In vertebrates, neural crest (NC) cells are a transient embryonic population that derive from the border between the neural plate and the non-neural ectoderm. They subsequently migrate throughout the embryo to form multiple derivates, including neurons and glia of the peripheral nervous system, melanocytes, and cartilage and bone of the face (Knecht and Bronner-Fraser,2002; Le Douarin,1982). NC cells form at the region of the neural plate border, and it has been demonstrated that interactions between the neural and non-neural ectoderm are required for cell fate acquisition (Mancilla and Mayor,1996; Mayor et al.,1995; Selleck and Bronner-Fraser,1995). Once specified, NC cells express foxd3, snail2, sox10, and ap-2alpha, and then express sox9, sox10, and crestin as they migrate to their final destination (Meulemans and Bronner-Fraser,2004). In zebrafish and Xenopus, Rohon-Beard (RB) sensory neurons are also born within the border domain and require a similar inductive mechanism as the NC, but do not migrate; RB neurons remain in the dorsal spinal cord and function as proprioceptive sensory neurons which mediate the mechanosensory touch response (Lamborghini,1980; Rossi et al.,2008). RB neurons express several genes required for development of all primary neurons including huC, neurog1, dlx3b/4b, neuroD and islet1, and these genes are also required for RB neurons (Rossi et al.,2009).
The molecular mechanisms by which NC cells and RB sensory neurons are specified and differentiate remain unclear, but several genes that are required for these processes have been identified. Both cell fates require BMP signaling for their induction, where an intermediate level of BMP signaling is required at the neural plate border (Neave B and Patient,1997; Nguyen et al.,1998,2000). Notch signaling is also important for the segregation of NC cells and RB sensory neurons. These cell fates form within the same equivalence domain at the neural plate border and RB neuron cell fate is promoted at the expense of NC cell fates when Notch signaling is defective. This was demonstrated in several zebrafish mutations, including mind bomb and deltaA (Cornell and Eisen,2000; Itoh et al.,2003; Jiang et al.,1996) where Notch signaling is reduced. The gene regulatory network downstream of these primary inducers is currently being assembled, and includes many transcription factors required in specific lineages. For NC cells, sox10, a member of the Sry-related transcription factor family, plays an important role in NC development. Mutations in both mouse and zebrafish have defects in pigmentation and peripheral nervous system derivatives (Britsch et al.,2001; Carney et al.,2006; Dutton et al.,2001; Herbarth et al.,1998). Islet1 is a LIM homeodomain transcription factor that plays a role in the specification and determination of a specific motorneuron subtype (Hutchinson and Eisen,2006; Pfaff et al.,1996). Although islet1 is expressed in RB sensory neurons, not much is known about the role of this protein in RB development.
Previous studies have identified the transcription factor, prdm1a, in the development of both NC and RB neurons. Zebrafish embryos with a mutation in prdm1a, known as narrowmindedm805 mutants, fail to develop RBs or display an escape response when touched on the trunk (Artinger et al.,1999; Hernandez-Lagunas et al.,2005). In addition, NC cells and their derivatives are reduced. The predicted Prdm1a protein in narrowminded mutants contains no DNA binding domain, thus preventing the ability of the protein to act as a transcription factor. prdm1a has been implicated in the development of several other cell types as well, including cells in craniofacial cartilage, slow twitch muscle cells, germ cells and immune cells (Bikoff et al.,2009; John and Garrett-Sinha,2009). In these systems, Prdm1a acts as a canonical transcriptional repressor to regulate cell fate (Ohinata et al.,2005). A conditional knockout of Blimp1 in the mouse, which is a homolog of prdm1a in zebrafish, has defects in posterior forelimb, secondary heart field, sensory vibrissae and, importantly, in caudal pharyngeal arches in which NC cells contribute to the head mesenchyme (Robertson et al.,2007). In zebrafish, prdm1a functions at several points in embryogenesis, including: during gastrulation, formation of head structures, and fin development (Mercader et al.,2006; Wilm and Solnica-Krezel,2005), and as a Hedgehog regulated switch between slow twitch and fast twitch muscle development (Baxendale et al.,2004; Elworthy et al.,2008; von Hofsten et al.,2008). The expression of prdm1a begins at midgastrulation at the neural plate border and continues to be expressed until around the six somite stage. By 24 h post fertilization (hpf), prdm1a is expressed in a large domain covering the posterior pharyngeal arches, which give rise to the posterior viscerocranium (Birkholz et al.,2009; Hernandez-Lagunas et al.,2005; Wilm and Solnica-Krezel,2005). While it is clear that prdm1a is required for the development of RB neurons and NC cells, the genes that act downstream of prdm1a are still unknown.
To begin to understand the genetic hierarchy of Prdm1a in NC, RB neuron and pharyngeal arch development, we determined the differential gene expression profiles between wild type and prdm1a mutant embryos focusing on 25 hpf because it is a key time point in the development of both NC cells and RB neurons. Using microarray analysis, we have identified potential downstream effectors of Prdm1a in the development of NC and RB neuron development. Further, in rescue experiments, we find that sox10 is a primary effector of prdm1a in the NC, while islet1 lies downstream of Prdm1a in the development of RB neurons.
Placing prdm1a in the Neural Crest and RB Sensory Neuron Gene Regulatory Network
To determine the genetic hierarchy of Prdm1a in the NC, RB and pharyngeal arch domain, we performed microarray analysis comparing gene expression in whole wildtype and prdm1a mutant zebrafish embryos at 25 hpf. Analysis revealed a large number of differentially expressed genes, including several genes specifically down- or upregulated in the NC, RB and pharyngeal arch domains. Based on its role as a transcriptional repressor, we expected that loss of Prdm1a would result in upregulated genes. Because we observe downregulated genes in the microarray, it suggests that Prdm1 may activate genes or that prdm1a represses genes that encode repressors. We confirmed the results of the microarray analysis by whole mount in situ hybridization. Here, we report our findings for a small number of genes expressed within the pharyngeal arches, RB neurons or in the NC (Table 1). We found in total 796 genes that are significantly (P < 0.05) upregulated and 1,197 genes that are significantly downregulated in prdm1a mutants at 25 hpf.
Table 1. Quantification of In Situ Hybridization Results
Mating between two heterozygote narrowminded fish will result in ∼25% embryos displaying a mutant phenotype. The percent affected describes the observed number of mutant embryo in which we observe either up or down regulated expression which is close to the expected number and between 22 and 27%.
Among the genes we identified as downregulated in prdm1a mutants, several known regulators of trunk NC cells and RB neurons. Islet1 is normally expressed in RB neurons, motor neurons, and a small subset of P2 interneurons (Inoue et al.,1994). In prdm1a mutant embryos, islet1 expression was lost specifically in the RB neuron domain, but its expression increased in the interneuron domain (Fig. 1a,b; arrows indicate interneurons). Islet2a and islet2b, both expressed in RB neurons of wild type embryos, exhibited decreased expression within the RB domain of prdm1a mutant embryos (Fig. 1c–f) (Appel et al.,1995) (Hutchinson and Eisen,2006). Similarly, expression of runx3, normally expressed in a subset of RB neurons (Horsfield et al.,2007; Park and Saint-Jeannet,2010) is severely reduced in prdm1a mutant embryos (Fig. 1g,h). The expression of several NC markers, including crestin (Luo et al.,2001) and rab32, a member of the ras oncogene family, was also reduced in prdm1a mutant embryos (Fig. 1i–l). Similarly, expression of the SRY-box containing gene 10 (sox10), normally strong in trunk and nonectomesenchymal cranial NC cells (Blentic et al.,2008; Dutton et al.,2001), was reduced dramatically in trunk NC cells of prdm1a mutant embryos (Fig. 1m,n).
Since prdm1a also functions in posterior pharyngeal arch development, we analyzed expression of genes in the craniofacial region that were downregulated in our microarray. At 25 hpf, hairy/enhancer-of-split related with YRPW motif 1(hey1) is expressed in the posterior somites (Winkler et al.,2003), pharyngeal arches, retina and fin mesenchyme. In prdm1a mutants, expression of hey1 was markedly reduced in each of these tissues (Fig. 2a–d). Integrin alpha 5 (itga5) is expressed throughout the pharyngeal arches (Crump et al.,2004) and caudal somites in wild type 25 hpf embryos, but was reduced in both the arches and caudal somites of prdm1a mutant embryos (Fig. 2e–h). distal-less homeobox gene 2a (dlx2a) is expressed throughout the pharyngeal arches (Kimmel and Eberhart,2008; Sperber and Dawid,2008) and fin mesenchyme. In prdm1a mutant embryos at 25 hpf, dlx2a was reduced in the anterior arches and absent in the posterior pharyngeal arches and dorsal fin mesenchyme (Fig. 2i–l) (Birkholz et al.,2009). Although reduced in trunk NC cells, crestin remained expressed throughout cranial NC cells in prdm1a mutant embryos (Fig. 2m–n). Expression of the chemokine Sdf1 and its receptor Cxcr4 were unaffected in the anterior arch in prdm1a mutant embryos (not shown; Olesnicky Killian et al.,2009); however, their expression in the lateral line, fin mesenchyme and caudal somites was reduced (Fig. 3a–d).
As expected based on Prdm1a's role as a transcriptional repressor, we observed increased expression of a number of genes in prdm1a mutants. Procollagen type IX, alpha 2 (col9a2), a chondrocyte specific marker (de Crombrugghe et al.,2000), is expressed in the pharyngeal arches and notochord of wild type embryos at 25 hpf. In prdm1a mutant embryos at 25 hpf, expression of col9a2 is increased within the pharyngeal arches and otic vesicle, with only a modest increase in expression in the caudal notochord (Fig. 4a–d). anterior gradient homolog 2 (agr2) is expressed within the otic vesicle and at low levels within the hatching gland of zebrafish embryos at 25 hpf (Shih et al.,2007). We found a dramatic increase in agr2 expression specifically within the hatching gland of prdm1a mutant embryos, while expression within the otic vesicle was unaffected (Fig. 4e,f). In addition, genes involved in muscle specification were upregulated in prdm1a mutants (Table 1). These results confirm genes previously known to be regulated by Prdm1a and identified new candidate Prdm1a target genes.
Sox10 is a Key Downstream Effector of prdm1a in the Neural Crest Development
sox10 has been shown to be integral to NC cell fate specification and subsequent differentiation in the zebrafish embryo. Expression of sox10 commences in premigratory NC and is required for differentiation of nonectomesenchymal cranial and trunk NC cells (Blentic et al.,2008; Carney et al.,2006; Dutton et al.,2001). Because sox10 expression is reduced in prdm1a mutant embryos, we tested whether sox10 is a downstream effector of prdm1a by asking whether sox10 can rescue the trunk NC cell defects of prdm1a mutant embryos, using crestin expression as a marker for NC cells. Injection of sox10 mRNA at the one-cell stage in wild type embryos results in an increase in NC cells throughout the trunk of the embryo at 25 hpf (Fig. 5a,b). Injection of sox10 mRNA into prdm1a mutant embryos rescues crestin-expressing NC cells throughout the trunk of the embryo (Fig. 5c,d). 26% (or 19 of 72) of prdm1a mutant embryos injected with control RNA exhibited mutant NC cell phenotype, comparable to uninjected mutants. However, following injection of sox10, only 9% (16 of 169) displayed mutant NC cell phenotype, indicating that sox10 rescued NC cells in the absence of prdm1a (Fig. 5n). To determine whether NC derivatives are also rescued, we examined pigment cells, which are decreased in prdm1a mutants. Injection of sox10 mRNA rescues NC-derived pigment cells in prdm1a mutant embryos, similar to what we observe for NC cells (Fig. 5e–j; 7% with 3 of 44 displaying the mutant phenotype). These results strongly suggest that sox10 is a key downstream effector of prdm1a in trunk NC cell specification. However, conserved domains within the sox10 enhancer region contained no canonical prdm1a binding sites, as previously described in prdm1a targets in mouse and zebrafish muscle (Lord et al.,2009; von Hofsten et al.,2008). This suggests that sox10 may not be a direct transcriptional target of prdm1a. rab32 was also decreased in prdm1a mutants; however, injection of rab32 mRNA did not rescue the NC cell phenotype (not shown).
Islet1 Functions Downstream of prdm1a in RB Sensory Neuron Development
Islet1 is expressed in RB neurons immediately after they appear (Inoue et al.,1994; Rossi et al.,2009) and depletion of islet1 via Morpholino injection results in a loss of primary motor neurons (Hutchinson and Eisen,2006). In these morphants, some RB sensory neurons fail to differentiate, suggesting that islet1 plays an important role in RB neuron development (J Eisen and S Hutchinson, personal communication). We therefore asked whether injection of islet1 mRNA can rescue RB neurons in prdm1a mutant embryos. In wildtype embryos, overexpression of islet1 mRNA does not induce more RB neurons (not shown). However, injection of islet1 mRNA into prdm1a mutant embryos at the one-cell stage results in a partial rescue of RB neurons at 25 hpf, assessed by antibody staining for the RB neuron marker, HNK-1 (Fig. 5k–m; quantification in Fig. 5o). Conserved domains within the islet1 enhancer contain two canonical prdm1a binding sites (GAAAG), suggesting that islet1 is a direct target of prdm1a. These results provide evidence that islet1 lies downstream and is likely a key effector of prdm1a in RB neuron development. By contrast, injection of islet2 or runx3 mRNAs did not rescue the RB phenotype of prdm1a mutants (not shown).
Prdm1a Overexpression Expands Sox10 and Islet1 Expression
Our rescue experiments provide evidence that sox10 and islet1 lie downstream of Prdm1a in the formation of NC cells and RB neurons, respectively. To determine the epistatic relationships, we asked whether overexpressing prdm1a in wild type zebrafish embryos increases expression of islet1 and sox10. Injection of prdm1a mRNA at the one-cell stage results in upregulation of both sox10 and islet1 expression at 25hpf. prdm1a overexpression results in ectopic cranial NC cells along the dorsal midline of the embryo (Fig. 6a,b; arrows). In the trunk, striking upregulation of sox10 expression appears in ectopic NC cells, which migrate as a sheet rather than in streams corresponding to each somite (Fig. 6c,d). prdm1a overexpression also increased islet1 expression specifically within the RB neuron domain, not within motorneurons or interneurons (Fig. 6e,f). These results are consistent with previous reports using crestin and HNK-1 staining to show increases in NC and RB neurons, respectively (Hernandez-Lagunas et al.,2005).
Our results demonstrate that Prdm1a functions upstream of islet1 in RB neuron development, while sox10 is the primary effector of prdm1a in NC development. Analysis of the prdm1a mutant phenotype in zebrafish reveals a variety of NC defects, including reduced peripheral nervous system derivatives and pigment cell number in addition to a complete loss of RB neurons (Artinger et al.,1999; Hernandez-Lagunas et al.,2005). Closer examination of prdm1a mutant embryos shows that cranial NC cells are initially reduced in number but at later time points recover to a number comparable to wild type. Nonetheless, the posterior pharyngeal arches fail to execute their normal developmental program, resulting in loss of the ceratobranchial skeletal elements (Birkholz et al.,2009). The results presented here confirm that cranial NC cells do reach the pharyngeal arches, as crestin expression in the head is unchanged between wild type and prdm1a mutants at 25 hpf. Instead, expression of genes that are important for condensing NC and craniofacial skeleton development, such as dlx2a and itga5, and the chondrocyte specific marker col9a2, are misregulated in prdm1a mutant embryos.
In contrast to cranial NC cells, the trunk crest cells in prdm1a mutant embryos remain reduced in prdm1a mutant embryos. This supports the idea that NC cell populations are differentially specified and maintained along the rostro-caudal axis. Other zebrafish mutants also show differential defects in NC specification along this axis. For example, mind bomb mutants have a more severe defect in NC development in the trunk then in the head (Itoh et al.,2003; Jiang et al.,1996). NC cells form normally in the cranial region of mind bomb mutant embryos whereas trunk NC are completely absent, replaced by an increase in RB sensory and other primary neurons (Cornell and Eisen,2000,2002). Other mutations such as foxd3, sox10, also show different affects along the rostro-caudal axis (Dutton et al.,2001; Li and Cornell,2007; Stewart et al.,2006). sox10 mutant embryos exhibit similar rostro-caudal defects in the nonectomesenchymal derivatives such as neurons, pigment, and glia. However, these mutant embryos show normal development of the craniofacial skeleton, a cranial NC derivative (Kelsh,2006). There are also examples of mutants that have similar affects along the entire rostrocaudal axis, such as ap-2 alpha, and embryos with a knockdown of ap-2 alpha + gamma via Morpholino injection, where both cranial and trunk NC are absent (Knight et al.,2003; Li and Cornell,2007; O'Brien et al.,2004). As Sox10 is reduced in prdm1a mutants, overexpression of prdm1a causes an increase in sox10 expression, suggesting that the prdm1a regulates sox10 during the development of NC cells. Using rescue experiments, we find that sox10 can rescue the prdm1a NC phenotype and thus lies downstream and is likely a key effector of Prdm1a in NC cell specification.
Consistent with prdm1a playing a role as a NPB specifier gene, RB sensory neurons, which are also derived from the NPB, are lost in prdm1a mutant embryos (Artinger et al.,1999; Hernandez-Lagunas et al.,2005; Roy and Ng,2004). We find that expression of runx3 and expression of members of the islet gene family including islet1, islet2a, and islet2b are lost within the RB domain of prdm1a mutant embryos. Interestingly, prdm1a mutant embryos also show an increase in islet1 expression within the ventral interneuron domain, suggesting that prdm1a might repress the interneuron cell fate and instead promote formation of RB sensory neurons. Rescue experiments suggest that islet1 plays a role downstream of Prdm1a in RB neuron specification, since islet1 expression in prdm1a mutants partially rescues RB neurons.
In conclusion, we have identified genes that play a role downstream of prdm1a in the specification of NC cells and RB neurons, within the developing zebrafish embryo. prdm1a is a key element in the gene regulatory networks responsible for both NC cells and RB sensory neurons.
Zebrafish were maintained according to Westerfield (1993) and staged by hours post fertilization (hpf) and morphology according to Kimmel (1995). The zebrafish prdm1a mutant has been previously described (Artinger et al.,1999; Birkholz et al.,2009; Hernandez-Lagunas et al.,2005; Rossi et al.,2009).
Single Embryo Genotyping and prdm1a Mutant Identification
Single embryo genotyping in prdm1a (narrowmindedm805) clutches was performed in the rescue experiments as previously described (Rossi et al.,2009). For microarray analysis and rescue experiments. Identification of mutants was based on phenotyping. At 25 hpf prdm1a mutant embryos can easily be identified by an obvious kinked tail and U shaped somite phenotype. Normally, a mating between two heterozygote narrow-minded fish will result in 25% mutant embryos. We scored embryos as rescued if we observed NC, pigment, or RB neurons in more than seven somite lengths. Table 1 describes the percentage of mutant embryos from such a cross and the number as expected of mutant embryo in which we observe either up or down regulated expression.
Embryo Manipulation and Analysis
Whole-mount in situ hybridization was adapted from Thisse and Thisse (1998) and Brent et al. (2003). Immunohistochemistry was performed as described (Ungos et al.,2003) and the following antibodies were used: HNK-1 antibody (Sigma) was used at a 1:1,000 dilution. For overexpression, the prdm1a ORF was cloned into the pCS2 vector. RNA was prepared using the mMessage mMachine capped RNA transcription kit (Ambion). About 60–100 pg of Capped RNA total was injected into one-cell-stage embryos together with rhodamine dextran for observation of efficiency of injection (molecular probes). For rescue experiments, 25 hpf prdm1a mutant embryos were identified by an obvious kinked tail and U-shaped somite phenotypes and/or by genotyping. About 100–200 pg of sox10 mRNA and 100–200 pg of islet1 mRNA were injected into one-cell stage embryos. At least three experiments in separate clutches were done for each experimental condition.
RNA was isolated from whole zebrafish embryos, three replicates each for wildtype and prdm1a mutant embryos, using the RNAqueous-Micro Micro Scale RNA Isolation kit (Ambion). Purity of each sample was determined based on the ratio of A260–A280. The integrity of total RNA samples was examined by Agilent 2100 Bioanalyzer. Total RNA was converted to double-stranded cDNA (ds-cDNA) using the cDNA synthesis kit (Affymetrix). ds-cDNA was purified and recovered using GeneChip sample cleanup module (Affymetrix). Three biological replicates were constructed for wildtype and prdm1a−/− embryos. In vitro transcription was performed to generate biotin-labeled cRNA using an RNA Transcript Labeling Kit (Affymetrix or Enzo, Farmingdale, NY). Biotin-labeled cRNA was purified using GeneChip sample cleanup module (Affymetrix). To ensure optimal hybridization to the oligonucleotide array, the cRNA was fragmented. Hybridization was performed by incubating 200 μL of the sample with Affymetrix Zebrafish GeneChip® arrays (Affymetrix, Santa Clara, CA). Arrays were read at a resolution of 2.5–3 microns using the GeneChip Scanner 3000 (Affymetrix).
All available raw gene expression data (probe-level) was taken from Affymetrix CEL files. The perfect-match (PM) data was background corrected, normalized, and summarized using the RMA (robust-multichip average) algorithm as previously described (Bolstad et al.,2003; Irizarry et al.,2003a,b).
A principal component analysis was performed on the normalized data to identify any outlier samples. No distinct outlier was determined. Prior to performing statistical analysis between the two groups, the normalized dataset was filtered for low variance genes across all samples. Genes with zero statistical variance to a variance P-value of 0.01 across all samples are considered to be flat, with no change in gene expression, and were filtered out from the dataset.
An ANOVA was performed on the filtered dataset to determine statistically significant gene regulation between the two experimental groups. The data was log base 2 (log2) transformed prior to running the ANOVA. After the ANOVA, the log2 ratios were converted into a linear scale fold change. P-values were calculated determining the most statistically significant gene changes. The dataset was then sorted by P-values and then fold changes to identify the genes with the most robust up- and downregulation between the two experimental groups.
We would like to thank ZIRC for reagents; Bruce Appel, Linda Barlow, and Lee Niswander for helpful comments on the manuscript; and Morgan Singleton for extraordinary fish care.