Gli1 is a target of Sonic hedgehog that induces ventral neural tube development

Great progress has been made in recent years in the identification of signaling molecules that mediate embryonic inductions. However, much less is known of the molecular mechanisms that interpret the inducing information in responding cells, including the identification of transcriptional regulators that orchestrate the induced differentiation programs.

Within the nervous system, the differentiation of the floor plate and other ventral neural tube cell types depends on induction from the underlying mesodermal cells of the notochord (van Straaten et al., 1988; van Straaten and Hekking, 1991; Placzek et al., 1990, 1993; Yamada et al., 1991; Ruiz i Altaba, 1992; Goulding et al., 1993). Sonic hedgehog (Shh) secreted from the notochord is necessary and sufficient to induce ventral cell types (Riddle et al., 1993; Krauss et al., 1993; Echelard et al., 1993; Roelink et al., 1994; Chang et al., 1994; Ruiz i Altaba et al., 1995a; Ekker et al., 1995; Martí et al., 1995a; Chiang et al., 1996). Floor plate cells and ventral neurons can be induced in vitro by different concentrations of Shh (Roelink et al., 1994, 1995; Martí et al., 1995a; Tanabe et al., 1995; Hynes et al., 1995; Ericson et al., 1996) and ectopic expression of Shh within the neural tube leads to the induction of floor plate cells (Krauss et al., 1993; Echelard et al., 1993; Roelink et al., 1994; Ruiz i Altaba et al., 1995a). In addition, absence of the notochord or loss of function of Shh leads to severe embryonic deficiencies including the lack of ventral neural tube cell types (Yamada et al., 1991; Ruiz i Altaba, 1992; Placzek et al., 1993; Martí et al., 1995a; Chiang et al., 1996; Erickson et al., 1996; Roessler et al., 1996; Belloni et al., 1996).

An important step in floor plate development in response to Shh appears to be the expression of the winged-helix transcription factor HNF-3β (Ruiz i Altaba et al., 1993a, 1995a; Monaghan et al., 1993; Strähle et al., 1993; Ang et al., 1993; Sasaki and Hogan, 1993). Ectopic expression of HNF-3β, like that of its early functional homolog in frogs Pintallavis (Ruiz i Altaba and Jessell, 1992), leads to ectopic floor plate differentiation within the neural tube (Ruiz i Altaba et al., 1993b, 1995a; Sasaki and Hogan, 1994). In frog embryos, ectopic HNF-3β expression induces the expression of floor plate markers including F-spondin, Shh and HNF-3β itself, and ectopic expression of Shh induces the expression of F-spondin, HNF-3β and Shh itself (Ruiz i Altaba et al., 1993b, 1995a; Roelink et al., 1994).

Previous studies have indicated a conservation in the cytoplasmic transduction of the hh signal from flies to vertebrates (Fan et al., 1995; Goodrich et al., 1996; Marigo et al., 1996a-d; Hahn et al., 1996; Johnson et al., 1996; Stone et al., 1996; Hammerschmidt et al., 1996; Epstein et al., 1996; Concordet et al., 1996). Genes involved in interpreting the Shh signal could regulate HNF-3β function during floor plate induction. In Drosophila, Cubitus interruptus (ci) encodes a putative transcription factor with five zinc fingers that acts downstream of all other known components in the hh transduction pathway regulating the expression of hh target genes (Orenic et al., 1990; Eaton and Kornberg, 1990; Forbes et al., 1993; Slusarski et al., 1995; Perrimon, 1995; Sánchez-Herrero et al., 1996; Domínguez et al., 1996; Alexandre et al., 1996). Homologs of ci in other species include three vertebrate genes: Gli (hereafter referred as Gli1), Gli2 and Gli3 (Kinzler et al., 1987, 1988; Kinzler and Vogelstein, 1990; Ruppert et al., 1988, 1990; Walterhouse et al., 1993; Hui et al., 1994) and the nematode tra-1 gene involved in sex determination (Zarkower and Hodgkin, 1992).

Gli1 was originally identified as an amplified nuclear oncogene in human gliomas and sarcomas (Kinzler et al., 1987; Roberts et al., 1989) and the mouse Gli genes are expressed in many embryonic cell types in overlapping domains (Walter-house et al., 1993; Hui et al., 1994). Loss of function of Gli3 in humans leads to a series of defects described as the Greig cephalopolysyndactyly syndrome (Vortkamp et al., 1991) and Gli3 mutant mouse (extra-toes) embryos show defects in limb and forebrain development that mimic the human syndrome (Johnson, 1967; Schimmang et al., 1992; Hui and Joyner, 1993; Franz, 1994). However, it is not known whether the Gli genes participate in the induction of neural pattern by Shh.

In this paper we show that the Gli genes display distinct yet overlapping domains of expression during embryogenesis and have different functions. Gli1 is normally expressed in midline neural plate cells during gastrulation, it is transcriptionally activated by Shh signaling and Gli1, but not Gli3, can mimic the inductive effects of Shh on neural tube pattern. Together, these results strongly support a role of Gli1 as a target and mediator of Shh signaling in floor plate and ventral neuronal differentiation.

Xenopus laevis embryos were obtained by standard methods (e.g. Ruiz i Altaba, 1993) and staged according to Nieuwkoop and Faber (1967). Exogastrulae were obtained as previously described (Ruiz i Altaba, 1992). Microinjection of plasmid DNA was performed as described (Ruiz i Altaba, 1993) introducing 20-200 pg of supercoiled DNA. The injections were localized to the animal-most region to deliver the plasmids to prospective neuroectodermal cells (Ruiz i Altaba et al., 1995a). Lineage tracing with rhodamine-lysine-dextran (M_r>70 000; Molecular Probes) was performed by coinjecting RLDx in water at 25 mg/ml with plasmid DNA. COS-7 cells (ATCC) were cultured and transfected by standard procedures with lipofectamine (GIBCO) and examined 24-48 hours after transfection. CD1 mouse embryos were obtained at the desired embryonic (E) day by natural matings from the Skirball Institute transgenic facility or Taconic Animal Farms. E0.5 was counted as the morning after conception.

A partial mouse Gli1 cDNA (Hui et al., 1994) was used as a probe to screen a frog neurula stage (stage 17) cDNA library (Kintner and Melton, 1987) at moderate stringency. Several positive clones were identified and classified by Southern blot and sequence analyses. The longest cDNA clones from the stage 17 library screen were cloned into the EcoRI site of pBluescript (Stratagene) yielding pfGli1 (3.6 kb), pfGli2 (4 kb) and pfGli3 (4 kb).To isolate the 5′ sequences missing from the fGli1 clone, RACE reactions were performed with primer 5′AGAGATGGGCTGATAGTTCC3′ yielding 300 additional nucleotides. This product was then used to isolate a partial cDNA clone containing the entire 5′ end from the stage 17 cDNA library at a frequency of 3.5×10⁻⁶. The full-length clone was obtained by joining the original fGli1 cDNA to the 5′ partial cDNA at their NruI sites. Overlapping sequences showed that the two cDNAs derive from the same gene.

cDNAs for injection were subcloned either into pcDNA1-Amp (Invitrogen) yielding subclones named pCMV-followed by the name of the cDNA or into pCS2MT, a vector that contains 6 myc tags in frame (Turner and Weintraub, 1994; Rupp et al., 1994), kindly provided by D. Turner. Expression of the myc-tagged cDNA fragments is driven by CMV regulatory sequences as in the case of pcDNA 1-Amp.

The full-length frog Gli1 cDNA was myc-tagged by cloning a sequenced 480 bp Pwo PCR EcoRI-XhoI product of the 5′ end of the coding sequence into pCS2MT followed by the insertion of the rest of the cDNA as a BclI-XhoI fragment, yielding pmyc-fGli1.

A myc tag deletion of pmyc-fGli1 was obtained by digesting this plasmid with EcoRI and ClaI, blunt ending and religation. In this clone, the start codon is the original one of Gli1.

A truncated (5′ deletion) version of frog Gli1 missing the first 224 amino acids was myc-tagged by cloning the EcoRI partial cDNA (clone 16) into the EcoRI site of pCS2MT yielding pmyc-fGli1-5′Δ. The VP16->fGli1 clone was made by joining a sequenced PCR product of the VP16 transactivator domain containing an ATG (kindly provided by Dr Ed Ziff) using primers (5′AACGGATCCACCATG-GTCGCGTACAGCCGCGCGCGTACG3′ and 5′AACGGATCCCC-CACCGTACTCGTCAATTCC3′) to the BamHI site in the polylinker of pcDNA1-Amp upstream of the fGli1 partial cDNA cloned into the EcoRI site. The ribosome binding site and initiator ATG in primer 1 are underlined. The nucleotides in between the BamHI and the EcoRI sites in the polylinker of the vector added the aa sequence GST-SNGRQCAG upstream of the start of the fGli1 sequence.

The human Gli1 (Kinzler et al., 1987) and Gli3 (Ruppert et al., 1990) cDNAs (kindly provided by Dr Bert Vogelstein) yielded pCMV-hGli1 and pCMV-hGli3.

Human Gli3 was myc tagged by cloning the coding region of this cDNA from aa 18 to the end into the pCS2-MT vector yiedling pmyc-hGli3.

Cubitus interruptus cDNAs (Orenic et al., 1990) were kindly provided by Dr Robert Holmgren and Dr Tom Kornberg. The intronless cDNA clone (kindly provided by Dr Kornberg) yielded pCMV-Ci.

Whole-mount in situ hybridization with digoxigenin-labeled antisense RNA probes was carried out in frog embryos according to Harland (1991) using maleic acid buffer and in mouse embryos according to Conlon and Herrmann (1993). Labeled embryos were sectioned in a microtome at 10-14 μm after embedding in paraplast/tissue prep (Fisher). Labeled mouse embryos were embedded in tissue tek and frozen after equilibrating in 30% sucrose in 0.1 M phosphate buffer. 50 μm sections were made in a cryostat and collected in glass slides (Superfrost; Fisher), briefly rinsed in PBS (∼0.5-1 minute) and mounted in Aquamount (Lerner Labs.). In situ hybridization in tissue sections with radioactively labeled RNA probes was performed as described (Hui and Joyner, 1993). Photographs were taken with a Zeiss Axiophot microscope and images transferred to Photoshop.

Antisense RNA probes were made as follows: pCMV-fGli1 (clone #16) was cut with BamHI and transcribed with SP6 RNA polymerase. pCMV-fGli2 (clone #21) was cut with BamHI and transcribed with SP6 RNA polymerase. pBluescript-fGli3 (clone #10) was cut with NotI and transcribed with T3 RNA polymerase. Frog

Shh (pfhh4 clone; Ruiz i Altaba et al., 1995a) was cut with NotI and transcribed with T3 RNA polymerase. A 1.4 kb PCR subclone of the frog F-spondin cDNA (Ruiz i Altaba et al., 1993a) in pCRII (Invitrogen) was cut and transcribed with SP6 polymerase. Frog neural-specific β-tubulin (N-tubulin) clone (Richter et al., 1988) was cut with BamHI and transcribed with T3 RNA polymerase. Pintalla vis pF5 clone (Ruiz i Altaba and Jessell, 1992) was cut with HindIII and transcribed with T7 RNA polymerase. Mouse Gli1 clone (Hui et al., 1994) was cut with NotI and transcribed with T3 RNA polymerase. Mouse Gli2 clone (Hui et al., 1994) was cut with HindIII and transcribed with T3 RNA polymerase. Mouse Gli3 clone (Hui et al., 1994) was cut with HindIII and transcribed with T7 RNA polymerase. Mouse Shh clone (Echelard et al., 1993) was cut with HindIII and transcribed with T3 RNA polymerase. A mouse HNF-3β clone (Sasaki and Hogan, 1993) was cut with HindIII and transcribed with T7 RNA polymerase.

Primary antibodies were used in PBS with 0.1% Triton and 10% heat-inactivated goat serum. Antibodies against human Gli1 were raised in rabbits immunized with a bacterially derived fusion protein containing the zinc finger DNA-binding domain (aa 255-408). Specific antibodies were purified by affinity chromatography through an affigel (Bio Rad)-Gli fusion protein column and used at 1/100. This antisera does not recognize frog Gli proteins. Rabbit polyclonal antibodies against frog HNF-3β protein (Ruiz i Altaba et al., 1995a) were used at 1/8000. Affinity-purified rabbit polyclonal antibodies against serotonin (purchased from Dr Hadassa, Columbia University) were used at 1/1500. Rabbit polyclonal antibodies against Nkx2.1 (TTF-1, Lazzaro et al., 1991; kindly provided by Dr Di Lauro) were used at 1/5000. Mouse 9E10 anti-myc monoclonal antibody was purchased from Santa Cruz, Inc. and used at 1/500. Monoclonal antibodies against the SC35 protein (Fu and Maniatis, 1990; kindly provided by T. Maniatis) were used as hybridoma supernatant at 1/5.

Embryos were fixed in MEMFA (Patel et al., 1989) for 20 minutes at room temperature. Preincubations, secondary antibodies, washes and peroxidase reactions were as described (Ruiz i Altaba et al., 1995a). Embryos were then dehydrated, cleared, sectioned and photographed as those used for whole-mount in situ hybridization. COS-7 cells were fixed in MEMFA for 5 minutes. Fluorochrome-coupled secondary antibodies (Boehringer Mannheim; Tago) were used at 1/200. Fluorescent images were photographed with an Axiophot microscope or collected with a Princeton Instruments cooled CCD (KAF 1400 chip) camera.

We have isolated cDNAs for the three Gli genes from frog embryos and analysed their embryonic expression patterns as compared to those of Shh, Pintalla vis/HNF-3β and N-tubulin. Among the Gli genes, homology in the zinc finger region was very high (between 82% and 96% identity) although the first zinc finger showed less conservation, as only fingers 3-5 bind DNA (Pavletich and Pabo, 1993), and was used to classify the frog cDNAs (Fig. 1A) together with sequences outside of the DNA-binding domain (not shown). The deduced frog and human Gli1 proteins showed 47% identity in the N-terminal region, 88% identity in the zinc finger region and 24% identity in the C-terminal region (Fig. 1B). The frog Gli1 and human Gli3 proteins showed 22% identity in the entire N-terminal region, 93% identity in the zinc finger region and 24% identity in the C-terminal region. In contrast, partial N-terminal sequences of fGli3 (not shown) and hGli3 showed 73% identity whereas their zinc finger regions showed 92% identity. Similarly, partial N-terminal sequences of fGli2 (not shown) and mouse Gli2 showed 71% identity and their zinc finger DNA-binding domains were 94% identical. Together, this shows that the identified frog Gli1 gene is the homolog of the human and mouse Gli1 genes.

Expression of Gli1 and Gli3 was first detected at low levels in the animal cap region of early gastrula-stage (stage ∼10.5) embryos, a time when Shh expression was undetectable and that of Pintallavis was evident in the organizer region (Fig. 2A-E). By stage ∼11.5, Gli1 RNA was detected in deep midline neural plate cells overlying the notochord (Fig. 2G,Q) in a similar manner to that of Pintallavis and Shh at later stages (Fig. 2P; Ruiz i Altaba and Jessell, 1992; Ruiz i Altaba et al., 1995a). At stage 11.5 Pintalla vis and Shh RNAs were expressed in the underlying notochord (Fig. 2F,H). In contrast, Gli2 expression was detected widely in the rest of the neural plate with higher levels in the future hindbrain region (Fig. 2I). Similarly, Gli3 expression was absent from the midline and highest in the anterior and lateral edges of the neural plate (Fig. 2J). Gli3 was also expressed in the posterior-ventral mesoderm (Fig. 2J and not shown).

By midgastrula stages (stage ∼12.5), expression of Gli2 and Gli3 was similar but expression of Gli1 was no longer detected in midline cells but rather in cells immediately adjacent to the midline (Fig. 2V). The position of Gli1⁺ cells adjacent to the midline appeared to be transiently coincident with the earliest differentiating N-tubulin⁺ primary medial neurons (Fig. 2W,X;Chitnis et al., 1995). The lateral-to-midline expression of Gli1 and its previous midline expression were dependent on the persistent presence of an underlying notochord as Gli1 expression was not detected in the neural ectoderm of complete exogastrula (from stages ∼11.5-13; Fig. 2U).

At midneurula stages (stage ∼14), Shh and Pintalla vis were expressed in midline neural plate cells, which will form the floor plate of the neural tube (Fig. 2K,M). Gli gene expression was maintained (Fig. 2L,N,O) with that of Gli1 highest in cells lateral to the midline (Fig. 2L,R) and that of Gli2 homogeneously in the neural plate outside of the midline (Fig. 2S), although the posterior neural plate showed lower levels of Gli2 expression (Fig. 2N). Gli1⁺ cells flanking the midline appeared to overlap with the medial boundary of Gli2 expression (Fig. 2L,N). Gli3 showed highest expression in the anterior neural ridge region (Fig. 2O). More posteriorly, there was a graded expression of Gli3 from the lateral edges to the midline (Fig. 2O,T). Expression of the Gli genes in somitic mesoderm occurred in a manner similar to that in the neural plate: Gli1 closest to the notochord, Gli2 throughout and Gli3 in a graded manner with highest levels distal from the midline (Fig. 2R-T and not shown).

In tadpoles (stages ∼32-36), the Gli genes showed partly overlapping patterns of expression (Fig. 2Y-Za). In the brain, expression of Gli1 was detected prominently in a diencephalic stripe similar to that of Shh (Figs 2Y, 6C). Gli2 was prominently expressed in the telencephalon, dorsal and ventral anterior diencephalon and in cells close to the dorsal midbrain/hindbrain junction and dorsal hindbrain (Fig. 2Z). High Gli3 expression was found throughout the dorsal aspect of the neural tube especially at the diencephalic/midbrain and midbrain/hindbrain junctions (Fig. 2Za).

Along the D-V axis of the posterior neural tube, expression of Gli1 was confined to the ventricular zone and there was no expression in the floor plate (Fig. 2Zd), where HNF-3β and Shh are detected (Fig. 2Zb, Zc). Expression of Gli2 and Gli3 was found in the dorsal ventricular zone (Fig. 2Ze, Zf) with low levels in the roof plate. Gli gene expression was reciprocal to that of N-tubulin in differentiated neurons distal from the ventricle (Fig. 2Zg).

Expression of three Gli genes has been previously examined in mouse embryos (Walterhouse et al., 1993; Hui et al., 1994). However, these studies did not resolve their early neural expression.

In the posterior neural tube of E8.5-8.75 mouse embryos, the three Gli genes were found to be expressed throughout the neural plate (Fig. 3A-C) except that Gli3 expression was very low or absent at the ventral midline. At this time, posterior midline neural plate cells are beginning to express HNF-3β (Fig. 3D) and Shh expression is still confined to the notochord (Fig. 3E). In more anterior (and thus older) regions of the neural plate, the expression of all three Gli genes was maintained throughout the neural plate with the exception of the floor plate (Fig. 3F-H).

In anterior regions of the closed neural tube of E9.5 embryos, Gli gene expression was further restricted (Fig. 3I-K). Gli1 transcripts were confined to ventricular cells adjacent to the floor plate (Fig. 3I) whereas Gli2 and Gli3 mRNAs were detected in the dorsal ventricular zone and immediately adjacent cells (Fig. 3J,K). At this time, the expression of both HNF-3β and Shh is detected in the floor plate and Shh transcripts are also present in the notochord (Fig. 3L,M). In the spinal cord of E11.5 embryos, Gli1 expression was confined to two patches in the ventral ventricular zone (Fig. 3N) and that of Shh to the floor plate and at much lower levels to the motor neuron pools (Fig. 3O). Thus, in both frog and mouse embryos, Gli gene expression is preferentially detected in precursor cells that are first found throughout the neural plate and later in the ventricular zone. Moreover, expression in midline cells ceases when these cells begin to express floor plate markers such as HNF-3β. However, all three Gli genes are coexpressed in the early mouse neural plate with clear coexpression of Gli1 and Gli2 at the ventral midline whereas only Gli1 is expressed in the prospective floor plate in frog embryos.

Given the distinct expression of the Gli1 gene in midline neural plate cells at the time of floor plate induction in frog embryos, we tested whether gain of function of Gli1 could mimic gain of function of Shh. We assayed for ventral neural tube differentiation by injecting plasmids driving the expression of cDNAs into the animal pole of one cell in 2-cell frog embryos, which results in the unilateral mosaic expression of injected genes in the ectoderm including the neural tube. Injected embryos at the tadpole stage were then labeled with specific antibodies to HNF-3β, a ventral marker (Fig. 6A) that is ectopically expressed after Shh injection (Ruiz i Altaba et al., 1995a).

Embryos injected with plasmids driving the expression of the full-length frog Gli1 open reading frame displayed ectopic HNF-3β expression within the neural tube as well as in the epidermal ectoderm (Table 1; Figs 4, 6D and not shown). To test for protein production, this clone contained myc tags fused in frame to the N terminus of Gli1. Myc-tagged Gli1 protein (myc-fGli1) was localized primarily in the cytoplasm of transfected COS cells with low nuclear levels although injected frog blastomeres showed both cytoplasmic and nuclear labeling (Fig. 5A,B).

To test whether the myc tags were altering the activity of the Gli1 protein, a deletion clone was made leaving only the intact Gli1 open reading frame. Tadpoles injected with this construct also showed ectopic expression of HNF-3β (Table 1; Fig. 4) demonstrating the ability of the normal endogenous Gli1 protein to induce HNF-3β expression. As controls, injection of a scrambled fGli1 cDNA (fGli1-inv) and uninjected controls did not show ectopic HNF-3β (Fig. 6A; Table 1). Because cells at the ventral midline of the neural tube do not inherit plasmids injected into the animal-most region of 2-cell embryos (Dale and Slack, 1987; Ruiz i Altaba et al., 1995a), aberrant migration or proliferation of endogenous floor plate cells cannot underlie the appearance of HNF-3β⁺ cells at ectopic locations.

The analysis of the subcellular localization of injected and transfected Gli1 protein showed it was localized to both the nucleus and cytoplasm (Fig. 5A,B). We tested for the requirement of cytoplasmic Gli1 for HNF-3β-inducing activity by targeting Gli1 to the nucleus. To this end, we created a myc-tagged fGli1 protein carrying an SV40 nuclear localization signal in its N terminus. Injection into frog embryos and transfection of this construct into COS cells showed that the NLS-myc-fGli1 protein was nuclear (Fig. 5C,D). Injected tadpoles showed a very high incidence of ectopic HNF-3β expression (Table 1; Fig. 4). The ability of nuclear Gli1 to mimic the effects of Gli1 suggests that it functions in the nucleus even though stable protein is found predominantly in the cytoplasm.

To test for a possible role of Gli1 as a transcriptional regulator, we first identified sequences outside the DNA-binding domain responsible for function. An N-terminal truncated myc-tagged protein (myc-fGli1N′Δ) was stable (not shown) but was unable to induce ectopic HNF-3β expression (Table 1; Fig. 4). This deletion was then used to make a chimera in which the strong transactivating domain of VP16 (Campbell et al., 1984; Triezenberg et al., 1988) was fused in frame to fGli1N′Δ. Injection of this chimeric protein (VP16-fGli1) also resulted in the ectopic expression of HNF-3β (Table 1; Fig. 4). The ability of VP16-fGli1 to mimic myc-fGli1 strongly suggests that Gli1 normally acts as a positive transcriptional transactivator.

To extend our findings with fGli1, we tested the activities of other members of the Gli gene family in our in vivo assay. Injection of the human Gli1, but not Gli3, myc-Gli3 or ci, cDNAs led to the ectopic expression of HNF-3β (Figs 4, 6B,E; Table 1). Ectopic HNF-3β expression in hGli1-injected embryos was detected in the neural tube as well as in the non-neural ectoderm (Fig. 6H,I).

Because the human cDNAs were obtained from glioma cells (Kinzler et al., 1987), the different activities of the Gli1 and Gli3 cDNAs could be due to differential expression from the injected plasmids as amplification has been proposed to be the mechanism of action of Gli1 in oncogenesis (Kinzler et al., 1987). Indeed, the induction of ectopic HNF-3β expression by Gli1 proteins was concentration dependent (Table 1 and not shown). However, quantitative RT-PCR revealed a similar expression on average of all tested genes from injected plasmids (not shown), indicating that their differential action is not due to differential expression.

To ascertain that the inactivity of Gli3 was not due to absence of protein, we analysed the expression of hGli1 and hGli3 proteins from the injected cDNAs. The human Gli1 and Gli3 cDNAs (Kinzler et al., 1987; Ruppert et al., 1990) directed the expression of stable protein as determined by labeling with specific antibodies against hGli1 and the production of myc-tagged Gli3 protein (Fig. 5 and not shown). hGli1 accumulated preferentially in the nucleus (Fig. 5E,F), like HNF-3β (Fig. 5I), although some cells showed also cytoplasmic labeling (Fig. 5F,K). Myc-tagged hGli3 protein was primarily cytoplasmic with a low level of nuclear labeling (Fig. 5G,H,K). The localization of cytoplasmic Gli proteins in COS cells was partly coincident with that of tubulin and suggestive of an association with the cytoskeleton (Fig. 5L and not shown). Within the nucleus, hGli1 was often detected in subdomains that were always distinct from those containing the splicing factor SC35 (Fig. 5J; Fu and Maniatis, 1990), unlike other zinc finger proteins having dual transcriptional and splicing functions (Larsson et al., 1995).

These results together with the identical abilities of the endogenous frog and human glioma Gli1 proteins in terms of HNF-3β induction indicate that Gli1 proteins are normally active and functionally equivalent. In the following assays, we have therefore used different Gli1 proteins interchangeably and hGli3 as a negative control.

Ectopic expression of HNF-3β in embryos injected with Gli1 proteins suggested the ectopic differentiation of floor plate cells. To further test this, we assayed injected embryos for the expression of Shh mRNA as a second floor plate marker of the posterior neural tube (Fig. 6C). Widespread expression of Gli1 proteins, but not hGli3, resulted in the ectopic transcription of Shh within the tadpole neural tube (Fig. 6F,G and not shown; Table 2). Similarly, the floor-plate-specific marker F-spondin (Fig. 6J, Ruiz i Altaba et al., 1993b) was induced within the tadpole neural tube by injection of Gli1 but not Gli3 proteins (Fig. 6K,L and not shown; Table 2). These results demonstrate that widespread expression of Gli1 leads to the ectopic differentiation of floor plate cells.

To determine the requirement of the presence of an endogenous floor plate for the ectopic differentiation of floor plate cells by Gli1, we injected hGli1 into embryos that were then induced to develop as complete exogastrulae, which lack an endogenous floor plate (Ruiz i Altaba, 1992, 1994). Injection of hGli1 led to ectopic HNF-3β expression in patches within the neural ectoderm of complete exogastrulae, in a position distal from the junction with mesoderm (20%, n=25; not shown), indicating that floor plate differentiation by Gli1 is independent of the presence of an endogenous floor plate.

Floor plate cells acquire the floor-plate-inducing properties of the notochord (Placzek et al., 1993). We therefore tested for the ability of newly induced floor plate cells in hGli1-injected embryos to induce, in turn, floor plate development in adjacent cells. To distinguish between ectopic floor plate cells derived from injected blastomeres and those that may be induced by homeogenetic induction, we coinjected a lineage tracer (rhodamine-lysine-dextran, RLDx) with hGli1 plasmids. Because RLDx diffuses more rapidly than plasmid DNA (as assessed by the distribution of RLDx and plasmid-derived myc-tagged proteins in coinjection assays; not shown), all cells that inherit DNA are labeled by RLDx but not the opposite. For this analysis, we screened histological sections of injected embryos for regions displaying ectopic HNF-3β expression and a mosaic pattern of RLDx distribution. In these areas, there was a similar incidence of RLDx-labeled and unlabeled HNF-3β⁺ cells (Fig. 6M,N). Non-RLDx-labeled HNF-3β⁺ cells were always in close proximity to RLDx-labeled HNF-3β⁺ cells, indicating that newly induced floor plate cells are functional and can themselves induce adjacent cells towards floor plate differentiation.

Ectopic expression of HNF-3β, Shh and F-spondin were detected throughout the A-P axis with similar frequencies (Fig. 6) taking into consideration the bias towards anterior expression given the site of injection into the animal pole according to the fate map (Dale and Slack, 1987). Expression of Gli1, like Shh (Ruiz i Altaba et al., 1995a), leads to the differentiation of floor plate cells in the forebrain, a region that lacks a normal floor plate suggesting that neural A-P fates are not irreversibly fixed. This alteration of A-P values was striking as D-V identities are normally constrained by the A-P position of precursor cells (Jacobson, 1964; Yamada et al., 1991; Ruiz i Altaba, 1994; Simon et al., 1995, Ericson et al., 1995).

Along the D-V axis, the distribution of ectopic HNF-3β expression sites showed also that there are no restrictions to floor plate differentiation driven by Gli1 proteins (Fig. 6M-P). However, the roof plate showed a high incidence of ectopic sites (Fig. 6N) and similar results were obtained using Shh mRNA as a marker in the hindbrain (not shown). These results show that all cells can acquire a floor plate phenotype but that roof plate cells are the most competent to do so in response to Gli1 function.

If Gli1 can mimic the actions of Shh, injected embryos should display the ectopic differentiation of ventral neuronal cell types. In the ventral midbrain, HNF-3β is expressed in the floor plate and in cells located away from the ventricular zone showing a rounded nucleus which suggests that they are differentiated neurons (asterisks in Fig. 6P). Embryos injected with hGli1 displayed ectopic cells in the midbrain resembling such putative neurons.

To directly test whether widespread expression of Gli1 could induce the ectopic differentiation of a defined cell type, we assayed for the expression of the neurotransmitter serotonin (5HT) in injected embryos as a marker of raphespinal neurons. In tadpole (stage ∼36) embryos, 5HT⁺ cells are normally found clustered in the ventral region of rhombomere 1 adjacent to the floor plate (Fig. 7A; van Mier et al., 1986; Ruiz i Altaba and Jessell, 1991). Embryos injected with hGli1 displayed the ectopic differentiation of 5HT⁺ neurons (Fig. 7B-E; Table 2). These neurons were frequently located near the dorsal midline and displayed axons with aberrant pathways. Ectopic differentiation of 5HT⁺ neurons outside of the anterior hindbrain was observed in the dorsal diencephalon in only two embryos (not shown). Ectopic 5HT⁺ neurons were never observed in Gli3-injected or uninjected control embryos (Table 2). It is not clear, however, whether the ectopic differentiation of these neurons is induced directly by injected Gli1 or secondarily via an induced ectopic floor plate.

Shh can induce the differentiation in vitro of ventral neurons from different anteroposterior levels of the neural tube (Ericson et al., 1995). We therefore tested for the ability of Gli1 to induce the differentiation of ventral cell types in the forebrain, a region that lacks a floor plate. We chose to examine the expression of Nkx2 homeoproteins expressed by ventral telencephalic and diencephalic neurons (Fig. 7F; Lazzaro et al., 1991; Price et al., 1992; Ericson et al., 1995), the expression of which is responsive to Shh (Barth and Wilson, 1995; Ericson et al., 1995). Ectopic Nkx2.1 protein was detected in dorsal telencephalic regions of embryos injected with hGli1 or fGli1 proteins (Fig. 7G,H; Table 2), demonstrating the ability of Gli1 to induce ventral forebrain development in vivo.

Floor plate development begins during gastrulation as the notochord underlies and induces midline neural plate cells, but widespread expression of Shh or HNF-3β cannot induce ectopic midline (floor plate) development within the neural plate at these early stages (Ruiz i Altaba et al., 1995a). To investigate whether similar restrictions affect the action of Gli1, we analysed midline marker expression in injected neurulae (stage ∼14-16). In these embryos, there was no ectopic expression of HNF-3β in the neural plate of hGli1- or hGli3-injected embryos although a fraction of embryos injected with hGli1 displayed expression in the non-neural ectoderm (10%; n=20).

Injected embryos were then analyzed for the expression of two midline neural plate markers: Pintallavis and Shh (Ruiz i Altaba and Jessell, 1992; Dirksen and Jamrich, 1992; Knöchel et al, 1992; Ruiz i Altaba et al., 1995a; Ekker et al., 1995). hGli1 and VP16-fGli1 proteins, but not hGli3, were able to induce the ectopic expression of these markers at neural plate stages (Table 3). However, their expression occurred in scattered cells mostly but not exclusively in non-neural, epidermal ectoderm (Fig. 8A,B). Similar results were obtained in Shh-injected embryos (Table 3; Ruiz i Altaba et al., 1995a). Restrictions that operate to prevent ectopic floor plate differentiation by Shh and Pintallavis and at neural plate stages also appear to affect Gli1 function. Nevertheless, epidermal ectoderm can respond to express midline genes.

We also tested for the expression of the endogenous Gli1 after injection of the human gene. Injected embryos did not display ectopic Gli1 (Table 3) suggesting that Gli1 does not positively autoregulate at these early stages.

Because the expression of Gli1 is detected in midline cells, we tested the possibility that Gli1 is a target of Shh induction. Embryos injected with Shh displayed the ectopic expression of Gli1 (Fig. 8C; Table 3). As expected, cells expressing Gli1 ectopically were found mostly in epidermal ectoderm. Gli1 induction by injected Shh was detected both at stages when Gli1 is restricted to the midline proper and later on when its expression is found in cells adjacent to the midline (Fig. 8C and not shown).

In this study, we have investigated whether the zinc finger genes of the Gli family function in Shh signaling. The coincident or adjacent patterns of expression of Gli1 and Shh in different tissues, including the neural ectoderm and somitic mesoderm, suggest an involvement of Gli1 in the interpretation of the Shh signal. These results therefore indicate a conservation in the response of ci/Gli1 to hh/Shh signals in insects and vertebrates. Within the neural plate, Gli1 is expressed in midline cells becoming the floor plate when Shh is expressed by the underlying notochord. At later stages, Gli1 is expressed in cells becoming ventral neurons adjacent to the midline when Shh is expressed by the floor plate at the midline. This, together with the induction of Gli1 by injected Shh suggests that it normally functions downstream of secreted Shh although some late aspects of Gli1 expression could be independent of Shh signaling.

Our results show that widespread expression of the endogenous frog or the human glioma Gli1 proteins induces ectopic ventral pattern in the neural tube in vivo, mimicking our previous results with widespread expression of Shh (Ruiz i Altaba et al., 1995a). The activity of nuclear Gli1 together with the ability of the VP16 transactivating domain to restore function to an N-terminal truncated fGli protein strongly suggest that, like Drosophila ci (Alexandre et al., 1996), Gli1 normally acts as a positive transactivator. In the posterior neural tube, Gli1 may therefore function to mediate the induction of floor plate cells and immediately adjacent ventral cell types acting both as a mediator and a target of Shh signaling (Fig. 9). The ectopic expression of Nkx2 proteins in the dorsal telencephalon of Gli1-injected embryos together with the normal expression of Gli1 in the forebrain in a manner similar to that of Shh also implicate Gli function in ventral forebrain patterning.

Floor plate development in frogs begins as the nascent notochord induces the overlying neuroectoderm to express floor plate markers such as Pintallavis and later Shh (Ruiz i Altaba, 1992; Ruiz i Altaba and Jessell, 1992; Dirksen and Jamrich, 1992; Ruiz i Altaba et al., 1993b; 1995a; Ekker et al., 1995). Our data suggest a pathway in which an early consequence of secreted Shh from the notochord is the transcription of Gli1 and the downstream activation of floor plate and medial (ventral) neuronal determination genes (Fig. 9).

Transcription of HNF-3β in neural cells is induced by notochord signals in the absence of protein synthesis (Ruiz i Altaba et al., 1995b). This must be mediated by proteins already present in the neural ectoderm before Gli1 is transcribed in response to Shh signaling. One possibility to account for this effect is that Shh signaling directly activates early HNF-3β transcription. However, the normal expression pattern of Gli1 predicts that it would be already expressed in explanted neural plate pieces similar to those induced in the presence of cycloheximide (Ruiz i Altaba et al., 1995b). Because Gli1 transcripts are first detected throughout the prospective neural plate, the dorsal animal cap, preexisting Gli1 protein may be activated in midline cells in response to initial Shh induction. This could be a general mechanism; in mice the three Gli genes are coexpressed throughout the early neural plate where initial Gli gene transcription would also appear to be independent of Shh signaling.

Gli1 does not appear to positively autoregulate and is only transiently expressed in deep midline cells that later will express Pintallavis and Shh and are fated to give rise to the floor plate. It is possible that high levels of HNF-3β/Pintallavis or an early floor plate protein repress Gli1 gene transcription. Autorepression is unlikely as Gli transcription is maintained in other cell types in which it may be independent of Shh. Signaling by Shh in floor plate cells and in adjacent ventral neuronal precursors may therefore lead to a regulatory feedback loop in which the activation of Gli1 is followed by its repression (Fig. 9).

At neural plate stages, embryos injected with Gli1, like those injected with Shh or HNF-3β (Ruiz i Altaba et al., 1995a), failed to show ectopic midline marker expression within the neural plate. Thus, in addition to loss of competence to become floor plate by neural cells (Placzek et al., 1993), there must be a mechanism that normally prevents floor plate differentiation throughout the neural plate (Ruiz i Altaba et al., 1995a). This mechanism may play an important role in neural patterning as the superimposition of the endogenous pattern of expression of Shh, bhh and chh shows that the entire neural plate expresses hh factors (Ekker et al., 1995). In contrast, at neural tube stages, Gli1, unlike HNF-3β? Pintallavis and Shh, is sufficient to induce floor plate differentiation throughout the neural tube in vivo (Ruiz i Altaba and Jessell, 1992; Ruiz i Altaba et al., 1993a, 1995a). This, together with their normal patterns of expression, suggest that Pintallavis/ HNF-3β act downstream of Gli1 and that their induction of floor plate cells may require a cofactor. Restrictions to Shh could reflect spatial restrictions on Gli1 expression or function.

The pattern of expression of Gli2 in cells close to the neural plate midline suggests that, like Gli1, it could be involved in mediating some of the inductive effects of Shh signaling on non-ventral midline cells. However, later in development, Gli2 would appear to have a Shh-independent function as it is coexpressed with Gli3 in the dorsal ventricular zone, a region that is unaffected by Shh from the notochord or floor plate.

The inability of hGli3 to induce ventral neural tube development is likely to reflect a similar function of fGli3 and is consistent with its predominant expression in dorsal regions. Gli3 expression is first detected in the animal cap and it is rapidly repressed from midline neural plate cells, possibly representing an early step in the progression of medial cells towards ventral development. Because expression of Gli3 is coincident or adjacent to sites of BMP4 expression, such as the animal cap, posterior/ventral mesoderm and the dorsal neural tube (Nishimatsu et al., 1992; Dale et al., 1992; Fainsod et al., 1994; Hemmati-Brivanlou and Thomsen, 1995; Liem et al., 1995), it is possible that the dorsal-to-ventral graded expression of Gli3 in the neural plate is related to BMP signaling from epidermal ectoderm. If so, Gli3 could act negatively on Shh signaling.

Gli3 could have a similar function in the developing limb bud and somites, acting in an opposite manner to Gli1. In the somites, the pattern of expression of the three Gli genes mimics that in the neural plate. Gli1 is expressed close to the notochord, which secretes Shh, Gli3 is expressed close to the lateral plate, which secretes BMP4, and these two signals have opposite effects on somite patterning (Fan and Tessier-Lavigne, 1994; Fan et al., 1995; Pourquié et al., 1995, 1996). In the limb bud, Gli1 is expressed close to the zone of polarizing activity (ZPA), Gli3 is expressed more generally but it is absent from the ZPA (Marigo et al., 1996d) and loss of function of Gli3 leads to the ectopic expression of Shh anteriorly (Masuya et al., 1995).

Within the vertebrate Gli family, there appears to be a diversification of function from that of ci in flies, as Gli1 and Gli3 have distinct patterns of expression and functional properties. Moreover, ci is unable to induce ectopic floor plate differentiation suggesting that, whereas both Gli1 and ci may mediate Shh/Hh signaling, the exact molecular mechanisms involved in these processes are not identical.

Gli1 was originally found as an amplified gene in a human glioma cell line (Kinzler et al., 1987) and it can transform fibroblasts in cooperation with E1A (Ruppert et al., 1991). However, it is not known if hGli1 is sufficient to cause tumor formation in vivo. Patched is a Shh receptor (Stone et al., 1996; Marigo et al., 1996c; Chen and Struhl, 1996) that is mutated in patients with the basal cell nevus syndrome, which predisposes them to develop basal cell carcinomas (Hahn et al., 1996; Johnson et al., 1996). Because patched may normally act negatively on Shh signaling (e.g. Forbes et al., 1993), its tumor suppressor activity raises the possibility that deregulated Shh signaling leads to skin cancer. Consistent with this, we note that the Shh signaling pathway can be activated in epidermal ectoderm by ectopic expression of Shh (Ruiz i Altaba et al., 1995a) or Gli1 (this work) as determined by the expression of HNF-3β, a Gli1-target gene.

We are grateful to Gord Fishell, Ruth Lehmann, Ed Ziff, Nadia Dahmane, Jessica Treisman, Rachel Brewster, Will Talbot, Alex Joyner and Alex Schier for discussion and comments on the manuscript. We thank E. Ziff, R. Di Lauro, R. Holmgren, A. Joyner, T. Kornberg, T. Maniatis and B. Vogelstein for reagents and G. Fishell for help with computer imaging. K. P. was a recipient of a National Research Service Award with A. Joyner. This work was supported by Skirball Institute start up funds to A. R. A.