vegf3 MOs differed at 8/25 nucleotide positions, demonstrating the specificity of the morphant phenotype. However, unlike results obtained in P. lividus (Röttinger et al., 2008), skeletogenesis was not eliminated in embryos injected with 2 mM splice-blocking Lv-fgfa MO, and these morphants formed extensive, although truncated, skeletons (Fig. 2C,C′). Embryos injected with as much as 4 mM Lv-fgfa MO also formed skeletal elements (Fig. 2D′), although these embryos showed toxicity effects due to the high MO concentration (Fig. 2D).

Following the same splice-blocking strategy used by Röttinger et al. (Röttinger et al., 2008), we designed the Lv-fgfa MO to overlap the exon 2-intron 2 boundary, leading to the exclusion of exon 2, which contains the highly conserved FGF domain (Fig. 2E). We demonstrated the effectiveness of the MO by RT-PCR analysis of 200 control embryos and 200 embryos injected with 1, 2 or 4 mM Lv-fgfa MO and observed the expected shift in the size of the Lv-fgfa transcript corresponding to an exclusion of exon 2 in embryos injected with all concentrations of Lv-fgfa MO (Fig. 2F, bright bands). At the MO concentrations used in our subsequent analysis (2 mM), we did not detect normally spliced Lv-fgfa mRNA by RT-PCR. The structure of the mis-spliced form of the mRNA lacking exon 2 was confirmed by sequencing. Embryos injected with an Lv-fgfa translation-blocking MO at a range of concentrations likewise formed bilaterally symmetrical, truncated skeletal elements (supplementary material Fig. S3B,D).

We examined the distribution of PMCs in Lv-vegf3 and Lv-fgfa morphants by 6a9 immunostaining. At the late gastrula stage, PMCs in control embryos were arranged in a subequatorial ring with VLCs (Fig. 3A, arrow). PMCs in Lv-vegf3 morphants did not form VLCs (Fig. 3B,B′), although most of the cells remained localized in the vegetal hemisphere. By contrast, PMCs in Lv-fgfa morphants formed a well-patterned ring comparable to that of control embryos (Fig. 3C,C′). At the pluteus stage, PMCs in control embryos were aligned along a well-developed skeleton (Fig. 3D,D′), whereas PMCs in Lv-vegf3 morphants were scattered primarily within the vegetal region of the embryo (Fig. 3E,E′). Strikingly, in Lv-vegf3 morphants, PMCs were highly fragmented at the pluteus stage (Fig. 3E, inset), but not at earlier stages (Fig. 3B, inset). We showed by a TUNEL assay that this fragmentation was not due to apoptosis (supplementary material Fig. S4). In Lv-fgfa morphants at the pluteus stage, PMCs were arranged in a pattern similar to that observed in control embryos, although defects in the morphology of the skeleton were apparent (Fig. 3F,F′).

We tested whether signaling through the VEGF and FGF pathways regulated the expression of Lv-vegf3 or Lv-fgfa. Lv-vegf3 expression was strongly upregulated in Lv-vegf3 morphants (Fig. 4A,B). We also observed a striking downregulation of Lv-fgfa in PMCs, from the mesenchyme blastula stage throughout later development (Fig. 4C,D). Lv-fgfa was expressed in an expanded territory in the ectoderm at the late gastrula stage in Lv-vegf3 morphants (Fig. 4D), although this expression domain narrowed to a pattern comparable to that of controls by the prism stage (data not shown). Upon Lv-fgfa knockdown, Lv-vegf3 was slightly upregulated, an effect that was more pronounced at mesenchyme blastula (Fig. 4E) than at late gastrula (Fig. 4F) stages. Furthermore, as previously described in P. lividus (Röttinger et al., 2008), Lv-fgfa expression was strongly upregulated in the PMCs and ectoderm of Lv-fgfa morphants (Fig. 4G,H). The increase in Lv-vegf3 and Lv-fgfa mRNA levels in the corresponding morphants is likely to reflect negative feedback of each pathway on transcription, although we cannot exclude the possibility that mRNA stability is affected (directly or indirectly) by the MOs.

We also examined two other genes expressed in the ectoderm that have been shown to play a role in regulating skeletogenesis: otp (Di Bernardo et al., 1999; Cavalieri et al., 2003) and pax2/5/8 (Cavalieri et al., 2011). Lv-pax2/5/8 expression was slightly downregulated in Lv-vegf3 morphants (supplementary material Fig. S5A,A′) and strongly downregulated in Lv-fgfa morphants (supplementary material Fig. S5C,C′). Lv-otp was unaffected by the knockdown of either Lv-vegf3 or Lv-fgfa (supplementary material Fig. S5B,B′,D,D′). As Lv-fgfa knockdown produced only modest defects in skeletogenesis and had no detectable effect on PMC migration, we focused our attention on further elucidating the role of VEGF signaling in PMC morphogenesis.

VEGF MO injection blocks VEGF signaling throughout embryogenesis. To determine the temporal requirements for VEGF signaling, we used a second-generation VEGFR inhibitor, axitinib, which specifically blocks VEGF signaling at low (nM) concentrations (Hu-Lowe et al., 2008; Bhargava and Robinson, 2011). L. variegatus embryos cultured in 75 nM axitinib from early cleavage developed no skeletal elements (Fig. 5B,B′), a phenotype identical to that of VEGF morphants (Fig. 2B). As in Lv-vegf3 morphants, PMCs in axitinib-treated embryos did not form VLCs (Fig. 5D,D′). At the pluteus stage, PMCs in axitinib-treated embryos exhibited the same fragmented phenotype observed in Lv-vegf3 morphants (Fig. 5F). Similarly, PMC migration and skeletogenesis were inhibited in S. purpuratus embryos cultured in 50 nM axitinib from early cleavage (supplementary material Fig. S2C,C′,F,F′). These findings confirmed that axitinib phenocopies VEGF knockdown.

PMC migration can be considered to occur in two phases. During the initial phase (early gastrulation), PMCs disperse from the site of ingression and arrange themselves in the subequatorial ring. During the second phase (late gastrulation), a chain of PMCs migrates from each VLC towards the animal pole. These cells produce the dorsoventral, anterolateral, and recurrent rods of the larval skeleton. To test whether VEGF signaling is required for the second phase of PMC migration, we cultured embryos in axitinib from the early gastrula stage, when PMCs were in the process of forming the subequatorial ring (Fig. 6A,A′). We fixed control and axitinib-treated embryos at the late gastrula stage and assessed the distribution of PMCs by immunostaining. PMCs in DMSO-treated control embryos migrated normally and formed two chains of cells that extended from the VLCs towards the animal pole (Fig. 6C,C′). No strands of PMCs migrated towards the animal pole in axitinib-treated embryos, however, despite the fact that VLCs were visible in 80% of embryos (Fig. 6B,B′). Based on these observations, we conclude that VEGF signaling is required for both the early and late phases of PMC migration.

Upon Lv-vegf3 knockdown or VEGFR inhibition by axitinib treatment, PMCs did not disperse randomly through the blastocoel but remained mostly in the vegetal hemisphere. To examine whether this behavior might be the result of VEGF-independent guidance cues in the blastocoel, we displaced PMCs throughout the blastocoel in control (Fig. 7A,A′) and axitinib-treated embryos at the mesenchyme blastula stage and examined the distribution of PMCs 5 hours later. In 28/30 cases, PMCs in control embryos returned to the vegetal region and formed a subequatorial ring, usually with two chains of cells extending towards the animal pole (Fig. 7C,C′). By contrast, in 30/30 axitinib-treated embryos, many PMCs remained in the animal half of the embryo (Fig. 7B,B′). Comparable results were obtained when PMCs in Lv-vegf3 morphants were scattered in the same manner (data not shown). Aggregates of PMCs remained scattered throughout the blastocoel of axitinib-treated embryos even 10 hours post-surgery (Fig. 7D,D′), whereas extensive, well-patterned skeletal elements were visible in control embryos (Fig. 7E,E′). We therefore found no evidence of a VEGF-independent mechanism that targets PMCs to the vegetal hemisphere.

In axitinib-treated embryos and Lv-vegf3 morphants, PMCs dispersed from the site of ingression and migrated along the basal surface of the ectoderm, indicating that cell motility per se was not blocked. We also used an in vitro assay to examine PMC motility (Fig. 8). Control PMCs migrated actively on fibronectin-coated coverslips and formed aggregates, as did PMCs that were isolated from axitinib-treated embryos and cultured continuously in the presence of the inhibitor (Fig. 8A). Axitinib treatment resulted in a very small, statistically insignificant decrease in average PMC velocity (P=0.1298) (Fig. 8B). These results support the view that VEGF signaling has a specific role in PMC guidance but is not required for PMC motility.

Because PMC pathfinding is likely to be mediated by filopodia, we analyzed the effect of VEGF signaling on the number and length of filopodia. PMCs in control embryos showed a stereotypical elongated morphology with filopodia that extended mostly from the two ends of each cell (Fig. 9A). PMCs in axitinib-treated embryos, however, were more rounded and had fewer, shorter filopodia (Fig. 9B). PMCs in control embryos (n=57) had ∼5 filopodia per cell, of an average length of 11.5 μm. By contrast, PMCs in axitinib-treated embryos (n=63) had ∼2 filopodia per cell, of an average length of 4.8 μm (Fig. 9C,D). These differences between control and axitinib-treated embryos were highly significant (P=7.68×10^-8 and P=5.15×10^-29 for filopodia number and length, respectively). These findings strongly suggest that VEGF signaling regulates aspects of filopodial dynamics, i.e. the formation of these cell protrusions and/or their stability.

Filopodia also mediate PMC-PMC fusion (Hodor and Ettensohn, 2008). We used a dye-transfer assay to test whether PMC fusion was blocked in Lv-vegf3 morphants. In 20/21 morphant embryos, dextran spread from a small number of transplanted, labeled PMCs to all host PMCs by the late gastrula stage (Fig. 10A,A′), demonstrating that PMC fusion occurred. At the pluteus stage, the dextran label was still visible in the fragmented PMCs of Lv-vegf3 morphant embryos (Fig. 10B,B′).

Why do PMCs fail to produce skeletal elements when VEGF signaling is blocked? One possibility is that VEGF regulates biomineralization directly, by regulating the expression of biomineralization genes. Alternatively, VEGF might regulate biomineralization indirectly, by affecting the ability of PMCs to migrate directionally and accumulate in the VLCs, where local ectodermal cues other than VEGF might induce biomineral secretion. To distinguish between these possibilities, we treated embryo cultures with axitinib at various stages of development (Fig. 11). The frequencies of morphant phenotypes that we observed upon drug addition at each stage of development are shown in Table 1. Embryos treated with axitinib beginning at the hatched blastula (Fig. 11B,B′), mesenchyme blastula (Fig. 11C,C′) or early gastrula (Fig. 11D,D′) stages developed no skeletal elements when control embryos reached the pluteus stage (Fig. 11A,A′). Embryos treated with axitinib from the mid-gastrula (Fig. 10E,E′), late gastrula (Fig. 11F,F′) or prism (Fig. 11G,G′) stages developed truncated skeletal elements.

Surprisingly, we noted that in embryos cultured in axitinib from the mid- or late gastrula stages, specific skeletal rods were affected to varying degrees. The body rods and recurrent rods extended normally in the dorsal region of the embryo (Fig. 11E′,F′, arrows, Fig. 12) whereas the postoral and anterolateral rods were severely truncated (Fig. 12). The postoral and anterolateral rods grow via the activity of a plug of PMCs at their tips; these PMC clusters were apparent in axitinib-treated embryos, although the rods failed to elongate normally (supplementary material Fig. S6).

We also examined the reversibility of the effects of VEGF inhibition on skeletogenesis. Embryos were cultured in axitinib from early cleavage and the inhibitor was washed out at various stages of development (supplementary material Fig. S7). Most embryos recovered fully from the effects of blocking VEGF signaling if axitinib was removed prior to the end of gastrulation (supplementary material Fig. S7B′-E′, Table S1), whereas embryos cultured in axitinib post-gastrulation had limited, if any, ability to form skeletal elements after removal of the drug (supplementary material Fig. S7F-H′, Table S1).

Because VEGF signaling regulates biomineralization directly (Figs 11, 12), we tested the role of this pathway in regulating the expression of genes in the PMC GRN. S. purpuratus embryos were used because the availability of a sequenced genome facilitated the design of a Nanostring probe set (supplementary material Table S2). Blocking VEGF signaling in S. purpuratus either with an Sp-vegf3 translation-blocking MO or by culturing embryos in 50 nM axitinib from early cleavage perturbed PMC migration and differentiation, as in L. variegatus (supplementary material Fig. S2). Differences in mRNA levels between control embryos and Sp-vegf3 morphants or embryos cultured in axitinib from early cleavage were measured at the late gastrula stage (48 hours postfertilization) (supplementary material Table S3). Three independent trials were conducted for both Sp-vegf3 morphants and axitinib-treated embryos.

We found that 19/90 genes (21%) showed a change in expression of greater than 50% in Sp-vegf3 morphants, the great majority of which were effector genes with proven or predicted roles in biomineralization (Fig. 13A; supplementary material Table S3). Similar results were obtained with axitinib-treated embryos, in which 21/90 genes (23%) showed a change of 50% or more (Fig. 13B; supplementary material Table S3). Of all genes that showed a change in expression of 50% or more, 14 were common to both the MO and axitinib-treated samples (74% and 67% of the affected genes, respectively). Most regulatory genes showed little or no change in both sample sets, an observation which is consistent with the fact that VEGF signaling does not regulate PMC specification.

Growth factors play a crucial role in regulating mesoderm morphogenesis during gastrulation, but the specific growth factors involved and the cell behaviors they control vary across species. For example, in the chick embryo, mesoderm migration is regulated by a combination of signals (PDGF, FGF and VEGF) (Chuai and Weijer, 2009), whereas our results indicate that in the sea urchin [which lacks PDGF and PDGF receptor genes (Lapraz et al., 2006)] PMC migration is controlled predominantly by one growth factor (VEGF3). In several organisms, it has been shown that a single growth factor influences multiple cell behaviors, although the details vary. During Xenopus gastrulation, PDGF regulates the orientation, radial intercalation and directional migration of mesodermal cells (Nagel et al., 2004; Damm and Winklbauer, 2011). PDGF signaling also regulates cell polarization and the formation of cellular processes by mesendodermal cells in zebrafish, but apparently not their directional migration (Montero et al., 2003). In Drosophila, FGF regulates several mesoderm cell behaviors, including EMT, intercalation, spreading, and attachment to the epithelium prior to migration (Wilson and Leptin, 2000; Winklbauer and Müller, 2011). In sea urchins, VEGF3 regulates the formation of cellular processes and directional migration. In contrast to mesodermal cells in Drosophila and Xenopus, however, the ingression and spreading of PMCs is independent of growth factor-mediated signaling. In addition, VEGF signaling is not required for PMC translocation or fusion, but only for guidance. We hypothesize that the effect of VEGF3 on the number and length of PMC filopodia is associated in some way with the defect in pathfinding, although the relationship is unclear.

In vertebrate endothelial cells, VEGF regulates filopodial extension and acts a chemoattractant to direct cell migration (Gerhardt et al., 2003; Gerhardt, 2008; Shamloo et al., 2008). These findings, and the observation that PMCs accumulate at ectopic sites of VEGF expression (Duloquin et al., 2007), indicate that VEGF3 functions as a chemoattractant in the sea urchin. Our finding that Lv-vegf3 expression is initially confined to a ring of ectoderm at the vegetal pole of the blastula and subsequently resolves to the ectoderm overlying the VLCs might account for the observation that vegetal directional cues are established even before PMC ingression (Ettensohn and McClay, 1986) and for the gradual emergence of the two PMC clusters during gastrulation (Malinda et al., 1995; Gustafson and Wolpert, 1999). Our findings further suggest that VEGF3 signaling is required for the second phase of PMC migration, during which chains of PMCs extend from the VLCs towards the animal pole. At present, owing to embryo to embryo variability in the timing of VEGF3 expression in the animal hemisphere during late gastrulation, we do not know whether a VEGF3 concentration gradient directs this phase of PMC migration or whether VEGF3 signaling at the VLCs renders the cells competent to respond to other directional cues.

Although VEGF signaling is required for biomineral formation (Duloquin et al., 2007), it has been unclear whether this is a secondary effect of disrupting the normal pattern of PMC migration, which might prevent PMCs from receiving a different ectodermal signal. In this regard, the ectoderm overlying sites of skeletal rod growth is a local source of signaling factors other than VEGF3, including FGFA and WNT5. Using axitinib to inhibit VEGF signaling late in development, after PMC migration is largely complete, we have demonstrated an effect of VEGF signaling on biomineralization that is separable from its role in PMC patterning. It was shown previously that photoablation of the ectoderm overlying the postoral rods inhibits the growth of these rods, pointing to local, ectoderm-derived cues that regulate skeletogenesis at late embryonic stages (Ettensohn and Malinda, 1993). At the pluteus stage, vegf3 and vegfr-10-Ig are expressed in the PMCs and ectoderm, respectively, at the tips of the elongating postoral and anterolateral rods (Fig. 1) (Duloquin et al., 2007). We have observed a striking selective inhibition of the elongation of exactly these rods when VEGF signaling is blocked late in development (Fig. 12). VEGF is not required, however, for the elongation of skeletal rods that form in the aboral region of the embryo.

The formation of biomineral is a complex, regulated process. It is possible that VEGF3 controls a feature of PMC motility and/or secretion that is required in the cell biology of biomineralization. Our findings, however, point to a more direct mechanism. PMC specification and differentiation are regulated by an elaborate transcriptional network (Rafiq et al., 2012). VEGF signaling controls the expression of numerous spicule matrix genes and other major protein components of the spicule (Fig. 13; supplementary material Table S3). The onset of expression of some of these genes precedes the expression of vegfr-10-Ig in the PMCs (Illies et al., 2002; Duloquin et al., 2007; Rafiq et al., 2012) and the expression of most genes was not fully blocked by inhibiting VEGF3, suggesting that VEGF signaling is not required for their initial activation but to maintain their expression. We favor the view that an initial, cell-autonomous phase of GRN deployment in the PMC lineage gives way to a later, signal-dependent phase during which local VEGF3 maintains the expression of a subset of biomineralization genes in specific regions of the PMC syncytium at levels compatible with skeletal growth. It is noteworthy that although the great majority of genes regulated by VEGF3 signaling have predicted or proven roles in biomineralization, not all genes encoding biomineralization proteins are regulated by VEGF signaling. For example, p16 and members of the msp130 gene family were not sensitive to VEGF signaling. Further analysis of the transcriptional regulatory control of these genes will be necessary to account for their differential responses to VEGF3.

Knapp et al. (Knapp et al., 2012) recently demonstrated concentration-dependent effects of recombinant VEGF3 on skeletal growth by cultured PMCs. Relatively high levels of VEGF3 favored growth in the a-axis (the formation of triradiate spicules), whereas lower levels of VEGF3 favored growth in the c-axis (the formation of linear rods that extend perpendicular to the plane of the triradiate spicule). These observations might be relevant to in vivo skeletogenesis, as vegf3 mRNA appears to be expressed at higher levels during gastrulation, when the triradiate spicules are synthesized, whereas levels are lower at the prism and pluteus stages, when other skeletal rods (e.g. the postoral and anterolateral rods) extend in the direction of the c-axis.

Our data and those of Röttinger et al. (Röttinger et al., 2008) point to negative feedback mechanisms that regulate fgfa and vegf3 in the ectoderm. We confirm that MO knockdown of FGFA enhances the expression of fgfa in the ectoderm and show a similar dependence of vegf3 expression on VEGF3. Because the only known embryonically expressed VEGF receptor, VEGFR-10-Ig, is restricted to PMCs, the effect of VEGF3 on vegf3 expression in the ectoderm is likely to be indirect. We report that the expression of fgfa in PMCs (but not in the ectoderm) is dependent on VEGF3, an effect that might be mediated directly by VEGFR-10-Ig. This provides evidence of cross-talk between these two pathways, which function synergistically in other systems (Tomanek et al., 2010). Cavalieri et al. (Cavalieri et al., 2011) showed that the tripartite motif-containing protein STRIM1 regulates the expression of fgfa in the ectoderm, but not in PMCs, highlighting the separate control of fgfa expression in these two tissues. STRIM1 also regulates otp in the ectoderm whereas VEGF3 and FGFA do not, suggesting that multiple pathways regulate gene expression in the ectoderm overlying the VLCs.

Our studies and those of Duloquin et al. (Duloquin et al., 2007) demonstrate a conserved role for VEGF3 in regulating PMC migration and differentiation in three species of sea urchin. Recent work suggests that evolutionary shifts in VEGF3 expression have been associated with changing patterns of skeletogenesis within the phylum (Morino et al., 2012). In contrast to the highly conserved function of VEGF3, the role of FGFA may vary among species, as this molecule has been implicated in PMC migration and skeletogenesis in P. lividus (Röttinger et al., 2008). In L. variegatus, we observed skeletal phenotypes similar to those reported by Röttinger and co-workers when we injected high concentrations of MO that also appeared to produce general toxic effects. At lower doses, all detectable FGF mRNA was mis-spliced (Fig. 2), but the overall morphology of the embryos was similar to that of controls, and we observed no effects on PMC migration and only subtle effects on skeletal growth. Thus, there might be technical issues that account for the reported differences in the phenotypes of FGF morphants in these two species, and the functions of both VEGF3 and FGFA might be conserved among echinoids.

The downstream components of the VEGF signaling pathway in PMCs are unknown. In other cell types, VEGF receptors signal via the MAPK and PI3 kinase pathways (Schlessinger, 2000), both of which are essential for skeletogenesis in the sea urchin embryo. Bradham et al. (Bradham et al., 2004) showed that inhibiting PI3K blocks the elongation of skeletal rods, but not PMC specification, cell migration, the extension of filopodia, cell fusion or skeletal initiation. The MAPK pathway, by contrast, regulates early PMC specification, thereby affecting all aspects of PMC morphogenesis (Fernandez-Serra et al., 2004; Röttinger et al., 2004). It has been suggested that MAPK signaling may regulate skeletogenesis by two distinct pathways: a Ras-independent pathway necessary for PMC ingression and a Ras-dependent pathway required for PMC migration and skeletogenesis (Fernandez-Serra et al., 2004; Röttinger et al., 2004). As the VEGF pathway does not regulate PMC ingression, it might function by a Ras/MAPK-dependent mechanism. Further research will be necessary to determine the molecular effectors of VEGF signaling in PMCs and the linkages between this pathway and the skeletogenic GRN.

We thank Kiran Rafiq for designing the Nanostring codeset.