Wnt1 regulates neurogenesis and mediates lateral inhibition of boundary cell specification in the zebrafish hindbrain

The formation of localised sources of signals that regulate cell differentiation underlies the temporal and spatial patterning of many tissues during embryogenesis. In a number of tissues, this is a component in the progressive refinement of pattern in which following an initial subdivision into distinct regional domains, local interactions induce a signalling centre at the interface of domains. For example, in the wing imaginal disc of Drosophila, signalling molecules expressed at the anteroposterior and dorsoventral compartment boundaries (Dpp and Wg, respectively) have a crucial role in long-range patterning of the tissue(Irvine and Rauskolb, 2001). An analogous situation occurs in the vertebrate neural epithelium in which a signalling centre expressing Fgf8 and Wnt1, the isthmic organiser, forms at the interface of the midbrain and hindbrain(Liu and Joyner, 2001; Rhinn and Brand, 2001; Wurst and Bally-Cuif, 2001). In order for such localised signals to correctly pattern the adjacent tissue,it is essential that the interface at which the signalling source forms is sharp and straight, and that the location and number of signalling cells are precisely regulated (Dahmann and Basler,1999). Important insights into underlying mechanisms have come from studies in Drosophila that have revealed receptor-ligand systems that regulate the restriction of cell intermingling, and that control the formation of signalling centres at boundaries(Irvine and Rauskolb,2001).

In vertebrates, the segmentation of the hindbrain to form rhombomeres provides an amenable model for studying regional specification and the formation of precise patterns of cell differentiation. Each rhombomere has a distinct anteroposterior identity regulated by Hox gene expression that underlies the generation of segmentally organised neurons(Lumsden and Krumlauf, 1996). The sharp and straight interface between each rhombomere is stabilised by the restriction of cell intermingling by Eph receptor-ephrin signalling(Fraser et al., 1990; Mellitzer et al., 1999; Xu et al., 1999). Interactions between adjacent rhombomeres induce the formation of boundary cells at the segment interfaces that have a distinct morphology from non-boundary cells and express a number of specific molecular markers(Guthrie et al., 1991; Guthrie and Lumsden, 1991; Heyman et al., 1995; Heyman et al., 1993; Lumsden and Keynes, 1989; Xu et al., 1995). Because boundary cells have reduced cell proliferation and interkinetic nuclear migration compared with other neural epithelial cells(Guthrie et al., 1991), they may act as a non-motile cell population that stabilises the interface of segments. A further potential role of boundaries is suggested by studies of cell organisation in the zebrafish hindbrain that reveal an anteroposterior organisation of neuronal and glial cell types within each rhombomere(Hanneman et al., 1988; Metcalfe et al., 1986; Trevarrow et al., 1990). For example, primary reticulospinal neurons are located at the centre of rhombomeres, and a GFAP-expressing glial cell `curtain' forms adjacent to segment boundaries (Trevarrow et al.,1990). The formation of this stereotypical pattern can most easily be explained by a role of segment boundaries as signalling centres that regulate the positioning of cell differentiation.

In recent work, we found that expression of radical fringe(rfng), a modulator of Notch signalling, occurs in boundary cells in the zebrafish hindbrain, and that Delta ligands are expressed in non-boundary regions (Cheng et al., 2004; Qiu et al., 2004). Activation of Notch receptor in boundary cells regulates their affinity such that they remain segregated to segment interfaces, and mediates lateral inhibition that prevents the premature neurogenesis of boundary cells(Cheng et al., 2004). Rfng function is required to upregulate wnt1 expression in hindbrain boundary cells but not in the roof plate(Cheng et al., 2004). To identify molecular mechanisms that may mediate potential roles of hindbrain boundaries in patterning, we set out to analyse functions of the Wnt1 signalling molecule. We show that Wnt1 regulates proneural and deltagene expression in non-boundary regions in the zebrafish hindbrain, and that this mediates lateral inhibition that prevents the spreading of hindbrain boundaries. Wnt1 acts in a regulatory network that is strikingly similar to that occurring at the dorsoventral compartment boundary in the Drosophila wing imaginal disc.

Wild-type zebrafish embryos were obtained by natural spawning and raised at 28°C, as described (Westerfield,1994). Control and injected embryos were stage-matched based on morphology (number of somites, pigmentation of eye)(Kimmel et al., 1995).

Morpholino oligonucleotides (MO) were purchased from Gene Tools (Oregon,USA). One- to four-cell blastomeres were microinjected with 0.5-2.5 pmol of MO. The following MO sequences were used:

• wnt1 MO, AGCAACGCGAGAACCCGCATGATAT;

• asha MO, CCATCTTGGCGGTGATGTCCATTTC;

• ashb MO, TCGTAGCGACGACAGTTGCCTCCAT;

• ngn1 MO, ATACGATCTCCATTGTTGATAACCT;

• deltaA MO, CTTCTCTTTTCGCCGACTGATTCAT;

• rfng MO, as described previously(Cheng et al., 2004); and

• tcf3b MO, as described previously(Dorsky et al., 2003).

Capped RNAs encoding full-length chick Wnt1 (cWnt1) and stabilisedβ-catenin (β-cat), lacking 87 amino acids that include the phosphorylation site for GSK3 (Domingos et al., 2001), were synthesised as described previously(Xu et al., 1995). To achieve mosaic expression, 0.2-1.0 ng of RNA was injected into one cell at the two- to eight-cell stage. In the case of embryos injected with both wnt1 MO and β-catenin, the morpholino was injected at the one-cell stage, andβ-catenin RNA was injected into one cell at the eight-cell stage,together with membrane-targeted GFP.

In situ hybridisation probes have previously been described as follows: rfng (Cheng et al.,2004; Qiu et al.,2004), foxb1.2 (Moens et al., 1996), wnt1(Molven et al., 1991), asha and ashb (Allende and Weinberg, 1994), ngn1(Blader et al., 1997; Kim et al., 1997; Korzh et al., 1998), p27^xic1-a (Geling et al., 2003), cyclind1(Yarden et al., 1995), deltaA and deltaD(Haddon et al., 1998), dbx1a (Fjose et al.,1994), tbx20 (Ahn et al., 2000), and gfap(Nielsen and Jorgensen, 2003). Digoxigenin-UTP labeled riboprobes were synthesized according to the manufacturer's instructions (Roche) and in situ hybridisation was performed as described previously (Xu et al.,1994). The colour reaction was carried out using NBT/BCIP substrate (Roche). Embryos were then re-fixed in paraformaldehyde and mounted for photography on a Zeiss Axiovision microscope, or processed for immunohistochemistry.

Primary antibodies used are as follows: anti-EphA4 [1:500(Irving et al., 1996)],anti-HuC/HuD (1:200, Molecular Probes), anti-phosphohistone H3 (1:250,Upstate) and anti-neurofilament (RMO-44, 1:200, Zymed). Antibodies were diluted in PBS-Tween containing 2% goat serum, and embryos were blocked in 5%goat serum. Detection of primary antibodies was carried out using Alexa Fluor-488, -594 or -647 goat anti-mouse or anti-rabbit IgG (1:500, Molecular Probes), or HRP-conjugated goat anti-rabbit IgG (1:250, Dako). Fluorescent images were captured using a Leica TCS SP2 confocal microscope.

Previous studies have revealed important roles of wnt1 at the midbrain/hindbrain organiser (Liu and Joyner, 2001; Wurst and Bally-Cuif, 2001) and roof plate(Megason and McMahon, 2002). We find that in addition to being detected at these sites, wnt1transcripts are present in hindbrain boundary cells in zebrafish embryos. wnt1 transcripts are first detected in hindbrain boundaries from 14 hours of development (not shown), and at 16.5 and 18 hours are present in the dorsal half of all boundaries (Fig. 1A-C). At these stages, wnt1 expression occurs in the roof plate of rhombomeres r2, r3 and r5(Fig. 1C). By 24 hours, wnt1 expression in the hindbrain occurs in the dorsal roof plate of all segments, except r1, and continues to occur in the dorsal part of hindbrain boundaries (Fig. 1D). We compared the expression of wnt1 with rfng and foxb1.2 (mariposa) that are expressed along the whole dorsoventral extent of hindbrain boundaries. Expression of rfng in two rows of boundary cells that flank the interface of segments(Cheng et al., 2004; Qiu et al., 2004) is maintained throughout the period of wnt1 expression(Fig. 1E-H). foxb1.2is expressed at low levels in non-boundary regions, and upregulated at hindbrain boundaries (Cheng et al.,2004; Moens et al.,1996) (Fig. 1I-L). As upregulation of foxb1.2 expression at boundaries does not require rfng function, or vice-versa (Fig. 2J, and data not shown), these genes are independent markers of boundary cells.

We examined whether wnt1 has a role in the formation of hindbrain boundaries by using an antisense morpholino oligonucleotide (MO) to block translation of wnt1 mRNA, and then assessed boundary marker expression. We find that at 18 hours, rfng expression appears unchanged in wnt1 MO-injected embryos compared with uninjected embryos (Fig. 2A,B). By contrast, only 1.5 hours later at 19.5 hours of development, rfngexpression is broader at all boundaries in wnt1 MO embryos, and in some embryos extends to the centre of the rhombomeres(Fig. 2C,D). A similar broadening of rfng expression(Fig. 2E,F) and of foxb1.2 boundary expression (Fig. 2G,H) is observed in wnt1 MO embryos at 26 hours. Notably, ectopic boundary marker expression does not occur in rhombomere 4(r4; Fig. 2D,F). Ectopic boundary marker expression does not occur following injection of control morpholino with scrambled sequence (data not shown).

In recent work, we have shown that modulation of Notch activation by Rfng is required for expression of wnt1 at hindbrain boundaries(Cheng et al., 2004). As roof plate expression of wnt1 is unaffected by rfng knockdown, this allowed us to assess the role of the boundary expression of wnt1. We find that in rfng MO embryos, the boundary expression domains of rfng and of foxb1.2 expand(Fig. 2I,J), as happens in wnt1 MO embryos. Taken together with the wnt1 MO experiments, these findings suggest that the boundary expression of wnt1 is required to prevent boundary marker expression from broadening. It is possible that rfng is also required for the expression of other boundary signals that contribute to the restriction of boundary spreading.

Previous studies have suggested that tcf3b, a mediator of canonical Wnt signalling, is required for the formation of hindbrain boundaries, as tcf3b MO-injected zebrafish embryos lacked morphological hindbrain boundaries, and foxb1.2 expression at boundaries appeared to be absent (Dorsky et al., 2003). However, the uniform expression of foxb1.2observed in this study could be due to ectopic hindbrain boundary expression,rather than an absence of boundary expression and persistence of the non-boundary expression of foxb1.2. We therefore examined the role of tcf3b in more detail. Knockdown of tcf3b causes an expansion of the expression domains of both rfng and foxb1.2(Fig. 2K,L), as occurs in wnt1 knockdowns. Our observation of the same phenotype following knockdown of two distinct genes in the same pathway is strong evidence for the specificity of this effect of the morpholino oligonucleotides, and suggests that the action of Wnt1 in the hindbrain is mediated by Tcf3b.

There are several possible explanations for the expansion of boundary marker expression following knockdown of wnt1 or tcf3b. Because hindbrain boundary markers are upregulated by interactions between cells with distinct segmental identity at segment interfaces(Guthrie and Lumsden, 1991),the broadening of boundary marker expression could be due to a mixing or interleaving of interfaces. However, we detect no difference from uninjected embryos in the formation of sharp segmental expression domains of hoxb1a and krox20 in wnt1 MO embryos in which ectopic boundary marker expression has occurred (data not shown). An alternative possibility is that decreased wnt1 function leads to a major increase in boundary cell proliferation and loss of non-boundary cells during the period of boundary marker expression expansion. To examine this, we detected phospho-histone H3, a marker of mitosis, and found that during boundary marker expansion at 19 hours there is a decrease in mitotic index(Fig. 2M,N,Q), both of boundary cells (42% decrease) and of non-boundary cells (30% decrease). Furthermore, we find that cyclinD1, a known downstream transcriptional target of Wnt1 signalling associated with cell proliferation in the spinal cord(Megason and McMahon, 2002),is expressed at elevated levels in hindbrain boundaries(Fig. 2O). In wnt1 MO embryos, cyclinD1 expression at boundaries is decreased(Fig. 2P), suggestive of a role of Wnt1 in enabling rather than constraining boundary cell proliferation. We therefore favour a third explanation for the broader expression domains of hindbrain boundary markers: that Wnt1 signalling via Tcf3b is required to prevent non-boundary cells from upregulating boundary markers.

A potential mechanism of Wnt1 action was suggested by the expression patterns of achaete-scute homologue (ash) and neurogenin(ngn) proneural genes, and of Notch ligands in the hindbrain(Cheng et al., 2004). At 18 hours, the proneural genes asha and ngn1 are expressed in cells in the dorsal and ventral half of each rhombomere, respectively, but are excluded from boundaries. At the same stage, ashb is expressed at high levels in cells in r4 (Fig. 3A-C). Subsequently, proneural gene expression resolves into stripes that become established by 24 hours and persist until at least 48 hours of development. For example, at 26 hours ashb and ngn1are expressed in presumptive neuroblasts adjacent to rhombomere boundaries,with weaker expression throughout the ventricular zone except at boundaries(Fig. 3G,H). At this stage, asha expression occurs in scattered cells but is excluded from boundaries (Fig. 3F), and from 30 hours is found in stripes adjacent to boundaries (data not shown). Expression of these genes at 24-48 hours is patterned along the dorsoventral axis, with ngn1 being most ventral, ashb medial, and asha most dorsal. We found a similar progression in the expression patterns of delta genes, which are downstream targets of proneural genes, and of p27^Xic1-a, which mediates cdk inhibition and cell cycle exit in neuroblasts (Carruthers et al., 2003; Geling et al.,2003; Ohnuma et al.,1999). At 18 hours, expression of deltaA, deltaD and p27^Xic1-a occurs throughout the rhombomeres, except in boundary cells, with stronger expression seen at lateral locations(Fig. 3D,E; Fig. 4A). By 26 hours, stripes of high level deltaA, deltaD and p27^Xic1-a expression have formed adjacent to the boundaries, with weaker expression in the ventricular zone in rhombomere centres (Fig. 3I,J and data not shown).

These expression patterns are reminiscent of the situation at the dorsoventral boundary of the Drosophila wing disc, where Notch activation at the boundary upregulates Wingless (Wg) expression, which in the anterior compartment signals to neighbouring cells to upregulate achaete-scute complex (as-c) proneural genes that activate Delta expression (Diaz-Benjumea and Cohen, 1995; Kim et al.,1995; Rulifson and Blair,1995). As Delta cell-autonomously blocks Notch activation, this pathway mediates lateral inhibition that prevents spreading of the Wgboundary expression domain (de Celis and Bray, 2000; Rulifson et al.,1996). Our finding that expression of the homologous gene families in the zebrafish hindbrain has the same spatial relationships could provide an explanation for the spreading of boundaries when Wnt1 signalling is blocked. We therefore tested first whether there is an equivalent regulatory hierarchy in the zebrafish hindbrain as in the Drosophila wing disc, and second, whether proneural and delta genes mediate the lateral inhibition of boundary formation by Wnt1.

Analysis of deltaD gene expression in wnt1 MO embryos revealed that at 18 hours neurogenesis appears unchanged from uninjected embryos, but by 19.5 hours there is a major reduction in the number of expressing cells (Fig. 4A-D). Similarly, there is a major decrease in the number of cells expressing p27^Xic1-a in wnt1 MO embryos(Fig. 4E,F). These results suggest that Wnt1 is not required for early neurogenesis, but is required for neurogenesis from 19.5 hours onwards. In support of this, we find that there are fewer differentiated neurons (marked by expression of Hu; Fig. 4Q,R) at 25 hours in wnt1 MO embryos compared with in uninjected embryos. Counting of the number of neural epithelial cells in hindbrain segments at 19 hours reveals that wnt1 MO embryos have approximately 85% of the cell number of uninjected embryos. Because many precursor cells remain, this decrease does not account for the major deficiency in neurogenesis. The width and morphology of the hindbrain was severely affected in wnt1 knockdowns at later stages due to the major depletion of the mantle layer of post-mitotic neurons.

To determine whether Wnt1 signalling via Tcf3b is required for generation of all neurons or of specific subsets of neurons after 19.5 hours, we analysed expression of proneural and delta genes that mark distinct populations of differentiating cells. We found that by 24 hours, wnt1or tcf3b knockdown led to a major decrease in the number of cells expressing deltaA (Fig. 3I, Fig. 4G,H), ashb (Fig. 4I-K), ngn1 (Fig. 4M-O), asha and deltaD (not shown) throughout r2, r3, r5 and r6. The greater decrease in neurogenesis following tcf3b knockdown compared with wnt1 knockdown (Fig. 4G,H) may be due to a contribution of other Wnts acting via Tcf3b. It is notable that knockdown of wnt1 or tcf3b consistently had a weaker effect on neurogenesis in the spinal cord, in the hindbrain posterior to r6, and in r4 (Fig. 4F,J-L,N,O). Importantly, we find that decreased proneural and delta gene expression occurs following knockdown of rfngthat leads to loss of wnt1 expression at hindbrain boundaries(Fig. 4L and data not shown).

To analyse in more detail whether Wnt1 is selectively required for generation of specific neural cell types, we analysed the organisation of the hindbrain at 48 hours, when there is a well-characterised pattern of neuronal and glial cell types (Trevarrow et al.,1990). We found that following wnt1 knockdown, generation of primary reticulospinal neurons located at rhombomere centres is unaffected(Fig. 5A,B), motor neurons form but in lower numbers (Fig. 5C,D), and there is a major decrease in the number of dbx1a-expressing neurons adjacent to segment boundaries, whereas dbx1a expression still occurs in progenitor cells(Fig. 5E,F,I-L). Furthermore, wnt1 knockdown leads to ectopic formation of GFAP-expressing glial cells that in uninjected embryos are located adjacent to hindbrain boundaries(Fig. 5G,H). This pattern is consistent with the requirement for Wnt1 only for later neurogenesis(subsequent to 18 hours of development) revealed by analysis of deltagene expression (Fig. 4A-D):most reticulospinal neurons are born prior to 15 hours, motor neurons differentiate between 16-20 hours, and neurogenesis adjacent to hindbrain boundaries occurs between 22-48 hours(Chandrasekhar et al., 1997; Mendelson, 1986; Trevarrow et al., 1990).

Based on previous studies of delta gene regulation(Haenlin et al., 1994; Hans and Campos-Ortega, 2002; Heitzler et al., 1996), it seemed likely that the reduction of delta gene expression in wnt1 MO and tcf3b MO embryos is due to the decreased expression of proneural genes. Indeed, in either single or multiple knockdowns of asha, ashb and ngn1, we found that deltaA and deltaD expression is strongly reduced (compare Fig. 4P with Fig. 3I; data not shown).

Gain-of-function experiments suggest that in the mouse spinal cord activation of the Wnt/β-catenin pathway promotes proliferation and inhibits differentiation of neuronal progenitors(Megason and McMahon, 2002; Zechner et al., 2003). We therefore further tested whether Wnt1 promotes neurogenesis in the hindbrain by analysis of the effects of ectopic expression of Wnt1 or stabilisedβ-catenin (lacking sequences required for GSK3-induced degradation). We found that overexpression of Wnt1 (Fig. 4S,T) or of stabilised β-catenin(Fig. 4U,V) led to an increase in the number of differentiated neurons in comparison to uninjected control embryos (Fig. 4Q). Furthermore,expression of stabilised β-catenin in wnt1 MO embryos(Fig. 4W,X) rescues the decrease in neurogenesis in embryos injected with wnt1 MO(Fig. 4R). These findings suggest that Wnt1 promotes rather than inhibits neuronal differentiation in the hindbrain.

We next analysed whether, analogous to the situation in the Drosophila wing disc, proneural and delta gene function is required to repress ectopic boundary marker expression. We find that expansion of rfng and foxb1.2 expression occurs in asha MO embryos, to a lesser extent in ashb MO injected embryos, and with a stronger phenotype in asha/ashb double knockdowns(Fig. 6A-D,I-K). Knockdown of ngn1 does not lead to hindbrain boundary spreading, but synergises in a double knockdown with ashb (Fig. 6E,F). These synergistic effects suggest that proneural genes overlap in restricting boundary spreading. As found in wnt1 and tcf3b knockdowns, boundary spreading does not occur in r4 in multiple knockdowns of proneural genes. Remarkably this was the case even following triple knockdown of asha, ashb and ngn1, which almost completely blocks neurogenesis (Fig. 6G). Finally, we analysed the effect of decreased deltagene function. Hindbrain boundary expansion is not detected in homozygous after eight mutant embryos that have a null mutation in the deltaD gene (data not shown). However, knockdown of deltaA leads to a major expansion of rfng and foxb1.2 boundary expression(Fig. 6H,L).

Previous work has shown that a distinct population of boundary cells forms at the interface of odd and even hindbrain segments(Guthrie and Lumsden, 1991; Lumsden and Keynes, 1989), but little is known regarding the functions of these cells or mechanisms that regulate their formation. One potential role of boundary cells is to inhibit mixing across the interfaces of segments(Guthrie et al., 1991), and consistent with this recent work has identified a role of Notch signalling in regulating an affinity difference between boundary cells and non-boundary cells (Cheng et al., 2004). Here, we have found that expression of Wnt1 at hindbrain boundaries regulates neurogenesis and prevents the spreading of boundaries. Hindbrain boundaries therefore have an essential role in the control of cell differentiation in the zebrafish hindbrain.

A key question is the nature of the relationship between Wnt1 and the control of neurogenesis in the zebrafish hindbrain. A number of studies have found that activation of the Wnt/β-catenin pathway promotes neural precursor proliferation and inhibits neuronal differentiation(Chenn and Walsh, 2002; Megason and McMahon, 2002; Zechner et al., 2003). For example, Wnt1 expression in the roof plate of the mouse spinal cord acts via upregulation of cyclinD1 to drive cell proliferation, which expands the neuronal precursor population and inhibits neuronal differentiation in dorsal regions (Megason and McMahon,2002). Indeed, the observation that neurogenesis occurs first in the centre of zebrafish hindbrain segments(Trevarrow et al., 1990) is suggestive of the situation in the spinal cord in which neurogenesis is initiated distal from the source of wnt1 expression. However, we find that rather than suppressing neurogenesis, after 18 hours of development wnt1 is required for neurogenesis in the zebrafish hindbrain.

One possible explanation for our findings is that, as in the spinal cord, wnt1 promotes cell proliferation, and that wnt1 knockdown leads to a subsequent lack of neurogenesis in the hindbrain because of a depletion of precursor cells available to differentiate into neurons at 19 hours and later stages. Indeed, wnt1 knockdown leads to decreased cyclinD1 expression and cell proliferation in the hindbrain. However,there is only an approximate 15% decrease in the number of neural epithelial cells at 19 hours in wnt1 knockdowns, with many neural precursors remaining that in r2, r3, r5 and r6 downregulate proneural gene expression and upregulate boundary cell markers. As boundary cells are themselves neural progenitors (with delayed differentiation), the decrease in neurogenesis following wnt1 knockout is in large part due to a switch of progenitors from non-boundary to boundary identity. In support of an inhibition of neuronal differentiation rather than loss of neural progenitors,we observe that wnt1 knockdown leads to inhibition of the differentiation of dbx1a-expressing neuronal precursors. Furthermore,the ectopic differentiation of GFAP-expressing glial cells in rhombomere centres following wnt1 knockdown could be due to a block in neuronal differentiation of precursors (Nieto et al., 2001), and/or to the ectopic formation of hindbrain boundary cells. By contrast, proneural gene expression is maintained in the spinal cord, with a sharp anterior boundary at the r6/r7 interface in wnt1knockdowns, suggestive of a distinct role of wnt1 in neurogenesis in the spinal cord compared with in the hindbrain. Intriguingly, the spatial relationship between wnt1 and cyclinD1 expression is different in the spinal cord and hindbrain. In the spinal cord, cyclinD1 is expressed in the dorsal part of neural tube adjacent to the wnt1-expressing roof plate, whereas in the hindbrain cyclinD1 expression occurs at high levels in boundary cells,suggesting that Wnt1 has an autocrine as well as a paracrine role.

We therefore favour a model in which, in addition to a role in regulating cell proliferation, Wnt1 promotes neurogenesis in the hindbrain. Because in wnt1 or tcf3b knockdowns, proneural gene expression is decreased and boundary markers upregulated throughout r2, r3, r5 and r6, Wnt1 seems to act throughout these segments to enable neurogenesis. The expression of wnt1 in both the roof plate and hindbrain boundaries raises the question of the relative contribution of these signalling sources. We find that loss of the boundary expression but not roof plate expression of wnt1 in rfng knockdowns leads to boundary spreading and decreased neurogenesis. As it is likely that the roof plate expression of Wnt1 also contributes to the promotion of neurogenesis and inhibition of hindbrain boundary spreading, expression by hindbrain boundaries may achieve a threshold level of signal required to drive sufficient proneural gene expression to block boundary spreading.

Our findings could be explained by two alternative roles of Wnt1 signalling. First, that Wnt1 induces a proneural state of non-boundary cells and is a permissive factor for neurogenesis, whereas other factors control the spatial patterns of differentiation. In support of this are the observations that neurogenesis occurs adjacent to the full dorsoventral extent of hindbrain boundaries, whereas wnt1 expression is restricted to the dorsal part of hindbrain boundaries, and that neurogenesis does not occur in a stripe adjacent to wnt1 expression in the roof plate of the hindbrain. A second possibility is that in addition to Wnt1 being required for a proneural cell state through the segment, high levels of Wnt1 induce the differentiation of neurons adjacent to boundaries. This situation is similar to the Drosophila wing disc, in which cells adjacent to the source of Wg express proneural genes and, as a consequence, undergo mitotic arrest and differentiate (Couso et al.,1994; Johnston and Edgar,1998; Phillips and Whittle,1993). Consistent with this, recent studies in the mouse neocortex and in in vitro cell culture have found that Wnt signalling promotes neurogenesis and directly regulates neurogenin 1 gene expression(Hirabayashi et al., 2004; Israsena et al., 2004; Otero et al., 2004). Furthermore, Wnt signalling may have stage-specific effects, in which it initially promotes proliferation and later induces neuronal differentiation(Hirabayashi et al., 2004). However, the spatial pattern of neurogenesis in the hindbrain cannot be explained based solely on Wnt1 signalling; for example, since neurogenesis adjacent to boundaries persists as wnt1 expression becomes increasingly dorsally restricted in boundaries. Therefore, other factors are required to cooperate with or overlap functionally with any instructive role of Wnt1 in spatial patterning of neuronal differentiation.

During completion of this work, another study has analysed roles of Wnt signals in the hindbrain (Riley et al.,2004). There are significant differences in the conclusions of these authors compared with the current work. First, it is proposed that Wnt signals from hindbrain boundaries have a purely organising role in the anteroposterior pattern of neurogenesis, whereas we demonstrate a temporal requirement for Wnt signals in neurogenesis that underlies a selective loss of later neurons that differentiate adjacent to boundaries. The data of these authors are consistent with a loss rather than a spatial disorganisation of neurogenesis following knockdown of wnt genes or of tcf3b. Second, Riley et al. (Riley et al.,2004) propose that Wnt signals are required to maintain hindbrain boundaries, whereas we show that they are required to prevent spreading of boundaries. This reflects that these authors did not have appropriate markers available to identify ectopic boundary cells, although, in agreement with our findings, GFAP-expressing glial cells that normally flank boundaries were observed in ectopic locations following decreased Wnt activity. A third difference is that Riley et al. (Riley et al., 2004) observe only subtle phenotypic changes in a deletion mutant that removes wnt1 and wnt10b, but knockdown of further wnt genes (or of tcf3b) leads to a similar effect on neurogenesis as we observe following knockdown of wnt1. The reason for the quantitative difference is unclear, but might be due to the large deletion that removes wnt1 and wnt10b also removing a gene(s) that has a compensatory effect on neurogenesis.

A potential pathway by which Wnt1 might inhibit ectopic expression of boundary cell markers was suggested by the similarity of gene expression patterns in the zebrafish hindbrain to those adjacent to the dorsoventral boundary of the Drosophila wing imaginal disc. In the wing imaginal disc, expression of Fringe and Serrate in the dorsal compartment, and of Delta in the ventral compartment, leads to a stripe of Notch activation at the dorsoventral boundary(Panin et al., 1997), and Notch activation upregulates wg expression(Diaz-Benjumea and Cohen,1995; Kim et al.,1995; Rulifson and Blair,1995). Wg protein acts on adjacent cells in the anterior compartment to upregulate expression of as-c proneural genes, which specify a post-mitotic sensory hair cell fate, and upregulate Delta gene expression (Couso et al., 1994; Johnston and Edgar, 1998; Phillips and Whittle, 1993). Delta acts cell autonomously to inhibit Notch activation, and because Notch activation is required to activate wg expression, this mediates a lateral inhibition that prevents spreading of the wg expression domain (de Celis and Bray,2000; Rulifson et al.,1996) (Fig. 7A).

After 18 hours of development, expression of ash and ngnproneural genes, and of delta genes, becomes restricted to stripes adjacent to hindbrain boundaries. We find that there is the same regulatory hierarchy in the hindbrain (Fig. 7B) as in the Drosophila wing disc: the modulation of Notch by Rfng upregulates wnt1 in boundary cells(Cheng et al., 2004), knockdown of wnt1 or tcf3b leads to a major decrease in the number of cells expressing the ash and ngn proneural genes, and knockdown of ash or ngn leads to a decrease in delta gene expression. Finally, we find that knockdown of ash,ngn or delta gene function leads to spreading of hindbrain boundary marker expression.

We have thus uncovered a striking parallel with the mechanisms that in the Drosophila wing disc control formation of sensory hair cells and prevent ectopic boundary cell formation. As the zebrafish hindbrain and Drosophila wing are not homologous structures, we propose that this reflects an independent recruitment during evolution of a regulatory network of genes that underlies an activity of boundaries both in patterning adjacent tissue and in preventing the spread of the signalling source. There are several differences in the Notch-Wnt-proneural network between the hindbrain and wing imaginal disc; for example, in the expression patterns of fng and Notch ligands that establish a stripe of Notch activation at boundaries. This may be due to the larger number of fng genes in vertebrates that have distinct expression domains in hindbrain segments and boundaries (Cheng et al., 2004; Prince et al., 2001), and may have diverged in their function (Dale et al., 2003). Another difference is that in the hindbrain, Wnt1 has a long-range role in promotion of neurogenesis, whereas in the wing imaginal disc, neural cells only form immediately adjacent to the Wg signalling source.

Taken together with previous work, our findings reveal a regulatory loop between boundary cells and non-boundary regions that stabilises the pattern of each cell population via bidirectional lateral inhibition(Fig. 7B). Rfng-mediated modulation of Notch activation upregulates wnt1 expression in boundary cells (Cheng et al.,2004). Notch activation also regulates the affinity properties of boundary cells (Cheng et al.,2004), thus maintaining their segregation to the interfaces of segments. Wnt1 expressed by boundary cells promotes proneural and delta gene expression in non-boundary regions, which enables neuronal differentiation and laterally inhibits the spread of boundary marker expression. In addition, roof-plate expression of Wnt1, which is independent of Rfng function, contributes to the promotion of neurogenesis, but is not sufficient to prevent hindbrain boundary spreading. Delta expression by non-boundary cells activates Notch in boundary cells, and thus laterally inhibits boundary cells from expressing proneural genes and undergoing neuronal differentiation (Cheng et al.,2004).

As wnt1 knockdown affects neurogenesis only after 18 hours of development, this model predicts that other signals from boundaries regulate neurogenesis and restrict boundary spreading at earlier stages of differentiation in the hindbrain. In addition, it is striking that specifically in r4 there is less downregulation of proneural gene expression following wnt1 or tcf3b knockdown, suggesting that other factors regulate neurogenesis in r4. This continued proneural gene expression in r4 could provide an explanation for why there is no boundary spreading into r4 following wnt1 or tcf3b knockdown. However, we find that even following an almost complete block of neurogenesis in a triple knockdown of proneural genes, hindbrain boundary spreading did not occur into r4. The mechanism underlying the restriction of boundary spreading in r4 is not known,but a potential functional significance is suggested by the findings that this rhombomere is the first to form and differentiate, and acts as an early signalling centre to pattern adjacent segments(Maves et al., 2002). This early role of r4 may require that other signals regulate the restriction of boundary spreading and neurogenesis independently of the later upregulation of wnt1 gene expression in boundaries.

Finally, our findings have implications for mechanisms by which boundary cells form at the interface of adjacent hindbrain segments. An attractively simple model is that boundary cells are specified by interactions that only occur at segment interfaces; for example, by complementary segmental expression of a cell surface receptor/ligand system. However, our finding that after blocking the Wnt1/Tcf3 pathway most or all cells (except in r4) can express boundary markers shows that boundary cell formation does not need to be located adjacent to the segment interface. We therefore propose that interactions at segment interfaces instead provide a bias that ensures that boundary cell specification is initiated at that location, and that lateral inhibition then prevents further cells from becoming boundary cells.

We thank James Briscoe, Alex Gould, Jean-Paul Vincent, Qiling Xu and Francois Guillemot for helpful discussions, and Nobue Itasaki, Qiling Xu,Isabel Campos, Derek Stemple and the RZPD for reagents. This work was supported by the Medical Research Council, the Human Frontier Science Program and the Agency of Science, Technology and Research, Singapore.