Positive and negative regulations by FGF8 contribute to midbrain roof plate developmental plasticity

The roof plate (RP) is a specialized structure that extends on the dorsal midline of the neural tube along its entire anteroposterior axis. The RP forms at the site of neural fold closure during development(Liem et al., 1995). It constitutes a signaling center that influences dorsoventral patterning of the neural tube, specification of dorsal neuronal types, and axonal guidance across the dorsal midline (Liem et al.,1997; Lee et al.,1998). RP development has been best studied in the spinal cord,where its specification relies on interactions between inductive signals of the TGFβ family produced by the adjacent epidermal ectoderm and intrinsic homeodomain transcription factors (Liem et al., 2000; Chizhikov and Millen, 2004b; Chizhikov and Millen, 2004c). Less is known about RP development in the anterior brain, where modulations of intrinsic or extrinsic factors involved in neural tube patterning along the AP axis could influence local RP differentiation(Bach et al., 2003). In chick embryos, the RP of the midbrain is highly plastic. Experimental perturbations of the midbrain neuroepithelium that do not directly involve the dorsal midline result in the disappearance of the RP at later stages or in its reorientation or bifurcation (Marin and Puelles, 1994; Bally-Cuif and Wassef, 1994; Crossley et al.,1996; Alexandre and Wassef,2003).

Growth and patterning of the midbrain-hindbrain (MH) domain of the neural tube depend on the activity of a signaling center called the isthmic organizer(IsO), located at the constriction or isthmus that links the midbrain and hindbrain. Fibroblast growth factor 8 (FGF8), a diffusible molecule secreted by the IsO, is one of the major mediators of its organizing and growth-inducing properties (Crossley et al., 1996; Martinez et al.,1999). Insertion of a FGF8-soaked bead in the MH domain or the posterior forebrain induces the formation of a supernumerary IsO, the development of an ectopic MH junction and also of an ectopic RP. The influence of the IsO, whether endogenous or FGF8 bead induced, on the midbrain RP is puzzling. It seems to promote RP elongation and maturation when an ectopic source of FGF8 is inserted into the midbrain(Bally-Cuif and Wassef, 1994; Crossley et al., 1996) or to impair its differentiation when rotation of the midbrain vesicle brings the anterior RP closer to the IsO or when the isthmic node is ablated(Marin and Puelles, 1994; Alexandre and Wassef,2003).

In the spinal cord, BMP signaling controls RP formation through the induction of the competence factors Lmx1a and Lmx1b(Chizhikov and Millen, 2004a; Chizhikov and Millen, 2004b; Chizhikov and Millen, 2004c). Wnt1 has also been implicated in later aspects of RP differentiation(Shimamura et al., 1994; Amoyel et al., 2005). Lmx1b and Wnt1 are also targets of FGF8 signaling widely expressed in the caudal midbrain. At later stages of development, Lmx1b and Wnt1 expression becomes confined to the RP and floor plate and to a ring of cells marking the MH boundary. A cue to the unexplained behavior of the midbrain RP may therefore reside in crossregulatory interactions between the BMP and FGF pathways on common downstream genes essential for RP development. The aim of the present study was to understand better these interactions. We find that FGF8 orchestrates midbrain RP differentiation by finely adjusting BMP, GDF and activin signaling in time and space.

White Leghorn chick (Morizeau) and Japanese quail (La Caille de Chanteloup)embryos were operated on between somite stages (ss) 9 and 14, and fixed 1-3 days later between Hamburger and Hamilton stages 14 and 23 (HH 14-23)(Hamburger and Hamilton,1951). Some embryos were also examined after long survival, mainly 4.5, 7 and 12 days after surgery. The methods for performing small ablations,homotopic and heterotopic isochronic grafts(Bally-Cuif and Wassef, 1994; Alexandre and Wassef, 2003),bead implantation (Martinez et al.,1999), in situ hybridization, immunocytochemistry and cryostat sections (Bally-Cuif and Wassef,1994) were as previously described with minor modifications. Rabbit anti-active caspase 3 (Promega) was diluted 1/250, mAb QCPN(Developmental studies hybridoma bank) was diluted 1/10 and mouse anti-AFP(Q-biogene) was diluted 1/2000.

The following vectors were used for in situ hybridization: ChWnt1and QWnt1 are species specific, although they weakly crossreact(Bally-Cuif and Wassef, 1994); Gdf7 (Lee et al.,1998) (gift from K. Lee), Lmx1b(Matsunaga et al., 2002) (gift from N. Nakamura), follistatin(Graham and Lumsden, 1996)(gift from F. Giudicelli), noggin(Hirsinger et al., 1997) (gift from C. Freitas), Id3 (Kee and Bronner-Fraser, 2001) (gift from M. Bronner-Fraser), and ActA and ActB (Merino et al., 1999) (gift from J. M. Hurle).

Heparin acrylic (Sigma) or Affigel Blue (BioRad) beads were soaked in a drop of recombinant protein (R&D, 16-20 beads/μl, unless otherwise specified) at the following concentrations: FGF8, 0.1, 0.2 and 0.4 mg/ml;BMP7, 1 mg/ml; GDF7, 0.5 mg/ml; noggin, 1 mg/ml; follistatin, 0.5 and 1 mg/ml;FGF8/GDF7, 0.2/0.5 mg/ml; activin A, 1 and 2.5 μg/ml). Formatederivatized AG-1 X2 beads (BioRad) were used to deliver the FGFR1 inhibitor SU5402(Promega, 4 mg/ml in DMSO); control beads were soaked in PBS or DMSO. The beads were incubated overnight in a moist chamber at 4°C and rinsed three times before insertion. When heparin acrylic beads soaked in PBS were inserted in the caudalmost region of the midbrain, they sometimes induced a relocation of the MH boundary or widened the Wnt1 expression domain, thus mimicking the effect of an FGF8 bead. We interpreted this behavior as resulting from a change in local FGF8 signaling resulting from the diffusion of heparin, which is known to be a potent co-factor of FGF8, or from a modification of the shape and time course of the gradient of FGF8, which may be accumulated and released by the heparin bead. PBS-soaked heparin acrylic beads had no noticeable effect elsewhere.

The heads of HH14-15 embryos were isolated from the rest of the body at the level of the otic vesicles then separated into two halves by cutting the dorsal and ventral midlines. The two halves remained together, one left unperturbed as control. The caudal part of explants, including the isthmic region, was ablated to remove the endogenous source of FGF8. FGF8 signaling was also modulated through implantation of FGF8, PBS, SU5402 or DMSO beads in the caudal midbrain. The explants were placed ventricular side down on floating membranes (de Diego et al.,2002) and cultured for 6 or 18 hours before fixation and processing for in situ hybridization.

The chick Lmx1b expression vector [pMiw-Lmx1b(Matsunaga et al., 2002); a gift from K. Nakamura] was derived from the pMiwIII vector containing regulatory sequences from the Rous sarcoma virus enhancer and chickenß-actin promoter. In ovo electroporation on stage HH10 chick embryos was performed as described (Funahashi et al.,1999). A GFP-GPI expression vector(Keller et al., 2001) (gift of D. Henrique) derived from pEGFP-N1 (Clontech) was co-electroporated with pMiw-LMX1B to check for the efficiency of transfection.

In the midbrain, an extensive reorganization of RP patterning may be induced by seemingly unrelated experimental conditions, as schematically represented in Fig. 1. We have shown previously (Louvi et al.,2003; Alexandre and Wassef,2003) that FGF8 beads inserted in the midbrain may induce locally a short RP segment. The hypothesis that FGF8 derived from the IsO may initiate development of the endogenous RP is, however, difficult to reconcile with the observation that expression on the midline of several BMP family members is transiently downregulated at the level of the IsO(Louvi et al., 2003). In the midbrain, BMP signaling is still required after neural tube closure for RP differentiation. Beads impregnated with the BMP inhibitor noggin inserted in the dorsal midbrain of HH10 embryos prevent Gdf7 expression on the midline and RP development (7/16, Fig. 2A-A″). However, even in the most affected embryos(Fig. 2A′,A″), a short RP segment still develops at both ends of the midbrain. Thus other factors may compensate for BMP downregulation at the anterior and posterior poles of the midbrain. Local factors on the midline may also potentiate RP marker induction by FGF8. In general, the ectopic RP segment induced by FGF8 links the bead to the dorsal midline (Fig. 2B). When the FGF8 induced RP is reduced to a short segment near the bead (Alexandre and Wassef,2003) (arrows in Fig. 2C), a small deflection is also induced opposite the bead on the midline RP (arrowhead in Fig. 2C). This indicates that, even if dispensable, co-factors on the midbrain dorsal midline potentiate the inductive activity of FGF8 on RP differentiation. To identify genetic factors involved in midbrain RP development, we examined another experimental situation where an ectopic RP is induced.

We examined grafts of naive dorsal midbrain neuroepithelium implanted across the host midline. We have shown previously(Alexandre and Wassef, 2003)that ectopic RP segments were induced in these transplants as extensions of the host RP. The induced RP formed a solid row of cells usually of uniform width. RP induction could be initiated both anteriorly and posteriorly in the graft (16/63, Fig. 3A,A′,A″). In general, the large dorsal midbrain transplants grafted perpendicular to the midline did not grow much(Louvi et al., 2003). Most transplants did not develop a RP structure. In these transplants, Wnt1 expression was maintained or induced in a wide domain comprising the whole graft (9/26, Fig. 3B,B′) or a large part of the graft (11/26, Fig. 3,C,C′,D,D′,D″). Lmx1b expression was upregulated in the grafts (10/10). In the transplants, the domain of Lmx1b expression was always included within the Wnt1 domain(4/4, Fig. 3B). Gdf7and Lmx1b expression coincided in most cases (4/6, not shown). Sometimes (2/6) part of the Lmx1b domain contained only scattered cells expressing Gdf7 (not shown). Induction of an ectopic RP in the transplant was accompanied by a downregulation of this widespread expression of Wnt1 and Lmx1b. Thus, ectopic transplantation stabilizes the expression of the RP competence factors Lmx1b and Wnt1which is normally transient in the lateral midbrain. These factors have been shown essential for RP differentiation in the spinal cord(Chizhikov and Millen, 2004a; Chizhikov and Millen, 2004b; Chizhikov and Millen,2004c).

Lmx1b and Wnt1 are competence factors for RP development in the spinal cord (Chizhikov and Millen,2004a; Chizhikov and Millen,2004b; Chizhikov and Millen,2004c). Both are targets of FGF8 and, in addition to their dorsal(and ventral) midline expression, are expressed in a wide domain of the caudal midbrain (Fig. 4A,B). We wondered whether expression of Gdf7 could be induced by Lmx1b on the anterior midbrain midline. Overexpression of Lmx1b in the dorsal midbrain was obtained by electroporation. Confirming the report of Matsunaga et al.(Matsunaga et al., 2002), Wnt1 but not Gdf7 was already strongly induced 10 hours after Lmx1b electroporation (Fig. 4D,D′,D″). Gdf7 was induced in a few scattered cells 24 hours after Lmx1b electroporation in the midbrain(Fig. 4D). High levels of Gdf7 expression were not detected before 34 hours after electroporation, when Gdf7 was expressed in large cell strands or patches in the lateral midbrain (Fig. 4E,F). Interestingly, Gdf7 induction by Lmx1bwas less efficient in the caudal midbrain (arrowheads in Fig. 4E,E′). Thus,initiation of Gdf7 expression on the anterior midbrain midline could be triggered by Lmx1b and Wnt1, which are both present much earlier on the midline.

In addition to Gdf7, the expression of other markers of the mature roof plate [Bmp5, Bmp7 (Louvi et al., 2003), noggin(Fig. 5A) and Id3(Fig. 5B)] is delayed in the caudal midbrain. Insertion of a FGF8 bead dorsally in the caudal midbrain prolonged this transient downregulation of Gdf7 expression(Fig. 5C-E). This suggests that, in situ, FGF8 not only promotes RP formation but also inhibits its maturation. To test if this dual influence could also be detected during the process of ectopic RP induction by FGF8 beads, we examined the expression of early (Wnt1) and late (Gdf7) RP markers 24 and 30 hours after FGF8 bead insertion. One day after bead implantation, a row of cells prefiguring the ectopic RP expressed Wnt1 along its entire length (in red in Fig. 5F,F′)bridging the midline to the FGF8 bead. Gdf7 expression was intense near the midline but became fainter more distally (in purple in Fig. 5F). In slightly older embryos (Fig. 5G), both Wnt1 and Gdf7 expressions in the induced RP reached the bead. Thus, maturation of the induced and endogenous RP was similarly delayed near the source of FGF8, suggesting that FGF8, in situ, regulates the caudal progression of RP differentiation. As mentioned above, Gdf7 induction by ectopic electroporated Lmx1b was less efficient in the caudal midbrain than elsewhere, which also suggests the existence of an isthmus-derived inhibitory influence that prevents or delays Gdf7induction by Lmx1b.

Thus, FGF8 both promotes RP differentiation by inducing a wide expression of the competence factors Lmx1b and Wnt1 and delays RP differentiation. As illustrated in Fig. 5H, any imbalance between these two FGF8 activities could result in enlargements or gaps in the developing RP structure. The short-range mechanisms of RP auto induction described below may partly obviate this problem.

Because BMP signals are involved at successive stages of RP development, we wondered if a BMP family member could mediate this RP homeogenetic signal.

Insertion of BMP7-soaked beads at stage HH9-10 affected the medial domain of the RP but did not induce Gdf7(Fig. 6A) or Wnt1(Fig. 6B,B′) locally,indicating that BMP7 is not directly involved in the homeogenetic behavior of the RP. The thin sheet of cells that develops between the two RP halves resembled the neural component of the choroid plexus but it did not express choroid plexus markers such as transthyretin (TTR, Fig. 6B,B′) or BMP5 (not shown). Downregulation of BMP signaling through the insertion of a noggin-soaked bead locally impaired RP differentiation(Fig. 6C,C′), but did not result in the formation of a choroid plexus.

In contrast to BMP7, GDF7-soaked beads acted locally. Gdf7expression was induced (Fig. 6D,E) at high (4/16) or moderate (9/16) levels, 24 hours after insertion of a GDF7 bead in the midbrain. The induction could involve the dorsal midline (Fig. 6E) or consist of smaller lateral cell patches(Fig. 6D). Interestingly,ectopic expression of Gdf7 was never detected at 48 hours, indicating that the induction was transient. Lmx1b was faintly or not induced by GDF7 beads (not shown), which could perhaps explain the lack of maintenance of ectopic Gdf7. Thus, autoregulation of Gdf7 could be an important component of the homeogenetic mechanism that promotes the extension of a pre-existing RP (Alexandre and Wassef,2003).

We asked whether FGF8 could widen the competence domain for GDF7 auto-activation. Gdf7 expression was widely induced, though at a low level, by beads soaked in a mixture of FGF8 and GDF7, and the refinement of RP differentiation into a compact linear structure was prevented(Fig. 6G,H). Overexpression of FGF8 and GDF7 together also induced a large number of activated caspase 3-immunoreactive cells (7/7; Fig. 6I), whereas beads soaked in GDF7 (1/5; Fig. 6F) or FGF8 (not shown)did not induce an increase in cell death. Thus, FGF8 signaling, by inducing cell death, precludes a rapid progression of the front of Gdf7-expressing cells, therefore stabilizing it [see the pattern of cell death in the caudal midbrain in Alexandre and Wassef(Alexandre and Wassef, 2003); Fig. 3].

Activins belong to the activin/nodal/TGFβ subgroup of the TGFβsuperfamily. At the difference of BMPs, which signal through SMADs1/5/8, the members of the activin subgroup signal through SMAD2/SMAD3. Two observations point to a possible function of activins in dorsal midbrain development. First, the neural folds of the MH domain express high levels of active SMAD2 in E8.5-E9.5 mouse embryos (de Sousa Lopes et al., 2003). Second, follistatin a high-affinity activin inhibitor, is expressed in a dynamic pattern in the MH domain (see below). We first examined if the pattern of expression of activins between HH9 and HH18 is consistent with a possible function in dorsal midbrain patterning and RP differentiation. Although activin A expression was not detected above background levels, we found that activin B is expressed in the midbrain-forebrain domain with a progressive anterior shift in its domain of expression. Activin B is first expressed at low level throughout the midbrain vesicle at HH10 (Fig. 7A). Its expression increases and becomes restricted to the mid-forebrain junction at stage HH12-13 (Fig. 7B). By stage HH18, activin B expression becomes confined to the pretectum(Fig. 7C). Overexpression of activin through the insertion of activin A-soaked beads induced a local increase in Lmx1b expression 7 and 24 hours later(Fig. 7D,E). Although clear,the induction of Lmx1b expression by activin does not compare with its widespread and extremely rapid (less than 3 hours) upregulation by FGF8(Adams et al., 2000). Two apparently opposite modifications of the RP were observed 2 days after activin bead insertion. In about half of the cases (8/18, Fig. 7F,H,H′), a thin RP bifurcation was induced between the bead and dorsal midline. In 3/18 embryos,the expression of RP markers was destabilized in the caudal midbrain(Fig. 7G). Both phenotypes could be observed in the same embryo (Fig. 7H). At later stages (E7.5, 6 days after bead insertion), most embryos lacked a RP in the posterior midbrain and the locally induced RP could no longer be detected (not shown). Thus, activin could be a major player in midbrain RP lability, because increasing activin signaling positively modulates RP extension but also results in its destabilization.

Follistatin binds activins A and B with 1000- and 100-fold higher affinities, respectively, than BMPs(Thompson et al., 2005), and inhibits their binding to activin receptors. At HH11, a small region at the midbrain forebrain junction is delineated by two follistatin domains(Fig. 7I), flanking an intense midline spot of Gdf7 expression(Fig. 7J). Follistatin expression progresses posteriorly in a decreasing gradient between stages HH11 and HH18 (Fig. 7K). Overexpression of follistatin at HH9-11 did not impair RP differentiation. Affigel Blue beads soaked in follistatin and inserted on the midline slightly widened the Gdf7 expression domain (8/24; Fig. 7L). This modest effect of follistatin overexpression contrasts with that of noggin beads and indicates that follistatin has little influence on BMP signaling in the midbrain. The anterior high/posterior low gradient of expression of follistatin is suggestive of a negative modulation by FGF8. In order to test this, FGF8 signaling in the midbrain was modulated in several ways. When FGF8 beads were inserted in the midbrain at stage HH10, the caudal progression of follistatin expression was prevented around the beads(Fig. 7M). At later survival times, follistatin expression was, however, normally turned on caudally beyond the circular range of action of the FGF8 bead (not shown). Downregulation of FGF8 signaling was obtained through the use of SU5402, a FGF signaling inhibitor. Follistatin expression was increased on the side of SU5402-soaked beads inserted in vivo at stage HH10 (Fig. 7N). The downregulation of follistatin expression around the FGF8 beads could reflect the transformation of the midbrain neuroepithelium into cerebellum and isthmus. In order to manipulate FGF8 signaling in older embryos, we used MH explants. The MH region of the neural tube of stage HH14-15 embryos was cut on the dorsal and ventral midlines, the two halves were cultured side by side on floating membranes, one serving as control. Follistatin expression was increased compared with the control side 6 hours after ablation of rhombomere 1, which suppressed the posterior source of FGF8(Fig. 7O,P). Insertion of beads soaked in SU5402 or DMSO in the posterior midbrain of explants gave similar results as in vivo, although less consistent. Thus, follistatin expression responds rapidly to variations in FGF8 signaling, allowing it to control activin signaling indirectly.

We have previously identified (Alexandre and Wassef, 2003) a patterning mechanism depending on the IsO and mimicked by an FGF8 source that locally induces the formation of a short RP segment. However, a posterior trigger for midbrain RP differentiation was not easy to reconcile with the initial anterior expression of markers of RP maturation. In addition, the long-range RP duplications induced by FGF8 beads or IsO grafts or the failure of RP differentiation observed after 180°rotation of the midbrain vesicle remained unexplained. The aim of the present study was to characterize new regulations involved in midbrain RP formation that could relate to its plasticity. We show here that activin dynamically expressed at the midbrain-forebrain junction acts as a potent modulator of RP differentiation.

RP differentiation (Liem et al.,1995; Liem et al.,1997; Furuta et al.,1997; Bach et al.,2003) (present work noggin bead treatment) and the dorsal midline expression of Lmx1b (Chizhikov and Millen, 2004b; Liu et al.,2004) generally depend on BMP signals. However, FGF8 controls the expression of genes such as Lmx1b and Wnt1, which act as competence factors and general markers of the RP. Changes in FGF8 signaling rapidly up- or downregulate Lmx1b expression in the caudal midbrain. In chick, we confirmed the observation of Adams et al.(Adams et al., 2000) that FGF8-soaked beads broadly increased Lmx1b in less than 3 hours. Conversely, in zebrafish, a 4 hour treatment with SU5402 completely abolished expression of Lmx1b in the caudal midbrain including the dorsal midline (O'Hara et al., 2005). Midline expression of Lmx1b was not affected in the anterior midbrain. Thus, on the caudal midbrain midline, FGF8 is the major regulator of Lmx1b expression which, in addition, is not maintained by BMP signaling. This is consistent with the observation that the level of expression of several ligands and targets of the BMP pathway is transiently downregulated on the caudal midbrain midline(Louvi et al., 2003) (this work). Taken together, these observations indicate that expression of Lmx1b in the caudal midbrain is under the control of FGF8, which locally abolishes the neural folds competence for RP differentiation. Thus,similar to the action of ectopic FGF8 beads, endogenous FGF8 may direct the differentiation of the caudal midbrain RP at the center of the field it organizes, independently of the site of early neural tube closure. It is interesting to note that the same behavior, pattern reorganization around the bead and RP duplication, is also induced by FGF8 beads inserted into the forebrain (Crossley et al.,2001). This strengthens the idea that regulatory mechanisms involved in FGF8 patterning may be co-opted for axis formation.

As schematized in Fig. 5Hand Fig. 8A, the induction of a wide RP competent domain in the caudal midbrain is counterbalanced by the inhibitory activity of FGF8 on a variety of RP differentiation markers. The sharp progression front of RP maturation illustrated in Fig. 5H and Fig. 8A, however, is an oversimplification. Individual RP markers are downregulated at variable distances from the FGF8 source at the MH junction(Louvi et al., 2003). FGF8 induces a battery of negative regulators (reviewed by Thisse and Thisse, 2005) that could slow down the normal process of differentiation in the RP, as it does in the adjacent neuroepithelium. The complex behavior of the midbrain RP suggests that midbrain-specific targets of FGF8, such as E1/E2, which are known to be potent inhibitors, or Pax2, which acts as activator or inhibitor depending on cellular context, could preferentially affect specific steps of RP maturation. First expressed independently of FGF8, Pax2is later regulated by FGF8 signaling(Crossley et al., 1996) and it is required for Lmx1b induction by FGF8(O'Hara et al., 2005). Pax2 is known to interact directly with LMX1B(Marini et al., 2005), which could explain why induction of Gdf7 by Lmx1b is inhibited in the caudal midbrain.

In several systems (Sun et al.,2002; Dubrulle et al.,2001; Delfini et al.,2005; Akai et al.,2005; Mathis et al.,2001; Diez del Corral et al.,2002; Delfino-Machin et al.,2005), a local pool of undifferentiated progenitors is maintained by high levels of FGF signaling, while the establishment of the polarity of the neuroepithelium and the progression of cell differentiation are synchronized by the decreasing gradient in FGF signaling(Lee et al., 1997). RP differentiation seems to be linked to the establishment or maintenance of planar cell polarity in the midbrain. Convergent extension movements induced by FGF8 are observed at the midline(Millet et al., 1996; Louvi et al., 2003; Alexandre and Wassef, 2003). They may be important to prevent discontinuities between the Lmx1bexpression domains controlled by FGF8 and BMPs. Convergent extension also repels the Gdf7-expressing cells beyond the influence of high FGF8 signaling, thus rescuing them from death. It is interesting to note that discontinuities in midbrain polarity induced by neuroepithelium rotation are rapidly regulated in the isthmic region(Martinez and Alvarado-Mallart,1990), but that elsewhere the transplants tend to adopt or maintain the signature of the isthmic midbrain [Lmx1b and Wnt1, this work; Pax2(Vieira et al., 2006)], which is considered to be less differentiated, and thus become competent to develop a RP (see Fig. 8C). However,signals from the host RP, possibly mediated by GDF7, are essential to initiating the formation of an ectopic RP.

Follistatin is often considered to be an inhibitor of BMP signaling, but it plays distinct roles, depending on the context, and interacts with members of several groups of TGFβ family ligands, including BMPs, GDFs and activins. Interestingly, we find that BMPs, GDF and activin ligands all play distinct roles in midbrain RP development. BMPs are involved as RP competence factors,GDF7 in auto-activation of RP differentiation, and activin B in RP expansion and stabilization. Muscle development is controlled by BMP7 and myostatin,both of which are modulated by follistatin, but in different ways. Follistatin binds BMP7 reversibly with low affinity. It converts the muscle growth-inhibiting effect of BMP7 into a strong stimulatory one that is blocked by noggin. Therefore, as follistatin does not prevent BMP7 binding to its receptor (Iemura et al.,1998), it has been suggested that follistatin influences BMP7 binding to its receptors. Follistatin binding to myostatin/GDF8 seems to completely prevent receptor activation(Amthor et al., 2002; Amthor et al., 2004). Because follistatin binds TGFβ ligands with distinct affinities and differentially affects interaction with their receptors, its progressive expansion across the midbrain may be important to modulate their function. Follistatin binds to activin with a much higher affinity than to other TGFβ ligands. The dynamic expression patterns of follistatin and activin suggest that they tightly regulate each other's availability in the midbrain. It remains unclear if the destabilization of the posterior midbrain RP that we observe after insertion of activin beads relates to an increase in activin signaling or whether sequestration by activin impairs other functions of follistatin.

Similar mechanisms are involved in RP differentiation in the anterior midbrain and in the spinal cord. FGF8 signaling prevents their normal deployment in the caudal midbrain through its modulation of the function of several members of the TGFβ superfamily. The need for a progressive transition between these two modes of regulation may increase the vulnerability of RP differentiation to experimental manipulation.

We thank L. Bally-Cuif for critical reading of the manuscript, Boris Barbour for checking the English, Rosette Goiame for technical help and many colleagues for providing plasmids or antibodies. The QCPN monoclonal antibody was obtained from the Developmental Studies Hybridoma Bank. This work was supported by ARC and ACI research grants to M.W. P.A. was supported by fellowships from FCT, Fondation des Treilles and FRM, and I.B. was supported by ARC.