Histone deacetylase 1 is required to repress Notch target gene expression during zebrafish neurogenesis and to maintain the production of motoneurones in response to hedgehog signalling

Neurogenesis is regulated by intercellular interactions involving the transmembrane-spanning proteins Notch and Delta (reviewed by Campos-Ortega, 1993; Artavanis-Tsakonas et al.,1999). In vertebrate and invertebrate embryos, neuronal precursor cells are born within small proneural clusters as a result of Notch- and Delta-dependent lateral inhibitory interactions (reviewed by Chan and Jan, 1999). Delta-bound Notch protein is cleaved intracellularly and the intracellular fragment promotes transcription of genes encoding transcriptional repressors of the hairy/Enhancer of split [E(spl)] family. Within a Notch-expressing cell, E(spl) proteins repress genes required for the adoption of a neuronal fate, including components of the achaete-scute complex and other bHLH genes such as neurogenin and neuroD. Homologues of Drosophila neurogenic genes such as notch, delta and hairy/E(spl) have been identified in many model vertebrates. In the zebrafish, three delta homologues have been described(Dornseifer et al., 1997; Appel and Eisen, 1998; Haddon et al., 1998), and seven different Hairy/E(spl) (her) genes have been characterised to date(www.ensembl.org/Danio_rerio).

The adoption of a neuronal fate in vertebrate embryos involves the sequential activation of bHLH proneural genes (reviewed by Brunet and Ghysen, 1999; Chitnis, 1999). In the mouse,achaete-scute orthologues such as Mash1 (Ascl1 - Mouse Genome Informatics) are required early on in neural cells to enable them to acquire a neuronal precursor identity (Cau et al., 1997). Expression of the murine bHLH genes neurogenin(Ngn1) and Neurod1 (previously known as neuroD) requires Mash1 activity (Cau et al.,1997). In the zebrafish, both ash1a and ngn1 are required for specification of epiphysial neurones(Cau and Wilson, 2002). ngn1 is also required for the formation of Rohon-Beard sensory neurones in the spinal cord (Cornell and Eisen, 2002), and later on in development for specification of dorsal root sensory ganglia (Andermann et al., 2002).

In the mouse embryo, the E(spl) protein Hes1 represses multiple proneural genes. Targeted inactivation of Hes1 causes upregulation of Mash1 and Ngn1, leading to accelerated neurogenesis and a decrease in the number of later born neurones(Ishibashi et al., 1995). Conversely, ectopic expression of Hes1 prevents neuronal differentiation (Ishibashi et al.,1994). The zebrafish orthologue of Hes1, her6, is expressed in the developing CNS (Pasini et al., 2001), suggesting that it may perform a function related to that of Hes1. Murine Hes5, in contrast to Hes1, is not required for repression of Mash1 during early neurogenesis but instead functions synergistically with Hes1 at a later stage to repress Ngn1(Cau et al., 2000). The zebrafish orthologue of Hes5, her4, similarly represses ngn1in the neural plate (Takke et al.,1999). Taken together, these studies indicate that different E(spl) homologues perform distinct roles during vertebrate neurogenesis. Increased understanding of the distinct functions of E(spl) genes, together with a better appreciation of how they are regulated, promise to yield important new insights into the molecular mechanisms controlling neurogenesis.

Covalent chromatin modifications play key roles in regulating eukaryotic gene expression (reviewed by Strahl and Allis, 2000). The acetylation of core histones on N-terminal lysines is a major determinant of the transcriptionally active state of many genes and chromatin-associated histone acetyltransferases (HATs) are known to perform essential functions in embryonic development. By contrast, the recruitment of histone deacetylase (HDAC) enzymes to specific genes presages their transcriptional silencing. The results of these and other studies suggest that chromatin modifying enzymes may be involved in the establishment and maintenance of cell memory during embryonic development (reviewed by Turner, 2002). Indeed,targeted deletion of murine hdac1 reduces embryonic growth, leading to morphological abnormalities in the head and allantois(Lagger et al., 2002). However, few other insights into the roles of vertebrate HDACs in embryonic development have emerged to date. Here, I have exploited a mutation in zebrafish hdac1 to investigate the function of this gene in the developing nervous system. The results reveal for the first time a primary, in vivo requirement for hdac1 to maintain vertebrate neurogenesis and evidence is presented that this occurs via repression of Notch targets,including her6, the zebrafish orthologue of murine Hes1. Moreover, I demonstrate that expression of proneural genes and neuronal specification are severely impaired in distinct CNS territories of mutant embryos within which strong ectopic expression of her6 is observed. Although the hdac1 mutant hindbrain is segmented, the patterning of post-mitotic neurones and glia within each rhombomere is disorganized. In addition, hdac1 mutants fail to maintain the responsiveness of hindbrain neural precursor cells to hedgehog signalling, which results in the specification of very few branchiomotor neurones. Taken together, these results reveal a surprisingly specific requirement for hdac1 to maintain neurogenesis and enable neuronal fates to be realised in the zebrafish CNS.

hdac1^hi1618 mutant fish were raised from embryos kindly supplied by Dr S. Farrington and Professor N. Hopkins at MIT, USA. mindbomb mutant fish were provided by Dr T. Whitfield; smo^hi1640 mutant and Isl1-GFP transgenic fish were provided by Prof. P. W. Ingham, from stocks maintained at the University of Sheffield.

Homozygous hdac1^hi1618 mutants were distinguished from siblings at 22-24 hours post-fertilization (hpf) by their reduced anterior hindbrain development. Homozygous smo^hi1640 mutants were distinguished from siblings at 24 hpf by their U-shaped somites. Homozygous mib mutants were distinguished from siblings by their abnormal trunk morphology.

Full-length hdac1 cDNA clones were amplified by RT-PCR using the forward primer 5′-ggc agg cgc agg ctg taa tt-3′ and the reverse primer 5′-atg cat cca gga gga ctg gc-3′, based on the hdac1 cDNA sequence deposited in the EMBL database (Accession Number,AF506201).

A morpholino (MO; Gene Tools, LLC) was designed to block translation initiation of hdac1 mRNA, using DNA sequence obtained from the EMBL database (Accession Number, AF506201) and verified by independent cDNA cloning: hdac1-MO: 5′-ttg ttc ctt gag aac tca gcg cca t-3′. A second MO identical in sequence to hdac1-MO apart from four mismatching nucleotides (highlighted in bold), was used to control for non-specific effects of MO injection: Control-MO: 5′-ttg ctc gtt gag aac tct gca cca t-3′

MOs were microinjected into zebrafish embryos at the one- to two-cell stage in a volume of ∼2 nl, at a final concentration of 0.3 mM in water. Neither the control MO nor an irrelevant sequence MO known from previous studies to be biologically inert (5′-cct ctt acc tca gtt aca att tat a-3′)exhibited any observable effects on embryonic development after microinjection.

To modulate the expression level of shh in vivo, in vitro-synthesised capped mRNA encoding zebrafish Shh was microinjected into embryos at the one- to two-cell stage, at a dose of 100 pg/embryo along with the appropriate dose of MO.

For histological analysis, embryos were fixed in 4% paraformaldehyde,embedded in paraffin wax, then 8 μm sections were taken and stained with Haematoxylin and Eosin. Immunohistochemistry was performed using standard procedures (Schulte-Merker,2003), which incorporated a 5 minute permeabilisation step with ice-cold trypsin for embryos older than 28 hpf (omitted for Isl1 staining). Antibodies were used at the following dilutions: anti-Isl1 monoclonal (39.4D5;Developmental Studies Hybridoma Bank, Iowa, USA), 1:500; anti-Phospho-H3 polyclonal (Upstate), 1:500; anti-Hu (BD Biosciences), 1:500; anti-GFAP (a kind gift of Dr J. Clarke, UCL) (Nona et al., 1989), 1:160. Primary antibody binding was visualized with peroxidase- or FITC-conjugated secondary antibodies.

Digoxigenin-labelled probes were prepared as recommended by the manufacturer (Roche). Whole-mount in situ hybridisation was performed using standard procedures (Oxtoby and Jowett,1993). Details of the probes used are available on request.

A recessive mutation of zebrafish hdac1 was recently reported to cause embryonic lethality (Golling et al.,2002). However, few details of the defects caused by loss of hdac1 function were described in that report, and so a systematic analysis of the hdac1 mutant phenotype was initiated in order to gain insights into the requirements for hdac1 function during embryonic development. The expression pattern of hdac1 during embryonic development was first documented using probes prepared from a full-length hdac1 cDNA clone. Fig. 1 shows that hdac1 is maternally expressed, with transcripts being detectable at all developmental stages analysed and in all regions of the embryo. Particularly strong zygotic hdac1 expression is evident in the anterior CNS of embryos at the 18-somite stage and 24 hpf,suggesting that there might be a specific requirement for hdac1function in the developing brain. As development continues, hdac1transcripts selectively accumulate throughout the brain and spinal cord(Fig. 1).

Until 22-24 hpf, embryos homozygous for the hdac1 mutant allele hdac1^hi1618 are morphologically indistinguishable from unaffected siblings. However, at this time, specific abnormalities of hindbrain development first become apparent. Although the correct number of rhombomeres are formed, the anterior rhombomeres fail to undergo the characteristic mediolateral expansion typical of the hindbrain in sibling embryos. Subsequently, ventricular expansion in the midbrain also fails. By 48 hpf, homozygous mutant embryos adopt a curled-down shape and their CNS is significantly smaller than that of unaffected siblings(Fig. 2A,B). Several additional morphological abnormalities are also apparent, including the absence of pectoral fins, the presence of a weakly beating heart lacking an atrio-ventricular valve, and the absence of epithelial projections forming the semi-circular canals in the developing otic vesicle(Fig. 2C-H). Histological analysis confirmed that the hdac1 mutant brain is correctly subdivided into forebrain, midbrain and hindbrain, but all of these structures, as well as the spinal cord, are smaller than in unaffected sibling embryos (Fig. 2C,D). Transverse sections through the hindbrain and spinal cord also revealed abnormalities in their dorsoventral patterning (Fig. 2E-H). To determine whether the reduced dimensions of the hindbrain reflects reduced proliferation of neural cells, phospho-H3-positive mitotic cells were identified and counted in the hindbrain of hdac1^hi1618 homozygous mutant and sibling embryos(Saka and Smith, 2001) over a period from 25 to 38 hpf (Fig. 3; Table 1). At 25 hpf, cell proliferation in the hdac1^hi1618 mutant hindbrain is significantly lower than that in unaffected siblings. However, by 33 hpf, the mitotic indices of hdac1^hi1618 mutants and siblings are similar and they remain so at 38 hpf(Fig. 3; Table 1). Thus, although there is a transient proliferation deficit in the hdac1^hi1618mutant hindbrain at 25 hpf, the normal proliferation rate of neural precursor cells is subsequently regained and maintained independently of hdac1function.

To explore the possibility that the morphological abnormalities in the CNS of hdac1^hi1618 mutants are caused by primary patterning abnormalities within the neuroepithelium, the expression patterns of genes involved in neural patterning, such as pax2a, epha4, hoxb1, hoxb4 and fgf8, were investigated. As exemplified by pax2a and epha4 in Fig. 4,expression of all of these genes in the CNS is essentially unperturbed by loss of hdac1 function, indicating that the mechanisms responsible for regional specification of the neuraxis do not require zygotic hdac1function (Fig. 4A-D).

To investigate the possibility that the hdac1^hi1618mutation affects neurogenesis, the expression patterns of the proneural genes ash1b (Allende and Weinberg,1994) and ngn1 (Blader et al., 1997) were compared in hdac1^hi1618mutants and siblings. At 25 hpf, expression of ash1b (ashb -Zebrafish Information Network) and ngn1 is strikingly reduced throughout the CNS of hdac1^hi1618 mutants(Fig. 4E-H), demonstrating that neurogenesis is significantly compromised by loss of hdac1function.

In view of the evidence that hdac1 functions primarily in the transcriptional repression of target genes (reviewed by Ng and Bird, 2000), it seemed reasonable to postulate that the downregulation of proneural gene expression observed in hdac1^hi1618 mutants could be an indirect consequence of derepressing other genes that are themselves direct targets of hdac1-mediated repression. Proneural genes are well-characterised targets of Notch-dependent transcriptional repression by members of the hairy/E(spl)family of bHLH proteins (reviewed by Campos-Ortega, 1993; Chitnis, 1999). In the mammalian CNS, the Notch target gene Hes1 is required during neurogenesis to co-ordinately repress transcription of both Mash1 and Ngn1 (Cau et al.,2000). This raised the possibility that its zebrafish orthologue, her6 (Pasini et al.,2001), might be aberrantly expressed in hdac1^hi1618 mutant embryos. At 26 hpf, her6 is weakly expressed in the lateral hindbrain of wild-type embryos, but by 33 hpf,transcripts have accumulated in two distinct lateral stripes running caudally from the rhombic lip (Fig. 5A,C). In 26 hpf hdac1^hi1618 mutant embryos, her6 is expressed in the medial hindbrain, being particularly abundant in rhombomeres 5 and 6, and this aberrant pattern persists at 33 hpf(Fig. 5B,D). In wild-type embryos, the proneural genes ash1b and ngn1 exhibit distinct, partially overlapping expression patterns in the hindbrain(Fig. 5G,I,M,O). However, in the hdac1 mutant hindbrain, the abundance of ash1b and ngn1 transcripts is dramatically reduced both at 26 hpf and 33 hpf(Fig. 5H,J,N,P). Transverse sections through rhombomere 5 further reveal the complementary nature of the her6 expression domain compared with those of ash1b and ngn1, in both hdac1^hi1618 mutant and unaffected sibling embryos (Fig. 5). Thus,loss of hdac1 function causes an increase in her6 transcript levels and suppresses expression of ash1b and ngn1(Fig. 5E,F). In the dorsal diencephalon of wild-type embryos, the expression domain of her6 is also complementary to those of ash1b and ngn1(Fig. 6), both at 26 hpf and at 33 hpf Moreover, the her6 expression domain is expanded in the dorsal diencephalon of hdac1^hi1618 mutants, whereas those of ash1b and ngn1 are correspondingly reduced or eliminated, at 26 and 33 hpf (Fig. 6). Taken together, these results demonstrate that hdac1 is required both to repress her6 and to promote expression of ash1b and ngn1 during the development of the hindbrain and dorsal diencephalon.

In order to investigate the degree to which loss of hdac1 function affected production of differentiated neurones in the hindbrain, the distribution of cells expressing the pan-neuronal Hu proteins was compared in hdac1^hi1618 mutant and sibling embryos, at 25, 34 and 38 hpf (Fig. 7). Within each rhombomere of the wild-type zebrafish hindbrain, Hu-positive post-mitotic neurones are arranged in a stereotyped segmental pattern that comprises a central cluster and an outer border separated by a layer of glial cells(Trevarrow et al., 1990)(Fig. 7). Strikingly, in the hdac1^hi1618 mutant hindbrain, Hu-expressing cells fail to adopt the characteristic segmental arrangement in each rhombomere, and instead they assemble into an unsegmented column on each side of the hindbrain(Fig. 7A-F). Between 25 and 38 hpf, the number of Hu-positive cells within the hdac1^hi1618 mutant hindbrain progressively increases,although in comparison to the hindbrain of sibling embryos, there are consistently fewer neurones. As the hdac1^hi1618 mutant hindbrain develops, most of the newly born Hu-positive neurones accumulate in rhombomeres 2, 3 and 4, whereas relatively few are found in rhombomeres 5 and 6 (Fig. 7E,F), in stark contrast to the situation in wild-type embryos(Fig. 7E,F). Glia are also specified in the hdac1 mutant hindbrain but they, like the Hu-expressing neurones, also fail to become organised into segmental arrays,and are distributed throughout the hindbrain in predominantly lateral locations, apparently intermingled with Hu-expressing neurones(Fig. 7G,H). Taken together,these results demonstrate that in the hindbrain hdac1 is specifically required for the production and segmental patterning of both neurones and glia.

In order to characterise the extent to which the hdac1^hi1618 mutation affected the specification of neuronal sub-types within the CNS, the development of Isl1-expressing epiphysial, branchiomotor and spinal cord neurones was compared in hdac1^hi1618 mutant and unaffected sibling embryos(Korzh et al., 1993). Epiphysial neurones express the Isl1 transcription factor once they have been specified within the epithalamic region of the dorsal diencephalon(Masai et al., 1997). Branchiomotor and spinal motoneurones are induced by hedgehog signalling in the hindbrain and spinal cord, respectively, and they also express Isl1 protein in response to such signals(Chandrasekhar et al., 1997). Intriguingly, homozygosity for the hdac1^hi1618 mutation eliminates Isl1 expression in the anterior half of the epiphysis(Fig. 8A,B), and prevents specification of almost all branchiomotor neuronal precursors except for two small clusters of trigeminal (nV) neurones in rhombomere 2 and two small groups of facial (nVII) precursors located in rhombomere 4(Fig. 8C,D). To determine whether the Isl1-positive cells that were specified in the hindbrain of hdac1-deficient embryos were able to differentiate fully into motoneurones with axons, a morpholino targeted against the translation start site of hdac1 mRNA was microinjected into embryos specifically expressing an Isl1-green fluorescent protein (GFP) transgene in branchiomotor neurones (Higashijima et al.,2000). At 30 hpf, hdac1-MO-injected embryos exhibited a pattern of four small clusters of Isl1-positive branchiomotor neurones identical to that observed in hdac1^hi1618 mutants, two of which (nV neurones) were located in rhombomere 2, and two of which (nVII neurones) were located in rhombomere 4 (compare Fig. 8C,D with 8E-H; and see Table 2). Moreover, whereas the population of GFP-positive motoneurones in control-MO injected embryos was considerably increased in size between 30 hpf and 36 hpf(Fig. 8E,F), the rudimentary pattern of four small clusters of GFP-positive branchiomotor neurones remained unchanged in hdac1-MO-injected embryos(Fig. 8G,H). No GFP-positive cells were ever observed in the caudal part of the hindbrain that normally gives rise to nIX and nX motoneurones. Nevertheless, the GFP-positive motoneurones that were present in hdac1-MO-injected embryos did appear to differentiate normally and axon outgrowth could be clearly observed(Fig. 8G,H). These results demonstrate that relatively few branchiomotor neurones are specified in hdac1-deficient embryos, which remain spatially restricted to rhombomeres located anterior to rhombomere 5, and they do not undergo tangential migration posteriorly. Remarkably, however, these cells differentiate into motoneurones with axons that project anteriorly.

In the developing spinal cord of hdac1^hi1618 mutant embryos, the number of Isl1-positive motoneurones is considerably reduced in comparison with that observed in unaffected siblings(Fig. 8I,J). Similarly, hdac1-MO-injected embryos exhibited a substantially reduced population of spinal motoneurones in comparison with control MO-injected embryos (Fig. 8K,L). To a lesser degree, the population of Rohon-Beard sensory neurones is also reduced in the dorsal spinal cord of hdac1 mutants and morphants(Fig. 8I,L). Thus, loss of hdac1 function profoundly impairs neuronal specification in the epiphysis, hindbrain and spinal cord, which is consistent with the observed widespread reduction of proneural gene expression and ectopic expression of the Notch target gene her6 in the CNS of hdac1 mutants.

Notch signalling is essential for proper transcriptional activation of many E(spl) genes during neurogenesis (for a review, see Chitnis, 1999). In vitro studies indicate that Notch-mediated transcriptional activation of E(spl) genes is antagonised by a protein complex that contains Hdac1(Kao et al., 1998). However,the in vivo requirements for hdac1 functions in repression of E(spl) genes, particularly in relation to the effects of Notch signalling, have not been defined. The mind bomb (mib)mutation profoundly impairs Notch signalling such that extensive, premature neuronal differentiation occurs throughout the embryonic CNS(Itoh et al., 2003). To test the hypothesis that hdac1 function is required for repression of her6 in the developing CNS, and that Notch signalling is required to relieve this repression, her6 expression was analysed in the mib mutant under conditions where levels of hdac1 activity were either unperturbed or reduced by hdac1-MO microinjection. Homozygosity for the mib mutation significantly reduces the abundance of her6 transcripts both in the dorsal diencephalon and the hindbrain(Fig. 9), confirming that notch signalling is essential for proper expression of her6. In stark contrast, microinjection of the hdac1-MO into either mibmutant or sibling embryos caused a dramatic derepression of her6 both in the dorsal diencephalon and in hindbrain rhombomeres 5 and 6(Fig. 9, arrows; Table 3). These results demonstrate that hdac1 does indeed act as a repressor of her6 in the hindbrain and dorsal diencephalon, and also that the repressive effect of hdac1 on her6 is normally alleviated by Notch signalling. In further confirmation of these findings, expression of the proneural gene ngn1in the CNS was strictly dependent on wild-type levels of hdac1activity, irrespective of whether hdac1-deficient embryos were homozygous for the mib mutation or not(Fig. 9I,L). Finally,immunostaining for Isl1 protein revealed that loss of hdac1 function severely impaired neuronal specification in both the epiphysis and the hindbrain in a mib-independent manner(Fig. 9M-T; Table 3). Reduced levels of hdac1 eliminated Isl1 expression in the anterior epiphysis, both in mib siblings and mutants (Fig. 9M-P). The hindbrain of hdac1 morphants developed with two pairs of Isl1-positive cell clusters in r2 and r4 and none posterior to r4, as was observed in hdac1^hi1618 mutants(Fig. 9S). By contrast,supernumerary Isl1-positive cells were found throughout the hindbrain of mib mutants (Fig. 9R). However, microinjection of the hdac1-MO into mib mutant embryos suppressed this widespread, ectopic neurogenesis and instead restricted the specification of Isl1-positive cells to the two pairs of cell clusters in rhombomeres 2 and 4 that are characteristic of hdac1^hi1618 mutants(Fig. 9T).

The results described above demonstrate that hdac1 is required to repress Notch targets in order to promote proneural gene expression and to specify several different neuronal cell types, including motoneurones. However, formation of motoneurones in the zebrafish also requires ventral midline-derived hedgehog signals (Chen et al., 2001; Varga et al.,2001; Lewis and Eisen,2001). It was therefore of interest to investigate the relationship between hedgehog signalling and the requirement for hdac1 function in motoneurone specification. In situ hybridisation analysis for expression of shh and the shh target gene ptc1revealed that hdac1 does not play a major role in determining the expression patterns of these genes in motoneurone-forming regions because shh is expressed in ventral midline cells in both hdac1^hi1618 mutant and sibling embryos, and expression of the shh target ptc1 is relatively unperturbed both in the CNS and paraxial mesoderm (Fig. 10A-D). A localised reduction in expression of both shhand ptc1 is found more anteriorly, however, in the zona limitans intrathalamica of the forebrain (zli, Fig. 10A-D). In wild-type embryos, the level of shh expression is strictly limiting for the number of branchiomotor neurones that are specified because over-expression of shh induces supernumerary branchiomotor neurones (Chandrasekhar et al., 1998). Although specification of the majority of branchiomotor neurones requires hdac1, small clusters of hdac1-independent motoneurones do form in rhombomeres 2 and 4 of hdac1-deficient embryos. Therefore, to determine whether shhoverexpression could expand the population of branchiomotor neurones formed in hdac1-deficient embryos, shh mRNA was microinjected into Isl1-GFP transgenic embryos with either a hdac1-MO or a control MO. Microinjection of shh mRNA along with the control MO induced many supernumerary branchiomotor neurones in the hindbrain of Isl1-GFP embryos. However, no supernumerary branchiomotor neurones were formed when the hdac1-MO was co-injected with shh mRNA and instead only the rudimentary pattern of branchiomotor neurones that is characteristic of hdac1^hi1618 mutants was observed(Fig. 10; Table 4). Thus, although the hedgehog pathway is active in hdac1-deficient embryos(Fig. 10A-D) and there are many mitotically active cells in the hindbrain(Fig. 3; Table 1), very few Isl1-expressing motoneurones are produced in response to hedgehog signals(Fig. 10E-L). To confirm that the branchiomotor neurones that did form in hdac1-deficient embryos were nevertheless dependent on hedgehog signalling, the effect of knocking down hdac1 function in smoothened mutants(Chen et al., 2001; Varga et al., 2001) was analysed. At 26 hpf, the first cells to detectably express Isl1 protein in the hindbrain of control-MO-injected, unaffected sibling embryos were a small group of rostrally located trigeminal (nV) motoneurones in rhombomere 2(Fig. 11A). Although loss of smoothened function completely abolishes the specification of these first-detected branchiomotor neurones(Fig. 11C), loss of hdac1 function had no observable effect on their formation(Fig. 11B). By 32 hpf,control-MO-injected, unaffected siblings had developed a properly expanded and appropriately positioned set of branchiomotor neurones(Fig. 11E), whereas once more,no Isl1-positive cells were detectable in the hindbrain of smoothenedmutants (Fig. 11G). As observed previously, at 32 hpf hdac1-deficient embryos developed only the two further small clusters of facial (nVII) motoneurones in r4 in addition to the trigeminal (nV) population first detected in r2 at 24 hpf(Fig. 11F), all of which were strictly smoothened-dependent(Fig. 11H). Taking the results described in Figs 10 and 11 together, it can be concluded that the hedgehog signalling pathway is intact in hdac1mutants, but sustained production of hedgehog-dependent motoneurones, although initially normal, is not efficiently maintained even when the level of hedgehog expression is experimentally increased. Therefore, consistent with its function as an antagonist of Notch signalling, these results reveal that hdac1 is required in the hindbrain to maintain the production of branchiomotor neurones in response to hedgehog signalling.

Previous studies have demonstrated that Hdac1 enzymes are deployed in a variety of developmental contexts to mediate transcriptional silencing. In the mouse, targeted mutation of Hdac1 derepressed CDK inhibitor genes leading to reduced cell proliferation and post-gastrulation embryonic lethality (Lagger et al.,2002). In Drosophila, a loss-of-function mutation in the hdac1 orthologue Rpd3 causes a pair-rule phenotype(Mannervik and Levine, 1999),whereas in C. elegans, mutation of hda-1 causes post-embryonic defects in gonadogenesis and vulval development(Dufourcq et al., 2002). In zebrafish, mutation of hdac1(Golling et al., 2002) causes a phenotype in which, as shown here, distinct aspects of the development of several organs and tissues are affected.

Histological analysis of the CNS in hdac1 mutants initially indicated that neurogenesis may be sensitive to loss of hdac1function. The ensuing analysis demonstrated unequivocally that hdac1is indeed required to promote neurogenesis and evidence is presented that this is accomplished by transcriptional repression of Notch-activated target genes such as her6. Her6 is the orthologue of mammalian Hes1,which is required for Notch-driven repression of Mash1 and Ngn1 (Cau et al.,2000). Consistent with these observations, transcripts of the proneural genes ash1b and ngn1 were almost completely absent in hdac1-mutant embryos, suggesting that, as for Hes1 in mammals,Her6 is a specific repressor of multiple proneural genes. Derepression of her6 was greatest in the dorsal diencephalon and hindbrain rhombomeres 5 and 6, two regions of the CNS where neuronal specification, as revealed by Isl1 immunostaining, was particularly severely affected(Fig. 8). Intriguingly, her6 derepression caused by hdac1 deficiency was epistatic to the mind bomb neurogenic phenotype, both in the dorsal diencephalon and in rhombomeres 5 and 6 of the hindbrain. These observations unequivocally show that a function of Notch signalling is to relieve hdac1-mediated transcriptional repression of Notch targets in the CNS, which is consistent with studies in other species(Kao et al., 1998; Barolo and Posakony, 2002). Interestingly, loss of hdac1 function did not cause widespread derepression of her6 throughout the embryo. Instead, her6derepression was strongest in CNS territories undergoing neuronal specification, implying that hdac1-mediated transcription silencing is selective for neural patterning processes, perhaps through interactions with neural-specific repressors. The results described here also imply that the products of Notch target genes such as the her6 transcriptional repressor can function in an hdac1-independent manner, possibly through redundant interactions with other class I HDACs.

In Drosophila, Notch target genes are actively repressed by DNA-bound Suppressor of Hairless [Su(H)], which recruits co-repressor molecules such as groucho and CtBP to the target locus via interactions with the Hairless protein (Barolo et al.,2002). Other studies in cultured mammalian cells previously found that the Su(H) orthologue CBF1 bound the HDAC1-containing SMRT complex(Kao et al., 1998). Moreover,in Xenopus animal cap explants, the broad specificity HDAC inhibitor trichostatin A caused a two-fold increase in expression of the Notch target gene ESR1 (an orthologue of zebrafish her4) in response to overexpression of Xdelta1 mRNA(Kao et al., 1998), suggesting a role for histone deacetylase in repression of Notch targets. The experiments reported here extend these observations significantly by demonstrating that in the zebrafish embryonic CNS, hdac1 is required for repression of Notch targets in a manner that still renders these genes inducible by Notch signalling. No vertebrate orthologue of the hairless gene has yet been described, and the SMRT protein can interact directly with the orthologue of Su(H) (Kao et al., 1998). However, proteins such as SHARP and SKIP may also modulate connectivity between DNA-bound Su(H) and Hdac1 in vertebrate embryos(Oswald et al., 2002). It will now be of great interest to determine whether interactions between Hdac1,SHARP, SKIP and Su(H) orthologues are essential for repressing Notch targets during neurogenesis in the zebrafish embryo.

Immunostaining for expression of Hu proteins and GFAP revealed that formation of neurones and glia was highly abnormal in the hindbrain of hdac1 mutant embryos (Fig. 7). There are fewer post-mitotic neurones in hdac1 mutant embryos than in siblings at each of the three stages analysed, and the characteristic segmental organisation of these neurones with their associated glia is lost. However, loss of hdac1 function does not cause a general arrest of CNS growth and development, because although cell proliferation in the hdac1 mutant hindbrain is reduced at 25 hpf in comparison to the situation in sibling embryos, it regains its normal proliferative capacity by 33 hpf and a similarly high level of mitotic activity persists at 38 hpf. Moreover, throughout the period from 25 hpf to 38 hpf, the size of the Hu-positive neuronal population in the hindbrain progressively increases. It is possible that the transient reduction in cell proliferation within the hdac1 mutant hindbrain specifically affects a particular step of neurogenesis, because the defects in neurogenesis are irreversible and become more profound with time, whereas cell proliferation within the hindbrain recovers. It is also possible that hdac1 mutants selectively accumulate proliferating neural precursors as a direct consequence of derepressing Notch target genes(Solecki et al., 2001). Future experiments will investigate these possibilities. However, taken together with the finding that patterning markers such as epha4 are properly segmentally expressed in the hindbrain(Fig. 4), the observed abnormalities clearly illustrate that hdac1 is required to efficiently couple neurogenesis to the mechanisms determining segmental patterning of the hindbrain.

Hdac1-deficient embryos exhibited a range of defects in specification of neuronal subtypes, as revealed by immunostaining for Isl1 and by confocal microscopy of an Isl1-GFP transgenic line. In the dorsal diencephalon of hdac1 mutants, there was a striking deficit of anterior epiphysial neurones in a position corresponding to the diencephalic territory within which strong expression of her6 and extinction of proneural gene expression was observed (Figs 6, 8). In the hindbrain, nascent clusters of Isl1-positive trigeminal (nV) and facial (nVII) motoneurones were produced in rhombomeres 2 and 4 of hdac1-deficient embryos, but these clusters failed to expand properly and no additional branchiomotor neurones were formed. Nevertheless, those nV (r2) and nVII (r4) motoneurones that were specified persisted in their original positions within the hindbrain and they produced correctly oriented axons that are characteristic of properly differentiated neurones. Interestingly, there was no evidence of tangential migration caudally (Chandrasekhar et al.,1997), by branchiomotor neurones born in rhombomere 4, into rhombomeres 5 and 6 of the hdac1 mutant hindbrain, where her6 was strongly expressed. It remains unclear whether this abnormality solely reflects a tangential migration defect in nVII neuronal precursors of hdac1 mutants that were born in rhombomere 4, or whether the observed defect is also the consequence of a failure to specify nVII motoneurones from separate precursors originating in rhombomeres 5 and 6.

Within the trunk, loss of hdac1 function caused a substantial reduction in the size of the spinal motoneurone population throughout the spinal cord, as revealed by Isl1 immunostaining, and a milder effect on the number of Rohon-Beard cells was also evident(Fig. 8), indicating that neuronal specification can be initiated throughout the length of the spinal cord but in the absence of hdac1 function it is not efficiently maintained.

All branchiomotor neurones require hedgehog signalling for specification and at the onset of this developmental process, hdac1-deficient and wild-type embryos were indistinguishable(Fig. 11). However, later phases of branchiomotor specification were defective in hdac1-deficient embryos, and this could not be averted by overexpression of shh (Figs 10, 11). Thus, hdac1 is required to maintain the response of neural precursors in the hindbrain to hedgehog signals. Impairment of this response could be a direct consequence of increased Notch target gene expression in hdac1-deficient embryos. Previous work demonstrated that expression of the proneural gene ngn1in the developing CNS is positively regulated by hedgehog signalling(Blader et al., 1997), raising the possibility that loss of hdac1 function directly inhibits the response to hedgehog signalling by preventing proneural gene expression. Recent studies in Drosophila have demonstrated that Notch signalling also prevents the hedgehog-mediated activation of collier in the wing margin (Glise et al., 2002). As vertebrate homologues of collier have previously been implicated in control of neurogenesis (Bally-Cuif et al., 1998; Dubois et al.,1998), it is conceivable that derepression of Notch targets such as her6 in the hdac1 mutant hindbrain could directly inhibit hedgehog-mediated expression of collier homologues. This possibility will now be investigated. Hdac1 protein has been found in physical association with numerous transcriptional repressors, and may be required for the deacetylation of histones associated with many different target genes in neural cells. Nevertheless, the results presented here unveil her6 as a likely direct target of hdac1-mediated transcriptional repression and imply that the her6 locus is hyperacetylated in hdac1mutant embryos. Deacetylated core histones are substrates for lysine-methylation by a large family of SET-domain-containing histone methyltransferases (for a review, see Turner, 2002), and so the her6 locus of wild-type embryos may also exhibit histone methylation patterns that are under-represented in hdac1 mutant embryos. These possibilities will now be investigated.

I am grateful to Professor Nancy Hopkins and Dr Sarah Farrington at MIT for generously providing the hdac1^hi1618 mutant; to Professor Hitoshi Okamoto (RIKEN) for the Isl1-GFP line; and to Professor Philip Ingham and Dr Tanya Whitfield for sharing their mutant stocks and for critically reading earlier versions of the manuscript. I also thank Julian Lewis (CRUK,London), Jonathan Clarke and Adam Guy (UCL, London), Kate Hammond, Claire Allen, Sarah Baxendale, and other members of the Whitfield and Ingham laboratories for technical advice, plasmids and gifts of other materials; and Fiona Browne, Lisa Gleadall and Matthew Green for fish care. Funding for confocal microscopy was provided by Yorkshire Cancer Research.