Regulation of the neural patterning activity of sonic hedgehog by secreted BMP inhibitors expressed by notochord and somites

In vertebrate embryos the specification of neural cell types is initiated by the actions of inductive factors derived from mesodermal and ectodermal tissues that flank the neural plate and neural tube (Tanabe and Jessell, 1996; Lumsden and Krumlauf, 1996). At caudal levels of the neural tube that give rise to the spinal cord and hindbrain, the patterning of ventral cell types depends on the actions of the secreted signaling factor Sonic hedgehog (Shh; Marti et al., 1995; Chiang et al., 1996; Ericson et al., 1996; Hammerschmidt et al., 1997). The long-range diffusion of Shh from the notochord and floor plate has been proposed to establish a ventral-to-dorsal gradient of Shh activity within the ventral neural tube that directs subsequent patterns of neurogenesis (Ericson et al., 1997b; Briscoe et al., 2000).

The possibility that graded Shh signaling activity controls ventral neural fates is supported by several lines of evidence. First, Shh activity can be detected at a distance from ventral sources of Shh synthesis (Ericson et al., 1996), suggesting that the active autoproteolytic amino-terminal fragment of Shh, Shh-N (Porter et al., 1995), is transferred over many cell diameters within the ventral neural epithelium. Second, recombinant Shh-N acts at several threshold concentrations in vitro to induce distinct subsets of ventral neurons (Roelink et al., 1995; Ericson et al., 1997a; Briscoe et al., 2000). Third, the Ptc gene is expressed in a ventral-to-dorsal gradient within the ventral neural tube (Goodrich et al., 1996; Marigo and Tabin, 1998; Ericson et al., 1997b). Ptc encodes a ligand binding subunit of the Shh receptor whose expression appears to be activated as a direct function of the level of hedgehog signaling (Marigo et al., 1996; Ingham, 1998; Murone et al., 1999), and thus its graded expression is indicative of a gradient of Shh activity. Finally, ectopic expression of an activated form of Smo, a gene encoding the signaling transducing subunit of the Shh receptor (Stone et al., 1996; Quirk et al., 1997), has been reported to induce ventral cell types in a cell-autonomous manner (Hynes et al., 2000), supporting the idea (Ericson et al., 1996) that Shh acts directly on target cells to specify ventral cell fates.

The gradient of Shh signaling activity evident in the ventral neural tube could reflect the graded distribution of Shh protein itself, or be achieved through a more complex process that involves the action of additional cell surface or secreted proteins that modify neural cell responses to Shh. One class of secreted proteins that has been shown to exert a critical role in the patterning of the neural tube is the bone morphogenetic protein (BMP) family (Lee and Jessell, 1999). The secretion of BMPs by the epidermal ectoderm and the roof plate is required for the induction of neural crest cells and dorsal interneurons (Basler et al., 1993; Liem et al., 1995, 1997; Lee et al., 1998, 2000; Barth et al., 1999; Nguyen et al., 2000). Although studies on the roles of BMPs in the spinal cord have focused primarily on the control of dorsal cell fates, prospective ventral neural cells may also be exposed to BMP signals prior to or coincident with their exposure to Shh. Cells of the neural plate differentiate from a region of the ectoderm that initially express high levels of BMPs (Fainsod et al., 1994; Streit et al., 1998; Streit and Stern, 1999), and thus newly formed neural plate cells may be exposed to low levels of BMP protein at stages after BMP gene expression itself has been down-regulated.

BMP and Shh signals appear to have opponent and antagonistic functions in the control of cell fate along the dorsoventral axis of the neural tube. Dorsal neural progenitors normally give rise to neural crest cells and dorsal interneurons but their exposure to Shh inhibits the differentiation of these dorsal cell types (Yamada et al., 1991; Liem et al., 1995). Conversely, the Shh-mediated differentiation of floor plate cells and motor neurons is blocked by exposure to high BMP levels (Basler et al., 1993; Arkell and Beddington, 1997). Thus, the coincident exposure of prospective ventral neural cells to Shh and BMP could regulate the patterning of ventral neural cell types.

BMP signaling is itself subject to regulation by secreted factors that bind to BMPs and block their activity (Smith, 2000). Some of these inhibitory factors have been shown to bind to selective subsets of BMP-related proteins. Noggin binds to and inhibits the BMP2/4 and GDF5-GDF7 subclasses of BMPs (Zimmerman et al., 1996; Lee et al., 1999), whereas follistatin preferentially inhibits members of the BMP5-BMP7 subclass and also inhibits activin signaling (Nakamura et al., 1990; Yamashita et al., 1995; Liem et al., 1997; Iemura et al., 1998). noggin, follistatin and other BMP inhibitors are expressed by axial and paraxial mesodermal cells that flank the neural tube (Graham and Lumsden, 1996; Connolly et al., 1997; Dale et al., 1999), raising the possibility the secretion of these inhibitors by mesodermal tissues regulates the ambient level of BMP signaling within neural cells.

In this study we have examined whether BMP signaling and its control by secreted BMP inhibitors regulates neural cell responses to Shh, and thus might contribute to the patterning of cell types in the ventral neural tube. Using an in vitro assay of neural cell differentiation we show that BMP signaling markedly alters neural cell responses to Shh signals, eliciting a ventral-to-dorsal switch in progenitor cell identity and neuronal fate. Conversely, exposure of neural cells to the secreted BMP inhibitor follistatin produces a dorsal-to-ventral switch in progenitor cell identity and neuronal fate in vitro and in vivo. These results suggest that the specification of ventral cell types normally depends on the integration of Shh and BMP signaling activities. The net level of BMP signaling, and thus the response of neural cells to Shh signaling, in turn may be regulated by the actions of follistatin and other secreted BMP inhibitors supplied by mesodermal cell types than flank the ventral neural tube.

Regions of the intermediate neural plate, ([i]) explants, were isolated from Hamburger and Hamilton (1951) (HH) stage 10 chick embryos as previously described (Yamada et al., 1993). Explants were cultured in a collagen gel (Vitrogen Inc) with Shh-N (Porter et al., 1995; Roelink et al, 1995; Ericson et al., 1996), follistatin (provided by National Hormone and Pituitary Program), noggin (gift from R. Harland), and BMP7-, or BMP4-transfected COS cell supernatant (see Basler et al., 1993; Lee et al., 1998; Alder et al., 1999). A myc-tagged pMT23-BMP7 was constructed using a mouse BMP2 (Derynck) proregion fused to the mature region of chick BMP7 (see Basler et al., 1993). One unit of BMP activity is defined as the threshold dilution of BMP-COS cell supernatant sufficient to induce the migration of >10 neural crest cells from [i] explants at 24 hours in vitro (see Liem et al., 1995). All experiments involving BMP7 activity derived from the same batch of BMP7-transfected COS cell supernatants.

A cDNA encoding Xenopus follistatin (a gift from A. Hemmati-Brivanlou) was cloned into a CMV expression vector, mixed in a ratio of 5:1 with a CMV-GFP expression vector (Clontech) and electroporated in ovo into the neural tube of HH stage 10-12 chick embryos. Embryos were allowed to develop for a further 48 hours, until HH stage 20-22, and then fixed and processed for immunohistochemistry.

Neural explants were processed as for immunocytochemistry as described by Yamada et al. (1993). Antibodies used were: mouse monoclonal anti-Pax6 and anti-Pax7 (Ericson et al, 1996), rabbit or guinea pig anti-Nkx6.1 (Briscoe et al., 1999), anti-Nkx2.2 (Ericson et al., 1997a), anti-HNF3β (Ericson et al., 1997a), anti-En1 (Ericson et al., 1997a), anti-Otx2 (Nothias et al., 1999), anti-Isl1/2 and anti-Isl2 (Tsuchida et al., 1994; Ericson et al., 1992, 1996, 1997). FITC- and Cy3-conjugated secondary antibodies were obtained from Jackson ImmunoResearch Inc. Images were collected on an MRC 1024 confocal microscope or on a Zeiss fluorescence microscope.

Whole-mount in situ hybridization histochemistry was performed as described by Thery et al. (1995) on 30 μm cryostat sections. Chick probes used for in situ hybridization were flik (Patel et al., 1996,), follistatin (Graham and Lumsden., 1996), noggin (Capdevila and Johnson, 1998), and chordin (Dale et al., 1999). A full length flik cDNA was cloned by PCR from the published sequence (Patel et al., 1996).

RT-PCR was performed as described by Tanabe et al. (1995). Primers used were:

S17 ribosomal protein (5′AGAAGGCGGCGCGGGTGATCATCG-3′ and 5′GTTTATTGTAAAAGCAACATAACG3′);

Isl-1 (5′TCAAACCTACTTTGGGGTCTTA3′ and 5′ATCGCCGG-GGATGAGCTGGCGGCT3′);

Ptc (5′TACCACGGGATACACGGGAATA3′ and 5′AATGG-GCTTTGCTCGGTTGC3′).

To begin to address whether the level of BMP activity can influence the response of neural cells to Shh signaling we monitored the expression of markers that define progenitor cell subclasses generated in the ventral neural tube of chick embryos. The identity of neural progenitor cells was defined on the basis of the expression of four homeodomain proteins, Pax7, Pax6, Nkx6.1 and Nkx2.2. Each protein marks a distinct progenitor (p) domain along the dorsoventral axis of the neural tube (see Fig. 1A; Ericson et al., 1997a; Briscoe et al., 1999, 2000). Pax7 is expressed by dorsal progenitor cells but is excluded from all ventral progenitors. Pax6 is expressed by ventral progenitors within the p0, p1, p2 and pMN progenitor domains. Nkx6.1 is expressed by cells within the p2, pMN and p3 progenitor domains. Nkx2.2 is expressed selectively by cells in the p3 progenitor domain (Fig. 1A). We also monitored the expression of HNF3β, a winged helix transcription factor expressed by floor plate cells (Ruiz i Altaba et al., 1995; Fig. 1A).

Previous studies have shown that the expression of Nkx6.1, Nkx2.2 and HNF3β can be induced in vitro at different Shh-N concentration thresholds. Nkx6.1 induction requires the lowest Shh-N level, Nkx2.2 a two-to three-fold higher level, and HNF3β a further two-to three-fold increase in Shh-N (Roelink et al., 1995; Ericson et al., 1997a; Briscoe et al., 1999; 2000 data not shown). Conversely, the expression of Pax7 is inhibited by Shh-N at a concentration approx. four-fold lower than that required for repression of Pax6 (see Ericson et al., 1997a). Thus, the differential patterns of expression of these five transcription factors serve as indicators of the response of neural progenitor cells to different levels of Shh signaling (Ericson et al., 1997b; Briscoe et al., 2000).

To test the influence of BMP signaling on the Shh-mediated control of neural progenitor cell identity we examined the expression of each of these transcription factors in HH stage 10 chick intermediate neural plate [i] explants (Yamada et al., 1993) grown in a defined concentration of Shh-N, with or without added BMPs. In these assays we routinely used BMP7 as a source of BMP, since all BMPs tested on [i] explants have qualitatively similar inductive activities (Liem et al., 1995; Lee et al., 1998). Consistent with these observations, we obtained similar results upon addition of BMP7 and BMP4 (data not shown).

In [i] explants grown alone for 24 hours, virtually all cells expressed Pax7 and Pax6, (Fig. 1B; data not shown), but expression of Nkx6.1, Nkx2.2 and HNF3β was not detected (Fig. 1C; data not shown; see Ericson et al., 1997b). Exposure of [i] explants to 2 nM Shh-N blocked Pax7 expression completely (Fig. 1D), permitted low level expression of Pax6 by virtually all cells (data not shown; Ericson et al., 1997a) and induced Nkx6.1 expression in ∼30% of cells (Fig. 1E). At this Shh-N concentration, no expression of Nkx2.2 and HNF3β was detected (data not shown; Ericson et al., 1997a). In contrast, in the presence of both 1.0 Unit BMP7 and 2 nM Shh-N, virtually all cells in [i] explants expressed Pax7 (Fig. 1F) and high levels of Pax6 (data not shown), and no expression of Nkx6.1 expression was detected (Fig. 1G). Thus, BMP signaling changes the response of progenitor cells to a fixed concentration of Shh, resulting in the conversion of progenitor cells from a ventral to a dorsal identity.

We next examined whether, in addition to causing a complete shift from ventral to dorsal progenitor identity, BMP exposure can also alter the subtype identity of ventral progenitor cells generated in response to a set level of Shh signaling. We defined a concentration of Shh-N (2.4 nM) that suppressed Pax7 expression, generated few Pax6⁺ cells and many Nkx2.2⁺ ventral progenitor cells (Fig. 2A-C), but resulted in the generation of few if any HNF3β⁺ cells (Fig. 2D). Coincident exposure of [i] explants to 0.25 Units BMP7 and 2.4 nM Shh-N resulted in an approx. 20-fold increase in the number of Pax6⁺ progenitor cells (Fig. 2F), and an approx. ten-fold reduction in the number of Nkx2.2⁺ progenitor cells (Fig. 2G). With this combination of factors, no Pax7⁺ cells were detected (Fig. 2E), indicating that progenitor cells have retained a generic ventral identity. These findings show that BMP signaling can change the subtype identity of ventral progenitor cells generated in response to a set level of Shh signaling activity.

Coincident exposure of newly generated caudal neural plate cells to Shh and BMPs has been shown to induce midline forebrain neural fates, at least in rat tissue (Dale et al., 1997). We were concerned therefore that cells induced in HH stage 10 chick [i] explants by coincident BMP and Shh exposure might lose their caudal identity, thus complicating interpretation of the observed changes in neural marker expression. However, coincident exposure of stage 10 chick [i] explants to Shh (4 nM) and BMP (1.0 Unit) did not result in expression of the anterior neural markers Otx2 and Nkx2.1 (Acampora et al., 1999; Sussel et al., 1999) (data not shown). Moreover, the Isl1⁺ neurons induced by this factor combination also coexpressed Isl2, a LIM homeodomain protein that is restricted to somatic motor neurons generated at caudal levels of the neuraxis (Ericson et al., 1997a; Varela-Echavarria et al.,1996; data not shown). These results indicate that the changes in neural marker expression by cells in stage 10 [i] explants exposed to Shh in the presence of BMPs result from an alteration in dorsoventral and not rostrocaudal identity.

Is the switch in the identity of ventral progenitor cells elicited by coincident BMP and Shh signaling accompanied by a change in neuronal fate? To test this, we assayed the generation of two classes of ventral neurons, motor neurons and V1 interneurons (Fig. 1A). In vitro, the level of Shh activity required to induce motor neurons is four-to six-fold greater than that required to induce V1 neurons (Ericson et al., 1997b). Exposure of [i] explants to 2 nM Shh-N generated approx. 450 Isl1/2⁺ motor neurons and approx. 20 En1⁺ V1 neurons (Fig. 3A,C). At this Shh-N concentration, no HNFβ⁺ floor plate cells were induced (see Fig. 6R). In the presence of 0.25 Units BMP7 and 2 nM Shh-N, the the number of Isl1/2⁺ motor neurons was reduced by approx. ten fold and the number of En1⁺ V1 neurons was increased approx. two to three fold (Fig. 3B,C). We detected no change in overall neuronal generation in the presence and absence of BMP7, as assessed by the number of neurons that express NeuN, a cell-type-independent neuronal marker (Mullen et al., 1992) (2 nM Shh-N: 212±25 NeuN⁺ neurons/explant; 2 nM Shh-N⁺ and 0.25 Units BMP7: 218±24 NeuN⁺ neurons, mean ± s.e.m., n=5 explants). Thus, the subtype identity of ventral neurons can also be switched in a ventral-to-dorsal direction by increasing the level of BMP signaling in the presence of a fixed level of Shh activity.

The ability of BMPs to switch the identity of ventral progenitor cells and postmitotic neurons in response to Shh signaling raised the question of the site of convergence of these two signaling pathways in responsive neural cells. We first examined whether BMPs modify cell responses to Shh by altering the availability or presentation of the Shh protein itself, or by modifying the response of neural target cells. We reasoned that the ability of BMPs to modify the fate of ventral neural cells under conditions in which Shh is no longer able to interact with its receptor would provide evidence that BMPs act on neural target cells. To test this idea we isolated HH stage 12 ventral neural explants containing a floor plate ([vf] explants) and grew them in the presence of a function-blocking anti-Shh monoclonal antibody (mAb 5E1) directed against the receptor-binding surface of Shh (Fuse et al., 1999; Pepinsky et al., 2000), with or without added BMP (Fig. 4A). Many ventral cells in stage 12[vf] explants grown alone expressed Nkx2.2 but not Pax7, indicating that these cells have been exposed to a high concentration of Shh (Fig. 4A). Stage 12 [vf] explants grown in the presence of anti-Shh IgG, without added BMP, lacked Nkx2.2 expression but did not express Pax7 (Fig. 4A). This result indicates that neural cells have been exposed to a level of Shh signaling sufficient to repress Pax7 but below the level required for the induction of Nkx2.2. Cells in stage 12 [vf] explants grown with anti-Shh IgG, in the additional presence of BMP7, also lacked Nkx2.2 expression, but under these conditions expressed high levels of Pax7 (Fig. 4A). These findings show that BMP7 can modify the response of ventral neural cells in the absence of available extracellular Shh protein, supporting the idea that BMPs act on neural cells to modify their response to Shh at a step downstream of the availability and presentation of extracellular Shh protein.

We next addressed whether BMPs modify a proximal step in the transduction of Shh signals within target neural cells. To test this possibility we examined the effect of BMPs on the Shh-induced expression of patched (ptc), a gene encoding a Shh receptor subunit that is likely to be induced as a direct response to Shh signaling (Marigo et al., 1996; Goodrich et al.,1996; Ingham, 1998). Low levels of ptc were detected in [i] explants both at the time of isolation and after culture for 48 hours (Fig. 4B; data not shown). The addition of 1 nM Shh to [i] explants for 48 hours resulted in an approx. ten-fold elevation in the level of ptc expression (Fig. 4A). In contrast, the coincident addition of 1 nM Shh-N and 1.0 Unit BMP7 for 48 hours resulted in a level of ptc expression only slightly greater than that obtained in [i] explants grown in the absence of Shh-N (Fig. 4B). Since ptc is likely to be a direct target of Shh signaling, the ability of BMPs to block the Shh-induced elevation in ptc expression provides one line of evidence that the modulatory action of BMPs on neural cell responses to Shh is exerted at a proximal step in the Shh signal transduction pathway.

The HNF3β gene has also been suggested to be a direct transcriptional target of Shh signaling in neural cells (Ruiz i Altaba et al., 1995). Consistent with this idea, the HNF3β promoter contains functional binding sites for Gli proteins, a class of zinc finger transcription factors that mediate certain of the signaling functions of Shh (Sasaki et al., 1997). We therefore addressed whether BMP signaling can also block the Shh-induced expression of HNF3β. Exposure of [i] explants to 4 nM Shh-N resulted in the induction of many HNF3β⁺ floor plate cells (Fig. 4D), and also in the generation of many Nkx2.2⁺ progenitors (Fig. 4C). Addition of 1.0 Unit BMP7 and 4 nM Shh-N to [i] explants almost completely blocked the induction of HNF3β⁺ cells (Fig. 4F), whereas many Nkx2.2⁺ cells were still generated (Fig. 4E). The ability of BMPs to block of the Shh-mediated induction of HNF3β expression provides further evidence that BMP signals modify neural cell responses to Shh through actions at a proximal step in the Shh signal transduction pathway.

The ability of BMPs to switch the fate of neural cells in response to Shh signaling led us to consider whether secreted BMP inhibitors might have a role in regulating the response of neural cells to Shh signaling. To begin to define which secreted BMP inhibitors might influence the initial patterning of ventral cell types, we analyzed the expression of genes encoding four secreted BMP inhibitors, follistatin, flik, chordin and noggin, at caudal levels of the neuraxis in stages 10-15 embryos.

follistatin was expressed at high levels by the unsegmented paraxial mesoderm adjacent to the caudal neural plate and neural folds (Fig. 5A) and was detected at high levels in the ventral region of the somitic mesoderm (Fig. 5B-D). The expression of follistatin by the somitic mesoderm persisted until at least stage 15 (data not shown). The expression of follistatin was also detected in the notochord, from stage 10 onwards (Fig. 5C; data not shown). Expression of flik, a follistatin-related gene, was detected at high levels in the notochord at an axial level underlying the closing neural plate (Fig. 5E). flik was also expressed transiently by cells at the midline of the neural plate at stage 10 (Fig. 5E), but not at later stages (data not shown; see Patel et al., 1996; Amthor et al., 1996). Both noggin and chordin were expressed by newly formed notochord (Fig. 5F,G; data not shown) and expression of both genes persisted until at least stage 15 (see Dale et al., 1999). Neither noggin nor chordin were expressed by cells in the paraxial mesoderm at these stages (data not shown), but chordin was expressed in midline cells of the neural plate and ventral neural tube in stage 10 embryos (data not shown; see Dale et al., 1999). These results extend previous observations on the expression of these genes in the chick embryo (Graham and Lumsden, 1996; Dale et al., 1999; Patel et al., 1996; Connolly et al., 1997). Thus, over the period that neural cells acquire their ventral positional identities, the genes encoding secreted factors that bind to and inhibit the signaling activity of BMPs and other TGFβ family members are expressed by both axial and paraxial mesodermal cells.

To test the role of secreted BMP inhibitors in ventral neural patterning we examined whether follistatin and noggin, two BMP inhibitors with distinct binding specificities, regulate the pattern of cell differentiation in [i] explants. We first examined whether the action of these BMP inhibitors alone, in the absence of Shh, was sufficient to induce the differentiation of ventral neural cell types. Addition of follistatin and noggin, alone, or in combination, to [i] explants did not suppress the expression of Pax7 and did not result in the induction of Nkx6.1, Nkx2.2 or HNF3β expression (data not shown; see also Pierani et al., 1999). Thus, these two BMP inhibitors do not mimic the ability of Shh to induce ventral progenitor cell types in vitro.

We next examined the effects of follistatin and noggin on responses to Shh signaling in [i] explants. Cells in [i] explants were exposed to a concentration of Shh-N (125 pM), that alone, was below the threshold for repression of Pax7 expression (Fig. 6A), and was insufficient to induce Nkx6.1, Nkx2.2 or HNF3β expression (Fig. 6B,C; data not shown). In contrast, addition of 125 pM Shh-N in the presence of 80 pM follistatin decreased by approx. ten fold the number of cells that expressed Pax7 (Fig. 6D) and induced the expression of Nkx6.1 in approx. 20% of neural cells (Fig. 6E). No induction of Nkx2.2 or HNF3β expression was detected (Fig. 6F; data not shown). This result shows that follistatin is able to modulate the sensitivity of progenitor cells to Shh signaling, converting dorsal to ventral progenitors. However, addition of Shh-N (125 pM) in the presence of noggin (80 pM: a concentration sufficient to block both BMP4 and epidermal ectoderm-derived signals, Liem et al., 1997), failed to repress Pax7 and did not induce the expression of Nkx6.1, Nkx2.2 or HNF3β (Fig. 6G-I; data not shown).

We next examined whether follistatin can alter the subtype identity of ventral progenitor cells generated in response to Shh signaling. To generate ventral progenitors we exposed [i] explants to 1 nM Shh-N, a concentration at which Pax7 was completely repressed (data not shown), at which Pax6 expression persisted at high levels (Fig. 6J), and at which neither Nkx2.2 nor HNF3β were induced (Fig. 6K,L). Addition of Shh-N (1 nM) in the presence of follistatin (80 pM) resulted in an approx. 20-fold decrease in the number of Pax6 cells (Fig. 6M), and led to the expression of Nkx2.2 by approx. 30% of cells (Fig. 6N). At these factor concentrations, only very few cells expressed HNF3β (Fig. 6O). In a separate set of experiments we exposed [i] explants to 2 nM Shh-N; a concentration that permitted expression of Pax6 (Fig. 6P; Ericson et al., 1997), but again was below the threshold for induction of Nkx2.2 or HNF3β (Fig. 6Q,R). Addition of 2 nM Shh-N in the presence of follistatin (80 pM) resulted in a virtually complete repression of Pax6 (Fig. 6S), and in the induction of Nkx2.2 expression by approx. 75% of neural cells (Fig. 6T). At these factor concentrations many cells were also induced to express HNF3β (Fig. 6U). Thus, follistatin can markedly enhance the response of neural cells to a set level of Shh signaling, as assessed by dorsal-to-ventral switch in the subtype identity of ventral progenitor cells.

In this latter set of experiments, the detection of HNF3β⁺ cells raised the issue of whether the secondary induction of floor plate cells, and accompanying Shh expression, could contribute to this change in ventral progenitor cell identity. However, the present experiments were analyzed at 24 hours and the induction of Shh expression in vitro is not detected before this time (Tanabe et al., 1995; Roelink et al., 1995; Pons and Marti, 2000). Thus, it is unlikely that induction of floor plate cells and the secondary neural expression of Shh contributes significantly to the observed changes in progenitor cell fate detected at 24h. Addition of noggin (80 pM) did not affect the response of cells to either 1 nM or 2 nM Shh-N (data not shown; see below).

We next determined whether follistatin exposure alters the subtype identity of ventral neurons generated in response to a set level of Shh signaling. To test this possibility, [i] explants were exposed to 0.5 nM Shh-N, in the absence or presence of follistatin, for 48h. Exposure of [i] explants to Shh-N alone generated approx. 250 En1⁺ V1 neurons, but only approx. 25 Isl1/2⁺ motor neurons, and no HNF3β⁺ floor plate cells (Fig. 7A,B,G). In contrast, [i] explants grown in the presence of 0.5 nM Shh-N and 80 pM follistatin generated approx. three-fold fewer En1⁺ V1 neurons and an eight-fold greater number of Isl1/2⁺ motor neurons (Fig. 7C,G). At this follistatin concentration, no HNF3β⁺ cells were induced (Fig. 7D,G), indicating that the change in subtype identity of ventral neurons is independent of floor plate induction. Addition of Shh-N (0.5 nM) in the presence of a higher concentration of follistatin, (400 pM), resulted in an almost complete suppression in the generation En1⁺ V1 neurons (Fig. 7E,G), and was associated with a further increase in the number of Isl1/2⁺ motor neurons (Fig. 7E,F,G). Moreover, at this concentration of follistatin, many HNF3β⁺ cells were induced (Fig. 7F,G). Thus coincident exposure of cells to Shh and follistatin can promote the generation of ventral cell types induced by an approx. ten-fold higher concentration of Shh alone (see Ericson et al., 1997b). Together, these findings provide evidence that modulation of endogenous BMP signaling within neural tissue by follistatin in vitro dramatically alters the identity of ventral progenitor cells and postmitotic neurons generated in response to a set level of Shh protein.

To examine whether the modulation of BMP signaling also affects ventral neural cell fates in vivo, we ectopically expressed follistatin in the chick neural tube. Vectors directing expression of follistatin and GFP were co-electroporated into one side of HH stage 10-12 chick neural tube, and embryos were permitted to develop in ovo until stages 20-22. Regions of ectopic follistatin expression were marked by expression of GFP (Fig. 8A and data not shown). The expression of Shh, the progenitor cell markers Pax7, Pax6, Nkx2.2, and the neuronal subtype markers Isl1/2, Lim1/2 and LH2A/B were examined.

We detected an approx. 70% decrease in the number of LH2A/B⁺ (D1) interneurons in the dorsal neural tube (Fig. 8C). This result provides evidence that dorsal neural expression of follistatin attenuates BMP signaling from the roof plate that is necessary for the induction of LH2A/B⁺ D1 neurons (Liem et al., 1997; Lee et al., 2000). The expression of Shh by floor plate cells was unaffected by unilateral follistatin expression (Fig. 8B). In regions of the ventral neural tube within which follistatin was ectopically expressed, we detected a marked dorsal-to-ventral shift in the identity of ventral neural progenitor cells and in neuronal fate (Fig. 8). On the electroporated side of the neural tube, the domain occupied by progenitor cells expressing low levels of Pax6 was shifted dorsally (Fig. 8D). In addition, the ventral boundary of Pax7 expression was shifted dorsally by three to five cell diameters (Fig. 8E). The ventralization of progenitor cell identity was accompanied by a similar ventralization in neuronal fate. We detected a dorsal expansion in the position of MN generation and a concomitant dorsal restriction in the position of generation of ventral Lim1/2⁺ (V0 and V1) interneurons (Fig. 8F; see Pierani et al., 1999). Together these findings provide evidence that follistatin expression in vivo results in a marked ventralization of ventral neural cell fates, in a manner independent of any change in Shh expression. These results support the idea that the modulation of endogenous BMP signaling alters the response of ventral progenitor cells to Shh signaling in vivo.

During the patterning of the neural tube, the level of Shh signaling activity appears to have a critical role in determining the identity of ventral progenitor cells and the fate of postmitotic neurons (Ericson et al., 1997b; Briscoe et al., 1999, 2000). The mechanisms that determine how neural cells respond to specific levels of Shh activity, however, remain poorly defined. The response of neural cells might be a direct reflection of the local concentration of active Shh protein within a specific domain of the ventral neural epithelium. Alternatively, cellular responses could depend on the actions of factors that activate distinct signaling pathways that modulate the Shh-signaling pathway within responsive cells. In this study we provide evidence that BMP signaling plays such a role. Our results lead to three main conclusions: (1) that prospective ventral neural plate cells are exposed to ongoing BMP activity, (2) that BMP signaling attenuates Shh signaling, and (3) that follistatin modulates BMP activity and sensitizes the response of neural cells to Shh. We discuss these findings in the context of the role of regulated BMP signaling in ventral neural patterning.

One conclusion that emerges from our studies is that the response of ventral neural progenitors to a specific level of Shh signaling activity is dependent on ambient BMP signaling. The exposure of neural progenitor cells to a fixed Shh concentration in the presence of BMPs causes a marked ventral-to-dorsal shift in the identity of neural progenitor cells and postmitotic neurons. Exposure of neural cells to BMPs has been shown previously to block the differentiation of motor neurons in vitro and floor plate cells in vivo (Basler et al., 1993; Liem et al., 1995; Arkell and Beddington, 1997), but these studies did not examine the identity of the cell types generated in response to BMP exposure.

The patterning roles of BMPs have been analyzed in most detail in the dorsal half of the neural tube (Lee and Jessell, 1999). Nevertheless, it seems likely that prospective ventral neural plate cells are also exposed to BMPs. Two lines of evidence support this idea. First, cells in neural plate explants initially express BMPs at very low levels (Liem et al., 1995; Streit et al., 1998), yet there is a marked elevation in BMP expression when neural explants are grown alone in vitro (Liem et al., 1995, 1997; our unpublished data). Since BMPs markedly induce BMP gene expression in neural cells (Liem et al., 1995), the elevation in BMP expression is likely to be initiated by BMP protein present in neural plate tissue at the time of its isolation. Second, and more persuasively, follistatin changes the response of neural cells to Shh signaling both in vitro and in vivo, eliciting a pronounced dorsal-to-ventral shift in progenitor cell identity and neuronal fate. The analysis of BMP mutant phenotypes in zebrafish embryos has also revealed an expansion in the domain of expression of ventral neural markers (Barth et al., 1999; Nguyen et al., 2000), consistent with the idea that the fate of cells in prospective ventral regions of the neural plate is regulated BMP signaling at neurula stages.

What might be the source of BMP protein in early neural plate tissue? Several BMP genes are expressed by ectodermal cells immediately preceding their differentiation into neural tissue (Fainsod et al., 1994; Streit et al., 1998; Streit and Stern, 1999). Thus, BMP protein is likely to persist in newly induced neural cells at levels sufficient to influence the patterning of early neural cells. In addition, certain BMP family members are expressed at later stages by ventral neural tube cells themselves (Bruneau and Rosa, 1997 Jones et al., 1994). Such ventral neural sources of BMPs could have a role in attenuating neural responses to Shh signaling after neural tube closure. BMPs are also expressed by roof plate cells at the dorsal midline of the neural tube (see Lee and Jessell, 1999). In mouse, genetic ablation of the roof plate eliminates its resident BMPs and causes a minor expansion in the domains of expression of ventral neural markers (Lee et al., 2000). Thus, BMPs derived from dorsal midline cells may have a minor influence on ventral patterning at stages after neural tube closure.

The mechanism by which BMPs regulate neural cell responses to Shh signaling is not known. Our findings indicate that BMPs act on target cells to modify their responses to Shh signaling, rather than regulating the availability or presentation of Shh protein itself. In addition, exposure of neural cells to BMPs blocks the Shh-mediated induction of HNF3β and Ptc, two genes that are thought to be induced as a direct response to hedgehog signaling (Ruiz i Altaba 1995; Sasaki et al., 1997; Ingham, 1998). Thus, BMPs are likely to act at a proximal step in the transduction of Shh signals within neural cells. BMP signal transduction typically involves the phosphorylation and nuclear translocation of Smad effector proteins (Whitman, 1998). Recent evidence suggests that Smads are sequestered in a protein complex that contains Gli proteins, a class of zinc finger transcription factors that transduce many Shh signals (Liu et al., 1999). Thus, Shh and BMP signaling may converge in neural cells at the level of a transcriptional regulatory complex that contains both Smad and Gli proteins.

The ability of BMPs to regulate the response of neural progenitor cells to Shh signaling raises the issue of how the net level of BMP activity is set within neural tissue. One prominent method for regulating BMP signaling is through the secretion of proteins that bind directly to BMPs or to BMP-receptor complexes, thus attenuating downstream transduction events (Piccolo et al., 1996; Zimmerman et al., 1996; Iemura et al., 1998).

Follistatin produces a dorsal-to-ventral shift in the response of neural plate cells to Shh signaling. In contrast, the exposure of neural plate cells to noggin, a BMP inhibitor with differing target specificity, did not change the response of neural progenitors to Shh. Follistatin inhibits activin and the BMP5-BMP7 subgroup, whereas noggin preferentially blocks signaling by the BMP2/4 and GDF5-GDF7 subgroups (Nakamura et al., 1990; Yamashita et al., 1995; Lee et al., 1998). Thus, our results suggest that in chick, neural plate cells are exposed primarily to follistatin-sensitive members of the TGFβ superfamily.

Genetic studies in mice have provided support for the idea that BMP antagonists play a role in ventral patterning in vivo. In noggin null mice there is a marked defect in ventral patterning, most notably, the absence of floor plate cells and motor neurons at caudal axial levels (McMahon et al., 1998). This defect in ventral neural differentiation is associated with the ectopic expression of Bmp4 within the ventral neural tube (McMahon et al., 1998). Thus, in mouse, noggin appears to function normally to prevent ventral neural expression of Bmps, in effect sensitizing neural cells to Shh signals. In chick neural tissue grown in vitro, however, noggin failed to modify the sensitivity of neural cells to Shh. The differing activities of noggin suggested by studies in chick and mouse could be related to the expression in these two species of a different spectrum of BMPs in and around the caudal neural tube (Liem et al., 1995, 1997; Lee et al., 1998).

Although ectopic expression of follistatin in chick neural tube cells in vivo markedly changes ventral neural pattern, the analysis of follistatin null mice has not revealed developmental defects indicative of a profound alteration in early neural patterning (Matzuk et al., 1995). One possible explanation for the lack of a developmental phenotype in follistatin mutants is that noggin serves as the predominant BMP inhibitor in mice, at least at this developmental stage. An alternative explanation is that flik, a follistatin-related gene expressed by the notochord and by midline neural cells (Patel et al., 1996), may have activities redundant with those of follistatin. Chick embryos treated in ovo with flik antisense oligonucleotides have been reported to lack a floor plate (Towers et al., 1999), a patterning defect that is consistent with a role for flik in promoting ventral cell type differentiation in the chick embryo.

Our results do not exclude that factors other than BMPs also contribute to the control of neural cell responses to Shh activity. Shh signaling has been shown to induce the ventral neural expression of HIP, a gene encoding a surface membrane protein that binds to Shh and appears to attenuate its signaling activity in non-neural tissues (Chuang and McMahon, 1999). Vitronectin, an extracellular matrix protein is also induced by Shh in ventral neural cells (Martinez-Morales et al., 1997), and has been reported to bind to Shh and regulate its activity (Pons and Marti, 2000). Neurally expressed peptides have also been suggested to regulate Shh signaling, through their ability to elevate protein kinase A activity (Hynes et al., 1995; Epstein et al., 1996; Fan et al., 1995; Waschek et al., 1998).

Finally, our findings may be relevant to the general issue of how graded Shh signaling activity within the ventral neural tube controls the specification of neural cell fate. In particular they raise the possibility that regulated BMP signaling has a role in establishing a ventral-to-dorsal gradient of Shh activity within the neural tube. Several BMP inhibitors are expressed by cells in or around the ventral neural tube (Amthor et al., 1996; Connolly et al., 1997; Dale et al., 1999). The ventrally restricted expression of these BMP inhibitors would be expected to reduce the net level of BMP signaling in more ventral regions of the neural tube. Such ventral sources of BMP inhibitors may serve to sharpen a ventral-to-dorsal gradient of Shh signaling activity within the ventral neural tube. The establishment of a steep gradient of Shh activity is likely to be a necessary step in ventral patterning, given the evidence that five distinct neuronal subtypes are generated in the ventral neural tube in response to progressive two-to three-fold changes in the level of Shh activity (Ericson et al., 1997b). Thus, the interplay between Shh, BMPs and BMP inhibitors may be crucial in establishing a dorsoventral gradient of Shh signaling activity that underlies the generation of neuronal diversity within the ventral neural tube.

We thank B. Han for COS cell transfections, S. Brenner-Morton for antibodies, A. Graham for chick follistatin, K. Lee for the Chordin in situ picture, R. Johnson for chick noggin, K. Basler for chick BMP7, R. Harland and J. De Jesus-Escobar for Xenopus noggin protein, and the National Hormone and Pituitary Program for follistatin. We thank A. Hemmati-Brivanlou for discussions, K. Lee and B. Novitch for advice and comments on the paper and K. MacArthur and I. Schieren for help in its preparation. K. L. was supported by NIH Grant 5T32GM07367 (Medical Science Training Program) and by Columbia University. J. B. was a Research Associate of the Howard Hughes Medical Institute. T. M. J. is supported by grants from the NIH and is an Investigator for the Howard Hughes Medical Institute.