Zebrafish zic2 controls formation of periocular neural crest and choroid fissure morphogenesis

Abstract

The vertebrate retina develops in close proximity to the forebrain and neural crest-derived cartilages of the face and jaw. Coloboma, a congenital eye malformation, is associated with aberrant forebrain development (holoprosencephaly) and with craniofacial defects (frontonasal dysplasia) in humans, suggesting a critical role for cross-lineage interactions during retinal morphogenesis. ZIC2, a zinc-finger transcription factor, is linked to human holoprosencephaly. We have previously used morpholino assays to show zebrafish zic2 functions in the developing forebrain, retina and craniofacial cartilage. We now report that zebrafish with genetic lesions in zebrafish zic2 orthologs, zic2a and zic2b, develop with retinal coloboma and craniofacial anomalies. We demonstrate a requirement for zic2 in restricting pax2a expression and show evidence that zic2 function limits Hh signaling. RNA-seq transcriptome analysis identified an early requirement for zic2 in periocular neural crest as an activator of alx1, a transcription factor with essential roles in craniofacial and ocular morphogenesis in human and zebrafish. Collectively, these data establish zic2 mutant zebrafish as a powerful new genetic model for in-depth dissection of cell interactions and genetic controls during craniofacial complex development.

INTRODUCTION

The key portion of the vertebrate eye, the neural retina, begins its development as an integral part of the forebrain primordium. It evaginates to form bilateral optic vesicles that connect to the forebrain via the optic stalks (OS). Optic vesicles then fold into cup-like structures that briefly remain open at the site adjacent to OS, termed the choroid fissure (Bazin-Lopez et al., 2015; Gestri et al., 2012; Kwan, 2014; Schmitt and Dowling, 1994). The edges of the choroid fissure come together and fuse during normal development; failure of this process results in uveal coloboma, estimated to occur once in every 5,000 births (Williamson and FitzPatrick, 2014). Coloboma is a significant cause of congenital blindness, found in 3–11% of blind children (Onwochei et al., 2000). Despite its prevalence and debilitating consequences, the underlying molecular defects that cause coloboma are not well understood.

The choroid fissure forms in a complex environment that includes the adjacent forebrain and neural crest (NC)-derived mesenchymal cells on their way to becoming skeletal and vascular elements of the face and jaw. In zebrafish, NC cells that migrate in the anterior streams around the optic vesicle form the neurocranium (ethmoid plate and the trabeculae) (Schilling et al., 1999; Wada et al., 2005). The retina and OS signal to the anterior NC, directing it to its destinations (Eberhart et al., 2008; Kish et al., 2011; Swartz et al., 2011). There is emerging evidence for a reciprocal interaction, whereby NC cells signal back to the eye and brain (reviewed in (Bazin-Lopez et al., 2015; Gestri et al., 2012; Le Douarin et al., 2007). In humans, significant comorbidity has been reported between frontonasal dysplasia and coloboma (Wu et al., 2007), suggesting that NC plays a specific role in choroid fissure morphogenesis. The importance of this mechanism has only recently come to light and needs robust genetic models to be fully understood.

Dysregulation in several signaling pathways has been implicated in coloboma, with Hedgehog (Hh) signaling arguably the best characterized (Williamson and FitzPatrick, 2014). Disruption of Hh signaling also causes forebrain anomalies called holoprosencephaly (HPE)(Roessler et al., 1996; Roessler and Muenke, 2010), and facial dysmorphologies that range from hypertelorism (increased distance between eye orbits) attributed to increased Hh signaling to orofacial clefting caused by Hh reduction (Brugmann et al., 2010; Gongal et al., 2011; Young et al., 2010). To understand how Hh signaling controls these processes, it is necessary to examine the downstream effectors of Hh signaling in each developmental context.

zic2, a member of the conserved Zic family of zinc-finger transcription factors, is one such effector. zic2 plays a key role in brain morphogenesis, as indicated by the high incidence of zic2 mutations in human patients with HPE (Brown et al., 2001; Brown et al., 1998; Ribeiro et al., 2012; Roessler et al., 2009; Solomon et al., 2010). Extensive studies in mouse and Xenopus have demonstrated essential roles for zic2 in early neural development, namely, neural crest specification and neural tube closure (Elms et al., 2004; Elms et al., 2003; Houtmeyers et al., 2016; Nagai et al., 2000; Nagai et al., 1997; Nakata et al., 1997, 1998; Nyholm et al., 2009; Teslaa et al., 2013; Warr et al., 2008; Ybot-Gonzalez et al., 2007). Zic2 is also required for specification of embryonic stem cells, where it functions as an enhancer-binding cofactor in concert with the Mbd3-NuRD chromatin remodeling complex (Luo et al., 2015). Later in development, zic2 function is required for correct migration of cortical neurons (Murillo et al., 2015), for cerebellar granular neuron differentiation (Frank et al., 2015) and as a key determinant of ipsilateral vs contralateral projection in retinal ganglion cells (Escalante et al., 2013; Garcia-Frigola et al., 2008; Herrera et al., 2003),. In hypomorphic zic2 mice, defective retinal morphogenesis has been reported, but not characterized (Herrera et al., 2003). The underlying mechanism of its functions during development have been elusive until recently, when zic2 was found to inhibit canonical Wnt signaling (Fujimi et al., 2012; Pourebrahim et al., 2011), and to control forebrain morphogenesis via a direct interaction with Smad2/3 in the Nodal signal transduction pathway (Houtmeyers et al., 2016). To fully understand the mechanism of zic2 functions in the context of the developing embryo, it is essential that we dissect these functions further, and in more than one model organism.

Here we report that zebrafish with genetic lesions in zebrafish zic2 orthologs, zic2a and zic2b, develop with profound retinal and craniofacial anomalies, similar to those observed after transient depletion of zic2 by antisense morpholino oligos (Sanek et al., 2009; Teslaa et al., 2013). We show that zic2 function is required for the correct morphogenesis of the OS and for juxtaposing the edges of the fissure to allow its subsequent closure. We demonstrate a requirement for zic2 in restricting pax2a expression at the OS/ventral retina border, and show evidence of increased Hh signaling in the absence of zic2 function. Using RNA-seq-based transcriptome analysis, we confirm an early requirement for zic2 function in NC-derived pharyngeal and periocular neural crest, and identify a novel role for zic2 as a transcriptional activator of Alx1, a paired homeobox transcription factor with key functions in craniofacial and ocular morphogenesis in human and zebrafish embryos (Dee et al., 2013; Uz et al., 2010). Collectively, these data establish zic2 mutant zebrafish as a powerful new genetic model for in-depth dissection of the complex inter-lineage cell interactions and genetic controls during craniofacial complex development.

RESULTS

Zebrafish zic2 orthologs function redundantly during retinal and craniofacial morphogenesis

Zebrafish zic2 orthologs zic2a and zic2b, the only members of the Zic gene family duplicated in teleosts, reside on chromosomes 9 and 1, respectively. To build a genetic model of zic2-linked HPE in zebrafish, we set out to establish lines mutant at both loci. Toward this end, we obtained a mutagenic gene-trap insertion in the first coding exon of zic2a, zic2a^gbt133, isolated in a screen by Clark et al. (Clark et al., 2011) (Fig. S1). zic2a^gbt133 homozygous embryos develop normally for the first 2 weeks (data not shown), likely due to functional redundancy with its ortholog, zic2b. We used targeted mutagenesis with TALEN and CRISPR/Cas9 to generate frame-shift alleles at two distinct sites in the first exon of the zic2b locus (see Materials and Methods for details). Three mutant alleles were isolated that contained a 27-nt insertion (zic2b¹¹²⁷) or a 16-nt insertion (zic2b¹¹¹⁶) at the CRISPR target site, and a 4-nt deletion (zic2b^t104) at the TALEN target site (Fig. S1). zic2b homozygous mutants developed normally and were viable as adults despite the predicted absence of full-length zic2b protein; in contrast, ~5% of the embryos derived from a double heterozygous (zic2a+/−; zic2b+/−) incross exhibited retinal coloboma by 48 hpf (Fig. 1A–F; Table S1). The expected proportion of zic2a;zic2b homozygous mutants (zic2 mutants) is 6.25%. Affected embryos frequently presented with mild coloboma, defined here as a relatively small gap present in only one of the two retinae (Fig. 1B, G). Unexpectedly, a subset of embryos with coloboma exhibited periocular hemorrhage and edema, indicative of vascular deficits (Fig. 1C, F).

Figure 1. Zebrafish Zic2 is required during retinal morphogenesis.

A: normal retinal morphology. B: retina exhibiting mild coloboma (*). C: retina exhibiting severe coloboma with periocular hemorrhaging (**). D: normal retinal morphology. E: bilateral coloboma in a severely affected embryo (**). F: mild, unilateral coloboma in an affected embryo. G: Penetrance and expressivity of coloboma is increased in progeny that lack maternal zic2b, derived from zic2a^gbt133/+; zic2b^t104/zic2b^t104 parents, compared to those from double heterozygous (zic2a^gbt133/+; zic2b^t104/+) parents (see Table S1 for details). H, I: Both CRISPR- and TALEN-induced mutant alleles of zic2b are tightly associated with coloboma in MZ-zic2 embryos. Embryos in A–C are at 2–3 dpf, shown in lateral views, anterior to the left. Embryos in D–F are at 4 dpf, shown in anterior views, dorsal at the top.

We next asked if the maternal function of zic2 played a role during retinal development by assessing embryonic phenotypes in progeny from a cross between zic2a+/−; zic2b−/− parents, 25% of which are predicted to be zic2 mutants. 25% of these embryos exhibited coloboma by 2 dpf (Table S1); coloboma was primarily severe, i.e. bilateral with large ventral gaps (Fig. 1C, E, G), consistent with a requirement for maternal zic2b during retinal morphogenesis. To confirm that zic2 mutations were responsible for abnormal retinal morphogenesis, we genotyped representative embryos with and without coloboma individually (Fig. S2). This analysis showed that the majority of embryos with coloboma were zic2 mutants (going forward, we will refer to zic2 mutants derived from zic2b-/-mothers as MZ-zic2 mutants). Coloboma was also occasionally observed in maternally depleted embryos with one wildtype allele of zic2a (Fig 1H, I).

By 5 dpf, all embryos with coloboma exhibited profoundly hypoplastic craniofacial cartilages, both in the neurocranium and pharyngeal arches, and severe periocular and cranial edema (Fig. 2A, B; Table S2). Similar defects were observed in embryos produced by heterozygous parents and those from zic2a+/−; zic2b−/− parents (data not shown). Genotypic analysis revealed that this phenotype was restricted to zic2 mutants and embryos with one wildtype allele of zic2a (Fig. 2C,D). Collectively, these data clearly demonstrate a requirement for zygotic zic2 function in the developing retina and craniofacial cartilages, and an early contribution of maternal zic2b function to retinal morphogenesis.

Figure 2. Zebrafish Zic2 is required for craniofacial cartilage development.

A: normal neurocranium and branchial arches. B: hypoplastic, disorganized craniofacial cartilages in a zic2 mutant. C: Craniofacial defects are enriched in zic2 mutants derived from double heterozygous parents. D: In embryos that lack maternal zic2b, craniofacial defects are observed in zic2 mutants and in embryos with one wildtype copy of zic2a. Cartilage was visualized by staining with Alcian Blue. Embryos at 5 dpf are shown in ventral views, anterior to the left.

zic2 restricts expression of pax2a in the optic stalk

We next asked if ventral retinal patterning and/or morphology were disrupted in zic2 mutants at 24 hpf, prior to the first appearance of overt coloboma. The choroid fissure and the OS are marked by patterned expression of several homeobox transcription factors, including pax2a (Macdonald et al., 1995; Mui et al., 2005; Take-uchi et al., 2003). We examined pax2a expression in zic2 embryos using whole mount in situ hybridization (WISH). Embryos derived from zic2a+/−;zic2b+/− parents exhibited normal expression of pax2a overall with the exception of the OS domain, which was mispatterned in 12% of the embryos (Fig. 3A, B). Post-hoc genotyping confirmed that all the embryos with mispatterned pax2a were homozygous for zic2b (Fig. 3C). All embryos with mispatterned pax2a exhibited aberrant ventral retina (Fig. 3D–F), i.e. were also zic2a^gbt133 homozygous. We next applied confocal microscopy to examine distribution of the Pax2a epitope in 24 hpf MZ-zic2 mutants. Normal retina expressed Pax2a in the restricted portion of the retina adjacent to the choroid fissure (Fig. 3G). In MZ-zic2 mutants, retinal edges of the choroid fissure expressed Pax2a, but were separated by a large gap. The OS were abnormally wide and contained aberrantly high pax2a signal (Fig. 3H). We also noted an intense concentration of F-actin at the choroid fissure in normal siblings and an absence thereof in mutant retina, consistent with aberrant morphogenesis. These observations are consistent with our previous finding that pa2a is ectopically expressed in embryos transiently depleted of zic2a (Sanek et al., 2009). When examined in ventral cross-sections, MZ-zic2 mutants exhibited aberrant expansion of Pax2a both in the ventral retina and in the pre-optic diencephalon contiguous with the OS (Fig. 3I, J). Pax2a-expressing diencephalon appeared dysmorphic, with thinner walls and larger lumen than the equivalent structure in the unaffected siblings (Fig. 3K, L). Collectively, these observations indicate an early requirement for zic2 function during morphogenesis of both the OS/choroid fissure boundary and the adjacent diencephalon.

Figure 3. Pax2a expression is aberrant in MZ-zic2 mutants at 1 dpf.

pax2a expression at 1 dpf was visualized in embryos derived from zic2a^gbt133/+; zic2b^uw1116/+ parents using WISH (A–F) or in progeny of zic2a^gbt133/+; zic2b^uwt104 parents using immunohistochemistry (G–L). A: normal pax2a expression in the ventral retina (*). B: mispatterned pax2a expression (*) was observed in 12 out of 103 embryos (12%, 2 expts.). C: Only zic2b homozygous embryos exhibit pax2a mispatterning. zic2a genotype was not tested because PCR genotyping was not robust after WISH. D–F: Embryos with mispatterned pax2a expression also exhibit coloboma, indicative of homozygosity for zic2a^gbt133. G, H: confocal stacks through representative retina of normal (G) and zic2 mutant (H) retina. I, J: confocal stacks through the ventral aspects of a normal (I) and zic2 mutant (J) diencephalon and retina. Arrowheads in H, J point to the aberrant optic stalk. In G–J, yellow = Pax2a, magenta = F-actin cytoskeleton visualized by phalloidin. K, L: single confocal sections through representative normal (K) and zic2 mutant (L) embryos, imaged ventrally at the level of choroid fissure. magenta = Pax2a; yellow = acetylated tubulin; cyan = nuclei visualized by DAPI. Embryos are shown in lateral views, anterior to the left (A–F) or anterior to the right (G, H); in ventral views with anterior at the top (I–L).

Hedgehog growth factors secreted from the ventral diencephalic midline pattern the OS and retina, partitioning it into three domains: the OS, ventral retina and dorsal retina (Ekker et al., 1995; Lee et al., 2008a; Lupo et al., 2005; Schimmenti et al., 2003; Varga et al., 2001; Wang et al., 2015). Since pax2a requires Hh signaling for its expression, pax2a expansion in zic2 mutants may indicate aberrant levels of Hh or an aberrant downstream transcriptional response. Zebrafish that lack the functional Hh receptor blowout/ptc1 due to mutations show increased levels Hh signaling and develop with incompletely penetrant coloboma; this defect is efficiently rescued by exposure to low levels of cyclopamine, a small molecule that inhibits Hh signaling (Lee et al., 2008a). Cyclopamine treatment also rescues coloboma caused by knockdown of sox4 or sox11, transcription factors that function as inhibitory modulators of Hh signaling during retinal morphogenesis (Pillai-Kastoori et al., 2014; Wen et al., 2015). We reasoned that, if Hh signaling is expanded in zic2 mutants, inhibition of Hh signaling should reduce the penetrance and/or expressivity of coloboma. To test this prediction, we exposed progeny from zic2a+/−; zic2b−/− incrosses to low concentration of cyclopamine that was sufficient to rescue coloboma in blowout/ptc1 mutants (Lee et al., 2008a). When progeny from zic2a+/−; zic2b^t104 parents were exposed to cyclopamine during gastrulation and somitogenesis, they developed with significantly milder coloboma than their vehicle-treated siblings (Fisher Exact test, p <0.02, Fig. 4A–E, Fig. S3). The overall morphology of the embryos was not affected at this cyclopamine concentration (3–4.5 μM). To test specificity of the rescue, we asked if cyclopamine rescue is allele-independent. Progeny from zic2a+/−; zic2b^uw1116 parents exhibited significant alleviation of coloboma phenotype after cyclopamine treatment (Fisher Exact test, p <0.04; Fig. 4E, Fig. S3). In contrast, cyclopamine did not significantly affect coloboma penetrance or expressivity in zygotic zic2 mutants (Fig. 4F, Fig. S3). These findings are consistent with the notion that Hh signaling is de-repressed in the absence of functional zic2 and that this de-repression contributes to retinal dysmorphology in zic2 mutants.

Figure 4. Cyclopamine treatment reduces frequency and severity of coloboma in zic2 mutants.

A: normal retinal morphology; B: retina with mild coloboma; C: retina with moderate coloboma. D: Embryos were derived from zic2a^gbt133/+; zic2b ^t104 parental crosses and exposed to 3 or 4.5 μM cyclopamine (CyA) from 3–5 hpf until 24–26 hpf. In CyA-treated groups (3 expts; Fig. S3A), the proportion of embryos with coloboma was reduced significantly compared to vehicle-treated control siblings (Fisher’s Exact test, P < 0.001). Proportion of severely affected embryos among all embryos with coloboma was also decreased in CyA-treated siblings (Fisher Exact test, p<0.02). E: Embryos were derived from zic2a^gbt133/+; zic2b ^uw1116 parents, and treated starting at 3 hpf with 4.5 uM Cya. CyA-treated groups exhibited reduction in coloboma penetrance (Fisher’s Exact test p<0.04) compared to vehicle-treated control siblings (2 expts; Fig. S3B). F: Embryos were derived from zic2a^gbt133/+; zic2b ^uw1116/+ parents, and treated as in D with 4.5 uM CyA or vehicle starting at 3 hpf. Proportion of embryos with coloboma was not affected by exposure to cyclopamine (2 expts). Embryos with unilateral mild coloboma were scored as “mild”; embryos with bilateral mild coloboma were scored as “moderate”, and embryos with bilateral moderate coloboma were scored as “severe”.

Zic2 function is required for neural crest-derived craniofacial lineage formation

We next set out to identify downstream effectors of zic2, i.e. genes whose transcription levels depend on zic2 function, using RNA-seq transcriptomic analysis (Fig. 5A). Embryos derived from zic2a+/−; zic2b+/− parents were sorted by coloboma at 25–28 hpf, when coloboma first manifests in zic2 mutants. RNA was isolated from individual embryos without pooling to increase the statistical power of analysis and subjected to high-throughput sequencing (see Materials and Methods for details). RNA-seq was performed on 10 samples in total, 5 normal and 5 with coloboma. One of the wild type samples was determined to be a mutant and excluded from further analysis. This approach identified a large set of 362 genes differentially expressed in zic2 mutants (199 increased and 169 reduced) (Table S3). We used the online bioinformatics tool (Huang da et al., 2009) to sort the responsive gene list into categories based on tissue enrichment in zebrafish (ZFIN_ANATOMY). This analysis identified myotome/somite, craniofacial elements, and heart (enrichment scores 2.78, 2.24 and 1.71, respectively). Myotome/somite markers and heart markers were found primarily in the upregulated set, consistent with expression of zic2b in zebrafish somites (Toyama et al., 2004) and with the demonstrated function for mammalian Zic2 during myogenesis (Inoue et al., 2007; Pan et al., 2011). In contrast, craniofacial markers appeared among transcripts depleted in zic2 mutants. (Fig. 5A). This set included chondrogenic neural crest markers dlx1a, dlx2a, dlx4a and barx1, and xanthophore lineage markers aox5, gch2 and cax1 (references in Table S4). We have previously shown both lineages to be strongly dependent on zic2 function in morpholino assays (Teslaa et al., 2013). A small number of retinal markers were also reduced in zic2 mutants, namely, vax1 and vax2, Hh-regulated OS/ventral retina markers (Take-uchi et al., 2003), and atoh7, expressed in retinal ganglion cells (Masai et al., 2000). Notably, pax2a transcript levels were unchanged in zic2 mutants, perhaps because the mispatterning we have documented in Fig. 3 affects a small portion of its expression domain.

Figure 5. RNA sequencing transcriptome analysis identifies a set of Zic2-dependent targets.

A: Embryos derived from zic2a^gbt133/+; zic2b¹¹²⁷/+ parents were sorted by presence or absence of coloboma into zic2 mutants and sibling groups, respectively. RNA extracted from individual embryos (4 wildtype and 5 with coloboma) was used to prepare cDNA libraries for illumina high-throughput sequencing. Genes with assigned value of False-Discovery Rate below 0.05 were preliminarily selected. The heat-map color represents relative expression levels of differentially expressed genes; 4 out of 5 coloboma-representing libraries are shown to maintain visual balance with the 4 normal sibling samples. B, C: Representative sibling and zic2 mutant embryos derived from zic2a^gbt133/+;zic2b^uw1116/+ parents show normal dlx2a expression by WISH in the telencephalon and diencephalon. D: normal dlx2a expression in branchial arch primordia. E: depleted branchial arch dlx2a expression (arrowhead) was observed in 7 out of 127 (6%, 3 expts.) of embryos from this cross. F: Only zic2b⁻ homozygotes exhibited dlx2a reduction in branchial arch primordia. G: normal cldn11a expression adjacent to the optic stalk of embryo with normal retinal morphology. H: depleted cldn11a expression in zic2 mutant with coloboma. Embryos in B – E are shown in lateral views, anterior to the left. Embryos in G and H are shown in dorsal views, anterior to the left.

To validate these results, we used WISH on zic2 mutants and siblings. Dlx2a was specifically reduced in pharyngeal arch primordia at 24 hpf, but not in the tel- or diencephalon, in 6% of progeny from zic2a+/−;zic2b+/− parents (Fig. 5B–E), and this reduction was restricted to zic2b−/− embryos (Fig. 5F). Reduced expression of atoh7 and vax2 was also confirmed by WISH (Fig. S4).

Cldn11a, a tight junction component enriched in myelinating oligodendrocytes (Bronstein et al., 1997; Morita et al., 1999) was the most strongly depleted gene identified by RNA-seq (Fig. 5A). In zebrafish, cldn11a expression has been reported in vascular endothelium (Cannon et al., 2013). In contrast, we found cldn11a expression to be restricted to a small group of cells anterior to the retina, in close proximity to the choroid fissure, at 24 hpf (Fig. 5G). Consistent with RNA-seq results, cldn11a expression was nearly abolished in zic2 mutants assayed by WISH (Fig. 5H). Notably, cldn11 transcription requires zic2 function in the differentiating mammalian cerebellar ganglion cells (Frank et al., 2015).

Zic2 controls transcription of alx1 in the periocular neural crest

One of the candidate targets of zic2 identified by RNA-seq, Alx1, is expressed in chondrogenic neural crest and functions during retinal and craniofacial cartilage morphogenesis. Homozygous mutations in alx1 are associated with profound frontonasal dysplasia and microophthalmia in humans (Bertola et al., 2013; Uz et al., 2010), and zebrafish alx1 morphants develop with hypoplastic craniofacial cartilages and coloboma (Dee et al., 2013). WISH analysis corroborated depletion of alx1 transcript in zic2 mutants (Fig. 6A,B). We examined expression of alx1 during normal development to verify its restriction to neural crest. At 16 hpf, alx1 was expressed in the frontonasal neural crest, which forms the facial skeleton (Couly et al., 2002; Langenberg et al., 2008) (Fig. 6C, D). alx1 was expressed in the periocular region at 24 and 36 hpf (Fig. 6E, F) and in the ethmoid plate anlagen at 48 hpf (Fig. 6F). alx1 was also expressed in the prospective swim bladder, but this domain did not appear to be affected in zic2 mutants (not shown).

Figure 6. Alx1 is a novel target of zic2 in the periocular neural crest.

Embryos derived from zic2a^gbt133/+; zic2b^t104/zic2b^t104 parents were stained for alx1 expression by WISH. A: normal expression in periocular mesenchyme of sibling embryo. B: depleted expression in zic2 (arrow) in mutant embryo (39 out of 112 total, 2 expts). C–G: wild type embryos stained for alx1 expression by WISH. C–D: alx1 is expressed in frontonasal neural crest at 16 hpf. E, F: alx1 is expressed in periocular mesenchyme (*) at 24 hpf and 36 hpf. G: alx1 is expressed in the ethmoid plate (arrowhead) at 48 hpf. Embryos in A, B, C, and E are shown in lateral views, anterior to the left. Embryo in D is shown dorsally, anterior to the left. Embryos in F and G are shown in anterior views, ventral at the top.

To ask if alx1 depletion was indicative of a broader periocular mesenchyme deficit, we examined expression of crestin, an early general marker of neural crest (Langenberg et al., 2008) in MZ-zic2 mutants and siblings. We found the anterior stream of crestin-positive neural crest and its pharyngeal domain strongly reduced in zic2 mutants (Fig. 7A–E). In contrast, crestin-expressing cells in the trunk appeared unaffected (Fig. 7C, F).

Figure 7. Frontonasal and pharyngeal neural crest is depleted in MZ-zic2 mutants.

A–C: normal crestin expression in frontonasal and pharyngeal neural crest. D–F: depleted crestin expression in 14 of 55 embryos from a zic2a^gbt133/+; zic2b^t104/zic2b^t104 incross. Arrows point to periocular neural crest. Arrowhead points to pharyngeal arch expression. Embryos are shown in lateral views, anterior to the left.

Since crestin is not expressed in the ventral portion of periocular neural crest, we examined this NC population directly by high-resolution confocal imaging. MZ-zic2 mutants and their siblings were fixed at 24 hpf and stained for F-actin and DNA to visualize cell outlines and nuclei. In normal siblings, cells with mesenchymal appearance were observed in the periocular space adjacent to the choroid fissure; these were enriched in the proximal (closest to the OS) half of the optic cup (Fig. 8A–C). In contrast, the ventral periocular space was largely devoid of cells along the entire proximodistal axis of the mutant optic cup (Fig. 8D–F). Taken together, these data argue that zic2 plays a critical early role during periocular neural crest formation.

Figure 8. Ventral periocular neural crest is depleted in MZ-zic2 mutants.

Single confocal sections through optic cups of embryos derived from zic2a^gbt133/+; zic2b¹¹¹⁶/zic2b¹¹¹⁶ parents. A–C′: embryo with normal retinal morphology. D–F′: embryo with coloboma. Embryos were imaged in lateral mounts. Cyan = nuclei visualized by DAPI; yellow=F-actin cytoskeleton visualized by phalloidin. Arrowheads point to aberrant gap in the ventral retina (coloboma). Embryos are shown in lateral views, anterior to the right (A–C) or anterior to the left (E–G). B′, C′, E′, F′ are enlarged from B,C,E and F, respectively.

DISCUSSION

The data presented here establish the first genetic model of zic-linked birth defects in zebrafish and extends our understanding of how zic2 coordinates development of the craniofacial complex comprised of brain, retina and craniofacial cartilages. We show that zebrafish zic2a and zic2b function redundantly to promote Hh-dependent retinal morphogenesis, and demonstrate a requirement for both zygotic and maternal zic2 in controlling morphogenesis of the optic stalk and retina, and to restrict pax2a expression in this region. This study confirms a key early role for zic2 in neural crest formation and identifies a homeobox transcription factor Alx1 as a novel effector of zic2 function in the periocular neural crest.

Where does zic2 function to control eye morphogenesis?

Our transcriptome analysis identified an early requirement for zic2 function in a number of neural crest lineages, particularly in the cartilage precursors of the branchial arches and in periocular neural crest. Zebrafish zic2a and zic2b are predominantly expressed in the presumptive neural crest (Grinblat and Sive, 2001; Nyholm et al., 2007; Teslaa et al., 2013; Toyama et al., 2004); hence it is tempting to speculate that zic2 controls ventral retinal morphogenesis indirectly, via a primary function in neural crest. Alx1 is an attractive candidate effector of Zic2, since alx gene family members function in periocular neural crest to regulate retinal morphogenesis in human, mouse and zebrafish (Bertola et al., 2013; Dee et al., 2013; Lakhwani et al., 2010; Qu et al., 1999; Uz et al., 2010; Zhao et al., 1996).

Non-cell-autonomous roles for periocular neural crest in choroid fissure morphogenesis have been demonstrated in mouse mutants (Evans and Gage, 2005; Matt et al., 2008) and in zebrafish morpholino knockdowns (Lupo et al., 2011; McMahon et al., 2009). There is yet much to learn about the mechanisms of these cell interaction, e.g. which specific neural crest lineages are important, when and how they interact with the retina, and which genes direct these interactions. Addressing these questions in robust mutant-based models will be essential going forward, yet such models are currently few and far between. The zic2 mutant zebrafish is an important step toward filling this gap.

Genetic removal of maternal zic2b enhances penetrance and expressivity of retinal coloboma in zic2 mutant embryos (Fig.1), suggesting a potential requirement for zebrafish zic2 function during gastrulation. Zic2a and 2b are expressed in the gastrula mesoderm (Drummond et al., 2013; Grinblat and Sive, 2001; Toyama et al., 2004), including the prechordal plate, which induces formation of the hypothalamus in the ventral diencephalon (Mathieu et al., 2002; Pera and Kessel, 1997; Rubenstein et al., 1998). The hypothalamus subsequently becomes an important source of Shh and is required for regionalization of the anterior diencephalon and optic fissure (Shimamura and Rubenstein, 1997; Zhao et al., 2012). It is therefore feasible that zic2 functions in the prechordal plate during gastrulation to promote ventral diencephalic/hypothalamic specification and the establishment of the hypothalamic signaling center. This hypothesis is supported by the partial rescue of coloboma by Hh inhibition in MZ-zic2 mutants. Notably, mouse zic2 promotes formation of the embryonic organizer during gastrulation (Barratt et al., 2014; Houtmeyers et al., 2016; Warr et al., 2008).

zic2a, but not zic2b, is also expressed in a restricted domain in the distal OS (Nyholm et al., 2007; Sanek et al., 2009; Toyama et al., 2004), where it may function to promote choroid fissure formation cell-autonomously. However, zic2b function should then be largely dispensable for normal retinal morphogenesis; this prediction is not born out by our results, which instead point toward a strict requirement for zic2b and a somewhat relaxed requirement for zic2a function during retinal morphogenesis (Fig. 1).

zic2a and zic2b are also expressed in the retina itself, as is zic2 in higher vertebrates (Brown et al., 2003; Nagai et al., 1997; Toyama et al., 2004). While this expression begins relatively late, ~ 24 hpf, it may contribute to retinal morphogenesis cell-autonomously. The timing of the zic2 mutant deficits described here, which manifest by 24 hpf, argues against a cell-autonomous function of zic2 in the retina. However, the accumulated evidence in mouse models warrants a detailed examination of retinal function of zebrafish zic2 (Garcia-Frigola et al., 2008; Herrera et al., 2003; Lee et al., 2008b). zic2 mutants in combination with powerful methodologies available in zebrafish, e.g. transplant assays, tissue-specific transgenesis and high-resolution live imaging, provide a robust platform for testing these hypotheses efficiently in future studies.

What is the molecular mechanism of zic2 function in the developing retina?

Misregulation of PAX2 has been causally linked to coloboma in humans (reviewed in (Gregory-Evans et al., 2004) and in chick (Sehgal et al., 2008). Regardless of the cell type where zic2 exerts its primary function, it is likely that misregulation of pax2a in zic2 mutants demonstrated here and in zic2 morphants (Sanek et al., 2009) contributes to their retinal anomalies; for this reason, it will be important to ask if this mechanism is conserved in mouse models.

Our demonstration that cyclopamine exposure ameliorates zic2-linked coloboma supports, albeit indirectly, the idea that Hh signaling upregulation is responsible for retinal defects in zic2 mutants. pax2a, a target of Hh signaling, is mispatterned, but is not reduced overall in zic2 mutants. Likewise, upregulation of Hh signaling is not detectable at the level of whole-transcriptome gene expression of known direct targets of Hh at the diencephalic midline, ptc1 and nkx2.2 (Bergeron et al., 2008). These genes are also expressed normally in zic2 mutants when assayed by WISH (data not shown).

An alternate hypothesis, consistent with the apparent de-repression of Hh signaling in zic2 mutants posits that Zic2 regulates transcription of Hh pathway components in a small portion of the embryo, such that would not be detected by our whole-embryo transcriptome approach. For example, it is plausible that a sub-lineage of the periocular neural crest modulates Hh signaling by producing secreted Hh inhibitors or creating physical barriers for Hh diffusion. It is also possible that zic2 restricts pax2a transcription via a parallel, Hh-independent mechanism. If this were the case, cyclopamine acting through Hh signaling may counteract pax2a expansion in zic2 mutants, thereby alleviating severity of coloboma observed in MZ-zic2 mutants. Our data are also consistent with the possibility that zic2 controls cell movements of pax2a-expressing cells rather than pax2a transcription. While the apparent increase in the number of pax2a-positive cells in the mutant diencephalon (Fig. 3G–I) argues against this hypothesis, additional studies are needed to test these hypotheses.

We have likely missed important targets of zic2 in our whole-embryo transcriptome analysis. In other contexts, zic2 has been shown to directly modulate Nodal signaling and canonical Wnt signaling (Fujimi et al., 2012; Houtmeyers et al., 2016; Murgan et al., 2015; Pourebrahim et al., 2011) and may function in this capacity in the developing zebrafish. Exciting recent data identify Zic2 as a key co-factor for chromatin remodeling in embryonic stem cells (Luo et al., 2015), and may function in this capacity in the developing embryos. Nonetheless, the broad-stroke approach taken here has correctly identified a number of cell lineages that depend on zic2 function, among them periocular neural crest, which is necessary for the formation and subsequent closure of the choroid fissure and whose migration is guided by the optic vesicle and by the optic stalks (Eberhart et al., 2008; Langenberg et al., 2008). Despite its limitations, this approach has led us to identify a strong candidate effector of zic2 function in retinal and craniofacial development, alx1. Going forward, RNA-seq to MZ-zic2 embryos at earlier stages of development will allow identification of a more complete and focused set of proximal zic2 targets and effectors.

How does this work inform our understanding of mammalian HPE and related disorders?

Loss-of-function alleles of ZIC2 are found in 10% of patients with the HPE (Brown et al., 2005; Solomon et al., 2010). zic2-linked HPE is unusual for two reasons. First, its penetrance in human patients is the highest of the common HPE-linked genes, 87%; by comparison, penetrance of HPE in patients with Shh mutations is only 36% (Solomon et al., 2012). Second, in contrast to Shh-linked HPE, facial structures of zic2-linked HPE patients are largely normal, although their cerebral morphology ranges from microform to severe alobar (Solomon et al., 2010). This suggests that the developing human forebrain is very sensitive to reduction in ZIC2 levels during human embryogenesis (in contrast, duplication of the ZIC2-containing region does not disrupt human development (Jobanputra et al., 2012).

Mouse and zebrafish embryos are less sensitive to zic2 depletion, since they develop normally when heterozygous for loss-of-function alleles of zic2. Coloboma is the most obvious defect in zebrafish that lack zic2, but diencephalic deficits are also present, as indicated by dysmorphic preoptic diencephalon (Fig. 3), narrowing of forebrain midline in the MZ-zic2 mutants (Fig. 1) and reduction of ventral diencephalic marker expression in zic2 mutants (e.g. lhx6 and nkx2.1; Fig. S3). In contrast, homozygous zic2 mouse mutants develop with prominent forebrain defects (Elms et al., 2003; Nagai et al., 2000). Gongal et al (Gongal et al., 2011) have proposed that HPE and coloboma represent mild and severe aspects of the same phenotypic spectrum; by this token, it is likely that the overt differences between mouse and zebrafish mutant phenotypes reflect quantitative rather than qualitative differences in brain primordium architecture in teleosts vs mammals. This argument further emphasizes the need for in-depth analysis of zic2 functions in more than one model organism.

It is important to note that zic2 double mutant phenotypes largely, but not completely, replicate the phenotypes observed after morpholino-mediated knockdown of zic2a and zic2b individually. The biggest difference between the assays is observed in the anterior diencephalon, which forms normally in MZ-zic2 mutants (Fig. 3), but is disrupted in zic2a morphants (Sanek and Grinblat, 2008; Sanek et al., 2009; Teslaa et al., 2013). This difference may indicate genetic compensation by other members of the zebrafish Zic family that function during brain morphogenesis (Elsen et al., 2008; Maurus and Harris, 2009; Winata et al., 2013) and retinal morphogenesis (Maurus and Harris, 2009). More generally, this study demonstrates the ability of closely related orthologs to compensate for each other’s functions when disrupted chromosomally, but not via transient knockdown. These data will contribute to the collective efforts to understand the mechanisms that underlie the well-documented differences in outcomes of gene disruption through transient knockdown and chromosomal lesions in target genes (Kok et al., 2015; Rossi et al., 2015).

Collectively, our data identify a novel role for zic2 in frontonasal/periocular neural crest development and establish a new animal model of inherited coloboma with frontonasal dysplasia. These data suggest that ZIC2 mutations may contribute to human conditions other than HPE, e.g. frontonasal dysplasia. Human hereditary coloboma frequently presents unilaterally, an indication that modifiers (genetic or environmental) are important contributors to choroid fissure formation. We find that coloboma in zygotic zic2 mutants is incompletely penetrant and predominantly unilateral, making this model ideally suited for modifier screens to identify molecular pathways that interact with zic2. This model will also facilitate in-depth analysis of other key roles for zic2, e.g. their functions in post-mitotic neurons such as cerebellar granule neurons (Frank et al., 2015) and Cajal-Retzius cells (Escalante et al., 2013; Murillo et al., 2015) and its potential link to schizophrenia (Hatayama et al., 2011).

MATERIALS AND METHODS

Zebrafish strains and embryo manipulation

Adult zebrafish were maintained according to established methods (Westerfield, 1993). All experimental protocols using zebrafish were approved by the University of Wisconsin Animal Care and Use Committee, and carried out in accordance with the institutional animal care protocols. Embryos were obtained from natural matings and staged according to (Kimmel et al., 1995). The following mutant strains of zebrafish were used: zic2a ^GBT133 insertional mutant (Clark et al., 2011); zic2b ^UWt104 mutant, generated by TALEN mutagenesis in the course of the study; zic2b^UW1127 and zic2b^UW1116 mutants, generated by CRISPR/Cas9 mutagenesis in the course of this study. Double zic2 mutants were obtained by crossing zic2a ^GBT133/+ zebrafish to each of the three zic2 mutant allele carriers; F1s were selected by RFP fluorescence to identify zic2a^GBT133 carriers, which express RFP (Clark et al., 2011) and raised to adulthood. Adult zic2b+/− zebrafish were identified by PCR genotyping of genomic DNA extracted from tail clips.

Cyclopamine treatments were carried out as follows: Cyclopamine (AdipoGen) was diluted in DMSO and added to E3 to final concentrations of 3uM or 4.5uM; final DMSO concentration was adjusted to 0.5%. Embryos were placed in E3 with cyclopamine or 0.5%DMSO for vehicle-only control. Treatments were started at 3hpf (before maternal-zygotic transition), 5 hpf or 6hpf. Embryos were moved to fresh E3 at 24 hpf and allowed to develop until 3–4 dpf, when they were assayed for retinal morphology.

Engineered nuclease mutagenesis and high-resolution melt analysis (HRMA)

Zic2b TALEN was designed and synthesized by the Mutation Generation and Detection Core (MGD) Facility, University of Utah to the following target left and right sites in exon 1, respectively: 5′-TCCTCTTGCGCAGCCGAGG-3′ and 5′-GGGGTGTTGTCCACTGGCCG-3.

Design of zic2b CRISPR site 5′-GGTGGAGTTAAAAGTGGAGC-3′ in exon1, mutagenesis and founder identifications were carried out as previously described ((Sedykh et al., 2016), with the following HRMA primers pairs: TALEN site - 5′-TGGACAACACCCCATCTTCA-3′ and 5′-GGATGTTTGGAGAGCCGTGAT-3; CRISPR site - 5′-TATTCTGCGGCCGCTCTT3′ and 5′-GGAGTCGAATCCCCAAATC-3′.

Sequencing and PCR genotyping of zic2 mutant alleles

To determine zic2a genotype, DNA was extracted from individual embryos or adult fish and subjected to PCR with the following primers: gbt forward 5′-CCCCGTAATGCAGAAGAAGA-3′, gbt reverse 5′-GTCCAGCTTGATGTCGGTCT-3′, wt forward and wt reverse 5′-ATTCATGGAGCCGTACTGGTTGTG-3′ and 5′-TGTTACTGGACGCAGGGCATCAGTT-3′. (see Supp. Fig. 2 for details).

Zic2b PCR fragments identified as mutant by HRMA were subcloned via TA cloning into pGEMT-Easy (Promega) and sequenced to characterize the mutations. Subsequently, PCR followed by Metaphor gel electrophoresis was used to efficiently genotype individual embryos and adult fish, HRMA primer sequence above were used for CRISPR allele genotyping. TALEN allele genotyping used 5′-GGACAACACCCCATCTTC-3′ and 5′-CGGGGAAAAGTAGGTGAC-3′ (Supp. Fig. 2).

RNA-seq transcriptome analysis

Embryos derived from a zic2a^GBT133/+; zic2b^UW1127/+ incross were sorted by presence or absence of coloboma. RNA was prepared from each individual embryo using the RNAEasy kit (Qiagen) according to (de Jong et al., 2010). cDNA libraries were prepared using the TruSeq stranded mRNA library preparation protocol with poly-A selection and sequenced on the Illumina HiSeq2500. Gene-level read counts were estimated from the raw sequencing data using RSEM v1.2.18 (Li and Dewey, 2011) and Bowtie v1.1.1 (Langmead et al., 2009). The gene set used consisted of all genes classified as protein coding or lincRNA within the Ensembl v77 annotation of the Zv9 assembly of the zebrafish genome. RSEM was run with option “–forward-prob 0” to take into account that the RNA-seq libraries were strand-specific. A matrix of gene-level counts from all libraries was compiled and analyzed for differential gene expression using the R statistical language and environment (R core team, 2014). Specifically, the count matrix was first pre-normalized using the median normalization routine from the EBSeq v1.5.4 package (Leng et al., 2013). The normalized dataset was then filtered to exclude genes that did not show coverage of at least 10 counts in at least 1 library across the entire dataset. The edgeR v3.14.0 package, (Robinson et al., 2010) with internal normalization switched off, was subsequently applied to call for differential expression (DE). Genes with assigned value of False Discovery Rate (FDR) below 0.05 by edgeR were preliminarily selected. Since a) low-expressed genes tend to be artificially enriched in the list of genes called DE by statistical algorithms and b) DE genes expressed at higher levels have more biological relevance and follow-up potential, we applied an additional filter to the edgeR output by retaining genes that have expression exceeding a certain quantile (0.2) of genome-wide distribution of expression values in at least 60% of libraries representing the strain with a larger mean expression of that gene.

Immunohistochemistry, in situ hybridization and Alcian Blue staining

Embryos were fixed in 4% paraformaldehyde (PFA) in PBS, or in 4% PFA/0.25% glutaraldehyde, 5mM EGTA, 0.2% TritonX-100, 1xPBS for optimal phalloidin staining. After PFA/glutaraldehyde fix, embryos were treated with 100mM sodium borohydride to reduce auto-fluorescence. Primary antibodies were detected fluorescently with Alexa-labeled goat anti-mouse or goat anti-rabbit secondary antibodies. Embryos were mounted in VectaShield and imaged on an Olympus IX81 inverted confocal microscope with the Fluoview 1000 confocal package, using a 60x water immersion objective (NA 1.10), a 60x oil immersion objective (NA 1.35) or a 20x objective (NA 0.75).

Antibody/stain reagent	Source	Dilution
Rabbit anti-pax2a	GeneTex, cat# GTX128127	1:500
Mouse anti-acetylated tubulin	Sigma, T6793	1:400
Goat Anti-Mouse Alexa 488	Invitrogen, cat#A-11001	1:500
Goat Anti-Rabbit Alexa 568	Invitrogen, cat#A-11011	1:500
phalloidin Alexa 488 or 568	Molecular Probes, cat#A12379; A12380	1:100
DAPI	Molecular Probes, cat#D21490	1:5000

In situ hybridization was carried out as previously described (Gillhouse et al., 2004), using the following probes: pax2a (Hoyle et al., 2004); dlx2a (Akimenko et al., 1994); crestin (Luo et al., 2001); vax2 (Gross and Dowling, 2005); atoh7 (Masai et al., 2000). cldn11a was synthesized as a gBlocks® Gene Fragments (IDT) and TA-cloned into pGEMT-Easy (Promega). Full length alx1 cDNA was amplified from total mRNA of 24 hpf embryos by PCR with primers 5′-TTGAGACGAGGCCAGAGGAC-3′ and 5′-CCTGGCTCTGTGAATAATTACAAG-3′ primers using OneTaq One-Step RT-PCR kit (NEB), and TA-cloned into pGEMT-Easy (Promega). After WISH, embryos were mounted in 100% glycerol and imaged on Axioskop2 Plus (Zeiss) compound or Leica MZ FLIII stereo microscopes equipped with Leica DFC310 FX camera and LAS v4.0 software. For cartilage staining, zebrafish larvae were fixed at 5–6 dpf in 4% paraformaldehyde and stained with Alcian Blue according to (Kimmel et al., 1998). Samples were flat-mounted in glycerol for imaging as described above.

Highlights.

Zic2 controls choroid fissure morphogenesis in zebrafish.

Zic2 is required for periocular mesenchyme formation.

Zic2 activates transcription of alx1, a transcription factor with essential functions during craniofacial and retinal development.