Paleogenomics of echinoids reveals an ancient origin for the double-negative specification of micromeres in sea urchins

Jeffrey R. Thompsona,1, Eric M. Erkenbrackb, Veronica F. Hinmanc, Brenna S. McCauleyc,d, Elizabeth Petsiosa, and David J. Bottjera

aDepartment of Earth Sciences, University of Southern California, Los Angeles, CA 90089; bDepartment of Ecology and Evolutionary Biology, Yale University, New Haven, CT 06511; cDepartment of Biological Sciences, Carnegie Mellon University, Pittsburgh, PA 15213; and dHuffington Center on Aging, Baylor College of Medicine, Houston, TX 77030

Edited by Douglas H. Erwin, Smithsonian National Museum of Natural History, Washington, DC, and accepted by Editorial Board Member Neil H. Shubin January 31, 2017 (received for review August 2, 2016) Establishing a timeline for the evolution of novelties is a common, methods thus provide a rigorous methodology in which to examine unifying goal at the intersection of evolutionary and developmental gene expression datasets, and ultimately animal body plan evolu- biology. Analyses of gene regulatory networks (GRNs) provide the tion (11), within the context of evolutionary time. After genomic ability to understand the underlying genetic and developmental novelties underlying differential body plan development have been mechanisms responsible for the origin of morphological structures identified, we can then consider the rates at which these novelties both in the development of an individual and across entire evolution- arise, and the rates at which GRNs evolve. Achieving this end ary lineages. Accurately dating GRN novelties, thereby establishing requires an explicit timeline in which to explore GRN evolution. a timeline for GRN evolution, is necessary to answer questions In a phylogenetically informed, comparative framework, it is about the rate at which GRNs and their subcircuits evolve, and to possible to infer where on a phylogenetic tree and when, in deep tie their evolution to paleoenvironmental and paleoecological time, GRN innovations are likely to have first arisen. Using a changes. Paleogenomics unites the fossil record and all aspects paleogenomic approach (12, 13), it is possible to incorporate deep of deep time, with modern genomics and developmental biology time into an analysis of when GRN novelties arose and to infer the to understand the evolution of genomes in evolutionary time. Recent work on the regulatory genomic basis of development in regulatory interactions directing the development of extinct organ- cidaroid echinoids, sand dollars, heart urchins, and other nonmodel isms, thereby bringing forth a unique understanding of the evolution echinoderms provides an ideal dataset with which to explore GRN of GRNs and the body plans that they encode. Paleogenomics al- evolution in a comparative framework. Using divergence time lows for the dating of the appearance of apomorphic GRNs, their estimation and ancestral state reconstructions, we have determined subcircuits, and particular network linkages using a combination of the age of the double-negative gate (DNG), the subcircuit which the fossil record, statistically derived divergence dates, and com- specifies micromeres and skeletogenic cells in Strongylocentrotus parative analyses of robust experimental data from extant organ- purpuratus. We have determined that the DNG has likely been used isms. With reliable dates in hand, the task of determining rates of for euechinoid echinoid micromere specification since at least the GRN evolution is not far off. We set out to establish a rigorous Late Triassic. The innovation of the DNG thus predates the burst of framework for determining the timeline of GRN evolution, and to post-Paleozoic echinoid morphological diversification that began in test a recently proposed hypothesis (8) concerning the timing of the the Early Jurassic. Paleogenomics has wide applicability for the in- evolution of a key sea urchin GRN novelty, the double-negative tegration of deep time and molecular developmental data, and has specification of micromeres. wide utility in rigorously establishing timelines for GRN evolution. Echinoids, or sea urchins, represent an ideal model system for understanding the mechanistic basis of GRNs in development and evolution | evo-devo | euechinoid | cidaroid | gene regulatory networks for studying the evolution of development (14–16). Research on the early embryonic development of echinoids has revealed the regula- he investigation of gene regulatory networks (GRNs) in modern tory interactions that compose the circuitry of developmental GRNs Ttaxa allows for the understanding of evolutionary changes in the driving early development of the purple sea urchin Strongylocentrotus regulatory genome that have underpinned the evolution of new purpuratus (14). Importantly, echinoids also have an excellent fossil – morphological structures in deep time (1 4). Establishing a timeline record that dates back to Ordovician strata, more than 400 Mya (17). for the rates at which these novel structures arise, and the rate at The combination of a robust fossil record and detailed under- which the developmental GRNs that encode them evolve, lies at the standing of the early developmental GRNs in numerous species heart of evolutionary developmental biology (5). In recent years, identifying genetic regulatory differences between diverse organisms

has become more feasible with broader phylogenetic sampling of This paper results from the Arthur M. Sackler Colloquium of the National Academy of developmental and gene expression data across Metazoa (6, 7). Sciences, “Gene Regulatory Networks and Network Models in Development and Evolu- These new data provide insight into genomically encoded develop- tion,” held April 12–14, 2016, at the Arnold and Mabel Beckman Center of the National mental programs and the species-specific GRNs that direct animal Academies of Sciences and Engineering in Irvine, CA. The complete program and video recordings of most presentations are available on the NAS website at www.nasonline.org/ development in previously unexplored branches of the tree of life. Gene_Regulatory_Networks. Thus, studies comparing GRN subcircuit wiring in distantly diverged Author contributions: J.R.T. designed research; J.R.T. and B.S.M. performed research; J.R.T., taxa (8, 9) are paving the way for the study of GRN evolution. E.M.E., V.F.H., and E.P. analyzed data; and J.R.T., E.M.E., V.F.H., and D.J.B. wrote the paper. The arrival of these new data has introduced new problems, The authors declare no conflict of interest. however. Importantly, as comparative studies of developmental This article is a PNAS Direct Submission. D.H.E. is a guest editor invited by the GRNs are becoming more commonplace, it is critical to keep in Editorial Board. mind that simple pairwise comparisons between taxa violate sta- 1To whom correspondence should be addressed. Email: [email protected]. tistical assumptions of independence and must be carried out in an This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. explicit phylogenetic framework (10). Phylogenetic comparative 1073/pnas.1610603114/-/DCSupplemental.

5870–5877 | PNAS | June 6, 2017 | vol. 114 | no. 23 www.pnas.org/cgi/doi/10.1073/pnas.1610603114 Downloaded by guest on September 28, 2021 PAPER

makes echinoids an opportune group in which to implement inte- phylogenetic sampling of data indicating the presence or absence of COLLOQUIUM grated approaches to understanding GRN evolution. the DNG in numerous euechinoid and cidaroid echinoids, as well as The echinoid crown group comprises two clades, the euechinoids in other echinoderms, makes it an ideal candidate subcircuit to study and the cidaroids (18). The adult body plans of these two clades the tempo of GRN evolution. provide prime examples of differential morphological disparity Previously, the initial evolution of the DNG was associated with (19). Euechinoids have evolved numerous diverse morphologies the divergence of cidaroid and euechinoid echinoids, at least 268 throughout their evolutionary history and include morphologi- Mya (8, 21). However, this conclusion was reached outside the cally distinct clades, such as the bilaterally symmetrical sand context of a phylogenetic comparative framework and was based dollars and heart urchins. In contrast, cidaroids have shown re- on a comparison of the differential GRN circuitry responsible for markable morphological conservation, and the earliest fossil micromere specification in regular euechinoid and cidaroid echi- cidaroids are almost morphologically identical to cidaroids living noids given the most recent date at which euechinoid and cidaroid in the oceans today (20, 21). Comparative studies of GRN ar- echinoids could have diverged (21). There are multiple evolu- chitecture between cidaroids and euechinoids have revealed ex- tionary scenarios pertaining to the timing of the evolution of the tensive differences in the wiring of their early developmental DNG that could explain the presence of this circuitry in regular GRNs (8, 22, 23), and have served to inform our understanding euechinoids like S. purpuratus and its absence in cidaroids. By of the genomic underpinning of the myriad morphological dif- analyzing the presence or absence of the DNG within an explicit ferences in their embryonic and larval development (24, 25). phylogenetic framework, it is possible to estimate the age of particular nodes in deep time. We can then estimate the proba- The Double-Negative Gate bility that the DNG was present or absent at ancestral nodes using A striking example of a GRN subcircuit is the double-negative ancestral state reconstruction to determine the support for par- gate (DNG) that operates in the early development of S. purpuratus ticular hypotheses explaining the course of GRN evolution. (15, 26, 27). This GRN subcircuit uses a double-repression To determine when, on evolutionary timescales, the DNG mechanism that spatially localizes genes critical to primary mes- likely evolved, we set out to establish a framework for rigorously enchyme cell (PMC) specification to the micromeres of the 16- investigating a timeline of GRN evolution. We used comparative cell-stage euechinoid embryo. PMC genes [e.g., alx homeobox 1 phylogenetic estimates of echinoid relationships, ancestral state (alx1), delta, ets proto-oncogene 1/2 (ets1), tbrain (tbr)] are kept reconstructions, the fossil record, and a wide array of experi- silent in the embryo by Hesc, a globally expressed repressor. mental data concerning the presence or absence of the DNG in Asymmetric localization of maternal regulatory factors at the numerous echinoid taxa and other echinoderm outgroups. This EVOLUTION vegetal pole results in the up-regulation of a repressor of Hesc, paleogenomic approach allowed us to estimate the age of the paired-class micromere anti-repressor 1 (pmar1), specifically in the DNG and to establish a timeline for GRN evolution in echinoids. micromere lineage (14, 26). Cells that express pmar1 also express PMC-specific genes and are fated to become PMCs, which later Results construct the larval skeleton of the embryo. Holothurians Appear to Lack the DNG. The holothurian P. parvimensis Although the DNG was first identified in S. purpuratus, its activity was developed as an experimental developmental system only has widely been regarded to occur throughout the euechinoid order recently, once improved methods for spawning animals were Camarodonta (28). The now-classic experiment in which ectopic established (38, 39). Those earlier studies revealed that overexpression of pmar1 converts all embryonic cells to a skeletogenic P. parvimensis has an alx1-expressing PMC-like cell lineage that fate (26) has been duplicated in numerous camaradont euechinoids, produces a small larval spicule. This lineage arises in the veg- including Hemicentrotus pulcherrimus (29), Paracentrotus lividus (30), etal pole, and, before ingression, alx1 is coexpressed with and Lytechinus variegatus (31), suggesting that DNG function and holothurian tbr, ets1, and erythroblast transformation specific circuitry are conserved in these taxa. Furthermore, whole-mount in ets-related gene (erg) orthologs (38). The expression of hesc, situ hybridization data from the camarodonts H. pulcherrimus (32) and indeed any earlier mechanisms that might lead to the speci- and L. variegatus (33) reveal spatial distributions of hesc and its fication of this lineage at the vegetal pole of P. parvimensis,were downstream targets that are consistent with the existence of the DNG not determined, however. in these euechinoids. For the present study, we blasted hesc and pmar1 sequences It also has been shown that the DNG appears to be present in a obtained from multiple species of echinoderms against the number of euechinoid echinoid taxa outside the camaradonts (28, P. parvimensis transcriptomes (https://trace.ncbi.nlm.nih.gov/Traces/ 32). Micro1,apmar1 ortholog, localizes to the 16-cell-stage mi- sra/?study=SRP012968). These transcriptomes were obtained from cromeres of the clypeasteroid sand dollar Scaphechinus mirabilis RNAs of early gastrula and larval stages (38). This identified a (32) and the spatangoid heart urchin Echinocardium cordatum single predicted hesc transcript (E values of 9e-13 for SpHesc (28). Furthermore, hesc shows complementary yet nonoverlapping and 6e-15 for PmHesc). Phylogenetic analyses (SI Appendix, expression patterns with alx1, ets1,anddelta at the 10th cleavage Sensitivity Analyses) showed that this transcript is the direct stage in S. mirabilis (32), and roughly complementary expression ortholog of the sea star and sea urchin hesc genes, and thus we with alx1, tbr,andets1 at the blastula stage in E. cordatum and the named this gene Pp-hesc. No sequences with significant similarity stomopneustoid Glyptocidaris crenularis (28). to pmar1 or the brittle star pplx were identified, but we cannot Recent work has suggested that the DNG is absent in two species discount the possibility that additional sequence coverage, espe- of cidaroid echinoids, Eucidaris tribuloides (8) and Prionocidaris cially of earlier embryonic stages, might reveal an ortholog. baculosa (34). No evidence supporting the presence of pmar1 was We next used whole-mount in situ hybridization to determine found in the E. tribuloides genome, and both organisms show vari- the spatial expression of Pp-hesc. Transcripts are localized in the able, overlapping expression patterns of hesc and its downstream vegetal plate from hatching through to early invagination (SI targets in S. purpuratus, tbr,andets1 (8). These data strongly suggest Appendix, Fig. S1), and in a spotted pattern throughout the an- that the DNG is absent in cidaroid echinoids. In addition, both pmar1 imal ectoderm. hesc continues to be expressed in the archen- and DNG circuitry are absent from asteroids (35) and ophiuroids teron, and by the larval stage, transcripts are localized to the top (36), although the latter study identified a potential ortholog to of the archenteron and diffusely across the ectoderm (SI Ap- pmar1, pplx, that lacked its function as a repressor in the ophiuroid pendix, Fig. S1F). Expression of Pp-hesc is very reminiscent of the Amphiura filiformis. Here we present data on the expression of hesc expression of the orthologous gene in the sea star Patiria miniata from the holothurian Parastichopus parvimensis that also suggests that (shown for comparison in SI Appendix, Fig. S1). Thus, these data the DNG is absent in the sister group to echinoids (37). The wide cannot support a role for Hesc as a dominant repressor of tbr or

Thompson et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5871 Downloaded by guest on September 28, 2021 erg in P. parvimensis. Because Hesc in P. parvimensis does not owing to topological variation, mean ages and 95% CIs for nodes repress tbr, a key downstream target of Hesc repression in the not explicitly discussed in the main text are provided in SI Ap- DNG, we coded the DNG as absent in holothurians in our pendix, Figs. S7–S19 and Tables S8–S20. Across all trees, 95% ancestral state reconstructions. CIs on nodes that were directly calibrated by fossils are all rel- atively narrow, whereas 95% CIs on nodes not calibrated by Divergence Time Estimation. To constrain the age of the DNG with fossils, such as the Irregularia and Camarodonta, are wider. Of respect to evolutionary time, and also the age of particular nodes particular interest for questions regarding GRN evolution are onto which we reconstructed ancestral states, we integrated fossil the ages of particular nodes onto which we have reconstructed and molecular data using a fossil-calibrated relaxed molecular the presence or absence of the DNG. All of our analyses show a clock (40) to estimate times of divergence. Because genome- Carboniferous-Permian (∼265–342 Mya) divergence for the scale data are lacking for echinoids, we instead decided to ex- euechinoids and cidaroids, and a Triassic to Early Jurassic di- plicitly account for topological and branch length uncertainty in versification of basal euechinoid lineages. The mean divergence our downstream analyses. We thus estimated divergence times time of the Camarodonta is at the oldest in the Triassic and at using 15 alternative topologies (13 of which reached conver- the youngest in the Late Cretaceous, dependent on the topology gence and are discussed herein) resulting from the resolution of used to estimate divergence times (SI Appendix, Figs. S7–S19). polytomies existing in our consensus tree inferred from Bayesian The mean divergence time of the Irregularia is approximately phylogenetic analyses in Phylobayes 4.1 (41). We also estimated Jurassic in most analyses, although the 95% CI on this node in all divergence times on the best maximum likelihood (ML) tree, analyses is wide (>100 Ma). The node representing the most which we used to perform a number of sensitivity analyses (SI recent common ancestor (MRCA) of all analyzed cidaroids has a Appendix, Sensitivity Analyses). 95% CI between ∼8 and ∼230 Mya, putting the divergence of the The 95% credible intervals (CIs) and mean posterior diver- analyzed extant cidaroids in the Cenozoic or Mesozoic. This gence times for nodes that are informative for tracing the evo- node was not directly calibrated by fossils, which likely explains lution of the DNG on one of our 13 alternative topologies are the wide 95% CI. The mean age for the node representing the shown in Fig. 1, and mean ages and 95% CIs for all other di- most recent common ancestor of all extant euechinoids is Middle vergences are shown in SI Appendix, Figs. S7–S19 and Tables S8– Triassic in all alternative topologies, whereas the 95% CI ranges S20. Because some dated nodes are not present across all trees from Early Triassic or Latest Permian (∼252 Mya) to Late Triassic

Temnopleurus Temnotrema Microcyphus 95% credibile Interval on Salmaciella divergence time estimate Salmacis

Amblypneustes Camarodonta Mean posterior probability of Holopneustes Mespilia divergence time estimates Pseudechinus Genocidaris Mean posterior probability of Sphaerechinus Cytechinus DNG Absent or Present Tripneustes Paracentrotus Euechinoidea Psammechinus Strongylocentrotus Anthocidaris Lytechinus Carinacea

Arbacia * Coelopleurus Stomopneustes Echinocyamus Rumphia Echinodiscus Encope

Conolampas Irregularia Echinolampas Cassidulus Fellaster Echinoneus Archaeopneustes Meoma Brissopsis Echinocardium Spatangus Abatus Paleopneustes Plexechinus Aspidodiadema Caenopedina Centrostephanus Diadema Araeosoma Asthenosoma Calocidaris Cidaroidea Stereocidaris Prionocidaris Cucumaria Psychopetes Asterias Ophiocan Antedon

Mississippian Pennsylvanian Pridoli Early Ludlow Early Early Early Early Late Late Late Late Late Middle Middle Middle Middle Wenlock Eocene Series2 Series3 Terreneuvian Paleocene Miocene Oligocene Guadalupian Llandovery Lopingian Cisuralian Furongian Cambrian Ordovician Silurian Devonian Carboniferous Permian Triassic Jurassic Cretaceous Paleogene Neogene 0 70 10 80 60 40 20 90 50 30 100 280 260 190 130 540 530 520 510 500 490 480 470 450 430 420 410 400 380 360 340 330 320 310 290 270 250 230 200 180 160 140 110 460 440 390 370 350 240 220 170 150 120 300 210

Fig. 1. Fossil-calibrated time tree showing age of clades of crown group echinoids based on one of the 15 alternative topologies resulting from the Bayesian phylogenetic analysis and showing the probability of the presence or absence of the DNG on nodes of interest. Blue bars are 95% CIs on divergence times for nodes. Pie charts show mean PPs that the DNG was present (blue) or absent (red) at particular ancestral nodes in ancestral state reconstructions on 1,300 alternative topologies under priors on transition rates of U(0, 0.01). The taxa in blue have been demonstrated experimentally to use the DNG for micromere specification, whereas the taxa in red do not use the DNG. Taxonomic names represent higher-level echinoid taxonomic groupings. PPs for ancestral state reconstructions are presented in SI Appendix, Table S8. *Indicates Stomopneustoida + Arbacioida. The geological timescale was created in strap (62). Illus- trations of spatangoid silhouette, by Hans Hillewaert (vectorized by T. Michael Keesey); cidaroid silhouette, by Didier Descouens (vectorized by T. Michael Keesey); and regular euechinoid, from Frank Förster (based on a picture by Jerry Kirkhart, modified by T. Michael Keesey), are available for use under a Creative Commons Attribution-ShareAlike 3.0 Unported license (https://creativecommons.org/licenses/by-sa/3.0/). Clypeasteroid image, from Michael Site, is available for use under a Creative Commons Attribution-NonCommercial 3.0 Unported license (https://creativecommons.org/licenses/by-nc/3.0/). All images are from Phylopic (www.phylopic.org).

5872 | www.pnas.org/cgi/doi/10.1073/pnas.1610603114 Thompson et al. Downloaded by guest on September 28, 2021 PAPER

(∼220 Mya) on all topologies. The mean estimated diver- phylogeny is 0.5, with an equal probability of presence or absence of COLLOQUIUM gence times for the divergence between the echinothurioids (Ara- the DNG at each ancestral node. Using narrower priors, which may eosoma + Asthenosoma) and (aspidodiadematoids + pedinoids) + be more appropriate for modeling the presence or absence of the diadematoids are in the Late Triassic, with 95% CIs from the Late DNG, which is unlikely to have evolved and unevolved numerous Triassic (∼211 Mya) to Middle Triassic (∼246 Mya) on all 13 to- times throughout the evolutionary history of echinoids, resulted in pologies. Furthermore, the mean ages and CIs for the divergences more informative reconstruction of ancestral states. ML recon- between (aspidodiadematoids + pedinoids) and diadematoids struction of these ancestral states on all but four of the 1,300 trees and aspidodiadematoids and pedinoids reside within the Late resulted in transition rate parameters of <0.001 (Dataset S1), which Triassic. Results from analyses using the ML topology are is in line with the more informative reconstructed ancestral states broadly comparable (SI Appendix,Fig.S3) and are generally resulting from smaller priors on transition rates. Using a uniform insensitive to changes in model parameters and tree topology prior from 0 to 0.01 on transition rates resulted in relatively (SI Appendix,Figs.S4andS6). high mean posterior probabilities (PPs) in favor of the presence of the DNG in the last common ancestor of camarodonts (0.76), Ancestral State Reconstruction. Recent paleogenomic studies have of Irregularia (0.76), and of Strongylocentrotus and Paracentrotus used the fossil record to date the appearance of particular GRNs (0.97) (Fig. 1). in evolutionary time by determining the first appearance of the The DNG was present in the MRCA of camarodonts and morphological structures that are specified by these GRNs in the the Irregularia, with a mean PP of 0.79; in the MRCA of fossil record (12). There is no fossil record of embryonic sea ur- diadematoids and camarodonts, with a mean PP of 0.76; and in chins; thus, we used ancestral state reconstructions (42) to deter- the MRCA of camarodonts and stomopneustoids, with a mean mine the probability that the DNG was present or absent in the PP of 0.78. The MRCA of echinoids appears to have used the last common ancestor of particular clades. We analyzed two datasets for our ancestral state reconstructions. One of these DNG, with a mean PP of 0.65, whereas the MRCA of euechi- datasets maximized taxon sampling and included many more taxa noids appears to have used the DNG, with a mean PP of 0.76; for which experimental evidence was unknown than known. This however, the MRCA of extant cidaroids appears to have not dataset is based on 1,300 time-scaled trees (Divergence Time Es- used the DNG, with a mean PP of 0.95. These probabilities of timation), with 100 trees of differing branch lengths sampled at the reconstructed presence and absence of the DNG at select random from the posterior distribution of each of the 13 diver- nodes are plotted in Fig. 1. When using a slightly more diffuse gence time estimation analyses run using alternative topologies. In prior [U(0, 0.05)], the mean PPs of the presence or absence of the EVOLUTION this way, we explicitly account for phylogenetic uncertainty in our DNG at all reconstructed ancestral nodes approached 0.5. Under ancestral state reconstructions. The other dataset was limited to this prior, the mean PPs of the presence or absence of the DNG of taxa for which direct experimental evidence concerning the DNG most reconstructed ancestral nodes were between 0.45 and 0.55 was available, and the topology onto which this ancestral state The exceptions to these less informative reconstructions are the reconstruction was run reflected the ML estimates of topological node representing the MRCA of extant analyzed cidaroids, where relationships (Fig. 2 and SI Appendix,Fig.S5). the DNG was absent with a mean PP of 0.72, and the node rep- Our Bayesian ancestral state reconstructions using the taxo- resenting the MRCA of Strongylocentrotus and Paracentrotus, nomically inclusive dataset were run under a variety of priors on the which likely used the DNG with a mean PP of 0.75. transition rates for each state. Given that the presence or absence of Many of our mean PP values derived using the dataset that the DNG is unknown for many of the taxa in our tree, with more maximized taxon sampling are systematically close to 0.50, uninformative priors (uniform distribution with bounds 0 and 100), resulting from the large amount of taxa included in the analysis the probability of reconstructed ancestral states at all nodes in the for which no data on the DNG are available. This is especially

A B Holothuroids, Asteroids, Ophiuroids Cidaroid Skeletogenic Micromere-descendants Circuitry largely unknown, but lack DNG

PPrionocidarisrionocidaris

StomopneustesStomopneustes Adjacent cells Micromere-descendants

Euechinoid Double Negative Gate ParacentrotusParacentrotus C

StrongylocentrotusStrongylocentrotus

LytechinusLytechinus

EncopeEncope Micromeres

EchinocardiumEchinocardium

Fig. 2. Ancestral state reconstruction of the DNG. (A) Pruned analyses showing reconstructed ancestral presence or absence of the DNG at select nodes on the ML tree using ML estimates of branch lengths. Pie charts show the mean PPs, with red sections representing the probability of absence of the DNG and blue sections representing the reconstructed probability of the DNG being present at a given ancestral node. (B) Wiring diagram for extant cidaroid echinoids showing the probable interactions of key genes involved in the DNG in S. purpuratus. There is no double-negative repression, because there is no Pmar1 gene, and hesc does not repress tbrain or ets1, as it does in euechinoids. (C) Wiring diagram showing the interactions of genes involved in the DNG in S. purpuratus and other examined euechinoids.

Thompson et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5873 Downloaded by guest on September 28, 2021 the case when wider priors are used. This is likely because the novelty unique to all euechinoid echinoids (1, 8, 21), and our tips with missing data are treated as having an equal probability ancestral state reconstructions have revealed mean PPs that of the presence or absence of the DNG in ancestral state re- strongly identify the DNG as a novelty unique to all extant constructions (42). Using a pruned dataset (Fig. 2), which in- euechinoids. Nonetheless, our analysis also fails to demonstrate cluded only taxa from orders for which direct experimental unambiguously that the last common ancestor of echinoids (in- evidence regarding the presence or absence of the DNG exists, cluding cidaroids) did not use the DNG, and instead used the resulted in PPs farther from 0.50 and more representative of the mechanism present in extant cidaroids (8, 21). Our results in- current state of knowledge with regard to taxonomic sampling of stead slightly favor the alternative hypothesis, that the DNG was developmental data. This analysis was run using the ML branch used in the MRCA of extant echinoids. We are hesitant about lengths as opposed to a time-scaled tree, which was able to more endorsing this interpretation, however, given that PPs for both adequately resolve ancestral states with a wider prior [U(0, 100)]. the presence and absence of the DNG at this node are closer to This dataset had high mean PPs (Fig. 2 and SI Appendix, Table 0.5 than at any other nodes examined. S3), and showed that the MRCA of the camarodonts used the The use of ancestral state reconstruction provides a sound sta- DNG, with a mean PP of 0.91, and that the DNG was present in tistical framework within which to explore the evolution of GRNs. the MRCA of the Irregularia + Camadodonta, with a mean PP Although we previously made statements regarding GRN evolution of 0.82, and also in the MRCA of the Carinacea, with a mean PP from cross-species comparisons in a few taxa (8, 21), here we of 0.83. The last common ancestor of echinoids showed a mean demonstrate that statistical analysis of GRN evolution within the PP of 0.41 and 0.59, respectively, in favor of the absence or context of a phylogeny provides a more rigorous framework within presence of the DNG. Using a tree topology based on mor- which to examine their evolution. Whereas a most parsimonious phology (18) with this pruned dataset also revealed similarly high interpretation of our data would have led us to make binary posterior mean PPs, supporting the presence of the DNG in the statements regarding the presence or absence of the DNG at par- MRCAs of the camarodonts, the Carinacea, the Irregularia, and ticular ancestral nodes, our approach allows us to assign probabil- the Camarodonta + Stomopneustidae (SI Appendix, Table S3 ities to our conclusions while taking into account uncertainty in and Fig. S5). phylogenetic topology and branch lengths. Surprisingly, within this statistical framework, we cannot definitively determine whether or Discussion not the DNG was used in the MRCA of all extant echinoids. This The Antiquity of the DNG. Our data strongly favor a euechinoid result indicates that inferences following from cross-species com- origin of the DNG, which appears to have evolved at the latest parisons of GRNs are at best one of many evolutionary scenarios, sometime before the Late Triassic (∼220–230 Mya depending and that caution should be exercised regarding definitive statements on the topology used). Support is greatest for the presence of about evolutionary changes in gene regulation associated with di- the DNG in the MRCA of camarodonts and in the MRCA of vergence of taxa. As more data on the utilization of the DNG in Paracentrotus and Strongylocentrotus, which is in line with other diverse echinoid clades become available, and as the echinoid recently published studies showing marked conservation in the phylogeny is improved on with new phylogenomic techniques, our timing of gene expression between camarodonts (43). Our results confidence regarding the phylogenetic origin, and antiquity, of the also favor the presence of the DNG over its absence in the last DNG will improve. At the very least, we are able to say that the common ancestor of early branching euechinoid clades. This DNG is an ancient GRN subcircuit that likely evolved in or before indicates that the DNG was likely used in the MRCA of extant the MRCA of extant euechinoids and has been conserved and used euechinoids. The pruned dataset in particular shows strong throughout the euechinoid tree of life and across at least 220 My of support for a pre-Carinacea age of the DNG; however, this echinoid evolution. dataset provides limited data concerning the non-Carinacea euechinoids, given the absence of direct experimental evidence The DNG and Echinoid Macroevolutionary Trends. Obtaining age from these early diverging euechinoid taxa. Future experimental estimates for the DNG allows us to compare this GRN innova- work in these euechinoids, such as the diadematoids, will aug- tion with other events in the macroevolutionary history of echi- ment phylogenetic sampling and further resolve the age of the noids. It was recently shown that crown group echinoids DNG in the euechinoid lineage. In the last common ancestor of underwent heightened rates of evolution during the Early Ju- echinoids, our Bayesian analyses show a mean PP of 0.65 for the rassic (20), which, given the results of our analysis, postdate the presence of the DNG at this node under our smallest prior. Our origin of the DNG during or before the Late Triassic. This burst analyses using the ML topology also show mean PPs slightly of diversification is tightly linked to the diversification of irreg- favoring the presence of the DNG over its absence at this node. ular echinoids and other euechinoid clades that, according to our Because mean PPs at this node showed the lowest support analysis, likely used the DNG. Because the DNG functions to among all nodes examined for the presence of the DNG, we can specify skeletogenic cells in developing embryos, its correct only tentatively interpret our findings as indicating that the DNG spatiotemporal deployment is required for embryonic and larval was used in the MCRA of all extant echinoids. Given the much skeletogenesis. That such a critical function is dependent on the higher mean PPs at the MRCA of extant euechinoids, however, DNG may have even resulted in the existence of “fail safe” it seems more likely that the DNG evolved at some point be- regulatory wiring of blimp1 and hesc, wherein Blimp1 represses tween the divergence of euechinoids and cidaroids in the hesc in the absence of Pmar1 (44). Given that the DNG appears Carboniferous-Permian and the MRCA of extant euechinoids in to have been conserved across ∼220 My of echinoid evolution, it the Triassic. We also note that that our results at the more basal is possible that the DNG is so critical for early indirect devel- euechinoid nodes are more sensitive to prior choice compared opment in euechinoids that it has undergone stabilizing selec- with those at more recently branching nodes (SI Appendix, Table tion, although this has yet to be demonstrated empirically. S21), and broader phylogenetic sampling in the future should The conserved role of the DNG in micromere specification, which decrease the sensitivity of results at these nodes to prior choice. appears to predate the diversification of irregular euechinoids, Given the low support for either the presence (mean PP, 0.65) also indicates that the regulatory mechanism of micromere spec- or absence (mean PP, 0.35) in the node representing the diver- ification very early in development has been constrained, whereas gence of euechinoids and cidaroids, we are unable to confidently aspects of later larval and adult morphology, which show high attribute the evolution of the DNG to the divergence of the degrees of disparity (20, 45), have been relatively free to vary. euechinoids and cidaroids as we have done in the past (8, 21). Thus, the euechinoid double-negative specification of micromeres Previous authors have characterized the DNG as an evolutionary and skeletogenesis offers a striking example of genomically encoded

5874 | www.pnas.org/cgi/doi/10.1073/pnas.1610603114 Thompson et al. Downloaded by guest on September 28, 2021 PAPER

developmental constraint early in development, whereas high de- a large polytomy in which the relationships of many of these major echinoid COLLOQUIUM grees of plasticity have prevailed in the evolution of later larval and clades to one another are unresolved. An analysis in Phylobayes using the adult morphology. This is also in stark contrast to the cidaroids, same model parameters and convergence criteria was also run using only the which although exhibiting much more constraint in their adult body 18s and 28s sequence data (Dataset S4). This topology retained most of the same clades as the topology resulting from inference using all three genes, plans (20), display high levels of morphological variability in early and was equally unresolved (SI Appendix,Fig.S21). We thus chose to use the development. Whereas euechinoids have an invariant four micro- topology resulting from inference using all three genes as a backbone for meres, cidaroids contain a variable number, even within a single topologies used in divergence time estimation and ancestral state re- species (8, 24, 25). This apparent differential constraint at different construction (Divergence Time Estimation and Ancestral State Reconstruction). times in development found in these two clades represents an in- Given that we explicitly account for phylogenetic uncertainty in our ancestral triguing avenue for future research. state reconstruction, we would not expect inference using polytomy resolu- tions of the two-gene consensus tree to seriously alter our results. Because the “Fossilized” GRNs. Paleogenomics allows us to frame hypotheses goal of this analysis was not explicitly to reconstruct echinoid phylogeny, but regarding the nature of embryonic micromere specification in rather to examine the evolution of the DNG within a phylogenetic context, the fossil echinoids. Because it has been shown that direct devel- phylogenetic uncertainty evident from our Bayesian analysis is accounted for in our ancestral state reconstructions (see below). opment in echinoids was not widespread until the Cretaceous (46), our analyses indicate that most indirectly developing eue- Divergence Time Estimation. The ML tree with the highest log-likelihood was chinoids, which compose the majority of the echinoid fossil re- used for divergence time estimation in the MCMCTREE package in PAML (56). cord since the Triassic, likely used the DNG. When put into the After removal of the outgroup taxa, a global clock model (clock = 1in context of environmental and ecological change undergone by BASEML) with the root of the tree (euechinoid-cidaroid divergence) calibrated echinoids through the Cretaceous-Paleogene mass extinction, was used to estimate a suitable prior for the substitution rate μ in BaseML. We which had profound effects on echinoid size, feeding strategy, thus used a gamma-Dirichlet distribution with shape parameter α = 1.79 and and biogeographic distribution (47), and further Cenozoic cli- scale parameter β = 70 (rgene_gamma = 1.79, 70 in MCMCTREE). Parameters α 2 matic changes (48), our findings suggest that the DNG has been and β of the gamma Dirichlet prior on σ , specifying the variability of rates very robust to evolutionary change over the course of the past across branches, were set to 1 and 2.68, respectively. The birth death priors were set to λ = 1, μ = 1, and ρ = 0(BDParas= 1 1 0). Divergence time estimation 220 My. Paleogenomics offers the opportunity to compare pre- using the ML topology was done with MCMCTREE using the approximate sumptive evolutionary changes in GRNs with paleoclimatic and likelihood calculation of dos Reis and Yang (57). We chose to use the inde- paleoenvironmental changes. We assert that this is a promising pendent rates model (58, 59). For our time priors, we used nine uniform EVOLUTION and intriguing avenue of future research, and that it is only one constraints, most of which focused on the earlier branching nodes of the tree. of the many doors that can be opened by paleogenomic ap- MCMCTREE uses soft constraints (40), and for all of our divergence time priors, proaches to studying the evolution of GRNs. 2.5% of the distribution served as a soft tail on both the minimum and maximum constraints. Two independent MCMC chains were run for 5,000,000 Methods iterations with a sample drawn every 50 iterations. The first 50,000 runs were Phylogenetic Tree Construction. For our dataset, we used previously published discarded as burn in. This resulted in a total of 100,001 samples in each chain. mitochondrial 16s, 18s, and 28s small subunit rRNA sequence data (49–51), Following analyses, all resultant MCMC samples were examined in Tracer which comprises the most comprehensive known molecular dataset with version 1.6 (60) to check for convergence. respect to echinoid taxonomic sampling. Although such a dataset for in- Fossil calibrations for analyses in MCMCTREE were compiled and justified “ ” ference of topology has limited use compared with more widely used phy- with up-to-date absolute dates and the first attempt at using best practices logenomic datasets, we accounted for uncertainty in topology and branch (61) to rigorously assemble well-supported fossil calibrations for use in prior length in our downstream ancestral state reconstructions. We used the construction for divergence time estimation in echinoids. Our compilation is concatenated aligned dataset of Smith et al. (51) with gaps removed, leaving not exhaustive, given our primary goal of rigorously calibrating nodes to- 3,230 sites (Datasets S2 and S3). For our ML analysis, we determined the best- ward the base of the echinoid tree, where most discussion regarding the fitting substitution model using Modeltest 3.7 (52) in PAUP* (53), which, origin of the DNG has been focused. A complete list of fossil calibrations among the models available in MCMCTREE (which we used to estimate di- with justifications is provided in SI Appendix, Fossil Calibrations for Priors on vergence times on our ML topology), identified GTR+Γ as the best fit. The Divergence Time Estimates and Datasets S5 and S6. outgroup taxa were constrained to represent the recently supported We also ran sensitivity analyses to check the robustness of our results to Asterozoa hypothesis (37). Phylogenetic trees were constructed using variations in prior, model choice, and tree topology. Our results were gen- RaxMLversion 8 (54). We ran 20 ML analyses, with the tree with the highest erally robust to changes in model parameters. The results of our sensitivity likelihood used for ancestral state reconstructions and divergence time es- analyses are shown in SI Appendix, Fig. S6 and discussed in SI Appendix, timation. This tree, showing bootstrap support from 1,000 replicates, is Sensitivity Analyses. shown in SI Appendix, Fig. S2. Because our Bayesian majority rule consensus tree was unresolved, we es- Our ML topology differs from previously published estimates (51) and the timated divergence times using multiple phylogenetic topologies resulting most up-to-date morphological phylogeny (18), predominantly in the from resolution of our Bayesian majority rule consensus tree. Nodes in the placement of a clade consisting of the stomopneustoids and arbacioids consensus tree with a PP <0.25 were collapsed. After this, all irregular echinoids plotting as sister group to a clade of Camarodonta + Irregularia. Because this were constrained to form a clade, based on morphological evidence (18), and topology differs with morphology and previous molecular estimates, which divergence times were estimated on the 15 possible resolutions of the either put arbacioids as a sister group to the camarodonts or Arbacioida + resulting tree. Divergence times were estimated using Phylobayes 4.1 (41) Stomopneustoida as a sister group to the camarodonts, we decided to run 20 under the model of Drummond et al. (58), where the rate on each branch is an additional ML analyses using a constrained topology that reflects the most independent draw from a gamma distribution, using the CAT-GTR+ Γ model recent morphological phylogeny (figure 5 in ref. 18). We used this topology, for the substitution process. The birth death priors were set to λ = 1, μ = 1, and shown in SI Appendix, Fig. S4, for sensitivity analyses. ρ = 0. Fossil calibrations were soft and used variable combinations of the same In addition to our ML analyses, we performed a Bayesian analysis in calibrations as were used for divergence time estimation using the ML tree Phylobayes 4.1 (41) using the CAT-GTR+ Γ model (55). Two independent plus three calibrations for divergences outside of echinoids (SI Appendix, Fossil Markov chains were run until both reached convergence. The burn-in for Calibrations for Priors on Divergence Time Estimates and Datasets S7–S20). each chain was 2,000 points, and each chain was sampled every 10 points up Thus, each analysis used between 9 and 12 calibrations. to 20,000 points. Chains were run until the effective sample size for each For each topology, two independent chains were run with a burn-in of parameter was >300 and the relative difference between parameters of 1,000 points sampling every 10 points. MCMC convergence was assessed using each run was <0.1. Convergence of Markov chain Monte Carlo (MCMC) was tracecomp and bpcomp in Phylobayes. Chains were run until the effective assessed using the tracecomp and bpcomp commands in Phylobayes. sample size for each parameter was >100 and the relative difference be- The majority rule consensus tree resulting from our Bayesian analyses (SI tween parameters of each run was <0.3. Thirteen of the 15 analyses reached Appendix, Fig. S20) resolved many traditional clades supported by mor- convergence within time available, whereas the two analyses that failed to phology and previous molecular analyses, including the cidaroids, camar- converge were excluded from ancestral state reconstructions. Those two odonts, and two clades of irregular echinoids. However, this tree contained topologies exhibited relationships widely different from previous molecular

Thompson et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5875 Downloaded by guest on September 28, 2021 and morphological inferences of echinoid topology (18, 51), and thus their across the 1,300 time-calibrated trees resulting from divergence time esti- exclusion from ancestral state reconstruction likely had no effect on mation. In this way, phylogenetic uncertainty with respect to both branch our inferences. length and topology were accounted for in reconstruction (42) (Dataset S23). We ran these analyses under four different sets of priors on transition rates Ancestral State Reconstructions. Representatives from families (or orders in (SI Appendix, Table 21). We also ran this dataset using the ML branch the case of the Stomopneustoida) and nonechinoid echinoderm classes for lengths from the best ML tree inferred in RAxML and the ML estimates of which there is experimentally derived evidence in favor or in disfavor of the branches constrained on the morphological tree of Kroh and Smith (18) (SI presence of the double-negative repression of hesc by Pmar1 and of alx1, tbr, Appendix, Tables S1 and S2). The topology of the pruned dataset was or ets1 by Hesc were used to score extant taxa for the presence or absence of constrained in RaxML before branch length estimation, such that the to- the DNG. For instance, the cidaroid P. baculosa expresses hesc in the same pological relationships matched those of the ML tree inferred from the cells as tbr, alx1, and ets1 at the late blastula stage, which indicates that Hesc more complete dataset. All analyses were run in BayesTraits 2.0 (42) using could not be acting as a repressor of these genes, a canonical feature of the the MRCA command, which reconstructs probabilities of ancestral states at DNG, and as such, the DNG is not used in this taxon (34). When experimental the node representing the most recent common ancestor of the specified data showed precise nonoverlapping expression patterns and experimen- taxa (Datasets S21 and S22). We ran our MCMC analyses for 10,000,000 tally tested GRN linkages mirroring those of S. purpuratus, in which the DNG iterations with a burn-in of 2,000,000 iterations and sampling every 1,000 was first discovered, the DNG was scored as present in that taxon. Given our iterations. All analyses were run under varying priors to assess the sensi- interest in modeling the evolution of the DNG as it was first described from tivity of analyses to prior choice. S. purpuratus, we feel that coding the DNG as present/absent is appropriate for this analysis, because any difference in experimentally derived linkages, ACKNOWLEDGMENTS. This research is the outgrowth of an extremely especially evidence of nonrepression of alx1, tbr,orets1 by Hesc or of hesc fruitful collaboration between the authors and the late Eric H. Davidson, by Pmar1, allows us to code a taxon as absent and thereby maximize which was funded by US National Science Foundation Grant IOS1240626 (to phylogenetic coverage. D.J.B. and Eric H. Davidson). J.R.T. thanks M. Florence, K. Hollis, D. Levin, and We used two datasets for our analyses, one dataset that maximized J. Strotman at the US National Museum of Natural History and T. Ewin at the Natural History Museum, London, who provided access to collections that phylogenetic coverage but had low proportional coverage of experimentally house many of the specimens used in fossil calibration for divergence time derived presence and absence data, and a pruned dataset that included only estimation. We also thank J. Yu, Director of Computing Services at the taxa from families (or the Stomopneustoida) for which direct experimental University of Southern California’s Department of Earth Sciences, for com- evidence exists regarding the presence or absence of the DNG. The dataset puting time and assistance. The comments of two anonymous reviewers maximizing phylogenetic coverage was used to reconstruct ancestral states improved on an earlier version of this manuscript and are much appreciated.

1. Ettensohn CA (2009) Lessons from a gene regulatory network: Echinoderm skeleto- 23. Erkenbrack EM (2016) Divergence of ectodermal and mesodermal gene regulatory genesis provides insights into evolution, plasticity and morphogenesis. Development network linkages in early development of sea urchins. Proc Natl Acad Sci USA 113(46): 136(1):11–21. E7202–E7211. 2. McClay DR (2011) Evolutionary crossroads in developmental biology: Sea urchins. 24. Schroeder T (1981) Development of a “primitive” sea urchin (Eucidaris tribuloides): Development 138(13):2639–2648. Irregularities in the hyaline layer, micromeres, and primary mesenchyme. Biol Bull 3. Peter IS, Davidson EH (2011) Evolution of gene regulatory networks controlling body 161:141–151. plan development. Cell 144(6):970–985. 25. Wray GA, McClay DR (1988) The origin of spicule-forming cells in a “primitive” sea 4. Peter IS, Davidson EH (2016) Implications of developmental gene regulatory networks urchin (Eucidaris tribuloides) which appears to lack primary mesenchyme cells. inside and outside developmental biology. Curr Top Dev Biol 117:237–251. Development 103(2):305–315. 5. Davidson EH, Erwin DH (2006) Gene regulatory networks and the evolution of animal 26. Oliveri P, Carrick DM, Davidson EH (2002) A regulatory gene network that directs – body plans. Science 311(5762):796–800. micromere specification in the sea urchin embryo. Dev Biol 246(1):209 228. 6. Hejnol A, Lowe CJ (2015) Embracing the comparative approach: How robust phy- 27. Revilla-i-Domingo R, Oliveri P, Davidson EH (2007) A missing link in the sea urchin logenies and broader developmental sampling impacts the understanding of nervous embryo gene regulatory network: hesC and the double-negative specification of – system evolution. Philos Trans R Soc Lond B Biol Sci 370(1684):20150045. micromeres. Proc Natl Acad Sci USA 104(30):12383 12388. 7. Levin M, et al. (2016) The mid-developmental transition and the evolution of animal 28. Yamazaki A, Minokawa T (2015) Expession patterns of mesenchyme specification body plans. Nature 531(7596):637–641. genes in two distantly related echinoids, Glyptocidaris crenularis and Echinocardium – 8. Erkenbrack EM, Davidson EH (2015) Evolutionary rewiring of gene regulatory net- cordatum. Gene Expr Patterns 17(2):87 97. work linkages at divergence of the echinoid subclasses. Proc Natl Acad Sci USA 29. Yamazaki A, Ki S, Kokubo T, Yamaguchi M (2009) Structure-function correlation of micro1 for micromere specification in sea urchin embryos. Mech Dev 126(8-9): 112(30):E4075–E4084. 611–623. 9. Hinman VF, Nguyen AT, Cameron RA, Davidson EH (2003) Developmental gene reg- 30. Duloquin L, Lhomond G, Gache C (2007) Localized VEGF signaling from ectoderm to ulatory network architecture across 500 million years of echinoderm evolution. Proc mesenchyme cells controls morphogenesis of the sea urchin embryo skeleton. Natl Acad Sci USA 100(23):13356–13361. Development 134(12):2293–2302. 10. Hejnol A, Dunn CW (2016) Animal evolution: Are phyla real? Curr Biol 26(10): 31. Wu SY, McClay DR (2007) The Snail repressor is required for PMC ingression in the sea R424–R426. urchin embryo. Development 134(6):1061–1070. 11. Dunn CW, Luo X, Wu Z (2013) Phylogenetic analysis of gene expression. Integr Comp 32. Yamazaki A, Furuzawa Y, Yamaguchi M (2010) Conserved early expression patterns Biol 53(5):847–856. of micromere specification genes in two echinoid species belonging to the orders 12. Bottjer DJ, Davidson EH, Peterson KJ, Cameron RA (2006) Paleogenomics of echino- Clypeasteroida and Echinoida. Dev Dyn 239(12):3391–3403. derms. Science 314(5801):956–960. 33. Sharma T, Ettensohn CA (2010) Activation of the skeletogenic gene regulatory net- 13. Bottjer DJ (2016) Geobiology and palaeogenomics. Earth Sci Rev 164:182–192. work in the early sea urchin embryo. Development 137(7):1149–1157. 14. Davidson EH, et al. (2002) A genomic regulatory network for development. Science 34. Yamazaki A, Kidachi Y, Yamaguchi M, Minokawa T (2014) Larval mesenchyme cell 295(5560):1669–1678. specification in the primitive echinoid occurs independently of the double-negative 15. Oliveri P, Tu Q, Davidson EH (2008) Global regulatory logic for specification of an gate. Development 141(13):2669–2679. – embryonic cell lineage. Proc Natl Acad Sci USA 105(16):5955 5962. 35. McCauley BS, Weideman EP, Hinman VF (2010) A conserved gene regulatory network 16. Peter IS, Davidson EH (2009) Genomic control of patterning. Int J Dev Biol 53(5-6): subcircuit drives different developmental fates in the vegetal pole of highly divergent – 707 716. echinoderm embryos. Dev Biol 340(2):200–208. 17. Smith AB, Savill JJ (2001) Bromidechinus, a new Ordovician echinozoan (Echino- 36. Dylus DV, et al. (2016) Large-scale gene expression study in the ophiuroid Amphiura dermata), and its bearing on the early history of echinoids. Earth Env Sci T R So 92: filiformis provides insights into evolution of gene regulatory networks. Evodevo 7:2. – 137 147. 37. Telford MJ, et al. (2014) Phylogenomic analysis of echinoderm class relationships 18. Kroh A, Smith AB (2010) The phylogeny and classification of post-Palaeozoic echi- supports Asterozoa. Proc Biol Sci 281(1786):20140479. noids. J Syst Palaeontology 8:147–212. 38. McCauley BS, Wright EP, Exner C, Kitazawa C, Hinman VF (2012) Development of an 19. Gao F, et al. (2015) Juvenile skeletogenesis in anciently diverged sea urchin clades. embryonic skeletogenic mesenchyme lineage in a sea cucumber reveals the trajectory Dev Biol 400(1):148–158. of change for the evolution of novel structures in echinoderms. Evodevo 3(1):17. 20. Hopkins MJ, Smith AB (2015) Dynamic evolutionary change in post-Paleozoic echi- 39. Le HS, Schulz MH, McCauley BM, Hinman VF, Bar-Joseph Z (2013) Probabilistic error noids and the importance of scale when interpreting changes in rates of evolution. correction for RNA sequencing. Nucleic Acids Res 41(10):e109. Proc Natl Acad Sci USA 112(12):3758–3763. 40. Yang Z, Rannala B (2006) Bayesian estimation of species divergence times under a 21. Thompson JR, et al. (2015) Reorganization of sea urchin gene regulatory networks at molecular clock using multiple fossil calibrations with soft bounds. Mol Biol Evol 23(1): least 268 million years ago as revealed by oldest fossil cidaroid echinoid. Sci Rep 5: 212–226. 15541. 41. Lartillot N, Lepage T, Blanquart S (2009) PhyloBayes 3: A Bayesian software package 22. Erkenbrack EM, et al. (2016) Ancestral state reconstruction by comparative analysis of for phylogenetic reconstruction and molecular dating. Bioinformatics 25(17): a GRN kernel operating in echinoderms. Dev Genes Evol 226(1):37–45. 2286–2288.

5876 | www.pnas.org/cgi/doi/10.1073/pnas.1610603114 Thompson et al. Downloaded by guest on September 28, 2021 PAPER

42. Pagel M, Meade A, Barker D (2004) Bayesian estimation of ancestral character states 51. Smith AB, et al. (2006) Testing the molecular clock: Molecular and paleontological COLLOQUIUM on phylogenies. Syst Biol 53(5):673–684. estimates of divergence times in the Echinoidea (Echinodermata). Mol Biol Evol 43. Gildor T, Ben-Tabou de-Leon S (2015) Comparative study of regulatory circuits in two 23(10):1832–1851. sea urchin species reveals tight control of timing and high conservation of expression 52. Posada D, Crandall KA (1998) MODELTEST: Testing the model of DNA substitution. dynamics. PLoS Genet 11(7):e1005435. Bioinformatics 14(9):817–818. 44. Smith J, Davidson EH (2009) Regulative recovery in the sea urchin embryo and the 53. Swofford DL (2003) PAUP*: Phylogenetic Analysis Using Parsimony (* and Other stabilizing role of fail-safe gene network wiring. Proc Natl Acad Sci USA 106(43): Methods), Version 4 (Sinauer Associates, Sunderland, MA). 18291–18296. 54. Stamatakis A (2014) RAxML version 8: A tool for phylogenetic analysis and post- 45. Wray GA (1992) The evolution of larval morphology during the post-Paleozoic radi- analysis of large phylogenies. Bioinformatics 30(9):1312–1313. ation of echinoids. Paleobiology 18(3):258–287. 55. Lartillot N, Philippe H (2004) A Bayesian mixture model for across-site heterogeneities 46. Jeffery CH (1997) Dawn of echinoid nonplanktotrophy: Coordinated shifts in devel- in the amino-acid replacement process. Mol Biol Evol 21(6):1095–1109. opment indicate environmental instability prior to the K-T boundary. Geology 25(11): 56. Yang Z (2007) PAML 4: Phylogenetic analysis by maximum likelihood. Mol Biol Evol 991–994. 24(8):1586–1591. 47. Smith AB, Jeffery CH (1998) Selectivity of extinction among sea urchins at the end of 57. dos Reis M, Yang Z (2011) Approximate likelihood calculation on a phylogeny for the Cretaceous period. Nature 392(5):69–71. Bayesian estimation of divergence times. Mol Biol Evol 28(7):2161–2172. 48. Kroh A (2007) Climate changes in the Early to Middle Miocene of the Central Para- 58. Drummond AJ, Ho SYW, Phillips MJ, Rambaut A (2006) Relaxed phylogenetics and tethys and the origin of its echinoderm fauna. Palaeogeogr Palaeoclimatol dating with confidence. PLoS Biol 4(5):e88. Palaeoecol 253:169–207. 59. Rannala B, Yang Z (2007) Inferring speciation times under an episodic molecular clock. 49. Littlewood DTJ, Smith AB (1995) A combined morphological and molecular phylog- Syst Biol 56(3):453–466. eny for sea urchins (Echinoidea: Echinodermata). Philos Trans R Soc Lond B Biol Sci 60. Rambaut A, Suchard M, Drummond AJ (2013) Tracer version 1.6. Available at tree.bio. 347(1320):213–234. ed.ac.uk/software/tracer/. Accessed February 6, 2017. 50. Stockley B, Smith AB, Littlewood T, Lessios HA, Mackenzie-Dodds JA (2005) Phylo- 61. Parham JF, et al. (2012) Best practices for justifying fossil calibrations. Syst Biol 61(2): genetic relationships of spatangoid sea urchins (Echinoidea): Taxon sampling density 346–359. and congruence between morphological and molecular estimates. Zool Scr 34(5): 62. Bell MA, Lloyd GT (2014) strap: An R package for plotting phylogenies against stra- 447–468. tigraphy and assessing their stratigraphic congruence. Palaeontology 58(2):379–389. EVOLUTION

Thompson et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5877 Downloaded by guest on September 28, 2021