Two tcf3 genes cooperate to pattern the zebrafish brain

The blastoderm margin is the source of secreted molecules that regulate gene expression along the rostral-caudal axis of zebrafish(Woo and Fraser, 1997) andXenopus embryos. Multiple signals, including Wnts, FGFs and Activin/Nodal-related factors, cooperate to establish a gradient of caudalizing activity in the gastrula with its high end around the blastoderm margin and its low end near the animal pole(McGrew et al., 1997;Thisse et al., 2000). By the end of gastrulation the isthmic organizer at the midbrain-hindbrain boundary(MHB) becomes a source of additional Wnts and FGFs and the prechordal plate and anterior neural ridge become sources of Wnt antagonists(Eroshkin et al., 2002;Hashimoto et al., 2000;Houart et al., 2002;Kim et al., 2002;Shinya et al., 2000). These additional sources of caudalizing factors and their antagonists are thought to further modulate the shape of the caudalizing activity gradient in the anterior neural plate.

Analysis of zebrafish maternal-zygotic headless (hereafter simply called hdl) mutants has suggested that caudalizing factors, in particular Wnts, operate in the context of basal repression provided by this Tcf3 homolog (Kim et al.,2000). Canonical Wnt signaling facilitates the expression of downstream target genes through β-catenin, which associates with Lef/Tcf proteins that bind to DNA regulatory elements(Barker et al., 2000). Whenβ-catenin levels are low, Lef/Tcf proteins maintain target genes in a repressed state (Brannon et al.,1997). Although Lef/Tcf transcription factors can have dual roles in activation or repression of target genes, it appears that in vivo Lef1 has a primary role in activation, whereas Tcf3 has a primary role as a repressor(Kengaku et al., 1998;Houston et al., 2002).

Several studies have converged to provide evidence for the role of Wnt/β-catenin activity in defining discrete domains of gene expression along the rostral-caudal axis of the neural plate and in the subsequent establishment of rostral-caudal compartments of the vertebrate neural tube(Domingos et al., 2001;Erter et al., 2001;Hashimoto et al., 2000;Lekven et al., 2001;McGrew et al., 1995;Nordstrom et al., 2002;Shinya et al., 2000;van de Water et al., 2001;Houart et al., 2002;Kiecker and Niehrs, 2001;Kim et al., 2002). These studies have shown that exaggerated Wnt signaling leads to loss of rostral neural domains and expansion of more caudal neural domains, whereas reduced Wnt signaling leads to expansion of rostral neural domains and loss of more caudal domains. However, in some contexts β-catenin is not primarily required for activation of target genes but rather for antagonizing repression mediated by Tcf homologs (Chan and Struhl,2002; Houston et al.,2002). The primary role for Wnt/β-catenin signaling in rostral-caudal patterning thus remains unclear.

Consistent with the role of hdl in repressing genes that define relatively caudal domains, hdl mutants are characterized by expanded expression of genes that define the MHB domain. At the same time, the forebrain, whose specification is most dependent on the repression of caudal genes, is lost in hdl mutants. Interestingly, patterning defects are restricted to the rostral neurectoderm, leaving the hindbrain and spinal cord relatively unaffected. Many zygotic hdl mutants survive to adulthood,suggesting that other lef/tcf genes may limit the severity of phenotypes observed in these fish.

Previously, we identified a partial cDNA clone of a second zebrafishtcf3 gene, which we named tcf3b(Dorsky et al., 1999). We report here the full-length sequence of tcf3b, and show that although both hdl and tcf3b are expressed maternally and throughout development, there are important differences in their expression patterns,most notably during early gastrulation. By examining loss-of-function phenotypes and performing mRNA rescue experiments, we determine that both genes have unique and cooperative roles in early zebrafish development. By comparing the abilities of tcf3b and lef1 to suppress the caudalization in hdl mutants, we reveal functional differences between these lef/tcf family members in repressing caudal target genes. In addition, we show that Wnt8 function is primarily required in the neurectoderm for de-repression of caudal genes rather than for their activation. Finally, by analyzing changes in the shape of gene expression domains caused by reduction of tcf3b function in hdlmutants, we make specific predictions about the shape of the caudalizing activity gradient in the neurectoderm.

A partial cDNA clone of tcf3b from a phage library was used to design primers for 5′ RACE, which was performed using the SMART PCR Kit(Clontech). To obtain a full-length cDNA, we amplified the SMART cDNA library with primers corresponding to the 5′ and 3′ ends of the coding sequence using Expand polymerase (Roche) and inserted the PCR product into the vector pCS2p+ (Turner and Weintraub,1994). This clone was sequenced and submitted to GenBank(#AY221031).

PCR was performed on cDNA from various developmental stages, using the following primers and an annealing temperature of 50°C for 30 cycles. Products were run on a 2% agarose gel and stained with ethidium bromide.

• tcf3b-F 5′ AGGTGGCATTCGCTATCACG 3′

• tcf3b-R 5′ TTTGGTGGTCAGGGACAACG 3′

• max-L 5′ GCCGAAGAATGAGCGACAAC 3′

• max-R 5′ CTGCTGTGTGTGTGGTTTTTC 3′

In situ hybridization with digoxigenin-labeled mRNA probes was performed as described previously (Oxtoby and Jowett,1993). Probes for hdl, tcf3b, lef1 and wnt1 were made from full-length cDNAs. Digital images were processed with Adobe Photoshop software.

Other plasmids used to make in situ probes have been published previously:opl (Grinblat et al.,1998), pax2.1 (Krauss et al., 1991a), pax6(Krauss et al., 1991b),gbx1 (Itoh et al.,2002), gsc (Stachel et al., 1993), krox20(Oxtoby and Jowett, 1993),en2 (Ekker et al.,1992), isl1 (Inoue et al., 1994), and mar(Popperl et al., 2000). Double in situs using digoxigenin- and fluorescein-labeled RNA probes were performed as described (Jowett,2001).

Fixed embryos were soaked in 0.1 mg/ml AlexaFluor 594 phalloidin (Molecular Probes) for one hour at room temperature and rinsed in PBS. Embryos were mounted in glass coverslips and imaged on a Nikon PCM2000 confocal microscope.

We added 200 ng of hdl and tcf3b cDNAs in pCS2+(Turner and Weintraub, 1994)to TNT Quick Coupled reactions (Promega, Madison, WI). Morpholinos (MOs) were added to the reactions as indicated, and reactions were labeled with³⁵S Methionine. Following incubation, reactions were run on 10%acrylamide gels, dried and exposed overnight for autoradiography.

Zebrafish were raised and maintained under standard conditions. To collect maternal zygotic headless^m881 mutant embryos, heterozygous males and homozygous females were crossed(Kim et al., 2000).

MO antisense oligonucleotides were designed by and purchased from Gene Tools (Philomath, OR). The MO sequences are as follows:

• hdl: 5′ CTCCGTTTAACTGAGGCATGTTGGC 3′

• tcf3b: 5′ CGCCTCCGTTAAGCTGCGGCATGTT 3′

For both MOs, doses ranging from 500 pg-5 ng were injected. After examining phenotypes and embryo survival, 1 ng was chosen as the optimal dose for producing specific phenotypes. wnt8 MOs were kindly provided by Arne Lekven (Lekven et al.,2001).

For mRNA injections, transcripts were synthesized using the mMessage mMachine kit (Ambion). Expression constructs were made by inserting full-length cDNAs into pCS2+ (Turner and Weintraub, 1994). For rescue experiments, we injected approximately 1 ng MO with 10 pg hdl, tcf3b, Δtcf3b, lef1 orΔlef1 mRNA and 100 pg wnt1 mRNA.

We used 5′ RACE to screen a cDNA library from somitogenesis-stage zebrafish embryos with primers derived from a partial clone of tcf3b. By performing RT-PCR on the same cDNA library, we obtained a full-length open reading frame for tcf3b (Fig. 1A). The predicted Tcf3b protein is 82% identical to Hdl, its closest homolog. The overall homology between the two proteins suggests that Hdl and Tcf3b are both orthologs of Tcf3 in other vertebrates. By RT-PCR analysis, we showed that tcf3b is expressed maternally and throughout gastrulation and somitogenesis, similarly to hdl(Fig. 1B). Based on their substantial homology and similar temporal expression patterns, we hypothesized that the two genes could play cooperative roles during zebrafish development.

To examine possible sites of cooperation between the two genes, we performed in situ hybridizations for hdl and tcf3b at multiple developmental stages (Fig. 2). Although both genes are widely expressed maternally (not shown), we observed a sharp difference in hdl and tcf3bexpression soon after zygotic transcription begins. At shield stage,hdl is expressed broadly throughout the epiblast, whereastcf3b shows very low expression in this tissue(Fig. 2A,B). Following gastrulation, both hdl and tcf3b show specific expression in the rostral neurectoderm (Fig. 2C,D). Although both genes are expressed at a low level in the notochord, only hdl expression was seen in the tailbud(Fig. 2E,F). During somitogenesis, we observed expression of both genes throughout the developing brain (Fig. 2G,H). Caudally, we once again detected only hdl in the tailbud and presomitic mesoderm(Fig. 2I,J). At late somitogenesis stages, both genes are still expressed throughout the brain(Fig. 2K,L).

To investigate the function of tcf3b we used MO antisense oligonucleotides to inhibit mRNA translation(Heasman et al., 2000;Nasevicius and Ekker, 2000). By knocking down the function of both hdl and tcf3b, we hoped to uncover functional overlap between the two genes. Using expression plasmids that contained sequences complementary to the MOs, we found that each MO could specifically block translation of the targeted gene in vitro(Fig. 3A). Importantly, the sequences of the two genes differ by only four bases in the targeted region.

We first examined the phenotype produced by a MO targeted againsthdl, and found that injection of 1 ng of hdl MO at the one-cell stage can completely reproduce the hdl mutant phenotype(Fig. 3C). No obvious phenotypes were observed in trunk or tail regions, again consistent with genetic loss-of-function data. We conclude that our hdl MO can specifically block hdl gene function.

We next examined the phenotype following injection of one-cell embryos with 1 ng of a MO targeted against tcf3b. In contrast to hdl, we observed no gross morphological abnormalities through 72 hours post-fertilization (h.p.f.), except that the brain appeared slightly smaller and there was minor cardiac edema (Fig. 3D).

We then co-injected embryos with 1 ng each of hdl andtcf3b MOs. The size of the brain in many of these embryos was substantially smaller than with either MO alone, especially when examined from the ear to the rostral limit of the brain(Fig. 3E). These results suggest that the two genes may have cooperative roles in these tissues.

To determine whether hdl and tcf3b encode proteins with similar functions, we attempted to rescue the hdl phenotype by overexpressing tcf3b. We first titrated the dose of tcf3bmRNA to find a concentration that was insufficient to produce a phenotype when overexpressed. When 100 pg of hdl or tcf3b mRNAs were injected at the one-cell stage, we observed identical phenotypes that included cyclopia, short tails and somite defects (not shown). Because Wnt signaling can regulate the activity of Tcf3, it is difficult to interpret such overexpression phenotypes. Nevertheless, we hypothesized that at a gross level, the two genes may encode proteins with equivalent functions. At a dose of 10 pg, we observed no obvious defects in injected embryos, so we chose this amount for our rescue experiments.

Injecting 10 pg of tcf3b mRNA with 1 ng of hdl MO resulted in rescue of the hdl phenotype. Expression ofpax2.1, a marker for the MHB, was expanded rostrally in 89% (25/28)of embryos injected with the hdl MO alone(Fig. 3F,G). Co-injection oftcf3b mRNA eliminated the expansion of pax2.1 in 78% (25/32)of embryos examined (Fig. 3H). Injection of tcf3b was also able to rescue hdl morphology at 24 h.p.f. (Fig. 3I). Although only 8% of embryos injected with the hdl MO had eyes, this fraction increased to 91% when tcf3b mRNA was co-injected(Table 1). To further demonstrate that tcf3b mRNA suppressed the defects caused by loss ofhdl function, we injected tcf3b mRNA into hdlmutant embryos. Again, injection of tcf3b mRNA increased the number of embryos with eyes from 0% to 91% (57/64). We therefore conclude that ectopic expression of tcf3b can functionally replace hdl in patterning the embryonic brain.

Low-level overexpression of wnt1 causes caudalization of the neurectoderm resulting in a phenotype similar to hdl mutants(Dorsky et al., 1998). To test whether tcf3b function is sensitive to Wnt signaling, we examined how co-injection of wnt1 mRNA affected the ability of tcf3b to suppress the eyeless phenotype induced by the hdl MO. Co-injection ofwnt1 and tcf3b mRNAs and hdl MO resulted in only 4%of embryos having eyes (Table 1), indicating a lack of rescue. A truncated form oftcf3b that lacks the β-catenin binding domain(Δtcf3b) decreases expression of a β-catenin-dependent reporter transgene (Dorsky et al.,2002) (data not shown) and rescues the hdl MO phenotype in 78% of injected embryos (Table 1). However, rescue by Δtcf3b mRNA was not reversible by wnt1, as 79% of embryos injected with both mRNAs had eyes (Table 1). These data suggest that Hdl and Tcf3b both act as repressors in vivo, and that target genes can be de-repressed by Wnt signaling via the Tcf3 β-catenin binding domain.

Different Lef/Tcf factors are known to play distinct roles in development(Roel et al., 2002). To further characterize the functions of these proteins in activating and repressing target genes, we attempted to suppress the hdl MO phenotype with a third family member, lef1. In our experiments,neither lef1 (3% with eyes) nor Δlef1, a form lacking the β-catenin binding domain, (5% with eyes) could compensate for the loss of hdl function (Table 1). The failure of lef1 to suppress the hdl MO phenotype is consistent with its suggested primary role in activating rather than repressing target genes (Kengaku et al., 1998; Merrill et al.,2001).

Analysis of hdl mutant embryos showed that this gene plays an essential role in forebrain specification. However, hdl andtcf3b double MO injections resulted in a more severe phenotype than that produced by the hdl MO alone(Fig. 3E), suggesting a cooperative role for hdl and tcf3b in early development. Furthermore, tcf3b is expressed both maternally and zygotically,indicating that it may function in the early embryo. We therefore explored the possibility that the two tcf3 genes contribute to a common function during embryogenesis.

To examine rostral-caudal neural patterning, we compared the expression ofpax6, pax2.1 and gbx1, which respectively mark the future eye field and dorsal diencephalon, the MHB and the rostral hindbrain(Fig. 4). In hdlmutants, the size of the rostral pax6 expression domain was reduced in 83% (20/24) of the embryos (Fig. 4A2). Our observations of pax6 expression were restricted to the rostral neurectoderm, because mechanisms that determine its expression in the caudal neurectoderm remain poorly defined at this stage. There was also, as described earlier, a rostral expansion of the MHB domain marked bypax2.1 expression in 26% (7/27) of the embryos(Fig. 4B2). There was no obvious change, however, in the size of the gbx1 expression domain inhdl mutants (Fig. 4C2).

When 1 ng tcf3b MO was injected into wild-type embryos there was no significant change in the expression of pax6, pax2.1 andgbx1 (data not shown). However, when the same amount was injected into hdl mutants a range of phenotypes was seen that reflected further caudalization of the brain. In all hdl embryos injected withtcf3b MO, the rostral pax6 expression domain was completely lost (100%, 34/34) (Fig. 4A3). We observed a variable change in pax2.1, first extending rostrally to define a broad oval expression domain in the rostral neurectoderm, then becoming restricted to the rostral edge of this domain (84%, 21/25)(Fig. 4B3-4) or in more severely caudalized embryos lost completely (8%, 2/25)(Fig. 4B5). Initially,gbx1 expression expanded rostrally in an arc-like manner to enclose an oval domain (Fig. 4C3). In more severely caudalized embryos, gbx1 expression spread into the oval domain (43%, 9/21) (Fig. 4C4-5) and was eventually restricted to an arc-like domain near the rostral edge of the neural plate (Fig. 4C5).

A striking feature of caudalized embryos is the systematic manner in which expression of caudal genes expands rostrally to resemble the wild-type gene expression in these compartments. For example, pax2.1 expression expands in hdl mutants from its normal domain at the MHB to a rostral domain that resembles wild-type pax6 expression (compareFig. 4A1 andFig. 4B2). At tailbud stage,the diencephalic marker pax6 is expressed in a compartment that extends rostrally to enclose an unlabelled area(Fig. 4D). At the same time,pax2.1 and gbx1 are expressed in more caudal compartments where they define the MHB domain and rostral hindbrain, respectively(Fig. 4D). In caudalized embryos, pax6 expression is lost and pax2.1 expands rostrally within an oval domain that is surrounded by gbx1 expression(Fig. 4E, also compareFig. 4B3 andFig. 4C3). As described earlier, in many caudalized embryos pax2.1 is most prominently expressed in a rostral crescent within this oval domain(Fig. 4B4,Fig. 4E), resembling the wild-type expression of genes such as emx1 that define the prospective telencephalon (Houart et al.,2002).

We showed that although tcf3b can functionally replacehdl, a third family member, lef1, which is more likely to have a role in gene activation, cannot. The respective roles of hdland lef1 in repressing and activating target genes correlates with their complementary expression in the blastoderm at the shield stage(Fig. 5A,B). lef1 is expressed in a domain that overlaps with wnt8 at the ventrolateral blastoderm margin and where Wnt/β-catenin signaling is expected to be high. In contrast, lef1 expression is excluded from the prospective rostral neurectoderm, where hdl is expressed and where Wnt/β-catenin signaling is expected to be low. In embryos injected withhdl and tcf3b MOs, lef1 expression expands to cover most of the blastoderm at the shield stage(Fig. 5C). This indicates that caudalization of neurectoderm following loss of Tcf3 function is preceded by expanded lef1 expression at early gastrulation. Furthermore, it indicates that one role of Tcf3 is to restrict lef1 expression to the blastoderm margin during normal development. It is important to note thathdl and tcf3b MOs do not cause increased activation of aβ-catenin-dependent reporter transgene(Dorsky et al., 2002) (data not shown). This suggests that Wnt/β-catenin signaling may be able to antagonize repression by Tcf3, but it may not play a role in the direct activation of caudal genes.

To determine whether Wnts are essential for de-repression of caudal genes but not for their activation, we examined whether wnt8 is required for caudal gene expression in the absence of hdl function. Embryos injected with hdl MO showed a significant reduction in rostralpax6 expression, an expansion of the pax2.1 expression domain and a rostral shift in the position of these expression domains(Fig. 5D-I). In contrast,injection of wnt8 MO expanded rostral pax6 expression and shifted both pax6 and pax2.1 expression domains caudally(Fig. 5M-O), confirming previous studies that have shown loss of this Wnt homolog leads to expansion of rostral domains and reduction of caudal domains(Erter et al., 2001;Lekven et al., 2001). Reduction of Wnt8 function also caused dorsalization and subsequent broadening of the neurectoderm. When embryos were co-injected with hdl andwnt8 MOs they were still mildly dorsalized, however, pax6expression was reduced and pax2.1 was expanded, showing thatwnt8 is not required for caudalization in the absence of hdlfunction (Fig. 5J-L). These observations support the conclusion that Wnts primarily contribute to de-repression of caudal genes in the neurectoderm.

The analysis of embryos lacking both hdl and tcf3bfunction revealed the cooperative roles of these genes in rostral-caudal neural patterning. To determine the unique function of tcf3b we examined molecular markers for brain development following tcf3b MO injection. At the earliest stages of brain development, the basic patterning of injected embryos appeared normal. We examined the expression ofpax2.1 at bud stage and found no changes compared to uninjected embryos (not shown), indicating that rostral-caudal patterning is unaffected by the tcf3b MO. In addition, other MHB markers (wnt1 anden2) and dorsal/ventral patterning markers (pax2.1 andaxial) appear normal at 18 somites (not shown). In fact, we were unable to detect any effects on brain development until 24 h.p.f., when we examined hindbrain and MHB morphology.

We observed a severe loss of hindbrain rhombomere segmentation intcf3b MO-injected embryos at 24 h.p.f. This phenotype was never observed in either hdl mutant or MO-injected embryos, suggesting a unique function for tcf3b in hindbrain development. Analysis of hindbrain morphology by phalloidin staining revealed a lack of physical boundaries (Fig. 6A-D). At the molecular level, wnt1 is normally expressed at inter-rhombomere boundaries along the dorsolateral edge of the hindbrain(Fig. 6E). However, in injected embryos, wnt1 expression is noticeably absent from boundary areas and appears uniform or patchy along the hindbrain margin(Fig. 6F). Themariposa gene is expressed ventrally in the hindbrain, again localized to rhombomere boundaries (Fig. 6G). Following tcf3b MO injection, there is a uniform level of expression throughout the hindbrain(Fig. 6H). We also observed a defect in the closure of the dorsal MHB(Fig. 6I,J), although the position and identity of the MHB appear normal as marked by the expression ofen2 (Fig. 6K,L). Rhombomere identity also does not appear to be affected in these embryos,because patterning markers such as krox20 (rhombomeres 3 and 5) are expressed normally (Fig. 6M,N). Neurogenesis is also normal, as indicated by isl1, which marks ventral neurons of the hindbrain (Fig. 6O,P). These phenotypes indicate defects in brain morphogenesis,rather than in patterning or differentiation. We attempted to rescue these hindbrain defects by co-injection of both hdl and tcf3b mRNA with the MO, but were unable to restore normal morphology because of the fact that both mRNAs produced similar defects when overexpressed (not shown).

We have cloned a second zebrafish tcf3 gene and have shown that it functions with hdl in early brain patterning and later on its own in establishment of morphological boundaries in the neural tube. Although bothhdl and tcf3b are expressed maternally and in the rostral neurectoderm by late gastrulation, only hdl is widely expressed in the epiblast at the shield stage. This finding may explain why rostral-caudal neural patterning that takes place during early gastrulation is specifically affected when hdl function is lost.

Reduction of tcf3b function does not significantly affect early patterning in wild-type embryos, suggesting that hdl plays a more prominent role in this process. However, tcf3b can rescue thehdl mutant phenotype and reduction of tcf3b function leads to further caudalization of hdl mutant embryos. This suggests that low levels of Tcf3b present in the embryo during early gastrulation help limit the degree of caudalization caused by loss of Hdl function. Indeed, in zygotichdl mutants, the loss of rostral neural structures is minimal and persistence of maternal hdl and tcf3b transcripts permits some mutants to be grown up to adulthood.

The essential function of Hdl protein revealed in zebrafish(Kim et al., 2000) andXenopus (Brannon et al.,1997; Houston et al.,2002) is as a repressor. In contrast, Lef1 appears to be a more effective activator of target genes in vivo(Kengaku et al., 1998;DasGupta and Fuchs, 1999). Complementary roles of Hdl/Tcf3b and Lef1 in mediating repression and activation of target genes, respectively, are consistent with their complementary expression during early zebrafish development(Fig. 5A,B)(Dorsky et al., 2002), as well as their complementary roles in mouse and Xenopus development(Merrill et al., 2001;Roel et al., 2002).

Our experiments support a mechanism in which genes that are Hdl and Tcf3b targets in the neurectoderm do not require Wnt/β-catenin signaling for either their repression or their endogenous activation. Furthermore, in the absence of Tcf3 function we observe an increase in lef1 expression without a corresponding increase in expression of a β-catenin-responsive reporter (Dorsky et al.,2002). Other Lef/Tcf factors have been shown to activate targets in lymphocytes (Travis et al.,1991; van de Wetering et al.,1991), and Xenopus embryos(Labbe et al., 2000) in a Wnt/β-catenin-independent manner. Through this evidence, it is possible to conclude that Hdl and Tcf3b function in a Wnt-independent manner as well. However, data from our studies indicates that the developmental roles of these factors are closely linked to Wnt signaling in the embryo. First, we show that rescue of hdl mutants is reversible by Wnt signaling in a manner that requires β-catenin binding. Second, loss of Wnt8 function results in expansion of the same rostral genes that require Tcf3 function for their expression. Although Wnt8 may act through Lef1 to activate target genes in the ventrolateral mesoderm, we conclude that it primarily antagonizes Tcf3 function in the neurectoderm.

Wnt signals are only a part of the system that determines gene expression along the rostral-caudal axis. Multiple caudalizing factors, including Wnts,FGFs and Activin/Nodal-related factors, contribute to rostral-caudal patterning (McGrew et al.,1997; Altmann and Brivanlou,2001; Thisse et al.,2000). Wnt and TGFβ signals can operate synergistically through Lef1 to activate target genes(Nishita et al., 2000;Riese et al., 1997). In addition, the Wnt and MAPK pathways work synergistically to reduce repression of target genes mediated by Lef/Tcf homologs(Behrens, 2000;Meneghini et al., 1999;Rocheleau et al., 1999). Together these studies suggest a broad role for Lef/Tcf family members in coordinating the response to multiple signaling pathways.

Our results are consistent with the emerging view that a gradient of Wnt/β-catenin activity helps define discrete domains of gene expression in the neural plate. As mentioned above, we propose that Tcf3 represses targets of caudalizing factors, and β-catenin prevents Tcf3 from being effective as a repressor, resulting in a rostral-caudal gradient of effective Tcf3 repression (Fig. 7A,broken lines). This effect of the Tcf3 repression gradient could be represented by lowering the rostral end of a caudalizing gradient(Fig. 7B, broken purple line). Loss of basal Tcf3 function would thus raise the low end of the caudalizing gradient, decreasing its slope (Fig. 7C,D). In the context of this gradient, specific thresholds of caudalizing activity define discrete windows of gene expression along the rostral-caudal axis (Fig. 7B-D). Progressive loss of Tcf3 function would cause loss of rostral and expansion of caudal gene expression domains(Fig. 7C,D).

A one-dimensional representation of the caudalizing gradient helps illustrate the caudal to rostral shift in gene expression when Tcf3 function is reduced. However, it does not provide an adequate explanation for actual changes in the size and shape of gene expression domains in the neural plate,in particular why caudal genes such as gbx1 eventually expand rostrally around a pax2.1 expression domain.

The gbx1 gene is consistently expressed just caudal topax2.1, defining a compartment expected to depend on a slightly higher window of caudalizing activity. Expansion of the gbx1 domain around the pax2.1 domain in severely caudalized embryos suggests thatpax2.1 represents the low end of the caudalizing gradient in these embryos and that the surrounding gbx1 expression reflects a slightly higher caudalizing activity. Indeed, the arc-like rostral expansion ofgbx1 in caudalized embryos resembles the arc-like expression ofpax6 in wild-type embryos. These observations imply that the low end of the caudalizing gradient is not at the rostral edge of the neural plate,but rather in a slightly caudal domain that is surrounded by pax6 in wild-type embryos and surrounded by gbx1 in severely caudalized embryos.

We have shown here and in previous analysis of hdl mutants(Kim et al., 2000) that loss of Tcf3 function leads to changes in patterning that become evident by early gastrulation. Embryos treated with lithium chloride soon after the shield stage are caudalized in a manner similar to hdl mutants(van de Water et al., 2001;Kim et al., 2002). At shield stage the ventrolateral blastoderm margin in zebrafish is the source of caudalizing factors and corresponds with the highest β-catenin activity(Woo and Fraser, 1997;Dorsky et al., 2002). If caudalizing factors at the ventrolateral margin help establish the gradient of caudalizing activity, the low end of the gradient should be located at the furthest distance from the source, slightly dorsal to the animal pole(Fig. 7E). At the same time,BMP antagonists secreted at the dorsal margin define the prospective neurectoderm (Grinblat et al.,1998) (Fig. 7F). Different levels of caudalizing activity in the prospective neurectoderm are expected to define discrete domains of gene expression(Fig. 7G,H). The yellow, green and blue compartments defined by different thresholds of caudalizing activity illustrate how expression of pax6, pax2.1 and gbx1,respectively, might be determined in wild-type embryos. As described above,loss of Tcf3-mediated repression is expected to alter the shape of the caudalizing gradient (Fig. 7I,K) and thus alter pax6, pax2.1 and gbx1expression domains (Fig. 7J,L). This model provides a potential explanation for the arc-shaped early expression of pax6 in wild-type embryos and illustrates whygbx1 expression would expand rostrally around an oval pax2.1expression domain in severely caudalized embryos.

The shape of the caudalizing gradient can also be influenced by factors that inhibit function of Wnts. As gastrulation proceeds, the prechordal plate,which is the source of at least one secreted Wnt antagonist, Dkk1(Hashimoto et al., 2000;Shinya et al., 2000), might help define the low point of the Wnt activity gradient in the overlying rostral neurectoderm. Furthermore, during gastrulation the anterior neural ridge also becomes a source of a Wnt antagonist, Tlc(Houart et al., 2002), and it probably contributes to the pattern of Wnt-mediated derepression as gastrulation is completed. Clearly, other factors such as cell and tissue movements contribute to patterning of the neurectoderm throughout this process. However, for simplicity their contribution is not emphasized in our model, which represents a static view at the beginning of gastrulation.

Following injection of the tcf3b MO, we observed a marked defect in morphogenesis of the MHB and hindbrain rhombomeres. Although we were unable to rescue this phenotype by overexpressing either gene, we believe it is specific to tcf3b because we never observed hindbrain defects in other MO-injected embryos. Our data predict that Tcf3b might affect the expression of genes involved in hindbrain morphogenesis. One obvious target for further investigation would be the Ephrin/Eph family of receptor tyrosine kinases and ligands, which have been demonstrated to play a role in cell sorting and boundary formation in the hindbrain(Cooke et al., 2001;Lumsden, 1999).

The hdl gene plays a unique role in forebrain patterning during development. Likewise, injection of the tcf3b MO produced unique phenotypes in hindbrain and MHB morphogenesis. Because our rescue experiments indicate that hdl and tcf3b encode proteins that can function identically, some of these unique roles can be explained by nonoverlapping expression patterns of the two genes. This may be true in the hindbrain and MHB as well, where we observed subtle differences in the expression patterns of hdl and tcf3b(Fig. 2K,L). Alternatively, the two genes may encode proteins with different DNA targets or transcriptional cofactors in the hindbrain and MHB, and the function encoded by hdlmay be dispensable. Our inability to rescue the tcf3b MO phenotype with either gene leaves these possibilities open.

In some tissues in which either one or both genes are expressed, we observed no phenotype in our MO injections. For example, both hdl andtcf3b are expressed in the notochord, but no obvious notochord defects were seen in MO-injected embryos. The function of hdl in the tailbud and paraxial mesoderm is unclear as well, as neither MO-injected embryos nor hdl mutants exhibit patterning defects in these tissues. Loss of hdl and tcf3b function prior to gastrulation resulted in minimal effects on initial dorsal-ventral patterning. The most probable explanation for these results is that in zebrafish, other genes are able to compensate for hdl and tcf3b in these regions.

In this study, we have demonstrated specific and overlapping developmental roles for two zebrafish tcf3 genes. Our results suggest regions in the embryo where Tcf3 function may be important for patterning and morphogenesis. It will now be important to identify the transcriptional targets of Hdl and Tcf3b in these regions so the cellular responses to this pathway become clear. In addition, the biochemical differences between Lef/Tcf proteins must be further investigated, so that both their redundancies and distinct functions can be better understood.

R.I.D. was supported as an Associate of the HHMI while in the laboratory of R.T.M. R.T.M. is an Investigator of the HHMI. This work was also supported by JSPS research fellowships (M.I.). We thank Hans Meinhardt and Stephen Wilson for stimulating discussions.