The proneural gene ascl1a is required for endocrine differentiation and cell survival in the zebrafish adenohypophysis

The hypothalamic-hypophyseal axis constitutes a functional link between the nervous and the endocrine system. As a key regulator of hormone secretion, the hypophysis, also called pituitary gland, controls basic vital processes including growth, reproduction and overall body homeostasis. The mature pituitary consists of two functionally and anatomically distinct components:the neurohypophysis and the adenohypophysis, also named posterior and anterior pituitary lobe, respectively. In contrast to the neurohypophysis, which derives from ventral regions of the hypothalamus, the adenohypophysis originates from placodal ectoderm. At late gastrula and early segmentation stages, the placodal ectoderm is located at the interphase of the anterior neuroectoderm and the epidermal ectoderm, constituting a cell population that in several respects is the neural crest equivalent of the presumptive forebrain region. Anterior-most placodal ectoderm at the anterior neural ridge gives rise to the adenohypophysis, while more lateral and posterior regions of the placodal ectoderm give rise to sensory cells of nose, inner ear and lateral line, neurons of various cranial ganglia, and the lenses (for review,see Baker and Bronner-Fraser,2001).

The adenohypophyseal primordium itself develops into multiple endocrine cell types, classified by the hormones they produce, namely lactotropes,somatotropes, thyrotropes, gonadotropes, corticotropes and melanotropes. Despite some crucial differences in morphogenesis and architecture of the adenohypophysis (Herzog et al.,2003), the molecular machinery that controls pituitary organogenesis appears largely conserved among different vertebrate species,following a program integrating extrinsic and intrinsic control elements. Extrinsic signaling molecules shown to be essential for pituitary development both in mouse and zebrafish are the fibroblast growth factors Fgf3/Fgf10 from the ventral diencephalon (Herzog et al.,2004a; Ohuchi et al.,2000), and Sonic hedgehog from the oral ectoderm and/or the hypothalamus (Herzog et al.,2003; Sbrogna et al.,2003; Treier et al.,2001). Together with other signals such as Fgf8, Bone morphogenetic proteins (Bmps) and Wnts(Ericson et al., 1998; Treier et al., 1998; Treier and Rosenfeld, 1996),they provide trophic and positional cues to the adenohypophyseal primordium,regulating the spatial and temporal expression patterns of transcription factor genes within the primordium (Burgess et al., 2002; Burrows et al.,1999; Scully and Rosenfeld,2002; Zhu and Rosenfeld,2004). Identified intrinsic transcription factors in control of adenohypohyseal cell proliferation and differentiation belong to different protein classes, including LIM homeodomain proteins (Lhx3, Lhx4, Isl1),Bicoid-like homeodomain proteins (Pitx1, Pitx2), Sine oculis-related homeodomain proteins (Six1, Six3, Six6), Paired-like homeodomain proteins(Hesx1, Prop1), POU domain homeodomain proteins (Pit1; Pou1f1 - Zebrafish Information Network), zinc-finger proteins (Gata2), T-box proteins (Tbx19) and STAT proteins (Stat3).

By contrast, basic helix-loop-helix (bHLH) transcription factors have been less well studied in the context of pituitary development. Two genes encoding bHLH proteins have been reported to be expressed in the zebrafish, mouse and/or human pituitary: Mash1 and Neurod1(Allende and Weinberg, 1994; Ferretti et al., 2003; Guillemot and Joyner, 1993; Poulin et al., 2000; Wullimann and Mueller, 2002). They are homologues of proneural genes in Drosophila; Mash1 is a homologue of the achaete-scute (asc) complex genes, Neurod1 a homologue of atonal(Kageyama et al., 1997). Both Mash1 and Neurod1 are known to be prominent regulators of mammalian neurogenesis (Casarosa et al.,1999; Farah et al.,2000; Guillemot et al.,1993; Olson et al.,2001; Sommer et al.,1995). In the mouse adenohypophysis, Neurod1 expression is confined to corticotropes (Poulin et al., 2000; Poulin et al.,1997). Neurod1 has further been shown to bind to the proopiomelanocortin (Pomc) promoter, and to have a positive effect on Pomc transcription (Lamolet et al., 2004; Poulin et al.,2000; Poulin et al.,1997). Together, these findings suggested a requirement of Neurod1 for corticotroph differentiation. However, Neurod1-deficient mice just display delayed initiation of Pomc expression(Lamolet et al., 2004),indicating that Neurod1 is largely dispensable for corticotroph development,possibly because of functional redundancy with other transcription factors.

In contrast to Neurod1, the role of Mash1 in pituitary development remained elusive. Remarkably, aside from controlling neurogenesis in the central and peripheral nervous system (CNS, PNS), mouse Mash1 is crucial for the development of cells of the so-called diffuse neuroendocrine system, including adrenal medullary chromaffin cells(Huber et al., 2002), thyroid parafollicular C-cells (Lanigan et al.,1998) and pulmonary neuroendocrine cells(Ball, 2004; Borges et al., 1997; Ito et al., 2000). In light of these data and the common expression of Mash1 in the mammalian pituitary, it was speculated that Mash1 might play a similar role during the ontogeny of the adenohypophysis (Ferretti et al., 2003). However, genetic proof for such a function had been missing thus far.

Genetic screens in the zebrafish for mutations that disrupt adenohypophyseal growth hormone (gh) gene expression have led to the isolation of the pituitary absent (pia) mutant,characterized by a dearth of lactotrope-, corticotrope-, somatotrope- and thyrotrope-specific hormone expression during larval stages(Herzog et al., 2004b). Here,we report that the pia gene corresponds to zebrafish achaete-scute-like 1a (ascl1a; formerly called zash1a) (Allende and Weinberg,1994), providing genetic evidence for an essential and cell-autonomous role of this proneural gene during endocrine specification of adenohypophyseal cells. In the absence of Ascl1a, all adenohypophyseal cell types fail to express their cognate hormone genes, although only some of them become apoptotic. This is in contrast to the situation in fgf3mutants, where failed specification leads to the loss of all adenohypophyseal cells. In view of this, the ascl1a mutation allows us to distinguish two different populations of adenohypophyseal cells in which cell specification and cell maintenance are either coupled or independent from each other.

The pia^t25215 mutation was mapped to LG4 via bulked segregation analysis, using a panel of simple sequence length polymorphism(SSLP) markers and standard techniques(Geisler, 2002). Linkage assignments were confirmed and further refined by genotyping single mutant and wild-type embryos, which placed pia^t25215 close to SSLP marker z27201 (1 recombinant among 192 meiosis). In addition, we identified a new SSLP, amplifying a genomic fragment from contig BX511171.13 with primers 5′-GTACACTTGAAGCTTGTGCG-3′ and 5′-GTTTTCTGCACCAGAACCTG-3′. For this marker, we found no recombination among 192 meioses.

The acsl1a open reading frame is encoded by a single exon. To search for lesions in ascl1a, we amplified overlapping ascl1a fragments by PCR from genomic mutant and wild-type DNA samples, followed by sequencing in forward and reverse directions.

The pia^t25215 mutation generates a HindIII restriction site, which was used as a restriction fragment length polymorphism(RFLP) for genotyping. A 523 bp ascl1a fragment containing the polymorphic restriction site was amplified from genomic DNA of single embryos,using primers 5′-TCAGAGCATCCAACTCAGCC-3′ and 5′-CGAACGCTCAAAACCAGTTG-3′, followed by HindIII digest. On wild-type DNA, this results in two fragments of 203 and 320 bp, whereas the mutant PCR product is cleaved into three fragments of 320, 139 and 64 bp. For genotyping of pia homozygotes injected with ascl1a BAC DNA,we identified an RFLP in the 3′-UTR of ascl1a, distinguishing the wild-type allele from the pia^t25215 allele and the BAC-encoded gene. A 387 bp DNA fragment of BAC-injected embryos was amplified via PCR using primers 5′-AACAAGAGCTCCTGGACTTC-3′and 5′-CACGGTGTCGTGGAAAGTCT-3′, followed by StyI digest,which cleaved only the pia^t25215 ascl1a allele and the BAC-derived fragment (301 bp and 86 bp), but not the endogenous wild-type allele.

The antisense MO for ascl1a(5′-CATCTTGGCGGTGATGTCCATTTCG-3′; corresponding to nucleotides -4 to +21 of ascl1a cDNA) was obtained from Gene Tools and diluted in 1× Danieu's buffer to a final concentration of 0.033 pmol/nl. The MO was injected in volumes of 1 to 1.5 nl into embryos at the one- to four-cell stage, as described (Nasevicius and Ekker,2000).

To obtain an ascl1a expression construct, the ascl1a cDNA was cloned via SmaI/XhoI sites into pCS2+(Rupp et al., 1994) to yield pCS2-ascl1a. To test the efficiency of the ascl1a MO (see below), a fusion construct encoding Ascl1a protein tagged with six C-terminal Myc epitopes was generated by amplifying the ascl1a-coding region from pCS2-ascl1a and cloning it into the BamHI and SpeI sites of pCS2-3′MT (kind gift of U. Strähle). For ascl1a-VP16 fusion constructs, either the entire ascl1a-coding region without the termination codon (wild type), or bp 1-207 (mutant) were amplified via PCR, and cloned into the EcoRI site of pCS2+. The VP-16 activator sequence was excised from a pGMT-VP16 construct(Pogoda et al., 2000) and cloned 3′ to the ascl1a fragments. Capped mRNA was prepared after plasmid linearization, using the Message Machine kit (Ambion). Synthetic mRNA was injected into one- to four-cell stage embryos, as previously described (1.5 nl per embryo)(Hammerschmidt et al., 1999). For BAC injections, an ascl1a containing BAC was obtained from RZPD(ID HUKGB735K15264Q8), and injected at a concentration of 75 ng/μl into the cytoplasm of one-cell stage pia^t25215 +/- intercross progeny.

Embryos were injected with 800 pg ascl1a-myc mRNA with or without ascl1a MO, and embryonic protein extraction were collected at 8 hpf as described (Westerfield,1994). Protein samples were separated via SDS-PAGE on 12%Acrylamid/Bis-acrylamid gels, and blotted on Hybond P membranes (Amersham). Immunoblotting analyses were performed using either the anti-Myc antibody 9E10(Roche Diagnostics) or an anti-pan cadherin antibody (Sigma) as a loading control.

Single and double whole-mount in situ hybridization was performed and specimens were photographed as previously described(Hammerschmidt et al., 1996; Nica et al., 2004). Riboprobes of the following cDNAs were used as described previously: ascl1a and ascl1b (Allende and Weinberg,1994), elavl3 (Kim et al., 1997), crybb1(Chen et al., 2001), omp (Yoshida et al.,2002), neurod (Blader et al., 1997), shh(Krauss et al., 1993), nkx2.1a (Rohr and Concha,2000), fgf3 (Herzog et al., 2004a), lim3(Glasgow et al., 1997), pitx3 (Dutta et al.,2005), pit1 and gsu(Nica et al., 2004), and gh, tsh, pomc and prl(Herzog et al., 2003). Stained embryos were post-fixed, embedded in JB4 resin (Polysciences) and sectioned at 7 μm. Anti-Prl immunostaining was performed with rabbit anti-prolactin(Kawauchi et al., 1983; Sbrogna et al., 2003) and Cy3-labelled goat anti-rabbit antibodies (Jackson ImmunoResearch). Apoptotic cell corpses were visualized with the vital dye Acridine Orange, as previously described (Herzog et al.,2004a).

Wild-type and mutant larvae were fixed with 2.5% glutaraldehyde in PBS for 30 minutes each at ambient temperature and then on ice. After washing with PBS, the larvae were postfixed with 1% osmium tetroxide in 100 mM phosphate buffer pH 7.2 for 1 hour on ice, washed with H₂0, stained with 1%aqueous uranylacetate for 1 hour, dehydrated in a graded series of ethanol and finally embedded in Epon. Ultrathin sections were stained with uranyl acetate and lead citrate and viewed in Philips CM10 electron microscope. In parallel,Toluidine Blue stained Epon sections of 0.5 μm of the same specimen were prepared for light microscopy.

Donor embryos were injected at the one- to four-cell stage with biotin-dextran (Fig. 5A-C) or rhodamin-dextran (Fig. 5D-F)(Molecular Probes), and transplanted at the shield stage into recipient embryos from a cross of two pia heterozygous parents. Alternatively, h2a::h2a-GFP transgenics were used as donors(Fig. 5G-I)(Pauls et al., 2001). Chimeric embryos were analyzed either by in situ hybridization against pit1 or pomc, followed by anti-biotin staining with the Vectastain Elite ABC kit (Vector Laboratories), or by Acridine Orange or anti-Prl immunostaining,followed by fluorescent evaluation in two different channels. Genotyping of recipients was performed with genomic DNA extracted from clipped tails lacking wild-type donor cells, as described above. Implantations of Fgf3-loaded beads into embryos of pia^t25215 or lia^t24149intercrosses were carried out as described(Herzog et al., 2004a). Embryos were genotyped after photography as described above (pia) or by Herzog et al. (Herzog et al.,2004a) (lia). Treatments with the Fgfr inhibitor SU5402(Calbiochem) were carried out as previously described(Herzog et al., 2004a).

During a large-scale screen for zebrafish mutants with altered growth hormone (gh) expression, we had isolated one ENU-induced pituitary-absent (pia) mutation(Herzog et al., 2004b). Segregation linkage analysis revealed that pia maps to a defined interval on linkage group 4 (Fig. 1A, upper panel; see Materials and methods). One of the genes that had been previously mapped to this genomic location (see http://zfin.org/cgi-bin/mapper_select.cgi)was the zebrafish achaete-scute homologue ascl1a,which appeared to be a reasonable candidate, as it is expressed in the pituitary (Allende and Weinberg,1994; Wullimann and Mueller,2002). Sequencing of the ascl1a gene from genomic DNA of pia mutant embryos uncovered a C to A mutation at coding nucleotide 210, changing a tyrosine residue to a premature stop codon. This results in a truncated Ascl1a protein that lacks all amino acid residues after position 69,including the entire basic DNA-binding domain and the HLH domain(Fig. 1A, lower panel). The found ascl1 mutation generates an RFLP (see Materials and methods),which we used to genotype additional pia homozygotes, revealing no recombination between ascl1 and pia in 215 tested embryos(430 meioses; <0.2 cM).

To further confirm that the defects observed in pia are due to a loss of Ascl1a function, we phenocopied the pia defects in wild-type embryos by knocking down ascl1a with antisense morpholino oligonucleotides (MO) (Nasevicius and Ekker, 2000). The used MO targeting a sequence covering the ascl1a start codon efficiently blocked ascl1a translation,as revealed by anti-Myc western blot analysis of protein extracts from embryos injected with mRNA encoding an Ascl1a-Myc fusion protein(Fig. 1B). Injection of the acsl1a MO into wild-type embryos led to a complete loss of adenohypophyseal prolactin (prl, 93% affected, n=58) and pomc (86% affected, n=48) expression, a phenotype indistinguishable from that of pia mutants(Fig. 1C-H). Furthermore, we were able to rescue prl and pomc expression in piamutants upon injection of BAC DNA containing the ascl1a gene(Fig. 1I-K; n=7; and data not shown).

Finally, to investigate the severity of the pia^t25215mutation, we studied the effect of ascl1a overexpression on neurogenesis in the neural plate of zebrafish embryos. Although injection of mRNA encoding wild-type ascl1a had no effect (data not shown),injection of mRNA encoding a Ascl1a-VP16 fusion protein (0.6 ng/μl) led to a massive increase in the number of elavl3-positive neural precursor cells (Fig. 1M; 78/78) (compare with Bae et al., 2005; Kim et al., 1997). By contrast, the corresponding fusion between the VP16 transactivating domain and the truncated part of Ascl1a encoded by the pia^t25215allele was completely ineffective, even when injected at 60-fold higher molarity (12 ng/μl) (Fig. 1N; 0/42). In summary, these data indicate that the pituitary phenotype of pia mutants is caused by a null mutation in the zebrafish achaete scute homologue acsl1a.

To gain more detailed information about the spatiotemporal dynamics of adenohypophyseal ascl1a expression, we performed whole-mount in situ hybridization from early segmentation stages to 120 hpf. First expression of ascl1a in the pituitary placode could be detected at the 20-somite stage (18 hpf; data not shown), coincident with the onset of expression of other pituitary regulators such as lim3 (lhx3 - Zebrafish Information Network), pitx3 and pit1. At 22 hpf and 24 hpf, ascl1a was co-expressed with pitx3 in all adenohypophyseal precursor cells at the anterior neural ridge(Fig. 2A-D), a subset of which also displayed expression of prl(Fig. 2E,F). Consistently, at 26 hpf, ascl1a transcripts were uniformly distributed throughout the entire adenohypophyseal anlage, including anterior-most cells, which give rise to lactotropes and corticotropes, and more lateral-posterior cells, which most probably form somatotropes and thyrotropes(Fig. 2G) (cf. Herzog et al., 2004a; Nica et al., 2004). This uniform distribution of ascl1a transcripts persisted until 72 hpf(see Fig. 2H for 40 hpf). Then,expression became confined to anterior and posterior regions of the adenohypophysis, displaying a pattern resembling that of pomcexpression (Herzog et al.,2004a), whereas the medial domain, characterized by the expression of gh and tsh (Herzog et al., 2004a), became devoid of ascl1a expression(Fig. 2I,J). This spatial shift from a uniform to a restricted distribution of transcripts suggests that ascl1a might initially be required in all pituitary cells, while later playing more restricted roles in particular cell types only. At all investigated stages (24 hpf -120 hpf), ascl1a in addition to the adenohypopyseal primodium itself was also expressed in adjacent cells of the diencephalon (Fig. 2C-J), such as the posterior ventral hypothalamus (pvh; Fig. 2I,J).

Most probably owing to the additional round of genome duplication that occurred during teleost evolution(Postlethwait et al., 1998),the zebrafish has a second ascl1 gene, named ascl1b(Allende and Weinberg, 1994). To study whether partial redundancy between Ascl1a and its paralog Ascl1b might account for the slightly weaker phenotype of pia mutants compared with fgf3 mutants, we also analyzed the expression of ascl1b (Allende and Weinberg,1994). However, at all stages examined (24-120 hpf), ascl1b expression was excluded from the adenohypophysis(Fig. 2K,L; and data not shown), indicating that the regulation of adenohypophysis development has been dedicated to the ascl1a paralog only.

Thus far, the pituitary defects of pia mutants had been analyzed only briefly, revealing the absence of gh, pomc, prl and thyroid-stimulating hormone (tsh) transcripts at late larval stages (120 hpf) (Herzog et al.,2004b). We extended such expression analyses, looking at additional markers and at earlier stages of development. The placodal ectoderm, from which the zebrafish pituitary originates, specifies from late gastrulation through mid segmentation stages, and is set up normally in pia mutants, as revealed by in situ hybridization against eya1 transcripts (data not shown) (cf. Herzog et al., 2004a). Specification of the pituitary itself starts around 19 hpf, as indicated by expression initiation of adenohypophysis-specific marker genes, such as the pan-pituitary marker lim3(Glasgow et al., 1997). At 25 hpf, lim3 expression was present in pia mutants, but the staining was very diffuse and strongly reduced compared with wild-type siblings (Fig. 3A). More moderately reduced expression was also observed for pitx3(Dutta et al., 2005; Zilinski et al., 2005),another marker of the entire adenohypophyseal anlage(Fig. 3C), and for ascl1a itself (Fig. 3E). Interestingly, expression levels of all three genes appeared to recover during later stages of development, when the pituitary of pia mutants was of reduced size (see also below); however, lim3,pitx3 and ascl1a hybridization signals were of normal intensity(Fig. 3B,D,E, inset).

By contrast, expression of other regulator genes was completely lacking in pia mutants from earliest stages onwards. Thus, neurod,another bHLH gene that like ascl1a is expressed throughout the adenohypophysis of wild-type embryos(Blader et al., 1997; Mueller and Wullimann, 2002),was absent in the pituitary of pia mutants at all investigated stages(Fig. 3F; and data not shown). Similarly, expression of pit1, which encodes a POU domain transcription factor required for transcriptional activation of prl,gh and tsh, and for repression of pomc(Nica et al., 2004), was absent in pia mutants at 20 hpf, shortly after its expression initiation in wild-type embryos (data not shown) and at all investigated later stages (see Fig. 3G,H for 25 hpf and 72 hpf).

Early defects in pituitary formation were also visible at the morphological level. Using Nomarski optics, the adenohypophysis can normally be recognized as a separate organ from 25 hpf onwards. However, in pia mutants,adenohypophyseal borders appeared much less distinct than in their wild-type sibling embryos (Fig. 4A,B; 26 hpf). Together with the decreased intensity of early marker gene expression,this suggests that ascl1a is crucial for the proper onset of adenohypophysis specification and organ formation.

We also studied the expression of the different hormone genes at early time points of pituitary development. In zebrafish, terminal differentiation of the different adenohypophyseal cell types occurs in sequential steps between 22 hpf and 48 hpf (Herzog et al.,2003). The first cells that undergo terminal differentiation are the lactotropes, indicated by the expression of prolactin(prl), which is initiated at 22 hpf. The next lineages to differentiate are the corticotropes and melanotropes, which express pomc from 24 hpf onwards (Herzog et al., 2003). The somatotropes, marked by expression of gh (Herzog et al.,2003), the thyrotropes, marked by co-expression of tshand gsu (Herzog et al.,2003; Nica et al.,2004), and the gonadotropes, marked by expression of gsu(Nica et al., 2004), start to differentiate between 36 and 42 hpf. In pia mutants, no prltranscripts could be detected at 26 hpf(Fig. 1C,D), and no ghor tsh transcripts at 48 hpf (Fig. 3I,K). In addition, transcripts of pomc and gsuwere absent from pia mutant pituitaries at 26 hpf (data not shown)and at 48 hpf (Fig. 1F,G; Fig. 3J). In summary, this indicates that Ascl1a is required for the initiation of cognate hormone genes of all adenohypophyseal cell types.

In a previous study, we have shown that in fgf3 mutant zebrafish embryos, failed pituitary specification is followed by destruction of the entire organ, driven by apoptosis of non-differentiated adenohypophyseal cells between 28 and 32 hpf (Herzog et al.,2004a). As early adenohypophyseal specification seems similarly disrupted in pia mutants, we determined apoptosis rates in pia mutant pituitaries, performing Acridine Orange staining at various stages of development. At 24 hpf, no Acridine Orange-positive cells could be detected at the anterior neural ridge, where the adenohypophyseal primordium is located (data not shown; two genotyped mutants). However, at 27 hpf (data not shown; four genotyped mutants) and more dramatically at 30 hpf(Fig. 3E-H; three genotyped mutants), mutant embryos displayed significantly increased numbers of apoptotic cells within the adenohypophyseal anlage. At this stage, the mutant organ was morphologically much more distinct than at 26 hpf (compare Fig. 4G with 4B), although smaller and flatter compared with the pituitary of 30 hpf wild-type siblings(compare Fig. 4G with 4E). In addition to the adenohypophysis itself, Acridine Orange-positive signals were found in regions adjacent to the organ(Fig. 4H). These signals most probably represent debris of apoptotic cells that have been extruded from the adenohypophysis, as previously shown in fgf3 mutants by cell-tracing experiments (Herzog et al.,2004a). At 36 hpf, the number of dying cells in piamutants was significantly decreased compared with the situation at 30 hpf(three genotyped mutants; Fig. 4I,J), while no apoptotic adenohypophyseal cells at all could be detected at 48 hpf (data not shown; three genotyped mutants), and at 52 hpf(two genotyped mutants; Fig. 4K,L). In contrast to fgf3 mutants(Herzog et al., 2004a),however, pia mutants of these late stages had a morphologically distinct adenohypophysis of smaller size(Fig. 4C,D), indicating that a subset of non-differentiated adenohypophyseal cells must have survived the transient phase of cell death. Together, these results indicate that in addition to, or as a consequence of, its role during cell differentiation Ascl1a is also required for the survival of most, but not all adenohypophyseal cells.

As mentioned above, surviving cells in the pituitary rudiment of later stage pia mutants displayed rather normal levels of lim3,pitx3 and ascl1a transcripts(Fig. 3B,D,E). In addition,they lacked expression of genes marking other derivatives of the placodal ectoderm such as crystallin bb1 (lens tissue; Fig. 3L) or omp(olfactory epithelium; Fig. 3M), ruling out trans-differentiation to these other placodal cell types, and strongly suggesting that pituitary identity was maintained to some extent. To gain further insight into the nature of cells in the pituitary rudiment, we carried out transmission electron microscopy. At 120 hpf, the pituitary of pia mutants showed a neurohypophyseal compartment of rather normal size and morphology (compare Fig. 4M,N for wild type with Fig. 4O,P for pia). However, striking differences were apparent in the adenohypophysis. In wild-type siblings, most, if not all, adenohypophyseal cells contained secretory vesicles. By morphology, at least two types of secretory vesicles could be distinguished (Fig. 4Q,R), most probably reflecting different hormone-producing cell types (Nica et al., 2006). By contrast, the adenohypophysis of pia mutants not only contained fewer cells (Fig. 4O,P); ∼90% of the cells also displayed a rather primitive ultrastructure. Thus, their cytoplasm lacked an elaborated endoplasmic reticulum and secretory vesicles,whereas the chromatin of the nucleus was very homogeneous(Fig. 4S). However, a few adenohypohyseal cells of pia mutants did contain a particular type of secretory vesicles (Fig. 4T),although in smaller numbers than in wild-type siblings, and despite the absence of hormone synthesis. This suggests that pia mutant adenohypophyseal cells in their entirety are not fully differentiated,although they differ in their exact specification state.

As described above, ascl1a is expressed not only in adenohypophysis itself but also in adjacent tissues, such as the ventral diencephalon, the source of signaling proteins such as Shh or Fgf3, both of which are required for proper pituitary development(Herzog et al., 2004a; Herzog et al., 2003; Sbrogna et al., 2003). Thus,Ascl1a might either have an essential intrinsic role in adenohypophyseal cells themselves, or might influence adenophypophysis indirectly via the diencephalon. In order to distinguish between these possibilities, we generated chimeric embryos. For chimeras with donor cells in the pituitary,labeled cells were transplanted at early gastrula stages into regions of host embryos ventral to the animal pole, whereas for telencephalic or diencephalic clones, cells were transplanted right at the animal pole, or slightly animal of the dorsal shield, respectively (compare with Herzog et al., 2004a). Chimeric embryos were raised to 26-48 hpf, and stained either for pit1 transcripts to evaluate early adenohypophyseal specification,with Acridine Orange to evaluate cell survival, or for Prl protein to evaluate terminal endocrine differentiation. When wild-type cells ended up in the telencephalon, which is located adjacent to the adenohypophyseal anlage early,or in the ventral diencephalon, which is in close proximity to the adenohypophysis later (Herzog et al.,2004a), no induction of adenohypophyseal pit1 expression was obtained (16/16 embryos; Fig. 5B). By contrast, when wild-type donor cells were located within the adenohypophysis of 26 hpf pia mutant hosts, most of them were pit1 positive (4/5 embryos; Fig. 5C; pit1-negative wild-type cells most probably represent corticotropes, melanotropes or gonadotropes). Similarly, at 32 hpf,transplanted wild-type cells within the adenohypophysis of pia mutant hosts all were Acridine Orange negative, while many of the host pituitary cells were apoptotic (nine embryos; Fig. 5D,E). In reverse, some of the ascl1a morphant cells transplanted into wild-type pituitaries became apoptotic, while the host tissue was completely Acridine Orange negative (11 embryos; Fig. 5F). Finally, many of the wild-type cells transplanted into pia mutant pituitaries contained pomc transcripts (four embryos; data not shown) or Prl protein at 48 hpf (eight embryos; Fig. 5I),while pia mutant cells were Pomc- and Prl-negative(Fig. 5H,I; and data not shown). In summary, this strongly suggests that Ascl1a is required in a cell-autonomous fashion in the adenophypophysis itself to promote early steps of specification, cell survival and terminal endocrine differentiation. By contrast, ascl1a expression in telencephalon and diencephalon appears to be dispensable for proper adenohypophysis development, consistent with the normal expression of all investigated marker genes in the forebrain of mutant embryos (nkx2.1a, Fig. 3N; fgf3, Fig. 6C,D; shh, data not shown).

The findings that fgf3 and ascl1a mutants display similar pituitary phenotypes, and that Fgf3 acts as an extrinsic signal(Herzog et al., 2004a) and Ascl1a as an intrinsic transcription factor, suggests that Ascl1a might act downstream of Fgf3 signaling to mediate at least some of its effects. This notion is further supported by the lack of adenohypophyseal ascl1aexpression in fgf3/lia mutants at 28 hpf(Fig. 6A,B), before the appearance of massive cell death (Herzog et al., 2004a), although in reverse, diencephalic fgf3expression is normal in ascl1a/pia mutants(Fig. 6C,D). Similarly,adenohypophyseal ascl1a expression was absent in 28 hpf wild-type embryos after treatment with the Fgf receptor inhibitor SU5402 from 18-22 hpf(data not shown). To investigate more directly whether Ascl1a is required to mediate diencephalic Fgf3 signaling, we implanted Fgf3-loaded beads into the diencephalon of embryos obtained from pia^+/- intercrosses at 18 hpf, briefly before first adenohypophyseal defects in piamutants become visible. Such beads caused a significant normalization of adenohypophyseal ascl1a expression in pia mutants at 28 hpf(4/6; Fig. 6F). However, they failed to rescue pit1 expression in pia mutants of the same age (5/5; Fig. 6I), in contrast to the striking rescue obtained in control implantations into fgf3/lia fish (6/6; Fig. 6J). This indicates that in the absence of Ascl1a, Fgf3 signaling can activate the expression of ascl1a itself, but not the expression of downstream genes, consistent with the notion that Fgf3 requires Ascl1a to mediate at least some of its effects on the adenohypophysis.

In Drosophila, bHLH proteins encoded by the so-called proneural genes of the achaete-scute complex (AS-C; achaete, scute,asense and lethal of scute) and atonal are not only required for the development of neurons and sensory cells (for reviews, see Calleja et al., 2002; Jan and Jan, 1994; Skeath and Carroll, 1994), but also in other cell types that arise by singling out from clusters of progenitor cells, such as mesodermal and endodermal precursors(Carmena et al., 1995; Tepass and Hartenstein,1995).

Also in vertebrates, Asc-like (Ascl) proteins and members of the Atonal superfamily, such as Math/Neurod1/neurogenin(Ledent et al., 2002), have not only been described in the context of neural differentiation(Westerman et al., 2003). Thus, mouse Mash1 is not only required for the specification of neurons and sensory cells of the CNS or PNS, but also for the differentiation of neuroendocrine cells in the lung (Ball,2004; Borges et al.,1997; Ito et al.,2000), the thyroid (Lanigan et al., 1998) and the adrenal gland(Huber et al., 2002). Furthermore, Mash2 is required for trophoblast formation in the placenta(Guillemot et al., 1994; Tanaka et al., 1997),Mash3/Sgn1 for salivary gland duct cells(Yoshida et al., 2001), and neurogenin 3, Neurod1 and Math1 for diverse endocrine cells of the gastrointestinal system (Jenny et al.,2002; Naya et al.,1997; Yang et al.,2001) and the pancreas islets(Gradwohl et al., 2000; Naya et al., 1997). Although some of these cell types represent specialized neurons, deriving from the neural crest (Lanigan et al.,1998), others lack any sign of neuronal specification, and are supposed to derive from non-ectodermal tissues, such as the endoderm in the case of the endocrine pancreatic cells(Percival and Slack,1999).

Here, we extend this list of Ascl-dependent cell types, showing that zebrafish Ascl1 is required for endocrine differentiation of adenohypophyseal cells. The adenohypophysis derives from the placodal ectoderm, in close proximity to the olfactory epithelium(Dubois and ElAmraoui, 1995). In mouse, the ascl1a homologue Mash1 is required for early olfactory development (Guillemot et al.,1993), while in zebrafish ascl1a mutants, the olfactory epithelium develops normally, indicated by unaffected expression of omp (Fig. 3). This is most probably due to the presence of the ascl1a paralog ascl1b in this tissue (Fig. 2) (Allende and Weinberg,1994). Alternatively, it might be due to functional redundancy between Ascl1a and other proneural bHLH proteins, similar to the situation described for zebrafish Ascl1a and neurogenin 1 during epiphysial neurogenesis(Cau and Wilson, 2003).

It is unclear why, in vertebrates, Ascl proteins in addition to neurons are particularly required for the specification of endocrine fates. In the case of the adenohypophysis and the olfactory epithelium, it might result from their common developmental and evolutionary origin. Thus, both tissues derive from the placodal ectoderm, and according to a current model, they have evolved from a common chemoreceptive structure already present in Bilaterian ancestors(De Velasco et al., 2004; Gorbman, 1995).

In line with the proposed common origin of adenohypophysis and neural structures, we found striking similarities in the modes of Ascl1 action during endocrine differentiation in the zebrafish adenohypophysis and neurogenesis in Drosophila and mouse. In Drosophila, the function of asc members is divided into distinct developmental steps(Brunet and Ghysen, 1999; Jan and Jan, 1994; Westerman et al., 2003). The first is the establishment of neural precursors out of a population of progenitors (proneural function), the second is the specification of specific neural fates (subtype specification step). The same is true for Mash1 during neural development in the mouse. Thus, in the subventricular zone of the ventral telencephalon, Mash1 is required for the formation of a pool of neural precursor cells (Casarosa et al.,1999), whereas in autonomic ganglia, the olfactory epithelium and some regions of the brain, it activates the expression of specific neuronal differentiation genes and initiates neuronal differentiation(Cau et al., 1997; Guillemot et al., 1993; Hirsch et al., 1998; Parras et al., 2002).

A similar biphasic proneural-like and subtype-specification function also appears to apply for Ascl1a in the zebrafish adenohypophysis. Thus, in the absence of Ascl1a function, all adenohypophyseal cell types fail to differentiate, in line with a general role of Ascl1a during the specification of pituitary precursor cells. Actually, Ascl1a is the first transcription factor shown to be required for the proper differentiation of all adenohypophyseal cell types, whereas mutants in all other previously described transcription factors only lack particular adenohypophyseal lineages (for a review, see Zhu and Rosenfeld,2004).

In somatotropes and thyrotropes, ascl1a is switched off after endocrine differentiation and the initiation of cognate hormone gene expression (Fig. 2). This indicates that, in these cell types, ascl1a is exclusively involved in earlier specification steps, such as the activation of pit1, the expression of which fails to be initiated in ascl1a mutants(Fig. 3). By contrast, ascl1a expression persists in the Pomc lineage(Fig. 2), suggesting that in these cell types, Ascl1a might be involved in later, subtype-specific differentiation processes. Interestingly, the zebrafish pomc promoter contains E-boxes, consensus binding sites for bHLH transcription factors(Liu et al., 2003), suggesting that Ascl1a might be involved in the transcriptional activation of the pomc gene. In summary, it appears that Ascl1a regulates adenohypophysis development at multiple levels, acting during both early and terminal steps of the transcriptional control cascade.

In addition to cell specification processes, ascl1a is required for the survival of some, but not all adenohypophyseal cells, allowing us to distinguish two cell populations: one in which cell differentiation and cell survival are strictly coupled; and another in which cells can survive in an undifferentiated state (Fig. 4). In mutant embryos, cell death occurs during a rather narrow time window (2-3 hours), several hours after the failed differentiation of lactotropes and corticotropes, but before somatotropes and thyrotropes would differentiate under wild-type conditions(Herzog et al., 2003). These surviving cells of ascl1a-deficient pituitaries have adenohypophyseal character, as revealed by the presence of lim3, pitx3 and ascl1a transcripts at 72 hpf (Fig. 3), and their ultrastructural morphology at 120 hpf, which resembles that of undifferentiated adenohypophyseal precursor cells(Fig. 4). In the mouse olfactory epithelium, where Mash1 plays a subtype-specific role to drive final steps of neural differentiation (see above), loss of Mash1 leads to massive cell death; however, apoptosis only starts after the initiation of the differentiation arrest (Cau et al.,1997). Assuming that the same is true for the zebrafish adenohypophysis, this would mean that apoptosis (between 30 and 32 hpf) occurs only in the early differentiating lineages (corticotropes, melanotropes and lactotropes; 24-28 hpf). Along the same lines, surviving cells would represent somatotrope and thyrotrope precursors, which normally differentiate later(42-48 hpf). However, their differentiation in the pituitary rudiment of ascl1a mutants would be blocked because of the absence of Pit1(Fig. 3), which is absolutely required for tsh and gh hormone gene expression(Nica et al., 2004). Clearly,this is only one of several possible interpretations, and further experiments,such as transgene-driven re-introduction of pit1 gene products into the pituitary rudiment of pia mutants will be necessary to prove this notion. In addition, it appears that even within this population of surviving cells, Ascl1a might have differential effects, as suggested by the different ultrastructure of adenohypophyseal cells in older mutants(Fig. 4).

In mouse, out of all bHLH proteins, only Neurod1 has been described in the context of pituitary development. In particular, Neurod1 has been shown to bind and activate the Pomc promoter, acting in concert with a Pitx homeodomain protein and another bHLH factor(Horton et al., 1999; Poulin et al., 2000; Poulin et al., 1997; Westerman et al., 2003). To search for bHLH proteins that might act in parallel or downstream of Ascl1a,we studied the expression pattern of other zebrafish achaete scuteand atonal homologues. However, we only found neurod(Fig. 3) to be expressed in the adenohypophysis, whereas ascl1b, neurod2, ngn1, atho1.1, atho1.2,atho2a/ndr1a, atho2b/ndr1b, ath3/neurod4,atho4/ngn3 and ath5 were not (H.M.P. and M.H., unpublished). Interestingly, in ascl1a mutants, neurod expression was absent at all investigated stages, suggesting that it acts downstream of Ascl1a. However, in contrast to Neurod1-deficient mice, which show moderate defects during corticotroph differentiation(Lamolet et al., 2004), we failed to observe any abnormalities during zebrafish pituitary development when knocking down zebrafish Neurod with specific antisense morpholino oligonucleotides (H.M.P. and M.H., unpublished). This indicates that Neurod is neither an essential partner, nor an essential downstream mediator of Ascl1a during zebrafish pituitary development. To address whether Neurod might nevertheless mediate Ascl1a function, e.g. in redundancy with another as yet unidentified factor, it will be necessary to generate transgenic lines to drive Ascl1a-independent adenohypophyseal neurod expression,investigating whether forced neurod expression is sufficient to rescue the adenohypophyseal defects of ascl1a mutants.

A similar transgenic approach will be necessary to provide ultimate proof for a role of Ascl1a downstream of Fgf3 signaling from the ventral diencephalon. For several reasons, such a role seems quite likely. First, fgf3 mutants show a similar combination of failed specification and apoptosis of adenohypophyseal cells(Herzog et al., 2004a), as described here for ascl1a mutants. Second, fgf3 mutants lack adenohypophyseal ascl1a expression, and third, implantation of Fgf3-loaded beads into ascl1a mutants can enhance adenohypophyseal ascl1a expression, while genes downstream of Ascl1a such as pit1 do not respond. Clearly, without the aforementioned rescue of fgf3 mutants by forced ascl1a expression, these data do not rule out that Ascl1a acts in parallel to, rather than downstream of, Fgf3. In any case, Fgf3 must have other mediators in addition to Ascl1a, given that the pituitary in fgf3 mutants is more severely affected than in ascl1a mutants, with apoptosis of all rather than a subset of adenohypophyseal cells.

We are very grateful to Carmen Sonntag (M.H.) and Brigitte Sailer (H.S.)for excellent technical assistance, and to the other members of both the Freiburg screening group (M. Schorpp, D. Diekhoff, T. Franz, M. Leicht, E. Nold, T. Nolting, C. Riegger and W. Wiest) and the Tuebingen 2000 screen consortium (MPI for Developmental Biology: F. van Bebber, E. Busch-Nentwich,R. Dahm, H.-G. Frohnhöfer, H. Geiger, D. Gilmour, S. Holley, J. Hooge, D. Jülich, H. Knaut, F. Maderspacher, H.-M. Maischein, C. Neumann, T. Nicolson, C. Nüsslein-Volhard, U. Schönberger, C. Seiler, C. Söllner, M. Sonawane, A. Wehner, C. Weiler; Exelixis Germany GmbH: P. Erker, H. Habeck, U. Hagner, C. Hennen, E. Kaps, A. Kirchner, T. Koblitzek, U. Langheinrich, C. Loeschke, C. Metzger, R. Nordin, J. Odenthal, M. Pezzuti, K. Schlombs, J. deSatana-Stamm, T. Trowe, G. Vacun, B. Walderich, A. Walker and C. Weiler) for carrying out the screen during which the pia allele was isolated. In addition, we thank H. Kawauchi (chum salmon Prolactin antiserum), Alexander Reugels (h2a::h2a-GFP), Uwe Strähle(pCS2-3′MT; neuroD), Eric Weinberg (ascl1a, ascl1b), Christine and Bernard Thisse (fgf3), Philip Ingham(shh), Klaus Rohr (nkx2.1a), Chi-Yao Chang (gh),Igor Dawid (lim3), Sandy Chen (crybb1), Masayoshi Mishina(omp), Wolfgang Driever (elavl3) and Zoltan Varga(pitx3) for sending fish lines or reagents. The work was supported by the Max-Planck society and by SFB 592.