Gene regulation of adult skeletogenesis in starfish and modifications during gene network co-option

The larval skeleton of the echinoderm is believed to have been acquired through co-option of a pre-existing gene regulatory network (GRN); that is, the mechanism for adult skeleton formation in the echinoderm was deployed in early embryogenesis during echinoderm diversification. To explore the evolutionary changes that occurred during co-option, we examined the mechanism for adult skeletogenesis using the starfish Patiria pectinifera. Expression patterns of skeletogenesis-related genes (vegf, vegfr, ets1/2, erg, alx1, ca1, and clect) suggest that adult skeletogenic cells develop from the posterior coelom after the start of feeding. Treatment with inhibitors and gene knockout using transcription activator-like effector nucleases (TALENs) suggest that the feeding-nutrient sensing pathway activates Vegf signaling via target of rapamycin (TOR) activity, leading to the activation of skeletogenic regulatory genes in starfish. In the larval skeletogenesis of sea urchins, the homeobox gene pmar1 activates skeletogenic regulatory genes, but in starfish, localized expression of the pmar1-related genes phbA and phbB was not detected during the adult skeleton formation stage. Based on these data, we provide a model for the adult skeletogenic GRN in the echinoderm and propose that the upstream regulatory system changed from the feeding-TOR-Vegf pathway to a homeobox gene-system during co-option of the skeletogenic GRN.


Results
Presumptive adult skeletogenic mesenchyme cells emerge around the posterior coelom in the starfish P. pectinifera. It is still unclear where adult skeletogenic cells are derived from in starfishes, including P. pectinifera, although later development in starfish has been well described 17,18 . Therefore, we observed adult skeleton formation until 7 days postfertilization (dpf), when skeletal rudiments were first observed on the dorsal and left sides of the stomach. At 2 dpf, bipinnaria larvae that started feeding had two types of coelomic pouches: a posterior enterocoel (PE, arrowhead in Fig. 1a) and bilateral coelomic pouches around the pharynx (arrows in Fig. 1a). By 3 dpf, the bilateral coelomic pouches had extended posteriorly, and the left coelomic pouch was attached to the PE (Fig. 1b,  b'; left coelomic pouch and PE are shown in blue with dotted lines in Fig. 1b'). By 5 dpf, the posteriorly extended left coelomic pouch have fused with PE to form the left posterior coelom, and the upper part of the bilateral coelomic pouches had fused around the mouth (arrowhead in Fig. 1c), while the mesenchyme cell population had emerged on the dorsal and left sides of larvae, i.e., near the left posterior coelom (dotted circle in Fig. 1c; enlarged image is shown in Fig. 1d). At approximately 7 dpf, the larvae had developed into brachiolaria larvae (Fig. 1e). A few adult skeletal rudiments first emerged on the dorsal and left sides of the larvae (arrowheads in Fig. 1f); skeletal fragments were also formed on the right side later.
Vegfr, ets1/2, erg, and alx1 are expressed in the posterior coelom in starfish. We next investigated the expression of skeletogenesis-related genes (vegf, vegfr, ets1/2 [an ortholog of sea urchin ets1], erg, alx1, ca1 [carbonic anhydrase 1], and clect [c-lectin]) at 3, 5, and 7 dpf by whole-mount in situ hybridization (WMISH) (Fig. 2). Although we previously showed the expression of some of these genes 7,15,16 , to clarify the temporal expression profile, we reexamined gene expression patterns in a single batch because the temporal expression  www.nature.com/scientificreports/ patterns appeared to vary slightly among batches, possibly due to differences in feeding conditions or genetic backgrounds.
Feeding-nutrient sensing pathway regulates adult skeletogenesis in starfish. The feedingnutrient sensing pathway is considered to be an additional indispensable factor for the formation of adult rudiments in indirectly developing echinoderms. In the indirectly developing starfish, the larval stage at which adult rudiments form cannot be observed without feeding 19,20 . In addition, a previous study on sea urchins suggests that adult rudiment formation requires the activity of target of rapamycin (TOR) 21 , which has evolutionarily conserved functions in eukaryotic cell growth and metabolism with external environmental signals, including nutrients 22 . Thus, we observed skeleton formation in larvae with no feeding and those treated with an inhibitor of the TOR signaling pathway, rapamycin (Fig. 3). The starfish larvae were treated with rapamycin from 2 dpf until 7 dpf, and the treated larvae were fed from 2 dpf. At 7 dpf, both nonfeeding and rapamycin-treated larvae showed smaller body sizes than control larvae (intact larvae with feeding and DMSO-treated larvae, respectively) ( Fig. 3a1-c1). Most control starfish larvae formed skeleton rudiments in each experiment at 7 dpf (arrowheads in Fig. 3a2), but almost none of the nonfeeding or rapamycin-treated larvae formed skeletons at 7 dpf (Fig. 3b2,c2,b4,c4). Notably, among nonfeeding larvae, a small proportion of larvae formed adult skeletal rudiments, but all of the larvae with skeletons were derived from one batch (13 of 29 larvae in this batch) among the five batches examined. Aggregation of mesenchyme cells on the posterior-dorsal side was observed in the control larvae (dotted circle in Fig. 3a2) but not in the other experimental larvae (Fig. 3b2,c2,b5,c5). Nonetheless, similar to the control larvae, the nonfeeding and rapamycin-treated larvae had a certain number of mesenchyme cells expressing the mesenchyme cell marker MC5 23 ( Fig. 3a3-c3).
We next examined the expression patterns of vegf, vegfr, ets1/2, erg, alx1, ca1 and clect in the nonfeeding larvae and larvae treated with rapamycin ( Fig. 4). At 7 dpf, the expression of all genes other than ets1/2 was not detected in the nonfeeding larvae (Fig. 4b1,b2,b4-b7), unlike in the control larvae (Fig. 4a1,a2,a4-a7), suggesting that feeding is essential for the expression of these genes. In contrast, ets1/2 expression was detected in the posterior coeloms in approximately half of larvae (Fig. 4b3). In most of the rapamycin-treated larvae, no expression was detected of any of the genes examined ( Fig. 4c1-c7), suggesting that TOR signaling is required for the expression of skeletogenesis-related genes, including ets1/2.
Vegf signaling plays indispensable roles in adult skeletogenesis of starfish. We also examined the role of Vegf signaling in adult skeleton formation using the Vegfr inhibitor axitinib (Fig. 3d1-d5). It has been demonstrated that axitinib inhibits larval skeleton formation in sea urchins 24 . We treated starfish larvae with axitinib from 2 dpf. By 7 dpf, the axitinib-treated larvae were similar in size to the control DMSO-treated larvae ( Fig. 3a1 and d1). The axitinib-treated larvae showed no mesenchyme cell aggregation or skeletal fragments in the posterior-dorsal region (Fig. 3d2,d4,d5), but MC5-positive mesenchyme cells were still observed throughout the larvae, like in the control larvae ( Fig. 3a3 and d3).
The expression of vegfr, ca1, and clect was not detected by WMISH in any larvae treated with axitinib ( Fig. 4d2,d6,d7). Some axitinib-treated larvae showed erg and alx1 expression (11 of 18 larvae for erg and two of nine larvae for alx1), but the signal level in axitinib-treated larvae was much lower than that in control larvae ( Fig. 4d4 and d5 compared to a4 and a5). In contrast, no obvious difference was observed in vegf or ets1/2 expression between control and axitinib-treated larvae ( Fig. 4d1 and d3 compared to a1 and a3). These observations suggest that Vegf signaling regulates erg, alx1, ca1 and clect during the early phase of adult skeleton formation in starfish. We also suggest that Vegf signaling is required for the expression of vegfr itself, as demonstrated in the sea urchin embryo 25 .
To further confirm the requirement of Vegf signaling in adult skeleton formation in starfish, TALEN-mediated knockout of vegfr was performed. We designed TALENs targeting the Vegfr exon region encoding its tyrosine kinase domain, which is a core domain for intercellular signal transduction 26 , and injected mRNA encoding the right and left arms of the TALENs into eggs of P. pectinifera.
To verify the genomic cleavage of the TALEN target site, we first cloned and sequenced the genomic fragments including the target site using genomes extracted from two larvae from each experimental group at 2 dpf (see Supplementary We cultured the rest of the experimental larvae until 7 dpf. Most of the control larvae formed adult skeletons (39 of 46 larvae, Fig. 5a; arrowheads in Fig. 5b), while the proportion of larvae with skeletons was substantially lower in the vegfr-knockout larvae (three of 31 larvae, Fig. 5g,h). In vegfr-knockout larvae, mesenchyme cells were observed, as in the control larvae, but aggregation of these cells around the posterior coeloms was observed www.nature.com/scientificreports/ in fewer knockout larvae (three of 31 larvae) than control larvae (40 of 46 larvae). No other differences were observed in larval morphology between vegfr-knockout and control larvae at 7 dpf. We investigated the expression patterns of some putative downstream genes of Vegf signaling, including alx1, ets1/2, ca1 and clect ( Fig. 5c-f,i-l), and found similar effects on gene expression patterns to those of axitinib treatment. Five of six control larvae showed alx1 expression in the posterior coelom and clusters of skeletogenic mesenchyme cells (Fig. 5c), whereas two of eight vegfr-knockout larvae showed a positive WMISH signal of alx1. Note that the expression was detected in much a smaller number of cells (Fig. 5i). This incomplete suppression of alx1 expression in vegfr-knockout larvae probably reflected mosaic effect of TALEN-mediated knockout. All of the control and vegfr-knockout larvae (five larvae each) showed ets1/2 expression in posterior coeloms and/or mesenchyme cells, but the expression level was lower in the knockout larvae (Fig. 5d,j). Ca1 and clect were also expressed in mesenchyme cell clusters around posterior coeloms in most of the control larvae (nine of 11 larvae for ca1; all six larvae for clect; Fig. 5e,f), while these expression levels were substantially lower in vegfr-knockout larvae (none of seven larvae for ca1; three of five larvae retained expression of clect, but in much a smaller number of cells, Fig. 5k,l). The above phenotypic effects of TALEN-mediated vegfr knockout were quite similar Observations of skeleton formation in the experimental starfish larvae. Morphology and skeleton formation were observed in control larvae (a1-a3), nonfeeding larvae (b1-b5), rapamycin (rapa)-treated larvae (c1-c5), and axitinib (axi)-treated larvae (d1-d5) at 7 dpf. (a1-d1, a2-d2) Living larvae. In the panel a2, the arrowhead indicates an adult skeletal rudiment and the dotted circle indicates aggregated mesenchyme cells. (a3-d3) Fluorescence images of larvae examined by immunohistochemistry using the mesenchyme-specific marker MC5. The green signal shows MC5 expression, whereas the red signal shows Chaetoceros calcitrans in the stomach or autofluorescence. At 7 pdf, the ratio of larvae with adult skeletons (b4-d4) and that of larvae with aggregated mesenchyme cells in the posterior dorsal region (b5-d5) were evaluated. Feeding larvae and DMSO-treated larvae were used as controls for nonfeeding larvae and inhibitor-treated larvae, respectively. Scale bars: 50 μm. M: mol/L. www.nature.com/scientificreports/  No localized expression of the phbA and phbB genes are observed during early adult skeletogenesis in starfish. In the sea urchin larval skeletogenic GRN, Pmar1 is one of the upstream key regulators, and we previously suggested that two Pmar1-related proteins, PhbA and PhbB, function as upstream regulators in the endomesoderm specification of starfish during embryogenesis 27 . To reveal whether these two Phb proteins control adult skeleton formation, we examined the expression of phbA and phbB until 7 dpf by WMISH and quantitative PCR (qPCR) (Fig. 6). For both phb genes, no obvious WMISH signals were detected at the coeloms, mesenchyme cells, or other cells from 3 to 7 dpf (Fig. 6a-h), although localized expression was detected during embryonic stages (Fig. 6a,e). We also performed qPCR to detect expression of phb genes. During embryonic stages, the mRNA level of both genes reached a peak at the midblastula stage (10 h postfertilization [hpf]) (Fig. 6i,j). By contrast, the expression levels of both phb genes were very low from 3 to 7 dpf (Fig. 6a,b); their expression levels during this period were 0.27-0.45% and 0.01-0.02% of their maximum expression levels at 10 hpf for phbA and phbB, respectively.
Vegf and TOR signaling are not required for expression of alx1 and ets1 in larval skeleton formation in the sea urchin. To estimate the evolutionary modifications in the GRN associated with cooption of skeletogenesis, we examined the effects of the Vegf and TOR signaling pathways on larval skeletogenesis in the sea urchin H. pulcherrimus, which belongs to the euechinoid group (Fig. 7). A previous study using another TOR inhibitor, PP242, in sea urchin embryos demonstrated that the TOR signaling pathway controls cyclin B mRNA translation during early cleavage stages 28 , but the role of TOR signaling in later development is still unknown. In sea urchin embryos, Vegf signaling is required for larval skeleton formation; however, this signaling pathway does not activate skeletogenic regulatory genes such as alx1. Zygotic expression of sea urchin vegf and vegfr starts later than the onset of alx1 or ets1 expression during embryonic development. Vegfr expression is regulated by alx1 and ets1 instead 25,29 .
In H. pulcherrimus embryos treated with rapamycin (500 nM or 2500 nM), no effects on larval skeleton formation were observed. The embryos were treated beginning at the 2-cell stage, and the cleavage of blastomeres tended to be delayed in the embryos treated with a high dose; however, primary mesenchyme cells (PMCs), www.nature.com/scientificreports/ which compose the skeletogenic cell population emerging before gastrulation, and larval skeletons were formed in almost all embryos (Fig. 7b1-b3,d).
We examined the expression of alx1, ets1, vegf and vegfr at the blastula and/or gastrula stages in rapamycintreated embryos by WMISH and found that, for all the genes, the expression levels were comparable to those in the control embryos at the blastula and gastrula stages (Fig. 7e1-e6,f1-f6). A portion of rapamycin-treated blastulae (five of 15 embryos) rather showed expansion of the alx1-expressing region (Fig. 7f1). Thus, we obtained no evidence that TOR controls Vegf signaling or regulatory genes such as alx1 and ets1 during sea urchin larval skeletogenesis.
Regarding Vegf signaling, we confirmed the previous results obtained in other sea urchins 24,25 using embryos of H. pulcherrimus; that is, axitinib inhibited the formation of larval skeletons (Fig. 7c2,c3,d) but not the formation of PMCs (Fig. 7c1). In the axitinib-treated embryos, expression level of vegfr mRNA was substantially www.nature.com/scientificreports/ lower than that in the control embryos (Fig. 7g6,e6), whereas the expression patterns of the other genes were not obviously affected at either stage ( Fig. 7g1-g5). This result is consistent with a previous study showing that alx1 and ets1 expression at the gastrula stage is not downregulated in axitinib-treated embryos 30 . These observations suggest that Vegf signaling is not required for the expression of alx1 and ets1 at the early and later embryonic stages in the sea urchin.

Discussion
Previous gene expression analyses of sea urchins and starfish suggested that certain larval skeletogenic GRN components identified in sea urchins, including Vegf signaling genes, are also associated with the formation of the adult skeleton in echinoderms 12,16,31 . In this study, we provide further supporting evidence and propose a model of the evolutionary change associated with the co-option of GRNs in the evolution of morphological novelties.

Development of adult skeletogenic cells in the starfish.
Our observations provide insight into the cell lineage of adult skeletogenic cells in starfish. In P. pectinifera, a mesenchyme cell population was seen on the dorsal side of the posterior coelom by approximately 5 dpf. When the expression of skeletogenesis-related genes was observed, vegfr, ets1/2 and erg were detected in some of the cells included in the left posterior coelom at 3 dpf, suggesting that some of the progenies develop into adult skeletogenic cells (Fig. 8a). As Yajima 32 demonstrated that, in euechinoids, secondary mesenchyme cells (SMCs), which are a non-larval skeletogenic mesenchyme population that emerges later than PMCs during embryogenesis, contribute to adult skeletogenesis, the timing of epithelial-mesenchymal transition of adult skeletogenic cells seems to be different between starfish and sea urchins, i.e., it occurs after or before the onset of feeding, respectively.
GRN model for adult skeletogenesis in the starfish. The hypothetical GRN model for adult skeletogenesis in starfish is shown in the left panel in Fig. 8b. Our observations suggest that adult skeleton formation in starfish requires feeding, TOR activity, and the Vegf signaling pathway, and the requirement of Vegf signaling was also confirmed by a TALEN-based knockout experiment. WMISH analysis in nonfeeding and rapamycintreated starfish larvae suggested that the TOR-mediated feeding-nutrient sensing pathway controls adult skeletogenesis upstream of the GRN. Together with the previous studies in larvae of C. elegans 33 and octopus 34 , our www.nature.com/scientificreports/ data suggests that the feeding-TOR pathway is conserved in bilaterians. Of note, in nonfeeding larvae, unlike in TOR-inhibited larvae, the expression of ets1/2 in the posterior coelom was detected in approximately half of the larvae. Therefore, it seems that the expression of ets1/2 is regulated by TOR pathway, but independently of feeding. On the other hand, inhibition of Vegf signaling caused the downregulation of genes other than vegf and ets1/2, indicating that Vegf signaling regulated by the TOR signaling pathway is required for the activation of at least some skeletogenic regulatory genes, such as alx1 and erg. Since clear downregulation of ets1/2 expression was not observed in Vegf signaling-inhibited larvae, we propose that activation of ets1/2 is independent of Vegf signaling.
One of the notable findings in the present study is that Vegf signaling is involved in different morphogenetic processes during larval and adult skeleton formation. During larval skeletogenesis in euechinoids, vegfr is expressed in skeletogenic mesenchyme cells (i.e., PMCs) after their ingression into the blastocoel, and it is required for mesenchyme cells to move to the correct position and form larval skeletons 25 . In contrast, starfish larvae treated with the Vegf signaling inhibitor showed few alx1-or erg-positive cells in the posterior coelom, suggesting that specification of adult skeletogenic cells were suppressed by the Vegf signaling inhibitor. In addition, Vegf signaling differentially affects one of the key skeletogenic regulatory genes, alx1, in sea urchin larval skeletogenesis and starfish adult skeletogenesis; Vegf controls alx1 expression only during adult skeletogenesis in starfish, while Vegf is dispensable for expression of sea urchin alx1 expression in the normal development. Based on these data, we propose the following adult skeletogenic process in starfish: the TOR-mediated feeding-nutrient sensing pathway activates vegf in the posterior ectoderm of the larvae, and Vegf signaling from the ectoderm to the posterior coelom activates regulatory genes such as alx1 and erg in the posterior coelom-mesenchyme lineage, resulting in the formation of adult skeletogenic mesenchyme cells. In contrast, ets1/2 appears to be regulated by the TOR pathway independent of Vegf signaling ( Fig. 8a; left panel in Fig. 8b).
Evolutionary modifications in the skeletogenic GRN during co-option. We compared the adult skeletogenic pathway in the starfish with the larval skeletogenic pathway in the sea urchin to assess the evolutionary changes in GRN associated with the co-option of skeletogenesis (Fig. 8b). A double-repression system consisting of two repressors, Pmar1 and HesC, a so-called double-negative gate, activates larval skeletogenic regulatory genes in euechinoid sea urchins 35 (Fig. 8b, right panel); this gate is not responsible for adult skeleton formation in sea urchins because pmar1 gene expression was not detected around the adult skeleton formation site in S. purpuratus 12 . In this study, localized expression of the pmar1-related genes phbA and phbB was not detected after the onset of feeding in starfish larvae, which supports the idea that these genes do not control adult skeletogenesis. Moreover, our previous study demonstrated that a double-repression system consisting of Pmar1 and an unknown repressor other than HesC activates alx1 and ets1/2 in cidaroid Prionocidaris baculosa, which is considered to possess ancestral characteristics of sea urchins 27 (Fig. 8b, center panel). Therefore, we propose that the upstream system was changed from the nutrient-Vegf signaling pathway to a double-negative gate with Pmar1 and an unknown repressor when the larval skeleton was acquired by co-option of the adult skeletogenic system. It should be noted that, based on the experiments using PMC-removed euechinoid embryos, the recent study also proposed that the regulation of alx1 by Vegf signaling is ancestral mode for skeletogenic cell formation in the echinoderm 36 .
Our data suggest that the hierarchy of Vegf signaling in the skeletogenic GRN differs between larval skeletogenesis and adult skeletogenesis, occurring either upstream or downstream of alx1, respectively, although Vegf signaling is still required for both types of skeleton formation. A recent study on the beetle suggests that many original components are not included in the co-opted GRN 37 , and similarly, it was also suggested that a few regulatory genes, such as the t-box gene tbr, are included in only the larval skeletogenic GRN and not the adult GRN 12 . These facts indicate the flexibility of GRNs in nature, and future studies will be needed to understand how novel characteristics evolved in such flexible GRNs.

Methods
Gametes and embryos. The collection and handing of gametes of P. pectinifera and H. pulcherrimus were performed according to Koga et al. 15 Embryos of these species were cultured in artificial seawater (MARINE ART BR, Osaka Yakken Co., Ltd.) at 22 °C and 14 °C, respectively. Starfish larvae were cultured according to the method described previously 38 and fed Chaetoceros calcitrans after 2 dpf. WMISH and immuno-staining. The fixation, hybridization and staining of larvae or embryos were performed as described previously 27 . The sequence of P. pectinifera erg (Pp-erg) was obtained using PCR with primers that corresponds the erg sequence in another starfish P. miniata, and the orthology was confirmed by phylogenetic analysis using RAxML 8.2.12 39 (see Supplementary Fig. S2 online). The P. pectinifera clect gene (Pp-clect) was obtained from the transcriptome assembly 38 -TGT TGT TGT TGC TCT CAT  TAT TGT T-3′; and Hp-vegfr-R, 5′-TTC CAT TCA ATG TCA TTA CTC TGT G-3′. RNA probes for the other genes were prepared as described previously 15,16,27 . The MC5 antibody was used for the visualization of mesenchyme cells in the starfish larvae; fixation and staining were performed according to the method of Hamanaka et al. 23 In the larvae examined using RNA probes of both phb genes, a WMISH signal was detected in the intestine of www.nature.com/scientificreports/ larvae (Fig. 6c,d,g,h), but we do not consider this signal to be phb-specific, as mentioned previously 38 , because it is frequently detected with other RNA probes in this species.
TALEN-mediated gene knockout. The sequence of vegfr was obtained from the transcriptome data of the starfish P. pectinifera 38 . We selected tyrosine kinase domain of vegfr as the target of TALENs, and the location of the domain was predicted by the NCBI Conserved Domain Search 40 (https:// www. ncbi. nlm. nih. gov/ Struc ture/ cdd/ wrpsb. cgi). The target sites of TALENs were finally determined using TAL Effector Nucleotide Targeter 2.0 (https:// tale-nt. cac. corne ll. edu/ node/ add/ talen). The designed TALEN pairs were constructed onto the TALEN backbone vector 41 by the Golden Gate method 42 using the Planinum TALEN kit 43 . We used the mMESSAGE mMACHINE T3 Transcription Kit (Invitrogen) for mRNA transcription. TALEN mRNA was introduced into eggs by microinjection according to the method of Saito et al. 44 We injected 1000 ng/µl mRNA each of the TALEN right and left arms into the unfertilized eggs of P. pectinifera to make vegfr-knockout embryos, and 2000 ng/µl mRNA of the right-arm TALEN was used as a control. To verify the cleavages of the target site, we extracted the genomes of two 2-dpf larvae from each experimental group using Nucleo Spin Tissue (MACHEREY-NAGEL), and amplified genomic fragments including the target site by PCR. We then cloned these fragments with pGEM-T Easy Vector System (Promega), and sequenced 5 and 18 clones from the genomes of control and vegfr-TALEN larvae, respectively. The sequences of the primers used were as follows: F1, 5′-ACC TGC CAT ATG ATC CTA AGT GGG AGT TCC -3′; F2, 5′-GAC AAG TCA CCG TGT TCA TAT TCA CAC TCA -3′; R1, 5′-GAG TTC AGT CAT CAG TGC TTT CCT CTC CAC -3′ and R2, 5′-TGA ACC TAC TGC GTC CTT CTA GTG AAG CTG -3′. In addition, we confirmed mCherry fluorescence at approximately 24 hpf, suggesting that the TALEN mRNAs were correctly translated. Larvae were cultured as described above until 7 dpf.

Data availability
The datasets supporting the conclusions of this article are included within the article and its Supplementary information file, and available on reasonable request. The sequences for Pp-erg and Pp-clect genes are shown in the Supplementary data online.