The zebrafish forkhead transcription factor Foxi1 specifies epibranchial placode-derived sensory neurons

Fish breeding and maintaining were performed as previously described(Guo et al., 1999b). Embryos were raised at 28.5°C and staged according to Kimmel et al.(Kimmel et al., 1995).

AB/EK female fish carrying the no soul mutation were crossed to wild-type WIK male fish, and F₁ progeny were raised to adulthood. Genomic DNA was extracted from pools of wild-type sibling and mutant embryos,and PCR reactions were performed using microsatellite marker primers(Knapik et al., 1998). The genetic linkage of no soul to microsatellite markers was established by detection of differential amplification of wild-type and mutant DNA pools(Knapik et al., 1998), and microsatellite marker Z1400 yielded two recombinants and marker Z13938 yielded 23 recombinants out of 276 individual no soul mutant embryos tested.

For mutation detection in no soul, primers specific for the foxi1 cDNA were used to amplify genomic DNA from pools (∼10) of the no soul mutant and wild-type sibling embryos. PCR products from mutant and wild-type sibling embryos (two independent sets) were directly sequenced by automated cycle sequencers (ABI).

Embryos carrying an islet-GFP (green fluorescent protein)transgene were anesthetized with 0.01% ethyl-m-aminobenzoate methanesulphonate(Sigma), and mounted in 3% methyl cellulose. They were viewed and photographed at different stages using a Bio-Rad confocoal microscope. After viewing, they were immediately returned to the 28.5°C incubator. The genotype of photographed embryos was subsequently confirmed by genotyping with tightly linked polymorphic Z markers.

Digoxigenin- or fluorescein-labeled antisense RNA probes were prepared from linearized templates using RNA labeling reagents (Boehringer Mannheim). Hybridization and detection with antidigoxigenin or anti-fluorescein antibodies was done as previously described(Guo et al., 1999b). For two-color in situ hybridization, two RNA probes (digoxigenin- or fluorescein-labeled) were hybridized simultaneously, and developed sequentially with purple and red substrates (Roche). After staining, embryos were cleared with glycerol, either whole-mounted or sectioned for viewing. Immunostaining was performed as previously described(Guo et al., 1999b). TUNEL staining was carried out using the ApopTag kit according to the manufacturer's instructions.

Capped sense RNA was synthesized by in vitro transcription from linearized pCS2 plasmids. DNA plasmids or in vitro transcribed RNA for either foxi1 or lacZ were microinjected at the 1- to 8-cell stage. At appropriate stages, they were fixed with 4% paraformaldehyde and processed for staining. Morpholino antisense oligonucleotide specific for foxi1 was provided by A. Fritz, the sequence is: 5′-TAATCC-GCTCTCCCTCCAGAAACAT-3′. Gene Tools, LLC standard control oligo was used. Oligos were injected at a concentration of 2-5 μg/μl.

A comparative model of the winged helix domain was constructed by MODELLER(Sali and Blundell, 1993),relying on its high (41-62%) sequence identities to the template structures of genesis (Marsden et al.,1997), FREAC-11 (van Dongen et al., 2000), and AFX (Weigelt et al., 2000) determined by NMR spectroscopy. The native and mutant 7-residue segments (57-63) were modeled de novo in the context of the rest of the comparative model (Fiser et al., 2000).

In a molecular marker-based genetic screen for mutations that affect the development of dopaminergic and noradrenergic neurons, we have recovered a mutant named no soul, which selectively disrupts one group of noradrenergic cells, the archassociated CA (AAC) cells(Guo et al., 1999b). Since AAC cells are thought to originate from epibranchial placodes, we examined if the epibranchial placode-derived visceral sensory neurons are affected by the mutation. We crossed the GFP, under the control of the islet 1promoter/enhancer, into the nosoul mutant background. The islet 1-GFP transgene allowed direct visualization of the epibranchial placode-derived sensory neurons, cranial motor neurons and their associated nerves, in living zebrafish(Higashijima et al., 2000). Time-lapse confocal analysis detected the emerging VIIth (geniculate) ganglia in the 26-hour postfertilization (hpf) wild-type sibling embryo (Fig. 1A,C). In the no soul mutant embryo, the VIIth ganglia were not detectable(Fig. 1B,D). By 48 hpf, while all three epibranchial-placode derived ganglia were observed in the wild-type embryo (Fig. 1E,G),all were missing in the no soul mutant embryo except the most distal division of the Xth ganglia, which appeared to be reduced (Fig. 1F,H). The defects persisted to later stages, suggesting that the no soulmutant suffers permanent deficits of epibranchial placode-derived visceral sensory neurons (Fig. 1I,J). In contrast to visceral sensory neurons, the cranial motor neurons appeared to develop normally in the no soul mutant (Fig. 1B,D,F,H,J). Taken together, these analyses indicate that the no soul mutation causes an eradication of the VIIth and IXth and a reduction of the Xth epibranchial placode-derived visceral sensory neurons.

In addition to Islet-GFP analysis, we also stained embryos with the monoclonal antibody mAb16 A11, which recognizes a specific epitope of the pan-neuronally expressed Hu proteins and labels cranial sensory ganglia in vertebrates (Marusich et al.,1994; Wakamatsu and Weston,1997). We found that consistent with the analysis with Islet-GFP,nascent neurons of the VIIth (geniculate) and IXth (petrosal) ganglia are absent, and the Xth (nodose) ganglia are reduced in the no soulmutant (Fig. 1K,L). In addition to epibranchial placode-derived ganglia, the antibody also labels other cranial ganglia including the trigeminal and lateral line ganglia(Raible and Kruse, 2000). The trigeminal ganglia (V), which are composed of cutaneous sensory neurons that innervate the face and jaws and other parts of the nervous system, appeared normal in the no soul mutant (Fig. 1K,L). The lateral line ganglia, which are involved in sensing water movement,however, are differentially affected in the no soul mutant: those that are in close association with the VIIth and IXth cranial nerves appeared to be absent, those in association with the Xth cranial nerve appeared to be reduced, whereas those that are near the Vth (trigeminal) ganglia appeared to be preserved (Fig. 1K,L). In addition to cranial sensory neuronal deficits, the no soul mutant is defective in subsets of pharyngeal arches(Guo et al., 1999b) and show variable defects of the otic vesicles (data not shown). They do not develop swim bladders and die around 6 dpf. Taken together, these analyses indicate that the no soul gene is required for the development of subsets of cranial sensory neurons and craniofacial structures. For the rest of the manuscript, we focused our analysis on the epibranchial placode-derived visceral sensory neurons to elucidate the mechanism of action of the no soul gene.

To identify the gene that is disrupted by the no soul mutation, we genetically mapped the mutation to linkage group (LG) 12, between polymorphic microsatellite markers Z1400 and Z13938. Primers for both markers amplified two distinct fragments using genomic DNA prepared from both parents and wild-type sibling embryos, while amplified only one of two fragments in DNA from no soul mutant embryos, indicating genetic linkage (data not shown). A winged helix domain-containing gene of the Foxi1 subfamily, which was identified in an expression cDNA screen(Kudoh et al., 2001), has been located at this interval by radiation hybrid mapping, and found to be essential for otic development (Solomon et al., 2003). Given the otic phenotypes we observed in the no soul mutant and the closeness in genetic distance between no soul and foxi1, we sought to determine if foxi1 is disrupted in the no soul mutant. We cloned and sequenced foxi1 from the no soul mutant embryos, as well as from their wild-type siblings. This analysis revealed that the no soul mutant carries a single substitution from serine to proline at amino acid 194 (Fig. 2A,C). This substitution was never found in wild-type siblings or in foxi1from other species (Fig. 2C). Furthermore, serine-194 is an absolutely conserved residue in helix H3, the recognition helix of the winged helix DNA binding domain(Gajiwala and Burley, 2000). Secondary structure prediction and comparative protein structure modeling(Sali and Blundell, 1993; Fiser et al., 2000) indicated that the change of the serine-194 to the helix-breaking proline residue shortens the critical recognition helix of the winged helix domain, which in turn is likely to destroy its ability to bind to DNA(Fig. 2B). It is probable that approximately one half of the crucial residues in the H3 recognition helix are not able to interact with DNA: the number of atomic contacts at less than 4.5 angstroms between the ab initio-modeled 7-residue segment and DNA in the no soul mutant is less than half of that in the wild type. In summary, both the experimental evidence and theoretical considerations provide strong evidence that the no soul mutation disrupts foxi1.

Scanning of sequence databases revealed that the no soul/foxi1sequences are most similar to the Xenopus foxi1c gene (47.3% overall identity), human HFH3 (46.9% overall identity) and mouse fkh10 (46.6% overall identity). The greatest similarity lies in the winged helix domain, whereas only limited similarity is detected outside this region (Fig. 2C). To verify that disruption of foxi1 is the cause of visceral sensory neuronal defects observed in the no soul mutant, we injected wild-type embryos with a morpholino (MO) antisense oligonucleotide(Nasevicius and Ekker, 2000)specific to foxi1. This foxi1-MO oligo but not the control oligo produced a phenotype that is indistinguishable from that of no soul: foxi1-MO-injected embryos had variable otic defects and were defective in subsets of pharyngeal arches (data not shown). Furthermore,the visceral sensory neuronal phenotype was also indistinguishable from that of the no soul mutant (compare Fig. 2D with Fig. 1F) in all the injected embryos (n=120), thus confirming that no soul is indeed foxi1. This result also suggests that the no soul mutation is likely to be a strong to complete loss-of-function allele of foxi1.

Embryological experiments have mapped the origin of cranial placodes to the border of neural plate and future epidermis(Baker and Bronner-Fraser,2001). At the tailbud stage (∼10 hpf), foxi1expression is found in bilaterally symmetric domains at the boundary of lateral neural plate/epidermis (Fig. 3A), suggesting that it is expressed in lateral cranial placodal progenitor cells. As the neural tube forms, the foxi1-expressing domains become elongated and by 24 hpf, the expression of foxi1 is somewhat reduced but still detectable (Fig. 3D). By 36 hpf, its expression is still present lateral to the neural tube (Fig. 3G). Cross sectioning showed foxi1 expression in the ectodermal layer. Weak expression was also detected in a few internal cells, which most likely represent delaminating neuronal precursors(Fig. 3G′). By 48 hpf, foxi1 expression domain has moved ventrolaterally, and appears to be in the pharyngeal arch primordia (Fig. 3M). Cross sectioning detected foxi1 expression again in the ectodermal layer (Fig. 3M′). Thus, the expression pattern of foxi1 is consistent with its role in cranial sensory neuron and pharyngeal arch development.

As mentioned earlier, visceral sensory neuronal lineage expresses two transcription regulators: ngn and phox2a. We compared their expression to that of foxi1 to determine their spatial and temporal relationships. A single zebrafish ngn gene, the ngn 1 is expressed in cranial sensory ganglia(Andermann et al., 2002; Blader et al., 1997; Cornell and Eisen, 2002; Korzh et al., 1998). Zebrafish phox2a/soulless was originally discovered in the same genetic screen that identified the no soul mutation, and is expressed in the epibranchial placode-derived visceral sensory ganglia(Guo et al., 1999a). In tailbud stage embryos, where foxi1 expression is detected in the lateral cranial placode progenitor cells, neither neurogenin nor phox2a is detected in this region (Fig. 3B,C),indicating that foxi1 expression precedes that of both ngnand phox2a. At 24 hpf, expression of both ngn and phox2a was initiated in the visceral sensory neuronal lineage (Fig. 3E,F)and such expression was more pronounced by 36 hpf (Fig. 3H,I). At this stage, foxi1, ngn, and phox2a are expressed in the vicinity of one another (Fig. 3J-L). The expression of ngn is also detected in other cranial sensory ganglia (Fig. 3E,H). By 48 hpf, ngn is weakly detected in a few cells and phox2ais expressed in many differentiating visceral sensory neurons (Fig. 3N,3O). Double in situ hybridization with foxi1 and ngn(Fig. 3P) or foxi1 and phox2a (Fig. 3Q,3R)confirmed that the foxi1-expressing domain overlaps with those of ngn and phox2a.

To further look into the relationship between foxi1, ngn and phox2a, we examined their expression in the no soul mutant. Since foxi1 expression precedes that of both ngn and phox2a foxi1 may be required to initiate their expression in the visceral sensory neuronal lineage. Alternatively, foxi1 may be required to maintain the integrity of progenitor cells, that is, in the absence of foxi1 activity, the progenitor cells disappear or are transformed to other cell types, thus, precluding their expression of ngn and phox2a and subsequent acquisition of visceral sensory neuronal identity. To look into these possibilities, we compared the state of progenitor cells and their expression of ngn and phox2a. Since the mutated no soul/foxi1 gene still gave rise to stable transcript, we used the foxi1 RNA to mark the presence of progenitor cells. At 24 hpf, the foxi1-expressing progenitor cells are present in the no soul mutant(Fig. 4B). In fact, they appeared to express foxi1 at a higher level compared to their wild-type siblings, suggesting that foxi1 negatively regulates its own expression (Fig. 4A-D). While the progenitor cells are still present in the no soul mutant at this stage, we detected no ngn or phox2a expression in the epibranchial placodes (Fig. 4E-H). Double in situ hybridization with foxi1 and ngn or foxi1 and phox2a further confirmed that while foxi1-expressing progenitor cells are still present, they fail to initiate the expression of ngn and phox2a(Fig. 4I-L). Consistent with the lack of ngn and phox2a expression, sectioning analysis of foxi1 and phox2a double-labeled embryos revealed that foxi1-expressing epibranchial placode progenitor cells fail to delaminate and subsequently fail to express phox2a(Fig. 4M-P). Taken together,this analysis suggests that in the absence of foxi1 activity,visceral sensory progenitor cells fail to initiate the expression of ngn and phox2a. In addition, the ngn expression in lateral line placodes was also absent in the no soul mutant(Fig. 4F), suggesting that foxi1 is also required to regulate the development of these placodes.

While foxi1-expressing progenitor cells are present in the 24 hpf no soul mutant embryos, by 36 hpf, it is clear that the anteriorly located foxi1-expressing progenitor cells were no longer present in the mutant (Fig. 5A,B). At this and later stages, the ngn- and phox2a-expressing ganglia were absent in the VIIth and IXth ganglionic regions, and much reduced in the Xth (Fig. 5C-F,I,J). The expression of phox2b, a downstream target gene of phox2ain some cell types and a regulator of phox2a in other cell types(Pattyn et al., 1997) is abolished in the VIIth and IXth ganglia. However, its expression in the Xth ganglion is largely unaffected by the mutation (Fig. 5G,H),suggesting that phox2b may be independent of foxi1 and phox2a in at least subsets of the nodose ganglion. Taken together,these analyses favor the hypothesis that foxi1 is required to initiate ngn and phox2a expression in the epibranchial placodes, and the failure to do so leads to the disappearance of progenitor cells.

To look into the disappearance of foxi1-expressing progenitor cells in the no soul mutant, we carried out TUNEL labeling to determine if these cells undergo apoptosis. As described above, the foxi1-expressing progenitor cells are still present in the no soul mutant at 24 hpf (Fig. 4B). At this stage, the pattern of cell death appeared comparable in the mutants and their wild-type siblings(Fig. 6A-D). However, 4 hours later, at 28 hpf, we observed an increased number of TUNEL⁺ cells in the vicinity of the VIIth and IXth ganglionic regions in the no soul mutant (Fig. 6E-H). Some TUNEL⁺ cells posterior to the otic vesicle and in the vicinity of the Xth ganglia were also detected. By 36 hpf, when the foxi1-expressing progenitor cells are absent in the region of the VIIth and IXth ganglionic regions, TUNEL⁺ cells were no longer detected in these regions. This is consistent with the observation that dead cell are cleared quite rapidly in zebrafish(Cole and Ross, 2001). Instead, we saw increased TUNEL⁺ cells in the Xth ganglionic region(Fig. 6I-L). This analysis suggests that upon the failure to initiate their normal developmental program,the progenitor cells undergo apoptosis.

Given that no soul/foxi1 is required to initiate the normal development of visceral sensory neurons, we next asked if it is sufficient to do so. We injected wild-type foxi1 RNA or DNA into one- to eight-cell stage zebrafish embryos and analyzed the expression of ngn and phox2a by in situ hybridization. At low concentrations (∼10 ng/μl for injected DNA constructs or ∼50 ng/μl for injected RNA),the injected embryos appeared morphologically normal, and we did not observe ectopic expression of either ngn or phox2a (data not shown; n=100). However, when we increased the concentration of foxi1 (∼30 ng/μl for injected DNA constructs or 200 ng/μl for RNA), ectopic phox2a⁺ (∼50%, n=50) and ngn⁺ (∼40%, n=50) cells were detected (Fig. 7C,D). In addition, the injected embryos showed morphological deformity. To determine whether the ectopic phox2a and ngn expression is due to foxi1 or is secondary to morphological deformity, we carried out double in situ hybridization with foxi1 and phox2a or foxi1 and ngn. Indeed, ectopic phox2a⁺and ngn⁺ cells were located in the region that ectopically expressed foxi1 (Fig. 7E,F). These data suggest that nosoul/foxi1 is sufficient to induce phox2a⁺ and ngn⁺ cells at ectopic locations. It is worth noting that the ectopic phox2a⁺ and ngn⁺ cells were largely confined to the yolk surface (nonneural ectoderm), and only a small number of cells became phox2a⁺ or ngn⁺ within the ectopic foxi1-expressing domains(Fig. 7C-F). These observations are consistent with the fact that not all endogenous foxi1-expressing cells became ngn⁺ and phox2a⁺, and further suggest that foxi1 needs to cooperate with additional signals to determine visceral sensory neurons.

The development of visceral sensory neurons represents a simple model system to examine the molecular mechanisms that control progenitor cell commitment and differentiation. Here we report that through a systematic genetic screen, we have identified a mutation named no soul, in which the development of visceral sensory neurons is severely disrupted. We further show that no soul encodes Foxi1, a winged helix domain-containing transcription regulator that is expressed in the placodal progenitor cells. Our analyses suggest that Foxi1 activity is required for the coordinated expression of both neuronal fate and subtype identity in the visceral sensory lineage: in the absence of foxi1 activity, the placodal progenitor cells fail to express both ngn and phox2a, and subsequently undergo apoptosis. Furthermore, ectopic foxi1 expression induces ectopic phox2a⁺ and ngn⁺ cells,suggesting that at least in some cellular contexts, foxi1 is sufficient to promote visceral sensory neuronal development(Fig. 7). Taken together, our work identifies an important upstream regulatory molecule that controls both pan-neuronal character and subtype identity in visceral sensory neurons(Fig. 8) and supports the emerging view that multi-faceted transcriptional cascades are crucial for the generation of diverse neuronal types from initially multipotent neural progenitor cells (Anderson and Jan,1997; Edlund and Jessell,1999).

Differential regulation of gene expression is essential to the control of development. In the case of epibranchial placodes, a few transcription factors have been identified, including the bHLH factor ngn, the paired homeodomain factor phox2a, and the homeodomain protein tlx(Andermann and Weinberg, 2001; Qian et al., 2001) that appears to act later in the differentiation of these neurons. Studies of ngn and phox2a reveal that ngn is required to determine generic neuronal identity (e.g. the ability to outgrow axons and dendrites) but does not affect the subtype identity (e.g. neurotransmitter phenotype and connectivity pattern) in the epibranchial placode-derived sensory neurons, whereas phox2a is essential for the subtype differentiation of epibranchial placode-derived visceral sensory neurons but not their pan-neuronal identity. These studies suggest that generic neuronal fate and neuronal subtype identity may be controlled by different subprograms in vertebrates (Anderson, 1999; Bertrand et al., 2002). They also raise the question as to what the upstream regulatory factors are and whether they would be required for the coordinated expression of these subprograms.

Whereas factors like olig2(Mizuguchi et al., 2001; Novitch et al., 2001; Zhou et al., 2001), Mash1 and phox2b (Pattyn et al., 1999) have been identified that coordinate pan-neuronal fate and subtype identity of motor neurons and autonomic ganglia respectively,molecules that carry out such function in the cranial sensory neuronal lineage have not been identified. Our study suggests that foxi1 is one such factor. In the absence of foxi1 activity, while foxi1-expressing progenitor cells are still present in the no soul mutant, they express neither ngn nor phox2a. Thus, foxi1 is required to coordinate ngn-mediated pan neuronal fate and phox2a-mediated subtype identity in placodal progenitor cells. Since foxi1 encodes a putative transcription regulator, a simple mechanism would be that foxi1 could directly activate the transcription of ngn and phox2a. The fact that the foxi1-expressing domain is much broader than that of ngn and phox2a suggests that foxi1 needs to cooperate with yet unidentified factors to turn on the expression of ngn and phox2a. Bmp7 from the pharyngeal endoderm has been shown to be an inducing factor for epibranchial placodes(Begbie et al., 1999). It will be important in the future to determine if foxi1 is regulated by Bmp7 in zebrafish.

It is worth noting that in the absence of foxi1 activity, placodal progenitor cells apparently undergo apoptosis. This observation suggests that foxi1 is involved in suppressing the apoptotic pathway in these progenitor cells. Since not all foxi1-expressing cells undergo apoptosis in the no soul mutant, it appears unlikely that foxi1 directly represses pro-apoptotic genes. Rather, apoptosis may be an alternative pathway when the normal developmental program is not initiated. Whereas it is not well understood how progenitor cells choose to differentiate or undergo apoptosis in vertebrates, it has been reported that progenitor cells can either undergo autonomic neurogenesis or apoptosis in culture in response to different concentrations of TGFβ(Hagedorn et al., 2000),suggesting that progenitor cells can adopt distinct fates in response to different extrinsic signals.

Winged helix domain (also known as the forkhead domain) was originally found in the Drosophila forkhead gene and the rat hepatocyte nuclear factor 3 (HNF-3), which function in determining terminal structures in the Drosophila embryo(Weigel et al., 1989), and in controlling gene expression in the liver in mice(Lai et al., 1990),respectively. Subsequently more than 100 winged helix domain-containing genes have been identified in species ranging from yeast to human, which serve such diverse functions as early patterning, differentiation and survival (refer to http://www.biology.pomona.edu/fox.html). X-ray and solution NMR structures of a number of winged helix proteins have been determined. These studies identify the H3 recognition helix as the most important DNA-binding portion of a winged helix(Gajiwala and Burley, 2000). Mutations in this region completely abolish DNA binding(Clevidence et al., 1993). Here we show that the no soul mutation lies in the H3 helix, changing a conserved serine to proline. Our comparative modeling analysis strongly suggests that the no soul mutation would disrupt DNA binding and thus the regulation of downstream target genes. Furthermore, the morpholino experiment phenocopied the no soul mutant phenotype. Taken together,we believe that the no soul mutation probably represents a severe to complete loss of function allele of foxi1.

Based on the similarity within the winged helix domain, the winged helix domain-containing proteins are further classified into more than 15 subfamilies. Relatively little is known about the expression pattern and function of the foxi1 subfamily. Our study with no souldemonstrates that foxi1 is required for the development of zebrafish epibranchial placode-derived visceral sensory neurons. The expression of Xenopus foxi1c has been recently reported(Pohl et al., 2002). Similar to our findings, the Xenopus Foxi1c is initially expressed in an epidermal ring around the neural field, and subsequently, is exclusively localized to placodal precursor cells. The targeted disruption of a probable homologue, Fkh10 (Foxi1), in mice leads to defects in inner ear development (Hulander et al.,1998). Otic defects were also observed in the no soulmutant (our unpublished observation). Furthermore, otic defects were well characterized in the hearsay mutation, which is another allele of foxi1 (Solomon et al.,2003). It will be interesting to determine whether Fkh10or a related forkhead gene is required for the development of visceral sensory neurons in mice.

Visceral sensory neurons include the geniculate, petrosal and nodose ganglia, which have distinct but also overlapping connectivity patterns. Fate mapping experiments show that they all derive from the epibranchial placodes(Baker and Bronner-Fraser,2001). However, in the no soul mutant, we saw a differential effect on these neurons: whereas the geniculate and petrosal neurons fail to develop, the nodose ganglia are partially spared. Interestingly, nodose ganglia were also less affected in mice with targeted disruption of ngn 2 (Fode et al.,1998) and phox2a(Morin et al., 1997). These analyses suggest that although the three distal ganglia share a developmental origin, different mechanisms may operate in their determination. Interestingly, we observed that unlike geniculate and petrosal ganglia, which express phox2a prior to phox2b, nodose ganglion initiates phox2b expression prior to that of phox2a. Therefore, it is possible that the commitment and differentiation of at least subsets of nodose ganglion is under the control of yet unidentified regulatory hierarchies. Alternatively, other neural progenitor populations are able to compensate for the loss of epibranchial placode-derived nodose progenitor cells. For example,compensation from cardiac neural crest cells has been previously reported in animals whose nodose precursors have been ablated(Harrison et al., 1995).

We are grateful to Keely Solomon, Drs Andrea Fritz and Igor Dawid for sharing information and reagents prior to publication. We wish to thank Dr Vladimir Korzh for the zebrafish neurogenin 1 plasmid, Drs Cori Bargmann, Yuh Nung Jan, Bingwei Lu and John Rubenstein for helpful comments on the manuscript, and members of the Guo laboratory for discussions. This work was supported by the Howard Hughes Medical Institute Biomedical Research Support Program grant to UCSF (#5300246). E.L.S. was supported by UCSF Neuroscience Postdoctoral Training Grant. S.G. is a Searle scholar, a Burroughs Wellcome Fund New Investigator, and a Fellow of the David and Lucile Packard Foundation.