vhnf1 and Fgf signals synergize to specify rhombomere identity in the zebrafish hindbrain

The vertebrate hindbrain is organized into segments along the anterior-posterior axis, and the process by which these segments are generated follows a sequence of progressive partitioning of the tissue. During and just after gastrulation, a broad domain of tissue is set aside as the future posterior hindbrain and this domain is then subdivided into protorhombomeres,lineage-restricted compartments that presage formation of rhombomeres. The seven or eight rhombomeres that appear as transient bulges along the anterior-posterior axis of the hindbrain have unique identities. These identities are the basis for development of specified neurons within the hindbrain and of specialized migrating cranial neural crest cells(Moens and Prince, 2002; Trainor and Krumlauf,2001).

By late gastrula stages, cells that will contribute to the hindbrain are already committed to that fate (Woo and Fraser, 1998). At this time, a broad region of the presumptive posterior neurectoderm is distinguished by the expression of genes including hoxa1, hoxb1 and meis3(Alexandre et al., 1996; Kolm and Sive, 1995; Murphy and Hill, 1991; Prince et al., 1998; Sagerstrom et al., 2001; Salzberg et al., 1999). Further, because the anterior boundary of hoxA1 gene expression probably lies at the future r3-r4 break and only the posterior hindbrain tissue up to the r3-r4 boundary is dependent on retinoic acid (RA), the presumptive caudal hindbrain (r4-r8) appears to compose an early, distinct domain (Dupe and Lumsden, 2001; Gavalas and Krumlauf,2000).

The posterior hindbrain is subsequently subdivided, as indicated by restricted gene expression, including krox20 (egr2 –Zebrafish Information Network) in r3 and r5, and valentino(val)/kreisler/mafB in r5 and r6. Both krox20 and val functions are required for the correct expression of some of the rhombomere-specific Hox genes(Frohman et al., 1993; Giudicelli et al., 2003; Manzanares et al., 1999; Prince et al., 1998; Seitanidou et al., 1997). In turn, Hox gene expression domains delineate presumptive rhombomeres, and Hox gene function is required for the development of neurons and other cells produced within each rhombomere (Lumsden and Krumlauf, 1996; Moens and Prince, 2002; Trainor and Krumlauf, 2001). For instance, hoxb1 is expressed in future rhombomere 4 (r4) and is sufficient to provide ectopic r4 neuronal morphology (Bell et al., 1999; Vlachakis et al., 2001). In combination with hoxa1, hoxb1 is required for normal development of presumptive r4 in mice and zebrafish(Gavalas et al., 1998; McClintock et al., 2002; Rossel and Capecchi, 1999; Studer et al., 1998). Similarly, the Hox paralog group 3 genes are expressed in r5 and r6 and are required for formation of specific neurons and mesenchymal neural crest-derived structures (Manley and Capecchi, 1997).

Recent findings identify some additional factors required for posterior hindbrain segmentation. In the chick, fibroblast growth factor (Fgf) signals have been shown to be sufficient for ectopic induction of krox20 and mafB/kreisler in caudal hindbrain neuroepithelium and neural crest, and drug-based inhibition of Fgf signaling results in inhibition of krox20 and mafB/kreisler within their normal expression domains (Marin and Charnay,2000). In the zebrafish, Fgf signals emanating from the anterior hindbrain are required to initiate expression of posterior hindbrain gene expression, in particular val, krox20 and hoxb3. Loss of both fgf3 and fgf8 functions together results in a loss of r5 and r6 identity (Maves et al.,2002; Walshe et al.,2002). Loss of function of the gene variant hepatocyte nuclear factor 1 (vhnf1; tcf2 – Zebrafish Information Network) in the zebrafish results in small ears and loss of val and krox20 (r5) expression (Sun and Hopkins, 2001). Although vhnf1 knockout mice have been made, the role of vhnf1 in murine hindbrain development has not been studied (Barbacci et al., 1999; Coffinier et al., 1999).

It remains to be defined how the broad domain of gastrula stage posterior neuroectodermal gene expression is subdivided into individual rhombomeric domains. Partially explaining this, we show here that vhnf1 is expressed in a broad domain during gastrulation and that it is required for differentiation of caudal hindbrain rhombomeres by two distinct mechanisms. Through an obligate synergy with Fgf signals, vhnf1 promotes expression of val, thereby promoting r5 and r6 identity. In addition, vhnf1 represses hoxb1a expression independently of Fgf function, thereby limiting r4 identity to the appropriate narrow domain.

Fish were raised and bred according to standard methods(Westerfield, 1995). Lines used were: AB, Tübingen Long Fin, val^338b(Moens et al., 1996). We recently identified a novel allele of vhnf1: vhnf1^wi408 (E. Wiellette et al., unpublished). The molecular identity of our allele was confirmed by complementation crosses in which vhnf1^wi408 failed to complement the day 3 kidney defects and the lethality of both vhnf1^hi2169 (null) and vhnf1^hi548 (hypomorph)(Sun and Hopkins, 2001). In addition, vhnf1^wi408 fails to complement the ear defect of vhnf1^hi2169. Expression of vhnf1 transcript is completely missing at gastrulation and tailbud stages in 1/4 of the embryos derived from vhnf1^wi408 heterozygous parent crosses.

pCS2+ plasmids with cDNA containing vhnf1(Sun and Hopkins, 2001), val (Moens et al.,1998) or lacZ were linearized and transcribed using the mMessage mMachine kit (Ambion). Capped mRNA concentration was measured and RNAs were injected in the following final amounts: vhnf1, 25 pg; lacZ, 25 pg; val, 5 pg. β-Galactosidase (β-gal)was visualized after fixation of embryos overnight in BT fix(Westerfield, 1995) at 4°C by washing in PBT and then staining in β-gal stain buffer (1× PBS,4 mM MgCl₂, 3 mM K₄[Fe(CN)₆], 3 mM K₃[Fe(CN)₆]) + 0.2% X-gal at room temperature. Embryos were analysed first by β-gal stain, followed by dechorionation,dehydration in methanol overnight and then the standard in situ method (see below).

To knock down the functions of the fgf3 and fgf8 genes,morpholino oligomers (MOs) targeted to the translation start sites(Raible and Brand, 2001) were injected into 1-2 cell embryos. The final concentrations used were: 2.5 ng of each MO (Fig. 3C,D) or 1 ng of each MO (Fig. 5). In each case,two controls were performed in which a double concentration of one oligo or the other was used and, in each case, the strongest effect was observed by injection of the combination of MOs directed against both fgftranscripts. For analysis of earlier staged embryos(Fig. 3E,F), 0.8 ng of each MO was injected, with an unrelated negative control MO used to make 1.6 ng total MO in the injections of either fgf3MO or fgf8MO alone.

Standard methods for hybridization and for single and two-color labeling were used and have been described elsewhere(Sagerstrom et al., 1996). Probes used were: vhnf1 (Sun and Hopkins, 2001), krox20(Oxtoby and Jowett, 1993), hoxb1a, hoxb3, hoxb4 (Prince et al., 1998), myod(Weinberg et al., 1996), val (Moens et al.,1998), hoxb1b(Alexandre et al., 1996), fgf8 (Reifers et al.,1998), no tail(Schulte-Merker et al.,1992).

Reticulospinal neurons were labeled in embryos fixed at 48 hours of development, using 1:50 dilution of the primary antibody RMO44(anti-Neurofilament; Zymed Laboratories #13-0500), as described(Waskiewicz et al., 2001). Localization of RMO44 was visualized using a 1:50 dilution of FITC-α-mouse (Zymed). The brains were partially dissected and mounted for visualization by confocal microscopy.

Beads were prepared and coated in mouse Fgf8b protein (R&D Systems) [or bovine serrum albumen (BSA)] as described in Reifers et al. (Reifers et al.,2000). vhnf1 or val mRNA was injected at the 2-cell stage and the embryos were allowed to grow until the shield stage. Injected embryos were placed in the lid of a small Petri dish lightly coated with 3%methylcellulose and covered with normal Ringer's solution(Westerfield, 1995). A needle prepared as for injection was used to tear a small hole in the ectoderm and sharp forceps were then used to pick up a single bead and push it into the incision. The needle was then used to push the bead farther under the ectoderm. When all embryos on a dish lid were treated, the lid was placed in a standard Petri plate, which was then flooded with Embryo Medium(Westerfield, 1995). Embryos were left untouched until the appropriate stage for fixation, at which point they were gently removed from the methylcellulose and transferred for fixation, β-gal staining and in situ hybridization (as above).

Consistent with previously published data, we find that vhnf1^wi408 mutant embryos show defects in the expression of genes normally expressed in rhombomeres 4, 5, 6 and 7 (r4-r7)(Sun and Hopkins, 2001). Expression of Hox genes that distinguish r2 and r3 are unaffected in the vhnf1^wi408 mutant background (data not shown). However,expression of hoxb1a, which is normally limited to r4(Fig. 1A), is expanded into the posterior hindbrain in mutant embryos (Fig. 1B) (Sun and Hopkins,2001). hoxb3 expression, which is normally limited to r5 and r6 (Fig. 1C), is completely missing in vhnf1^wi408 mutants(Fig. 1D). Finally, hoxB4 expression, which normally has its anterior boundary at the r6-r7 limit and extends into the spinal column(Fig. 1E), is misexpressed in vhnf1^wi408 mutants such that the anterior boundary of expression is not sharp and is more posterior(Fig. 1F). krox20expression, which is normally present in r3 and r5, is missing in r5 in vhnf1^wi408 mutants. Although the r5 phenotype is 100%penetrant, vhnf1^wi408 mutants show variable levels of krox20 expression in r5 (compare Fig. 1B,D,F).

The earliest defect that we have been able to identify in vhnf1^wi408 mutants is a lack of valentino(val) expression, which normally begins at the end of gastrulation[10 hours post-fertilization (hpf)] in presumptive r5 and r6(Fig. 1G,H). In addition, by the beginning of somitogenesis, it is apparent that the r5 stripe of krox20 expression is severely reduced in vhnf1^wi408 mutants(Fig. 1I,J).

Morphological defects are consistent with the observed pattern formation defects. The reticulospinal neurons that develop in the hindbrain are visible by 48 hours of development, including the large Mauthner neurons, which are derived from r4 (Fig. 1K). vhnf1^wi408 mutants develop excess Mauthner neurons in parallel with loss of r5+r6-derived neurons(Fig. 1L). Although the normal r4-derived Mauthner neuron pair is always present in the correct location, the extra Mauthner neurons in vhnf1^wi408 mutants are routinely observed both within the normal locale for Mauthner neurons and in more posterior locations. In addition to the neuron identity changes, the otic vesicle is small and round (data not shown)(Sun and Hopkins, 2001). Together with published data, the gene misexpression and morphological defect data are consistent with a transformation of rhombomere identity from r5+r6 to r4 during late gastrula and tailbud stages.

It has been shown that vhnf1 expression is present in the hindbrain by tailbud stage and that the anterior boundary of vhnf1expression lies within the presumptive r5 domain by the four-somite stage(Sun and Hopkins, 2001). Because vhnf1 regulates val expression before tailbud stage,we wanted to generate an accurate description of the timing and initial localization of vhnf1 expression. An excellent marker of relative position in the gastrulating embryo is hoxb1b expression, which appears at early gastrula stages, and which has a sharp anterior border(zebrafish hoxb1b is the same as mouse hoxa1)(Alexandre et al., 1996). It is not clear what region of the hindbrain will be derived from this early domain of Hoxb1a expression, but it is likely to map at or near the future r3-r4 boundary. In situ hybridization of hoxb1b and vhnf1 shows that localized hindbrain expression of vhnf1 begins after (not shown)and with a more posterior boundary than that of hoxb1b(Fig. 2A). Co-localization of fgf8 and vhnf1 transcripts shows that vhnf1expression begins before that of fgf8 in the hindbrain (not shown)and that fgf8 expression is more anterior than that of vhnf1(Fig. 2B). There is a persistent gap of two or three cells between these fgf8 and vhnf1 expression domains. fgf8 expression abuts hoxb1b expression throughout late gastrula and tailbud stages(Fig. 2C).

Expression of krox20 begins in r5 at the beginning of somitogenesis (Oxtoby and Jowett,1993). At this stage, vhnf1 expression overlaps the posterior half of the r5 krox20 stripe(Fig. 2D), a relative position that is maintained through the four-somite stage(Sun and Hopkins, 2001). hoxb1a expression begins at 90% epiboly, initially throughout the posterior neural plate (not shown) (Prince et al., 1998). hoxb1a and vhnf1 expression overlap significantly at the end of gastrulation(Fig. 2E). However, during the tailbud stage, hoxb1a expression decreases in the posterior of its expression domain and, by the two-somite stage, it is limited to a stripe in presumptive r4 (Prince et al.,1998). At this stage, there is a gap between hoxb1aexpression and vhnf1 expression(Fig. 2F). In summary, vhnf1 expression begins at midgastrula stages, and its anterior boundary is limited from an early stage to a location that lies within the r5 krox20 expression domain (Fig. 2G).

To understand better the molecular interactions that allow vhnf1to generate r5+r6 identity, we considered other defects that cause a similar phenotype. fgf3 and fgf8 are expressed in the presumptive anterior hindbrain (Fig. 2G),and their partially redundant functions can be ablated by injection of MOs(Raible and Brand, 2001). Loss of Fgf signaling in the hindbrain results in loss of r5+r6 identity(Maves et al., 2002; Walshe et al., 2002). However,although loss of vhnf1 function results in expansion of r4 identity(Fig. 3A,B), loss of Fgf signals gives no expansion of hoxb1a(Fig. 3C)(Maves et al., 2002; Walshe et al., 2002). In this fgf loss-of-function background, hoxB4 expression does not expand to the anterior, suggesting that there is no posteriorization of the tissue (Fig. 3D).

To test whether an epistatic relationship exists between vhnf1 and Fgf signals, synthetic double mutants were made. Injection of fgf3 +fgf8 MOs into embryos derived from vhnf1^wi408carrier parents produced embryos all of the same phenotype, with hoxb1a expression limited to the r4 domain and hoxB4expression limited to the r6-r7 boundary(Fig. 3C,D). Because the loss of the combination of vhnf1 function and Fgf signals gives the same phenotype as loss of Fgf signals alone, Fgf signaling appears to be epistatic to vhnf1 function in hindbrain pattern formation.

To determine whether the epistasis between vhnf1 and Fgfs results from regulation of transcript levels, expression of each gene was tested in the other mutant background. Knockdown of Fgf signaling by injection of a mix of fgf3 + fgf8 MOs results in no change in the anterior-posterior pattern of vhnf1 expression(Fig. 3E). The effectiveness of the MOs to knock down Fgf function in the posterior hindbrain was monitored by loss of r5 krox20 expression in sibling embryos(Fig. 3F). Embryos generated from vhnf1^wi408 heterozygous parents all show the same pattern of fgf8 and fgf3 transcript expression at the one-somite stage, and so vhnf1 function is not required for normal fgf8 or fgf3 expression at this stage(Fig. 3G; not shown). Thus, Fgf signals function epistatically to vhnf1 to generate rhombomere identity in the posterior hindbrain, and this epistatic relationship is not based on regulation of RNA levels during gastrulation(Fig. 3H).

Although both Fgf signals and vhnf1 function are required to specify r5+r6, neither factor alone is sufficient. Ectopic vhnf1induces r5 identity only in the r4 domain, and ectopic Fgf has a limited ability to induce r5+r6 identity at late somitogenesis and primarily within the caudal hindbrain (Maves et al.,2002; Sun and Hopkins,2001). Therefore, the combination of these two factors was tested for an enhanced ability to induce r5+r6 identity.

vhnf1 RNA was injected into one cell at the two-cell stage. Injected embryos were grown to shield stage, at which point an Fgf8-coated bead was inserted. Embryos were then grown to approximately the three-somite stage and fixed (Fig. 4A). Injection of vhnf1 RNA and implantation of an Fgf8-coated bead results in strong induction of krox20 expression limited to the cells expressing vhnf1 and lying close to the Fgf8 bead (18/20 embryos). This effect is particularly strong throughout the neural plate(Fig. 4B), and is also robust in non-neural ectoderm (Fig. 4C). Embryos in which a BSA-coated bead was implanted near cells overexpressing vhnf1 show no induction of krox20 expression(6/6 embryos) (Fig. 4D). The Fgf8 bead alone does not induce any krox20 expression (10/10 embryos)(Fig. 4E), unless the bead lies adjacent to the normal domain of krox20 expression (not shown). To determine whether induced krox20 expression has r5-specific identity,induction of val expression was also examined and the combination of vhnf1 expression with the Fgf8 bead was found to result in induction of val expression (5/5 embryos)(Fig. 4F). It appears that the combination of vhnf1 and Fgf8 is sufficient for induction of early r5+r6 gene expression.

To determine whether induction of ectopic krox20 in response to vhnf1+Fgf8 is mediated by val function, the experiment was carried out in embryos produced by val heterozygous parents. Embryos were characterized as val mutants by their lack of r5 krox20expression in the uninjected half of the embryo. val mutant embryos show no induction of krox20 expression in cells expressing vhnf1 and located near the implanted Fgf8 bead(Fig. 4G) (7/7 embryos),whereas their wild-type siblings show strong induction of krox20 (not shown). Thus, ectopic induction of krox20 expression in the presence of vhnf1 and Fgf8 is dependent on val function.

To determine whether Fgf8 and val can synergize in a manner similar to Fgf8 and vhnf1, val RNA was injected and Fgf8-coated beads were added to the embryos. In this case, no induction of krox20expression is observed (15/19 embryos)(Fig. 4H), although weak krox20 induction is observed in a few cases in which the bead is located close to the endogenous krox20 domain (4/19 embryos). Thus,Fgf8+val is insufficient to activate ectopic krox20expression, suggesting that vhnf1 is specifically required to make hindbrain cells competent to respond to Fgf8 signals. In addition, this result shows that val RNA alone is insufficient to induce ectopic krox20 expression (Fig. 4H). Therefore, it is likely that additional factors induced by Fgf+vhnf1 are required in collaboration with val to activate krox20 expression.

To test whether vhnf1+Fgf8 can induce caudal hindbrain identity other than r5+r6, the expression of hoxb1a, a presumptive r4 marker,was analysed. hoxb1a expression is partially repressed in the region where vhnf1 expression overlaps endogenous hoxb1a expression(10/10 embryos) and is not ectopically induced in regions where the Fgf8 bead lies close to the vhnf1 expression (4/4 embryos)(Fig. 4I). Fgf+vhnf1is therefore insufficient to induce ectopic r4 identity.

Fgf8 can induce mesoderm identity(Griffin et al., 1998; Rodaway et al., 1999) and mesoderm is a source of neural posteriorizing signals including Fgfs(Koshida et al., 1998; Kudoh et al., 2002; Woo and Fraser, 1997). The expression of no tail (ntl) was examined to determine whether mesoderm was induced by Fgf8 under these experimental conditions and could mediate the synergy between vhnf1 and Fgf8. However, no ectopic ntl expression is observed in most embryos (13/15)(Fig. 4J), although, when the bead lies deep under the epiblast and close to the prechordal plate, ectopic ntl is observed close to the beads in deep tissue regions (2/15 embryos). Therefore, the co-operativity of Fgf signals and vhnf1function appears to occur within the ectoderm. In summary, Fgf signals synergize with vhnf1 to activate expression of r5+r6-specific genes,including the activation of val and krox20 expression(Fig. 4K).

The expansion of hoxb1a in vhnf1^wi408 mutants(Fig. 1B), paired with the suppression of hoxb1a by injection of vhnf1 RNA(Fig. 4I), suggests that one of the ways in which vhnf1 functions is to repress r4-specific hoxb1a expression. Therefore, we further examined hoxb1aexpression in response to ectopic vhnf1. Expression of hoxb1a at 90-100% epiboly is normal when vhnf1 is overexpressed (Fig. 5A). However, by the six-somite stage, expression of hoxb1a is completely repressed by overexpression of vhnf1, and krox20 expression appears in a single broad band (32/32 embryos)(Fig. 5B). Overexpression of lacZ alone has no effect on hoxb1a or krox20expression (17/17 embryos) (Fig. 5C). Thus, ectopic vhnf1 is able to repress hoxb1a expression in presumptive r4.

To test whether the repression of hoxb1a by vhnf1 occurs through val function, the ability of vhnf1 to repress hoxb1a in a val mutant background was tested. Ectopic vhnf1 is able to repress hoxb1a expression in this background (10/10 embryos) (Fig. 5D). val mutant embryos injected with lacZ do not show repression of hoxb1a (7/7 embryos) (data not shown). No krox20 expression remains after vhnf1 overexpression in val mutants, indicating that ectopic vhnf1 represses r3-specific krox20 as well as r4-specific hoxb1a. Therefore,it appears that ectopic vhnf1 transforms both presumptive r3 and r4 towards the r5 identity autonomously of val function.

To test whether the repression of hoxb1a by vhnf1requires the co-function of Fgf signals, the ability of ectopic vhnf1to repress hoxb1a in a fgf-compromised background was tested. Embryos were injected with MOs directed against fgf3 and fgf8 along with the vhnf1 RNA. Ectopic vhnf1represses hoxb1a expression in the absence of Fgf signals (22/22 embryos) (Fig. 5E) but co-injection of lacZ with the fgf MOs does not reduce the extent of hoxb1a expression (59/59 embryos)(Fig. 5F). These data imply that vhnf1 functions independently of Fgf signals to repress hoxb1a expression. Because knockdown of fgfs based on the MOs used in this experiment does not lead to complete repression of r5 krox20 expression, it is interesting that there is no remaining krox20 expression in the tissue injected with vhnf1 RNA and anti-fgf MOs. This might reflect a synergistic effect that results from repression of fgf8 expression by vhnf1(Fig. 5G,H), which might give a stronger loss of fgf functions and therefore a complete repression of r5 krox20 expression. fgf8 is expressed throughout presumptive anterior hindbrain tissue at tailbud stage(Reifers et al., 1998), and vhnf1 appears to be sufficient to repress anterior hindbrain identity during gastrulation (Fig. 5G).

The expression of multiple genes is dependent on vhnf1 function,including val, krox20 and hoxb3. Of these, val is the earliest expressed, and loss of val function also results in loss of krox20 in presumptive r5 and loss of hoxb3 expression(Moens et al., 1996; Prince et al., 1998). To test whether vhnf1 is independently required for regulation of krox20 and hoxb3, val RNA was injected into vhnf1^wi408 mutant embryos and expression of target genes was analyzed. Expression of krox20 is unaffected in wild-type embryos by overexpression of val (Fig. 6A). However, injection of val RNA is sufficient to rescue r5 krox20 expression in vhnf1^wi408 mutant embryos in the injected side of the embryo (7/8 embryos) and it is striking that krox20 expression induced by val RNA is limited to the presumptive r5 domain (Fig. 6B). lacZ RNA alone has no ability to rescue krox20 expression (Fig. 6C).

In contrast to the strong rescue of krox20 expression by val RNA, expression of hoxb3 is infrequently rescued,although hoxb3 expression is occasionally observed after valinjection into vhnf1^wi408 embryos(Fig. 6D-F). However, the frequency of rescue is lower (5/14 embryos) than observed for krox20rescue, suggesting that the val RNA injected is not sufficient to rescue to hoxb3 expression. This result could be due to a requirement for vhnf1 functions not mediated by val function or to a requirement for a higher concentration of val RNA for hoxb3expression.

We have shown that vhnf1 functions in the zebrafish hindbrain to generate individual rhombomere identity in the caudal hindbrain. vhnf1 functions through two mechanisms to specify rhombomeres 5 and 6, first by synergizing with Fgf signals to activate r5+r6-specific genes and second by limiting hoxb1a expression to presumptive r4 independently of fgf3 and fgf8 function.

The data presented here define the mechanism by which the initially broad expression domains of the future hindbrain are subdivided(Fig. 7). Soon after hoxb1b expression appears with an anterior limit near the future r3-r4 boundary, vhnf1 expression begins, with a more posterior limit of expression, probably within the future r5 region. It is not clear how vhnf1 expression is initiated or what limits its anterior boundary. Expression of vhnf1 is sensitive to retinoic acid (RA) (E. Wiellette and H. Sive, unpublished) and RA receptor binding sites have been identified in the mouse vhnf1 promoter(Power and Cereghini, 1996),so it is possible that RA directly regulates vhnf1 transcription. However, it is also possible that the observed RA sensitivity is mediated by hoxb1b function; when Hox gene function is removed from zebrafish by making embryos devoid of pbx gene functions, vhnf1transcription is not initiated (Waskiewicz et al., 2002).

In late gastrula stages, fgf3 and fgf8 expression begins in the anterior hindbrain (Maves et al.,2002; Reifers et al.,1998; Walshe et al.,2002), and the combination of Fgf signals with vhnf1 is sufficient to initiate expression of val and krox20. It seems likely, based on the demonstrated significance of the presumptive r4 domain in caudal hindbrain pattern formation(Maves et al., 2002; Walshe et al., 2002), that the significant Fgf signals received by vhnf1-expressing cells are secreted from the anterior hindbrain domain. Consistent with this, the cellular response as measured by val expression does not begin until after the anterior hindbrain expression of fgf3 and fgf8 begins.

vhnf1 function is required for activation of val and krox20 expression. However, the anterior boundary of vhnf1expression lies posterior to the anterior boundary of krox20 r5 expression, which is comparable to that of val(Moens et al., 1998). Because vhnf1 encodes a putative transcription factor, it seems likely that Vhnf1 protein is not acting directly to regulate the transcription of val or krox20 but might rather be regulating transcription of an extracellular signal, which works at a distance of one or two cell diameters to control the expression of val and krox20.

Also, at the end of gastrulation, hoxb1a expression appears in a broad domain of the posterior neural plate, with an anterior boundary similar to that of hoxb1b. Restriction of hoxb1a expression to presumptive r4 during somitogenesis is conserved in mouse(Murphy and Hill, 1991), and downregulation of hoxb1a might be necessary to allow normal development of r5-r7. Although hoxb1a expression is activated throughout the posterior hindbrain in a vhnf1-insensitive manner, hoxb1a can be repressed by ectopic vhnf1 starting at tailbud stage, the same stage at which hoxb1a transcripts are first downregulated outside of r4. hoxb1a repression in future r5-r7 is probably brought about at least in part by the function of vhnf1,independent of Fgf and val functions. The separate roles of vhnf1, as an activator in conjunction with Fgf signals and as a repressor independent of Fgf signals, might reflect distinct molecular interactions, either with cofactor proteins or with DNA. Two different forms of Vhnf1 protein, which result from alternative splicing, have been characterized as having different DNA binding affinities and transactivation strengths (Ringeisen et al.,1993). It is possible that these isoforms provide the different functional specificities in the caudal hindbrain.

Although one of the central functions of vhnf1 in hindbrain pattern formation is activation of val expression, loss of vhnf1 has a more severe phenotype in the forming hindbrain than loss of val function. Loss of val results in the production of a narrowed `rX' domain in place of r5 and r6(Moens et al., 1996). Like the mis-specified r5 and r6 domain in vhnf1 mutant embryos, the val mutant rX domain does not express hoxb3, and the posterior boundary of hoxb1a and the anterior boundary of hoxB4 expression are similarly indistinct(Prince et al., 1998). However, the rX domain of val mutants is significantly narrower than the combination of r5 and r6 domains, whereas vhnf1 mutants show no apparent reduction of tissue. In addition, the reticulospinal neurons in val mutants are correctly specified(Moens et al., 1996).

The genetic distinctions between vhnf1 and val are paralleled by differences in molecular capacities. The combination of val+Fgf is not sufficient to induce r5 identity outside the r5 domain, whereas the embryo is broadly sensitive to vhnf1+Fgf function. Conversely, Fgf+vhnf1 cannot induce ectopic krox20in a val mutant background, suggesting that each transcription factor has unique and necessary functions in hindbrain pattern formation. Although overexpression of val in the vhnf1^wi408 mutant background results in recovery of krox20 expression, this appears only in the r5 domain, with no ectopic krox20 expression detected. This suggests that only limited domains are competent to respond to val function, potentially based on the presence of a cofactor.

Knockout of the murine vhnf1 gene results in early death as a result of failure to form visceral endoderm(Barbacci et al., 1999; Coffinier et al., 1999). As yet, no studies of later loss-of-function of vhnf1 have been published. However, vhnf1 is expressed in the mouse hindbrain in a broad domain that lies close to the otic placode, a position that is similar to the domain of expression in zebrafish(Barbacci et al., 1999; Coffinier et al., 1999). In addition, vhnf1 expression is detected before the onset of kreisler expression (Cordes and Barsh, 1994), suggesting that vhnf1 could have a conserved role in regulation of kreisler expression in the developing mouse hindbrain. Furthermore, expression of mouse hoxb1, the most likely functional homolog of zebrafish hoxb1a, is downregulated in the most posterior hindbrain at a time soon after initiation of vhnf1transcription initiation (Murphy and Hill,1991). Finally, although fgf8 transcript expression is restricted to r1 and the midbrain-hindbrain boundary (MHB) in mice, fgf3 expression is observed in r4 at the same time that kreisler expression is induced in presumptive r5+r6, and before the upregulation of fgf3 in the r5+r6 domain(Cordes and Barsh, 1994; Joyner et al., 2000; Mahmood et al., 1996). Thus,it is likely that the restriction of rhombomere-specific identities in the caudal hindbrain of the mouse follows a molecular mechanism similar to the one we have described for zebrafish.

Fgf signals are reused throughout development and yet the cellular response to the signal varies based on time and location. In the developing zebrafish brain, fgf8 function is required not only for pattern formation in the caudal hindbrain but also for formation of the MHB(Reifers et al., 1998). In the posterior hindbrain, Fgf signaling results in activation of valexpression, whereas, at the MHB, Fgf signaling results in activation of gbx2, fkd3 and spry4(Reim and Brand, 2002). During late gastrulation stages, the anterior hindbrain expresses fgf3 and fgf8 in a domain that lies between the forming MHB and the posterior hindbrain. Thus, it appears that the anterior hindbrain domain provides a source of Fgf signals for both the MHB and posterior hindbrain. However, the molecular and morphological results of this signaling are different.

Various features of Fgf signal transduction might provide distinct cellular responses in the MHB and posterior hindbrain, including different Fgf receptor interactions and negative feedback regulation of Fgf signal transduction. One potential distinction in Fgf signaling outcome is the presence of intracellular cofactors. We have shown here that the expression of vhnf1 in or near cells receiving Fgf signals is sufficient to promote activation of posterior hindbrain gene expression. Similarly, it has been shown that the presence of pou2 in cells receiving Fgf signals is required for activation of MHB target genes including gbx2, fkd3 and spry4 (Reim and Brand,2002). Thus, it is possible that one of the ways in which cells generate a differential response to Fgf signals is through the presence of a transcriptional cofactor such as pou2 or vhnf1, which provides promoter selection specificity. Further work to characterize the unique cellular responses to Fgf signals in the presence of pou2 or vhnf1 will help determine the role of transcription factors as mediators of signaling specificity.

We thank members of the Sive lab for helpful comments. In addition, we would like to thank L. Maves and C. Moens for helpful discussions, and members of the Moens lab for sending plasmids, protocols and fish. Finally, we thank N. Hopkins and Z. Sun for the vhnf1 cDNA and for communicating results prior to publication. We are indebted to B. Pratt for excellent fish husbandry. Supported by NIH HD08704 to E.L.W. and NIH MH059942 to H.S.