Geometric control of ciliated band regulatory states in the sea urchin embryo

The spatial specification of the ciliated band (CB) from the sea urchin embryo presents a uniquely challenging regulatory problem. During late embryogenesis this band consists of ciliated cells that facilitate feeding and locomotion in the larva (Fig. 1A). Not only do these cells overlie axonal tracts descending from apical neurons, but also the CB per se constitutes a neurogenic territory (Yaguchi et al., 2010; Angerer et al., 2011). Long before differentiation of these cell types occurs, prior to gastrulation the future position of the CB is foreshadowed by expression of the gene onecut, erroneously referred to earlier as hnf6 (Howard-Ashby et al., 2006), in a trapezoidal CB pattern four to five cells wide (Fig. 1B,C) (Otim et al., 2004; Poustka et al., 2004,, 2007). When viewed from the oral side (as in Fig. 1C) the band of onecut expression can be seen to border the oral ectoderm on either side, to abut the endoderm vegetally, and in its animalmost portion to include the oral part of the apical neurogenic field.

Previous studies concerning the spatial specification of this band of gene expression did not address its heterogeneous nature and have primarily focused on the influence of signals in positioning the band, in particular TGFβ family ligands and their antagonists, including Nodal and Bmp2/4 (Duboc et al., 2004,, 2008; Lapraz et al., 2009; Saudemont et al., 2010; Yaguchi et al., 2010; Angerer et al., 2011). However, it is not possible to elucidate from signal responses alone the transcriptional inputs into onecut, nor into the other regulatory genes that, as we show below, are also expressed in parallel within the CB. It is their transcriptional inputs, encoded directly in the genomic target site sequences of these genes, that govern their spatial expression. The genomically encoded CB gene expression pattern can only be understood by identifying these inputs. Spatial control of the CB gene expression pattern is unlikely to operate in a simple manner. As it traverses the embryo, this band forms the boundary between distinct, Boolean embryonic domains, each expressing a different regulatory state and each destined toward a different developmental fate. The various (indirect) effects of different signals and their antagonists on the disposition and breadth of the CB suggest that several different repressors might be responsible for confining its boundaries. The signals received in given territories can be expected to activate positively or negatively acting regulatory genes, but the diverse territories bordering the CB express unique regulatory states that result from different gene regulatory networks (GRNs) (Su et al., 2009; Li et al., 2012,, 2014; Ben-Tabou de-Leon et al., 2013; Materna et al., 2013); consequently, regional regulatory inputs into the CB are also likely to differ. The regulatory system affecting overall CB disposition in the embryo could thus be predicted a priori to be complex. But just how complex, as revealed in the following studies, we were nonetheless surprised to discover.

Here we show that as gastrulation commences the CB regulatory state includes more genes than onecut alone and, more significantly, that gene expression in the CB is mosaic. Each of the four regions of the CB expresses a unique regulatory state by 24 hours post fertilization (hpf), yet all express onecut and z166 (supplementary material Fig. S1F,J). Therefore, multiple boundaries of CB gene expression have to be accounted for, in that each of the four regulatory states of the band confronts a unique pair of distinct GRNs operating across its borders.

It is interesting to consider a priori the regulatory constraints on CB positioning. Each of the territories it borders has a distinct, known regulatory state. Indeed, as Fig. 2 shows, the expression domains of a number of regulatory genes specific to the territories that border the CB directly abut the transcriptional territories where onecut is expressed. Fig. 2A,B provide a map of CB subdomains that we utilize in the following, while Fig. 2D-I show double RNA in situ hybridizations that illustrate this point: Fig. 2D,F demonstrate that the aboral ectoderm regulator irxa directly abuts the onecut stripe in subdomains 1 and 2; Fig. 2G,I demonstrate that the oral ectoderm regulator gsc directly abuts the onecut pattern in all subdomains; and Fig. 2E,H illustrate by reference to the Veg2 endoderm regulator foxa that subdomain 4 is coincident with Veg1 ectoderm but excluded from Veg1 endoderm (see legend). With respect to the domain of onecut expression per se, which is the subject of Fig. 2, there are two possible regulatory scenarios. Upstream regulatory genes expressed in the same pattern as onecut, or in respective portions of its overall pattern, could drive its expression; or, it could obey a broadly distributed activator and have its boundaries set by the cis-regulatory action of repressors emanating from the various abutting domains. In the following we exclude the first of these alternatives and demonstrate the second (additional evidence pertaining to onecut per se will be presented in a forthcoming cis-regulatory study). With respect to the control of gene expression within the CB, again several possibilities present themselves. The transcription factor onecut could provide the spatial information required for the expression of additional CB genes. Alternatively, these genes might be regulated independently of onecut, or a combination of both possibilities might occur, depending on the gene in question. As it turns out, the last is correct, although the second scenario is the more prominent control strategy.

Almost all regulatory genes (here, genes encoding sequence-specific transcription factors) expressed in the embryo up to the stage of gastrulation relevant to this work are known, and are included in our experimental network analyses. To our knowledge, the expression matrix shown in Fig. 3 for the four subdomains of the 35 hpf CB is complete, or nearly so, with respect to the genes expressed within the various territories of the CB, except for subdomain 1. That is, mature GRNs have been published for all the territories of the embryo that border the CB, except for the apical domain through which subdomain 1 passes (Peter et al., 2012; Ben-Tabou de-Leon et al., 2013; Li et al., 2013,, 2014). Spatial expression patterns of all the regulatory genes that are transcribed within the CB up to this time are summarized in Fig. 3, which also includes expression data pertaining to genes from immediately neighboring territories (see Fig. 3A,B for simplified embryo maps). The primary RNA in situ hybridization data on which the summary is based are shown in supplementary material Fig. S1.

The matrix in Fig. 3C provides the explicit regulatory states of the four regions of the CB up to mid-gastrula, delineated in Fig. 2A,B and Fig. 3B. Remarkably, of the many genes accounted for in this expression matrix, only three are expressed throughout the extent of the CB, as is onecut, namely: z166, which is expressed in the aboral mesoderm earlier in development (supplementary material Fig. S1J) (Ransick and Davidson, 2012); otxβ1/2, which is expressed in the endoderm earlier in development (Yuh et al., 2004; Peter and Davidson, 2011); and foxg, which is expressed broadly in the oral ectoderm earlier in development (Li et al., 2014). All of the other regulators transcribed in the CB are expressed only within particular subdomains. The CB is thus, in regulatory terms, a mosaic composed of four separate regulatory states: the oral apical CB (subdomain 1); the bilateral animal lateral CB (subdomain 2); the vegetal lateral CB (subdomain 3); and the vegetal oral CB (subdomain 4). These regulatory states are the sum of the active genes indicated in the respective columns of Fig. 3C. All are complex, indicating different regulatory functions within each subdomain.

The Nanostring nCounter codeset used for the following measurements contains probes for most Strongylocentrotus purpuratus genes that encode transcription factors and signaling ligands expressed up to mid-gastrula stage, except for uncharacterized zinc-finger genes, although all zinc-finger genes encoding proteins orthologous to known transcription factors were included (181 genes in total; for further details see supplementary material Methods and Table S1). The Nanostring instrument, as utilized in this study, provides simultaneous quantitative assessment of the numbers of molecules of each regulatory mRNA species found in control embryos, as compared with embryos of the same batch in which translation of a given regulatory gene has been blocked by introduction of morpholino substituted antisense oligonucleotides (MASOs). Representative results are shown in Fig. 4 (each experiment was multiply replicated, and both splice-blocking and translation-blocking MASOs were evaluated; see Materials and Methods).

Fig. 4A demonstrates that a large fraction of the genes comprising the regulatory states of the CB are responsive to suppression of SoxB1 levels. SoxB1 is a pan-ectodermal transcription factor that is known to affect many other regulatory genes of the oral and aboral ectoderm GRNs (Saudemont et al., 2010; Ben-Tabou de-Leon et al., 2013; Li et al., 2013,, 2014). The genes observed to be downregulated by soxb1 MASO are indicated by red dots and in black text (indicating specific relevance to this work) or in gray text (chiefly other ectodermal genes). As these analyses show, SoxB1 controls its own level of expression through negative feedback on the soxb1 gene, so that soxb1 MASO produces a large increase in the prevalence of its own mRNA. Furthermore, SoxB1 evidently represses otxβ1/2 transcription. However, most of its target genes utilize SoxB1 as a transcriptional activator. Two logical consequences follow from these observations. First, since SoxB1 is cleared from the endomesoderm during the blastula stage, which is a prerequisite for endomesodermal development to proceed (Kenny et al., 1999,, 2003), its target genes can only be expressed at normal levels in ectodermal, CB, or apical domain cells. This provides a spatial constraint on the expression of SoxB1 target genes adjacent to CB subdomains 3 and 4, both of which abut the Veg1 endoderm. Second, since soxb1 is expressed everywhere except within the endomesoderm at the time the CB becomes spatially established as a trapezoidal stripe of gene expression (Fig. 1 and Fig. 2A-C), it cannot provide spatial boundary information for subdomains 1 and 2, nor for the oral and aboral boundaries of subdomain 3, nor for the oral boundary of subdomain 4. All of these boundaries must therefore be established either by regionally expressed repressors or through positive control via an activator that is in turn confined by gastrula stage to the CB, such as onecut. To distinguish between these alternatives, we examined the consequence of blocking onecut expression on CB subdomain gene expression, again utilizing Nanostring assays. The other three genes expressed throughout the CB, i.e. z166, otxβ1/2 and foxg, are either not uniquely expressed there (as noted above) or their expression within the CB is subsequent to the expression of onecut. Therefore, these genes are unlikely candidates to execute early CB-specific spatial control functions. Interference with z166 expression has no effect on the abundance of any regulatory gene transcript represented in the Nanostring codeset (not shown), and we did not consider its role further.

Despite its precocious activation in the CB, a notable result is that the onecut gene does not affect the expression of any of the regulatory genes specifically expressed in any of the CB subdomains (Fig. 4B), except for onecut itself, z166 and otxβ1/2. This conclusion was cross-corroborated with two different types of MASO targeted against onecut mRNA (translation blocker and splice blocking). Expression of the deadringer (dri) gene is also downregulated by onecut MASO, as previously observed (Saudemont et al., 2010), but expression of dri is not exclusively confined to the CB by 35 hpf, as it is then transcribed throughout the oral ectoderm in addition to the CB. Only later in development does dri mRNA clear entirely from the oral ectoderm and become CB specific. The other effects of the onecut MASO are marginal, or they concern genes expressed in other embryonic territories (shown in gray in Fig. 4B). The onecut transcript is present maternally (supplementary material Fig. S2A) and is expressed prior to its zygotic expression in the CB (Otim et al., 2004; Oliveri et al., 2008). Since injected onecut MASO is present throughout development, early effects will be recovered here as well as CB-specific effects. Taken together, these results rule out the alternative scenario whereby the spatial pattern of CB gene expression in the four subdomains, each with its own boundaries, is regulated by Onecut. It is clear that this transcription factor does not execute the function of a pioneer gene responsible for delineating the spatial pattern of all subsequent CB gene expression.

A function that it does perform, however, is indicated in Fig. 4C,D. Here we see that if expression of the Onecut target gene otxβ1/2 is blocked, the level of foxg transcript is significantly decreased. Previous studies have shown that otxβ1/2 positively autoregulates (Yuh et al., 2004), and this can be seen as well in Fig. 4C. In turn, if expression of foxg is blocked (Fig. 4D) the only relevant effect is downregulation of otxβ1/2. Thus, onecut activates otxβ1/2, which in turn activates foxg, and foxg feeds back on otxβ1/2. This feedback loop will be self-perpetuating and, as seen in many other contexts, it can be predicted that it will act to stabilize the CB regulatory state, assuming that the genes of this feedback loop act upon other CB targets later in development. The effect of onecut on foxg is thus most likely indirect.

Prior work (Duboc et al., 2010; Li et al., 2013,, 2014) has identified several repressors of genes expressed in the CB, and these are specifically listed in the following section. Additional repressive relationships required for given boundaries of CB gene expression are shown in Fig. 5. The initial tier of images in Fig. 5 reveals the response of a onecut cis-regulatory construct to the oral repressor Gsc, substantiating at the DNA level an original report that onecut expression is excluded from the oral ectoderm by this repressor (Saudemont et al., 2010). In controls the mosaic expression of the onecut:GFP cis-regulatory module reporter in CB cells is shown in green, superimposed on an RNA in situ hybridization image of endogenous onecut transcript in red (Fig. 5Aa,a′). But if a cluster of Gsc target sites in this construct is mutated, GFP expression spreads to cells all over the oral face of the embryo (Fig. 5Ab,b′). Thus, we confirm that Gsc directly excludes onecut expression from the oral ectoderm, presumably as early as 24 hpf.

Fig. 5B shows representative results with different repressors that affect the boundaries of expression of subdomain 2 genes. In Fig. 5Ba we see that fgf9 expression is excluded from CB subdomain 1 by Foxq2; in Fig. 5Bb that fgf9 expression is excluded from the oral face by Not; and in Fig. 5Bc that fgf9 expression is excluded from subdomains 3 and 4 (Veg1 ectoderm) by Eve. In Fig. 5Bd we see that myc expression is excluded from the oral face by Not; and in Fig. 5Be that myc expression is also excluded from subdomain 4 by Eve. Similarly, tlx expression is excluded from the oral ectoderm by Not (Fig. 5Bf), from the aboral ectoderm by Msx (Fig. 5Bg), and from subdomains 3 and 4 by Eve (Fig. 5Bh). univin expression is also excluded from the aboral ectoderm by Msx (Fig. 5Bi). The results shown in Fig. 5 are all those in which spatial exclusion was observed, and negative results in which knockdown of regulatory gene expression had no effect on CB gene transcription domains are not shown. Repressions affecting expression of the key subdomain gene pax2/5/8 are illustrated in Fig. 5C. Here we see that expression of this gene is excluded from subdomain 4 and the oral face by Not; from the region vegetal to subdomain 3 (i.e. the Veg1 endoderm) by Hox11/13b; and from the aboral ectoderm by an unknown repressor downstream of Bmp2/4 signaling (Fig. 5Ca, Cb and Cc, respectively). Analysis of gene expression patterns (supplementary material Fig. S1) and kinetics (supplementary material Fig. S2) suggests that all of these spatial repressors operate as described above, from 24 hpf until 36 hpf.

A comprehensive list of interactions affecting CB genes of each of the four subdomains is assembled in Table 1. This includes the new observations reproduced in Figs 4 and 5, as well as evidence for specific interactions published previously (Saudemont et al., 2010; Li et al., 2013,, 2014). We summarize all of these results in Fig. 6. Here, the four regulatory state subdomains of the CB are represented individually. Impinging on the boundaries are the repressors for which there is evidence that, in their absence, that specific boundary is abrogated, so that expression of the indicated CB gene spreads into the normal domain of expression of the repressor. Obviously, the roster of repressors is not complete, as we lack the identities of the negative inputs that prevent ectopic expression of several of the genes. For instance, in subdomain 2, although we know that Msx represses tlx and univin expression on the aboral side, as does Irxa repress onecut, Msx does not repress emx, fgf9 or myc expression on the aboral side and whether this function is executed by Irxa remains unknown. A general feature revealed in this diagram is that the inside boundaries of all four domains, where they abut the oral ectoderm, are defined by only two oral ectoderm repressors, whereas the outer boundaries are controlled by diverse repressors (some of which remain to be identified). Note that in Fig. 6 there is one boundary that is not set negatively, and that is the vegetal boundary of onecut expression in subdomains 3 and 4. Here, the probable regulatory limitation precluding more vegetal expression is the absence of SoxB1, which, as discussed above, is cleared in the process of endomesoderm specification.

When the inputs of Fig. 5 and the CB subdomain regulatory states of Fig. 3 are integrated with the published ectoderm GRNs (Ben-Tabou de-Leon et al., 2013; Li et al., 2013,, 2014), as well as with not yet published but publically available gene interaction data (http://sugp.caltech.edu/endomes/#EctodermNetwork), a general clarification emerges. It becomes clear why in earlier studies CB boundaries and position within the embryo were found to be affected by perturbation of signaling (for references see Introduction). The reason is that the genes encoding the repressors that establish all of these regulatory state boundaries are direct and indirect transcriptional targets of Nodal and Bmp2/4 signals, in known GRN circuits. Thus, the genomic code that specifies the location of the CB and its boundaries resides physically in the cis-regulatory modules of the CB regulatory genes shown within the boxes of Fig. 6. Only one of these genes, onecut, has been the subject of direct cis-regulatory mutational experimentation to prove the point (Fig. 5A and our unpublished results), but from the other known circuits in which these same repressors operate it is unlikely, for the following reasons, that any of the repressive interactions pictured in Fig. 6 are indirect linkages.

First, it is known that all of these factors also act elsewhere as repressors: in addition to the above references, additional data are available for Not (Materna et al., 2013), Irxa (Saudemont et al., 2010) and Gsc (Angerer et al., 2001; Saudemont et al., 2010). Second, since they act as repressors, in order to function indirectly while retaining the end result of repressing the CB target genes this would require three tandem repressors in sequence, a feature that we have never seen in our GRNs, and neither the expression kinetics of these repressors (supplementary material Fig. S2) nor the wiring of the GRNs as thus far established is consistent with this possibility.

A further clarification is summarized diagrammatically in the GRN map of Fig. 7. Here we see that all of the repressors that generate the boundaries for CB subdomains 1, 2, 3 and 4 emanate from the GRN elements across those respective boundaries. In Fig. 7 the oral and aboral ectoderm GRNs themselves are not shown in detail, as they are all to be found elsewhere, and Fig. 7 merely includes the linkages just upstream of the respective repressors. Only the interactions within each of the four CB subdomains are explicitly complete. Note that all subdomains display the otxβ1/2-foxg feedback ‘motor’; downstream of Onecut expression. The main point illustrated in Fig. 7, however, is that the CB boundaries are, in each subdomain, the specific output of the GRNs creating the regional regulatory states and not the output of the CB subdomains themselves. The circuitry in Fig. 7 shows explicitly the code that results in the observed CB regulatory state geometry. Furthermore, as observed throughout ectodermal GRNs (Li et al., 2014), the general design principle is boundary formation by spatial repressors, which sharply confine the expression of regulatory genes driven by broadly distributed activators.

A general import of developmental GRNs is that spatial gene expression always demands a remarkable multiplicity of genomic regulatory transactions. This applies to the complex design of individual cis-regulatory modules that execute spatial gene expression, as in the recent illuminating example of the sparkling enhancer (Swanson et al., 2010). Even more generally, the same statement applies to the complex information-processing functions of network subcircuits that control spatial gene expression functions by setting boundaries and by installing exclusive spatial regulatory states (Peter and Davidson, 2015). Development of the CB displays both cis-regulatory and subcircuit levels of information processing – initially cis-regulatory, in the control of the trapezoid stripe of onecut gene expression (and the same could potentially be implied for the similar pattern of z166 expression). We describe how the processing of multiple inputs produces this pattern in a forthcoming cis-regulatory study of the onecut gene. Here, we focus on the network interaction level of control as it affects four cohorts of gene expression, each of which is confined to the single CB stripe by different combinations of repressive regulators. The spatial coincidence between the expression of the different regulatory genes of each CB subdomain is due to their response to a common set of repressors, some of which are also used by onecut. This is also the explanation for the coincidence between the CB stripe and the expression of all the later subdomain genes. One such example is provided by the Gsc repressor, which this and earlier work (Saudemont et al., 2010) showed prevents onecut expression in the oral ectoderm, but as we see here the same repressor controls other genes of all four CB subdomains as well.

These encoded, parallel, regulatory relationships ensure the common internal boundaries of the CB regulatory state genes. We can see quantitatively the minimal complexity of this multilevel spatial control system; minimal in that, as Fig. 6 shows, we do not yet know the repressive inputs to several of the CB genes. There are already eight different repressors at play that are required (Fig. 6 and ancillary references), probably directly, to make the CB regulatory state pattern by interacting with the cis-regulatory modules of at least a dozen regulatory genes. This example is paradigmatic: when deconvolved into its logical elements (Table 1) the control system consists of as many individual encoded interactions as there are specific regulatory jobs to be done.

Microinjection of Strongylocentrotus purpuratus zygotes was performed according to well-established protocols (McMahon et al., 1985). Eggs were fertilized in situ and zygotes injected (1 pl per zygote) with onecut:GFP cis-regulatory reporter construct or MASOs as follows: onecut reporter construct was injected at 1 ng/μl together with 10 ng HindIII-digested genomic carrier DNA in nuclease-free water; MASOs were injected at 300 µM.

The wild-type GFP reporter construct contained 311 bp of non-coding sequence, which comprises an enhancer located within the second intron of the onecut gene. This DNA sequence element, followed by the GFP coding sequence, was positioned upstream of the onecut basal promoter by means of fusion PCR. The mutated version was generated by replacing 52 bp of the enhancer harboring a cluster of consensus Gsc binding sites.

The efficacy and specificity of six out of ten MASOs used in this work were authenticated in previous studies, namely those targeting soxb1, bmp2/4, eve, hox11/13b (Peter and Davidson, 2011; Li et al., 2014), foxq2 (Yaguchi et al., 2008) and not (Li et al., 2012). Multiple alternative foxg MASOs were tested, but only a single translation blocker proved useful: 5′-ACTTCTTGCTAAATACCAAGGCGGA-3′. For msx, two translation blockers proved equally useful and produced similar results: 5′-TCGCT-TCAACAGTAATCAAGGATGA-3′ and 5′-TGCACGTCGATTCGATA-GAAGAAAA-3′; data shown in Fig. 5 were obtained using the first of these. For otxβ1/2, a single translation blocker proved useful: 5′-AATGGTGTAAGCCATGCTCGCTACC-3′. For knockdown of onecut expression, we identified both a splice-blocking (5′-CAAGTTTTTG-ACTGACTTACCAGCT-3′) and a translation-blocking (5′-AGCCAAC-TAACTCACTTGAAAGCAT-3′) MASO, which gave similar results, cross-validating one another.

Embryos were manually harvested at the desired developmental stage. Samples were prepared containing 100 embryos in 15 μl RLT Plus buffer (Qiagen) with 2-mercaptoethanol (1:100) and stored at −70°C. Once thawed, 5 μl lysate was processed following the nCounter manufacturer's instructions (Nanostring Technologies). Detailed information concerning the Nanostring probe set utilized in this study, normalization of mRNA raw code counts, and threshold parameters, is given in supplementary material Methods and Table S1. Relevant Nanostring data have been deposited at the Dryad Digital Repository under data identifier doi:10.5061/dryad.mb804.

Whole-mount RNA in situ hybridization was performed following our previously published method (Ransick, 2004). The probes used in this study were complementary to almost the entire coding sequence of all mRNAs targeted, the sequences of which are available at http://www.spbase.org:3838/quantdev (Tu et al., 2014).

We thank Eric Erkenbrack for the microphotograph shown in Fig. 1A and Klara Stefflova for assistance throughout figure preparation.